linux/drivers/gpu/drm/i915
Daniel Vetter 17e1df07df drm/i915: fix wait_for_pending_flips vs gpu hang deadlock
My g33 here seems to be shockingly good at hitting them all. This time
around kms_flip/flip-vs-panning-vs-hang blows up:

intel_crtc_wait_for_pending_flips correctly checks for gpu hangs and
if a gpu hang is pending aborts the wait for outstanding flips so that
the setcrtc call will succeed and release the crtc mutex. And the gpu
hang handler needs that lock in intel_display_handle_reset to be able
to complete outstanding flips.

The problem is that we can race in two ways:
- Waiters on the dev_priv->pending_flip_queue aren't woken up after
  we've the reset as pending, but before we actually start the reset
  work. This means that the waiter doesn't notice the pending reset
  and hence will keep on hogging the locks.

  Like with dev->struct_mutex and the ring->irq_queue wait queues we
  there need to wake up everyone that potentially holds a lock which
  the reset handler needs.

- intel_display_handle_reset was called _after_ we've already
  signalled the completion of the reset work. Which means a waiter
  could sneak in, grab the lock and never release it (since the
  pageflips won't ever get released).

  Similar to resetting the gem state all the reset work must complete
  before we update the reset counter. Contrary to the gem reset we
  don't need to have a second explicit wake up call since that will
  have happened already when completing the pageflips. We also don't
  have any issues that the completion happens while the reset state is
  still pending - wait_for_pending_flips is only there to ensure we
  display the right frame. After a gpu hang&reset events such
  guarantees are out the window anyway. This is in contrast to the gem
  code where too-early wake-up would result in unnecessary restarting
  of ioctls.

Also, since we've gotten these various deadlocks and ordering
constraints wrong so often throw copious amounts of comments at the
code.

This deadlock regression has been introduced in the commit which added
the pageflip reset logic to the gpu hang work:

commit 96a02917a0
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date:   Mon Feb 18 19:08:49 2013 +0200

    drm/i915: Finish page flips and update primary planes after a GPU reset

v2:
- Add comments to explain how the wake_up serves as memory barriers
  for the atomic_t reset counter.
- Improve the comments a bit as suggested by Chris Wilson.
- Extract the wake_up calls before/after the reset into a little
  i915_error_wake_up and unconditionally wake up the
  pending_flip_queue waiters, again as suggested by Chris Wilson.

v3: Throw copious amounts of comments at i915_error_wake_up as
suggested by Chris Wilson.

Cc: stable@vger.kernel.org
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-09-09 11:26:03 +02:00
..
dvo_ch7xxx.c drm/i915: dvo_ch7xxx: fix vsync polarity setting 2013-07-25 16:10:22 +02:00
dvo_ch7017.c drm/i915/dvo: implement get_hw_state 2012-09-06 07:58:52 +02:00
dvo_ivch.c drm/i915/dvo: implement get_hw_state 2012-09-06 07:58:52 +02:00
dvo_ns2501.c drm/i915/dvo: implement get_hw_state 2012-09-06 07:58:52 +02:00
dvo_sil164.c drm/i915/dvo: implement get_hw_state 2012-09-06 07:58:52 +02:00
dvo_tfp410.c drm/i915/dvo: implement get_hw_state 2012-09-06 07:58:52 +02:00
dvo.h Merge branch 'drm-next' of git://people.freedesktop.org/~airlied/linux 2012-10-03 23:29:23 -07:00
i915_debugfs.c drm/i915: Report requested frequency alongside current frequency in debugfs 2013-09-03 11:12:28 +02:00
i915_dma.c drm/i915: Delay disabling of VGA memory until vgacon->fbcon handoff is done 2013-09-06 23:27:03 +02:00
i915_drv.c drm/i915: split PCI IDs out into i915_drm.h v4 2013-09-03 19:17:56 +02:00
i915_drv.h drm/i915: fix hpd work vs. flush_work in the pageflip code deadlock 2013-09-04 17:34:02 +02:00
i915_gem_context.c drm/i915: mm_list is per VMA 2013-08-08 14:06:58 +02:00
i915_gem_debug.c drm/i915: Fix #endif comment 2013-08-09 10:45:52 +02:00
i915_gem_dmabuf.c drm/i915: Pin pages whilst mapping the dma-buf 2013-09-03 19:17:58 +02:00
i915_gem_evict.c drm/i915/vma: Correct use after free in eviction 2013-08-23 14:52:21 +02:00
i915_gem_execbuffer.c drm/i915: fix up the relocate_entry refactoring 2013-09-03 19:18:01 +02:00
i915_gem_gtt.c drm/i915: Use Write-Through cacheing for the display plane on Iris 2013-08-22 13:31:38 +02:00
i915_gem_stolen.c drm/i915: Skip stolen region initialisation if none is reserved 2013-09-05 14:49:50 +02:00
i915_gem_tiling.c drm/i915: plumb VM into bind/unbind code 2013-08-08 14:04:20 +02:00
i915_gem.c drm/i915: Hold an object reference whilst we shrink it 2013-09-05 14:47:59 +02:00
i915_gpu_error.c drm/i915: Embed the ring->private within the struct intel_ring_buffer 2013-09-03 19:17:55 +02:00
i915_ioc32.c UAPI: (Scripted) Convert #include "..." to #include <path/...> in drivers/gpu/ 2012-10-02 18:01:07 +01:00
i915_irq.c drm/i915: fix wait_for_pending_flips vs gpu hang deadlock 2013-09-09 11:26:03 +02:00
i915_reg.h drm/i915: enable trickle feed on Haswell 2013-09-03 19:17:57 +02:00
i915_suspend.c Linux 3.10 2013-07-18 12:03:29 +02:00
i915_sysfs.c drm/i915: Adjust available RPS information through sysfs for vlv 2013-09-03 11:10:55 +02:00
i915_trace_points.c
i915_trace.h drm/i915: plumb VM into bind/unbind code 2013-08-08 14:04:20 +02:00
i915_ums.c drm/i915: scrap register address storage 2013-06-10 19:54:14 +02:00
intel_acpi.c i915: fix ACPI _DSM warning 2013-08-05 19:04:05 +02:00
intel_bios.c drm/i915: Organize VBT stuff inside drm_i915_private 2013-05-10 21:56:46 +02:00
intel_bios.h drm/i915: set CPT FDI RX polarity bits based on VBT 2013-04-18 09:43:31 +02:00
intel_crt.c drm/i915: Apply the force-detect VGA w/a to Valleyview 2013-09-03 11:10:55 +02:00
intel_ddi.c drm/i915: Track pfit enable state separately from size 2013-09-08 21:59:13 +02:00
intel_display.c drm/i915: Track pfit enable state separately from size 2013-09-08 21:59:13 +02:00
intel_dp.c drm/i915: allow package C8+ states on Haswell (disabled) 2013-08-23 14:52:33 +02:00
intel_drv.h drm/i915: Track pfit enable state separately from size 2013-09-08 21:59:13 +02:00
intel_dvo.c drm/i915/dvo: use native encoder ->mode_set callback 2013-08-04 21:25:21 +02:00
intel_fb.c drm/i915: Export intel_framebuffer_fini 2013-08-06 20:08:50 +02:00
intel_hdmi.c Merge tag 'drm-intel-next-2013-08-23' of git://people.freedesktop.org/~danvet/drm-intel into drm-next 2013-08-30 09:47:41 +10:00
intel_i2c.c drm/i915: allow package C8+ states on Haswell (disabled) 2013-08-23 14:52:33 +02:00
intel_lvds.c drm/i915: Fix pipe config warnings when dealing with LVDS fixed mode 2013-09-03 19:18:00 +02:00
intel_modes.c drm/i915: Add "Automatic" mode for the "Broadcast RGB" property 2013-01-20 13:09:44 +01:00
intel_opregion.c drm/i915: try not to lose backlight CBLV precision 2013-09-06 09:41:20 +02:00
intel_overlay.c drm/i915: Add VM to pin 2013-08-05 19:04:09 +02:00
intel_panel.c drm/i915: Track pfit enable state separately from size 2013-09-08 21:59:13 +02:00
intel_pm.c drm/i915: Track pfit enable state separately from size 2013-09-08 21:59:13 +02:00
intel_ringbuffer.c drm/i915: Embed the ring->private within the struct intel_ring_buffer 2013-09-03 19:17:55 +02:00
intel_ringbuffer.h drm/i915: Embed the ring->private within the struct intel_ring_buffer 2013-09-03 19:17:55 +02:00
intel_sdvo_regs.h drm/i915: clear the entire sdvo infoframe buffer 2012-10-24 15:12:48 +02:00
intel_sdvo.c drm/i915: handle sdvo input pixel multiplier correctly again 2013-09-04 17:34:03 +02:00
intel_sideband.c drm/i915: change VLV IOSF sideband accessors to not return error code 2013-05-23 23:25:42 +02:00
intel_sprite.c drm/i915: enable trickle feed on Haswell 2013-09-03 19:17:57 +02:00
intel_tv.c drm/i915/tv: Use native encoder->mode_set callback 2013-08-04 21:25:22 +02:00
intel_uncore.c drm/i915: sanitize forcewake registers on reset 2013-09-03 11:10:54 +02:00
Makefile drm/i915: Colocate all GT access routines in the same file 2013-07-25 15:21:50 +02:00