1 files changed, 47 insertions, 7 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 53539ceeda16..019581577603 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -609,6 +609,13 @@ void amdgpu_device_detect_runtime_pm_mode(struct amdgpu_device *adev)
 	int bamaco_support;
 
 	adev->pm.rpm_mode = AMDGPU_RUNPM_NONE;
+	if (pci_is_thunderbolt_attached(adev->pdev) ||
+	    dev_is_removable(&adev->pdev->dev)) {
+		dev_info(adev->dev,
+			 "Runtime PM disabled for externally attached device\n");
+		return;
+	}
+
 	bamaco_support = amdgpu_device_supports_baco(adev);
 
 	switch (amdgpu_runtime_pm) {
@@ -2523,7 +2530,11 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev)
 	if (r)
 		goto init_failed;
 
-	amdgpu_ttm_enable_buffer_funcs(adev);
+	/* If SDMA is not brought up during hwini, the ttm buffer funcs enablement
+	 * is delayed after reset-on-init completes.
+	 */
+	if (amdgpu_ip_member_of_hwini(adev, AMD_IP_BLOCK_TYPE_SDMA))
+		amdgpu_ttm_enable_buffer_funcs(adev);
 
 	/* Don't init kfd if whole hive need to be reset during init */
 	if (adev->init_lvl->level != AMDGPU_INIT_LEVEL_MINIMAL_XGMI) {
@@ -3733,6 +3744,14 @@ static void amdgpu_device_sys_interface_fini(struct amdgpu_device *adev)
 	amdgpu_ptl_sysfs_fini(adev);
 }
 
+static bool
+amdgpu_device_should_register_switcheroo(struct amdgpu_device *adev, bool px)
+{
+	return !pci_is_thunderbolt_attached(adev->pdev) &&
+	       (px || (!dev_is_removable(&adev->pdev->dev) &&
+		       apple_gmux_detect(NULL, NULL)));
+}
+
 /**
  * amdgpu_device_init - initialize the driver
  *
@@ -4183,8 +4202,7 @@ fence_driver_init:
 
 	px = amdgpu_device_supports_px(adev);
 
-	if (px || (!dev_is_removable(&adev->pdev->dev) &&
-				apple_gmux_detect(NULL, NULL)))
+	if (amdgpu_device_should_register_switcheroo(adev, px))
 		vga_switcheroo_register_client(adev->pdev,
 					       &amdgpu_switcheroo_ops, px);
 
@@ -4349,8 +4367,7 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev)
 
 	px = amdgpu_device_supports_px(adev);
 
-	if (px || (!dev_is_removable(&adev->pdev->dev) &&
-				apple_gmux_detect(NULL, NULL)))
+	if (amdgpu_device_should_register_switcheroo(adev, px))
 		vga_switcheroo_unregister_client(adev->pdev);
 
 	if (px)
@@ -5037,6 +5054,31 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,
 		amdgpu_fence_driver_force_completion(ring, fence);
 	}
 
+	/*
+	 * MES scheduler rings have no drm scheduler, so they are missed by the
+	 * loop above. Realign their polling fence too (one per XCC), otherwise the
+	 * first post-reset submission polls forever on a stale seq. sched.ready is
+	 * only set while the driver owns the ring.
+	 */
+	for (i = 0; i < AMDGPU_MAX_MES_INST_PIPES; i++) {
+		struct amdgpu_ring *mes_ring = &adev->mes.ring[i];
+
+		if (mes_ring->fence_drv.initialized && mes_ring->sched.ready)
+			amdgpu_fence_driver_force_completion(mes_ring, fence);
+	}
+
+	/*
+	 * KIQ rings are polling-fence/no_scheduler like MES, so realign their
+	 * fence too (one ring per XCC), otherwise the first post-reset KIQ
+	 * submission polls forever on a stale seq.
+	 */
+	for (i = 0; i < AMDGPU_MAX_GC_INSTANCES; i++) {
+		struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[i].ring;
+
+		if (kiq_ring->fence_drv.initialized && kiq_ring->sched.ready)
+			amdgpu_fence_driver_force_completion(kiq_ring, fence);
+	}
+
 	amdgpu_fence_driver_isr_toggle(adev, false);
 
 	r = amdgpu_reset_prepare_hwcontext(adev, reset_context);
@@ -5137,8 +5179,6 @@ int amdgpu_device_reinit_after_reset(struct amdgpu_reset_context *reset_context)
 				if (r)
 					goto out;
 
-				amdgpu_ttm_enable_buffer_funcs(tmp_adev);
-
 				r = amdgpu_device_ip_resume_phase3(tmp_adev);
 				if (r)
 					goto out;