Re: [PATCH 4.4 20/87] ALSA: pcm: Allow aborting mutex lock at OSS read/write loops

[Takashi Iwai]

On Wed, 14 Feb 2018 17:20:23 +0100,
Ben Hutchings wrote:
> 
> On Mon, 2018-02-12 at 09:34 +0100, Takashi Iwai wrote:
> > On Wed, 24 Jan 2018 00:35:48 +0100,
> > Ben Hutchings wrote:
> > > 
> > > On Mon, 2018-01-15 at 13:34 +0100, Greg Kroah-Hartman wrote:
> > > > 4.4-stable review patch.ÂÂIf anyone has any objections, please let me know.
> > > > 
> > > > ------------------
> > > > 
> > > > From: Takashi Iwai <tiwai@xxxxxxx>
> > > > 
> > > > commit 900498a34a3ac9c611e9b425094c8106bdd7dc1c upstream.
> > > > 
> > > > PCM OSS read/write loops keep taking the mutex lock for the whole
> > > > read/write, and this might take very long when the exceptionally high
> > > > amount of data is given.ÂÂAlso, since it invokes with mutex_lock(),
> > > > the concurrent read/write becomes unbreakable.
> > > > 
> > > > This patch tries to address these issues by replacing mutex_lock()
> > > > with mutex_lock_interruptible(), and also splits / re-takes the lock
> > > > at each read/write period chunk, so that it can switch the context
> > > > more finely if requested.
> > > 
> > > [...]
> > > > @@ -1414,18 +1417,18 @@ static ssize_t snd_pcm_oss_write1(struct
> > > > Â			xfer += tmp;
> > > > Â			if ((substream->f_flags & O_NONBLOCK) != 0 &&
> > > > Â			ÂÂÂÂtmp != runtime->oss.period_bytes)
> > > > -				break;
> > > > +				tmp = -EAGAIN;
> > > > Â		}
> > > > + err:
> > > > +		mutex_unlock(&runtime->oss.params_lock);
> > > > +		if (tmp < 0)
> > > > +			break;
> > > > Â		if (signal_pending(current)) {
> > > > Â			tmp = -ERESTARTSYS;
> > > > -			goto err;
> > > > +			break;
> > > > Â		}
> > > > +		tmp = 0;
> > > > Â	}
> > > > -	mutex_unlock(&runtime->oss.params_lock);
> > > > -	return xfer;
> > > > -
> > > > - err:
> > > > -	mutex_unlock(&runtime->oss.params_lock);
> > > > Â	return xfer > 0 ? (snd_pcm_sframes_t)xfer : tmp;
> > > > Â}
> > > 
> > > [...]
> > > 
> > > Some of the "goto err" statements in the loop are conditional on tmp <=
> > > 0, but if tmp == 0 this will no longer terminate the loop.ÂÂIs that
> > > intentional or a bug?
> > 
> > The patch rather fixes the endless loop: the signal_pending() check is
> > added after goto err, so that it aborts the loop properly.
> 
> Let me rephrase then: if snd_pcm_oss_write2() returns 0, does that
> imply that signal_pending() is true?  If there is any other reason that
> it could return 0, then this appears to introduce a bug.

In some condition (depending on the plugin / conversion and partial
write) it may return zero, but practically seen it doesn't happen in
the whole this loop.


Takashi