Today my notebook (AMD Renoir iGPU + Nvidia GTX 1650 dGPU) did not resume from suspend. After a hard reboot, I checked journalctl for errors, and found this (not sure it is directly related, though):
I have the exact same issue with Fedora 39 running default Gnome 45 on Wayland using the 545.29.06 driver.
Basically any screen connected to the NVIDIA dGPU works when I’m at the login screen and remains working after log on.
However when I try to connect a screen to the NVIDIA dGPU after I’ve logged in, the screen will remain black and I’ll get the same error as the OP:
This means I have to keep my external screen turned on forever from the moment it lights up at the login screen.
I cannot suspend my laptop, switch inputs on the monitor or have it turned off by my screensaver.
Basically anything that causes the screen to disconnect and reconnect will trigger the kernel panic and turns the screen black.
When I go back to the login screen it will light up again (so I’ll have to lose all of my work everytime this happens…)
[76732.070625] ==================================================================
[76732.070627] BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
Hi just wanted to clarify that the black screen when the monitor disconnects and reconnects was actually caused by an unofficial patch for Gnome’s mutter.
I think the kernel panic which I was experiencing when a connecting a new screen is solved by applying this patch:
At least it didn’t happen so far after applying the patch.
This is happening on NixOS 24.11 unstable as well, on my HP Omen 15 (2020) en0010ca.
AMD+Nvidia hybrid graphics, running nvidia version 555.58.02, Gnome 46.
The patch doesn’t work, neither does nvidia-drm.modeset=1 nvidia-drm.fbdev=1
EDIT: I should note that my computer freezes when this happens, and it triggers whenever I plug in my external monitor, no matter which port (both are connected to the dGPU)
I just opened this based on your comment as it is exactly my experience, but not sure if it is the same issue as the actual subject of this thread, and @hubofeverything seems to imply NVIDIA has already fixed this, but have just not released the fix yet except via manual installation. For those of us who can only load NVIDIA drivers from the repos this is a shame, because the crashes lose work, data and time each time. We are getting around a crash every hour or 2, how about you? Is there any way you have found to reduce the frequency?
@hubofeverything - if you are still watching this thread, could you please confirm that the you are still doing fine with 560.31.02 so NVIDIA (do they watch these threads?) have this assurance they have already fixed it and can perhaps be tempted to release some version with this fix to some repo (so there is no manual steps required for normal users or people who for any reason do not use alternative installation/update methods outside of what their distro provides by default)?
I’ve done some additional testing, and the result is pretty interesting.
I’m running nixos 24.05, and to fix the issue, I’ve used the zen kernel from the unstable branch instead of the stable branch.
I’m not entirely sure, but previous testing may have suggested that this issue is only mitigated when I’m on the kernel on the nixos-unstable branch and R560, and not either or.
The kernel version for nixos-24.05 is 6.10.1, and for nixos-unstable it’s 6.10.5, so I don’t think that’s the main reason.
However, there’s some kernel config changes between the two. Maybe it’s some sort of combination of a crash that involves a kernel config?
… wrt “either or” we stay uptodate with latest fedora 40 kernels/nvidia drivers (currently at 6.10.5-200.fc40.x86_64/555.58.02) as soon as they are available in the main repos and have had this problem for a couple of weeks, with various combinations.
In fact, this is the 2nd time I am trying to reply as the crash/reboot happened again, in the middle of the first try!!
… thanks @hubofeverything for causing me to find out that NVIDIA are finally able to move to open drivers!! (BTW, system still crashes/reboots with same BUG: KFENCE with latest Fedora kernal update 6.10.6-200.fc40.x86_64 with R550). Here is an article about the move top open drivers for those like me who do not follow such things as long as everything seems to be working (which, of course, is the case usually these days).
IIUC, version R560 is when the open-drivers will come into effect for GeForce, so that is good news and hopefully, eventually, just a standard Fedora dnf update will fix this pretty serious problem.
In the meantime, following the instructions given in the article for Fedora, sudo dnf module install nvidia-driver:open-dkms gives
Error: Problems in request:
missing groups or modules: nvidia-driver:open-dkms
Once I work out what this means and what to do about it (all hints/advice very welcome!), happy to try out anything that helps the devs here test stuff on my specific setup (as long as you have the patience to give step by step instructions, as I do not develop in this domain).
Anyway, congratulations to NVIDIA, this seems like a good move and sustainable policy for staying ahead!
Here with the same output from dmesg. Fedora 40 with 555.58.02 on GeForce RTX 3050 Ti.
This bug is quite dangerous in Blender as there is a small chance of any mouse movement from the viewport attempting to render the scene to then cause the system to freeze (if Blender is installed as package) or the application to freeze (if Blender is installed as a Flatpak). I have been able to reproduce this consistently in Blender 4.2.0. I set a box with lights with the render engine set to Cycles with noise threshold set to 0.99 and viewport pixel size set to 8x. This setup maximizes the chance of it being triggered. Simply move the camera for around a minute.
Using Wayland seems to be important as I have not been able to reproduce the issue on X11. Can anyone verify this?
A temporary fix is to downgrade from 555 to 470 if you cannot upgrade above 555 yet. It seems to not happen on 470.
If one has been already made, are there any patches for this issue on 555 that anyone can verify works?
BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
Use-after-free read at 0x00000000a1480d4d (in kfence-#248):
nv_dma_release_sgt+0x29/0x70 [nvidia]
@robin.dusky since this is reproducible on the latest feature branch release (560.35.03), it would be nice to report it on the official feedback thread for 560.x drivers (if you haven’t done already), so they can properly track it. I’m still on 555.58.02, and for my daily use, it is performing well, so I will delay the upgrade to 560.x as much as I can, hoping it will stabilize (they’re making good progress as far as I can see, but there’s still some ground to cover).
Aug 29 13:09:12 fedora kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
Aug 29 13:09:12 fedora kernel: Use-after-free read at 0x000000005b920d10 (in kfence-#12):
Aug 29 13:09:12 fedora kernel: nv_dma_release_sgt+0x29/0x70 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv043722rm+0x67/0xd0 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv038493rm+0xc7/0x430 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv038458rm+0x6b/0x130 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv004292rm+0xd/0x20 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv006122rm+0x1b/0xb0 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv018080rm+0x59c/0x680 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv048181rm+0xb3/0xe0 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv049956rm+0xb0/0x180 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv049955rm+0x50b/0x660 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv048073rm+0xdd/0x190 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv048074rm+0x41/0x70 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv000587rm+0x4a/0x60 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv000745rm+0x20f/0xe00 [nvidia]
Aug 29 13:09:12 fedora kernel: rm_ioctl+0x7f/0x400 [nvidia]
Aug 29 13:09:12 fedora kernel: nvidia_unlocked_ioctl+0x53b/0x8d0 [nvidia]
Aug 29 13:09:12 fedora kernel: __x64_sys_ioctl+0x94/0xd0
Aug 29 13:09:12 fedora kernel: do_syscall_64+0x82/0x160
Aug 29 13:09:12 fedora kernel: entry_SYSCALL_64_after_hwframe+0x76/0x7e
Aug 29 13:09:12 fedora kernel:
Aug 29 13:09:12 fedora kernel: kfence-#12: 0x00000000cd93353a-0x00000000f97c0ada, size=384, cache=kmalloc-512
Aug 29 13:09:12 fedora kernel: allocated by task 3668 on cpu 11 at 1188.240062s:
Aug 29 13:09:12 fedora kernel: nv_drm_gem_prime_import_sg_table+0x2d/0xb0 [nvidia_drm]
Aug 29 13:09:12 fedora kernel: drm_gem_prime_import_dev+0x93/0x180
Aug 29 13:09:12 fedora kernel: drm_gem_prime_fd_to_handle+0xe7/0x220
Aug 29 13:09:12 fedora kernel: drm_ioctl_kernel+0xb0/0x100
Aug 29 13:09:12 fedora kernel: drm_ioctl+0x28b/0x540
Aug 29 13:09:12 fedora kernel: __x64_sys_ioctl+0x94/0xd0
Aug 29 13:09:12 fedora kernel: do_syscall_64+0x82/0x160
Aug 29 13:09:12 fedora kernel: entry_SYSCALL_64_after_hwframe+0x76/0x7e
Aug 29 13:09:12 fedora kernel:
Aug 29 13:09:12 fedora kernel: freed by task 3668 on cpu 29 at 1188.254434s:
Aug 29 13:09:12 fedora kernel: nv_dma_release_sgt+0x49/0x70 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv043722rm+0x67/0xd0 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv038493rm+0xc7/0x430 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv038458rm+0x6b/0x130 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv004292rm+0xd/0x20 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv006122rm+0x1b/0xb0 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv018080rm+0x59c/0x680 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv048181rm+0xb3/0xe0 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv049956rm+0xb0/0x180 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv049955rm+0x50b/0x660 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv048073rm+0xdd/0x190 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv048074rm+0x41/0x70 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv000587rm+0x4a/0x60 [nvidia]
Aug 29 13:09:12 fedora kernel: _nv000745rm+0x20f/0xe00 [nvidia]
Aug 29 13:09:12 fedora kernel: rm_ioctl+0x7f/0x400 [nvidia]
Aug 29 13:09:12 fedora kernel: nvidia_unlocked_ioctl+0x53b/0x8d0 [nvidia]
Aug 29 13:09:12 fedora kernel: __x64_sys_ioctl+0x94/0xd0
Aug 29 13:09:12 fedora kernel: do_syscall_64+0x82/0x160
Aug 29 13:09:12 fedora kernel: entry_SYSCALL_64_after_hwframe+0x76/0x7e
Aug 29 13:09:12 fedora kernel:
Aug 29 13:09:12 fedora kernel: CPU: 29 PID: 3668 Comm: gnome-shell Tainted: P B W OE 6.10.6-200.fc40.x86_64 #1
If this is not what you need, please advise specific steps and will try to get back quickly.
NB this one did not cause a reboot but often it does (or appears to, as there is no error recorded after it but before crash).
Thks for reaching out @amrits . I still see the use-after-free error on the logs, even with 560.35.03, but in my case at least, they don’t seem to cause much problem:
ago 29 08:42:20 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 21:00:38 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 20:58:26 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 16:40:05 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 16:32:50 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 16:03:12 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 16:02:58 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 15:57:19 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 15:51:49 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 15:50:46 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 15:50:00 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 15:45:15 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 14:24:48 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 11:56:33 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 11:55:15 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 11:40:16 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 11:31:53 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 11:20:32 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 11:15:43 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 11:13:37 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 11:10:42 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 11:09:14 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 10:52:58 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 10:47:36 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 10:33:47 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 10:20:53 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 10:19:31 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 10:10:47 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 09:33:04 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 09:28:28 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 09:27:34 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 09:20:50 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 09:20:46 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 09:17:45 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 09:17:42 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 28 09:17:39 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
ago 27 19:01:35 fedoracosta kernel: BUG: KFENCE: use-after-free read in nv_dma_release_sgt+0x29/0x70 [nvidia]
I don’t have a concrete use case to report, but it seems somewhat related to suspend/resume (and 99% of the time I have an external monitor plugged to the GTX 1650 via HDMI). Scratch that, it happens much more frequently than the usual suspend/resume cycles, there must be some other cause.