Message ID | s5ha8afmbpu.wl-tiwai@suse.de (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On 01/25/2017 03:54 PM, Takashi Iwai wrote: > On Wed, 25 Jan 2017 13:28:11 +0100, > Vlastimil Babka wrote: >> >> Hi, >> >> my desktop randomly experiences workqueue lockups on boot with >> openSUSE Tumbleweed kernels 4.9.x, installed around >> Christmas. Previously I had a (badly maintained) Gentoo installation >> with 4.4 IIRC, so I can't say if the kernel has regressed, or the >> major userspace changes exposed different timing of stuff. > > If the lockup can be reproduced easily, could you check whether the > old kernel shows the issue? I don't remember of any big changes in > ca0132 driver in 4.x kernels. It'd be helpful even just checking > an openSUSE Leap 42.1 or 42.2 kernel. > >> This is how the workqueue lockup looks like: > (snip) >> kernel: [<ffffffffc0c20501>] dspio_read+0x51/0x70 [snd_hda_codec_ca0132] >> kernel: [<ffffffffc0c20566>] ca0132_process_dsp_response+0x46/0x160 >> [snd_hda_codec_ca0132] >> kernel: [<ffffffffc0c02fe5>] call_jack_callback.isra.1+0x25/0xa0 [snd_hda_codec] >> kernel: [<ffffffffc0c033c6>] snd_hda_jack_unsol_event+0x66/0x80 [snd_hda_codec] >> kernel: [<ffffffffc0bfd077>] hda_codec_unsol_event+0x17/0x20 [snd_hda_codec] >> kernel: [<ffffffffc0b86193>] process_unsol_events+0x63/0x70 [snd_hda_core] > > This is the code path that runs when the codec chip (CA0132) receives > an unsolicited event with a specific tag (0x16). It means the DSP > communication going. Oh, so it is actually the unused Creative card after all. Wonder what "jack" event it processes, since no jack is plugged in... > Possibly the bug is due to the recursive runtime PM handling. Could > you check the patch below? Hmm, so the issue didn't happen when rebooting with this patch on top of current kernel-source stable branch (i.e. 4.9.5). But then I did a full poweroff by mistake, and now I can't reproduce it even with the original kernel. Before the poweroff it persisted over each reboot today, so perhaps the card was in some specific state and now it's not... Might be also related to dual boot with Win10 and whatever its driver does to it and it persists over reboot? I'll keep using the nonpatched kernel until I hit the problem again and then try to test the patched kernel more times. Thanks so far! Vlastimil > > thanks, > > Takashi > > --- > diff --git a/sound/pci/hda/patch_ca0132.c b/sound/pci/hda/patch_ca0132.c > --- a/sound/pci/hda/patch_ca0132.c > +++ b/sound/pci/hda/patch_ca0132.c > @@ -4417,12 +4417,14 @@ static void ca0132_process_dsp_response(struct hda_codec *codec, > struct ca0132_spec *spec = codec->spec; > > codec_dbg(codec, "ca0132_process_dsp_response\n"); > + snd_hda_power_up_pm(codec); > if (spec->wait_scp) { > if (dspio_get_response_data(codec) >= 0) > spec->wait_scp = 0; > } > > dspio_clear_response_queue(codec); > + snd_hda_power_down_pm(codec); > } > > static void hp_callback(struct hda_codec *codec, struct hda_jack_callback *cb) >
On Wed, 25 Jan 2017 18:03:38 +0100, Vlastimil Babka wrote: > > On 01/25/2017 03:54 PM, Takashi Iwai wrote: > > On Wed, 25 Jan 2017 13:28:11 +0100, > > Vlastimil Babka wrote: > >> > >> Hi, > >> > >> my desktop randomly experiences workqueue lockups on boot with > >> openSUSE Tumbleweed kernels 4.9.x, installed around > >> Christmas. Previously I had a (badly maintained) Gentoo installation > >> with 4.4 IIRC, so I can't say if the kernel has regressed, or the > >> major userspace changes exposed different timing of stuff. > > > > If the lockup can be reproduced easily, could you check whether the > > old kernel shows the issue? I don't remember of any big changes in > > ca0132 driver in 4.x kernels. It'd be helpful even just checking > > an openSUSE Leap 42.1 or 42.2 kernel. > > > >> This is how the workqueue lockup looks like: > > (snip) > >> kernel: [<ffffffffc0c20501>] dspio_read+0x51/0x70 [snd_hda_codec_ca0132] > >> kernel: [<ffffffffc0c20566>] ca0132_process_dsp_response+0x46/0x160 > >> [snd_hda_codec_ca0132] > >> kernel: [<ffffffffc0c02fe5>] call_jack_callback.isra.1+0x25/0xa0 [snd_hda_codec] > >> kernel: [<ffffffffc0c033c6>] snd_hda_jack_unsol_event+0x66/0x80 [snd_hda_codec] > >> kernel: [<ffffffffc0bfd077>] hda_codec_unsol_event+0x17/0x20 [snd_hda_codec] > >> kernel: [<ffffffffc0b86193>] process_unsol_events+0x63/0x70 [snd_hda_core] > > > > This is the code path that runs when the codec chip (CA0132) receives > > an unsolicited event with a specific tag (0x16). It means the DSP > > communication going. > > Oh, so it is actually the unused Creative card after all. Wonder what > "jack" event it processes, since no jack is plugged in... > > > Possibly the bug is due to the recursive runtime PM handling. Could > > you check the patch below? > > Hmm, so the issue didn't happen when rebooting with this patch on top > of current kernel-source stable branch (i.e. 4.9.5). But then I did a > full poweroff by mistake, and now I can't reproduce it even with the > original kernel. Before the poweroff it persisted over each reboot > today, so perhaps the card was in some specific state and now it's > not... Might be also related to dual boot with Win10 and whatever its > driver does to it and it persists over reboot? I'll keep using the > nonpatched kernel until I hit the problem again and then try to test > the patched kernel more times. Thanks so far! The code path is related with the runtime PM, so it's likely depending on the device state, e.g. long-time pause or such. I don't think Win 10 plays a role, but who knows. In anyway, let me know if this helps. Basically I can merge it even for now, as the fix shouldn't give a regression. But of course it'd be better to have a test result :) thanks, Takashi
On 01/25/2017 06:06 PM, Takashi Iwai wrote: > The code path is related with the runtime PM, so it's likely depending > on the device state, e.g. long-time pause or such. I don't think Win > 10 plays a role, but who knows. > > In anyway, let me know if this helps. Basically I can merge it even > for now, as the fix shouldn't give a regression. But of course it'd > be better to have a test result :) OK so unfortunately it now happened with the patch.
On Mon, 30 Jan 2017 16:02:38 +0100, Vlastimil Babka wrote: > > On 01/25/2017 06:06 PM, Takashi Iwai wrote: > > The code path is related with the runtime PM, so it's likely depending > > on the device state, e.g. long-time pause or such. I don't think Win > > 10 plays a role, but who knows. > > > > In anyway, let me know if this helps. Basically I can merge it even > > for now, as the fix shouldn't give a regression. But of course it'd > > be better to have a test result :) > > OK so unfortunately it now happened with the patch. Hm, do you get the very same stack trace? It's strange because azx_rirb_get_response() has the timeout of 1 second in its loop, so it can't wait forever. Takashi
diff --git a/sound/pci/hda/patch_ca0132.c b/sound/pci/hda/patch_ca0132.c --- a/sound/pci/hda/patch_ca0132.c +++ b/sound/pci/hda/patch_ca0132.c @@ -4417,12 +4417,14 @@ static void ca0132_process_dsp_response(struct hda_codec *codec, struct ca0132_spec *spec = codec->spec; codec_dbg(codec, "ca0132_process_dsp_response\n"); + snd_hda_power_up_pm(codec); if (spec->wait_scp) { if (dspio_get_response_data(codec) >= 0) spec->wait_scp = 0; } dspio_clear_response_queue(codec); + snd_hda_power_down_pm(codec); } static void hp_callback(struct hda_codec *codec, struct hda_jack_callback *cb)