[PATCH v7 3/7] drm/xe/pm: avoid reclaim when arming PME wakeup

[Vinod Govindapillai <vinod.govindapillai@intel.com>]

pci_enable_wake() evaluates ACPI methods and allocates with GFP_KERNEL.
xe's shrinker calls xe_pm_runtime_get() and blocks until the runtime
suspend transition completes, so fs_reclaim sits above the runtime PM
lockmap (primed in xe_pm_runtime_lockdep_prime()). An allocation that
enters reclaim from the suspend path can therefore deadlock against a
task already holding fs_reclaim and waiting on that transition.

Wrap the call in memalloc_noreclaim_save()/restore() so the allocation
is satisfied from reserves and never enters reclaim.

Assisted-by: GitHub_Copilot:claude-opus-5
Signed-off-by: Vinod Govindapillai <vinod.govindapillai@intel.com>
---
 drivers/gpu/drm/xe/xe_pci.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c
index 3e7ee67c46f9..ddea264b16fd 100644
--- a/drivers/gpu/drm/xe/xe_pci.c
+++ b/drivers/gpu/drm/xe/xe_pci.c
@@ -1384,11 +1384,14 @@ 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;
 	bool pme_enabled;
 	int err;
 
+	flags = memalloc_noreclaim_save();
 	pme_enabled = xe->pme.capable && !xe->d3cold.allowed &&
 		      pci_enable_wake(pdev, PCI_D3hot, true) == 0;
+	memalloc_noreclaim_restore(flags);
 
 	xe_pm_update_pme_enabled(xe, pme_enabled);
 
@@ -1404,7 +1407,10 @@ static int xe_pci_runtime_suspend(struct device *dev)
 	err = xe_pm_runtime_suspend(xe);
 	if (err) {
 		if (pme_enabled) {
+			flags = memalloc_noreclaim_save();
 			pci_enable_wake(pdev, PCI_D3hot, false);
+			memalloc_noreclaim_restore(flags);
+
 			xe_pm_update_pme_enabled(xe, false);
 		}
 
-- 
2.43.0