Now that all users of snd_dma_continuous_data() is gone, let's drop
this ugly (and dangerous) way.
After this commit, SNDRV_DMA_TYPE_CONTINUOUS may take the standard
device pointer instead of the hacked pointer by the macro above, and
the memalloc core refers to the coherent_dma_mask of the given
device like other SNDRV_DMA_TYPE. It's still allowed to pass NULL
there, and in that case, the allocation is performed always in the
normal zone.
For SNDRV_DMA_TYPE_VMALLOC, the device pointer is simply ignored.
Link: https://lore.kernel.org/r/20220823115740.14123-5-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The recent change for memory allocator replaced the SG-buffer handling
helper for x86 with the standard non-contiguous page handler. This
works for most cases, but there is a corner case I obviously
overlooked, namely, the fallback of non-contiguous handler without
IOMMU. When the system runs without IOMMU, the core handler tries to
use the continuous pages with a single SGL entry. It works nicely for
most cases, but when the system memory gets fragmented, the large
allocation may fail frequently.
Ideally the non-contig handler could deal with the proper SG pages,
it's cumbersome to extend for now. As a workaround, here we add new
types for (minimalistic) SG allocations, instead, so that the
allocator falls back to those types automatically when the allocation
with the standard API failed.
BTW, one better (but pretty minor) improvement from the previous
SG-buffer code is that this provides the proper mmap support without
the PCM's page fault handling.
Fixes: 2c95b92ecd ("ALSA: memalloc: Unify x86 SG-buffer handling (take#3)")
BugLink: https://gitlab.freedesktop.org/pipewire/pipewire/-/issues/2272
BugLink: https://bugzilla.suse.com/show_bug.cgi?id=1198248
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20220413054808.7547-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This is a second attempt to unify the x86-specific SG-buffer handling
code with the new standard non-contiguous page handler.
The first try (in commit 2d9ea39917) failed due to the wrong page
and address calculations, hence reverted. (And the second try failed
due to a copy&paste error.) Now it's corrected with the previous fix
for noncontig pages, and the proper sg page iteration by this patch.
After the migration, SNDRV_DMA_TYPE_DMA_SG becomes identical with
SNDRV_DMA_TYPE_NONCONTIG on x86, while others still fall back to
SNDRV_DMA_TYPE_DEV.
Tested-by: Alex Xu (Hello71) <alex_y_xu@yahoo.ca>
Tested-by: Harald Arnesen <harald@skogtun.org>
Link: https://lore.kernel.org/r/20211017074859.24112-4-tiwai@suse.de
Link: https://lore.kernel.org/r/20211109062235.22310-1-tiwai@suse.de
Link: https://lore.kernel.org/r/20211116073358.19741-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This reverts commit 2d9ea39917.
We've got a regression report showing that the audio got broken the
device over AMD IOMMU. The conversion assumed the wrong pointer /
page mapping for the indirect mapping case, and we need to correct
this urgently, so let's revert it for now.
Fixes: 2d9ea39917 ("ALSA: memalloc: Convert x86 SG-buffer handling with non-contiguous type")
Reported-and-tested-by: Alex Xu (Hello71) <alex_y_xu@yahoo.ca>
Link: https://lore.kernel.org/r/20211104180846.16340-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
We've had an x86-specific SG-buffer handling code, but now it can be
merged gracefully with the standard non-contiguous DMA pages.
After the migration, SNDRV_DMA_TYPE_DMA_SG becomes identical with
SNDRV_DMA_TYPE_NONCONTIG on x86, while others still fall back to
SNDRV_DMA_TYPE_DEV.
The remaining problem is about the SG-buffer with WC pages: the DMA
core stuff on x86 doesn't treat it well, so we still need some special
handling to manipulate the page attribute manually. The mmap handler
for SNDRV_DMA_TYPE_DEV_SG_WC still returns -ENOENT intentionally for
the fallback to the default handler.
Link: https://lore.kernel.org/r/20211017074859.24112-4-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Following to the addition of non-contiguous pages, this patch adds the
new contiguous non-coherent page allocation to the standard memalloc
helper. Like the previous non-contig type, this non-coherent type is
also directional and requires the explicit sync, too. Hence the
driver using this type of buffer may need to set
SNDRV_PCM_INFO_EXPLICIT_SYNC flag to the PCM hardware.info as well,
unless it's set up in the managed mode.
Link: https://lore.kernel.org/r/20211017074859.24112-3-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch adds the support for allocation of non-contiguous DMA pages
in the common memalloc helper. It's another SG-buffer type, but
unlike the existing one, this is directional and requires the explicit
sync / invalidation of dirty pages on non-coherent architectures.
For this enhancement, the following points are changed:
- snd_dma_device stores the DMA direction.
- snd_dma_device stores need_sync flag indicating whether the explicit
sync is required or not.
- A new variant of helper functions, snd_dma_alloc_dir_pages() and
*_all() are introduced; the old snd_dma_alloc_pages() and *_all()
kept as just wrappers with DMA_BIDIRECTIONAL.
- A new helper snd_dma_buffer_sync() is introduced; this gets called
in the appropriate places.
- A new allocation type, SNDRV_DMA_TYPE_NONCONTIG, is introduced.
When the driver allocates pages with this new type, and it may require
the SNDRV_PCM_INFO_EXPLICIT_SYNC flag set to the PCM hardware.info for
taking the full control of PCM applptr and hwptr changes (that implies
disabling the mmap of control/status data). When the buffer
allocation is managed by snd_pcm_set_managed_buffer(), this flag is
automatically set depending on the result of dma_need_sync()
internally. Otherwise, if the buffer is managed manually, the driver
has to set the flag explicitly, too.
The explicit sync between CPU and device for non-coherent memory is
performed at the points before and after read/write transfer as well
as the applptr/hwptr syncptr ioctl. In the case of mmap mode,
user-space is supposed to call the syncptr ioctl with the hwptr flag
to update and fetch the status at first; this corresponds to CPU-sync.
Then user-space advances the applptr via syncptr ioctl again with
applptr flag, and this corresponds to the device sync with flushing.
Other than the DMA direction and the explicit sync, the usage of this
new buffer type is almost equivalent with the existing
SNDRV_DMA_TYPE_DEV_SG; you can get the page and the address via
snd_sgbuf_get_page() and snd_sgbuf_get_addr(), also calculate the
continuous pages via snd_sgbuf_get_chunk_size().
For those SG-page handling, the non-contig type shares the same ops
with the vmalloc handler. As we do always vmap the SG pages at first,
the actual address can be deduced from the vmapped address easily
without iterating the SG-list.
Link: https://lore.kernel.org/r/20211017074859.24112-2-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
SNDRV_DMA_TYPE_DEV_UC and SNDRV_DMA_TYPE_DEV_UC_SG are incorrectly
named as if they were for the uncached memory, while actually we set
the pages as write-combined. Rename them to reflect the right
attribute.
Acked-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20210802072815.13551-3-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This is a preparation for allowing devres usages more widely in
various sound drivers. As a first step, this patch adds a new
allocator function, snd_devm_alloc_pages(), to manage the allocated
pages via devres, so that the pages will be automagically released as
device unbinding.
Unlike the old snd_dma_alloc_pages(), the new function returns
directly the snd_dma_buffer pointer. The caller needs NULL-check for
the allocation error appropriately.
Also, since a real device pointer is mandatory for devres,
SNDRV_DMA_TYPE_CONTINUOUS or SNDRV_DMA_TYPE_VMALLOC type can't be used
for this function.
Link: https://lore.kernel.org/r/20210715075941.23332-2-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The recent refactoring of memalloc stuff removed the inclusion of
asm/page.h for simplicity, but it turned out this caused a compile
error due the lack of PAGE_SIZE definition on some architectures.
Do a partial revert for recovering from that.
Fixes: 37af81c599 ("ALSA: core: Abstract memory alloc helpers")
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/r/202106101858.PfXMMuAa-lkp@intel.com
Link: https://lore.kernel.org/r/20210610110935.10393-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch moves the mmap handling code into the common memalloc
handler. It allows us to reduce the memory-type specific code in PCM
code gracefully.
Link: https://lore.kernel.org/r/20210609162551.7842-5-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch introduces the ops table to each memory allocation type
(SNDRV_DMA_TYPE_XXX) and abstract the handling for the better code
management. Then we get separate the page allocation, release and
other tasks for each type, especially for the SG buffer.
Each buffer type has now callbacks in the struct snd_malloc_ops, and
the common helper functions call those ops accordingly. The former
inline code that is specific to SG-buffer is moved into the local
sgbuf.c, and we can simplify the PCM code without details of memory
handling.
Link: https://lore.kernel.org/r/20210609162551.7842-4-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
We have a few helper functions for making the access to the buffer
address easier on SG-buffer. Those are specific to the buffer that is
allocated with SG-buffer type, and it makes hard to use both SG and
non-SG buffers in the same code.
This patch adds a few simple checks and lets the helpers to deal with
both SG- and continuous buffers gracefully. It's a preliminary step
for the upcoming patch that mimics the buffer type on the fly.
Link: https://lore.kernel.org/r/20200615160045.2703-4-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch adds the vmalloc buffer support to ALSA memalloc core. A
new type, SNDRV_DMA_TYPE_VMALLOC was added.
The vmalloc buffer has been already supported in the PCM via a few own
helper functions, but the user sometimes get confused and misuse
them. With this patch, the whole buffer management is integrated into
the memalloc core, so they can be used in a sole common way.
Link: https://lore.kernel.org/r/20191105080138.1260-3-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Based on 1 normalized pattern(s):
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
extracted by the scancode license scanner the SPDX license identifier
GPL-2.0-or-later
has been chosen to replace the boilerplate/reference in 1334 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Allison Randal <allison@lohutok.net>
Reviewed-by: Richard Fontana <rfontana@redhat.com>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190527070033.113240726@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
snd_malloc_pages() and snd_free_pages() are merely thin wrappers of
the standard page allocator / free functions. Even the arguments are
compatible with some standard helpers, so there is little merit of
keeping these wrappers.
This patch replaces the all existing callers of snd_malloc_pages() and
snd_free_pages() with the direct calls of the standard helper
functions. In this version, we use a recently introduced one,
alloc_pages_exact(), which suits better than the old
snd_malloc_pages() implementation for our purposes. Then we can avoid
the waste of pages by alignment to power-of-two.
Since alloc_pages_exact() does split pages, we need no longer
__GFP_COMP flag; or better to say, we must not pass __GFP_COMP to
alloc_pages_exact(). So the former unconditional addition of
__GFP_COMP flag in snd_malloc_pages() is dropped, as well as in most
other places.
Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
We used to pass NULL to memory allocators for ISA devices due to
historical reasons. But we prefer rather a proper device object to be
assigned, so let's fix it by replacing snd_dma_isa_data() call with
card->dev reference, and kill snd_dma_isa_data() definition.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In some cases (mainly for x86), we need the DMA coherent buffer with
non-cached pages. Although this has been done in each driver side
like HD-audio and intel8x0, it can be done cleaner in the core memory
allocator.
This patch adds the new types, SNDRV_DMA_TYPE_DEV_UC and
SNDRV_DMA_TYPE_DEV_UC_SG, for allocating such non-cached buffer
pages. On non-x86 architectures, they work as same as the standard
SNDRV_DMA_TYPE_DEV and *_SG.
One additional change by this move is that we can assure to pass the
non-cached pgprot to the vmapped buffer, too. It eventually fixes the
case like non-snoop mode without mmap access on HD-audio.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The recent fix moved the inline snd_sgbuf_aligned_pages() outside the
ifdef, and this triggered a build error on some architectures due to
the undefined PAGE_SIZE, as spotted by 0day bot.
Fix it by adding the missing header inclusion.
Fixes: 4cae99d9b5 ("ALSA: memalloc: declare snd_sgbuf_aligned_pages() unconditionally")
Reported-by: kbuild test robot <lkp@intel.com>
Cc: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Mark Brown <broonie@kernel.org>
Make this helper inline function available for all platforms. This
helps solve 0-day compilation issues when CONFIG_SND_DMA_SGBUF is not
defined.
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Mark Brown <broonie@kernel.org>
Drop the superfluous #ifndef check in memalloc.h that had been put
just for allowing building the alsa-driver kernel modules externally.
Since the external build was discontinued years ago, let's clean up
the old kludges.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Nowadays we have CMA for obtaining the contiguous memory pages
efficiently. Let's kill the old kludge for reserving the memory pages
for large buffers. It was rarely useful (only for preserving pages
among module reloading or a little help by an early boot scripting),
used only by a couple of drivers, and yet it gives too much ugliness
than its benefit.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
When running a 32bit kernel the hda_intel driver is still reporting
a 64bit dma_mask if the HW supports it.
From sound/pci/hda/hda_intel.c:
/* allow 64bit DMA address if supported by H/W */
if ((gcap & ICH6_GCAP_64OK) && !pci_set_dma_mask(pci, DMA_BIT_MASK(64)))
pci_set_consistent_dma_mask(pci, DMA_BIT_MASK(64));
else {
pci_set_dma_mask(pci, DMA_BIT_MASK(32));
pci_set_consistent_dma_mask(pci, DMA_BIT_MASK(32));
}
which means when there is a call to dma_alloc_coherent from
snd_malloc_dev_pages a machine address bigger than 32bit can be returned.
This can be true in particular if running the 32bit kernel as a pv dom0
under the Xen Hypervisor or PAE on bare metal.
The problem is that when calling setup_bdle to program the BLE the
dma_addr_t returned from the dma_alloc_coherent is wrongly truncated
from snd_sgbuf_get_addr if running a 32bit kernel:
static inline dma_addr_t snd_sgbuf_get_addr(struct snd_dma_buffer *dmab,
size_t offset)
{
struct snd_sg_buf *sgbuf = dmab->private_data;
dma_addr_t addr = sgbuf->table[offset >> PAGE_SHIFT].addr;
addr &= PAGE_MASK;
return addr + offset % PAGE_SIZE;
}
where PAGE_MASK in a 32bit kernel is zeroing the upper 32bit af addr.
Without this patch the HW will fetch the 32bit truncated address,
which is not the one obtained from dma_alloc_coherent and will result
to a non working audio but can corrupt host memory at a random location.
The current patch apply to v3.13-rc3-74-g6c843f5
Signed-off-by: Stefano Panella <stefano.panella@citrix.com>
Reviewed-by: Frediano Ziglio <frediano.ziglio@citrix.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
It turned out that we can't use gen_pool_*() functions on archs
without CONFIG_GENERIC_ALLOCATOR (resulting in missing symbols), since
linux/genalloc.h doesn't provide dummy functions for all. We'd be
able to fix linux/genalloc.h size, but I take an easier path for
now...
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Now it's quite common that an SoC contains its on-chip internal RAM.
By using this RAM space for DMA buffer during audio playback/record,
we can shutdown the voltage for external RAM to save power.
So add new DEV type with iram malloc()/free() and accordingly modify
current default mmap() for the iram circumstance.
Signed-off-by: Nicolin Chen <b42378@freescale.com>
Reviewed-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Some audio drivers are calling snd_dma_continuous_data(GFP_KERNEL)
which makes "sparse" give a warning:
$ make C=2 M=sound/usb modules
...
sound/usb/6fire/pcm.c:625:25: warning: cast from restricted gfp_t
sound/usb/caiaq/audio.c:845:41: warning: cast from restricted gfp_t
sound/usb/usx2y/usbusx2yaudio.c:997:54: warning: cast from restricted gfp_t
sound/usb/usx2y/usbusx2yaudio.c:1001:54: warning: cast from restricted gfp_t
sound/usb/usx2y/usx2yhwdeppcm.c:774:54: warning: cast from restricted gfp_t
sound/usb/usx2y/usx2yhwdeppcm.c:778:54: warning: cast from restricted gfp_t
Add __force to the cast to silence the warning.
Signed-off-by: Antonio Ospite <ao2@amarulasolutions.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Passing struct snd_dma_buffer pointer instead, so that they work no
matter whether real SG buffer is used or not.
This is a preliminary work for the HD-audio DSP loader code.
Signed-off-by: Ian Minett <ian_minett@creativelabs.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Using SG-buffers with dma_alloc_coherent() is often very inefficient
on non-coherent architectures because a tracking record could be
allocated in addition for each dma_alloc_coherent() call.
Instead, simply disable SG-buffers but just allocate normal continuous
buffers on non-supported (currently all but x86) architectures.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Most hardwares have limited buffer-descriptor table length. This
also restricts the max buffer size of the sound driver.
For example, snd-hda-intel has 1MB buffer size limit, and this is
because it can have at most 256 BDL entries. For supporting larger
buffers, we need to allocate larger pages even for sg-buffers.
This patch changes the sgbuf allocation code to try to allocate
larger pages first. At each head of the allocated pages, the
number of allocated pages is stored in the lowest bits of the
corresponding entry of the table addr field. This change isn't
visible as long as the driver uses snd_sgbuf_get_addr() helper.
Also, the patch adds a new function, snd_pcm_sgbuf_get_chunk_size().
This returns the size of the chunk on continuous pages starting at
the given position offset. If the chunk reaches to a non-continuous
page, it returns the size to the boundary.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
Clean up SG-buffer helper functions and macros. Helpers take substream
as arguments now.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Jaroslav Kysela <perex@perex.cz>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
