wait_for_completion_timeout() never returns a <0 value. It returns either
on timeout or a positive value (at least 1, or number of jiffies left
till timeout)
So, fix the error handling path and return -ETIMEDOUT should a timeout
occur.
Fixes: b0823ee35c ("spi: Add spi driver for Socionext SynQuacer platform")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Acked-by: Jassi Brar <jaswinder.singh@linaro.org>
Link: https://lore.kernel.org/r/c2040bf3cfa201fd8890cfab14fa5a701ffeca14.1676466072.git.christophe.jaillet@wanadoo.fr
Signed-off-by: Mark Brown <broonie@kernel.org>
Merge series from Dhruva Gole <d-gole@ti.com>:
* Reset the CMD_CTRL Register, without which read/writes in STIG mode
were failing in some cases. The issue came to light while using STIG
Mode for small reads.
* Also add a flag that can allow us to do direct reads but distinguish
direct writes, thus enabling us to disable writes in DAC mode in some
cases that require it. (Like to write to some connected Flash registers)
* Fix register reads in STIG mode and also use STIG mode while reading flash
registers.
Currently if you try to read a register while in STIG mode there is no
support for ADDR and thus naturally a register never gets read from the
flash.
This patch series has been tested on a TI AM625-SK-EVM with both a quad
spi nor flash (s25hs) and OSPI NOR Flash (s28hs).
Output of ltp-ddt test, "DD_RW_ERASESIZE_UBIFS" run with s25hs512t flash:
...
[ 2.334068] spi-nor spi0.0: s25hs512t (65536 Kbytes)
[ 2.339185] 7 fixed-partitions partitions found on MTD device
fc40000.spi.0
[ 2.346158] Creating 7 MTD partitions on "fc40000.spi.0":
[ 2.351555] 0x000000000000-0x000000080000 : "ospi.tiboot3"
[ 2.358344] 0x000000080000-0x000000280000 : "ospi.tispl"
[ 2.364788] 0x000000280000-0x000000680000 : "ospi.u-boot"
[ 2.371311] 0x000000680000-0x0000006c0000 : "ospi.env"
[ 2.377519] 0x0000006c0000-0x000000700000 : "ospi.env.backup"
[ 2.384419] 0x000000800000-0x000003fc0000 : "ospi.rootfs"
[ 2.390890] 0x000003fc0000-0x000004000000 : "ospi.phypattern"
..snip..
Test Start Time: Wed Jan 11 21:14:31 2023
-----------------------------------------
Testcase Result Exit Value
-------- ------ ----------
OSPI_S_FUNC_DD_RW_ERASESIZE_UBIFS PASS 0
-----------------------------------------------
Total Tests: 1
Total Skipped Tests: 0
Total Failures: 0
Kernel Version: 6.2.0-rc1-00040-g700d796a94e0-dirty
Machine Architecture: aarch64
Hostname: am62xx-evm
Group some variables based on their sizes to reduce hole and avoid padding.
On x86_64, this shrinks the size from 144 to 128 bytes.
Turn 'timestamped' into a bitfield so that it can be easily merged with
some other bifields and move 'error'.
This should have no real impact on memory allocation because 'struct
spi_transfer' is mostly used on stack, but it can save a few cycles
when the structure is initialized or copied.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Link: https://lore.kernel.org/r/93a051da85a895bc6003aedfb00a13e1c2fc6338.1676370870.git.christophe.jaillet@wanadoo.fr
Signed-off-by: Mark Brown <broonie@kernel.org>
Fix the issue where some flash chips like cypress S25HS256T return the
value of the same register over and over in DAC mode.
For example in the TI K3-AM62x Processors refer [0] Technical Reference
Manual there is a layer of digital logic in front of the QSPI/OSPI
Drive when used in DAC mode. This is part of the Flash Subsystem (FSS)
which provides access to external Flash devices.
The FSS0_0_SYSCONFIG Register (Offset = 4h) has a BIT Field for
OSPI_32B_DISABLE_MODE which has a Reset value = 0. This means, OSPI 32bit
mode enabled by default.
Thus, by default controller operates in 32 bit mode causing it to always
align all data to 4 bytes from a 4byte aligned address. In some flash
chips like cypress for example if we try to read some regs in DAC mode
then it keeps sending the value of the first register that was requested
and inorder to read the next reg, we have to stop and re-initiate a new
transaction.
This causes wrong register values to be read than what is desired when
registers are read in DAC mode. Hence if the data.nbytes is very less
then prefer STIG mode for such small reads.
[0] https://www.ti.com/lit/ug/spruiv7a/spruiv7a.pdf
Signed-off-by: Dhruva Gole <d-gole@ti.com>
Link: https://lore.kernel.org/r/20230125081023.1573712-5-d-gole@ti.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Setup the Addr bit field while issuing register reads in STIG mode. This
is needed for example flashes like cypress define in their transaction
table that to read any register there is 1 cmd byte and a few more address
bytes trailing the cmd byte. Absence of addr bytes will obviously fail
to read correct data from flash register that maybe requested by flash
driver because the controller doesn't even specify which address of the
flash register the read is being requested from.
Signed-off-by: Dhruva Gole <d-gole@ti.com>
Link: https://lore.kernel.org/r/20230125081023.1573712-4-d-gole@ti.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Create new flag inorder to avoid playing with use_direct_mode
flag currently being used throughout the driver.
Disable DAC write if auto polling is disabled or CQSPI_DISABLE_DAC_MODE
is set.
Signed-off-by: Dhruva Gole <d-gole@ti.com>
Link: https://lore.kernel.org/r/20230125081023.1573712-3-d-gole@ti.com
Signed-off-by: Mark Brown <broonie@kernel.org>
If one leaves the CQSPI_REG_CMDCTRL in an unclean state this may cause
issues in future command reads. This issue came to light when some flash
reads in STIG mode were coming back dirty.
Signed-off-by: Dhruva Gole <d-gole@ti.com>
Link: https://lore.kernel.org/r/20230125081023.1573712-2-d-gole@ti.com
Signed-off-by: Mark Brown <broonie@kernel.org>
sparse tool report warnings: drivers/spi/spi-bcm63xx-hsspi.c:197:31:
sparse: sparse: cast from restricted __be16.
The controller requires big endian 16 bit data. Define an intermediate
u16 value and use __be16 piointer dereferncing for the data to avoid
directly casting to u16 and sparse warning.
Fixes: 85a84a6169 ("spi: bcm63xx-hsspi: Endianness fix for ARM based SoC")
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202302110438.sQwQnU54-lkp@intel.com/
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230211032759.72967-1-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
The newer BCMBCA SoCs such as BCM6756, BCM4912 and BCM6855 include an
updated SPI controller that add the capability to allow the driver to
control chip select explicitly. Driver can control and keep cs low
between the transfers natively. Hence the dummy cs workaround or prepend
mode found in the bcm63xx-hsspi driver are no longer needed and this new
driver is much cleaner.
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230209200246.141520-15-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
In general the controller supports SPI dual mode operation but the
particular SPI flash dual io read op switches from single mode in cmd
phase to dual mode in address and data phase. This is not compatible
with prepend operation where cmd and address are sent out through the
prepend buffer and they must use same the number of io pins.
This patch disables these SPI flash dual io read ops through the mem_ops
supports_op interface. This makes sure the SPI flash driver selects the
compatible read ops at run time.
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230209200246.141520-14-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Currently exec_op is always required if controller driver provides
mem_ops. But some controller such as bcm63xx-hsspi may only need to
implement other operation like supports_op and use the default
execution operation. This patch removes this restriction.
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230209200246.141520-13-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Due to the controller limitation to keep the chip select low during the
bus idle time between the transfer, a dummy cs workaround was used when
this driver was first upstreamed to the kernel. It basically picks the
dummy cs as !actual_cs so typically dummy cs is 1 when most of the case
only cs 0 is used in the board design. Then invert the polarity of both
cs and tell the controller to start the transfers using dummy cs.
Assuming both cs are active low before the inversion, effectively this
keeps dummy cs high and actual cs low during the transfer and workaround
the issue.
This workaround implies that dummy cs 1 pin has to be set to chip
selection function in the pinmux when the transfer clock is above
25MHz. The old chips likely have default pinmux set to chip select on
the dummy cs pin so it works but this is not case for the new Broadband
BCA chips and this workaround stop working. This is specifically an
issue to support SPI NAND and SPI NOR flash because these flash devices
can typically run at or above 100MHz.
This patch utilizes the prepend feature of the controller to combine the
multiple transfers in the same message to a single transfer when
possible. This way there is no need to keep clock low between transfers
and solve the issue without any hardware requirement.
Multiple transfers within a SPI message may be combined into one
transfer if the following are all true:
* One or more half duplex write transfer in single bit mode
* Optional full duplex read/write at the end
* No delay and cs_change between transfers
Most of the SPI device meets this requirements such as SPI NOR,
SPI NAND flash, Broadcom SPI voice card and etc. For any SPI message
that does not meet the above requirement to combine the transfers, we
switch to original dummy cs mode but limit the clock rate to the safe
25MHz. This is the default auto transfer mode and it makes sure all the
SPI message can be supported automatically under the hood.
This patch also adds the driver sysfs node xfer_mode to provide
the option for overriding the default auto mode and force it to dummy cs
or prepend mode.
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230209200246.141520-12-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Currently the driver always sets the controller to dual data bit mode
for both tx and rx data in the profile mode control register even for
single data bit transfer. Luckily the opcode is set correctly according
to SPI transfer data bit width so it does not actually cause issues.
This change fixes the problem by setting tx and rx data bit mode field
correctly according to the actual SPI transfer tx and rx data bit width.
Fixes: 142168eba9 ("spi: bcm63xx-hsspi: add bcm63xx HSSPI driver")
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230209200246.141520-11-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
The kernel SPI interface includes the cs_change flag that alters how
the CS behaves.
If we're in the middle of transfers, it tells us to unselect the
CS momentarily since the target device requires that.
If we're at the end of a transfer, it tells us to keep the CS
selected, perhaps because the next transfer is likely targeted
to the same device.
We implement this scheme in the HSSPI driver in this change.
Prior to this change, the CS would toggle momentarily if cs_change
was set for the last transfer. This can be ignored by some or
most devices, but the Microchip TPM2 device does not ignore it.
With the change, the behavior is corrected and the 'glitch' is
eliminated.
Signed-off-by: Kursad Oney <kursad.oney@broadcom.com>
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230209200246.141520-10-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
For SPI controller that implements transfer_one_message, it needs to
insert the delay that required by cs change event between the transfers.
Add a wrapper for the local function _spi_transfer_cs_change_delay_exec
and export it for SPI controller driver to use.
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230209200246.141520-9-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Modern intel hardware uses controllers that work in hardware
sequencing mode. In this mode, the controller exposes a subset
of operations, like read, write and erase, making it easier
and less error-prone for use.
On the other hand, most of the controllers handled by the
platform driver use software sequencing that exposes the
entire set of operations i.e. exposes the low level SPI-NOR
opcodes to the software for use.
Update PCI and Platform help text with this information.
Signed-off-by: Mauro Lima <mauro.lima@eclypsium.com>
Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Link: https://lore.kernel.org/r/20230210164158.211065-1-mauro.lima@eclypsium.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Merge series from Xiangsheng Hou <xiangsheng.hou@mediatek.com>:
This patch series split from bellow series which pick-up spi relevant patches
https://lore.kernel.org/all/20230130030656.12127-1-xiangsheng.hou@mediatek.com.
This series add MediaTek MT7986 SPI NAND controller support, add read latch
latency, smaple delay adjust and add optional nfi_hclk.
Add snfi sample delay and read latency adjustment which can get
from dts property.
Signed-off-by: Xiangsheng Hou <xiangsheng.hou@mediatek.com>
Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Link: https://lore.kernel.org/r/20230201020921.26712-5-xiangsheng.hou@mediatek.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Add optional nfi_hclk which is needed for MT7986.
Signed-off-by: Xiangsheng Hou <xiangsheng.hou@mediatek.com>
Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Link: https://lore.kernel.org/r/20230201020921.26712-3-xiangsheng.hou@mediatek.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Polling mode provides better throughput in general by avoiding the
interrupt overhead as the maximum data size one interrupt can handle is
only 512 bytes. So switch to polling mode as the default mode but add
a driver sysfs option wait_mode to allow user manually changing the mode
at run time between interrupt and polling. Also add driver banner
message when the driver is loaded successfully.
When test on a Broadcom BCM47622(ARM A7 dual core) reference board with
WINBOND W25N01GV SPI NAND chip at 100MHz SPI clock using the MTD speed
test suite, it shows about 15% improvement on the write and 30% on
the read:
** Interrupt mode **
mtd_speedtest: MTD device: 0 count: 16
mtd_speedtest: MTD device size 134217728, eraseblock size 131072, page
size 2048, count of eraseblocks 1024, pages per eraseblock 64, OOB size
64
mtd_test: scanning for bad eraseblocks
mtd_test: scanned 16 eraseblocks, 0 are bad
mtd_speedtest: testing eraseblock write speed
mtd_speedtest: eraseblock write speed is 3072 KiB/s
mtd_speedtest: testing eraseblock read speed
mtd_speedtest: eraseblock read speed is 6690 KiB/s
mtd_speedtest: testing page write speed
mtd_speedtest: page write speed is 3066 KiB/s
mtd_speedtest: testing page read speed
mtd_speedtest: page read speed is 6762 KiB/s
mtd_speedtest: testing 2 page write speed
mtd_speedtest: 2 page write speed is 3071 KiB/s
mtd_speedtest: testing 2 page read speed
mtd_speedtest: 2 page read speed is 6772 KiB/s
** Polling mode **
mtd_speedtest: MTD device: 0 count: 16
mtd_speedtest: MTD device size 134217728, eraseblock size 131072, page
size 2048, count of eraseblocks 1024, pages per eraseblock 64, OOB size
64
mtd_test: scanning for bad eraseblocks
mtd_test: scanned 16 eraseblocks, 0 are bad
mtd_speedtest: testing eraseblock write speed
mtd_speedtest: eraseblock write speed is 3542 KiB/s
mtd_speedtest: testing eraseblock read speed
mtd_speedtest: eraseblock read speed is 8825 KiB/s
mtd_speedtest: testing page write speed
mtd_speedtest: page write speed is 3563 KiB/s
mtd_speedtest: testing page read speed
mtd_speedtest: page read speed is 8787 KiB/s
mtd_speedtest: testing 2 page write speed
mtd_speedtest: 2 page write speed is 3572 KiB/s
mtd_speedtest: testing 2 page read speed
mtd_speedtest: 2 page read speed is 8806 KiB/s
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230207065826.285013-8-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
HSSPI controller uses big endian for the opcode in the message to the
controller ping pong buffer. Use cpu_to_be16 to properly handle the
endianness for both big and little endian host.
Fixes: 142168eba9 ("spi: bcm63xx-hsspi: add bcm63xx HSSPI driver")
Signed-off-by: Kursad Oney <kursad.oney@broadcom.com>
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Link: https://lore.kernel.org/r/20230207065826.285013-7-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
New compatible string brcm,bcmbca-hsspi-v1.0 is introduced based on dts
document brcm,bcm63xx-hsspi.yaml. Add it to the driver to support this
new binding.
Signed-off-by: William Zhang <william.zhang@broadcom.com>
Link: https://lore.kernel.org/r/20230207065826.285013-6-william.zhang@broadcom.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Modern CPUs exposes this controller as PCI device that only uses
hardware sequencing capabilities which is safer than software
sequencing.
Leave the platform driver as *DANGEROUS* and update help text since
most of these controllers are using software sequencing.
Signed-off-by: Mauro Lima <mauro.lima@eclypsium.com>
Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Link: https://lore.kernel.org/r/20230206183143.75274-2-mauro.lima@eclypsium.com
Signed-off-by: Mark Brown <broonie@kernel.org>
This unlocks twice due to a merge issue.
Fixes: 35bba23e67 ("Merge remote-tracking branch 'spi/for-6.2' into spi-6.2")
Signed-off-by: Dan Carpenter <error27@gmail.com>
Link: https://lore.kernel.org/r/Y+ETCk7XKfAy8yWQ@kili
Signed-off-by: Mark Brown <broonie@kernel.org>
Add Device IDs to the module table for the following controllers:
- 9da4 Cannon Lake 300 Series On-Package
- a2a4 200 Series/Z370 Chipset Family SPI Controller
- 9d24 Intel® 200 Series Chipset Family (Including Intel® X299),
Intel® Z370 Intel® H310C,B365,
also Intel® B460 and H410 Chipset Platform Controller Hub
Signed-off-by: Mauro Lima <mauro.lima@eclypsium.com>
Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Link: https://lore.kernel.org/r/20230201205455.550308-3-mauro.lima@eclypsium.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Some private data fields have to change from bxt_info to cnl_info.
Here is the list of Device IDs with the respective documentation taken for
validation:
0xa0a4 - Intel® 500 Series Chipset Family On-Package PCH
0x02a4 - Intel® 400 Series Chipset Family On-Package PCH
0x06a4 - Intel® 400 Series Chipset Family Platform Controller Hub
0x34a4 - Intel® 495 Chipset Family On-Package
0xa3a4 - Intel® B460 and H410 Chipset
According to documentation Broxton controller type has five PR registers.
Signed-off-by: Mauro Lima <mauro.lima@eclypsium.com>
Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Link: https://lore.kernel.org/r/20230201205455.550308-2-mauro.lima@eclypsium.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Merge series from Amit Kumar Mahapatra <amit.kumar-mahapatra@amd.com>:
In preparation for supporting devices with multiple chip selects add an
interface for accessing the chip selects via a function.
Supporting multi-cs in spi core and spi controller drivers would require
the chip_select & cs_gpiod members of struct spi_device to be an array.
But changing the type of these members to array would break the spi driver
functionality. To make the transition smoother introduced four new APIs to
get/set the spi->chip_select & spi->cs_gpiod and replaced all
spi->chip_select and spi->cs_gpiod references in spi core with the API
calls.
While adding multi-cs support in further patches the chip_select & cs_gpiod
members of the spi_device structure would be converted to arrays & the
"idx" parameter of the APIs would be used as array index i.e.,
spi->chip_select[idx] & spi->cs_gpiod[idx] respectively.
Suggested-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Amit Kumar Mahapatra <amit.kumar-mahapatra@amd.com>
Reviewed-by: Michal Simek <michal.simek@amd.com>
Link: https://lore.kernel.org/r/20230119185342.2093323-2-amit.kumar-mahapatra@amd.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Selecting a symbol with additional dependencies requires
adding the same dependency here:
WARNING: unmet direct dependencies detected for MUX_MMIO
Depends on [n]: MULTIPLEXER [=y] && OF [=n]
Selected by [y]:
- SPI_DW_BT1 [=y] && SPI [=y] && SPI_MASTER [=y] && SPI_DESIGNWARE [=y] && (MIPS_BAIKAL_T1 || COMPILE_TEST [=y])
Drop the 'select' here to avoid the problem. Anyone using
the dw-bt1 SPI driver should make sure they include the
mux driver as well now.
Fixes: 7218838109 ("spi: dw-bt1: Fix undefined devm_mux_control_get symbol")
Fixes: abf0090753 ("spi: dw: Add Baikal-T1 SPI Controller glue driver")
Link: https://lore.kernel.org/all/20221218192523.c6vnfo26ua6xqf26@mobilestation/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Serge Semin <fancer.lancer@gmail.com>
Link: https://lore.kernel.org/r/20230130140156.3620863-1-arnd@kernel.org
Signed-off-by: Mark Brown <broonie@kernel.org>
When calling spidev_message() from the one of the ioctl() callbacks, the
spi_lock is already taken. When we then end up calling spidev_sync(), we
get the following splat:
[ 214.047619]
[ 214.049198] ============================================
[ 214.054533] WARNING: possible recursive locking detected
[ 214.059858] 6.2.0-rc3-0.0.0-devel+git.97ec4d559d93 #1 Not tainted
[ 214.065969] --------------------------------------------
[ 214.071290] spidev_test/1454 is trying to acquire lock:
[ 214.076530] c4925dbc (&spidev->spi_lock){+.+.}-{3:3}, at: spidev_ioctl+0x8e0/0xab8
[ 214.084164]
[ 214.084164] but task is already holding lock:
[ 214.090007] c4925dbc (&spidev->spi_lock){+.+.}-{3:3}, at: spidev_ioctl+0x44/0xab8
[ 214.097537]
[ 214.097537] other info that might help us debug this:
[ 214.104075] Possible unsafe locking scenario:
[ 214.104075]
[ 214.110004] CPU0
[ 214.112461] ----
[ 214.114916] lock(&spidev->spi_lock);
[ 214.118687] lock(&spidev->spi_lock);
[ 214.122457]
[ 214.122457] *** DEADLOCK ***
[ 214.122457]
[ 214.128386] May be due to missing lock nesting notation
[ 214.128386]
[ 214.135183] 2 locks held by spidev_test/1454:
[ 214.139553] #0: c4925dbc (&spidev->spi_lock){+.+.}-{3:3}, at: spidev_ioctl+0x44/0xab8
[ 214.147524] #1: c4925e14 (&spidev->buf_lock){+.+.}-{3:3}, at: spidev_ioctl+0x70/0xab8
[ 214.155493]
[ 214.155493] stack backtrace:
[ 214.159861] CPU: 0 PID: 1454 Comm: spidev_test Not tainted 6.2.0-rc3-0.0.0-devel+git.97ec4d559d93 #1
[ 214.169012] Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
[ 214.175555] unwind_backtrace from show_stack+0x10/0x14
[ 214.180819] show_stack from dump_stack_lvl+0x60/0x90
[ 214.185900] dump_stack_lvl from __lock_acquire+0x874/0x2858
[ 214.191584] __lock_acquire from lock_acquire+0xfc/0x378
[ 214.196918] lock_acquire from __mutex_lock+0x9c/0x8a8
[ 214.202083] __mutex_lock from mutex_lock_nested+0x1c/0x24
[ 214.207597] mutex_lock_nested from spidev_ioctl+0x8e0/0xab8
[ 214.213284] spidev_ioctl from sys_ioctl+0x4d0/0xe2c
[ 214.218277] sys_ioctl from ret_fast_syscall+0x0/0x1c
[ 214.223351] Exception stack(0xe75cdfa8 to 0xe75cdff0)
[ 214.228422] dfa0: 00000000 00001000 00000003 40206b00 bee266e8 bee266e0
[ 214.236617] dfc0: 00000000 00001000 006a71a0 00000036 004c0040 004bfd18 00000000 00000003
[ 214.244809] dfe0: 00000036 bee266c8 b6f16dc5 b6e8e5f6
Fix it by introducing an unlocked variant of spidev_sync() and calling it
from spidev_message() while other users who don't check the spidev->spi's
existence keep on using the locking flavor.
Reported-by: Francesco Dolcini <francesco@dolcini.it>
Fixes: 1f4d2dd45b ("spi: spidev: fix a race condition when accessing spidev->spi")
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
Tested-by: Max Krummenacher <max.krummenacher@toradex.com>
Link: https://lore.kernel.org/r/20230116144149.305560-1-brgl@bgdev.pl
Signed-off-by: Mark Brown <broonie@kernel.org>
Due to using the u16 type in the min_t() macros the SPI transfer length
will be cast to word before participating in the conditional statement
implied by the macro. Thus if the transfer length is greater than 64KB the
Tx/Rx FIFO threshold level value will be determined by the leftover of the
truncated after the type-case length. In the worst case it will cause the
dramatical performance drop due to the "Tx FIFO Empty" or "Rx FIFO Full"
interrupts triggered on each xfer word sent/received to/from the bus.
The problem can be easily fixed by specifying the unsigned int type in the
min_t() macros thus preventing the possible data loss.
Fixes: ea11370fff ("spi: dw: get TX level without an additional variable")
Reported-by: Sergey Nazarov <Sergey.Nazarov@baikalelectronics.ru>
Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Link: https://lore.kernel.org/r/20230113185942.2516-1-Sergey.Semin@baikalelectronics.ru
Signed-off-by: Mark Brown <broonie@kernel.org>
Merge series from Hector Martin <marcan@marcan.st>:
Commits f6c911f330 ("spi: dt-bindings: Introduce
spi-cs-setup-ns property") and 33a2fde5f7 ("spi: Introduce
spi-cs-setup-ns property") introduced a new property to represent the
CS setup delay in the device tree, but they have some issues:
- The property is only parsed as a 16-bit integer number of nanoseconds,
which limits the maximum value to ~65us. This is not a reasonable
upper limit, as some devices might need a lot more.
- The property name is inconsistent with other delay properties, which
use a "*-delay-ns" naming scheme.
- Only the setup delay is introduced, but not the related hold and
inactive delay times.
This series fixes the issues and adds support for the two missing
properties. Please pull in the first 3 patches as fixes for 6.2, to
avoid introducing a problematic DT API in this release. The last two
patches can wait until 6.3, though are probably harmless to throw in
as fixes too, since they're trivial.
Now that we support parsing the setup time from the Device Tree, we can
also easily support the remaining hold and inactive time delay values.
Signed-off-by: Janne Grunau <j@jannau.net>
Signed-off-by: Hector Martin <marcan@marcan.st>
Link: https://lore.kernel.org/r/20230113102309.18308-4-marcan@marcan.st
Signed-off-by: Mark Brown <broonie@kernel.org>
65us is not a reasonable maximum for this property, as some devices
might need a much longer setup time (e.g. those driven by firmware on
the other end). Plus, device tree property values are in 32-bit cells
and smaller widths should not be used without good reason.
Also move the logic to a helper function, since this will later be used
to parse other CS delay properties too.
Fixes: 33a2fde5f7 ("spi: Introduce spi-cs-setup-ns property")
Signed-off-by: Janne Grunau <j@jannau.net>
Signed-off-by: Hector Martin <marcan@marcan.st>
Link: https://lore.kernel.org/r/20230113102309.18308-2-marcan@marcan.st
Signed-off-by: Mark Brown <broonie@kernel.org>
The newly added spi-cs-setup-ns doesn't really fit with the existing
property names for delays, rename it so that it does before it makes it
into a release and becomes ABI.
The two debug messages in spidev_open() dereference spidev->spi without
taking the lock and without checking if it's not null. This can lead to
a crash. Drop the messages as they're not needed - the user-space will
get informed about ENOMEM with the syscall return value.
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
Link: https://lore.kernel.org/r/20230106100719.196243-2-brgl@bgdev.pl
Signed-off-by: Mark Brown <broonie@kernel.org>
There's a spinlock in place that is taken in file_operations callbacks
whenever we check if spidev->spi is still alive (not null). It's also
taken when spidev->spi is set to NULL in remove().
This however doesn't protect the code against driver unbind event while
one of the syscalls is still in progress. To that end we need a lock taken
continuously as long as we may still access spidev->spi. As both the file
ops and the remove callback are never called from interrupt context, we
can replace the spinlock with a mutex.
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
Link: https://lore.kernel.org/r/20230106100719.196243-1-brgl@bgdev.pl
Signed-off-by: Mark Brown <broonie@kernel.org>
The two debug messages in spidev_open() dereference spidev->spi without
taking the lock and without checking if it's not null. This can lead to
a crash. Drop the messages as they're not needed - the user-space will
get informed about ENOMEM with the syscall return value.
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
Link: https://lore.kernel.org/r/20230106100719.196243-2-brgl@bgdev.pl
Signed-off-by: Mark Brown <broonie@kernel.org>
There's a spinlock in place that is taken in file_operations callbacks
whenever we check if spidev->spi is still alive (not null). It's also
taken when spidev->spi is set to NULL in remove().
This however doesn't protect the code against driver unbind event while
one of the syscalls is still in progress. To that end we need a lock taken
continuously as long as we may still access spidev->spi. As both the file
ops and the remove callback are never called from interrupt context, we
can replace the spinlock with a mutex.
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
Link: https://lore.kernel.org/r/20230106100719.196243-1-brgl@bgdev.pl
Signed-off-by: Mark Brown <broonie@kernel.org>
Merge series from Hector Martin <marcan@marcan.st>:
This brings the name of the cs-setup-ns parameter which was added during
the merge window into line with other delay properties.
From: Hector Martin <marcan@marcan.st>
Date: Wed, 04 Jan 2023 18:36:26 +0900
Message-Id: <20230104093631.15611-1-marcan@marcan.st>
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
Merge changes from Hector Martin <marcan@marcan.st>:
This brings the name of the cs-setup-ns parameter which was added during
the merge window into line with other delay properties.
