: no SPDIF output with vt82xx/ICEnsemble VT1616i
A BUGNOTE has been added to this bug.
======================================================================
https://bugtrack.alsa-project.org/alsa-bug/bug_view_page.php?bug_id=0000573
======================================================================
Reported By: vladimir
Assigned To:
======================================================================
Project: ALSA - driver Bug ID: 573 Category: PCI - via82xx Reproducibility: always Severity: major Priority: normal Status: new Distribution: Fedora Core 3 test 2 Kernel Version: 2.6.8-1.541, as shipped with FC3t2 ====================================================================== Date Submitted: 10-09-2004 00:59 CEST Last Modified: 10-13-2004 11:16 CEST
======================================================================
Summary: no SPDIF output with vt82xx/ICEnsemble VT1616i
Description:
With alsa 1.0.6 (in kernel), 1.0.7rc1, or CVS, I'm unable to reliably get SPDIF output to work -- it might work once out of a dozen or so times. The attached patch improves things -- ignoring the cached values in ac97_read_cache gives me working spdif playbac. The values I see printed with the printk added in ac97_read_cache by the patch are:
AC97_read_cache: register 0x002a has real value 0x01f1 cached 0x05f1AC97_read_cache: register 0x002a has real value 0x05f1 cached 0x01f1
which indicates a change in the read-only SPCV bit. Perhaps registers with read-only status bits shouldn't be cached? I'm not sure if that's the fix, or if there's a deeper issue here.
======================================================================
----------------------------------------------------------------------
vladimir - 10-11-2004 08:21 CEST
----------------------------------------------------------------------
Note, even with the patch applied, I've had the audio cut out on occasion (spdif only, analog always continues to work); sometimes mid stream, but mostly at the end of a song with no sound for the beginning of the next -- this is using mplayer in shuffle mode. Flipping through songs with next/previous a few times usually restores audio. I'm not sure if this cutting out is a further driver issue, or a faulty hardware issue.
----------------------------------------------------------------------
tiwai - 10-11-2004 12:54 CEST
----------------------------------------------------------------------
Above all, we need to check _who_ modifies SPCV bit. I don't see any codes to change SPCV in ac97 related stuff...
----------------------------------------------------------------------
vladimir - 10-11-2004 19:46 CEST
----------------------------------------------------------------------
It shouldn't get modified; it's a read-only status bit. More curious is that there is no code that actually looks at the SPCV bit at all, except for something in patch_cm9739.. I'm not yet willing to rule out faulty hardware, though; I've been trying to find a kernel/alsa combo that does confirmed working spdif out with someone else's me6000 to test.
----------------------------------------------------------------------
tiwai - 10-12-2004 14:31 CEST
----------------------------------------------------------------------
Yes, that's why I wrote in the above. As far as I understand from your bug report, the SPCV bit of the cache value is changed. That is, some code really writes this bit.
Could you try the attached patch?
This will show the code address of the caller which changes SPCV. You can know the caller by comparing it with /proc/kallsyms.
----------------------------------------------------------------------
vladimir - 10-12-2004 19:44 CEST
----------------------------------------------------------------------
I don't seem to have /proc/kallsyms (red hat seems to have disabled it?!). However, this is the output that I get:
ac97: SPCV changed: 05c1 -> 0181: called from ce8e07a2AC97_read_cache: register 0x002a has real value 0x01c5 cached 0x05c5
Comparing that address with the base address in /proc/modules, I get an offset of 0x37a2 inside snd_ac97_codec -- the disassembly here looks like:
/* make sure the proper powerdown bits are cleared */
if (ac97->scaps) {
3763: 8b 44 24 34 mov 0x34(%esp),%eax
3767: 83 78 70 00 cmpl $0x0,0x70(%eax)
376b: 74 35 je 37a2
<snd_ac97_mixer+0x912>
reg = snd_ac97_read(ac97, AC97_EXTENDED_STATUS);
376d: ba 2a 00 00 00 mov $0x2a,%edx
3772: e8 fc ff ff ff call 3773
<snd_ac97_mixer+0x8e3>
if (ac97->scaps & AC97_SCAP_SURROUND_DAC)
3777: 8b 4c 24 34 mov 0x34(%esp),%ecx
377b: 8b 51 70 mov 0x70(%ecx),%edx
377e: 0f b7 c0 movzwl %ax,%eax
3781: f6 c2 04 test $0x4,%dl
3784: 74 03 je 3789
<snd_ac97_mixer+0x8f9>
reg &= ~AC97_EA_PRJ;
3786: 80 e4 ef and $0xef,%ah
if (ac97->scaps & AC97_SCAP_CENTER_LFE_DAC)
3789: 80 e2 08 and $0x8,%dl
378c: 74 03 je 3791
<snd_ac97_mixer+0x901>
reg &= ~(AC97_EA_PRI | AC97_EA_PRK);
378e: 80 e4 d7 and $0xd7,%ah
snd_ac97_write_cache(ac97, AC97_EXTENDED_STATUS, reg);
3791: 0f b7 c8 movzwl %ax,%ecx
3794: ba 2a 00 00 00 mov $0x2a,%edx
3799: 8b 44 24 34 mov 0x34(%esp),%eax
379d: e8 fc ff ff ff call 379e
<snd_ac97_mixer+0x90e>
}
snd_ac97_proc_init(ac97);
37a2: 8b 44 24 34 mov 0x34(%esp),%eax
37a6: e8 fc ff ff ff call 37a7
<snd_ac97_mixer+0x917>
if ((err = snd_device_new(card, SNDRV_DEV_LOWLEVEL, ac97, &ops)) <
0) {
37ab: 68 3c 01 00 00 push $0x13c
37b0: 8b 4c 24 38 mov 0x38(%esp),%ecx
37b4: 8b 44 24 34 mov 0x34(%esp),%eax
37b8: ba 00 20 00 00 mov $0x2000,%edx
37bd: e8 fc ff ff ff call 37be
<snd_ac97_mixer+0x92e>
37c2: 5d pop %ebp
37c3: 89 c3 mov %eax,%ebx
37c5: 85 c0 test %eax,%eax
37c7: 79 0d jns 37d6
<snd_ac97_mixer+0x946>
snd_ac97_free(ac97);This looks ok; ac97->regs[AC97_EXTENDED_STATUS] contains junk before this (the 0x181 value), and the correct value is written into the cache. (The order of numbers in your printk had me confused for a bit, I thought it was transitioning from 0x5c1 to 0x0181 :) The correct value ends up in both the cache and the register at this point (confirmed via a _read after the _write_cache, it's 0x05c1).
Other than that, the value doesn't seem to get changed directly; the hardware just updates that bit based on the other SPDIF configuration settings... but even then, I don't understand why having the incorrect SPCV value in the cached reg value is bad, since nothing seems to look at it; unless the hardware cares if it actually got read?
----------------------------------------------------------------------
tiwai - 10-13-2004 11:16 CEST
----------------------------------------------------------------------
A good cactch. At least, the attached patch fixes the cache mismatch in the original comments.
I don't expect that this will fix your problem, though...
Do you have any datasheet for this ac97 chip?
Bug History
Date Modified Username Field Change
======================================================================
10-09-04 00:59 vladimir New Bug 10-09-04 00:59 vladimir File Added: ac97_codec-no-readcache.patch 10-09-04 00:59 vladimir Distribution => Fedora Core 3 test 2 10-09-04 00:59 vladimir Kernel Version => 2.6.8-1.541, as shipped with FC3t2 10-11-04 08:21 vladimir Bugnote Added: 0002070 10-11-04 12:54 tiwai Bugnote Added: 0002076 10-11-04 19:46 vladimir Bugnote Added: 0002083 10-12-04 14:31 tiwai Bugnote Added: 0002092 10-12-04 14:31 tiwai File Added: ac97-spcv-debug.diff 10-12-04 19:44 vladimir Bugnote Added: 0002107 10-13-04 11:16 tiwai Bugnote Added: 0002118
======================================================================
-------------------------------------------------------
This SF.net email is sponsored by: IT Product Guide on ITManagersJournal Use IT products in your business? Tell us what you think of them. Give us Your Opinions, Get Free ThinkGeek Gift Certificates! Click to find out more
http://productguide.itmanagersjournal.com/guidepromo.tmpl