[Question] Where to investigate DMAR interrupt-remap warning observed near four-B70 initialization?

Aperta
  • Qualcosa non va?
#1,054 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
35/100
Tipo di issue
Bug
Chiarezza
Da chiarire
Stato di attività
Attiva
Stack tecnologico
cpp, linux, ubuntu
Ambito
backend, operating-systems

Direzione di ricerca

The issue does not identify a compute-runtime file, test, or confirmed runtime version, and asks who owns the warning. Start by checking the kernel’s DMAR/IOMMU logs and PCI device inventory alongside the compute-runtime version and a minimal reproducer; the issue provides no specific commands or files to begin with. Done means identifying the appropriate owning team and the non-invasive evidence needed to triage the source.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Question Category

Compatibility

Question

On an Ubuntu 26.04.1 four-Intel-Arc-Pro-B70 system, we observed the Linux kernel report an interrupt-remapping source-ID verification failure around Level Zero/oneCCL initialization. Is this appropriate for compute-runtime triage, or should we report it to a kernel IOMMU, board/firmware or other team? Which specific non-invasive logs would help identify the source?

DMAR: DRHD: handling fault status reg 2
DMAR: [INTR-REMAP] Request device [00:1b.4] fault index 0x2 [fault reason 0x26] Blocked an interrupt request due to source-id verification failure

We are not asserting a compute-runtime or GPU-driver defect. The same pair appeared at boot and again outside the test container's lifetime. The printed 00:1b.4 did not correspond to an enumerated 0000:00:1b.4 in our inspection. Another PCI address, 0000:24:1b.4, is distinct and has not been identified as the requester.

Additional Notes

  • ASUS Pro WS W890E-SAGE SE, BIOS 0507; 4 x Arc Pro B70; Ubuntu 26.04.1, kernel 7.0.0-43-generic, kernel driver xe.
  • A synthetic, model-free four-process Torch XPU / oneCCL all-reduce test was stopped by our monitor after the pair appeared during startup. The retained output did not show a completed collective; this is not evidence the collective failed. The same pair was observed again around 55 seconds after the container was removed.
  • An earlier Ornith model service returned responses after the same warning. That does not prove the warning is harmless or that unrelated devices are unaffected.
  • Torch 2.11.0+xpu, Triton XPU 3.7.0; a locally adapted vLLM/oneCCL stack and rebuilt extension were present, but the selected oneCCL implementation has not been verified. A minimal portable reproducer and current compute-runtime driver version have not yet been established.
  • We searched this repository for DMAR and INTR-REMAP; we did not find a matching report. We have not verified that the installed GPU runtime is the latest version.

We would appreciate guidance on the correct owner and useful evidence before proposing any workaround or attributing the fault to a particular component.

Related environment discussion: https://github.com/steveseguin/b70-optimization-lab/issues/84

Lingua principale
C++
Stelle
1.5k
Fork
306
Nessuna PR unita negli ultimi 30g

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di intel/compute-runtime

Tutte le issue di intel/compute-runtime

Issue simili

Altre issue su C++

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.