Re: [PATCH v10 06/10] drm/xe/migrate: Introduce xe_migrate_reinit()

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] xe_migrate_reinit() fails to clear the VRAM-backed page table BOs (m->pt_bo and the root page table) after a PCIe FLR, leaving garbage PDEs/PTEs.
--

--- Patch [6]: [PATCH v10 06/10] drm/xe/migrate: Introduce xe_migrate_reinit() ---
commit 897712e0d4905c2317c45ed76acb56c7362ca8d9
Author: Raag Jadav <[email protected]>

drm/xe/migrate: Introduce xe_migrate_reinit()

This commit introduces the xe_migrate_reinit() helper to re-initialize the
migrate context after a PCIe FLR. Since the migrate exec queue and pt_bo live
on VRAM, their contents are lost during FLR and require re-initialization.

> diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
> index 75b83687f1b5d..eac91a4b78ed9 100644
> --- a/drivers/gpu/drm/xe/xe_migrate.c
> +++ b/drivers/gpu/drm/xe/xe_migrate.c

[ ... ]

> @@ -454,6 +454,18 @@ static int xe_migrate_lock_prepare_vm(struct xe_tile *tile, struct xe_migrate *m
>  	return err;
>  }
>  
> +/**
> + * xe_migrate_reinit() - Re-initialize a migrate context
> + * @m: The migration context
> + *
> + * Returns: 0 on success, negative error code otherwise.
> + */
> +int xe_migrate_reinit(struct xe_migrate *m)
> +{
> +	xe_migrate_prepare_vm(m->tile, m, m->q->vm, NULL);
> +	return xe_exec_queue_reinit(m->q);
> +}

[Severity: High]
Does this code leave garbage PDEs/PTEs in the VRAM-backed page table BOs
after a PCIe FLR?

Since the FLR garbles VRAM contents, and xe_migrate_prepare_vm() only writes
specific PDEs and PTEs rather than zeroing out the entire page table space,
the unwritten regions in m->pt_bo and the root page table will retain VRAM
garbage.

Could this cause the GPU page walker to speculatively fetch garbage PDEs/PTEs
(which may randomly have the Valid bit set), leading to unauthorized memory
accesses, fatal page faults, or GPU hangs?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=6