Public bug reported:

## Summary

Minimal, network-free reproduction:

```bash
sudo ffmpeg -f kmsgrab -device /dev/dri/card1 -framerate 30 -i - \
  -vf 'hwmap=derive_device=vaapi,scale_vaapi=format=nv12' \
  -c:v hevc_vaapi -rc_mode CBR -b:v 6M -profile:v main -g 60 -bf 0 \
  -t 2 -f null -
```

On a freshly booted system this hangs the GPU (3 hangs, `ecode 12:10:85dffffb`,
`ccs0`, active process `vf#0:0`). ffmpeg reports:

```
[Parsed_scale_vaapi_1 @ ...] Failed to start picture processing: 1 (operation 
failed).
[vf#0:0 @ ...] Error while filtering: Input/output error
```

The GNOME session then stops presenting: the CRTC stays `enable=yes/active=yes`
with a valid framebuffer, but mutter no longer commits new frames and the
desktop is black. It is recoverable with a VT switch (`chvt 1; chvt 2`), no
reboot required.

Reproduced **3/3 times** across two boots (twice via an RTMP output, once with
this `-f null` variant on a clean boot, first attempt).

## Environment

| Item | Value |
|---|---|
| Distribution | Ubuntu 26.10 |
| Kernel | 7.3.0-10-generic (Ubuntu 7.3.0-10.10) |
| GPU | Intel Arrow Lake-P [Arc Pro 130T/140T], PCI 8086:7d51 (rev 03) |
| Kernel driver | i915 (GuC `i915/mtl_guc_70.bin` 70.53.0, GT0 + GT1) |
| Mesa | 26.2.4-1ubuntu1 |
| libva | 2.24.1-2 (reports VA-API 1.24) |
| iHD | intel-media-va-driver 26.2.4+dfsg1-1 |
| ffmpeg | 8.1.2-2ubuntu2 (`--enable-libdrm`, kmsgrab, vaapi) |
| Session | GNOME Wayland/mutter, 1920x1200@60 |
| Scanout buffer | `XR30` (DRM_FORMAT_XRGB2101010, 10 bpc), modifier 
`0x010000000000000f` (`I915_FORMAT_MOD_4_TILED`) |

Note: the i915 error state labels the platform `METEORLAKE` while lspci reports
Arrow Lake-P; ARL reuses the MTL display/media IP, so this is expected.

## Isolation

Each test run separately on the same boot; "hang" counts kernel
`GPU HANG`/`Resetting chip` events.

| # | Pipeline | Result |
|---|---|---|
| A | `testsrc2` → `hwupload` → `hevc_vaapi` | pass, 0 hangs |
| A2 | `testsrc2` → `hwupload` → `scale_vaapi=w=1280:h=720` → `hevc_vaapi` | 
pass, 0 hangs |
| B | `kmsgrab` → `hwdownload,format=x2rgb10` | fails, no hang (`mmap EINVAL`, 
see below) |
| **C** | **`kmsgrab` → `hwmap=derive_device=vaapi` → `scale_vaapi` → 
`hevc_vaapi`** | **fail, 3 hangs** |

A2 shows VPP and the HEVC encoder both work on ordinary uploaded surfaces.
B shows the CPU-download path of the same buffer fails cleanly (no hang).
C is the only combination that hangs.

## Kernel log (test C, clean boot)

```
00:27:03 i915 0000:00:02.0: [drm] GPU HANG: ecode 12:10:85dffffb, in vf#0:0 
[5047]
00:27:23 Fence expiration time out i915-0000:00:02.0:vf#0:0[5047]:4!
00:27:28 i915 0000:00:02.0: [drm] GPU HANG: ecode 12:10:85dffffb, in vf#0:0 
[5047]
00:27:48 Fence expiration time out i915-0000:00:02.0:vf#0:0[5047]:6!
00:27:53 i915 0000:00:02.0: [drm] GPU HANG: ecode 12:10:85dffffb, in vf#0:0 
[5047]
```

No full GT reset message this time; the context-level resets alone were enough
to stop mutter from presenting. The ffmpeg pipeline took ~70 s to unwind.

Error state header:

```
GPU HANG: ecode 12:10:85dffffb, in vf#0:0 [5047]
Active process (on ring ccs0): vf#0:0 [5047]
Platform: METEORLAKE / Subplatform: 0x1
PCI ID: 0x7d51
Reset count: 0
```

## Secondary issue: `hwdownload` cannot map the buffer

```bash
sudo ffmpeg -f kmsgrab -device /dev/dri/card1 -framerate 30 -i - \
  -vf 'hwdownload,format=x2rgb10' -frames:v 1 -f null -
```

```
[AVHWFramesContext @ ...] Failed to map DRM object 4 to memory: 22.
[hwdownload @ ...] Failed to download frame: -22.
```

`format=bgr0` is rejected earlier (`Invalid output format bgr0 for hwframe
download`) because the buffer is 10-bit — expected. The `mmap` `EINVAL` on the
4-tiled scanout object is the interesting part; it makes the CPU fallback
unusable and leaves no safe path for `kmsgrab` on this platform.

## Analysis

* The failing batches belong to ffmpeg's filter-graph context (`vf#0:0`), i.e.
  they were submitted by libva/iHD on ffmpeg's behalf.
* ffmpeg's own error is at the VPP stage (`Failed to start picture processing`
  from `scale_vaapi`), immediately followed by the hang.
* Yet VPP works on ordinary surfaces (test A2), and the HEVC encoder works
  (tests A, A2). Only the surface imported from the 4-tiled `XR30` scanout
  buffer triggers the hang.
* The hang is reported on `ccs0` (compute), while VPP/VDENC normally run on the
  media GT. This suggests the import/copy path iHD uses for the external buffer
  (a compute/render-side operation) rather than the VPP or encoder kernel.
* The same buffer is also not CPU-mappable (secondary issue), i.e. it is an
  unusual buffer all the way down.

## Impact

* Any `kmsgrab` + VA-API workflow on this platform wedges the GPU.
* The desktop session becomes unusable (black screen) although the machine and
  SSH remain functional; recovery needs a VT switch or session restart.
* `kmsgrab` is therefore unusable here: the VA-API path hangs and the CPU path
  cannot map the buffer.

## Attachments

* `i915-error-state-0027.txt` — `/sys/class/drm/card1/error` from test C

## dmesg record

11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] Found meteorlake 
(device ID 7d51) integrated display version 14.00 stepping D0
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] VT-d active for gfx 
access
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: vgaarb: deactivate vga 
console
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] Using Transparent 
Hugepages
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: vgaarb: VGA decodes 
changed: olddecodes=io+mem,decodes=io:owns=io+mem
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] Finished loading DMC 
firmware i915/mtl_dmc.bin (v2.23)
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT0: GuC firmware 
i915/mtl_guc_70.bin version 70.53.0
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT0: GUC: submission 
enabled
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT0: GUC: SLPC enabled
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT0: GUC: RC enabled
11月 17 00:20:22 RUSTYBOX kernel: mei_gsc_proxy 
0000:00:16.0-0f73db04-97ab-4125-b893-e904ad0d5464: bound 0000:00:02.0 (ops 
i915_gsc_proxy_component_ops [i915])
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT1: GuC firmware 
i915/mtl_guc_70.bin version 70.53.0
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT1: HuC firmware 
i915/mtl_huc_gsc.bin version 8.5.4
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT1: HuC: 
authenticated for clear media
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT1: GUC: submission 
enabled
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT1: GUC: SLPC enabled
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT1: GUC: RC enabled
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] Protected Xe Path 
(PXP) protected content support initialized
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] Registered 4 planes 
with drm panic
11月 17 00:20:22 RUSTYBOX kernel: [drm] Initialized i915 1.6.0 for 0000:00:02.0 
on minor 1
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] Allocating fbdev: 
Stolen memory not preferred.
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT1: Loaded GSC 
firmware i915/mtl_gsc_1.bin (cv1.0, r102.1.15.1926, svn 1)
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GT1: HuC: 
authenticated for all workloads
11月 17 00:20:22 RUSTYBOX kernel: i915 0000:00:02.0: [drm] fb0: i915drmfb frame 
buffer device
11月 17 00:20:22 RUSTYBOX kernel: sof-audio-pci-intel-mtl 0000:00:1f.3: bound 
0000:00:02.0 (ops intel_audio_component_bind_ops [i915])
11月 17 00:27:03 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GPU HANG: ecode 
12:10:85dffffb, in vf#0:0 [5047]
11月 17 00:27:03 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GPU error state saved 
to /sys/class/drm/card1/error
11月 17 00:27:03 RUSTYBOX kernel: i915 0000:00:02.0: [drm] vf#0:0[5047] context 
reset due to GPU hang
11月 17 00:27:23 RUSTYBOX kernel: Fence expiration time out 
i915-0000:00:02.0:vf#0:0[5047]:4!
11月 17 00:27:28 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GPU HANG: ecode 
12:10:85dffffb, in vf#0:0 [5047]
11月 17 00:27:28 RUSTYBOX kernel: i915 0000:00:02.0: [drm] vf#0:0[5047] context 
reset due to GPU hang
11月 17 00:27:48 RUSTYBOX kernel: Fence expiration time out 
i915-0000:00:02.0:vf#0:0[5047]:6!
11月 17 00:27:53 RUSTYBOX kernel: i915 0000:00:02.0: [drm] GPU HANG: ecode 
12:10:85dffffb, in vf#0:0 [5047]
11月 17 00:27:53 RUSTYBOX kernel: i915 0000:00:02.0: [drm] vf#0:0[5047] context 
reset due to GPU hang

ProblemType: Bug
DistroRelease: Ubuntu 26.10
Package: linux-image-7.3.0-10-generic 7.3.0-10.10
ProcVersionSignature: Ubuntu 7.3.0-10.10-generic 7.3.0
Uname: Linux 7.3.0-10-generic x86_64
ApportVersion: 2.36.0-0ubuntu1
Architecture: amd64
AudioDevicesInUse:
 USER        PID ACCESS COMMAND
 /dev/snd/controlC0:  tmpuser    2014 F.... pipewire
                      tmpuser    2017 F.... wireplumber
 /dev/snd/seq:        tmpuser    2014 F.... pipewire
CasperMD5CheckResult: pass
CurrentDesktop: ubuntu:GNOME
Date: Tue Nov 17 00:44:16 2026
InstallationDate: Installed on 2026-11-08 (8 days ago)
InstallationMedia: Ubuntu 26.10 "Stonking Stingray" - Official amd64 (20261013)
MachineType: Default string Default string
ProcEnviron:
 LANG=zh_CN.UTF-8
 PATH=(custom, no user)
 SHELL=/bin/bash
 TERM=xterm-256color
 XDG_RUNTIME_DIR=<set>
ProcFB: 0 i915drmfb
ProcKernelCmdLine: BOOT_IMAGE=/boot/vmlinuz-7.3.0-10-generic 
root=UUID=e6826553-80c5-446a-81cf-2bb127ba5f47 ro 
crashkernel=2G-4G:320M,4G-32G:512M,32G-64G:1024M,64G-128G:2048M,128G-:4096M
SourcePackage: linux
UpgradeStatus: No upgrade log present (probably fresh install)
dmi.bios.date: 11/09/2025
dmi.bios.release: 5.32
dmi.bios.vendor: American Megatrends International, LLC.
dmi.bios.version: 5.32
dmi.board.asset.tag: Default string
dmi.board.name: Default string
dmi.board.vendor: Default string
dmi.board.version: Default string
dmi.chassis.asset.tag: Default string
dmi.chassis.type: 3
dmi.chassis.vendor: Default string
dmi.chassis.version: Default string
dmi.modalias: 
dmi:bvnAmericanMegatrendsInternational,LLC.:bvr5.32:bd11/09/2025:br5.32:svnDefaultstring:pnDefaultstring:pvrDefaultstring:rvnDefaultstring:rnDefaultstring:rvrDefaultstring:cvnDefaultstring:ct3:cvrDefaultstring:skuDefaultstring:pfaDefaultstring:
dmi.product.family: Default string
dmi.product.name: Default string
dmi.product.sku: Default string
dmi.product.version: Default string
dmi.sys.vendor: Default string

** Affects: linux (Ubuntu)
     Importance: Undecided
         Status: New


** Tags: amd64 apport-bug stonking wayland-session

** Attachment added: "/sys/class/drm/card1/error from test C"
   
https://bugs.launchpad.net/bugs/2174625/+attachment/6018501/+files/i915-error-state-0027.txt

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2174625

Title:
  GPU hang (ccs0) importing a 10-bit 4-tiled scanout framebuffer as a
  VA-API surface, then running VPP on it

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2174625/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to