Re: [PATCH] ASoC: soc-pcm: fix fe and be race when accessing substream->runtime - Pierre-Louis Bossart

From: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
To: Eugeniu Rosca <erosca@de.adit-jv.com>,
	Liam Girdwood <lgirdwood@gmail.com>,
	Mark Brown <broonie@kernel.org>, Jaroslav Kysela <perex@perex.cz>,
	Takashi Iwai <tiwai@suse.com>,
	alsa-devel@alsa-project.org, linux-kernel@vger.kernel.org
Cc: Yanmin Zhang <yanmin_zhang@linux.intel.com>,
	Eugeniu Rosca <roscaeugeniu@gmail.com>,
	Jiada Wang <jiada_wang@mentor.com>,
	Zhang Yanmin <yanmin.zhang@intel.com>,
	Ramesh Babu <ramesh.babu@intel.com>,
	Dean Jenkins <Dean_Jenkins@mentor.com>,
	Ramesh Babu B <ramesh.babu.b@intel.com>,
	xiao jin <jin.xiao@intel.com>,
	Cezary Rojewski <cezary.rojewski@intel.com>
Subject: Re: [PATCH] ASoC: soc-pcm: fix fe and be race when accessing substream->runtime
Date: Tue, 27 Sep 2022 09:51:46 +0200	[thread overview]
Message-ID: <bdf56f23-1019-b4c8-a11a-3934db432cf1@linux.intel.com> (raw)
In-Reply-To: <1664210154-11552-1-git-send-email-erosca@de.adit-jv.com>

On 9/26/22 18:35, Eugeniu Rosca wrote:
> From: xiao jin <jin.xiao@intel.com>
> 
> After start of fe and be, fe might go to close without triggering
> STOP, and substream->runtime is freed. However, be is still at
> START state and its substream->runtime still points to the
> freed runtime.
> 
> Later on, FE is opened/started again, and triggers STOP.
> snd_pcm_do_stop => dpcm_fe_dai_trigger
>                 => dpcm_fe_dai_do_trigger
>                 => dpcm_be_dai_trigger
>                 => dpcm_do_trigger
>                 => soc_pcm_trigger
>                 => skl_platform_pcm_trigger
> skl_platform_pcm_trigger accesses the freed old runtime data and
> kernel panic.
> 
> The patch fixes it by assigning be_substream->runtime in
> dpcm_be_dai_startup when be's state is START.

Can I ask on which kernel this patch was validated and on what platform?

We've done a lot of work since last year on DPCM states, and I wonder
the problem mentioned above actually exists on recent kernels.

Specifically, if the FE is closed, I don't get how the BE is not closed
as well. And if this problem is found on a recent kernel, then it should
be seen in the AVS driver as well, no?

> Signed-off-by: xiao jin <jin.xiao@intel.com>
> Signed-off-by: Zhang Yanmin <yanmin.zhang@intel.com>
> Signed-off-by: Eugeniu Rosca <erosca@de.adit-jv.com>
> ---
>  sound/soc/soc-pcm.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/sound/soc/soc-pcm.c b/sound/soc/soc-pcm.c
> index 4f60c0a83311..6ca1d02065ce 100644
> --- a/sound/soc/soc-pcm.c
> +++ b/sound/soc/soc-pcm.c
> @@ -1608,6 +1608,8 @@ int dpcm_be_dai_startup(struct snd_soc_pcm_runtime *fe, int stream)
>  		if (be->dpcm[stream].users++ != 0)
>  			continue;
>  
> +		be_substream->runtime = be->dpcm[stream].runtime;
> +
>  		if ((be->dpcm[stream].state != SND_SOC_DPCM_STATE_NEW) &&
>  		    (be->dpcm[stream].state != SND_SOC_DPCM_STATE_CLOSE))
>  			continue;
> @@ -1615,7 +1617,6 @@ int dpcm_be_dai_startup(struct snd_soc_pcm_runtime *fe, int stream)
>  		dev_dbg(be->dev, "ASoC: open %s BE %s\n",
>  			stream ? "capture" : "playback", be->dai_link->name);
>  
> -		be_substream->runtime = be->dpcm[stream].runtime;
>  		err = __soc_pcm_open(be, be_substream);
>  		if (err < 0) {
>  			be->dpcm[stream].users--;