Skip to contentSkip to conversation

Crash in the cs46xx driver when using alsaplayer

BUG ? Benny Sjostrand 2 participants 3 messages Aug 12, 2002

lore
Share & tools
lore
#0 lore raw

Hi!

Using the alsaplayer with the cs46xx driver causes a oops when closing the alsaplayer, playing mp3, wav, CD's etc with the alsaplayer works fine. All other applications that i've tested this dont happens OSS stuff like XMMS, aplay, acrecord, jackit.

Soiling all over the code with a lot of snd_printdd finally i've discovered the cause of the crash, and it's not cause of my patches to the cs46xx driver, it will probably happen even without CONFIG_SND_CS46XX_NEW_DSP defined.

So, when alsaplayer is closing the snd_cs46xx_playback_hw_free(...) is invoked, and after that the for some reason the snd_cs46xx_playback_copy(...) is invoked, and then runtime->dma_area == NULL which causes a kernel "oops" when copy_from_user(...) is invoked from snd_cs46xx_playback_copy(...)

Let's take a look at snd_cs46xx_playback_hw_free(...):

static int snd_cs46xx_playback_hw_free(snd_pcm_substream_t * substream) {
/*cs46xx_t *chip = snd_pcm_substream_chip(substream);*/ snd_pcm_runtime_t *runtime = substream->runtime; cs46xx_pcm_t *cpcm;

cpcm = snd_magic_cast(cs46xx_pcm_t, runtime->private_data, return
-ENXIO);

   if (runtime->dma_area != cpcm->hw_area)
       snd_pcm_lib_free_pages(substream);
  
   runtime->dma_area = NULL;
   runtime->dma_addr = 0;
   runtime->dma_bytes = 0;

return 0;
}

and then snd_cs46xx_playback_copy(...):

static int snd_cs46xx_playback_copy(snd_pcm_substream_t *substream,
int channel,
snd_pcm_uframes_t hwoff,
void *src,
snd_pcm_uframes_t frames)
{
snd_pcm_runtime_t *runtime = substream->runtime; /*cs46xx_t *chip = snd_pcm_substream_chip(substream); */ size_t hwoffb;
size_t bytes;
char *hwbuf;
cs46xx_pcm_t *cpcm = snd_magic_cast(cs46xx_pcm_t,
substream->runtime->private_data, return -ENXIO);

   hwoffb = hwoff << cpcm->shift;
   bytes = frames << cpcm->shift;
   hwbuf = runtime->dma_area + hwoffb;

if (copy_from_user(hwbuf, src, bytes)) /* It crash here when
runtime->dma_area == NULL */

       return -EFAULT;

spin_lock_irq(&runtime->lock); snd_cs46xx_playback_transfer(substream, frames); spin_unlock_irq(&runtime->lock);
return 0;
}

Where is BUG ?, or in cs46xx driver or in the ALSA PCM core somewhere ?, well, it's easy fixed in snd_cs46xx_playback_copy(...) doing a check. Then why it only happen with the alsaplayer, just no idea ....

suggestions ... ?? comments .... ??

/Benny

-------------------------------------------------------
This sf.net email is sponsored by: Dice - The leading online job board for high-tech professionals. Search and apply for tech jobs today!
http://seeker.dice.com/seeker.epl?rel_code=31

#1 lore raw
Quoted 69 lines · Benny Sjostrand

Hi!

Using the alsaplayer with the cs46xx driver causes a oops when closing the alsaplayer, playing mp3, wav, CD's etc with the alsaplayer works fine. All other applications that i've tested this dont happens OSS stuff like XMMS, aplay, acrecord, jackit.

Soiling all over the code with a lot of snd_printdd finally i've discovered the cause of the crash, and it's not cause of my patches to the cs46xx driver, it will probably happen even without CONFIG_SND_CS46XX_NEW_DSP defined.

So, when alsaplayer is closing the snd_cs46xx_playback_hw_free(...) is invoked, and after that the for some reason the snd_cs46xx_playback_copy(...) is invoked, and then runtime->dma_area == NULL which causes a kernel "oops" when copy_from_user(...) is invoked from snd_cs46xx_playback_copy(...)

Let's take a look at snd_cs46xx_playback_hw_free(...):

static int snd_cs46xx_playback_hw_free(snd_pcm_substream_t * substream)

{
    /*cs46xx_t *chip = snd_pcm_substream_chip(substream);*/
    snd_pcm_runtime_t *runtime = substream->runtime;
    cs46xx_pcm_t *cpcm;

    cpcm = snd_magic_cast(cs46xx_pcm_t, runtime->private_data, return 

-ENXIO);

    if (runtime->dma_area != cpcm->hw_area)
        snd_pcm_lib_free_pages(substream);
   
    runtime->dma_area = NULL;
    runtime->dma_addr = 0;
    runtime->dma_bytes = 0;

    return 0;
}

and then snd_cs46xx_playback_copy(...):

static int snd_cs46xx_playback_copy(snd_pcm_substream_t *substream,
int channel,
snd_pcm_uframes_t hwoff,
void *src,
snd_pcm_uframes_t frames)

{
    snd_pcm_runtime_t *runtime = substream->runtime;
    /*cs46xx_t *chip = snd_pcm_substream_chip(substream); */
    size_t hwoffb;
    size_t bytes;
    char *hwbuf;
    cs46xx_pcm_t *cpcm = snd_magic_cast(cs46xx_pcm_t, 

substream->runtime->private_data, return -ENXIO);

    hwoffb = hwoff << cpcm->shift;
    bytes = frames << cpcm->shift;
    hwbuf = runtime->dma_area + hwoffb;

    if (copy_from_user(hwbuf, src, bytes)) /* It crash here when 
runtime->dma_area == NULL */
        return -EFAULT;

    spin_lock_irq(&runtime->lock);
    snd_cs46xx_playback_transfer(substream, frames);
    spin_unlock_irq(&runtime->lock);
    return 0;
}
Quoted 5 lines

Where is BUG ?, or in cs46xx driver or in the ALSA PCM core somewhere ?, well, it's easy fixed in snd_cs46xx_playback_copy(...) doing a check. Then why it only happen with the alsaplayer, just no idea ....

suggestions ... ?? comments .... ??

playback_copy shouldn't be called after hw_free. I don't see any error in PCM core. The additional check would be dead code. It would be better to determine the real problem. Do you know the order of syscalls?

Jaroslav

-----
Jaroslav Kysela <[email protected]>
Linux Kernel Sound Maintainer
ALSA Project http://www.alsa-project.org
SuSE Linux http://www.suse.com

-------------------------------------------------------
This sf.net email is sponsored by: Dice - The leading online job board for high-tech professionals. Search and apply for tech jobs today!
http://seeker.dice.com/seeker.epl?rel_code=31

#2 lore raw
Quoted 11 lines
Quoted 8 lines

Where is BUG ?, or in cs46xx driver or in the ALSA PCM core somewhere ?, well, it's easy fixed in snd_cs46xx_playback_copy(...) doing a check. Then why it only happen with the alsaplayer, just no idea ....

suggestions ... ?? comments .... ??

playback_copy shouldn't be called after hw_free. I don't see any error in PCM core. The additional check would be dead code. It would be better to determine the real problem. Do you know the order of syscalls?

That's exactly my thought.
Adding a check in snd_cs46xx_playback_copy(...) works fine however it may not be the right solution to the problem.

It only happens to me with the "alsaplayer", so probably the alsaplayer is doing something special. The alsaplayer seems to have various threads:

10826 pts/1    S      0:00 alsaplayer -f 2048
10827 pts/1    S      0:00 alsaplayer -f 2048
10828 pts/1    S      0:00 alsaplayer -f 2048
10830 pts/1    S      0:00 alsaplayer -f 2048
10831 pts/1    S      0:02 alsaplayer -f 2048
10832 pts/1    S      0:00 alsaplayer -f 2048
10834 pts/1    RN     0:00 alsaplayer -f 2048
10835 pts/1    RN     0:00 alsaplayer -f 2048

Maybe the problem is some race condition, just a theory ....

Is there way to debug the PCM core finding out what happens ??

/Benny

-------------------------------------------------------
This sf.net email is sponsored by: Dice - The leading online job board for high-tech professionals. Search and apply for tech jobs today!
http://seeker.dice.com/seeker.epl?rel_code=31