Re: [PATCH v5 03/10] drm/xe/pm: avoid reclaim when arming PME wakeup

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Misuse of Memory Allocator API: using memalloc_noreclaim_save() improperly grants emergency memory reserves to ACPI methods.
--

commit 7547491908f27fd6d8049619891faf97cc3918e3
Author: Vinod Govindapillai <vinod.govindapillai@intel.com>

drm/xe/pm: avoid reclaim when arming PME wakeup

This commit wraps calls to pci_enable_wake() with memalloc_noreclaim_save()
and memalloc_noreclaim_restore() in xe_pci_runtime_suspend(). The goal is to
avoid deadlocks with the xe_shrinker by preventing allocations from entering
reclaim while holding the runtime PM lock.

> diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c
> index 9b486e8ad2b5c..4369f0d9efb1e 100644
> --- a/drivers/gpu/drm/xe/xe_pci.c
> +++ b/drivers/gpu/drm/xe/xe_pci.c
> @@ -1384,10 +1384,13 @@ static int xe_pci_runtime_suspend(struct device *dev)
>  {
>  	struct pci_dev *pdev = to_pci_dev(dev);
>  	struct xe_device *xe = pdev_to_xe_device(pdev);
> +	unsigned int flags;
>  	int err;
>  
> +	flags = memalloc_noreclaim_save();
>  	xe->pme.enabled = xe->pme.capable && !xe->d3cold.allowed &&
>  			  pci_enable_wake(pdev, PCI_D3hot, true) == 0;
> +	memalloc_noreclaim_restore(flags);

[Severity: Medium]
Could this use of memalloc_noreclaim_save() improperly grant emergency memory
reserves to ACPI methods?

Using memalloc_noreclaim_save() sets the PF_MEMALLOC flag for the current
task, which grants allocations access to the ALLOC_NO_WATERMARKS emergency
memory reserves. These reserves are strictly intended for tasks that are
actively freeing memory to guarantee forward progress.

Since ACPI methods inside pci_enable_wake() can perform arbitrary evaluation
and allocation, granting them emergency reserves risks depleting these
reserves.

To avoid the deadlock with xe_shrinker without abusing PF_MEMALLOC, would it
be safer to use memalloc_nofs_save() instead?

Stripping __GFP_FS causes can_backup to be false in xe_shrinker_scan(), which
makes the shrinker exit early without attempting to acquire the runtime PM
lock.

>  
>  	/*
>  	 * We hold an additional reference to the runtime PM to keep PF in D0
> @@ -1401,7 +1404,9 @@ static int xe_pci_runtime_suspend(struct device *dev)
>  	err = xe_pm_runtime_suspend(xe);
>  	if (err) {
>  		if (xe->pme.enabled) {
> +			flags = memalloc_noreclaim_save();
>  			pci_enable_wake(pdev, PCI_D3hot, false);
> +			memalloc_noreclaim_restore(flags);

[Severity: Medium]
Does this error path have the same issue with granting emergency memory
reserves to pci_enable_wake() via memalloc_noreclaim_save() as mentioned
above?

>  			xe->pme.enabled = false;
>  		}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260911080448.778316-1-vinod.govindapillai@intel.com?part=3