[alsa-devel] kernel oops related to the new HDA audio handling for DP MST

Wed Nov 27 20:46:36 CET 2019

Hi,

On 27-11-2019 20:40, Takashi Iwai wrote:
> On Wed, 27 Nov 2019 20:31:20 +0100,
> Hans de Goede wrote:
>>
>> Hi,
>>
>> On 27-11-2019 20:27, Takashi Iwai wrote:
>>> On Wed, 27 Nov 2019 20:24:49 +0100,
>>> Takashi Iwai wrote:
>>>>
>>>> On Wed, 27 Nov 2019 19:52:37 +0100,
>>>> Michał Matysiak wrote:
>>>>>
>>>>> On Wed, Nov 27, 2019 at 06:00:17PM +0100, Takashi Iwai wrote:
>>>>>> On Wed, 27 Nov 2019 16:57:04 +0100,
>>>>>> Takashi Iwai wrote:
>>>>>>>
>>>>>>> On Wed, 27 Nov 2019 16:46:00 +0100,
>>>>>>> Hans de Goede wrote:
>>>>>>>>
>>>>>>>> Hi Takashi,
>>>>>>>>
>>>>>>>> On 27-11-2019 15:39, Takashi Iwai wrote:
>>>>>>>>> On Wed, 27 Nov 2019 15:26:37 +0100,
>>>>>>>>> Michał Matysiak wrote:
>>>>>>>>>>
>>>>>>>>>> Hi
>>>>>>>>>>
>>>>>>>>>> Recently I've encountered this error and as Hans de Goede's request I'm
>>>>>>>>>> reporting this back to you. This happens while booting my laptop
>>>>>>>>>> connected to docking station and without using one.
>>>>>>>>>>
>>>>>>>>>> kernel: WARNING: CPU: 1 PID: 330 at sound/hda/hdac_component.c:290 snd_hdac_acomp_init+0xde/0x130 [snd_hda_core]
>>>>>>>>>> There are 2 more "cut here", but they're almost identical so I've only
>>>>>>>>>> included one in this email.
>>>>>>>>>>
>>>>>>>>>> Don't know what will be valuable to you, but I'm willing to help test
>>>>>>>>>> this and do what I'm told. So, how can I help?
>>>>>>>>>>
>>>>>>>>>> More info about this particular kernel and issue, that led to this is at:
>>>>>>>>>> https://bugzilla.redhat.com/show_bug.cgi?id=1757891
>>>>>>>>>>
>>>>>>>>>> dmesg output:
>>>>>>>>>>
>>>>>>>>>> Nov 26 18:05:45 kernel: microcode: microcode updated early to revision 0x2f, date = 2019-02-17
>>>>>>>>>> Nov 26 18:05:45 kernel: Linux version 5.4.0-0.rc8.git0.1.rhbz1757891.fc31.x86_64 (mockbuild at buildhw-10.phx2.fedoraproject.org) (gcc version 9.2.1 20190827 (Red Hat 9.2.1-1) (GCC)) #1 SMP Wed Nov 20 14:50:34 UTC 2019
>>>>>>>>>> Nov 26 18:05:45 kernel: Command line: BOOT_IMAGE=(hd0,gpt2)/vmlinuz-5.4.0-0.rc8.git0.1.rhbz1757891.fc31.x86_64 root=/dev/mapper/fedora-root ro resume=/dev/mapper/fedora-swap rd.lvm.lv=fedora/root rd.luks.uuid=luks-efd8b438-8f56-405a-8cea-88f83ca38d2b rd.lvm.lv=fedora/swap rhgb quiet
>>>>>>>>>> ...
>>>>>>>>>> ...
>>>>>>>>>> ...
>>>>>>>>>> Nov 26 18:06:16 kernel: ------------[ cut here ]------------
>>>>>>>>>> Nov 26 18:06:16 kernel: WARNING: CPU: 0 PID: 461 at sound/hda/hdac_component.c:290 snd_hdac_acomp_init+0xde/0x130 [snd_hda_core]
>>>>>>>>>
>>>>>>>>> This should have been already fixed by the recent commit
>>>>>>>>> 5a858e79c911330678b5a9be91a24830e94a0dc9
>>>>>>>>>        ALSA: hda - Disable audio component for legacy Nvidia HDMI codecs
>>>>>>>>> which is already included in Linus tree.  Please check it.
>>>>>>>>
>>>>>>>> Thanks, I've started a scratch kernel build with the relevant patches added,
>>>>>>>> for the Fedora users hitting this to test.
>>>>>>>>
>>>>>>>> The reason they started looking into their dmesg is that their nvidia GPU (hybrid gfx setup)
>>>>>>>> will no longer suspend with recent kernels, this is with a 5.4 kernel which already has the
>>>>>>>>
>>>>>>>> "ALSA: hda - Force runtime PM on Nvidia HDMI codecs"
>>>>>>>>
>>>>>>>> Fix and for good measure I've already given them a test kernel with the:
>>>>>>>> "ALSA: hda: Allow HDA to be runtime suspended when dGPU is not bound to a drive"
>>>>>>>>
>>>>>>>> patch added. But looking at the fix for the oops I'm not sure if fixing
>>>>>>>> the oops is also going to fix the issue with the dGPU no longer suspending?
>>>>>>>
>>>>>>> I guess it's irrelevant with that problem, as this kernel warning (not
>>>>>>> really an Oops) is just about skipping the multiple audio component
>>>>>>> registration.  And the audio component isn't used in nouveau side on
>>>>>>> 5.4.x at all, and it's just a placeholder.  But who knows the black
>>>>>>> magic behind the scene :)
>>>>>>
>>>>>> ... and if this still doesn't fix the problem, please check the
>>>>>> runtime PM state of all HD-audio codec devices and HD-audio controller
>>>>>> device.  Do all show the runtime-suspended but the power consumption
>>>>>> is high?  Or is some device blocked?
>>>>>>
>>>>>> Basically the audio controller corresponding to the dGPU should have
>>>>>> chip->bus.keep_power = 0, which allows the runtime PM.  This flag is
>>>>>> cleared at azx_vx_gpu_bound() only for the dGPU.  For the primary GPU,
>>>>>> we need to keep the link power unless the notification is done via the
>>>>>> audio component (like i915 or amdgpu).  I already submitted a patch to
>>>>>> enable the audio component for nouveau in the past, but it's ignored,
>>>>>> so far.
>>>>>>
>>>>>>
>>>>>> thanks,
>>>>>>
>>>>>> Takashi
>>>>>
>>>>> Hi Takashi
>>>>>
>>>>> On linux-next-20191127 warning indeed disappeared. Thanks!
>>>>>
>>>>> Rest of problems did not. This is my output from alsa-info.sh
>>>>> http://alsa-project.org/db/?f=91bb789a01f9eed92d0534fe8951619312b355da
>>>>> Don't know if it's helpful, so I'll leave it here. Power consumption is
>>>>> high, because runtime-suspended is not enabled/active (and cannot be
>>>>> forced) without removing nvidia's audio.
>>>>
>>>> What if you pass power_save=1 to snd-hda-intel option?
>>>>
>>>>     echo -n 1 > /sys/module/snd_hda_intel/parameters/power_save
>>>
>>> and the following, too:
>>>       echo -n 1 > /sys/module/snd_hda_intel/parameters/power_save_controller
>>>
>>> Otherwise the controller device will keep on.
>>
>> Thank you for helping us getting to the root of this.
>>
>> These both default to 1 resp. Y on Fedora kernels, so I do not think
>> that this will help, as they are both already 1/Y.
> 
> Well, the alsa-info.sh output Michal suggested shows differently:
>    http://alsa-project.org/db/?f=91bb789a01f9eed92d0534fe8951619312b355da
> 
>    !!Loaded sound module options
>    !!---------------------------
>    ....
>    power_save : 0
>    power_save_controller : N
> 
> If these values are true, maybe some desktop tuning modifies on the
> fly?

Oh, that might explain why the fixes for this are confirmed as working by
some of our users, but not by all.

Michal, are you perhaps using TLP or some similar software to tune /
improve power-management settings ?  Or do you perhaps have some
custom files under /etc/modprobe.conf.d messing with these settings?

Regards,

Hans