Skip to contentSkip to conversation

Pb with simultaneous SATA and ALSA I/O

DISCUSSION Dominique Dumont 6 participants 21 messages Sep 25, 2006 – Oct 14, 2006

lore
Share & tools
lore
#0 lore raw

Hello

I have some problem using the SPDIF output of my SB Live card while performing I/O on my SATA drive. (See [1] for the whole story on the ALSA mailing list)

My Yamaha amplifier does not recognize the AC3 (Dolby digital) stream from the sdpif plug while performing I/O on my SATA drive.

If I have a lot of I/O (e.g. running md5sum on a 4Gb file), the AC3 stream is completely broken.

If I have some I/O (e.g. reading a Hi-def movie), I get some AC3 drop-out even if the CPU is about 50%.

I have the same result with DTS output.

With PCM output, I've noticed a hi-frequency distortion, which means that the interaction between SATA and snd module occurs several thousands time per second.

My set up is:
- Debian Linux kernel 2.6.17
- Sound blaster SB Live 5.1 (snd_emu10k1 module)
- SATA drive (sata_sil and libata module)
- A7n8x deluxe mobo
- AMD XP 3200

So far I verified that:
- AC3 output works fine when SATA drive is left alone
- AC3 output works fine when running md5sum on a PATA drive
- DTS output works fine on the mobo SPDIF output (snd_intel8x0 module) even when running md5sum on the SATA drive. (cannot try AC3 stream because of Soundstorm chip :-( )
- Preemp kernel option does not fix the problem
- when running md5sum on SATA drive, alsa driver report a starvation (xrun) every few seconds, not thousands of time per second.

Could someone shed some light on this problem ?

What can I do to help debug this problem ?

Thanks

[1] http://www.mail-archive.com/[email protected]/msg17399.html

#1 lore raw
Quoted 39 lines · Dominique Dumont

Hello

I have some problem using the SPDIF output of my SB Live card while performing I/O on my SATA drive. (See [1] for the whole story on the ALSA mailing list)

My Yamaha amplifier does not recognize the AC3 (Dolby digital) stream from the sdpif plug while performing I/O on my SATA drive.

If I have a lot of I/O (e.g. running md5sum on a 4Gb file), the AC3 stream is completely broken.

If I have some I/O (e.g. reading a Hi-def movie), I get some AC3 drop-out even if the CPU is about 50%.

I have the same result with DTS output.

With PCM output, I've noticed a hi-frequency distortion, which means that the interaction between SATA and snd module occurs several thousands time per second.

My set up is:
- Debian Linux kernel 2.6.17
- Sound blaster SB Live 5.1 (snd_emu10k1 module)
- SATA drive (sata_sil and libata module)
- A7n8x deluxe mobo
- AMD XP 3200

So far I verified that:
- AC3 output works fine when SATA drive is left alone
- AC3 output works fine when running md5sum on a PATA drive
- DTS output works fine on the mobo SPDIF output (snd_intel8x0 module) even when running md5sum on the SATA drive. (cannot try AC3 stream because of Soundstorm chip :-( )
- Preemp kernel option does not fix the problem
- when running md5sum on SATA drive, alsa driver report a starvation (xrun) every few seconds, not thousands of time per second.

Quoted 7 lines

Could someone shed some light on this problem ?

What can I do to help debug this problem ?

Thanks

Have you tried using a different slot for the SB Live?

Signature · 5 lines
Francesco Peeters
----
GPG Key = AA69 E7C6 1D8A F148 160C  D5C4 9943 6E38 D5E3 7704
If your program doesn't recognize my signature, please visit
http://www.CAcert.org/index.php?id=3 to retrieve the Root CA certificate.
#2 lore raw
Quoted 1 line · Francesco Peeters

Have you tried using a different slot for the SB Live?

Yes. No change at all. (Sorry for the delay).

Cheers

#3 lore raw
Quoted 5 lines · Dominique Dumont
Quoted 1 line · Francesco Peeters

Have you tried using a different slot for the SB Live?

Yes. No change at all. (Sorry for the delay).

This is going to be a problem with the SATA driver not ALSA. I've heard that some motherboards do evil stuff like implementing legacy drive access modes using SMM which would cause dropouts without xruns reported.

Please report it on LKML.

Lee

#4 lore raw
Quoted 4 lines · Lee Revell

This is going to be a problem with the SATA driver not ALSA. I've heard that some motherboards do evil stuff like implementing legacy drive access modes using SMM which would cause dropouts without xruns reported.

Why do I hear distortion with PCM on SPDIF output and no distortion at all on analog front output ? (both with the SB Live card)

Quoted 1 line

Please report it on LKML.

Will do. (although your mail and this mail are cc'ed to LKLM)

Thanks

#5 lore raw
Quoted 9 lines · Dominique Dumont
Quoted 4 lines · Lee Revell

This is going to be a problem with the SATA driver not ALSA. I've heard that some motherboards do evil stuff like implementing legacy drive access modes using SMM which would cause dropouts without xruns reported.

Why do I hear distortion with PCM on SPDIF output and no distortion at all on analog front output ? (both with the SB Live card)

No idea. Maybe SPDIF playback is more sensitive to timing glitches. But the problem must be in the SATA driver not the emu10k1 driver.

Lee

#6 lore raw

Ar Iau, 2006-10-05 am 17:12 -0400, ysgrifennodd Lee Revell:

Quoted 5 lines
Quoted 2 lines

Why do I hear distortion with PCM on SPDIF output and no distortion at all on analog front output ? (both with the SB Live card)

No idea. Maybe SPDIF playback is more sensitive to timing glitches. But the problem must be in the SATA driver not the emu10k1 driver.

Nothing to do with the SATA drivers, remember the S/PDIF and PCM bits come off the same codec so the bits hitting the codec are the same and thats way way outside the realm sata can affect.

If it's electrical noise though then power fluctuations or similar could be to blame ?

#7 lore raw

PCM distorsion snapshots from SATA/ALSA conflict (was: [Alsa-user] Pb with simultaneous SATA and ALSA I/O)

Quoted 2 lines · Alan Cox

If it's electrical noise though then power fluctuations or similar could be to blame ?

To check this, I've dusted off my old oscilloscope.

I've used speaker-test program to generate a 500, 900 and 1000 Hz sine wave that I've sent on the spdif line (as PCM) and had them decoded by my Yamaha DSP-A1 amplifier.

I've enclosed only the snapshots to this mail, I hope it'll get through vger's majordomo.

Anyway, the oscilloscope shows that:
- a spike occurs every 330 useconds (about 3kHz)
(note: 330us is 15.85 times the period of the 48KHz spdif stream)
- the spike level roughly matches the level of the sine waves 330 useconds sooner

I may be wrong, but it looks like that, somewhere in the audio data stream, a buffer is read downstream before the correct value is written by upstream (delayed by SATA I/O ??). OTOH, 16 samples is a very short buffer...

If needed I can try other things like:
- loop back the spdif output to my mobo's input to get the actual values of the PCM stream sent to the Yamaha.
- try the oscilloscope on the analog output of the SB live card ( I couldn't hear any distorsion on the analog line, but it does mean there no problem on this side as the D/A converter may filter out these kind of short duration spikes)
- try another signal shape

Any thought ?

HTH

#8 lore raw

Re: PCM distorsion snapshots from SATA/ALSA conflict (was: [Alsa-user] Pb with simultaneous SATA and ALSA I/O)

(it's probably a bad idea to attach .jpgs to a LKML post - please post them on the web and send a URL)

Quoted 6 lines · Dominique Dumont

Anyway, the oscilloscope shows that:
- a spike occurs every 330 useconds (about 3kHz)
(note: 330us is 15.85 times the period of the 48KHz spdif stream)
- the spike level roughly matches the level of the sine waves 330 useconds sooner

You seem to be using a period size of 256 samples - any change if you use a larger period size?

Any change if you use setpci to increase LATENCY_TIMER for the SBLive! card?

Lee

#9 lore raw

Re: [Alsa-user] PCM distorsion snapshots from SATA/ALSA conflict

Quoted 2 lines · Lee Revell

(it's probably a bad idea to attach .jpgs to a LKML post - please post them on the web and send a URL)

(ok. But is there a site that will guarantee the availability of the jpg attachments when people are searching the lklm or alsa archives ? )

Quoted 2 lines

You seem to be using a period size of 256 samples - any change if you use a larger period size?

Unless I'm wrong, you want me to test the PCM output while modifying the --period option of speaker-test.

So I've tested this command:
speaker-test -D iec958 -c 2 -f 1000 -p 4096 -t sine -s 2

No change: I still have a lot of spikes in the sine wave.

Quoted 2 lines

Any change if you use setpci to increase LATENCY_TIMER for the SBLive! card?

To be sure I've done the correct tests, here are the commands are used:

$ lspci
[...]
01:07.0 Multimedia audio controller: Creative Labs SB Live! EMU10k1 (rev 0a)
01:07.1 Input device controller: Creative Labs SB Live! Game Port (rev 0a)
[...]
01:0b.0 RAID bus controller: Silicon Image, Inc. SiI 3112 [SATALink/SATARaid] Serial ATA Controller (rev 02)
$ sudo setpci -s 01:07.0 latency_timer=80
$ sudo setpci -s 01:07.0 latency_timer=f8
$ sudo setpci -s 01:0b.0 latency_timer=10

No change at all :-(

Cheers

#10 lore raw
In reply to Dominique Dumont
Quoted 7 lines · Dominique Dumont
Quoted 4 lines · Lee Revell

This is going to be a problem with the SATA driver not ALSA. I've heard that some motherboards do evil stuff like implementing legacy drive access modes using SMM which would cause dropouts without xruns reported.

Quoted 6 lines

Why do I hear distortion with PCM on SPDIF output and no distortion at all on analog front output ? (both with the SB Live card)

Quoted 1 line

Please report it on LKML.

Will do. (although your mail and this mail are cc'ed to LKLM)

Did you ever try the latency tracer? (See LKML archives for instructions)

Lee

#11 lore raw
Quoted 2 lines · Lee Revell

Did you ever try the latency tracer? (See LKML archives for instructions)

Yes I did (this time better than 2 or 3 days ago :-/ ).

- I compiled and booted 2.6.28-rt5 with latency tracer enabled
- I verified that I got latency trace enabled (seen trace in kern.log)

I only got some traces after running this (detail added for the sake of other newbies like me) :

   gandalf:/proc/sys/kernel# echo 0 > preempt_max_latency

Here's the max latency trace I got (note that I got a similar trace *before* running the test, so I'd say it's unrelated to the AC3 drop-out problem.):

preemption latency trace v1.1.5 on 2.6.18-rt5
--------------------------------------------------------------------
latency: 19 us, #58/58, CPU#0 | (M:rt VP:0, KP:0, SP:1 HP:1)

    -----------------
    | task: posix_cpu_timer-2 (uid:0 nice:0 policy:1 rt_prio:99)
    -----------------

                 _------=> CPU#            
                / _-----=> irqs-off        
               | / _----=> need-resched    
               || / _---=> hardirq/softirq 
               ||| / _--=> preempt-depth   
               |||| /                      
               |||||     delay             
   cmd     pid ||||| time  |   caller      
      \   /    |||||   \   |   /           
  <idle>-0     0Dnh2    0us : __trace_start_sched_wakeup (try_to_wake_up)
  <idle>-0     0Dnh2    0us : __trace_start_sched_wakeup <<...>-2> (0 0)
  <idle>-0     0Dnh2    0us : try_to_wake_up <<...>-2> (199 20)
  <idle>-0     0Dnh1    0us : preempt_schedule (try_to_wake_up)
  <idle>-0     0Dnh1    0us : wake_up_process (run_posix_cpu_timers)
  <idle>-0     0Dnh1    0us : note_interrupt (handle_level_irq)
  <idle>-0     0Dnh2    0us : enable_8259A_irq (handle_level_irq)
  <idle>-0     0Dnh2    1us : preempt_schedule (enable_8259A_irq)
  <idle>-0     0Dnh1    1us : preempt_schedule (handle_level_irq)
  <idle>-0     0Dnh1    2us : irq_exit (do_IRQ)
  <idle>-0     0Dnh1    3us : do_IRQ (b0101ade b 0)
  <idle>-0     0Dnh1    3us : handle_level_irq (do_IRQ)
  <idle>-0     0Dnh2    3us+: mask_and_ack_8259A (handle_level_irq)
  <idle>-0     0Dnh2    8us : preempt_schedule (mask_and_ack_8259A)
  <idle>-0     0Dnh2    8us : redirect_hardirq (handle_level_irq)
  <idle>-0     0Dnh2    9us : wake_up_process (redirect_hardirq)
  <idle>-0     0Dnh2    9us : try_to_wake_up (wake_up_process)
  <idle>-0     0Dnh2    9us : task_rq_lock (try_to_wake_up)
  <idle>-0     0Dnh3    9us : activate_task (try_to_wake_up)
  <idle>-0     0Dnh3    9us : sched_clock (activate_task)
  <idle>-0     0Dnh3    9us : __activate_task (activate_task)
  <idle>-0     0Dnh3    9us : __activate_task <<...>-2006> (59 1)
  <idle>-0     0Dnh3    9us : enqueue_task (__activate_task)
  <idle>-0     0Dnh3    9us : __trace_start_sched_wakeup (try_to_wake_up)
  <idle>-0     0Dnh3    9us : try_to_wake_up <<...>-2006> (140 20)
  <idle>-0     0Dnh2    9us : preempt_schedule (try_to_wake_up)
  <idle>-0     0Dnh2    9us : wake_up_process (redirect_hardirq)
  <idle>-0     0Dnh1   10us : preempt_schedule (handle_level_irq)
  <idle>-0     0Dnh1   10us : irq_exit (do_IRQ)
  <idle>-0     0Dnh1   11us : do_IRQ (b0101ade e 0)
  <idle>-0     0Dnh1   11us : handle_level_irq (do_IRQ)
  <idle>-0     0Dnh2   11us+: mask_and_ack_8259A (handle_level_irq)
  <idle>-0     0Dnh2   16us : preempt_schedule (mask_and_ack_8259A)
  <idle>-0     0Dnh2   16us : redirect_hardirq (handle_level_irq)
  <idle>-0     0Dnh2   16us : wake_up_process (redirect_hardirq)
  <idle>-0     0Dnh2   16us : try_to_wake_up (wake_up_process)
  <idle>-0     0Dnh2   16us : task_rq_lock (try_to_wake_up)
  <idle>-0     0Dnh3   17us : activate_task (try_to_wake_up)
  <idle>-0     0Dnh3   17us : sched_clock (activate_task)
  <idle>-0     0Dnh3   17us : __activate_task (activate_task)
  <idle>-0     0Dnh3   17us : __activate_task <<...>-861> (53 2)
  <idle>-0     0Dnh3   17us : enqueue_task (__activate_task)
  <idle>-0     0Dnh3   17us : __trace_start_sched_wakeup (try_to_wake_up)
  <idle>-0     0Dnh3   17us : try_to_wake_up <<...>-861> (146 20)
  <idle>-0     0Dnh2   17us : preempt_schedule (try_to_wake_up)
  <idle>-0     0Dnh2   17us : wake_up_process (redirect_hardirq)
  <idle>-0     0Dnh1   17us : preempt_schedule (handle_level_irq)
  <idle>-0     0Dnh1   17us : irq_exit (do_IRQ)
  <idle>-0     0Dn..   17us : __schedule (cpu_idle)
  <idle>-0     0Dn..   17us : profile_hit (__schedule)
  <idle>-0     0Dn.1   18us : sched_clock (__schedule)
   <...>-2     0D..2   18us : __switch_to (__schedule)
   <...>-2     0D..2   18us : __schedule <<idle>-0> (20 199)
   <...>-2     0...1   18us : trace_stop_sched_switched (__schedule)
   <...>-2     0D..1   18us : trace_stop_sched_switched <<...>-2> (0 0)
   <...>-2     0D..2   18us : __schedule (__schedule)

vim:ft=help

Cheers

#12 lore raw
Quoted 10 lines · Dominique Dumont
Quoted 2 lines · Lee Revell

Did you ever try the latency tracer? (See LKML archives for instructions)

Yes I did (this time better than 2 or 3 days ago :-/ ).

- I compiled and booted 2.6.28-rt5 with latency tracer enabled
- I verified that I got latency trace enabled (seen trace in kern.log)

Quoted 8 lines

I only got some traces after running this (detail added for the sake of other newbies like me) :

   gandalf:/proc/sys/kernel# echo 0 > preempt_max_latency

Here's the max latency trace I got (note that I got a similar trace *before* running the test, so I'd say it's unrelated to the AC3 drop-out problem.):

I think it must be an electrical noise issue.

Lee

#13 lore raw
In reply to Lee Revell

Ar Iau, 2006-10-05 am 16:45 -0400, ysgrifennodd Lee Revell:

Quoted 3 lines

I've heard that some motherboards do evil stuff like implementing legacy drive access modes using SMM which would cause dropouts without xruns reported.

They don't. SATA causes audio dropouts on some systems because its fast enough to starve the audio device of regular enough access to the PCI bus. If that is a problem the audio device should be tuning PCI latencies

#14 lore raw
Quoted 10 lines · Alan Cox

Ar Iau, 2006-10-05 am 16:45 -0400, ysgrifennodd Lee Revell:

Quoted 3 lines

I've heard that some motherboards do evil stuff like implementing legacy drive access modes using SMM which would cause dropouts without xruns reported.

They don't. SATA causes audio dropouts on some systems because its fast enough to starve the audio device of regular enough access to the PCI bus. If that is a problem the audio device should be tuning PCI latencies

OK. In fact the Windows driver and IIRC the OSS driver do tune PCI latencies. But that can't be the problem here if analog playback is unaffected. Sounds like electrical noise could be the issue...

Lee

#15 lore raw
In reply to Dominique Dumont
Quoted 36 lines · Dominique Dumont

I have some problem using the SPDIF output of my SB Live card while performing I/O on my SATA drive. (See [1] for the whole story on the ALSA mailing list)

My Yamaha amplifier does not recognize the AC3 (Dolby digital) stream from the sdpif plug while performing I/O on my SATA drive.

If I have a lot of I/O (e.g. running md5sum on a 4Gb file), the AC3 stream is completely broken.

If I have some I/O (e.g. reading a Hi-def movie), I get some AC3 drop-out even if the CPU is about 50%.

I have the same result with DTS output.

With PCM output, I've noticed a hi-frequency distortion, which means that the interaction between SATA and snd module occurs several thousands time per second.

My set up is:
- Debian Linux kernel 2.6.17
- Sound blaster SB Live 5.1 (snd_emu10k1 module)
- SATA drive (sata_sil and libata module)
- A7n8x deluxe mobo
- AMD XP 3200

So far I verified that:
- AC3 output works fine when SATA drive is left alone
- AC3 output works fine when running md5sum on a PATA drive
- DTS output works fine on the mobo SPDIF output (snd_intel8x0 module) even when running md5sum on the SATA drive. (cannot try AC3 stream because of Soundstorm chip :-( )
- Preemp kernel option does not fix the problem
- when running md5sum on SATA drive, alsa driver report a starvation (xrun) every few seconds, not thousands of time per second.

Quoted 3 lines

Could someone shed some light on this problem ?

What can I do to help debug this problem ?

Well i agree with the suggestion of trying a different PCI slot for the sb live. There is so much onboard stuff sharing interrupts on those boards that you might have problems because of that. Creative cards are not very good at dealing with anything other than ideal conditions from what I have gathered over the years. The manual for the board will tell you which IRQ goes to which slot, and I guess you want to avoid using a slot that shares with the SATA controller.

--
Len Sorensen

#16 lore raw
Quoted 7 lines · Lennart Sorensen

Well i agree with the suggestion of trying a different PCI slot for the sb live. There is so much onboard stuff sharing interrupts on those boards that you might have problems because of that. Creative cards are not very good at dealing with anything other than ideal conditions from what I have gathered over the years. The manual for the board will tell you which IRQ goes to which slot, and I guess you want to avoid using a slot that shares with the SATA controller.

It might not be interrupt related, it could be DMA starvation. This has been observed with some SATA controllers while testing the -rt patches. The symptom is that the latency traces show the machine going in "slow motion".

Dominique: try the -rt kernel, enable latency tracing and post the output.

Lee

#17 lore raw
Quoted 9 lines · Lee Revell
Quoted 8 lines

Well i agree with the suggestion of trying a different PCI slot for the sb live. There is so much onboard stuff sharing interrupts on those boards that you might have problems because of that. Creative cards are not very good at dealing with anything other than ideal conditions from what I have gathered over the years. The manual for the board will tell you which IRQ goes to which slot, and I guess you want to avoid using a slot that shares with the SATA controller.

Quoted 4 lines

It might not be interrupt related, it could be DMA starvation. This has been observed with some SATA controllers while testing the -rt patches. The symptom is that the latency traces show the machine going in "slow motion".

with the -rt patch, high resolution timer + dyn helps too in going in deep slow mode.

#18 lore raw
In reply to Lee Revell
Quoted 2 lines · Lee Revell

Dominique: try the -rt kernel, enable latency tracing and post the output.

Should I use 2.6.17 + rt to stay closer to my current setup or should I get 2.6.18 + rt ?

Thanks

#19 lore raw
Quoted 7 lines · Dominique Dumont
Quoted 2 lines · Lee Revell

Dominique: try the -rt kernel, enable latency tracing and post the output.

Should I use 2.6.17 + rt to stay closer to my current setup or should I get 2.6.18 + rt ?

I would say 2.6.18 + -rt - it's not as if 2.6.17 could be fixed at this point.

Lee

#20 lore raw
In reply to Lee Revell
Quoted 7 lines · Lee Revell

It might not be interrupt related, it could be DMA starvation. This has been observed with some SATA controllers while testing the -rt patches. The symptom is that the latency traces show the machine going in "slow motion".

Dominique: try the -rt kernel, enable latency tracing and post the output.

Done. I've tried with 2.6.18-rt5. CONFIG_LATENCY_TRACE is enabled.

Here are the results (still with running "ac3dec -C " and "md5sum *" on a SATA drive):
- I get no more ALSA xrun.
- /proc/latency_trace is empty
- dolby digital output is still considerably chopped.

Note that the dolby digital output works fine when:
- No I/O is done
- heavy I/O on pata HDD (md5sum *)
- heavy I/O on DVD reader (md5sum *)

Did I miss something with the latency_trace ?

Cheers