Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 5f8f635e authored by Jerome Glisse's avatar Jerome Glisse Committed by Alex Deucher
Browse files

drm/radeon: avoid deadlock in pm path when waiting for fence 



radeon_fence_wait_empty_locked should not trigger GPU reset as no
place where it's call from would benefit from such thing and it
actually lead to a kernel deadlock in case the reset is triggered
from pm codepath. Instead force ring completion in place where it
makes sense or return early in others.

Signed-off-by: default avatarJerome Glisse <jglisse@redhat.com>
Reviewed-by: default avatarChristian König <christian.koenig@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
parent 76903b96

drivers/gpu/drm/radeon/radeon.h

+1 −1
Original line number Diff line number Diff line
@@ -231,7 +231,7 @@ void radeon_fence_process(struct radeon_device *rdev, int ring); 
bool radeon_fence_signaled(struct radeon_fence *fence);

int radeon_fence_wait(struct radeon_fence *fence, bool interruptible);

int radeon_fence_wait_next_locked(struct radeon_device *rdev, int ring);

void radeon_fence_wait_empty_locked(struct radeon_device *rdev, int ring);

int radeon_fence_wait_empty_locked(struct radeon_device *rdev, int ring);

int radeon_fence_wait_any(struct radeon_device *rdev,

			  struct radeon_fence **fences,

			  bool intr);

drivers/gpu/drm/radeon/radeon_device.c

+11 −2
Original line number Diff line number Diff line
@@ -1164,6 +1164,7 @@ int radeon_suspend_kms(struct drm_device *dev, pm_message_t state) 
	struct drm_crtc *crtc;

	struct drm_connector *connector;

	int i, r;

	bool force_completion = false;



	if (dev == NULL || dev->dev_private == NULL) {

		return -ENODEV;
@@ -1206,8 +1207,16 @@ int radeon_suspend_kms(struct drm_device *dev, pm_message_t state) 


	mutex_lock(&rdev->ring_lock);

	/* wait for gpu to finish processing current batch */

	for (i = 0; i < RADEON_NUM_RINGS; i++)

		radeon_fence_wait_empty_locked(rdev, i);

	for (i = 0; i < RADEON_NUM_RINGS; i++) {

		r = radeon_fence_wait_empty_locked(rdev, i);

		if (r) {

			/* delay GPU reset to resume */

			force_completion = true;

		}

	}

	if (force_completion) {

		radeon_fence_driver_force_completion(rdev);

	}

	mutex_unlock(&rdev->ring_lock);



	radeon_save_bios_scratch_regs(rdev);

drivers/gpu/drm/radeon/radeon_fence.c

+14 −16
Original line number Diff line number Diff line
@@ -609,26 +609,20 @@ int radeon_fence_wait_next_locked(struct radeon_device *rdev, int ring) 
 * Returns 0 if the fences have passed, error for all other cases.

 * Caller must hold ring lock.

 */

void radeon_fence_wait_empty_locked(struct radeon_device *rdev, int ring)

int radeon_fence_wait_empty_locked(struct radeon_device *rdev, int ring)

{

	uint64_t seq = rdev->fence_drv[ring].sync_seq[ring];



	while(1) {

	int r;



	r = radeon_fence_wait_seq(rdev, seq, ring, false, false);

		if (r == -EDEADLK) {

			mutex_unlock(&rdev->ring_lock);

			r = radeon_gpu_reset(rdev);

			mutex_lock(&rdev->ring_lock);

			if (!r)

				continue;

		}

	if (r) {

			dev_err(rdev->dev, "error waiting for ring to become"

				" idle (%d)\n", r);

		if (r == -EDEADLK) {

			return -EDEADLK;

		}

		return;

		dev_err(rdev->dev, "error waiting for ring[%d] to become idle (%d)\n",

			ring, r);

	}

	return 0;

}



/**
@@ -854,13 +848,17 @@ int radeon_fence_driver_init(struct radeon_device *rdev) 
 */

void radeon_fence_driver_fini(struct radeon_device *rdev)

{

	int ring;

	int ring, r;



	mutex_lock(&rdev->ring_lock);

	for (ring = 0; ring < RADEON_NUM_RINGS; ring++) {

		if (!rdev->fence_drv[ring].initialized)

			continue;

		radeon_fence_wait_empty_locked(rdev, ring);

		r = radeon_fence_wait_empty_locked(rdev, ring);

		if (r) {

			/* no need to trigger GPU reset as we are unloading */

			radeon_fence_driver_force_completion(rdev);

		}

		wake_up_all(&rdev->fence_queue);

		radeon_scratch_free(rdev, rdev->fence_drv[ring].scratch_reg);

		rdev->fence_drv[ring].initialized = false;

drivers/gpu/drm/radeon/radeon_pm.c

+12 −3
Original line number Diff line number Diff line
@@ -234,7 +234,7 @@ static void radeon_set_power_state(struct radeon_device *rdev) 


static void radeon_pm_set_clocks(struct radeon_device *rdev)

{

	int i;

	int i, r;



	/* no need to take locks, etc. if nothing's going to change */

	if ((rdev->pm.requested_clock_mode_index == rdev->pm.current_clock_mode_index) &&
@@ -248,8 +248,17 @@ static void radeon_pm_set_clocks(struct radeon_device *rdev) 
	/* wait for the rings to drain */

	for (i = 0; i < RADEON_NUM_RINGS; i++) {

		struct radeon_ring *ring = &rdev->ring[i];

		if (ring->ready)

			radeon_fence_wait_empty_locked(rdev, i);

		if (!ring->ready) {

			continue;

		}

		r = radeon_fence_wait_empty_locked(rdev, i);

		if (r) {

			/* needs a GPU reset dont reset here */

			mutex_unlock(&rdev->ring_lock);

			up_write(&rdev->pm.mclk_lock);

			mutex_unlock(&rdev->ddev->struct_mutex);

			return;

		}

	}



	radeon_unmap_vram_bos(rdev);