Bug 103529 - [GM45] GPU hang with mpv fullscreen (bisected)
Summary: [GM45] GPU hang with mpv fullscreen (bisected)
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/DRI/i965 (show other bugs)
Version: 17.2
Hardware: x86 (IA32) Linux (All)
: medium normal
Assignee: Kenneth Graunke
QA Contact: Intel 3D Bugs Mailing List
Keywords: bisected, regression
Depends on:
Blocks: mesa-17.3
  Show dependency treegraph
Reported: 2017-10-31 17:11 UTC by Adesh
Modified: 2017-11-17 01:41 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Description Adesh 2017-10-31 17:11:14 UTC
Complete graphics freeze when watching videos in mpv fullscreen mode with hardware acceleration enabled (vaapi), which results in GPU hang. Video plays just fine in normal windowed mode.

After mpv crashes, following is printed on the terminal:
intel_do_flush_locked failed: Input/output errord: 47

I bisected mesa between 17.1.8 and 17.2.0 to find first bad commit to be 79f2a5541f92920d35d3621179f7377c97cc75e1

Relevant thread on Arch Linux forums: https://bbs.archlinux.org/viewtopic.php?id=230688
Comment 1 Emil Velikov 2017-10-31 18:07:07 UTC
This seems to be covering i965 DRI driver - reassigning.
Comment 2 Adesh 2017-10-31 19:17:06 UTC
Building tag 17.2.4 with the below change fixes the problem for me.

diff --git a/src/mesa/drivers/dri/i965/brw_clear.c b/src/mesa/drivers/dri/i965/brw_clear.c
index 5118d96c38..2c9f7d735f 100644
--- a/src/mesa/drivers/dri/i965/brw_clear.c
+++ b/src/mesa/drivers/dri/i965/brw_clear.c
@@ -239,7 +239,7 @@ brw_clear(struct gl_context *ctx, GLbitfield mask)
          mt->stencil_mt->r8stencil_needs_update = true;
-   if (mask & BUFFER_BITS_COLOR) {
+   if (brw->gen >= 6 && (mask & BUFFER_BITS_COLOR)) {
       brw_blorp_clear_color(brw, fb, mask, partial_clear,
       debug_mask("blorp color", mask & BUFFER_BITS_COLOR);
Comment 3 Matt Turner 2017-11-02 01:17:11 UTC
The bisection is

commit 79f2a5541f92920d35d3621179f7377c97cc75e1
Author: Jason Ekstrand <jason.ekstrand@intel.com>
Date:   Fri May 12 17:14:18 2017 -0700

    i965: Use BLORP for color clears on gen4-5
Comment 4 Emil Velikov 2017-11-08 17:42:03 UTC
Wondering if we couldn't merge the hunk in comment 2 (alongside a reference to this bug). It will address the mpv crash and we can resolve properly at a later stage?
Comment 5 Kenneth Graunke 2017-11-16 08:41:22 UTC

I was able to reproduce GPU hangs on Ironlake using mpv -hwdec=vaapi, and I believe I have a fix: https://patchwork.freedesktop.org/patch/188597/

If possible, could you try the patch and see if it fixes it for you?

If it doesn't work, could you please attach /sys/class/drm/card0/error to the bug report?

Comment 6 Adesh 2017-11-16 13:06:01 UTC
Kenneth, your patch fixes the issue for me (on 17.2.4).
Comment 7 Kenneth Graunke 2017-11-17 01:41:30 UTC
Fixed on master by:

commit 8f91aa35a54e127b68415376ef2b577ea8fc30f9
Author: Kenneth Graunke <kenneth@whitecape.org>
Date:   Wed Nov 15 22:40:16 2017 -0800

    i965: Upload invariant state once at the start of the batch on Gen4-5.
    We want to emit invariant state at the start of a render batch.  In the
    past, this more or less happened: a new batch flagged BRW_NEW_CONTEXT
    (because we don't have hardware contexts), which triggered the
    brw_invariant_state atom.  So, it would be emitted before any 3D
    drawing.  (Technically, there might be some BLT commands in the batch
    because Gen4-5 have a single combined render/BLT ring, but that should
    be harmless).
    With the advent of BLORP, this broke.  The first item in a batch might
    be a BLORP operation, which bypasses the normal draw upload path.  So,
    we need to ensure invariant state happens first.  To do that, we just
    upload it when creating a new batch.  On Gen6+ we'd need to worry about
    whether it's a RENDER or BLT batch, but because we have a combined ring,
    this approach should work fine on Gen4-5.
    Seems to fix GPU hangs when playing hardware accelerated video with
    mpv -hwdec=vaapi on Ironlake.
    Cc: mesa-stable@lists.freedesktop.org
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103529
    Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

Hopefully it should arrive in the stable branches soon.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.