Suspend regression on 3.7-rc with nouveau

[Herton Ronaldo Krzesinski]

Hi,

since 3.7-rc1 I'm unable to suspend my "Optimus" based laptop, suspend
aborts, this is the relevant log snippet with debugging enabled up to
tracing in nouveau:

[   65.941744] nouveau  [     DRM] suspending fbcon...
[   65.941746] nouveau  [     DRM] suspending display...
[   65.942108] nouveau  [     DRM] unpinning framebuffer(s)...
[   65.942176] nouveau  [     DRM] evicting buffers...
[   65.942749] ACPI handle has no context!
[   65.942941] nouveau  [     DRM] suspending client object trees...
[   65.942942] nouveau D[    1281] suspend running
[   65.942944] nouveau T[    1281] 0xffffffff:0xffffffff suspend children
[   65.942946] nouveau T[    1281] 0xffffffff:0xffffffff suspend running
[   65.942947] nouveau T[    1281] use(-1) == 1
[   65.942949] nouveau T[    1281] 0xffffffff:0xffffffff suspend completed
[   65.942949] nouveau D[    1281] suspend completed with 0
[   65.942950] nouveau D[     DRM] suspend running
[   65.942952] nouveau T[     DRM] 0xffffffff:0xffffffff suspend children
[   65.942953] nouveau T[     DRM] 0xffffffff:0xdddddddd suspend children
[   65.942954] nouveau T[     DRM] 0xdddddddd:0xcccc0000 suspend children
[   65.942956] nouveau T[     DRM] 0xcccc0000:0x000490b5 suspend children
[   65.942957] nouveau T[     DRM] 0xcccc0000:0x000490b5 suspend running
[   65.942959] nouveau T[    PCE0][0000:01:00.0][0x000090b5][ffff880132932920] use(-1) == 0
[   65.942961] nouveau T[    PCE0][0000:01:00.0][0x000090b5][ffff880132932920] suspending...
[   65.942962] nouveau T[    PCE0][0000:01:00.0] use(-1) == 1
[   65.942964] nouveau T[    PCE0][0000:01:00.0][0x0300c01b][ffff8801373fbc00] use(-1) == 0
[   65.942965] nouveau T[    PCE0][0000:01:00.0][0x0300c01b][ffff8801373fbc00] suspending...
[   65.942993] nouveau E[   PFIFO][0000:01:00.0] write fault at 0x0000000000 [PAGE_NOT_PRESENT] from PCOPY0/PCOPY0 on channel 0x003fe1b000
[   65.949760] ACPI handle has no context!
[   66.120887] e1000e 0000:00:19.0: wake-up capability enabled by ACPI
[   66.237094] i915 0000:00:02.0: power state changed by ACPI to D3hot
[   67.938439] nouveau E[   PFIFO][0000:01:00.0] channel 0 kick timeout
[   67.938441] nouveau E[   PFIFO][0000:01:00.0][0xc000906f][ffff880132ad2480] failed to detach PCE0 context, -16
[   67.938442] nouveau E[    PCE0][0000:01:00.0][0x0300c01b][ffff8801373fbc00] failed suspend, -16
[   67.938443] nouveau W[    PCE0][0000:01:00.0][0x000090b5][ffff880132932920] parent failed suspend, -16
[   67.938448] nouveau T[    PCE0][0000:01:00.0] use(+1) == 2
[   67.938450] nouveau E[     DRM] 0xcccc0000:0x000490b5 suspend failed with -16
[   67.938451] nouveau E[     DRM] 0xdddddddd:0xcccc0000 suspend failed with -16
[   67.938452] nouveau E[     DRM] 0xffffffff:0xdddddddd suspend failed with -16
[   67.938453] nouveau E[     DRM] 0xffffffff:0xffffffff suspend failed with -16
[   67.938455] nouveau D[     DRM] suspend completed with -16
[   67.938455] nouveau D[    1281] init running
[   67.938456] nouveau T[    1281] 0xffffffff:0xffffffff init running
[   67.938457] nouveau T[    1281] use(+1) == 2
[   67.938458] nouveau T[    1281] 0xffffffff:0xffffffff init children
[   67.938459] nouveau T[    1281] 0xffffffff:0xffffffff init completed
[   67.938460] nouveau D[    1281] init completed with 0
[   67.938461] nouveau  [     DRM] resuming display...
[   68.241664] pci_legacy_suspend(): nouveau_drm_suspend+0x0/0x270 [nouveau] returns -16
[   68.241683] dpm_run_callback(): pci_pm_suspend+0x0/0x140 returns -16
[   68.241685] PM: Device 0000:01:00.0 failed to suspend async: error -16
[   68.241810] PM: Some devices failed to suspend

I attached the almost full dmesg (gzipped). I did some investigation, and
tried to bisect, but code shuffled a lot and it isn't very useful, the
problem is well isolated on suspend path anyway. I started to have problems
with commit e193b1d42c390bf1bff7fa02a5a1202b98e75601, but with this commit
as HEAD the machine actually freezes when trying to suspend, probably at
nvc0_fence_suspend (not checked). But the code got changed a lot again
with commit ebb945a94bba2ce8dff7b0942ff2b3f2a52a0a69 (the failure seems
to be coming from nvc0_fifo_context_detach). With latest 3.7-rc5 I can't
suspend like the log shows, but at least the machine doesn't freeze.

-- 
[]'s
Herton

Attachment: dmesg-52.txt.gz
Description: Binary data