[26.04_linux-nvidia-bos] NVIDIA: VR: SAUCE: cxl: guard unlinked memdev endpoints

[nirmoy]

Summary

• Treat both NULL and error-valued cxlmd->endpoint pointers as unlinked before entering CXL HDM and region helper paths.
• Port the functional hdm.c and region.c changes from aff4ccee4530 onto BOS/Resolute.
• Omit the source drivers/cxl/pci.c hunk because this branch already guards cxl_reset_done() against both NULL and ERR_PTR endpoints.

Root cause

cxlmd->endpoint starts as ERR_PTR(-ENXIO) until endpoint-port registration links the memdev to a real cxl_port. The affected helpers checked only for NULL, allowing early CXL consumers to pass the error pointer into functions such as device_find_child().

The BOS region-management backport exposes these helpers before endpoint linkage, producing the observed NVIDIA probe failure and boot delay.

Validation

• Verified the guard source check fails on the unmodified BOS branch and passes after the port.
• git diff --check: pass.
• Strict checkpatch.pl: 0 errors, 0 warnings.
• Zero-context patch ID for the two changed files matches the corresponding subset of aff4ccee4530.
• Confirmed CONFIG_CXL_BUS, CONFIG_CXL_MEM, CONFIG_CXL_PCI, and CONFIG_CXL_REGION are enabled in the generated amd64 BOS configuration.
• Forced amd64 rebuild of drivers/cxl/core/hdm.o, region.o, and drivers/cxl/core/built-in.a: pass.

Tracking

• NVBug 6274048
• https://bugs.launchpad.net/ubuntu/resolute/+source/linux-nvidia-bos/+bug/2153819

[github-actions]

PR Validation Report

Patchscan ✅ No Missing Fixes

All cherry-picked commits checked — no missing upstream fixes found.

PR Lint ✅ All checks passed

Details

Checking 1 commits...

Cherry-pick digest:
┌──────────────┬──────────────────────────────────────────────────────────────────┬────────────┬─────────┬───────────────────────────┐
│ Local        │ Referenced upstream / Patch subject                              │ Patch-ID   │ Subject │ SoB chain                 │
├──────────────┼──────────────────────────────────────────────────────────────────┼────────────┼─────────┼───────────────────────────┤
│ 2c512ff1698a │ [SAUCE] cxl: guard unlinked memdev endpoints                     │ N/A        │ N/A     │ nirmoyd                   │
└──────────────┴──────────────────────────────────────────────────────────────────┴────────────┴─────────┴───────────────────────────┘

Lint: all checks passed.

[nirmoy]

BaseOS Kernel Review

Summary

No issues found across the reviewed commits.

Findings: no problems found

Latest watcher review: open review

Generated test plan: open test plan

Kernel deb build: successful (download debs, 4 files)

Head: 2c512ff1698a

This comment is maintained by nv-pr-bot. It is updated when the GitHub watcher publishes a newer review.

[nirmoy]

Strata/Vera boot smoke (2026-07-03)

• PR 482 ARM64 debs at 2c512ff1698a booted successfully on strata-vera-eb5-107 as 7.0.6-pr482-g2c512ff1698a using a one-shot GRUB entry; the persistent default remains the known 6.16.12 kernel.
• BMC serial and the post-boot journal showed no panic, Oops, device_find_child, or cxl_get_committed_decoder signature. Boot completed in 2m55.785s (firmware 2m2.365s, loader 8.764s, kernel 22.932s, userspace 21.721s).
• Four NVIDIA GPUs remained visible on PCI. The installed nvidia/600.09 DKMS source does not compile against 7.0.6 (__assign_str tracepoint API change), so no NVIDIA module was available and nvidia-smi could not run. No CXL memdevs were enumerated; therefore this is a boot smoke test, not a functional reproduction of the endpoint/region race.
• mlx5_core reported 80 firmware-internal-error lines on both the prior and test boots, so those messages are pre-existing rather than introduced by this change.

[arighi]

This looks like a fix that can go upstream, is there any plan to post this to the LKML?

Other than that, LGTM.

Acked-by: Andrea Righi <arighi@nvidia.com>