Bug Report: Wayland-only: dmaAllocMapping_GM107 VA mapping failures (does not reproduce on X11)

Driver versions affected: 580.105.08 and earlier versions (issue reproducible across multiple driver/kernel combinations)
Kernel: 6.17.9-2-cachyos (also observed on earlier kernels)
GPU: GeForce RTX 3080 Founders Edition (GA102, 10 GiB)
Driver type: nvidia-open kernel modules
Desktop session: KDE Plasma Wayland
Application affected: Jellyfin “Delfin” Electron client (NVDEC-accelerated video playback), Various GPU-intensive workloads (e.g., gaming)
ReBAR: Tested enabled and disabled — issue occurs in both cases
Bug report log: nvidia-bug-report.log.gz (506.8 KB)

Summary
The NVIDIA driver eventually begins emitting repeated kernel messages of the form:

NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.

This occurs reliably after ~30–45 minutes of NVDEC hardware-accelerated video playback, and also intermittently during other GPU-intensive workloads such as gaming.

Impact differs by workload:

  • During video playback: the video output turns blank, and the application must be restarted, but the desktop remains responsive.

  • During heavier GPU workloads (e.g., games): the same error sometimes leads to noticeable stalls, and in some cases a full system freeze (impossible to switch to another TTY).

The issue has been reproduced across multiple NVIDIA driver versions and multiple kernel versions, with and without ReBAR enabled.

No Xid codes are generated — only the repeated VA-space allocation failure message.

Reproduction Steps (most reliable via NVDEC)

  1. Start KDE Plasma Wayland session.

  2. Launch HW accelerated video playback (for example with Delfin, a Jellyfin client).

  3. Ensure NVDEC hardware acceleration is enabled.

  4. Play any long video.

  5. After ~30–45 minutes:

    • Video freezes and goes blank.

    • Application must be restarted.

    • Kernel logs begin spamming:

      NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.
      

Observed using:

sudo journalctl -k -f -g NVRM

Observed Behaviour

  • Playback failure and blank video output during NVDEC decoding.

  • Kernel repeatedly logs VA-space allocation errors.

  • No Xid codes or GPU resets logged.

  • Same log message appears during gaming or other heavy GPU tasks, where it may cause stalls or full system freezes.

  • Problem persists:

    • With ReBAR enabled (large BAR1 aperture).

    • With ReBAR disabled (classic 256 MiB BAR1).

    • Across several driver and kernel versions.

  • VRAM usage remains stable; issue appears unrelated to memory pressure.

Update: Issue also occurs with software decoding + does NOT reproduce under X11

I have additional information after further testing:

1. The dmaAllocMapping_GM107: can't alloc VA space for mapping error also occurs with hardware decoding completely disabled.

  • I disabled hardware video acceleration in the application.

  • I verified via nvidia-smi dmon -s u that NVDEC usage remained at 0% throughout playback.

  • After a longer period (larget timescale than previous tests, though I’m now inclined to think this is random), the same error eventually appeared in the kernel log.

This suggests the issue is not tied to NVDEC.

2. The issue does not reproduce under a Plasma (X11) session.

I switched from Plasma Wayland to Plasma X11 and tested:

  • multiple hours of continuous video playback

  • gaming workloads

  • other GPU-intensive tasks

The error never appeared, and no playback stalls or freezes occurred.

This strongly suggests that the issue is triggered specifically under Wayland.

3. Summary of updated findings

  • Error occurs even with software decoding only - unrelated to NVDEC.

  • Error occurs under Plasma Wayland during normal usage (video playback or gaming).

  • Error does NOT occur on X11, even after extended stress testing.

  • System remains stable on X11; no VA mapping failures observed.

Let me know if there is any additional logging I can enable to help isolate the problem

I see this as well with a similar setup (Arch 6.12.61-1-lts, Driver 580.105.08, GPU 5070 Ti, nvidia-open), and have also noticed it happening even without NVDEC issues.

Another trigger I’ve found is certain webpages in Chrome 143, even with hardware decoding forcibly disabled, such as this one (the most recent one that crashed my system): https://www.nokia.com/broadband-access/in-home-connectivity/home-wi-fi/wi-fi-beacon-g6-2/

I have reported this multiple times. Nvidia seem determined to ignore it.

Hi All,

Apologize for the delayed response, I have filed a bug 5762513 internally for tracking purpose.

Team will try to duplicate issue in lab and will get back if required any additional information.

I have the same issue while playing Hytale on wayland.

NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from pReuseMappingDb->pMapCb(pReuseMappingDb->pGlobalCtx, pAllocCtx, range, cachingFlags, &token, _reusemappingdbAddMappingCallback) @ mapping_reuse.c:273
NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.
NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.
NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.
NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.

It gets absolutely spammed, that even the journal can’t keep up.

$ journalctl -b -g NVRM | wc -l
3950544
$ journalctl -b -u systemd-journald
/dev/kmsg buffer overrun, some messages lost. (Dropped 24 similar message(s))
/dev/kmsg buffer overrun, some messages lost.
/dev/kmsg buffer overrun, some messages lost.
Missed 483 kernel messages
Missed 13 kernel messages
Missed 425 kernel messages
Missed 1124 kernel messages

Additional reproduction data - RTX 3090 Ti on Fedora 43 + COSMIC Wayland

I can confirm this issue occurs on driver 580.126.09 (latest production release as of Jan 2026), indicating the kvmalloc() memory allocation fix in this release does not address this specific bug.

System Configuration:

  • GPU: NVIDIA GeForce RTX 3090 Ti (GA102, 24GB)
  • Driver: 580.126.09 (proprietary, DKMS install)
  • Kernel: 6.18.5-200.fc43.x86_64
  • OS: Fedora 43
  • Desktop: COSMIC (System76 Wayland compositor)
  • Browser: Chromium

Reproduction:

  • Simply loading a webpage with media content (Nokia product page with embedded content)
  • Crash occurs within seconds, not requiring extended playback

Observed errors (journalctl):

[drm:__nv_drm_gem_nvkms_map [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000100] Failed to map NvKmsKapiMemory 0x000000009ebe4fc4
traps: Media[19678] trap invalid opcode ip:560a118419d6 sp:7f2c67f0c430 error:0 in chromium-browser

Impact:

  • Chrome’s GPU/Media process crashes with illegal instruction trap
  • Extended occurrence leads to full system freeze requiring hard reboot
  • Issue is intermittent but reproducible within normal browsing session

Notes:

  • Issue did NOT occur on driver 580.119.02 with akmod-nvidia (RPM Fusion) initially, but manifested after extended uptime
  • Upgraded to 580.126.09 hoping kvmalloc() fix would help - issue persists
  • ReBAR status: enabled in BIOS

Bug report attached: nvidia-bug-report-20260123-222838.log.gz


Files to attach:

  1. nvidia-bug-report-20260123-222838.log.gz (1.8 MB)

Can you confirm my report is related or different issue and if so kindly file a separate ticket?

welp I’m now getting it as well. it came out from absolutely nowhere.
been locked on 580.82.09 drivers for months, and now running gamescope -–backend sdl – command triggers this in dmesg:
[ 7587.307062] NVRM: dmaAllocMapping_GM107: Virtual address 0x6E30000 is not compatible with page size 0x200000 or page offset 0x0.
[ 7587.307273] gamescope-xwm[161783]: segfault at 80 ip 000055cffbced49e sp 00007efe42c78070 error 4 in gamescope[7449e,55cffbcaf000+382000] likely on CPU 12 (core 6, socket 0)
[ 7587.307280] Code: 00 48 29 c1 48 89 c8 48 89 55 e0 48 89 45 e8 48 8b 45 e0 48 8b 55 e8 48 89 10 90 90 c9 c3 55 48 89 e5 48 89 7d f8 48 8b 45 f8 <48> 8b 00 5d c3 90 55 48 89 e5 48 89 7d f8 48 8b 45 f8 48 8b 00 48

I also have this issue, though I just found this thread, I tried to report it in 580 and 590 feedback but was ignored.

I’m able to reproduce it using the nvidium mod, maintained fork, setting the memory limit to 8192 or above causes it, under that does not. I have a 3090 with 24GB of vram.

Here’s the two associated threads of other people with the same issue:

I was able to confirm that this issue occurs with both open and closed drivers, and was introduced in 580.126.09, the prior version 580.119.02 or 570/575 do not have this issue, and the latest 590.48.01 still has it.

[  +0.000003] NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.
[  +0.000003] NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from pReuseMappingDb->pMapCb(pReuseMappingDb->pGlobalCtx, pAllocCtx, range, cachingFlags, &token, _reusemappingdbAddMappingCallback) @ mapping_reuse.c:274
[  +0.002012] NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from reusemappingdbMap(&pBar1VaInfo->reuseDb, pMappingType, mapRange, pMemArea, cachingFlags) @ kern_bus_gm107.c:3187
[  +0.003045] NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.
[  +0.000031] NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from pReuseMappingDb->pMapCb(pReuseMappingDb->pGlobalCtx, pAllocCtx, range, cachingFlags, &token, _reusemappingdbAddMappingCallback) @ mapping_reuse.c:274
[  +0.000281] NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.

I can also reproduce it by running any local LLM which uses more than 8GB of vram, errors in dmesg and very slow startup, keeps trying to allocate until it succeeds which can take several minutes depending on how much vram.

I used to have this problem only sometimes when using NVDec on 580/590 drivers and only in certain applications (for example freetube would lock up and spam the following in the logs):

Apr 06 12:09:55 balrog kernel: NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.
Apr 06 12:09:55 balrog kernel: NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from pReuseMappingDb->pMapCb(pReuseMappingDb->pGlobalCtx, pAllocCtx, range, cachingFlags, &token, _reusemappingdbAddMappingCallback) @ mapping_reuse.c:273
Apr 06 12:09:55 balrog kernel: NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from reusemappingdbMap(&pBar1VaInfo->reuseDb, pMappingType, mapRange, pMemArea, cachingFlags) @ kern_bus_gm107.c:3186

However, since upgrading to 595 drivers the system is completely unusable. It happens with everything! Even opening a terminal fails (kitty opens but with a blank screen and hangs). All of this right after boot. Reverting back to 590 drivers makes (almost) everything functional again.


In my case the system is:

Arch Linux (6.19.9-arch1-1) / Hyprland (0.54.3-2) / GeForce RTX 3090

On 590.48.01-17 (rarely affected)
On 595.58.03-03 (everything crashing)

If it’s relevant, beyond normal set of apps running I do have whisper and sunshine loaded on boot (sunshine may initialize nvdec not sure, while whisper may do something similar with cuda?).

> nvidia-smi                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     
Mon Apr  6 15:48:35 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 590.48.01              Driver Version: 590.48.01      CUDA Version: 13.1     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 3090        Off |   00000000:01:00.0  On |                  N/A |
| 53%   40C    P0            110W /  350W |    2809MiB /  24576MiB |      5%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            1834    C+G   whisper-server                          226MiB |
|    0   N/A  N/A            2086      G   Hyprland                                424MiB |
|    0   N/A  N/A            2162      G   /usr/bin/64gram-desktop                   4MiB |
|    0   N/A  N/A            2163    C+G   /usr/bin/python                          87MiB |
|    0   N/A  N/A            2168      G   /usr/lib/firefox/firefox                221MiB |
|    0   N/A  N/A            2169      G   /usr/lib/firefox/firefox                260MiB |
|    0   N/A  N/A            2179      G   kitty                                    92MiB |
|    0   N/A  N/A            2217      G   Xwayland                                  6MiB |
|    0   N/A  N/A            2432    C+G   /usr/bin/swaync                         150MiB |
|    0   N/A  N/A            3208      G   /usr/lib/electron40/electron             63MiB |
|    0   N/A  N/A            3699      G   ...signal-desktop/signal-desktop         44MiB |
|    0   N/A  N/A            4685      G   ...share/Steam/ubuntu12_32/steam          8MiB |
|    0   N/A  N/A            5519      G   /opt/Heroic/heroic                       65MiB |
|    0   N/A  N/A            6209      G   ./steamwebhelper                         48MiB |
|    0   N/A  N/A            6249      G   ...am/ubuntu12_64/steamwebhelper         74MiB |
|    0   N/A  N/A            6311    C+G   /usr/bin/sunshine                       262MiB |
|    0   N/A  N/A            6421      G   /usr/lib/electron38/electron             59MiB |
|    0   N/A  N/A          597626      G   kitty                                   124MiB |
+-----------------------------------------------------------------------------------------+

I can reliably reproduce this issue with a specific trigger.

System:

  • GPU: GeForce RTX 3060 12GB
  • Driver: 595.58.03 (latest production branch)
  • Kernel: 6.19.12-1-cachyos
  • OS: CachyOS based on Arch Linux
  • Desktop: KDE Plasma 6.6.4 Wayland
  • Browser: Chromium-based (tested both Helium Browser and Google Chrome - same result)

Reproduction:

Navigating to https://m3.material.io/get-started reliably triggers the crash every time. The page loads briefly, then the tab freezes and the renderer process crashes with SIGILL. No extended usage or waiting required, it happens within seconds of the page loading.

Another page on the same site - https://m3.material.io/styles/shape/overview-principles - goes further and freezes the entire system, requiring a hard reboot. Unlike the get-started page which only crashes the renderer process, this one escalates to a full system hang immediately.

Relevant dmesg output:

kernel: NVRM: dmaAllocMapping_GM107: can't alloc VA space for mapping.
kernel: NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from pReuseMappingDb->pMapCb(pReuseMappingDb->pGlobalCtx, pAllocCtx, range, cachingFlags, &token, _reusemappingdbAddMappingCallback) @ mapping_reuse.c:274
kernel: resource: resource sanity check: requesting [mem 0x00000000bf4f0000-0x00000000c02fffff], which spans more than 0000:01:00.0 [mem 0xb0000000-0xbfffffff 64bit pref]
kernel: caller __nv_drm_gem_nvkms_map+0x94/0x110 [nvidia_drm] mapping multiple BARs
kernel: hrtimer: interrupt took 18635980 ns
kwin_wayland[1420]: Pageflip timed out! This is a bug in the nvidia-drm kernel driver

Coredump from renderer process:

Signal: 4 (ILL)
Stack trace of thread:
#0  0x0000562c191f5458 n/a (n/a + 0x0)

What I’ve tested:

  • --disable-gpu → no crash (confirms GPU path is the trigger)
  • Firefox with GPU acceleration → works fine on the same page

Additional findings:

All <video> elements on affected pages use preload="auto" autoplay muted loop, causing Chromium to decode all videos simultaneously on page load. The main page https://m3.material.io has 1 video and works fine. The https://m3.material.io/get-started page has 10 simultaneous autoplay videos and crashes the renderer. The https://m3.material.io/styles/shape/overview-principles page has 7 simultaneous autoplay videos and freezes the entire system. Firefox handles all three pages without issues.

Happy to provide any additional logs or run diagnostic commands if it helps narrow this down. The fact that this is 100% reproducible on a specific URL should hopefully make it easier to debug on your end.
nvidia-bug-report.log.gz (362.3 KB)