[alsa-devel] Backported sbxfi driver (UNTESTED!)

[William Pitcock]

On Sat, 2008-10-18 at 12:10 -0500, William Pitcock wrote:
> On Sat, 2008-10-18 at 17:06 +0200, Takashi Iwai wrote:
> > At Sat, 18 Oct 2008 10:06:20 +0000,
> > Xarteras wrote:
> > > 
> > > Takashi Iwai wrote:
> > > > At Fri, 17 Oct 2008 14:16:38 +0400,
> > > > The Source wrote:
> > > >> Takashi Iwai пишет:
> > > >>> At Fri, 17 Oct 2008 14:01:55 +0400,
> > > >>> The Source wrote:
> > > >>>   
> > > >>>> Takashi Iwai пишет:
> > > >>>>     
> > > >>>>> At Fri, 17 Oct 2008 13:58:20 +0400,
> > > >>>>> The Source wrote:
> > > >>>>>   
> > > >>>>>       
> > > >>>>>> Takashi Iwai пишет:
> > > >>>>>>     
> > > >>>>>>         
> > > >>>>>>> At Fri, 17 Oct 2008 11:57:08 +0400,
> > > >>>>>>> The Source wrote:
> > > >>>>>>>   
> > > >>>>>>>       
> > > >>>>>>>           
> > > >>>>>>>> Takashi Iwai пишет:
> > > >>>>>>>>     
> > > >>>>>>>>         
> > > >>>>>>>>             
> > > >>>>>>>>> At Thu, 16 Oct 2008 22:18:07 +0400,
> > > >>>>>>>>> The Source wrote:
> > > >>>>>>>>>   
> > > >>>>>>>>>       
> > > >>>>>>>>>           
> > > >>>>>>>>>               
> > > >>>>>>>>>>>> Ok. OpenAL with alsa also seem to cause problems.
> > > >>>>>>>>>>>>     
> > > >>>>>>>>>>>>         
> > > >>>>>>>>>>>>             
> > > >>>>>>>>>>>>                 
> > > >>>>>>>>>>>>                     
> > > >>>>>>>>>>> In both cases, check the period_size and buffer_size values (shown in
> > > >>>>>>>>>>> the kernel message, or /proc/asound/card0/pcm0p/sub0/hw_params).
> > > >>>>>>>>>>> And, try to aplay with these parameters, whether you get the similar
> > > >>>>>>>>>>> problem.
> > > >>>>>>>>>>>
> > > >>>>>>>>>>> 	% aplay -v --period-size=xxx --buffer-size=yyy foo.wav
> > > >>>>>>>>>>>
> > > >>>>>>>>>>>
> > > >>>>>>>>>>> Takashi
> > > >>>>>>>>>>>
> > > >>>>>>>>>>>   
> > > >>>>>>>>>>>       
> > > >>>>>>>>>>>           
> > > >>>>>>>>>>>               
> > > >>>>>>>>>>>                   
> > > >>>>>>>>>> I'm sorry, but any attemp to play file with ossplay results in complete 
> > > >>>>>>>>>> system hang with error:
> > > >>>>>>>>>> unable to handle NULL ponter dereference at address 
> > > >>>>>>>>>> 0000000000000008.....(hang, no more output).
> > > >>>>>>>>>> I tried many wav formats. So I can't get error log or period and buffer 
> > > >>>>>>>>>> sizes, sorry.
> > > >>>>>>>>>>     
> > > >>>>>>>>>>         
> > > >>>>>>>>>>             
> > > >>>>>>>>>>                 
> > > >>>>>>>>> Can anyone confirm to reproduce Oops with OSS apps (ossplay)?
> > > >>>>>>>>>
> > > >>>>>>>>> I'm wondering whether this has anything to do with the capture.
> > > >>>>>>>>> Can you record the sound, and change the capture mixer element properly?
> > > >>>>>>>>>
> > > >>>>>>>>>
> > > >>>>>>>>> thanks,
> > > >>>>>>>>>
> > > >>>>>>>>> Takashi
> > > >>>>>>>>>
> > > >>>>>>>>>   
> > > >>>>>>>>>       
> > > >>>>>>>>>           
> > > >>>>>>>>>               
> > > >>>>>>>> I checked mplayer. It uses period size 1024 instead of 4096 and 16384 
> > > >>>>>>>> buffer size (default). Sound is choppy (sound pauses is more frequent 
> > > >>>>>>>> when rate is lower).
> > > >>>>>>>> However an attempt to play the same file with the same period and buffer 
> > > >>>>>>>> sizes with aplay results in complete system hang.
> > > >>>>>>>>     
> > > >>>>>>>>         
> > > >>>>>>>>             
> > > >>>>>>> OK, that looks like a problem.  Looks like the timer resolution can be
> > > >>>>>>> short like that, or something racy in the timer handling.
> > > >>>>>>>
> > > >>>>>>> Can you check whether this happens with XXX_SYSTEM_TIMER, too?
> > > >>>>>>>
> > > >>>>>>> Or, does the patch below avoid the problem, at least?
> > > >>>>>>>
> > > >>>>>>>
> > > >>>>>>> thanks,
> > > >>>>>>>
> > > >>>>>>> Takashi
> > > >>>>>>>
> > > >>>>>>> diff --git a/sound/pci/sbxfi/sbxfi.c b/sound/pci/sbxfi/sbxfi.c
> > > >>>>>>> index 26a6cd3..5ceb228 100644
> > > >>>>>>> --- a/sound/pci/sbxfi/sbxfi.c
> > > >>>>>>> +++ b/sound/pci/sbxfi/sbxfi.c
> > > >>>>>>> @@ -277,6 +277,7 @@ static void sbxfi_rearm_timer(struct sbxfi *chip, int ticks)
> > > >>>>>>>  #else
> > > >>>>>>>  
> > > >>>>>>>  #define MAX_TICKS	((1 << 13) - 1)
> > > >>>>>>> +#define MIN_TICKS	1000		/* FIXME: really so? */
> > > >>>>>>>  
> > > >>>>>>>  static void sbxfi_init_timer(struct sbxfi *chip)
> > > >>>>>>>  {
> > > >>>>>>> @@ -287,6 +288,8 @@ static void sbxfi_set_timer(struct sbxfi *chip, int ticks)
> > > >>>>>>>  	LOG(2, "SET TIMER TICKS = %d\n", ticks);
> > > >>>>>>>  	if (ticks > MAX_TICKS)
> > > >>>>>>>  		ticks = MAX_TICKS;
> > > >>>>>>> +	else if (ticks < MIN_TICKS)
> > > >>>>>>> +		ticks = MIN_TICKS;
> > > >>>>>>>  	sbxfi_write(chip, TIMR, ticks | TIMR_IE | TIMR_IP);
> > > >>>>>>>  }
> > > >>>>>>>  static void sbxfi_stop_timer(struct sbxfi *chip)
> > > >>>>>>>
> > > >>>>>>>   
> > > >>>>>>>       
> > > >>>>>>>           
> > > >>>>>> After patch:
> > > >>>>>>
> > > >>>>>> Without system timer:
> > > >>>>>>
> > > >>>>>> aplay --period-size=1024 96000_16_Stereo.wav
> > > >>>>>> Plays fine
> > > >>>>>>
> > > >>>>>> aplay --period-size=1024 22050_16_Mono.wav
> > > >>>>>> BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
> > > >>>>>> Hang
> > > >>>>>>
> > > >>>>>> With system timer:
> > > >>>>>>
> > > >>>>>> aplay --period-size=1024 96000_16_Stereo.wav
> > > >>>>>> Superglitch. Each sample is played many times before advancing to next one.
> > > >>>>>>
> > > >>>>>> aplay --period-size=1024 22050_16_Mono.wav
> > > >>>>>> Instant reboot.
> > > >>>>>>     
> > > >>>>>>         
> > > >>>>> Just to be sure: you don't enable XXX_CONT_RATE, right?
> > > >>>>> It's known to be buggy, so disabled as default now.
> > > >>>>>
> > > >>>>>
> > > >>>>> Takashi
> > > >>>>>
> > > >>>>>   
> > > >>>>>       
> > > >>>> It is disabled for me too.
> > > >>>>     
> > > >>> And the patch didn't help?
> > > >>>
> > > >>>
> > > >>> Takashi
> > > >>>
> > > >>>   
> > > >> Only 96000Hz Stereo without system timer plays fine after patch.
> > > > 
> > > > Please elaborate?  You mean 22.5k still doesn't work with the patch?
> > > > Does 22.5kHz ever work with other parameters?  It'd be really helpful
> > > > if you get the full Oops log...
> > > > 
> > > > The patch affects only the emu20k1 timer.  The system timer stuff
> > > > isn't touched by it.
> > > > 
> > > > The reboot implies that it's unlikely the timer but the driver wrote
> > > > something wrong (unless you have the set up that the kernel traps and
> > > > automatically reboot).  But, it's nothing but a wild guess at this
> > > > point.
> > > > 
> > > > Anyway, I updated the driver code a bit again.  Please check the
> > > > latest one.
> > > 
> > > The latest change in the timer code really broke playback for me.
> > > Sound now stutters every 2 seconds.
> > > It seems to go back to normal when I put back the TIMR_IP flag that was 
> > > commented out in that patch.
> > 
> > Oh, thanks for noticing it.  I took the bit back now.
> > Looks like the bit is different from what I expected...
> 
> Works in mpg123 (use -r 96000) again, breaks in Audacious and mplayer...
> likely due to their non-blocking tickless model. Trying with various
> period_time and buffer_size settings does not help.

Figured it out. The hardware does not like period sizes larger than
4096. So we should manage a ringbuffer in the driver.

William
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://mailman.alsa-project.org/pipermail/alsa-devel/attachments/20081018/1f255c11/attachment.sig