Public bug reported:
Request:
Please SRU-backport upstream Linux commit
046045543e530605c441063535e7dca0075369a6 to the Ubuntu linux source package for
the Ubuntu 26.04 resolute 7.0 GA kernel line.
Commit:
drm/xe/pt: Reset current_op in xe_pt_update_ops_init()
Fixes: e8babb280b5e ("drm/xe: Convert multiple bind ops into single job")
The upstream commit is marked Cc: [email protected].
Affected system:
Ubuntu 26.04 resolute
Kernel: 7.0.0-31-generic
GPU: Intel Arc Pro B70, PCI ID 8086:e223, using the xe driver
Secure Boot is enabled, so a signed Ubuntu kernel is required.
Impact:
The host experiences a kernel Oops followed by soft lockups and RCU stalls.
Userland becomes unresponsive: SSH hangs, Frigate API hangs, and the host
requires a cold power-off to recover.
Trigger:
A llama.cpp SYCL VLM workload on the Arc Pro B70 causes heavy VRAM
bind/eviction activity through xe_vm_bind_ioctl. The incidents occurred on
2026-09-13 and 2026-09-15 with the same signature.
Why this commit matches:
The crash path goes through xe_vm_bind_ioctl, xe_bo_validate, ttm_bo_validate,
ttm_bo_alloc_resource, ttm_lru_walk_for_evict, ttm_bo_evict, xe_bo_move_notify,
xe_bo_trigger_rebind, and xe_pt_zap_ptes. A second Oops occurred in
ttm_resource_manager_next on the same eviction path.
The installed 7.0.0-31 xe.ko already appears to contain the earlier
xe_pt_zap_ptes_entry NULL guard, so this report is specifically
requesting the current_op retry fix in commit
046045543e530605c441063535e7dca0075369a6.
Primary Oops excerpt:
BUG: kernel NULL pointer dereference, address: 0000000000000018
CPU: 3 PID: 1919934 Comm: llama-server
Kernel: 7.0.0-31-generic
RIP: xe_pt_zap_ptes+0x97/0x110 [xe]
Call trace includes:
xe_vm_invalidate_vma
xe_bo_trigger_rebind
xe_bo_move_notify
ttm_bo_evict
ttm_lru_walk_for_evict
ttm_bo_alloc_resource
ttm_bo_validate
xe_bo_validate
xe_vm_bind_ioctl
Secondary Oops excerpt:
BUG: unable to handle page fault
CPU: 0 PID: 1942263 Comm: llama-server
RIP: ttm_resource_manager_next+0x116/0x300 [ttm]
Call trace includes:
ttm_resource_manager_first
ttm_lru_walk_for_evict
ttm_bo_alloc_resource
ttm_bo_validate
xe_bo_validate
xe_vm_bind_ioctl
Proposed fix:
Backport upstream commit 046045543e530605c441063535e7dca0075369a6 to the Ubuntu
26.04 resolute 7.0 GA linux kernel line.
Verification plan:
Boot a signed Ubuntu 7.0 kernel containing the patch.
Re-run the failing B70 SYCL VLM workload under sustained VRAM eviction pressure.
Confirm no xe_pt_zap_ptes Oops, ttm_resource_manager_next Oops, soft lockup,
RCU stall, or cold-power recovery requirement over a 24 to 72 hour soak.
Regression risk:
Low. The requested change is a small xe page-table operation fix already
upstream and marked for stable backport.
Attachments:
version.log
lspci-vvnn.log
linux2-kernel-oops-20260915.log
** Affects: linux (Ubuntu)
Importance: Undecided
Status: New
** Attachment added: "Kernel version, lspci -vvnn, and 2026-09-15 kernel Oops
excerpt from Linux2 Arc Pro B70 wedge."
https://bugs.launchpad.net/bugs/2167337/+attachment/6000408/+files/linux2-b70-launchpad-evidence-20260915.tar.gz
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2167337
Title:
SRU: backport drm/xe current_op fix 0460455 for Intel Arc Pro B70 host
wedge
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2167337/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
