Walking the dram->cs array was seen as accesses beyond the first array
item by the compiler. Instead, use the array index directly. This allows
for run-time bounds checking under CONFIG_UBSAN_BOUNDS as well. Seen
with GCC 13 with -fstrict-flex-arrays:
../sound/soc/kirkwood/kirkwood-dma.c: In function
'kirkwood_dma_conf_mbus_windows.constprop':
../sound/soc/kirkwood/kirkwood-dma.c:90:24: warning: array subscript 0 is outside array bounds of 'const struct mbus_dram_window[0]' [-Warray-bounds=]
90 | if ((cs->base & 0xffff0000) < (dma & 0xffff0000)) {
| ~~^~~~~~
Cc: Liam Girdwood <lgirdwood@gmail.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Jaroslav Kysela <perex@perex.cz>
Cc: Takashi Iwai <tiwai@suse.com>
Cc: alsa-devel@alsa-project.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20230127224128.never.410-kees@kernel.org
Signed-off-by: Mark Brown <broonie@kernel.org>
Merge series from Peter Ujfalusi <peter.ujfalusi@linux.intel.com>:
The following series will enable multi-stream support for playback and capture
streams.
Currently only a single PCM can be connected to a DAI, with the multi-stream
support it is possible to connect multiple PCMs to a single DAI.
To achieve this we need to make sure that DAIs/AIF are only set up once since
other stream could be connected to it later.
We also need to introduce reference or use counting for widgets to make sure
that they are not going to be destroyed while other streams are still using
them.
With the multi-stream support we also need to extend our current locking scheme
which worked well for simple paths.
Merge series from Astrid Rost <astrid.rost@axis.com>:
Add a generic way to create jack inputs for auxiliary jack detection
drivers (e.g. via i2c, spi), which are not part of any real codec.
The simple-card can be used as combining card driver to add the jacks,
no new one is required.
Create a jack (for input-events) for jack devices in the auxiliary
device list (aux_devs). A device which returns a valid value on
get_jack_type counts as jack device; set_jack is required
to add the jack to the device.
Add a generic way to create jack inputs for auxiliary jack detection
drivers (e.g. via i2c, spi), which are not part of any real codec.
The simple-card can be used as combining card driver to add the jacks,
no new one is required.
Create a jack (for input-events) for jack devices in the auxiliary
device list (aux_devs). A device which returns a valid value on
get_jack_type counts as jack device; set_jack is required
to add the jack to the device.
Signed-off-by: Astrid Rost <astrid.rost@axis.com>
Link: https://lore.kernel.org/r/20230123135913.2720991-3-astrid.rost@axis.com
Signed-off-by: Mark Brown <broonie@kernel.org>
The use_count of the swidget is protect by ALSA core PCM locking with the
exception when an associated kcontrol is changed.
It has been observed that a rightly timed kcontrol access during stream
stop can result of an attempt to send a control update to a widget which
has been freed up between the check of the use_count and the message
sending.
We need to protect the entire sof_widget_setup() and sof_widget_free()
execution to make it safe to rely on the use_count.
Move the code under an _unlocked() function and use a mutex to protect
the execution of the functions for concurrency.
On the control path we need to use the lock only for the kcontrol access,
the widget_kcontrol_setup() op is called with the lock already held.
Reported-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Link: https://lore.kernel.org/r/20230127120031.10709-18-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Use the started_count and paused_count to implement reference counting
when making decisions to start/stop/pause pipelines during the FE DAI
trigger. This is necessary to trigger the shared pipelines in the FE DAI
trigger properly.
With IPC4, the FE trigger will issue multiple pipeline state changes,
and the triggers are propagated downstream to connected pipelines by
the SOF driver - not the firmware. This creates a window for race
conditions where an FE trigger preempts another one, which results
in inconsistent pipeline states and refcounts.
This patch introduces a mutex lock for the pcm trigger that guarantees
that IPC4 state and resources are accessed in a serialized manner.
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Rander Wang <rander.wang@intel.com>
Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com>
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Link: https://lore.kernel.org/r/20230127120031.10709-15-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Add a new flag, skip_during_fe_trigger, to struct sof_ipc4_pipeline to
skip triggering pipelines in the FE DAI trigger. Set this flag for the
HDA DAI BE pipelines so that their BE pipeline will not be triggered in
the FE DAI trigger. Also, move the trigger handling for all commands
include START/PAUSE_RELEASE for the HDA DAI's to the backend DAI trigger
ops.
For the SSP/DMIC/SDW cases, remove the BE DAI trigger as they involve no
DMA operations and can be triggered in the FE DAI trigger. This is in
preparation to perform batch triggering of all pipelines for the non-HDA
case.
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Reviewed-by: Libin Yang <libin.yang@intel.com>
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com>
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Link: https://lore.kernel.org/r/20230127120031.10709-10-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
IPC3 and IPC4 have different requirements for the order in which the FE
CPU and BE CPU DAI trigger callbacks must be invoked. With a regular PCM
start/stop, pipeline widgets are set up during hw_params and freed
during hw_free.
But when the system is suspended when a PCM is running,
pipeline widgets are freed during the SUSPEND trigger callback for the
FE CPU DAI. In order to avoid freeing the pipeline widgets before the BE
CPU DAI trigger is executed, the trigger order was modified in previous
contributions in the PCM dai_link_fixup callback to make sure that the BE
CPU DAI trigger stop/suspend is always invoked before the FE CPU DAI
trigger. But this contradicts the firmware requirement for IPC4 w.r.t.
ordering of pipeline triggers.
So, remove the freeing of pipeline widgets during FE CPU DAI suspend
trigger and handle it during system suspend when the
tear_down_all_pipelines() IPC op is invoked. This will be followed up
with a patch to fix the trigger order for IPC4.
Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
Reviewed-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Reviewed-by: Péter Ujfalusi <peter.ujfalusi@linux.intel.com>
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Link: https://lore.kernel.org/r/20230127120031.10709-6-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
If panic_on_warn is set and compress stream(DPCM) is started,
then kernel panic occurred because card->pcm_mutex isn't held appropriately.
In the following functions, warning were issued at this line
"snd_soc_dpcm_mutex_assert_held".
static int dpcm_be_connect(struct snd_soc_pcm_runtime *fe,
struct snd_soc_pcm_runtime *be, int stream)
{
...
snd_soc_dpcm_mutex_assert_held(fe);
...
}
void dpcm_be_disconnect(struct snd_soc_pcm_runtime *fe, int stream)
{
...
snd_soc_dpcm_mutex_assert_held(fe);
...
}
void snd_soc_runtime_action(struct snd_soc_pcm_runtime *rtd,
int stream, int action)
{
...
snd_soc_dpcm_mutex_assert_held(rtd);
...
}
int dpcm_dapm_stream_event(struct snd_soc_pcm_runtime *fe, int dir,
int event)
{
...
snd_soc_dpcm_mutex_assert_held(fe);
...
}
These functions are called by soc_compr_set_params_fe, soc_compr_open_fe
and soc_compr_free_fe
without pcm_mutex locking. And this is call stack.
[ 414.527841][ T2179] pc : dpcm_process_paths+0x5a4/0x750
[ 414.527848][ T2179] lr : dpcm_process_paths+0x37c/0x750
[ 414.527945][ T2179] Call trace:
[ 414.527949][ T2179] dpcm_process_paths+0x5a4/0x750
[ 414.527955][ T2179] soc_compr_open_fe+0xb0/0x2cc
[ 414.527972][ T2179] snd_compr_open+0x180/0x248
[ 414.527981][ T2179] snd_open+0x15c/0x194
[ 414.528003][ T2179] chrdev_open+0x1b0/0x220
[ 414.528023][ T2179] do_dentry_open+0x30c/0x594
[ 414.528045][ T2179] vfs_open+0x34/0x44
[ 414.528053][ T2179] path_openat+0x914/0xb08
[ 414.528062][ T2179] do_filp_open+0xc0/0x170
[ 414.528068][ T2179] do_sys_openat2+0x94/0x18c
[ 414.528076][ T2179] __arm64_sys_openat+0x78/0xa4
[ 414.528084][ T2179] invoke_syscall+0x48/0x10c
[ 414.528094][ T2179] el0_svc_common+0xbc/0x104
[ 414.528099][ T2179] do_el0_svc+0x34/0xd8
[ 414.528103][ T2179] el0_svc+0x34/0xc4
[ 414.528125][ T2179] el0t_64_sync_handler+0x8c/0xfc
[ 414.528133][ T2179] el0t_64_sync+0x1a0/0x1a4
[ 414.528142][ T2179] Kernel panic - not syncing: panic_on_warn set ...
So, I reposition and add pcm_mutex to resolve lockdep error.
Signed-off-by: Shinhyung Kang <s47.kang@samsung.com>
Link: https://lore.kernel.org/r/016401d90ac4$7b6848c0$7238da40$@samsung.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Replace SNDRV_DMA_TYPE_DEV by SNDRV_DMA_TYPE_NONCOHERENT for pcm
buffer allocation. We measured about 18x performance improvement
when accessing the snd_pcm_mmap() buffer from userspace.
Eg: aarch64 with internal speaker
==
SNDRV_DMA_TYPE_DEV
frames_per_second=66.4557M/s
max_time_per_4096_frames=106.094u
time_per_4096_frames=61.6351us
SNDRV_DMA_TYPE_NONCOHERENT
frames_per_second=1028.58M/s
max_time_per_4096_frames=20.312u
time_per_4096_frames=3.98221us
Signed-off-by: Judy Hsiao <judyhsiao@chromium.org>
Link: https://lore.kernel.org/r/20221208084200.2136311-1-judyhsiao@chromium.org
Signed-off-by: Mark Brown <broonie@kernel.org>
