linux/drivers/mtd/ubi/io.c

1447 lines
42 KiB
C
Raw Normal View History

UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* Copyright (c) International Business Machines Corp., 2006
* Copyright (c) Nokia Corporation, 2006, 2007
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
* the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
* Author: Artem Bityutskiy (Битюцкий Артём)
*/
/*
* UBI input/output sub-system.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* This sub-system provides a uniform way to work with all kinds of the
* underlying MTD devices. It also implements handy functions for reading and
* writing UBI headers.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* We are trying to have a paranoid mindset and not to trust to what we read
* from the flash media in order to be more secure and robust. So this
* sub-system validates every single header it reads from the flash media.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* Some words about how the eraseblock headers are stored.
*
* The erase counter header is always stored at offset zero. By default, the
* VID header is stored after the EC header at the closest aligned offset
* (i.e. aligned to the minimum I/O unit size). Data starts next to the VID
* header at the closest aligned offset. But this default layout may be
* changed. For example, for different reasons (e.g., optimization) UBI may be
* asked to put the VID header at further offset, and even at an unaligned
* offset. Of course, if the offset of the VID header is unaligned, UBI adds
* proper padding in front of it. Data offset may also be changed but it has to
* be aligned.
*
* About minimal I/O units. In general, UBI assumes flash device model where
* there is only one minimal I/O unit size. E.g., in case of NOR flash it is 1,
* in case of NAND flash it is a NAND page, etc. This is reported by MTD in the
* @ubi->mtd->writesize field. But as an exception, UBI admits of using another
* (smaller) minimal I/O unit size for EC and VID headers to make it possible
* to do different optimizations.
*
* This is extremely useful in case of NAND flashes which admit of several
* write operations to one NAND page. In this case UBI can fit EC and VID
* headers at one NAND page. Thus, UBI may use "sub-page" size as the minimal
* I/O unit for the headers (the @ubi->hdrs_min_io_size field). But it still
* reports NAND page size (@ubi->min_io_size) as a minimal I/O unit for the UBI
* users.
*
* Example: some Samsung NANDs with 2KiB pages allow 4x 512-byte writes, so
* although the minimal I/O unit is 2K, UBI uses 512 bytes for EC and VID
* headers.
*
* Q: why not just to treat sub-page as a minimal I/O unit of this flash
* device, e.g., make @ubi->min_io_size = 512 in the example above?
*
* A: because when writing a sub-page, MTD still writes a full 2K page but the
* bytes which are not relevant to the sub-page are 0xFF. So, basically,
* writing 4x512 sub-pages is 4 times slower than writing one 2KiB NAND page.
* Thus, we prefer to use sub-pages only for EC and VID headers.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* As it was noted above, the VID header may start at a non-aligned offset.
* For example, in case of a 2KiB page NAND flash with a 512 bytes sub-page,
* the VID header may reside at offset 1984 which is the last 64 bytes of the
* last sub-page (EC header is always at offset zero). This causes some
* difficulties when reading and writing VID headers.
*
* Suppose we have a 64-byte buffer and we read a VID header at it. We change
* the data and want to write this VID header out. As we can only write in
* 512-byte chunks, we have to allocate one more buffer and copy our VID header
* to offset 448 of this buffer.
*
* The I/O sub-system does the following trick in order to avoid this extra
* copy. It always allocates a @ubi->vid_hdr_alsize bytes buffer for the VID
* header and returns a pointer to offset @ubi->vid_hdr_shift of this buffer.
* When the VID header is being written out, it shifts the VID header pointer
* back and writes the whole sub-page.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
#include <linux/crc32.h>
#include <linux/err.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
#include <linux/slab.h>
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
#include "ubi.h"
#ifdef CONFIG_MTD_UBI_DEBUG
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
static int paranoid_check_not_bad(const struct ubi_device *ubi, int pnum);
static int paranoid_check_peb_ec_hdr(const struct ubi_device *ubi, int pnum);
static int paranoid_check_ec_hdr(const struct ubi_device *ubi, int pnum,
const struct ubi_ec_hdr *ec_hdr);
static int paranoid_check_peb_vid_hdr(const struct ubi_device *ubi, int pnum);
static int paranoid_check_vid_hdr(const struct ubi_device *ubi, int pnum,
const struct ubi_vid_hdr *vid_hdr);
#else
#define paranoid_check_not_bad(ubi, pnum) 0
#define paranoid_check_peb_ec_hdr(ubi, pnum) 0
#define paranoid_check_ec_hdr(ubi, pnum, ec_hdr) 0
#define paranoid_check_peb_vid_hdr(ubi, pnum) 0
#define paranoid_check_vid_hdr(ubi, pnum, vid_hdr) 0
#endif
/**
* ubi_io_read - read data from a physical eraseblock.
* @ubi: UBI device description object
* @buf: buffer where to store the read data
* @pnum: physical eraseblock number to read from
* @offset: offset within the physical eraseblock from where to read
* @len: how many bytes to read
*
* This function reads data from offset @offset of physical eraseblock @pnum
* and stores the read data in the @buf buffer. The following return codes are
* possible:
*
* o %0 if all the requested data were successfully read;
* o %UBI_IO_BITFLIPS if all the requested data were successfully read, but
* correctable bit-flips were detected; this is harmless but may indicate
* that this eraseblock may become bad soon (but do not have to);
* o %-EBADMSG if the MTD subsystem reported about data integrity problems, for
* example it can be an ECC error in case of NAND; this most probably means
* that the data is corrupted;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* o %-EIO if some I/O error occurred;
* o other negative error codes in case of other errors.
*/
int ubi_io_read(const struct ubi_device *ubi, void *buf, int pnum, int offset,
int len)
{
int err, retries = 0;
size_t read;
loff_t addr;
dbg_io("read %d bytes from PEB %d:%d", len, pnum, offset);
ubi_assert(pnum >= 0 && pnum < ubi->peb_count);
ubi_assert(offset >= 0 && offset + len <= ubi->peb_size);
ubi_assert(len > 0);
err = paranoid_check_not_bad(ubi, pnum);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* Deliberately corrupt the buffer to improve robustness. Indeed, if we
* do not do this, the following may happen:
* 1. The buffer contains data from previous operation, e.g., read from
* another PEB previously. The data looks like expected, e.g., if we
* just do not read anything and return - the caller would not
* notice this. E.g., if we are reading a VID header, the buffer may
* contain a valid VID header from another PEB.
* 2. The driver is buggy and returns us success or -EBADMSG or
* -EUCLEAN, but it does not actually put any data to the buffer.
*
* This may confuse UBI or upper layers - they may think the buffer
* contains valid data while in fact it is just old data. This is
* especially possible because UBI (and UBIFS) relies on CRC, and
* treats data as correct even in case of ECC errors if the CRC is
* correct.
*
* Try to prevent this situation by changing the first byte of the
* buffer.
*/
*((uint8_t *)buf) ^= 0xFF;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
addr = (loff_t)pnum * ubi->peb_size + offset;
retry:
err = ubi->mtd->read(ubi->mtd, addr, len, &read, buf);
if (err) {
const char *errstr = (err == -EBADMSG) ? " (ECC error)" : "";
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err == -EUCLEAN) {
/*
* -EUCLEAN is reported if there was a bit-flip which
* was corrected, so this is harmless.
*
* We do not report about it here unless debugging is
* enabled. A corresponding message will be printed
* later, when it is has been scrubbed.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
dbg_msg("fixable bit-flip detected at PEB %d", pnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_assert(len == read);
return UBI_IO_BITFLIPS;
}
if (retries++ < UBI_IO_RETRIES) {
dbg_io("error %d%s while reading %d bytes from PEB %d:%d,"
" read only %zd bytes, retry",
err, errstr, len, pnum, offset, read);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
yield();
goto retry;
}
ubi_err("error %d%s while reading %d bytes from PEB %d:%d, "
"read %zd bytes", err, errstr, len, pnum, offset, read);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_dbg_dump_stack();
/*
* The driver should never return -EBADMSG if it failed to read
* all the requested data. But some buggy drivers might do
* this, so we change it to -EIO.
*/
if (read != len && err == -EBADMSG) {
ubi_assert(0);
err = -EIO;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
} else {
ubi_assert(len == read);
if (ubi_dbg_is_bitflip()) {
dbg_gen("bit-flip (emulated)");
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = UBI_IO_BITFLIPS;
}
}
return err;
}
/**
* ubi_io_write - write data to a physical eraseblock.
* @ubi: UBI device description object
* @buf: buffer with the data to write
* @pnum: physical eraseblock number to write to
* @offset: offset within the physical eraseblock where to write
* @len: how many bytes to write
*
* This function writes @len bytes of data from buffer @buf to offset @offset
* of physical eraseblock @pnum. If all the data were successfully written,
* zero is returned. If an error occurred, this function returns a negative
* error code. If %-EIO is returned, the physical eraseblock most probably went
* bad.
*
* Note, in case of an error, it is possible that something was still written
* to the flash media, but may be some garbage.
*/
int ubi_io_write(struct ubi_device *ubi, const void *buf, int pnum, int offset,
int len)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
int err;
size_t written;
loff_t addr;
dbg_io("write %d bytes to PEB %d:%d", len, pnum, offset);
ubi_assert(pnum >= 0 && pnum < ubi->peb_count);
ubi_assert(offset >= 0 && offset + len <= ubi->peb_size);
ubi_assert(offset % ubi->hdrs_min_io_size == 0);
ubi_assert(len > 0 && len % ubi->hdrs_min_io_size == 0);
if (ubi->ro_mode) {
ubi_err("read-only mode");
return -EROFS;
}
/* The below has to be compiled out if paranoid checks are disabled */
err = paranoid_check_not_bad(ubi, pnum);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/* The area we are writing to has to contain all 0xFF bytes */
err = ubi_dbg_check_all_ff(ubi, pnum, offset, len);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (offset >= ubi->leb_start) {
/*
* We write to the data area of the physical eraseblock. Make
* sure it has valid EC and VID headers.
*/
err = paranoid_check_peb_ec_hdr(ubi, pnum);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = paranoid_check_peb_vid_hdr(ubi, pnum);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
if (ubi_dbg_is_write_failure()) {
dbg_err("cannot write %d bytes to PEB %d:%d "
"(emulated)", len, pnum, offset);
ubi_dbg_dump_stack();
return -EIO;
}
addr = (loff_t)pnum * ubi->peb_size + offset;
err = ubi->mtd->write(ubi->mtd, addr, len, &written, buf);
if (err) {
ubi_err("error %d while writing %d bytes to PEB %d:%d, written "
"%zd bytes", err, len, pnum, offset, written);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_dbg_dump_stack();
ubi_dbg_dump_flash(ubi, pnum, offset, len);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
} else
ubi_assert(written == len);
if (!err) {
err = ubi_dbg_check_write(ubi, buf, pnum, offset, len);
if (err)
return err;
/*
* Since we always write sequentially, the rest of the PEB has
* to contain only 0xFF bytes.
*/
offset += len;
len = ubi->peb_size - offset;
if (len)
err = ubi_dbg_check_all_ff(ubi, pnum, offset, len);
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return err;
}
/**
* erase_callback - MTD erasure call-back.
* @ei: MTD erase information object.
*
* Note, even though MTD erase interface is asynchronous, all the current
* implementations are synchronous anyway.
*/
static void erase_callback(struct erase_info *ei)
{
wake_up_interruptible((wait_queue_head_t *)ei->priv);
}
/**
* do_sync_erase - synchronously erase a physical eraseblock.
* @ubi: UBI device description object
* @pnum: the physical eraseblock number to erase
*
* This function synchronously erases physical eraseblock @pnum and returns
* zero in case of success and a negative error code in case of failure. If
* %-EIO is returned, the physical eraseblock most probably went bad.
*/
static int do_sync_erase(struct ubi_device *ubi, int pnum)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
int err, retries = 0;
struct erase_info ei;
wait_queue_head_t wq;
dbg_io("erase PEB %d", pnum);
retry:
init_waitqueue_head(&wq);
memset(&ei, 0, sizeof(struct erase_info));
ei.mtd = ubi->mtd;
ei.addr = (loff_t)pnum * ubi->peb_size;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ei.len = ubi->peb_size;
ei.callback = erase_callback;
ei.priv = (unsigned long)&wq;
err = ubi->mtd->erase(ubi->mtd, &ei);
if (err) {
if (retries++ < UBI_IO_RETRIES) {
dbg_io("error %d while erasing PEB %d, retry",
err, pnum);
yield();
goto retry;
}
ubi_err("cannot erase PEB %d, error %d", pnum, err);
ubi_dbg_dump_stack();
return err;
}
err = wait_event_interruptible(wq, ei.state == MTD_ERASE_DONE ||
ei.state == MTD_ERASE_FAILED);
if (err) {
ubi_err("interrupted PEB %d erasure", pnum);
return -EINTR;
}
if (ei.state == MTD_ERASE_FAILED) {
if (retries++ < UBI_IO_RETRIES) {
dbg_io("error while erasing PEB %d, retry", pnum);
yield();
goto retry;
}
ubi_err("cannot erase PEB %d", pnum);
ubi_dbg_dump_stack();
return -EIO;
}
err = ubi_dbg_check_all_ff(ubi, pnum, 0, ubi->peb_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (ubi_dbg_is_erase_failure() && !err) {
dbg_err("cannot erase PEB %d (emulated)", pnum);
return -EIO;
}
return 0;
}
/* Patterns to write to a physical eraseblock when torturing it */
static uint8_t patterns[] = {0xa5, 0x5a, 0x0};
/**
* torture_peb - test a supposedly bad physical eraseblock.
* @ubi: UBI device description object
* @pnum: the physical eraseblock number to test
*
* This function returns %-EIO if the physical eraseblock did not pass the
* test, a positive number of erase operations done if the test was
* successfully passed, and other negative error codes in case of other errors.
*/
static int torture_peb(struct ubi_device *ubi, int pnum)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
int err, i, patt_count;
ubi_msg("run torture test for PEB %d", pnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
patt_count = ARRAY_SIZE(patterns);
ubi_assert(patt_count > 0);
mutex_lock(&ubi->buf_mutex);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
for (i = 0; i < patt_count; i++) {
err = do_sync_erase(ubi, pnum);
if (err)
goto out;
/* Make sure the PEB contains only 0xFF bytes */
err = ubi_io_read(ubi, ubi->peb_buf1, pnum, 0, ubi->peb_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err)
goto out;
err = ubi_check_pattern(ubi->peb_buf1, 0xFF, ubi->peb_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err == 0) {
ubi_err("erased PEB %d, but a non-0xFF byte found",
pnum);
err = -EIO;
goto out;
}
/* Write a pattern and check it */
memset(ubi->peb_buf1, patterns[i], ubi->peb_size);
err = ubi_io_write(ubi, ubi->peb_buf1, pnum, 0, ubi->peb_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err)
goto out;
memset(ubi->peb_buf1, ~patterns[i], ubi->peb_size);
err = ubi_io_read(ubi, ubi->peb_buf1, pnum, 0, ubi->peb_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err)
goto out;
err = ubi_check_pattern(ubi->peb_buf1, patterns[i],
ubi->peb_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err == 0) {
ubi_err("pattern %x checking failed for PEB %d",
patterns[i], pnum);
err = -EIO;
goto out;
}
}
err = patt_count;
ubi_msg("PEB %d passed torture test, do not mark it a bad", pnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
out:
mutex_unlock(&ubi->buf_mutex);
if (err == UBI_IO_BITFLIPS || err == -EBADMSG) {
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* If a bit-flip or data integrity error was detected, the test
* has not passed because it happened on a freshly erased
* physical eraseblock which means something is wrong with it.
*/
ubi_err("read problems on freshly erased PEB %d, must be bad",
pnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = -EIO;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return err;
}
/**
* nor_erase_prepare - prepare a NOR flash PEB for erasure.
* @ubi: UBI device description object
* @pnum: physical eraseblock number to prepare
*
* NOR flash, or at least some of them, have peculiar embedded PEB erasure
* algorithm: the PEB is first filled with zeroes, then it is erased. And
* filling with zeroes starts from the end of the PEB. This was observed with
* Spansion S29GL512N NOR flash.
*
* This means that in case of a power cut we may end up with intact data at the
* beginning of the PEB, and all zeroes at the end of PEB. In other words, the
* EC and VID headers are OK, but a large chunk of data at the end of PEB is
* zeroed. This makes UBI mistakenly treat this PEB as used and associate it
* with an LEB, which leads to subsequent failures (e.g., UBIFS fails).
*
* This function is called before erasing NOR PEBs and it zeroes out EC and VID
* magic numbers in order to invalidate them and prevent the failures. Returns
* zero in case of success and a negative error code in case of failure.
*/
static int nor_erase_prepare(struct ubi_device *ubi, int pnum)
{
int err, err1;
size_t written;
loff_t addr;
uint32_t data = 0;
/*
* Note, we cannot generally define VID header buffers on stack,
* because of the way we deal with these buffers (see the header
* comment in this file). But we know this is a NOR-specific piece of
* code, so we can do this. But yes, this is error-prone and we should
* (pre-)allocate VID header buffer instead.
*/
struct ubi_vid_hdr vid_hdr;
/*
* It is important to first invalidate the EC header, and then the VID
* header. Otherwise a power cut may lead to valid EC header and
* invalid VID header, in which case UBI will treat this PEB as
* corrupted and will try to preserve it, and print scary warnings (see
* the header comment in scan.c for more information).
*/
addr = (loff_t)pnum * ubi->peb_size;
err = ubi->mtd->write(ubi->mtd, addr, 4, &written, (void *)&data);
if (!err) {
addr += ubi->vid_hdr_aloffset;
err = ubi->mtd->write(ubi->mtd, addr, 4, &written,
(void *)&data);
if (!err)
return 0;
}
/*
* We failed to write to the media. This was observed with Spansion
* S29GL512N NOR flash. Most probably the previously eraseblock erasure
* was interrupted at a very inappropriate moment, so it became
* unwritable. In this case we probably anyway have garbage in this
* PEB.
*/
err1 = ubi_io_read_vid_hdr(ubi, pnum, &vid_hdr, 0);
if (err1 == UBI_IO_BAD_HDR_EBADMSG || err1 == UBI_IO_BAD_HDR ||
err1 == UBI_IO_FF) {
struct ubi_ec_hdr ec_hdr;
err1 = ubi_io_read_ec_hdr(ubi, pnum, &ec_hdr, 0);
if (err1 == UBI_IO_BAD_HDR_EBADMSG || err1 == UBI_IO_BAD_HDR ||
err1 == UBI_IO_FF)
/*
* Both VID and EC headers are corrupted, so we can
* safely erase this PEB and not afraid that it will be
* treated as a valid PEB in case of an unclean reboot.
*/
return 0;
}
/*
* The PEB contains a valid VID header, but we cannot invalidate it.
* Supposedly the flash media or the driver is screwed up, so return an
* error.
*/
ubi_err("cannot invalidate PEB %d, write returned %d read returned %d",
pnum, err, err1);
ubi_dbg_dump_flash(ubi, pnum, 0, ubi->peb_size);
return -EIO;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/**
* ubi_io_sync_erase - synchronously erase a physical eraseblock.
* @ubi: UBI device description object
* @pnum: physical eraseblock number to erase
* @torture: if this physical eraseblock has to be tortured
*
* This function synchronously erases physical eraseblock @pnum. If @torture
* flag is not zero, the physical eraseblock is checked by means of writing
* different patterns to it and reading them back. If the torturing is enabled,
* the physical eraseblock is erased more than once.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* This function returns the number of erasures made in case of success, %-EIO
* if the erasure failed or the torturing test failed, and other negative error
* codes in case of other errors. Note, %-EIO means that the physical
* eraseblock is bad.
*/
int ubi_io_sync_erase(struct ubi_device *ubi, int pnum, int torture)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
int err, ret = 0;
ubi_assert(pnum >= 0 && pnum < ubi->peb_count);
err = paranoid_check_not_bad(ubi, pnum);
if (err != 0)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (ubi->ro_mode) {
ubi_err("read-only mode");
return -EROFS;
}
if (ubi->nor_flash) {
err = nor_erase_prepare(ubi, pnum);
if (err)
return err;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (torture) {
ret = torture_peb(ubi, pnum);
if (ret < 0)
return ret;
}
err = do_sync_erase(ubi, pnum);
if (err)
return err;
return ret + 1;
}
/**
* ubi_io_is_bad - check if a physical eraseblock is bad.
* @ubi: UBI device description object
* @pnum: the physical eraseblock number to check
*
* This function returns a positive number if the physical eraseblock is bad,
* zero if not, and a negative error code if an error occurred.
*/
int ubi_io_is_bad(const struct ubi_device *ubi, int pnum)
{
struct mtd_info *mtd = ubi->mtd;
ubi_assert(pnum >= 0 && pnum < ubi->peb_count);
if (ubi->bad_allowed) {
int ret;
ret = mtd->block_isbad(mtd, (loff_t)pnum * ubi->peb_size);
if (ret < 0)
ubi_err("error %d while checking if PEB %d is bad",
ret, pnum);
else if (ret)
dbg_io("PEB %d is bad", pnum);
return ret;
}
return 0;
}
/**
* ubi_io_mark_bad - mark a physical eraseblock as bad.
* @ubi: UBI device description object
* @pnum: the physical eraseblock number to mark
*
* This function returns zero in case of success and a negative error code in
* case of failure.
*/
int ubi_io_mark_bad(const struct ubi_device *ubi, int pnum)
{
int err;
struct mtd_info *mtd = ubi->mtd;
ubi_assert(pnum >= 0 && pnum < ubi->peb_count);
if (ubi->ro_mode) {
ubi_err("read-only mode");
return -EROFS;
}
if (!ubi->bad_allowed)
return 0;
err = mtd->block_markbad(mtd, (loff_t)pnum * ubi->peb_size);
if (err)
ubi_err("cannot mark PEB %d bad, error %d", pnum, err);
return err;
}
/**
* validate_ec_hdr - validate an erase counter header.
* @ubi: UBI device description object
* @ec_hdr: the erase counter header to check
*
* This function returns zero if the erase counter header is OK, and %1 if
* not.
*/
static int validate_ec_hdr(const struct ubi_device *ubi,
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
const struct ubi_ec_hdr *ec_hdr)
{
long long ec;
int vid_hdr_offset, leb_start;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ec = be64_to_cpu(ec_hdr->ec);
vid_hdr_offset = be32_to_cpu(ec_hdr->vid_hdr_offset);
leb_start = be32_to_cpu(ec_hdr->data_offset);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (ec_hdr->version != UBI_VERSION) {
ubi_err("node with incompatible UBI version found: "
"this UBI version is %d, image version is %d",
UBI_VERSION, (int)ec_hdr->version);
goto bad;
}
if (vid_hdr_offset != ubi->vid_hdr_offset) {
ubi_err("bad VID header offset %d, expected %d",
vid_hdr_offset, ubi->vid_hdr_offset);
goto bad;
}
if (leb_start != ubi->leb_start) {
ubi_err("bad data offset %d, expected %d",
leb_start, ubi->leb_start);
goto bad;
}
if (ec < 0 || ec > UBI_MAX_ERASECOUNTER) {
ubi_err("bad erase counter %lld", ec);
goto bad;
}
return 0;
bad:
ubi_err("bad EC header");
ubi_dbg_dump_ec_hdr(ec_hdr);
ubi_dbg_dump_stack();
return 1;
}
/**
* ubi_io_read_ec_hdr - read and check an erase counter header.
* @ubi: UBI device description object
* @pnum: physical eraseblock to read from
* @ec_hdr: a &struct ubi_ec_hdr object where to store the read erase counter
* header
* @verbose: be verbose if the header is corrupted or was not found
*
* This function reads erase counter header from physical eraseblock @pnum and
* stores it in @ec_hdr. This function also checks CRC checksum of the read
* erase counter header. The following codes may be returned:
*
* o %0 if the CRC checksum is correct and the header was successfully read;
* o %UBI_IO_BITFLIPS if the CRC is correct, but bit-flips were detected
* and corrected by the flash driver; this is harmless but may indicate that
* this eraseblock may become bad soon (but may be not);
* o %UBI_IO_BAD_HDR if the erase counter header is corrupted (a CRC error);
* o %UBI_IO_BAD_HDR_EBADMSG is the same as %UBI_IO_BAD_HDR, but there also was
* a data integrity error (uncorrectable ECC error in case of NAND);
* o %UBI_IO_FF if only 0xFF bytes were read (the PEB is supposedly empty)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* o a negative error code in case of failure.
*/
int ubi_io_read_ec_hdr(struct ubi_device *ubi, int pnum,
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
struct ubi_ec_hdr *ec_hdr, int verbose)
{
int err, read_err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
uint32_t crc, magic, hdr_crc;
dbg_io("read EC header from PEB %d", pnum);
ubi_assert(pnum >= 0 && pnum < ubi->peb_count);
read_err = ubi_io_read(ubi, ec_hdr, pnum, 0, UBI_EC_HDR_SIZE);
if (read_err) {
if (read_err != UBI_IO_BITFLIPS && read_err != -EBADMSG)
return read_err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* We read all the data, but either a correctable bit-flip
* occurred, or MTD reported a data integrity error
* (uncorrectable ECC error in case of NAND). The former is
* harmless, the later may mean that the read data is
* corrupted. But we have a CRC check-sum and we will detect
* this. If the EC header is still OK, we just report this as
* there was a bit-flip, to force scrubbing.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
}
magic = be32_to_cpu(ec_hdr->magic);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (magic != UBI_EC_HDR_MAGIC) {
if (read_err == -EBADMSG)
return UBI_IO_BAD_HDR_EBADMSG;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* The magic field is wrong. Let's check if we have read all
* 0xFF. If yes, this physical eraseblock is assumed to be
* empty.
*/
if (ubi_check_pattern(ec_hdr, 0xFF, UBI_EC_HDR_SIZE)) {
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/* The physical eraseblock is supposedly empty */
if (verbose)
ubi_warn("no EC header found at PEB %d, "
"only 0xFF bytes", pnum);
dbg_bld("no EC header found at PEB %d, "
"only 0xFF bytes", pnum);
if (!read_err)
return UBI_IO_FF;
else
return UBI_IO_FF_BITFLIPS;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
/*
* This is not a valid erase counter header, and these are not
* 0xFF bytes. Report that the header is corrupted.
*/
if (verbose) {
ubi_warn("bad magic number at PEB %d: %08x instead of "
"%08x", pnum, magic, UBI_EC_HDR_MAGIC);
ubi_dbg_dump_ec_hdr(ec_hdr);
}
dbg_bld("bad magic number at PEB %d: %08x instead of "
"%08x", pnum, magic, UBI_EC_HDR_MAGIC);
return UBI_IO_BAD_HDR;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
crc = crc32(UBI_CRC32_INIT, ec_hdr, UBI_EC_HDR_SIZE_CRC);
hdr_crc = be32_to_cpu(ec_hdr->hdr_crc);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (hdr_crc != crc) {
if (verbose) {
ubi_warn("bad EC header CRC at PEB %d, calculated "
"%#08x, read %#08x", pnum, crc, hdr_crc);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_dbg_dump_ec_hdr(ec_hdr);
}
dbg_bld("bad EC header CRC at PEB %d, calculated "
"%#08x, read %#08x", pnum, crc, hdr_crc);
if (!read_err)
return UBI_IO_BAD_HDR;
else
return UBI_IO_BAD_HDR_EBADMSG;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
/* And of course validate what has just been read from the media */
err = validate_ec_hdr(ubi, ec_hdr);
if (err) {
ubi_err("validation failed for PEB %d", pnum);
return -EINVAL;
}
/*
* If there was %-EBADMSG, but the header CRC is still OK, report about
* a bit-flip to force scrubbing on this PEB.
*/
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return read_err ? UBI_IO_BITFLIPS : 0;
}
/**
* ubi_io_write_ec_hdr - write an erase counter header.
* @ubi: UBI device description object
* @pnum: physical eraseblock to write to
* @ec_hdr: the erase counter header to write
*
* This function writes erase counter header described by @ec_hdr to physical
* eraseblock @pnum. It also fills most fields of @ec_hdr before writing, so
* the caller do not have to fill them. Callers must only fill the @ec_hdr->ec
* field.
*
* This function returns zero in case of success and a negative error code in
* case of failure. If %-EIO is returned, the physical eraseblock most probably
* went bad.
*/
int ubi_io_write_ec_hdr(struct ubi_device *ubi, int pnum,
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
struct ubi_ec_hdr *ec_hdr)
{
int err;
uint32_t crc;
dbg_io("write EC header to PEB %d", pnum);
ubi_assert(pnum >= 0 && pnum < ubi->peb_count);
ec_hdr->magic = cpu_to_be32(UBI_EC_HDR_MAGIC);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ec_hdr->version = UBI_VERSION;
ec_hdr->vid_hdr_offset = cpu_to_be32(ubi->vid_hdr_offset);
ec_hdr->data_offset = cpu_to_be32(ubi->leb_start);
ec_hdr->image_seq = cpu_to_be32(ubi->image_seq);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
crc = crc32(UBI_CRC32_INIT, ec_hdr, UBI_EC_HDR_SIZE_CRC);
ec_hdr->hdr_crc = cpu_to_be32(crc);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = paranoid_check_ec_hdr(ubi, pnum, ec_hdr);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = ubi_io_write(ubi, ec_hdr, pnum, 0, ubi->ec_hdr_alsize);
return err;
}
/**
* validate_vid_hdr - validate a volume identifier header.
* @ubi: UBI device description object
* @vid_hdr: the volume identifier header to check
*
* This function checks that data stored in the volume identifier header
* @vid_hdr. Returns zero if the VID header is OK and %1 if not.
*/
static int validate_vid_hdr(const struct ubi_device *ubi,
const struct ubi_vid_hdr *vid_hdr)
{
int vol_type = vid_hdr->vol_type;
int copy_flag = vid_hdr->copy_flag;
int vol_id = be32_to_cpu(vid_hdr->vol_id);
int lnum = be32_to_cpu(vid_hdr->lnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
int compat = vid_hdr->compat;
int data_size = be32_to_cpu(vid_hdr->data_size);
int used_ebs = be32_to_cpu(vid_hdr->used_ebs);
int data_pad = be32_to_cpu(vid_hdr->data_pad);
int data_crc = be32_to_cpu(vid_hdr->data_crc);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
int usable_leb_size = ubi->leb_size - data_pad;
if (copy_flag != 0 && copy_flag != 1) {
dbg_err("bad copy_flag");
goto bad;
}
if (vol_id < 0 || lnum < 0 || data_size < 0 || used_ebs < 0 ||
data_pad < 0) {
dbg_err("negative values");
goto bad;
}
if (vol_id >= UBI_MAX_VOLUMES && vol_id < UBI_INTERNAL_VOL_START) {
dbg_err("bad vol_id");
goto bad;
}
if (vol_id < UBI_INTERNAL_VOL_START && compat != 0) {
dbg_err("bad compat");
goto bad;
}
if (vol_id >= UBI_INTERNAL_VOL_START && compat != UBI_COMPAT_DELETE &&
compat != UBI_COMPAT_RO && compat != UBI_COMPAT_PRESERVE &&
compat != UBI_COMPAT_REJECT) {
dbg_err("bad compat");
goto bad;
}
if (vol_type != UBI_VID_DYNAMIC && vol_type != UBI_VID_STATIC) {
dbg_err("bad vol_type");
goto bad;
}
if (data_pad >= ubi->leb_size / 2) {
dbg_err("bad data_pad");
goto bad;
}
if (vol_type == UBI_VID_STATIC) {
/*
* Although from high-level point of view static volumes may
* contain zero bytes of data, but no VID headers can contain
* zero at these fields, because they empty volumes do not have
* mapped logical eraseblocks.
*/
if (used_ebs == 0) {
dbg_err("zero used_ebs");
goto bad;
}
if (data_size == 0) {
dbg_err("zero data_size");
goto bad;
}
if (lnum < used_ebs - 1) {
if (data_size != usable_leb_size) {
dbg_err("bad data_size");
goto bad;
}
} else if (lnum == used_ebs - 1) {
if (data_size == 0) {
dbg_err("bad data_size at last LEB");
goto bad;
}
} else {
dbg_err("too high lnum");
goto bad;
}
} else {
if (copy_flag == 0) {
if (data_crc != 0) {
dbg_err("non-zero data CRC");
goto bad;
}
if (data_size != 0) {
dbg_err("non-zero data_size");
goto bad;
}
} else {
if (data_size == 0) {
dbg_err("zero data_size of copy");
goto bad;
}
}
if (used_ebs != 0) {
dbg_err("bad used_ebs");
goto bad;
}
}
return 0;
bad:
ubi_err("bad VID header");
ubi_dbg_dump_vid_hdr(vid_hdr);
ubi_dbg_dump_stack();
return 1;
}
/**
* ubi_io_read_vid_hdr - read and check a volume identifier header.
* @ubi: UBI device description object
* @pnum: physical eraseblock number to read from
* @vid_hdr: &struct ubi_vid_hdr object where to store the read volume
* identifier header
* @verbose: be verbose if the header is corrupted or wasn't found
*
* This function reads the volume identifier header from physical eraseblock
* @pnum and stores it in @vid_hdr. It also checks CRC checksum of the read
* volume identifier header. The error codes are the same as in
* 'ubi_io_read_ec_hdr()'.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* Note, the implementation of this function is also very similar to
* 'ubi_io_read_ec_hdr()', so refer commentaries in 'ubi_io_read_ec_hdr()'.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
int ubi_io_read_vid_hdr(struct ubi_device *ubi, int pnum,
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
struct ubi_vid_hdr *vid_hdr, int verbose)
{
int err, read_err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
uint32_t crc, magic, hdr_crc;
void *p;
dbg_io("read VID header from PEB %d", pnum);
ubi_assert(pnum >= 0 && pnum < ubi->peb_count);
p = (char *)vid_hdr - ubi->vid_hdr_shift;
read_err = ubi_io_read(ubi, p, pnum, ubi->vid_hdr_aloffset,
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->vid_hdr_alsize);
if (read_err && read_err != UBI_IO_BITFLIPS && read_err != -EBADMSG)
return read_err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
magic = be32_to_cpu(vid_hdr->magic);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (magic != UBI_VID_HDR_MAGIC) {
if (read_err == -EBADMSG)
return UBI_IO_BAD_HDR_EBADMSG;
if (ubi_check_pattern(vid_hdr, 0xFF, UBI_VID_HDR_SIZE)) {
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (verbose)
ubi_warn("no VID header found at PEB %d, "
"only 0xFF bytes", pnum);
dbg_bld("no VID header found at PEB %d, "
"only 0xFF bytes", pnum);
if (!read_err)
return UBI_IO_FF;
else
return UBI_IO_FF_BITFLIPS;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
if (verbose) {
ubi_warn("bad magic number at PEB %d: %08x instead of "
"%08x", pnum, magic, UBI_VID_HDR_MAGIC);
ubi_dbg_dump_vid_hdr(vid_hdr);
}
dbg_bld("bad magic number at PEB %d: %08x instead of "
"%08x", pnum, magic, UBI_VID_HDR_MAGIC);
return UBI_IO_BAD_HDR;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
crc = crc32(UBI_CRC32_INIT, vid_hdr, UBI_VID_HDR_SIZE_CRC);
hdr_crc = be32_to_cpu(vid_hdr->hdr_crc);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (hdr_crc != crc) {
if (verbose) {
ubi_warn("bad CRC at PEB %d, calculated %#08x, "
"read %#08x", pnum, crc, hdr_crc);
ubi_dbg_dump_vid_hdr(vid_hdr);
}
dbg_bld("bad CRC at PEB %d, calculated %#08x, "
"read %#08x", pnum, crc, hdr_crc);
if (!read_err)
return UBI_IO_BAD_HDR;
else
return UBI_IO_BAD_HDR_EBADMSG;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
err = validate_vid_hdr(ubi, vid_hdr);
if (err) {
ubi_err("validation failed for PEB %d", pnum);
return -EINVAL;
}
return read_err ? UBI_IO_BITFLIPS : 0;
}
/**
* ubi_io_write_vid_hdr - write a volume identifier header.
* @ubi: UBI device description object
* @pnum: the physical eraseblock number to write to
* @vid_hdr: the volume identifier header to write
*
* This function writes the volume identifier header described by @vid_hdr to
* physical eraseblock @pnum. This function automatically fills the
* @vid_hdr->magic and the @vid_hdr->version fields, as well as calculates
* header CRC checksum and stores it at vid_hdr->hdr_crc.
*
* This function returns zero in case of success and a negative error code in
* case of failure. If %-EIO is returned, the physical eraseblock probably went
* bad.
*/
int ubi_io_write_vid_hdr(struct ubi_device *ubi, int pnum,
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
struct ubi_vid_hdr *vid_hdr)
{
int err;
uint32_t crc;
void *p;
dbg_io("write VID header to PEB %d", pnum);
ubi_assert(pnum >= 0 && pnum < ubi->peb_count);
err = paranoid_check_peb_ec_hdr(ubi, pnum);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
vid_hdr->magic = cpu_to_be32(UBI_VID_HDR_MAGIC);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
vid_hdr->version = UBI_VERSION;
crc = crc32(UBI_CRC32_INIT, vid_hdr, UBI_VID_HDR_SIZE_CRC);
vid_hdr->hdr_crc = cpu_to_be32(crc);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = paranoid_check_vid_hdr(ubi, pnum, vid_hdr);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
p = (char *)vid_hdr - ubi->vid_hdr_shift;
err = ubi_io_write(ubi, p, pnum, ubi->vid_hdr_aloffset,
ubi->vid_hdr_alsize);
return err;
}
#ifdef CONFIG_MTD_UBI_DEBUG
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/**
* paranoid_check_not_bad - ensure that a physical eraseblock is not bad.
* @ubi: UBI device description object
* @pnum: physical eraseblock number to check
*
* This function returns zero if the physical eraseblock is good, %-EINVAL if
* it is bad and a negative error code if an error occurred.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
static int paranoid_check_not_bad(const struct ubi_device *ubi, int pnum)
{
int err;
if (!(ubi_chk_flags & UBI_CHK_IO))
return 0;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = ubi_io_is_bad(ubi, pnum);
if (!err)
return err;
ubi_err("paranoid check failed for PEB %d", pnum);
ubi_dbg_dump_stack();
return err > 0 ? -EINVAL : err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
/**
* paranoid_check_ec_hdr - check if an erase counter header is all right.
* @ubi: UBI device description object
* @pnum: physical eraseblock number the erase counter header belongs to
* @ec_hdr: the erase counter header to check
*
* This function returns zero if the erase counter header contains valid
* values, and %-EINVAL if not.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
static int paranoid_check_ec_hdr(const struct ubi_device *ubi, int pnum,
const struct ubi_ec_hdr *ec_hdr)
{
int err;
uint32_t magic;
if (!(ubi_chk_flags & UBI_CHK_IO))
return 0;
magic = be32_to_cpu(ec_hdr->magic);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (magic != UBI_EC_HDR_MAGIC) {
ubi_err("bad magic %#08x, must be %#08x",
magic, UBI_EC_HDR_MAGIC);
goto fail;
}
err = validate_ec_hdr(ubi, ec_hdr);
if (err) {
ubi_err("paranoid check failed for PEB %d", pnum);
goto fail;
}
return 0;
fail:
ubi_dbg_dump_ec_hdr(ec_hdr);
ubi_dbg_dump_stack();
return -EINVAL;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
/**
* paranoid_check_peb_ec_hdr - check erase counter header.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* @ubi: UBI device description object
* @pnum: the physical eraseblock number to check
*
* This function returns zero if the erase counter header is all right and and
* a negative error code if not or if an error occurred.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
static int paranoid_check_peb_ec_hdr(const struct ubi_device *ubi, int pnum)
{
int err;
uint32_t crc, hdr_crc;
struct ubi_ec_hdr *ec_hdr;
if (!(ubi_chk_flags & UBI_CHK_IO))
return 0;
ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (!ec_hdr)
return -ENOMEM;
err = ubi_io_read(ubi, ec_hdr, pnum, 0, UBI_EC_HDR_SIZE);
if (err && err != UBI_IO_BITFLIPS && err != -EBADMSG)
goto exit;
crc = crc32(UBI_CRC32_INIT, ec_hdr, UBI_EC_HDR_SIZE_CRC);
hdr_crc = be32_to_cpu(ec_hdr->hdr_crc);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (hdr_crc != crc) {
ubi_err("bad CRC, calculated %#08x, read %#08x", crc, hdr_crc);
ubi_err("paranoid check failed for PEB %d", pnum);
ubi_dbg_dump_ec_hdr(ec_hdr);
ubi_dbg_dump_stack();
err = -EINVAL;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
goto exit;
}
err = paranoid_check_ec_hdr(ubi, pnum, ec_hdr);
exit:
kfree(ec_hdr);
return err;
}
/**
* paranoid_check_vid_hdr - check that a volume identifier header is all right.
* @ubi: UBI device description object
* @pnum: physical eraseblock number the volume identifier header belongs to
* @vid_hdr: the volume identifier header to check
*
* This function returns zero if the volume identifier header is all right, and
* %-EINVAL if not.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
static int paranoid_check_vid_hdr(const struct ubi_device *ubi, int pnum,
const struct ubi_vid_hdr *vid_hdr)
{
int err;
uint32_t magic;
if (!(ubi_chk_flags & UBI_CHK_IO))
return 0;
magic = be32_to_cpu(vid_hdr->magic);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (magic != UBI_VID_HDR_MAGIC) {
ubi_err("bad VID header magic %#08x at PEB %d, must be %#08x",
magic, pnum, UBI_VID_HDR_MAGIC);
goto fail;
}
err = validate_vid_hdr(ubi, vid_hdr);
if (err) {
ubi_err("paranoid check failed for PEB %d", pnum);
goto fail;
}
return err;
fail:
ubi_err("paranoid check failed for PEB %d", pnum);
ubi_dbg_dump_vid_hdr(vid_hdr);
ubi_dbg_dump_stack();
return -EINVAL;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
/**
* paranoid_check_peb_vid_hdr - check volume identifier header.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* @ubi: UBI device description object
* @pnum: the physical eraseblock number to check
*
* This function returns zero if the volume identifier header is all right,
* and a negative error code if not or if an error occurred.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
static int paranoid_check_peb_vid_hdr(const struct ubi_device *ubi, int pnum)
{
int err;
uint32_t crc, hdr_crc;
struct ubi_vid_hdr *vid_hdr;
void *p;
if (!(ubi_chk_flags & UBI_CHK_IO))
return 0;
vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (!vid_hdr)
return -ENOMEM;
p = (char *)vid_hdr - ubi->vid_hdr_shift;
err = ubi_io_read(ubi, p, pnum, ubi->vid_hdr_aloffset,
ubi->vid_hdr_alsize);
if (err && err != UBI_IO_BITFLIPS && err != -EBADMSG)
goto exit;
crc = crc32(UBI_CRC32_INIT, vid_hdr, UBI_EC_HDR_SIZE_CRC);
hdr_crc = be32_to_cpu(vid_hdr->hdr_crc);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (hdr_crc != crc) {
ubi_err("bad VID header CRC at PEB %d, calculated %#08x, "
"read %#08x", pnum, crc, hdr_crc);
ubi_err("paranoid check failed for PEB %d", pnum);
ubi_dbg_dump_vid_hdr(vid_hdr);
ubi_dbg_dump_stack();
err = -EINVAL;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
goto exit;
}
err = paranoid_check_vid_hdr(ubi, pnum, vid_hdr);
exit:
ubi_free_vid_hdr(ubi, vid_hdr);
return err;
}
/**
* ubi_dbg_check_write - make sure write succeeded.
* @ubi: UBI device description object
* @buf: buffer with data which were written
* @pnum: physical eraseblock number the data were written to
* @offset: offset within the physical eraseblock the data were written to
* @len: how many bytes were written
*
* This functions reads data which were recently written and compares it with
* the original data buffer - the data have to match. Returns zero if the data
* match and a negative error code if not or in case of failure.
*/
int ubi_dbg_check_write(struct ubi_device *ubi, const void *buf, int pnum,
int offset, int len)
{
int err, i;
size_t read;
void *buf1;
loff_t addr = (loff_t)pnum * ubi->peb_size + offset;
if (!(ubi_chk_flags & UBI_CHK_IO))
return 0;
buf1 = __vmalloc(len, GFP_NOFS, PAGE_KERNEL);
if (!buf1) {
ubi_err("cannot allocate memory to check writes");
return 0;
}
err = ubi->mtd->read(ubi->mtd, addr, len, &read, buf1);
if (err && err != -EUCLEAN)
goto out_free;
for (i = 0; i < len; i++) {
uint8_t c = ((uint8_t *)buf)[i];
uint8_t c1 = ((uint8_t *)buf1)[i];
int dump_len;
if (c == c1)
continue;
ubi_err("paranoid check failed for PEB %d:%d, len %d",
pnum, offset, len);
ubi_msg("data differ at position %d", i);
dump_len = max_t(int, 128, len - i);
ubi_msg("hex dump of the original buffer from %d to %d",
i, i + dump_len);
print_hex_dump(KERN_DEBUG, "", DUMP_PREFIX_OFFSET, 32, 1,
buf + i, dump_len, 1);
ubi_msg("hex dump of the read buffer from %d to %d",
i, i + dump_len);
print_hex_dump(KERN_DEBUG, "", DUMP_PREFIX_OFFSET, 32, 1,
buf1 + i, dump_len, 1);
ubi_dbg_dump_stack();
err = -EINVAL;
goto out_free;
}
vfree(buf1);
return 0;
out_free:
vfree(buf1);
return err;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/**
* ubi_dbg_check_all_ff - check that a region of flash is empty.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* @ubi: UBI device description object
* @pnum: the physical eraseblock number to check
* @offset: the starting offset within the physical eraseblock to check
* @len: the length of the region to check
*
* This function returns zero if only 0xFF bytes are present at offset
* @offset of the physical eraseblock @pnum, and a negative error code if not
* or if an error occurred.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
int ubi_dbg_check_all_ff(struct ubi_device *ubi, int pnum, int offset, int len)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
size_t read;
int err;
void *buf;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
loff_t addr = (loff_t)pnum * ubi->peb_size + offset;
if (!(ubi_chk_flags & UBI_CHK_IO))
return 0;
buf = __vmalloc(len, GFP_NOFS, PAGE_KERNEL);
if (!buf) {
ubi_err("cannot allocate memory to check for 0xFFs");
return 0;
}
err = ubi->mtd->read(ubi->mtd, addr, len, &read, buf);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err && err != -EUCLEAN) {
ubi_err("error %d while reading %d bytes from PEB %d:%d, "
"read %zd bytes", err, len, pnum, offset, read);
goto error;
}
err = ubi_check_pattern(buf, 0xFF, len);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err == 0) {
ubi_err("flash region at PEB %d:%d, length %d does not "
"contain all 0xFF bytes", pnum, offset, len);
goto fail;
}
vfree(buf);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return 0;
fail:
ubi_err("paranoid check failed for PEB %d", pnum);
ubi_msg("hex dump of the %d-%d region", offset, offset + len);
print_hex_dump(KERN_DEBUG, "", DUMP_PREFIX_OFFSET, 32, 1, buf, len, 1);
err = -EINVAL;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
error:
ubi_dbg_dump_stack();
vfree(buf);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return err;
}
#endif /* CONFIG_MTD_UBI_DEBUG */