AI DISCLOSURE
This report was generated by an AI assistant (Codex) from logs and interactive
testing performed by me. I supplied the system exports and observed the
reported failures and passing tests. The assistant organized the evidence and
proposed interpretations; the underlying root cause has not been verified.
SUMMARY
Scrolling Instagram Reels in accelerated Google Chrome on native Wayland
reproducibly makes my Niri graphical session unusable. NVIDIA logs an
NvKmsKapiMemory mapping failure, a burst of Xid 32 across multiple GPU clients,
and a Chrome Media thread invalid-opcode trap. A TTY remains accessible,
although sometimes laggy. Closing Chrome and restarting Niri did not restore
usable graphics; a normal reboot from a working TTY restored operation.
Disabling hardware video decoding alone still reproduces this in a separate
Chrome profile with extensions disabled. Chrome through XWayland under the
same Niri session passes. Firefox native Wayland with GPU WebRender and
software video decoding also passes. The exact faulty component is unknown.
ENVIRONMENT (2026-11-01)
- Arch Linux, kernel 7.2.11-arch1-1.
- Gigabyte NVIDIA GeForce GTX 1060 3GB, GP106/Pascal, PCI 10de:1c02.
- Proprietary NVIDIA 580.196.07; nvidia-580xx-dkms 580.196.07-1;
nvidia-580xx-utils 580.196.07-1. - Google Chrome 156.0.8078.79; AUR google-chrome 156.0.8078.79-1.
This packages Google Chrome’s binary. Arch’s Chromium was not tested. - Niri 26.04 (8ed0da4), Noctalia 5.4.0-1.
- The same user-observed symptom also occurred previously on Hyprland.
Its version, driver/browser versions and matching logs were not recorded
for this report; this is historical context, not a controlled comparison. - xwayland-satellite 0.8.3-1; xorg-xwayland 24.1.14-1.
- Chrome uses ANGLE OpenGL; Vulkan disabled in the supplied GPU exports.
- Firefox comparison: 159.0, native Wayland, GPU WebRender; hardware video
decoding unavailable (FEATURE_FAILURE_VIDEO_DECODING_TEST_FAILED).
REPRODUCTION
- Start a Niri Wayland session with Noctalia using the proprietary driver.
- Run Chrome natively on Wayland with graphics acceleration enabled.
- Open Instagram, open a Reel, repeatedly scroll to subsequent Reels.
- A tab may freeze first; continued attempts can leave the graphical session
unusable. Initially this happened after about two Reels, but timing varies.
Exact Reel counts were not recorded in controlled runs.
The controlled decoder-disabled failing run used:
google-chrome-stable --user-data-dir=“$HOME/gpu-report/profiles/chrome” --no-first-run --disable-extensions --enable-logging=stderr --ozone-platform=wayland --disable-accelerated-video-decode https://www.instagram.com/
This profile was separate from my everyday profile. chrome://gpu confirmed
hardware compositing and software-only video decoding. The same test profile
was used for the passing XWayland run.
EXPECTED / ACTUAL
Expected: normal playback and scrolling without disrupting the desktop.
Actual: tab/media failure and NVIDIA errors across multiple clients; pointer
freeze or severe desktop stalls. Closing Chrome and restarting Niri did not
restore usable graphics. TTY access could work. Older attempts used SysRq
S/U/B; fresh attempts could be recovered by normal reboot from a working TTY.
COMPARISONS
- Chrome native Wayland, GPU acceleration enabled: FAIL. A frozen tab at
15:31:43 was followed by a GPU-wide failure and 99 observed Xid 32 entries
during a subsequent attempt at 15:34:59. - Chrome native Wayland, GPU compositing on, hardware video decoding off:
FAIL. NvKms mapping failure and 109 observed Xid 32 entries at 16:19:01,
followed by Xid 38. Separate profile, extensions disabled. - Chrome, all GPU acceleration off: PASS, approximately 15 minutes of
scrolling. No GPU export was saved for this control. - Chrome --ozone-platform=x11, GPU acceleration on: PASS. No reported lag
or matching kernel GPU faults. Approximately 35.5 minutes elapsed since
launcher start, including setup; actual scrolling duration was not timed.
This was XWayland inside Niri, not a full Xorg desktop. Hardware video
decoding was advertised; the actual per-video decoder was not captured. - Firefox native Wayland, GPU WebRender, software video decoding: PASS.
No reported lag or matching GPU faults during the approximate test interval.
Exact scrolling duration and Reel count were not recorded.
KEY LOG EVENTS (2026-11-01, CET / UTC+01:00)
15:31:43.115842 nvidia-drm: Failed to map NvKmsKapiMemory
15:31:43.146839 traps: Media[34042] trap invalid opcode … in chrome
15:34:59.087845 nvidia-drm: Failed to map NvKmsKapiMemory
15:34:59.090956 NVRM: Xid (PCI:0000:01:00): 32, … intr 00004000
15:34:59.108988 traps: Media[35292] trap invalid opcode … in chrome
16:19:01.018899 nvidia-drm: Failed to map NvKmsKapiMemory
16:19:01.023956 NVRM: Xid (PCI:0000:01:00): 32, … name=niri … intr 00004000
The complete NVIDIA report was collected after the first fresh failure and
before reboot. During its diagnostic probes, vulkaninfo was associated with
later Xid 38 at 15:41:00 and Xid 31 at 15:42:10. These later diagnostic events
should not be mistaken for the original Instagram trigger. A Niri page-flip
permission error around TTY switching may be incidental.
A monitoring heartbeat continued after the failure with approximately 13 GB
MemAvailable. This does not rule out GPU memory or PCIe problems.
Older October logs show similar Xid 32 errors on 580.178.04, kernel
7.2.8-arch1-2 and Chrome 154.0.8037.97. The exact historical trigger is
unverified; this is not a controlled driver-regression comparison.
WORKAROUND / LIMITATIONS
Chrome with --ozone-platform=x11 retains GPU acceleration and passed testing
so far. Firefox in its default native Wayland configuration also passed.
Finite passing tests do not establish permanent stability. The comparisons
narrow the trigger toward Chrome’s accelerated native Wayland path; they do
not establish whether NVIDIA, Chromium/ANGLE, Niri or their interaction is
responsible. Xid 32 also leaves PCIe/hardware causes open. No driver, kernel,
firmware or hardware settings were changed during these comparisons.
ATTACHMENTS
Filtered kernel faults, sanitized Chrome GPU exports, Firefox graphics
summary and relevant package versions are prepared in a small public bundle.
A complete nvidia-bug-report.log.gz from the failed state exists locally.
It contains identifying details and needs separate review before public
upload; it is intentionally absent from the public bundle. Raw originals and
full chronological investigation notes have been retained locally.
POTENTIALLY RELATED (not confirmed duplicates)
Pascal/580/Chromium Wayland Xid 32 report with a different initial Xid:
Broader Wayland mapping-failure thread, different GPUs/configurations:
REQUEST
Can NVIDIA investigate this reproducible native-Wayland trigger on Pascal/580
and advise which targeted diagnostics would distinguish buffer/mapping path
failures from PCIe/hardware problems? Is this related to an existing issue?
public-evidence.tar.gz (9.0 KB)