[alsa-devel] [PATCH] ASoC: dpcm: Fix race between FE/BE updates and trigger
DPCM can update the FE/BE connection states totally asynchronously from the FE's PCM state. Most of FE/BE state changes are protected by mutex, so that they won't race, but there are still some actions that are uncovered. For example, suppose to switch a BE while a FE's stream is running. This would call soc_dpcm_runtime_update(), which sets FE's runtime_update flag, then sets up and starts BEs, and clears FE's runtime_update flag again.
When a device emits XRUN during this operation, the PCM core triggers snd_pcm_stop(XRUN). Since the trigger action is an atomic ops, this isn't blocked by the mutex, thus it kicks off DPCM's trigger action. It eventually updates and clears FE's runtime_update flag while soc_dpcm_runtime_update() is running concurrently, and it results in confusion.
Usually, for avoiding such a race, we take a lock. There is a PCM stream lock for that purpose. However, as already mentioned, the trigger action is atomic, and we can't take the lock for the whole soc_dpcm_runtime_update() or other operations that include the lengthy jobs like hw_params or prepare.
This patch provides an alternative solution. This adds a way to defer the conflicting trigger callback to be executed at the end of FE/BE state changes. For doing it, two things are introduced:
- Each runtime_update state change of FEs is protected via PCM stream lock. - The FE's trigger callback checks the runtime_update flag. If it's not set, the trigger action is executed there. If set, mark the pending trigger action and returns immediately. - At the exit of runtime_update state change, it checks whether the pending trigger is present. If yes, it executes the trigger action at this point.
Reported-and-tested-by: Qiao Zhou zhouqiao@marvell.com Signed-off-by: Takashi Iwai tiwai@suse.de --- include/sound/soc-dpcm.h | 2 ++ sound/soc/soc-pcm.c | 72 +++++++++++++++++++++++++++++++++++++----------- 2 files changed, 58 insertions(+), 16 deletions(-)
diff --git a/include/sound/soc-dpcm.h b/include/sound/soc-dpcm.h index 2883a7a6f9f3..98f2ade0266e 100644 --- a/include/sound/soc-dpcm.h +++ b/include/sound/soc-dpcm.h @@ -102,6 +102,8 @@ struct snd_soc_dpcm_runtime { /* state and update */ enum snd_soc_dpcm_update runtime_update; enum snd_soc_dpcm_state state; + + int trigger_pending; /* trigger cmd + 1 if pending, 0 if not */ };
/* can this BE stop and free */ diff --git a/sound/soc/soc-pcm.c b/sound/soc/soc-pcm.c index 002311afdeaa..57277dd79e11 100644 --- a/sound/soc/soc-pcm.c +++ b/sound/soc/soc-pcm.c @@ -1522,13 +1522,36 @@ static void dpcm_set_fe_runtime(struct snd_pcm_substream *substream) dpcm_init_runtime_hw(runtime, &cpu_dai_drv->capture); }
+static int dpcm_fe_dai_do_trigger(struct snd_pcm_substream *substream, int cmd); + +/* Set FE's runtime_update state; the state is protected via PCM stream lock + * for avoiding the race with trigger callback. + * If the state is unset and a trigger is pending while the previous operation, + * process the pending trigger action here. + */ +static void dpcm_set_fe_update_state(struct snd_soc_pcm_runtime *fe, + int stream, enum snd_soc_dpcm_update state) +{ + struct snd_pcm_substream *substream = + snd_soc_dpcm_get_substream(fe, stream); + + snd_pcm_stream_lock_irq(substream); + if (state == SND_SOC_DPCM_UPDATE_NO && fe->dpcm[stream].trigger_pending) { + dpcm_fe_dai_do_trigger(substream, + fe->dpcm[stream].trigger_pending - 1); + fe->dpcm[stream].trigger_pending = 0; + } + fe->dpcm[stream].runtime_update = state; + snd_pcm_stream_unlock_irq(substream); +} + static int dpcm_fe_dai_startup(struct snd_pcm_substream *fe_substream) { struct snd_soc_pcm_runtime *fe = fe_substream->private_data; struct snd_pcm_runtime *runtime = fe_substream->runtime; int stream = fe_substream->stream, ret = 0;
- fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_FE; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_FE);
ret = dpcm_be_dai_startup(fe, fe_substream->stream); if (ret < 0) { @@ -1550,13 +1573,13 @@ static int dpcm_fe_dai_startup(struct snd_pcm_substream *fe_substream) dpcm_set_fe_runtime(fe_substream); snd_pcm_limit_hw_rates(runtime);
- fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_NO); return 0;
unwind: dpcm_be_dai_startup_unwind(fe, fe_substream->stream); be_err: - fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_NO); return ret; }
@@ -1603,7 +1626,7 @@ static int dpcm_fe_dai_shutdown(struct snd_pcm_substream *substream) struct snd_soc_pcm_runtime *fe = substream->private_data; int stream = substream->stream;
- fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_FE; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_FE);
/* shutdown the BEs */ dpcm_be_dai_shutdown(fe, substream->stream); @@ -1617,7 +1640,7 @@ static int dpcm_fe_dai_shutdown(struct snd_pcm_substream *substream) dpcm_dapm_stream_event(fe, stream, SND_SOC_DAPM_STREAM_STOP);
fe->dpcm[stream].state = SND_SOC_DPCM_STATE_CLOSE; - fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_NO); return 0; }
@@ -1665,7 +1688,7 @@ static int dpcm_fe_dai_hw_free(struct snd_pcm_substream *substream) int err, stream = substream->stream;
mutex_lock_nested(&fe->card->mutex, SND_SOC_CARD_CLASS_RUNTIME); - fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_FE; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_FE);
dev_dbg(fe->dev, "ASoC: hw_free FE %s\n", fe->dai_link->name);
@@ -1680,7 +1703,7 @@ static int dpcm_fe_dai_hw_free(struct snd_pcm_substream *substream) err = dpcm_be_dai_hw_free(fe, stream);
fe->dpcm[stream].state = SND_SOC_DPCM_STATE_HW_FREE; - fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_NO);
mutex_unlock(&fe->card->mutex); return 0; @@ -1773,7 +1796,7 @@ static int dpcm_fe_dai_hw_params(struct snd_pcm_substream *substream, int ret, stream = substream->stream;
mutex_lock_nested(&fe->card->mutex, SND_SOC_CARD_CLASS_RUNTIME); - fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_FE; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_FE);
memcpy(&fe->dpcm[substream->stream].hw_params, params, sizeof(struct snd_pcm_hw_params)); @@ -1796,7 +1819,7 @@ static int dpcm_fe_dai_hw_params(struct snd_pcm_substream *substream, fe->dpcm[stream].state = SND_SOC_DPCM_STATE_HW_PARAMS;
out: - fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_NO); mutex_unlock(&fe->card->mutex); return ret; } @@ -1910,7 +1933,7 @@ int dpcm_be_dai_trigger(struct snd_soc_pcm_runtime *fe, int stream, } EXPORT_SYMBOL_GPL(dpcm_be_dai_trigger);
-static int dpcm_fe_dai_trigger(struct snd_pcm_substream *substream, int cmd) +static int dpcm_fe_dai_do_trigger(struct snd_pcm_substream *substream, int cmd) { struct snd_soc_pcm_runtime *fe = substream->private_data; int stream = substream->stream, ret; @@ -1984,6 +2007,23 @@ out: return ret; }
+static int dpcm_fe_dai_trigger(struct snd_pcm_substream *substream, int cmd) +{ + struct snd_soc_pcm_runtime *fe = substream->private_data; + int stream = substream->stream; + + /* if FE's runtime_update is already set, we're in race; + * process this trigger later at exit + */ + if (fe->dpcm[stream].runtime_update != SND_SOC_DPCM_UPDATE_NO) { + fe->dpcm[stream].trigger_pending = cmd + 1; + return 0; /* delayed, assuming it's successful */ + } + + /* we're alone, let's trigger */ + return dpcm_fe_dai_do_trigger(substream, cmd); +} + int dpcm_be_dai_prepare(struct snd_soc_pcm_runtime *fe, int stream) { struct snd_soc_dpcm *dpcm; @@ -2027,7 +2067,7 @@ static int dpcm_fe_dai_prepare(struct snd_pcm_substream *substream)
dev_dbg(fe->dev, "ASoC: prepare FE %s\n", fe->dai_link->name);
- fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_FE; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_FE);
/* there is no point preparing this FE if there are no BEs */ if (list_empty(&fe->dpcm[stream].be_clients)) { @@ -2054,7 +2094,7 @@ static int dpcm_fe_dai_prepare(struct snd_pcm_substream *substream) fe->dpcm[stream].state = SND_SOC_DPCM_STATE_PREPARE;
out: - fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_NO); mutex_unlock(&fe->card->mutex);
return ret; @@ -2201,11 +2241,11 @@ static int dpcm_run_new_update(struct snd_soc_pcm_runtime *fe, int stream) { int ret;
- fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_BE; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_BE); ret = dpcm_run_update_startup(fe, stream); if (ret < 0) dev_err(fe->dev, "ASoC: failed to startup some BEs\n"); - fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_NO);
return ret; } @@ -2214,11 +2254,11 @@ static int dpcm_run_old_update(struct snd_soc_pcm_runtime *fe, int stream) { int ret;
- fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_BE; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_BE); ret = dpcm_run_update_shutdown(fe, stream); if (ret < 0) dev_err(fe->dev, "ASoC: failed to shutdown some BEs\n"); - fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO; + dpcm_set_fe_update_state(fe, stream, SND_SOC_DPCM_UPDATE_NO);
return ret; }
On Tue, 2014-11-04 at 16:52 +0100, Takashi Iwai wrote:
DPCM can update the FE/BE connection states totally asynchronously from the FE's PCM state. Most of FE/BE state changes are protected by mutex, so that they won't race, but there are still some actions that are uncovered. For example, suppose to switch a BE while a FE's stream is running. This would call soc_dpcm_runtime_update(), which sets FE's runtime_update flag, then sets up and starts BEs, and clears FE's runtime_update flag again.
When a device emits XRUN during this operation, the PCM core triggers snd_pcm_stop(XRUN). Since the trigger action is an atomic ops, this isn't blocked by the mutex, thus it kicks off DPCM's trigger action. It eventually updates and clears FE's runtime_update flag while soc_dpcm_runtime_update() is running concurrently, and it results in confusion.
Usually, for avoiding such a race, we take a lock. There is a PCM stream lock for that purpose. However, as already mentioned, the trigger action is atomic, and we can't take the lock for the whole soc_dpcm_runtime_update() or other operations that include the lengthy jobs like hw_params or prepare.
This patch provides an alternative solution. This adds a way to defer the conflicting trigger callback to be executed at the end of FE/BE state changes. For doing it, two things are introduced:
- Each runtime_update state change of FEs is protected via PCM stream lock.
- The FE's trigger callback checks the runtime_update flag. If it's not set, the trigger action is executed there. If set, mark the pending trigger action and returns immediately.
- At the exit of runtime_update state change, it checks whether the pending trigger is present. If yes, it executes the trigger action at this point.
Reported-and-tested-by: Qiao Zhou zhouqiao@marvell.com Signed-off-by: Takashi Iwai tiwai@suse.de
Acked-by: Liam Girdwood liam.r.girdwood@linux.intel.com
On Tue, Nov 04, 2014 at 04:52:28PM +0100, Takashi Iwai wrote:
DPCM can update the FE/BE connection states totally asynchronously from the FE's PCM state. Most of FE/BE state changes are protected by mutex, so that they won't race, but there are still some actions that
Applied, thanks.
participants (3)
-
Liam Girdwood
-
Mark Brown
-
Takashi Iwai