| Bug ID | 106957 |
|---|---|
| Summary | GPU runtime suspend broken since 4.17 |
| Product | DRI |
| Version | unspecified |
| Hardware | x86-64 (AMD64) |
| OS | Linux (All) |
| Status | NEW |
| Severity | normal |
| Priority | medium |
| Component | DRM/Radeon |
| Assignee | dri-devel@lists.freedesktop.org |
| Reporter | prg@cooco.de |
| CC | lukas@wunner.de |
Created attachment 140212 =
[details]
dmesg from 4.17.2
I got a notebook with some Intel IGP and a Radeon HD5650. Since commit
07f4f97d7b4bf325d9f558c5b58230387e4e57e0 the dGPU is constantly DynPwr'd. T=
his
still happens in 4.17.2, so the patch mentioned in #106597 doesn't help.
Hm, what does the following show? cat /sys/bus/pci/devices/0000:02:00.0/power/control # GPU cat /sys/bus/pci/devices/0000:02:00.0/power/runtime_status # GPU cat /sys/bus/pci/devices/0000:02:00.0/power/runtime_usage # GPU cat /sys/bus/pci/devices/0000:02:00.0/power/runtime_active_kids # GPU cat /sys/bus/pci/devices/0000:02:00.1/power/control # HDA cat /sys/bus/pci/devices/0000:02:00.1/power/runtime_status # HDA cat /sys/bus/pci/devices/0000:02:00.1/power/runtime_usage # HDA cat /sys/bus/pci/devices/0000:02:00.1/power/runtime_active_kids # HDA I can't see anything in dmesg that the HDA controller is bound to a driver. What does the following show? ls -l /sys/bus/pci/devices/0000:02:00.1/driver
# cat /sys/bus/pci/devices/0000:02:00.0/power/control =
# GPU
auto
# cat /sys/bus/pci/devices/0000:02:00.0/power/runtime_status # GPU
active
# cat /sys/bus/pci/devices/0000:02:00.0/power/runtime_usage # GPU
cat: '/sys/bus/pci/devices/0000:02:00.0/power/runtime_usage': No such file =
or
directory
# cat /sys/bus/pci/devices/0000:02:00.0/power/runtime_active_kids # GPU
cat: '/sys/bus/pci/devices/0000:02:00.0/power/runtime_active_kids': No such
file or directory=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=
=20=20=20=20=20=20=20=20=20=20=20=20
# cat /sys/bus/pci/devices/0000:02:00.1/power/control # HDA=
=20=20=20=20=20=20=20
auto=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=
=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=
=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=
=20=20=20
# cat /sys/bus/pci/devices/0000:02:00.1/power/runtime_status # HDA
active
# cat /sys/bus/pci/devices/0000:02:00.1/power/runtime_usage # HDA
cat: '/sys/bus/pci/devices/0000:02:00.1/power/runtime_usage': No such file =
or
directory
# cat /sys/bus/pci/devices/0000:02:00.1/power/runtime_active_kids # HDA
cat: '/sys/bus/pci/devices/0000:02:00.1/power/runtime_active_kids': No such
file or directory
# ls -l /sys/bus/pci/devices/0000:02:00.1/driver
lrwxrwxrwx 1 root root 0 Jun 19 18:04 /sys/bus/pci/devices/0000:02:00.1/dri=
ver
-> ../../../../bus/pci/drivers/snd_hda_intel
Okay so the HDA controller is bound to a driver and is runtime= active.=20 Naturally, if it's runtime active it'll keep the GPU awake. Question is wh= at's keeping it active. Could you check if there are any user space processes accessing the HDA controller: sudo lsof /dev/snd/controlC1 You got "No such file or directory" for some of the commands I ga= ve you because the kernel isn't compiled with CONFIG_PM_ADVANCED_DEBUG=3Dy. Could you ena= ble that option and try again? Thanks!
# cat /sys/bus/pci/devices/0000:02:00.0/power/runtime_usage 1 # cat /sys/bus/pci/devices/0000:02:00.0/power/runtime_active_kids 0 # cat /sys/bus/pci/devices/0000:02:00.1/power/runtime_usage 0 # cat /sys/bus/pci/devices/0000:02:00.1/power/runtime_active_kids 0 # lsof /dev/snd/controlC1 No output. Yes, I did run this as root.
Okay the HDA's runtime ref counter is 0 and it has no active c= hildren, so it should suspend. Chances are it doesn't because rpm_idle() fails for some reason. Could you try this debug patch that I had created for #106597 and post the dmesg output? https://bugs.freedesktop.org/attachment.cgi?id=3D139706&acti= on=3Dedit You can add "log_buf_len=3D10M ignore_loglevel" to the command li= ne to ensure that dmesg isn't truncated and contains all debug output.
Created attachment 140235=
[details]
dmesg with debug patch
I'm having a d=C3=A9j=C3=A0 vu here :-) [ 7.840122] snd_hda_intel 0000:02:00.1: azx_runtime_idle: !power_save_controller =3D 0, !azx_has_pm_runtime(chip) =3D 0, azx_bus(chip)->codec_powered =3D 0x1, !chip->running =3D 0 The single codec on this HDA controller is considered powered on, hence the= HDA controller refuses to runtime suspend. Same problem as in #106597. What does the following show: grep . /sys/bus/hdaudio/devices/hdaudioC1D0/widgets/*/power_caps And could you try this debug patch (in lieu of the other one) to narrow down the root cause further: https://bugs.freedesktop.org/attachment.cgi?id=3D139735&acti= on=3Dedit Thanks!
Created attachment=
140242 [details]
dmesg with second debug patch
# grep . /sys/bus/hdaudio/devices/hdaudioC1D0/widgets/*/power_caps
/sys/bus/hdaudio/devices/hdaudioC1D0/widgets/01/power_caps:0x00000009
| What | Removed | Added |
|---|---|---|
| Resolution | --- | FIXED |
| Status | NEW | RESOLVED |
Fixed with commit 57cb54e53bdd ("ALSA: hda - Force to lin=
k down at runtime
suspend on ATI/AMD HDMI") which is now queued for 4.18-rc2 and marked =
for
stable, it will probably appear in 4.17.3:
https://git.k=
ernel.org/tiwai/sound/c/57cb54e53bdd
It looks like no pull request was sent out for the sound subsy=
stem this week,
so I'm afraid the fix will not appear in mainline earlier than 4.18-rc3.
The fix landed in Linus' tree yesterday: https://git.kernel.or= g/linus/57cb54e53bdd