| Message ID | 20200415104928.86091-1-stephan@gerhold.net (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 9b5db059366ae2087e07892b5fc108f81f4ec189 |
| Headers | show |
| Series | ASoC: soc-pcm: dpcm: Only allow playback/capture if supported | expand |
> diff --git a/sound/soc/soc-pcm.c b/sound/soc/soc-pcm.c > index 454735f8fa92..77a680da366f 100644 > --- a/sound/soc/soc-pcm.c > +++ b/sound/soc/soc-pcm.c > @@ -2911,8 +2911,17 @@ int soc_new_pcm(struct snd_soc_pcm_runtime *rtd, int num) > int i; > > if (rtd->dai_link->dynamic || rtd->dai_link->no_pcm) { > - playback = rtd->dai_link->dpcm_playback; > - capture = rtd->dai_link->dpcm_capture; > + cpu_dai = asoc_rtd_to_cpu(rtd, 0); > + if (rtd->num_cpus > 1) { > + dev_err(rtd->dev, > + "DPCM doesn't support Multi CPU yet\n"); > + return -EINVAL; > + } > + > + playback = rtd->dai_link->dpcm_playback && > + snd_soc_dai_stream_valid(cpu_dai, SNDRV_PCM_STREAM_PLAYBACK); > + capture = rtd->dai_link->dpcm_capture && > + snd_soc_dai_stream_valid(cpu_dai, SNDRV_PCM_STREAM_CAPTURE); This commit introduces major regressions with SOF on CherryTrail and Broadwell: [ 25.705750] SSP2-Codec: ASoC: no backend playback stream [ 27.923378] SSP2-Codec: ASoC: no users playback at close - state it's likely due to the check for min_channels > 0 in snd_soc_dai_stream_valid(), which wasn't a requirement before. We are testing a fix [1] but other users of DPCM might be impacted. Mark, this commit is on your for-5.7 branch but not on for-next? Not sure which SHA1 to use for the Fixes: tag [1] https://github.com/thesofproject/linux/pull/2018/commits/4fa10638dca8aad7a320e85cc3e00b179b8de410
On Thu, Apr 16, 2020 at 03:51:23PM -0500, Pierre-Louis Bossart wrote: > > > diff --git a/sound/soc/soc-pcm.c b/sound/soc/soc-pcm.c > > index 454735f8fa92..77a680da366f 100644 > > --- a/sound/soc/soc-pcm.c > > +++ b/sound/soc/soc-pcm.c > > @@ -2911,8 +2911,17 @@ int soc_new_pcm(struct snd_soc_pcm_runtime *rtd, int num) > > int i; > > if (rtd->dai_link->dynamic || rtd->dai_link->no_pcm) { > > - playback = rtd->dai_link->dpcm_playback; > > - capture = rtd->dai_link->dpcm_capture; > > + cpu_dai = asoc_rtd_to_cpu(rtd, 0); > > + if (rtd->num_cpus > 1) { > > + dev_err(rtd->dev, > > + "DPCM doesn't support Multi CPU yet\n"); > > + return -EINVAL; > > + } > > + > > + playback = rtd->dai_link->dpcm_playback && > > + snd_soc_dai_stream_valid(cpu_dai, SNDRV_PCM_STREAM_PLAYBACK); > > + capture = rtd->dai_link->dpcm_capture && > > + snd_soc_dai_stream_valid(cpu_dai, SNDRV_PCM_STREAM_CAPTURE); > > This commit introduces major regressions with SOF on CherryTrail and > Broadwell: > > [ 25.705750] SSP2-Codec: ASoC: no backend playback stream > [ 27.923378] SSP2-Codec: ASoC: no users playback at close - state > > it's likely due to the check for min_channels > 0 in > snd_soc_dai_stream_valid(), which wasn't a requirement before. > > We are testing a fix [1] but other users of DPCM might be impacted. > Indeed, I actually ran into a similar problem myself for q6afe-dai: https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound.git/commit/?h=for-5.7&id=0c824ec094b5cda766c80d88c2036e28c24a4cb1 As mentioned in that commit message it was already broken on 5.7-rc1 for me, because of commit 0e9cf4c452ad ("ASoC: pcm: check if cpu-dai supports a given stream"). [2] With this commit it's more visible at least, you get a proper error instead of silently not calling hw_params() for example. :) [2]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=0e9cf4c452ad7e2776441cbac0b9983abaf17ff0 > Mark, this commit is on your for-5.7 branch but not on for-next? Not sure > which SHA1 to use for the Fixes: tag > > [1] https://github.com/thesofproject/linux/pull/2018/commits/4fa10638dca8aad7a320e85cc3e00b179b8de410
On Thu, Apr 16, 2020 at 03:51:23PM -0500, Pierre-Louis Bossart wrote: > Mark, this commit is on your for-5.7 branch but not on for-next? Not sure > which SHA1 to use for the Fixes: tag The commit you're fixing should be the commit you're fixing regardless of which branch it is on.
diff --git a/sound/soc/soc-pcm.c b/sound/soc/soc-pcm.c index 454735f8fa92..77a680da366f 100644 --- a/sound/soc/soc-pcm.c +++ b/sound/soc/soc-pcm.c @@ -2911,8 +2911,17 @@ int soc_new_pcm(struct snd_soc_pcm_runtime *rtd, int num) int i; if (rtd->dai_link->dynamic || rtd->dai_link->no_pcm) { - playback = rtd->dai_link->dpcm_playback; - capture = rtd->dai_link->dpcm_capture; + cpu_dai = asoc_rtd_to_cpu(rtd, 0); + if (rtd->num_cpus > 1) { + dev_err(rtd->dev, + "DPCM doesn't support Multi CPU yet\n"); + return -EINVAL; + } + + playback = rtd->dai_link->dpcm_playback && + snd_soc_dai_stream_valid(cpu_dai, SNDRV_PCM_STREAM_PLAYBACK); + capture = rtd->dai_link->dpcm_capture && + snd_soc_dai_stream_valid(cpu_dai, SNDRV_PCM_STREAM_CAPTURE); } else { /* Adapt stream for codec2codec links */ int cpu_capture = rtd->dai_link->params ?
At the moment, PCM devices for DPCM are only created based on the dpcm_playback/capture parameters of the DAI link, without considering if the CPU/FE DAI is actually capable of playback/capture. Normally the dpcm_playback/capture parameter should match the capabilities of the CPU DAI. However, there is no way to set that parameter from the device tree (e.g. with simple-audio-card or qcom sound cards). dpcm_playback/capture are always both set to 1. This causes problems when the CPU DAI does only support playback or capture. Attemting to open that PCM device with an unsupported stream type then results in a null pointer dereference: Unable to handle kernel NULL pointer dereference at virtual address 0000000000000128 Internal error: Oops: 96000044 [#1] PREEMPT SMP CPU: 3 PID: 1582 Comm: arecord Not tainted 5.7.0-rc1 pc : invalidate_paths_ep+0x30/0xe0 lr : snd_soc_dapm_dai_get_connected_widgets+0x170/0x1a8 Call trace: invalidate_paths_ep+0x30/0xe0 snd_soc_dapm_dai_get_connected_widgets+0x170/0x1a8 dpcm_path_get+0x38/0xd0 dpcm_fe_dai_open+0x70/0x920 snd_pcm_open_substream+0x564/0x840 snd_pcm_open+0xfc/0x228 snd_pcm_capture_open+0x4c/0x78 snd_open+0xac/0x1a8 ... ... because the DAI playback/capture_widget is not set in that case. We could add checks there to fix the problem (maybe we should anyway), but much easier is to not expose the device as playback/capture in the first place. Attemting to use that device would always fail later anyway. Add checks for snd_soc_dai_stream_valid() to the DPCM case to avoid exposing playback/capture if it is not supported. Signed-off-by: Stephan Gerhold <stephan@gerhold.net> --- This can be reproduced with q6asm-dai with a device tree configuration like: &q6asmdai { dai@0 { reg = <0>; direction = <Q6ASM_DAI_RX>; }; }; Then a simple "arecord output.wav" will result in the crash above, since it attempts to use that FE DAI for capture (even though it is limited to playback). --- sound/soc/soc-pcm.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-)