2023-06-25 16:44:49

by YE Chengfeng

Subject: [PATCH] ALSA: aloop: Fix &cable->lock deadlock issues

The timer loopback_jiffies_timer_function is executed under
bottom-half softirq context and require a spinlock, thus
other process context code requiring the same lock (i.e.,
loopback_trigger, loopback_pointer) can deadlock with the
timer if it is preempted while holding the lock.

Deadlock scenario:
loopback_trigger
-> spin_lock(&cable->lock);
<timer interrupt>
-> loopback_jiffies_timer_function
-> spin_lock_irqsave(&dpcm->cable->lock, flags);

Fix the potential deadlock by using spin_lock_irqsave.

Signed-off-by: Chengfeng Ye <[email protected]>
---
sound/drivers/aloop.c | 22 ++++++++++++----------
1 file changed, 12 insertions(+), 10 deletions(-)

diff --git a/sound/drivers/aloop.c b/sound/drivers/aloop.c
index a38e602b4fc6..8ee93f8581b4 100644
--- a/sound/drivers/aloop.c
+++ b/sound/drivers/aloop.c
@@ -379,6 +379,7 @@ static int loopback_trigger(struct snd_pcm_substream *substream, int cmd)
struct snd_pcm_runtime *runtime = substream->runtime;
struct loopback_pcm *dpcm = runtime->private_data;
struct loopback_cable *cable = dpcm->cable;
+ unsigned long flags;
int err = 0, stream = 1 << substream->stream;

switch (cmd) {
@@ -389,39 +390,39 @@ static int loopback_trigger(struct snd_pcm_substream *substream, int cmd)
dpcm->last_jiffies = jiffies;
dpcm->pcm_rate_shift = 0;
dpcm->last_drift = 0;
- spin_lock(&cable->lock);
+ spin_lock_irqsave(&cable->lock, flags);
cable->running |= stream;
cable->pause &= ~stream;
err = cable->ops->start(dpcm);
- spin_unlock(&cable->lock);
+ spin_unlock_irqrestore(&cable->lock, flags);
if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK)
loopback_active_notify(dpcm);
break;
case SNDRV_PCM_TRIGGER_STOP:
- spin_lock(&cable->lock);
+ spin_lock_irqsave(&cable->lock, flags);
cable->running &= ~stream;
cable->pause &= ~stream;
err = cable->ops->stop(dpcm);
- spin_unlock(&cable->lock);
+ spin_unlock_irqrestore(&cable->lock, flags);
if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK)
loopback_active_notify(dpcm);
break;
case SNDRV_PCM_TRIGGER_PAUSE_PUSH:
case SNDRV_PCM_TRIGGER_SUSPEND:
- spin_lock(&cable->lock);
+ spin_lock_irqsave(&cable->lock, flags);
cable->pause |= stream;
err = cable->ops->stop(dpcm);
- spin_unlock(&cable->lock);
+ spin_unlock_irqrestore(&cable->lock, flags);
if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK)
loopback_active_notify(dpcm);
break;
case SNDRV_PCM_TRIGGER_PAUSE_RELEASE:
case SNDRV_PCM_TRIGGER_RESUME:
- spin_lock(&cable->lock);
+ spin_lock_irqsave(&cable->lock, flags);
dpcm->last_jiffies = jiffies;
cable->pause &= ~stream;
err = cable->ops->start(dpcm);
- spin_unlock(&cable->lock);
+ spin_unlock_irqrestore(&cable->lock, flags);
if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK)
loopback_active_notify(dpcm);
break;
@@ -865,12 +866,13 @@ static snd_pcm_uframes_t loopback_pointer(struct snd_pcm_substream *substream)
struct snd_pcm_runtime *runtime = substream->runtime;
struct loopback_pcm *dpcm = runtime->private_data;
snd_pcm_uframes_t pos;
+ unsigned long flags;

- spin_lock(&dpcm->cable->lock);
+ spin_lock_irqsave(&dpcm->cable->lock, flags);
if (dpcm->cable->ops->pos_update)
dpcm->cable->ops->pos_update(dpcm->cable);
pos = dpcm->buf_pos;
- spin_unlock(&dpcm->cable->lock);
+ spin_unlock_irqrestore(&dpcm->cable->lock, flags);
return bytes_to_frames(runtime, pos);
}

--
2.17.1

2023-06-25 18:17:40

by Takashi Iwai

Subject: Re: [PATCH] ALSA: aloop: Fix &cable->lock deadlock issues

On Sun, 25 Jun 2023 18:22:56 +0200,
YE Chengfeng wrote:
>
> The timer loopback_jiffies_timer_function is executed under
> bottom-half softirq context and require a spinlock, thus
> other process context code requiring the same lock (i.e.,
> loopback_trigger, loopback_pointer) can deadlock with the
> timer if it is preempted while holding the lock.
>
> Deadlock scenario:
> loopback_trigger
> -> spin_lock(&cable->lock);
> <timer interrupt>
> -> loopback_jiffies_timer_function
> -> spin_lock_irqsave(&dpcm->cable->lock, flags);
>
> Fix the potential deadlock by using spin_lock_irqsave.

Similarly like the patch for snd-dummy, this change looks superfluous,
too.


thanks,

Takashi