Bug 16442 - AB-INT: linux: drm_atomic_helper: vblank wait timed out
Summary: AB-INT: linux: drm_atomic_helper: vblank wait timed out
Status: NEW
Alias: None
Product: Runtime Testing
Classification: QA/Testing
Component: oeqa/runtime (show other bugs)
Version: unspecified
Hardware: x86 Multiple
: Medium+ normal
Target Milestone: 6.2 M1
Assignee: Bruce Ashfield
QA Contact:
URL:
Whiteboard: AB-INT
Depends on:
Blocks:
 
Reported: 2026-10-01 12:08 UTC by Mathieu Dubois-Briand
Modified: 2026-11-11 15:51 UTC (History)
3 users (show)

See Also:
OS type for building Yocto: ---
Type of Regression: ---
Verified:
Documentation change: No (bug/feature does not impact docs)


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Mathieu Dubois-Briand 2026-10-01 12:08:40 UTC
We are seeing a new failure with the Linux kernel boot on qemu:


> -----------------------
> Central error: [    4.607421] ------------[ cut here ]------------
> ***********************
> [    1.429143]     TERM=linux
> [    1.497412] usb 2-2: new high-speed USB device number 3 using ehci-pci
> [    1.512300] udevd[156]: starting version 3.2.14
> [    1.516174] udevd[157]: starting eudev-3.2.14
> [    1.630813] input: QEMU QEMU USB Keyboard as /devices/pci0000:00/0000:00:1d.7/usb2/2-2/2-2:1.0/0003:0627:0001.0002/input/input2
> [    1.685861] hid-generic 0003:0627:0001.0002: input: USB HID v1.11 Keyboard [QEMU QEMU USB Keyboard] on usb-0000:00:1d.7-2/input0
> [    1.755803] EXT4-fs (vda): re-mounted cfec03de-a7fa-4cdb-a77c-5807dd93bb11.
> [    3.523804] Bluetooth: BNEP (Ethernet Emulation) ver 1.3
> [    3.525677] Bluetooth: BNEP filters: protocol multicast
> [    3.527499] Bluetooth: BNEP socket layer initialized
> [    4.607421] ------------[ cut here ]------------
> [    4.608801] [CRTC:37:crtc-0] vblank wait timed out
> [    4.610163] WARNING: /usr/src/kernel/drivers/gpu/drm/drm_atomic_helper.c:1921 at drm_atomic_helper_wait_for_vblanks.part.0+0x215/0x240, CPU#2: kworker/2:1/51
> [    4.614073] Modules linked in: bnep
> [    4.615083] CPU: 2 UID: 0 PID: 51 Comm: kworker/2:1 Not tainted 7.2.6-yocto-standard #1 PREEMPT(full) 
> [    4.617668] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.17.0-0-gb52ca86e094d-prebuilt.qemu.org 04/01/2014
> [    4.620767] Workqueue: events drm_fb_helper_damage_work
> [    4.622244] RIP: 0010:drm_atomic_helper_wait_for_vblanks.part.0+0x21d/0x240
> [    4.624148] Code: ff 84 c0 74 88 48 8b 7d 98 48 8d 75 a8 e8 5b 18 5a ff 45 85 ed 0f 85 f0 fe ff ff 48 8d 3d cb 7a 2b 01 49 8b 57 20 41 8b 77 58 <67> 48 0f b9 3a e9 d7 fe ff ff 48 89 5d d8 4c 89 65 e0 4c 89 6d e8
> [    4.629285] RSP: 0018:ffffb3d8801c7b88 EFLAGS: 00010246
> [    4.630751] RAX: ffff9cfa412f7438 RBX: 0000000000000000 RCX: 0000000000000202
Comment 1 Mathieu Dubois-Briand 2026-10-01 12:31:28 UTC
pkgman-non-rpm ubuntu2204-vk-1 master completed at 2026-09-30 03:51:15+00:00
https://autobuilder.yoctoproject.org/valkyrie/#/builders/67/builds/4750/steps/15/logs/stdio

qemux86-64-musl-ptest debian12-vk-3 mathieu/master-next completed at 2026-10-01 08:41:46+00:00
https://valkyrie.yocto.io/pub/non-release/20261001-53/testresults/qemux86-64-musl-ptest/
https://autobuilder.yoctoproject.org/valkyrie/#/builders/110/builds/1293/steps/15/logs/stdio
Comment 2 Randy MacLeod 2026-10-01 14:44:08 UTC
RP thinks we had a fix for this. 
Any ideas Bruce ?
Comment 3 Bruce Ashfield 2026-10-01 16:55:12 UTC
Thanks for the detailed trace .. easier to triage

This is a different warning from the one we recently fixed. We did
backport a "vblank wait timed out" fix to 7.2 (d8a1ec518ab72,
"drm/client: Avoid warning on vblank timeout during modeset client
waits", Krystian Kaniewski's patch), but that one only covers the
drm_client_modeset path and it swaps drm_crtc_wait_one_vblank() for a
non-warning variant during modeset client waits.

The warning in this report is a different site: it fires in
drm_atomic_helper_wait_for_vblanks(), reached via the
drm_fb_helper_damage_work workqueue, not the modeset client path. So
d8a1ec518ab72 does not silence this one.

Is that the fix you were thinking of ?

The underlying cause looks the same, though: on the emulated QEMU GPU,
drm_crtc_vblank_get() succeeds (vblank is advertised), so the helper
waits up to 100ms for the vblank counter to advance. Under
autobuilder/ptest host load the emulated vblank simply doesn't arrive in
time, so it legitimately times out and WARNs. It is a false positive on
virtual hardware, not a real driver fault; the CRTC comes up and boot
continues.

Because drm_atomic_helper_wait_for_vblanks() is a core exported helper
used by every atomic driver, we cannot blanket-silence it the way we did
the client path.

The cleaner fix is driver-side (the virtual GPU driver setting
no_vblank / not advertising vblank) or an upstream conditional. I will
trace which QEMU GPU driver is in play here and whether there is an
upstream change to pick up (rather than silencing the WARN)

Was my previous fix in the run that produced this warning ? I can't
tell from the kernel string.
Comment 4 Bruce Ashfield 2026-10-01 18:02:48 UTC
Followed up on the trace, and it is worth recording what I found.

This is a known upstream issue, actively discussed this year (syzbot +
dri-devel/LKML). The root cause lines up with what we see here: drivers
that simulate vblank with an hrtimer get that timer starved under heavy
CPU load, so the vblank never arrives and drm_atomic_helper_wait_for_vblanks()
legitimately times out and WARNs. On these runs that is exactly the
pressure: ptest load on a PREEMPT(full) kernel under emulation. It is a
scheduling-starvation symptom, not a driver fault; the CRTC comes up and
boot continues.

On whether there is a fix to pick up: there is an RFC
("drm/vblank: Replace WARN with drm_err for vblank wait timeouts") that
downgrades the WARN to drm_err in both drm_atomic_helper_wait_for_vblanks()
and drm_crtc_wait_one_vblank(). In review, Krystian Kaniewski asked to
narrow it and to suppress the warning only in the optional fbdev /
drm_client_modeset throttling path, and keep the WARN on the
driver-facing paths like drm_atomic_helper_wait_for_vblanks().

That narrow approach is precisely the patch we already carry
(d8a1ec518ab72). So the path this report hits is the one upstream has, so
far, deliberately chosen to keep warning on, and the broader downgrade
that would quiet it is still an unmerged RFC. In other words there isn't
a clean upstream change to backport for this specific WARN yet.

So the options, roughly:
 - wait for upstream to settle the RFC and pick up whatever lands (my
   preference. I'd rather not diverge on a core helper);
 - carry a local WARN->drm_err downgrade of just the atomic-helper hunk
   (works, but diverges from where upstream is heading);
 - treat it as benign/expected under emulated ptest load, since that is
   what it is.

One thing I could not pin down from the source alone: which GPU driver is
actually tripping it. The drivers we build in for qemux86-64 (bochs,
virtio-gpu) do not advertise vblank in their current form, so by rights
neither should reach this WARN. Which suggests a vblank-simulating
driver (vkms/vmwgfx style) is in play in the failing config.

Can someone post DRM init lines from the boot dmesg of a failing run
(the "[drm] ..." / driver-bind lines)? That would confirm the driver
and let me be sure we are reasoning about the right one.
Comment 5 Mathieu Dubois-Briand 2026-10-05 09:41:54 UTC
qemuarm64-ptest alma10-vk-arm1 mathieu/master-next completed at 2026-10-03 17:51:29+00:00
https://valkyrie.yocto.io/pub/non-release/20261003-100/testresults/qemuarm64-ptest/
https://autobuilder.yoctoproject.org/valkyrie/#/builders/61/builds/4661/steps/15/logs/stdio
Comment 6 Yoann Congal 2026-10-06 09:48:14 UTC
qemuriscv64-ptest ubuntu2504-vk-1 blacksail completed at 2026-10-05 15:05:31+00:00
https://autobuilder.yoctoproject.org/valkyrie/?#/builders/56/builds/2550
Comment 7 Yoann Congal 2026-10-06 09:52:46 UTC
(In reply to Bruce Ashfield from comment #4)
> Can someone post DRM init lines from the boot dmesg of a failing run
> (the "[drm] ..." / driver-bind lines)? That would confirm the driver
> and let me be sure we are reasoning about the right one.

From: https://valkyrie.yocto.io/pub/non-release/20261005-65/testresults/qemuriscv64-ptest/core-image-ptest-python3-click/qemu_boot_log.20261005115453

Extract:
| [    1.823520] bochs-drm 0000:00:03.0: enabling device (0000 -> 0002)
| [    1.827057] [drm] Found bochs VGA, ID 0xb0c5.
| [    1.864926] [drm] Framebuffer size 16384 kB @ 0x40000000, mmio @ 0x41009000.
| [    1.877344] [drm] Initialized bochs-drm 1.0.0 for 0000:00:03.0 on minor 0
| [    2.091702] bochs-drm 0000:00:03.0: [drm] fb0: bochs-drmdrmfb frame buffer device
| [    7.664680] ------------[ cut here ]------------
| [    7.665610] [CRTC:37:crtc-0] vblank wait timed out
| [    7.667861] WARNING: /usr/src/kernel/drivers/gpu/drm/drm_atomic_helper.c:1921 at drm_atomic_helper_wait_for_vblanks.part.0+0x1de/0x218, CPU#2: kworker/2:0/29
Comment 8 Mathieu Dubois-Briand 2026-10-08 06:49:38 UTC
musl-qemuarm64 stream9-vk-1 master completed at 2026-10-08 02:16:05+00:00
https://autobuilder.yoctoproject.org/valkyrie/#/builders/111/builds/3525/steps/16/logs/stdio
Comment 9 Mathieu Dubois-Briand 2026-10-16 08:18:29 UTC
qemux86-64 debian12-vk-1 master completed at 2026-10-16 02:55:01+00:00
https://autobuilder.yoctoproject.org/valkyrie/#/builders/68/builds/4974/steps/17/logs/stdio
Comment 10 Mathieu Dubois-Briand 2026-10-20 07:50:49 UTC
qemux86-64-ptest ubuntu2404-vk-1 mathieu/master-next completed at 2026-10-19 11:16:21+00:00
https://valkyrie.yocto.io/pub/non-release/20261019-62/testresults/qemux86-64-ptest/
https://autobuilder.yoctoproject.org/valkyrie/#/builders/73/builds/4772/steps/15/logs/stdio
Comment 11 Mathieu Dubois-Briand 2026-10-21 07:47:24 UTC
qemuriscv64-ptest ubuntu2204-vk-1 mathieu/master-next-tests&mathieu/master-next completed at 2026-10-20 23:10:15+00:00
https://valkyrie.yocto.io/pub/non-release/20261020-104/testresults/qemuriscv64-ptest/
https://autobuilder.yoctoproject.org/valkyrie/#/builders/56/builds/2636/steps/15/logs/stdio
Comment 12 Mathieu Dubois-Briand 2026-11-02 07:45:54 UTC
qemux86-64-musl-ptest fedora44-vk-3 mathieu/master-next completed at 2026-10-30 18:25:35+00:00
https://valkyrie.yocto.io/pub/non-release/20261030-99/testresults/qemux86-64-musl-ptest/
https://autobuilder.yoctoproject.org/valkyrie/#/builders/110/builds/1489/steps/15/logs/stdio
Comment 13 Mathieu Dubois-Briand 2026-11-11 15:51:20 UTC
qemux86-64-ptest alma8-vk-2 master&mathieu/master-next&mathieu/master-next-tests completed at 2026-11-10 15:43:03+00:00
https://valkyrie.yocto.io/pub/non-release/20261110-84/testresults/qemux86-64-ptest/
https://autobuilder.yoctoproject.org/valkyrie/#/builders/73/builds/4921/steps/16/logs/stdio