[Intel-gfx] [bisected] suspend broken by DRM fbdev name change on i915 IVB

[Johannes Stezenbach]

Hi,

yesterday I updated the kernel from 5.13.7 to 5.14.9 and found
it broke suspend-to-RAM. The machine displays a few messages on
text console after resume but hangs when switching to X11.

The hardware: Asus P8H77-V with Intel Core i5-3550 CPU,
display connected via DP.

  DP1 connected primary 1920x1200+0+0 (normal left inverted right x axis y axis) 520mm x 320mm
     1920x1200     59.95*+

I bisected it and the offending commit is the totally unlikely
and innocent looking

  commit b3484d2b03e4c940a9598aa841a52d69729c582a
  Author: Javier Martinez Canillas <javierm at redhat.com>
  Date:   Tue May 25 17:13:13 2021 +0200

    drm/fb-helper: improve DRM fbdev emulation device names

Now I'm running 5.14.9 with this commit reverted and
suspend works.

In /var/log/kern.log I found this on suspend:

[   34.002252][ T3455] ------------[ cut here ]------------
[   34.002256][ T3455] i915 0000:00:02.0: drm_WARN_ON((intel_de_read(dev_priv, intel_dp->output_reg) & (1 << 31)) == 0)
[   34.002274][ T3455] WARNING: CPU: 0 PID: 3455 at drivers/gpu/drm/i915/display/g4x_dp.c:431 intel_dp_link_down.isra.0+0x2e7/0x390
[   34.002285][ T3455] Modules linked in: kvm_intel kvm irqbypass ehci_pci xhci_pci ehci_hcd xhci_hcd
[   34.002304][ T3455] CPU: 0 PID: 3455 Comm: kworker/u8:27 Not tainted 5.14.9 #29
[   34.002309][ T3455] Hardware name: System manufacturer System Product Name/P8H77-V, BIOS 1905 10/27/2014
[   34.002312][ T3455] Workqueue: events_unbound async_run_entry_fn
[   34.002320][ T3455] RIP: 0010:intel_dp_link_down.isra.0+0x2e7/0x390
[   34.002326][ T3455] Code: 4c 8b 67 50 4d 85 e4 75 03 4c 8b 27 e8 d2 19 05 00 48 c7 c1 e8 53 8c 88 4c 89 e2 48 c7 c7 7b 83 89 88 48 89 c6 e8 d1 d0 52 00 <0f> 0b 48 83 c4 08 5b 5d 41 5c 41 5d 41 5e 41 5f c3 49 8b 7d 08 4c
[   34.002330][ T3455] RSP: 0018:ffffaa1dc1be3a88 EFLAGS: 00010282
[   34.002335][ T3455] RAX: 0000000000000000 RBX: ffffa15a45f28000 RCX: 0000000000000000
[   34.002338][ T3455] RDX: 0000000000000001 RSI: ffffffff8714af2f RDI: ffffffff8714af2f
[   34.002341][ T3455] RBP: 0000000080180344 R08: 00000007eab15173 R09: 0000000000000001
[   34.002344][ T3455] R10: 0000000000080000 R11: 0000000000000000 R12: ffffa15a40f87180
[   34.002347][ T3455] R13: ffffa15a464e0000 R14: ffffa15a593c6000 R15: 0000000000000001
[   34.002350][ T3455] FS:  0000000000000000(0000) GS:ffffa15d4f800000(0000) knlGS:0000000000000000
[   34.002354][ T3455] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   34.002357][ T3455] CR2: 000056000d53cf60 CR3: 000000015fa2a006 CR4: 00000000001706f0
[   34.002361][ T3455] Call Trace:
[   34.002373][ T3455]  g4x_post_disable_dp+0x2e/0x110
[   34.002380][ T3455]  intel_encoders_post_disable+0x73/0x80
[   34.002391][ T3455]  ilk_crtc_disable+0x96/0x3a0
[   34.002402][ T3455]  intel_old_crtc_state_disables+0x5c/0x110
[   34.002412][ T3455]  intel_atomic_commit_tail+0xdcc/0x1410
[   34.002434][ T3455]  intel_atomic_commit+0x332/0x3b0
[   34.002443][ T3455]  drm_atomic_helper_disable_all+0x175/0x190
[   34.002453][ T3455]  drm_atomic_helper_suspend+0xa6/0x200
[   34.002474][ T3455]  intel_display_suspend+0x23/0x50
[   34.002480][ T3455]  i915_drm_suspend+0x42/0xe0
[   34.002488][ T3455]  pci_pm_suspend+0x74/0x160
[   34.002496][ T3455]  ? pci_pm_freeze+0xb0/0xb0
[   34.002500][ T3455]  dpm_run_callback+0x6f/0x170
[   34.002512][ T3455]  __device_suspend+0x110/0x4b0
[   34.002521][ T3455]  async_suspend+0x1b/0x90
[   34.002529][ T3455]  async_run_entry_fn+0x2e/0x110
[   34.002535][ T3455]  process_one_work+0x2c9/0x550
[   34.002551][ T3455]  worker_thread+0x4f/0x3e0
[   34.002556][ T3455]  ? rescuer_thread+0x340/0x340
[   34.002563][ T3455]  kthread+0x14a/0x170
[   34.002569][ T3455]  ? set_kthread_struct+0x40/0x40
[   34.002578][ T3455]  ret_from_fork+0x22/0x30
[   34.002600][ T3455] ---[ end trace 2058ff589e8cbd78 ]---

and after resume:

[   44.513409][    C0] i915 0000:00:02.0: [drm] *ERROR* uncleared pch fifo underrun on pch transcoder A
[   44.514557][    C0] i915 0000:00:02.0: [drm] *ERROR* PCH transcoder A FIFO underrun
[   54.997256][ T3520] [drm:drm_atomic_helper_wait_for_flip_done] *ERROR* [CRTC:45:pipe A] flip_done timed out


Scratching my head about it I found one hint in
/usr/lib/pm-utils/sleep.d/98video-quirk-db-handler:

 using_kms() { grep -q -E '(nouveau|drm)fb' /proc/fb; }

So the ABI change in /proc/fb causes the pm-utils scripts to
skip the --quirk-no-chvt and apply other quirks,
/var/log/pm-suspend.log says:

  Running hook /usr/lib/pm-utils/sleep.d/98video-quirk-db-handler suspend suspend:
  No quirk database entry for this system, using default.
  /usr/lib/pm-utils/sleep.d/98video-quirk-db-handler suspend suspend: success.
  
  Running hook /usr/lib/pm-utils/sleep.d/98video-quirk-db-handler resume suspend:
  Saving last known working quirks: --quirk-vbe-post --quirk-dpms-on
                              --quirk-dpms-suspend --quirk-vbestate-restore
                              --quirk-vbemode-restore --quirk-vga-mode-3
  /usr/lib/pm-utils/sleep.d/98video-quirk-db-handler resume suspend: success.

whereas in the normal case it's

  Running hook /usr/lib/pm-utils/sleep.d/98video-quirk-db-handler suspend suspend:
  Kernel modesetting video driver detected, not using quirks.
  /usr/lib/pm-utils/sleep.d/98video-quirk-db-handler suspend suspend: success.
  
  Running hook /usr/lib/pm-utils/sleep.d/98video-quirk-db-handler resume suspend:
  /usr/lib/pm-utils/sleep.d/98video-quirk-db-handler resume suspend: success.


Johannes