ASoC: Intel: avs: DSP recovery and resume fixes
Subsystems (3)
Cezary Rojewski
maintainercover ASoC: Intel: avs: DSP recovery and resume fixes
Two fixes that are result of the recent discussions [1][2].
First adds missing locking around snd_pcm_stop() while the second fix sets substream state to DISCONNECTED if any suspend/resume related operation fails so that userspace has means to be aware that something went wrong during said operation.
[1]: https://lore.kernel.org/alsa-devel/[email protected]/
[2]: https://lore.kernel.org/alsa-devel/[email protected]/
Cezary Rojewski (2):
ASoC: Intel: avs: Lock substream before snd_pcm_stop()
ASoC: Intel: avs: Disconnect substream if suspend or resume fails
sound/soc/intel/avs/ipc.c | 2 ++ sound/soc/intel/avs/pcm.c | 10 ++++++++-- 2 files changed, 10 insertions(+), 2 deletions(-)
Signature · 1 line
2.25.1
Cezary Rojewski
maintainerPATCH 1/2 ASoC: Intel: avs: Lock substream before snd_pcm_stop()
snd_pcm_stop() shall be called with stream lock held to prevent any races between nonatomic streaming operations.
--- sound/soc/intel/avs/ipc.c | 2 ++ 1 file changed, 2 insertions(+)
sound/soc/intel/avs/ipc.c +2 -0
@@ -123,7 +123,9 @@ static void avs_dsp_recovery(struct avs_dev *adev)if (!substream || !substream->runtime)continue ; +snd_pcm_stream_lock (substream);snd_pcm_stop (substream,SNDRV_PCM_STATE_DISCONNECTED ); +snd_pcm_stream_unlock (substream); } } }
Signature · 1 line
2.25.1
Cezary Rojewski
maintainerPATCH 2/2 ASoC: Intel: avs: Disconnect substream if suspend or resume fails
In reply to Cezary RojewskiTo improve performance and overall system stability, suspend/resume operations for ASoC cards always return success status and defer the actual work.
Because of that, if a substream fails to resume, userspace may still attempt to invoke commands on it as from their perspective the operation completed successfully. Set substream's state to DISCONNECTED to ensure no further commands are attempted.
--- sound/soc/intel/avs/pcm.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)
sound/soc/intel/avs/pcm.c +8 -2
@@ -934,8 +934,11 @@ static int avs_component_pm_op(struct snd_soc_component *component, bool be, rtd =snd_pcm_substream_chip (data->substream);if (rtd->dai_link->no_pcm == be && !rtd->dai_link->ignore_suspend) { ret =op (dai, data); -if (ret <0 ) +if (ret <0 ) { + data->substream->runtime->status->state = +SNDRV_PCM_STATE_DISCONNECTED ;return ret; + } } }
@@ -944,8 +947,11 @@ static int avs_component_pm_op(struct snd_soc_component *component, bool be, rtd =snd_pcm_substream_chip (data->substream);if (rtd->dai_link->no_pcm == be && !rtd->dai_link->ignore_suspend) { ret =op (dai, data); -if (ret <0 ) +if (ret <0 ) { + data->substream->runtime->status->state = +SNDRV_PCM_STATE_DISCONNECTED ;return ret; + } } } }
Signature · 1 line
2.25.1
Re: [PATCH 2/2] ASoC: Intel: avs: Disconnect substream if suspend or resume fails
Quoted 19 lines · Cezary Rojewski
To improve performance and overall system stability, suspend/resume operations for ASoC cards always return success status and defer the actual work.
Because of that, if a substream fails to resume, userspace may still attempt to invoke commands on it as from their perspective the operation completed successfully. Set substream's state to DISCONNECTED to ensure no further commands are attempted.
Signed-off-by Cezary Rojewski <[email protected]>--- sound/soc/intel/avs/pcm.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)sound/soc/intel/avs/pcm.c +0
-0
sound/soc/intel/avs/pcm.c static int avs_component_pm_op(struct snd_soc_component *component, bool be, · line 939
Quoted 7 lines
sound/soc/intel/avs/pcm.c +3
-1@@ -934,8 +934,11 @@ static int avs_component_pm_op(struct snd_soc_component *component, bool be, rtd =snd_pcm_substream_chip (data->substream);if (rtd->dai_link->no_pcm == be && !rtd->dai_link->ignore_suspend) { ret =op (dai, data); -if (ret <0 ) +if (ret <0 ) { + data->substream->runtime->status->state = +SNDRV_PCM_STATE_DISCONNECTED ;
should runtime->state be used instead of runtime->status->state?
A quick grep shows the former seems more popular in drivers, status->seems to be only used in pcm_native.c?
Another plumbing question is whether it's actually ok to change the state of the runtime in a driver, are you not going to have locking issues? Very few drivers change the internal state and I wonder how legit/safe this is.
Quoted 18 lines
sound/soc/intel/avs/pcm.c +5
-1return ret; + } } }@@ -944,8 +947,11 @@ static int avs_component_pm_op(struct snd_soc_component *component, bool be, rtd =snd_pcm_substream_chip (data->substream);if (rtd->dai_link->no_pcm == be && !rtd->dai_link->ignore_suspend) { ret =op (dai, data); -if (ret <0 ) +if (ret <0 ) { + data->substream->runtime->status->state = +SNDRV_PCM_STATE_DISCONNECTED ;return ret; + } } } }
Cezary Rojewski
maintainerRe: [PATCH 2/2] ASoC: Intel: avs: Disconnect substream if suspend or resume fails
Quoted 15 lines · Pierre-Louis Bossart
sound/soc/intel/avs/pcm.c static int avs_component_pm_op(struct snd_soc_component *component, bool be, · line 939
Quoted 10 lines · Cezary Rojewski
sound/soc/intel/avs/pcm.c +3
-1@@ -934,8 +934,11 @@ static int avs_component_pm_op(struct snd_soc_component *component, bool be, rtd =snd_pcm_substream_chip (data->substream);if (rtd->dai_link->no_pcm == be && !rtd->dai_link->ignore_suspend) { ret =op (dai, data); -if (ret <0 ) +if (ret <0 ) { + data->substream->runtime->status->state = +SNDRV_PCM_STATE_DISCONNECTED ;should
runtime->statebe used instead ofruntime->status->state?
Quoted 7 lines
A quick grep shows the former seems more popular in drivers,
status->seemsto be only used inpcm_native.c?Another plumbing question is whether it's actually ok to change the state of the runtime in a driver, are you not going to have locking issues? Very few drivers change the internal state and I wonder how legit/safe this is.
Unless something new has been added/updated, there is no runtime->state field available. All the PCM code works with runtime->status->state.
component->resume() gets fired before any PCMs have a chance to be resumed. Userspace cannot access us _yet_. Similarly for suspend, component->suspend() is called once all the streams receive TRIGGER_SUSPEND and from then on, we're on device-pm level already.
Well, in regard to grep, very few drivers enter the recovery jungle. All of this is to help improve user experience when things go south. Unfortunately for us, restoring regmap is insufficient to get AudioDSP happy. Right now if things do go south, userspace still performs all of the PCM commands on the stream as nothing has happened - snd_soc_suspend/resume() return 0 in all cases.
Regards,
Czarek
Re: [PATCH 2/2] ASoC: Intel: avs: Disconnect substream if suspend or resume fails
Quoted 21 lines · Cezary Rojewski
Quoted 20 lines · Pierre-Louis Bossart
Quoted 5 lines · Cezary Rojewski
sound/soc/intel/avs/pcm.c +0
-0@@ -934,8 +934,11 @@ static int avs_component_pm_op(struct snd_soc_component *component, bool be, rtd = snd_pcm_substream_chip(data->substream);Quoted 7 lines
if (
rtd->dai_link->no_pcm== be && !rtd->dai_link->ignore_suspend) {
ret =op(dai, data);
- if (ret < 0)
+ if (ret < 0) {
+data->substream->runtime->status->state= +SNDRV_PCM_STATE_DISCONNECTED;should
runtime->statebe used instead ofruntime->status->state?A quick grep shows the former seems more popular in drivers,
status->seemsto be only used inpcm_native.c?
Quoted 8 lines
Quoted 4 lines
Another plumbing question is whether it's actually ok to change the state of the runtime in a driver, are you not going to have locking issues? Very few drivers change the internal state and I wonder how legit/safe this is.
Unless something new has been added/updated, there is no
runtime->statefield available. All the PCM code works withruntime->status->state.
cd sound/
git grep -c 'runtime->state'core/compress_offload.c:27core/oss/pcm_oss.c:21core/pcm.c:2core/pcm_compat.c:2core/pcm_lib.c:8core/pcm_native.c:50drivers/aloop.c:2firewire/bebob/bebob_pcm.c:2firewire/dice/dice-pcm.c:2firewire/digi00x/digi00x-pcm.c:2firewire/fireface/ff-pcm.c:2firewire/fireworks/fireworks_pcm.c:2firewire/motu/motu-pcm.c:2firewire/oxfw/oxfw-pcm.c:4firewire/tascam/tascam-pcm.c:2hda/hdmi_chmap.c:1isa/gus/gus_pcm.c:1soc/intel/skylake/skl-pcm.c:2soc/sh/rz-ssi.c:1usb/pcm.c:2usb/usx2y/usbusx2yaudio.c:1usb/usx2y/usx2yhwdeppcm.c:1
git grep -c 'status->state'core/pcm_compat.c:2core/pcm_native.c:4
Cezary Rojewski
maintainerRe: [PATCH 2/2] ASoC: Intel: avs: Disconnect substream if suspend or resume fails
Quoted 9 lines · Pierre-Louis Bossart
Quoted 2 lines · Cezary Rojewski
Unless something new has been added/updated, there is no
runtime->statefield available. All the PCM code works withruntime->status->state.cd
sound/git grep -c '
runtime->state'core/compress_offload.c:27
...
Quoted 3 lines
git grep -c '
status->state'core/pcm_compat.c:2core/pcm_native.c:4
I see now, thanks. Commit from late September "ALSA: pcm: Avoid reference to status->state" add a new field. Will update the code to use __snd_pcm_set_state() instead.
My base did not have it yet.
Re: [PATCH 2/2] ASoC: Intel: avs: Disconnect substream if suspend or resume fails
Quoted 17 lines · Cezary Rojewski
Quoted 9 lines · Pierre-Louis Bossart
Quoted 2 lines · Cezary Rojewski
Unless something new has been added/updated, there is no
runtime->statefield available. All the PCM code works withruntime->status->state.cd
sound/git grep -c '
runtime->state'core/compress_offload.c:27...
Quoted 3 lines
git grep -c '
status->state'core/pcm_compat.c:2core/pcm_native.c:4
Quoted 5 lines
I see now, thanks. Commit from late September "ALSA: pcm: Avoid reference to
status->state" add a new field. Will update the code to use__snd_pcm_set_state()instead.My base did not have it yet.
Right, it's relatively recent, and my point is that the helper does use stream locking when testing the same state you modify. Maybe that's a red herring but I thought I'd ask.
static void snd_pcm_set_state(struct snd_pcm_substream *substream,
snd_pcm_state_t state)
{
snd_pcm_stream_lock_irq(substream);
if (substream->runtime->state != SNDRV_PCM_STATE_DISCONNECTED)
__snd_pcm_set_state(substream->runtime, state);
snd_pcm_stream_unlock_irq(substream);
}Cezary Rojewski
maintainerRe: [PATCH 2/2] ASoC: Intel: avs: Disconnect substream if suspend or resume fails
Quoted 28 lines · Pierre-Louis Bossart
Quoted 17 lines · Cezary Rojewski
Quoted 9 lines · Pierre-Louis Bossart
Quoted 2 lines · Cezary Rojewski
Unless something new has been added/updated, there is no
runtime->statefield available. All the PCM code works withruntime->status->state.cd
sound/git grep -c '
runtime->state'core/compress_offload.c:27...
Quoted 3 lines
git grep -c '
status->state'core/pcm_compat.c:2core/pcm_native.c:4Quoted 5 lines
I see now, thanks. Commit from late September "ALSA: pcm: Avoid reference to
status->state" add a new field. Will update the code to use__snd_pcm_set_state()instead.My base did not have it yet.
Right, it's relatively recent, and my point is that the helper does use stream locking when testing the same state you modify. Maybe that's a red herring but I thought I'd ask.
Quoted 9 lines
static void snd_pcm_set_state(struct
snd_pcm_substream*substream,
snd_pcm_state_tstate){ snd_pcm_stream_lock_irq(substream); if (substream->runtime->state != SNDRV_PCM_STATE_DISCONNECTED) __snd_pcm_set_state(substream->runtime, state); snd_pcm_stream_unlock_irq(substream); }
Your question is a right one and this is the right time to talk about locking. Our current test results and my analysis show that locking is not needed (what isn't the case for the first patch in the series) but races such as this one are hard to reproduce. If I'm proven wrong, no problem updating the code on my side.