Jun 24 14:08:47 transmission1 kernel: [TTM] Buffer eviction failed
Jun 24 14:08:47 transmission1 kernel: qxl 0000:00:02.0: object_init failed for (4096, 0x00000001)
Jun 24 14:08:47 transmission1 kernel: [drm:qxl_alloc_bo_reserved [qxl]] *ERROR* failed to allocate VRAM BO
Jun 24 14:09:03 transmission1 kernel: [TTM] Buffer eviction failed
Jun 24 14:09:03 transmission1 kernel: qxl 0000:00:02.0: object_init failed for (4096, 0x00000001)
Jun 24 14:09:03 transmission1 kernel: [drm:qxl_alloc_bo_reserved [qxl]] *ERROR* failed to allocate VRAM BO
Jun 24 14:09:18 transmission1 kernel: [TTM] Buffer eviction failed
running journalctl my logs are full of these messages. Googling this error goes back as far as 2016-2017 era and it seems to have to do with qemu and video/spice or something.
People mention solving this by changing video modes to virtio, but it doesn’t seem to be an option using the TrueNAS gui.
I’ve got fresh installs of ubuntu 22.04 and 24.04 and the issue occurs on both.
I have this issue too. Have been wondering why my VM is crashing. I think it may be this. A bit of googling and this seems to be to do with KVM and QXL which it seems is another name for Spice. The workaround most are providing is to use something other than QXL which I doubt is possible on TrueNAS but could be wrong.
Interestingly I CAN still ssh to the VM, but at least the app itself (and I assume other things) are not able to function properly.
I have logged a ticket for this here, I would encourage you to go and vote for it with a me too or something and add your details so that IX has all the information they need.
As a workaround / test I have completely deleted the display device in the virtual machines devices section - I don’t usually need it anyway. Will see if that helps to confirm if it is indeed the QXL / Spice setup causing the crash.
As a bit of an update, it’s been a few days and it’s been rock solid since I deleted the display from the VM config. More time will need to pass to truly tell, but so far it looks like this is the issue.
It seems IX systems has elected to say they don’t care about our VM’s crashing. Perhaps it’s only a few of us but I thought that given where we isolated it to, they would be a little more interested that just quoting amateur hour upstream fix required.
I’d suggest log a ticket. And maybe say that if an upstream fix is required to make ‘your’ product work, to submit one rather than asking customers to become developers like the ticket above. Or perhaps you’ll get a different developer that isn’t so amateur. Some months down the track, I can confirm all errors have ceased since disabling QXL / Spice driver.
They have mentioned dropping SPICE in Fangtooth and going back to VNC.
Source is iX’s kris in the T3:
He gets into the SPICE specific bits at around 18:14 but I recommend you start listening at the linked timestamp or a bit earlier if you want more context to the discussion.
I still have these errors, also when using journalctl and over ssh session.
AFAICT from the web and the error, it’s a VRAM issue:
```
Dec 03 09:14:44 host.domain.org kernel: [drm:qxl_alloc_bo_reserved [qxl]] ERROR failed to allocate VRAM BO
Dec 03 09:14:44 host.domain.org kernel: qxl 0000:00:02.0: object_init failed for (4096, 0x00000001)
Dec 03 09:14:44 host.domain.org kernel: [TTM] Buffer eviction failed
Dec 03 09:14:29 host.domain.org kernel: [drm:qxl_alloc_bo_reserved [qxl]] ERROR failed to allocate VRAM BO
Dec 03 09:14:29 host.domain.org kernel: qxl 0000:00:02.0: object_init failed for (4096, 0x00000001)
Dec 03 09:14:29 host.domain.org kernel: [TTM] Buffer eviction failed
Dec 03 09:14:13 host.domain.org kernel: [drm:qxl_alloc_bo_reserved [qxl]] ERROR failed to allocate VRAM BO
Dec 03 09:14:13 host.domain.org kernel: qxl 0000:00:02.0: object_init failed for (4096, 0x00000001)
Dec 03 09:14:13 host.domain.org kernel: [TTM] Buffer eviction failed
```
On some hypervisor it’s possible to set an amount of VRAM for a given VM.
Not sure it’s the one thing or that it would help though, but it doesn’t seem possible on TN (25.04)
Actually, I do have the message in my logs constantly, even without using journalctl , which still fails on pressing G to get to the latest line, with that very error.
AFAIC search it comes from using Spice/QXL combo.
The only think I can think of with the VM management capacity offered by TN 25.04 is to simply remove the Display Device of my Ubuntu24.04 VM in question, which is headless anyway.
