[PATCH 3/3] drm/amdgpu: drop eviction in pm notifier callbacks

Alex Deucher alexander.deucher at amd.com
Thu May 1 18:34:57 UTC 2025


We can't evict VRAM while processes are running as they
will continue to submit work which causes buffers to
get pulled back into VRAM.

Unfortunately we need a way to still have swap while
processes are fozen if memory is limited.

Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4178
Fixes: 2965e6355dcd ("drm/amd: Add Suspend/Hibernate notification callback support")
Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 21 ++-------------------
 1 file changed, 2 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index a3e9f289e37c9..dd2a557f97634 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4907,40 +4907,23 @@ static int amdgpu_device_evict_resources(struct amdgpu_device *adev)
  * @data: data
  *
  * This function is called when the system is about to suspend or hibernate.
- * It is used to evict resources from the device before the system goes to
- * sleep while there is still access to swap.
+ * It is used to set the appropriate flags so that eviction can be optimized
+ * in the pm prepare callback.
  */
 static int amdgpu_device_pm_notifier(struct notifier_block *nb, unsigned long mode,
 				     void *data)
 {
 	struct amdgpu_device *adev = container_of(nb, struct amdgpu_device, pm_nb);
-	int r;
 
 	switch (mode) {
 	case PM_HIBERNATION_PREPARE:
 		adev->in_s4 = true;
-		r = amdgpu_device_evict_resources(adev);
-		/*
-		 * This is considered non-fatal at this time because
-		 * amdgpu_device_prepare() will also fatally evict resources.
-		 * See https://gitlab.freedesktop.org/drm/amd/-/issues/3781
-		 */
-		if (r)
-			drm_warn(adev_to_drm(adev), "Failed to evict resources, freeze active processes if problems occur: %d\n", r);
 		break;
 	case PM_SUSPEND_PREPARE:
 		if (amdgpu_acpi_is_s0ix_active(adev))
 			adev->in_s0ix = true;
 		else if (amdgpu_acpi_is_s3_active(adev))
 			adev->in_s3 = true;
-		r = amdgpu_device_evict_resources(adev);
-		/*
-		 * This is considered non-fatal at this time because
-		 * amdgpu_device_prepare() will also fatally evict resources.
-		 * See https://gitlab.freedesktop.org/drm/amd/-/issues/3781
-		 */
-		if (r)
-			drm_warn(adev_to_drm(adev), "Failed to evict resources, freeze active processes if problems occur: %d\n", r);
 		break;
 	}
 
-- 
2.49.0



More information about the amd-gfx mailing list