[PATCH 3/5] drm/amdgpu: take reset lock after halt_activities

From: Vitaly Prosyak <[email protected]>

halt_activities() calls drm_client_dev_suspend() which takes
clientlist_mutex. The userq restore path establishes the ordering:

  clientlist_mutex -> userq_mutex -> reset_domain->sem

Taking reset_domain->sem before halt_activities() inverts
clientlist_mutex -> reset_domain->sem and triggers:

  WARNING: possible circular locking dependency detected
  kworker/u128:0 is trying to acquire lock:
    (&dev->clientlist_mutex), at: drm_client_dev_suspend+0x2d/0xf0
  but task is already holding lock:
    (&reset_domain->sem), at: amdgpu_device_lock_reset_domain+0x1c/0x30 [amdgpu]
  Chain exists of:
    &dev->clientlist_mutex --> &userq_mgr->userq_mutex --> &reset_domain->sem

Move amdgpu_device_recovery_get_reset_lock() to after halt_activities().

Cc: Christian Koenig <[email protected]>
Cc: Alex Deucher <[email protected]>
Cc: Sunil Khatri <[email protected]>
Signed-off-by: Vitaly Prosyak <[email protected]>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index d7640da9f6de..c248d589f755 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -5786,19 +5786,23 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 	/* Cannot be called after locking reset domain */
 	amdgpu_ras_pre_reset(adev, &device_list);
 
-	/* We need to lock reset domain only once both for XGMI and single device */
-	amdgpu_device_recovery_get_reset_lock(adev, &device_list);
-
 	/* unmap all the mappings of doorbell and framebuffer to prevent user space from
 	 * accessing them
 	 */
 	unmap_mapping_range(adev->ddev.anon_inode->i_mapping, 0, 0, 1);
 	amdgpu_amdkfd_clear_kfd_mapping(adev);
 
+	/* Halt activities before taking reset lock to avoid circular dependency:
+	 * halt_activities -> drm_client_dev_suspend -> dc_lock, while display
+	 * IRQ path holds dc_lock and tries to acquire reset_domain->sem.
+	 */
 	amdgpu_device_halt_activities(adev, job, reset_context, &device_list,
 				      hive, need_emergency_restart);
 	if (need_emergency_restart)
 		goto skip_sched_resume;
+
+	/* Take reset lock after halt_activities to avoid AB-BA deadlock */
+	amdgpu_device_recovery_get_reset_lock(adev, &device_list);
 	/*
 	 * Must check guilty signal here since after this point all old
 	 * HW fences are force signaled.
-- 
2.54.0