sync code with last improvements from OpenBSD

sys/dev/pci/drm/amd/amdgpu/amdgpu.h

@@ -246,6 +246,7 @@ extern int amdgpu_num_kcq;
 
 #define AMDGPU_VCNFW_LOG_SIZE (32 * 1024)
 extern int amdgpu_vcnfw_log;
+extern int amdgpu_sg_display;
 
 #define AMDGPU_VM_MAX_NUM_CTX			4096
 #define AMDGPU_SG_THRESHOLD			(256*1024*1024)
@@ -287,6 +288,9 @@ extern int amdgpu_vcnfw_log;
 #define AMDGPU_SMARTSHIFT_MAX_BIAS (100)
 #define AMDGPU_SMARTSHIFT_MIN_BIAS (-100)
 
+/* Extra time delay(in ms) to eliminate the influence of temperature momentary fluctuation */
+#define AMDGPU_SWCTF_EXTRA_DELAY		50
+
 struct amdgpu_device;
 struct amdgpu_irq_src;
 struct amdgpu_fpriv;
@@ -1291,6 +1295,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 void amdgpu_device_pci_config_reset(struct amdgpu_device *adev);
 int amdgpu_device_pci_reset(struct amdgpu_device *adev);
 bool amdgpu_device_need_post(struct amdgpu_device *adev);
+bool amdgpu_sg_display_supported(struct amdgpu_device *adev);
 bool amdgpu_device_pcie_dynamic_switching_supported(void);
 bool amdgpu_device_should_use_aspm(struct amdgpu_device *adev);
 bool amdgpu_device_aspm_support_quirk(void);

sys/dev/pci/drm/amd/amdgpu/amdgpu_atomfirmware.c

@@ -101,39 +101,97 @@ void amdgpu_atomfirmware_scratch_regs_init(struct amdgpu_device *adev)
 	}
 }
 
+static int amdgpu_atomfirmware_allocate_fb_v2_1(struct amdgpu_device *adev,
+	struct vram_usagebyfirmware_v2_1 *fw_usage, int *usage_bytes)
+{
+	uint32_t start_addr, fw_size, drv_size;
+
+	start_addr = le32_to_cpu(fw_usage->start_address_in_kb);
+	fw_size = le16_to_cpu(fw_usage->used_by_firmware_in_kb);
+	drv_size = le16_to_cpu(fw_usage->used_by_driver_in_kb);
+
+	DRM_DEBUG("atom firmware v2_1 requested %08x %dkb fw %dkb drv\n",
+			  start_addr,
+			  fw_size,
+			  drv_size);
+
+	if ((start_addr & ATOM_VRAM_OPERATION_FLAGS_MASK) ==
+		(uint32_t)(ATOM_VRAM_BLOCK_SRIOV_MSG_SHARE_RESERVATION <<
+		ATOM_VRAM_OPERATION_FLAGS_SHIFT)) {
+		/* Firmware request VRAM reservation for SR-IOV */
+		adev->mman.fw_vram_usage_start_offset = (start_addr &
+			(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
+		adev->mman.fw_vram_usage_size = fw_size << 10;
+		/* Use the default scratch size */
+		*usage_bytes = 0;
+	} else {
+		*usage_bytes = drv_size << 10;
+	}
+	return 0;
+}
+
+static int amdgpu_atomfirmware_allocate_fb_v2_2(struct amdgpu_device *adev,
+		struct vram_usagebyfirmware_v2_2 *fw_usage, int *usage_bytes)
+{
+	uint32_t fw_start_addr, fw_size, drv_start_addr, drv_size;
+
+	fw_start_addr = le32_to_cpu(fw_usage->fw_region_start_address_in_kb);
+	fw_size = le16_to_cpu(fw_usage->used_by_firmware_in_kb);
+
+	drv_start_addr = le32_to_cpu(fw_usage->driver_region0_start_address_in_kb);
+	drv_size = le32_to_cpu(fw_usage->used_by_driver_region0_in_kb);
+
+	DRM_DEBUG("atom requested fw start at %08x %dkb and drv start at %08x %dkb\n",
+			  fw_start_addr,
+			  fw_size,
+			  drv_start_addr,
+			  drv_size);
+
+	if ((fw_start_addr & (ATOM_VRAM_BLOCK_NEEDS_NO_RESERVATION << 30)) == 0) {
+		/* Firmware request VRAM reservation for SR-IOV */
+		adev->mman.fw_vram_usage_start_offset = (fw_start_addr &
+			(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
+		adev->mman.fw_vram_usage_size = fw_size << 10;
+	}
+
+	if ((drv_start_addr & (ATOM_VRAM_BLOCK_NEEDS_NO_RESERVATION << 30)) == 0) {
+		/* driver request VRAM reservation for SR-IOV */
+		adev->mman.drv_vram_usage_start_offset = (drv_start_addr &
+			(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
+		adev->mman.drv_vram_usage_size = drv_size << 10;
+	}
+
+	*usage_bytes = 0;
+	return 0;
+}
+
 int amdgpu_atomfirmware_allocate_fb_scratch(struct amdgpu_device *adev)
 {
 	struct atom_context *ctx = adev->mode_info.atom_context;
 	int index = get_index_into_master_table(atom_master_list_of_data_tables_v2_1,
 						vram_usagebyfirmware);
-	struct vram_usagebyfirmware_v2_1 *firmware_usage;
-	uint32_t start_addr, size;
+	struct vram_usagebyfirmware_v2_1 *fw_usage_v2_1;
+	struct vram_usagebyfirmware_v2_2 *fw_usage_v2_2;
 	uint16_t data_offset;
+	uint8_t frev, crev;
 	int usage_bytes = 0;
 
-	if (amdgpu_atom_parse_data_header(ctx, index, NULL, NULL, NULL, &data_offset)) {
-		firmware_usage = (struct vram_usagebyfirmware_v2_1 *)(ctx->bios + data_offset);
-		DRM_DEBUG("atom firmware requested %08x %dkb fw %dkb drv\n",
-			  le32_to_cpu(firmware_usage->start_address_in_kb),
-			  le16_to_cpu(firmware_usage->used_by_firmware_in_kb),
-			  le16_to_cpu(firmware_usage->used_by_driver_in_kb));
-
-		start_addr = le32_to_cpu(firmware_usage->start_address_in_kb);
-		size = le16_to_cpu(firmware_usage->used_by_firmware_in_kb);
-
-		if ((uint32_t)(start_addr & ATOM_VRAM_OPERATION_FLAGS_MASK) ==
-			(uint32_t)(ATOM_VRAM_BLOCK_SRIOV_MSG_SHARE_RESERVATION <<
-			ATOM_VRAM_OPERATION_FLAGS_SHIFT)) {
-			/* Firmware request VRAM reservation for SR-IOV */
-			adev->mman.fw_vram_usage_start_offset = (start_addr &
-				(~ATOM_VRAM_OPERATION_FLAGS_MASK)) << 10;
-			adev->mman.fw_vram_usage_size = size << 10;
-			/* Use the default scratch size */
-			usage_bytes = 0;
-		} else {
-			usage_bytes = le16_to_cpu(firmware_usage->used_by_driver_in_kb) << 10;
+	if (amdgpu_atom_parse_data_header(ctx, index, NULL, &frev, &crev, &data_offset)) {
+		if (frev == 2 && crev == 1) {
+			fw_usage_v2_1 =
+				(struct vram_usagebyfirmware_v2_1 *)(ctx->bios + data_offset);
+			amdgpu_atomfirmware_allocate_fb_v2_1(adev,
+					fw_usage_v2_1,
+					&usage_bytes);
+		} else if (frev >= 2 && crev >= 2) {
+			fw_usage_v2_2 =
+				(struct vram_usagebyfirmware_v2_2 *)(ctx->bios + data_offset);
+			amdgpu_atomfirmware_allocate_fb_v2_2(adev,
+					fw_usage_v2_2,
+					&usage_bytes);
 		}
 	}
+
 	ctx->scratch_size_bytes = 0;
 	if (usage_bytes == 0)
 		usage_bytes = 20 * 1024;

sys/dev/pci/drm/amd/amdgpu/amdgpu_cs.c

@@ -287,7 +287,7 @@ static int amdgpu_cs_pass1(struct amdgpu_cs_parser *p,
 
 	if (!p->gang_size) {
 		ret = -EINVAL;
-		goto free_partial_kdata;
+		goto free_all_kdata;
 	}
 
 	for (i = 0; i < p->gang_size; ++i) {

sys/dev/pci/drm/amd/amdgpu/amdgpu_device.c

@@ -1351,6 +1351,32 @@ bool amdgpu_device_need_post(struct amdgpu_device *adev)
 	return true;
 }
 
+/*
+ * On APUs with >= 64GB white flickering has been observed w/ SG enabled.
+ * Disable S/G on such systems until we have a proper fix.
+ * https://gitlab.freedesktop.org/drm/amd/-/issues/2354
+ * https://gitlab.freedesktop.org/drm/amd/-/issues/2735
+ */
+bool amdgpu_sg_display_supported(struct amdgpu_device *adev)
+{
+	switch (amdgpu_sg_display) {
+	case -1:
+		break;
+	case 0:
+		return false;
+	case 1:
+		return true;
+	default:
+		return false;
+	}
+	if ((totalram_pages() << (PAGE_SHIFT - 10)) +
+	    (adev->gmc.real_vram_size / 1024) >= 64000000) {
+		DRM_WARN("Disabling S/G due to >=64GB RAM\n");
+		return false;
+	}
+	return true;
+}
+
 /*
  * Intel hosts such as Raptor Lake and Sapphire Rapids don't support dynamic
  * speed switching. Until we have confirmation from Intel that a specific host

sys/dev/pci/drm/amd/amdgpu/amdgpu_drv.c

@@ -185,6 +185,7 @@ int amdgpu_num_kcq = -1;
 int amdgpu_smartshift_bias;
 int amdgpu_use_xgmi_p2p = 1;
 int amdgpu_vcnfw_log;
+int amdgpu_sg_display = -1; /* auto */
 
 static void amdgpu_drv_delayed_reset_work_handler(struct work_struct *work);
 
@@ -929,6 +930,16 @@ module_param_named(num_kcq, amdgpu_num_kcq, int, 0444);
 MODULE_PARM_DESC(vcnfw_log, "Enable vcnfw log(0 = disable (default value), 1 = enable)");
 module_param_named(vcnfw_log, amdgpu_vcnfw_log, int, 0444);
 
+/**
+ * DOC: sg_display (int)
+ * Disable S/G (scatter/gather) display (i.e., display from system memory).
+ * This option is only relevant on APUs.  Set this option to 0 to disable
+ * S/G display if you experience flickering or other issues under memory
+ * pressure and report the issue.
+ */
+MODULE_PARM_DESC(sg_display, "S/G Display (-1 = auto (default), 0 = disable)");
+module_param_named(sg_display, amdgpu_sg_display, int, 0444);
+
 /**
  * DOC: smu_pptable_id (int)
  * Used to override pptable id. id = 0 use VBIOS pptable.

sys/dev/pci/drm/amd/amdgpu/amdgpu_object.c

@@ -348,17 +348,16 @@ int amdgpu_bo_create_kernel(struct amdgpu_device *adev,
  * @adev: amdgpu device object
  * @offset: offset of the BO
  * @size: size of the BO
- * @domain: where to place it
  * @bo_ptr:  used to initialize BOs in structures
  * @cpu_addr: optional CPU address mapping
  *
- * Creates a kernel BO at a specific offset in the address space of the domain.
+ * Creates a kernel BO at a specific offset in VRAM.
  *
  * Returns:
  * 0 on success, negative error code otherwise.
  */
 int amdgpu_bo_create_kernel_at(struct amdgpu_device *adev,
-			       uint64_t offset, uint64_t size, uint32_t domain,
+			       uint64_t offset, uint64_t size,
 			       struct amdgpu_bo **bo_ptr, void **cpu_addr)
 {
 	struct ttm_operation_ctx ctx = { false, false };
@@ -368,8 +367,9 @@ int amdgpu_bo_create_kernel_at(struct amdgpu_device *adev,
 	offset &= LINUX_PAGE_MASK;
 	size = roundup2(size, PAGE_SIZE);
 
-	r = amdgpu_bo_create_reserved(adev, size, PAGE_SIZE, domain, bo_ptr,
-				      NULL, cpu_addr);
+	r = amdgpu_bo_create_reserved(adev, size, PAGE_SIZE,
+				      AMDGPU_GEM_DOMAIN_VRAM, bo_ptr, NULL,
+				      cpu_addr);
 	if (r)
 		return r;
 

sys/dev/pci/drm/amd/amdgpu/amdgpu_object.h

@@ -285,7 +285,7 @@ int amdgpu_bo_create_kernel(struct amdgpu_device *adev,
 			    u32 domain, struct amdgpu_bo **bo_ptr,
 			    u64 *gpu_addr, void **cpu_addr);
 int amdgpu_bo_create_kernel_at(struct amdgpu_device *adev,
-			       uint64_t offset, uint64_t size, uint32_t domain,
+			       uint64_t offset, uint64_t size,
 			       struct amdgpu_bo **bo_ptr, void **cpu_addr);
 int amdgpu_bo_create_user(struct amdgpu_device *adev,
 			  struct amdgpu_bo_param *bp,

sys/dev/pci/drm/amd/amdgpu/amdgpu_ttm.c

@@ -1574,6 +1574,23 @@ static void amdgpu_ttm_fw_reserve_vram_fini(struct amdgpu_device *adev)
 		NULL, &adev->mman.fw_vram_usage_va);
 }
 
+/*
+ * Driver Reservation functions
+ */
+/**
+ * amdgpu_ttm_drv_reserve_vram_fini - free drv reserved vram
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * free drv reserved vram if it has been reserved.
+ */
+static void amdgpu_ttm_drv_reserve_vram_fini(struct amdgpu_device *adev)
+{
+	amdgpu_bo_free_kernel(&adev->mman.drv_vram_usage_reserved_bo,
+						  NULL,
+						  NULL);
+}
+
 /**
  * amdgpu_ttm_fw_reserve_vram_init - create bo vram reservation from fw
  *
@@ -1595,11 +1612,34 @@ static int amdgpu_ttm_fw_reserve_vram_init(struct amdgpu_device *adev)
 	return amdgpu_bo_create_kernel_at(adev,
 					  adev->mman.fw_vram_usage_start_offset,
 					  adev->mman.fw_vram_usage_size,
-					  AMDGPU_GEM_DOMAIN_VRAM,
 					  &adev->mman.fw_vram_usage_reserved_bo,
 					  &adev->mman.fw_vram_usage_va);
 }
 
+/**
+ * amdgpu_ttm_drv_reserve_vram_init - create bo vram reservation from driver
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * create bo vram reservation from drv.
+ */
+static int amdgpu_ttm_drv_reserve_vram_init(struct amdgpu_device *adev)
+{
+	uint64_t vram_size = adev->gmc.visible_vram_size;
+
+	adev->mman.drv_vram_usage_reserved_bo = NULL;
+
+	if (adev->mman.drv_vram_usage_size == 0 ||
+	    adev->mman.drv_vram_usage_size > vram_size)
+		return 0;
+
+	return amdgpu_bo_create_kernel_at(adev,
+					  adev->mman.drv_vram_usage_start_offset,
+					  adev->mman.drv_vram_usage_size,
+					  &adev->mman.drv_vram_usage_reserved_bo,
+					  NULL);
+}
+
 /*
  * Memoy training reservation functions
  */
@@ -1622,14 +1662,15 @@ static int amdgpu_ttm_training_reserve_vram_fini(struct amdgpu_device *adev)
 	return 0;
 }
 
-static void amdgpu_ttm_training_data_block_init(struct amdgpu_device *adev)
+static void amdgpu_ttm_training_data_block_init(struct amdgpu_device *adev,
+						uint32_t reserve_size)
 {
 	struct psp_memory_training_context *ctx = &adev->psp.mem_train_ctx;
 
 	memset(ctx, 0, sizeof(*ctx));
 
 	ctx->c2p_train_data_offset =
-		roundup2((adev->gmc.mc_vram_size - adev->mman.discovery_tmr_size - SZ_1M), SZ_1M);
+		roundup2((adev->gmc.mc_vram_size - reserve_size - SZ_1M), SZ_1M);
 	ctx->p2c_train_data_offset =
 		(adev->gmc.mc_vram_size - GDDR6_MEM_TRAINING_OFFSET);
 	ctx->train_data_size =
@@ -1647,9 +1688,10 @@ static void amdgpu_ttm_training_data_block_init(struct amdgpu_device *adev)
  */
 static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
 {
-	int ret;
 	struct psp_memory_training_context *ctx = &adev->psp.mem_train_ctx;
 	bool mem_train_support = false;
+	uint32_t reserve_size = 0;
+	int ret;
 
 	if (!amdgpu_sriov_vf(adev)) {
 		if (amdgpu_atomfirmware_mem_training_supported(adev))
@@ -1665,18 +1707,18 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
 	 * Otherwise, fallback to legacy approach to check and reserve tmr block for ip
 	 * discovery data and G6 memory training data respectively
 	 */
-	adev->mman.discovery_tmr_size =
-		amdgpu_atomfirmware_get_fw_reserved_fb_size(adev);
-	if (!adev->mman.discovery_tmr_size)
-		adev->mman.discovery_tmr_size = DISCOVERY_TMR_OFFSET;
+	if (adev->bios)
+		reserve_size =
+			amdgpu_atomfirmware_get_fw_reserved_fb_size(adev);
+	if (!reserve_size)
+		reserve_size = DISCOVERY_TMR_OFFSET;
 
 	if (mem_train_support) {
 		/* reserve vram for mem train according to TMR location */
-		amdgpu_ttm_training_data_block_init(adev);
+		amdgpu_ttm_training_data_block_init(adev, reserve_size);
 		ret = amdgpu_bo_create_kernel_at(adev,
 					 ctx->c2p_train_data_offset,
 					 ctx->train_data_size,
-					 AMDGPU_GEM_DOMAIN_VRAM,
 					 &ctx->c2p_bo,
 					 NULL);
 		if (ret) {
@@ -1688,14 +1730,14 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
 	}
 
 	ret = amdgpu_bo_create_kernel_at(adev,
-				adev->gmc.real_vram_size - adev->mman.discovery_tmr_size,
-				adev->mman.discovery_tmr_size,
-				AMDGPU_GEM_DOMAIN_VRAM,
-				&adev->mman.discovery_memory,
+				adev->gmc.real_vram_size - reserve_size,
+				reserve_size,
+				&adev->mman.fw_reserved_memory,
 				NULL);
 	if (ret) {
 		DRM_ERROR("alloc tmr failed(%d)!\n", ret);
-		amdgpu_bo_free_kernel(&adev->mman.discovery_memory, NULL, NULL);
+		amdgpu_bo_free_kernel(&adev->mman.fw_reserved_memory,
+				      NULL, NULL);
 		return ret;
 	}
 
@@ -1788,6 +1830,14 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 		return r;
 	}
 
+	/*
+	 *The reserved vram for driver must be pinned to the specified
+	 *place on the VRAM, so reserve it early.
+	 */
+	r = amdgpu_ttm_drv_reserve_vram_init(adev);
+	if (r)
+		return r;
+
 	/*
 	 * only NAVI10 and onwards ASIC support for IP discovery.
 	 * If IP discovery enabled, a block of memory should be
@@ -1804,21 +1854,18 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 	 * avoid display artifacts while transitioning between pre-OS
 	 * and driver.  */
 	r = amdgpu_bo_create_kernel_at(adev, 0, adev->mman.stolen_vga_size,
-				       AMDGPU_GEM_DOMAIN_VRAM,
 				       &adev->mman.stolen_vga_memory,
 				       NULL);
 	if (r)
 		return r;
 	r = amdgpu_bo_create_kernel_at(adev, adev->mman.stolen_vga_size,
 				       adev->mman.stolen_extended_size,
-				       AMDGPU_GEM_DOMAIN_VRAM,
 				       &adev->mman.stolen_extended_memory,
 				       NULL);
 	if (r)
 		return r;
 	r = amdgpu_bo_create_kernel_at(adev, adev->mman.stolen_reserved_offset,
 				       adev->mman.stolen_reserved_size,
-				       AMDGPU_GEM_DOMAIN_VRAM,
 				       &adev->mman.stolen_reserved_memory,
 				       NULL);
 	if (r)
@@ -1910,14 +1957,16 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)
 	/* return the stolen vga memory back to VRAM */
 	amdgpu_bo_free_kernel(&adev->mman.stolen_vga_memory, NULL, NULL);
 	amdgpu_bo_free_kernel(&adev->mman.stolen_extended_memory, NULL, NULL);
-	/* return the IP Discovery TMR memory back to VRAM */
-	amdgpu_bo_free_kernel(&adev->mman.discovery_memory, NULL, NULL);
+	/* return the FW reserved memory back to VRAM */
+	amdgpu_bo_free_kernel(&adev->mman.fw_reserved_memory, NULL,
+			      NULL);
 	if (adev->mman.stolen_reserved_size)
 		amdgpu_bo_free_kernel(&adev->mman.stolen_reserved_memory,
 				      NULL, NULL);
 	amdgpu_bo_free_kernel(&adev->mman.sdma_access_bo, NULL,
 					&adev->mman.sdma_access_ptr);
 	amdgpu_ttm_fw_reserve_vram_fini(adev);
+	amdgpu_ttm_drv_reserve_vram_fini(adev);
 
 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
 

sys/dev/pci/drm/amd/amdgpu/amdgpu_ttm.h

@@ -79,7 +79,8 @@ struct amdgpu_mman {
 	/* discovery */
 	uint8_t				*discovery_bin;
 	uint32_t			discovery_tmr_size;
-	struct amdgpu_bo		*discovery_memory;
+	/* fw reserved memory */
+	struct amdgpu_bo		*fw_reserved_memory;
 
 	/* firmware VRAM reservation */
 	u64		fw_vram_usage_start_offset;
@@ -87,6 +88,11 @@ struct amdgpu_mman {
 	struct amdgpu_bo	*fw_vram_usage_reserved_bo;
 	void		*fw_vram_usage_va;
 
+	/* driver VRAM reservation */
+	u64		drv_vram_usage_start_offset;
+	u64		drv_vram_usage_size;
+	struct amdgpu_bo	*drv_vram_usage_reserved_bo;
+
 	/* PAGE_SIZE'd BO for process memory r/w over SDMA. */
 	struct amdgpu_bo	*sdma_access_bo;
 	void			*sdma_access_ptr;

sys/dev/pci/drm/amd/amdgpu/amdgpu_virt.c

@@ -391,7 +391,6 @@ static void amdgpu_virt_ras_reserve_bps(struct amdgpu_device *adev)
 		 */
 		if (amdgpu_bo_create_kernel_at(adev, bp << AMDGPU_GPU_PAGE_SHIFT,
 					       AMDGPU_GPU_PAGE_SIZE,
-					       AMDGPU_GEM_DOMAIN_VRAM,
 					       &bo, NULL))
 			DRM_DEBUG("RAS WARN: reserve vram for retired page %llx fail\n", bp);
 

sys/dev/pci/drm/amd/display/amdgpu_dm/amdgpu_dm.c

@@ -351,6 +351,19 @@ static inline bool is_dc_timing_adjust_needed(struct dm_crtc_state *old_state,
 		return false;
 }
 
+static inline void reverse_planes_order(struct dc_surface_update *array_of_surface_update,
+					int planes_count)
+{
+	int i, j;
+	struct dc_surface_update surface_updates_temp;
+
+	for (i = 0, j = planes_count - 1; i < j; i++, j--) {
+		surface_updates_temp = array_of_surface_update[i];
+		array_of_surface_update[i] = array_of_surface_update[j];
+		array_of_surface_update[j] = surface_updates_temp;
+	}
+}
+
 /**
  * update_planes_and_stream_adapter() - Send planes to be updated in DC
  *
@@ -367,6 +380,8 @@ static inline bool update_planes_and_stream_adapter(struct dc *dc,
 						    struct dc_stream_update *stream_update,
 						    struct dc_surface_update *array_of_surface_update)
 {
+	reverse_planes_order(array_of_surface_update, planes_count);
+
 	/*
 	 * Previous frame finished and HW is ready for optimization.
 	 */
@@ -1621,6 +1636,8 @@ static int amdgpu_dm_init(struct amdgpu_device *adev)
 		}
 		break;
 	}
+	if (init_data.flags.gpu_vm_support)
+		init_data.flags.gpu_vm_support = amdgpu_sg_display_supported(adev);
 
 	if (init_data.flags.gpu_vm_support)
 		adev->mode_info.gpu_vm_support = true;

sys/dev/pci/drm/amd/display/dc/core/amdgpu_dc.c

@@ -1079,6 +1079,7 @@ static void disable_dangling_plane(struct dc *dc, struct dc_state *context)
 	struct dc_state *dangling_context = dc_create_state(dc);
 	struct dc_state *current_ctx;
 	struct pipe_ctx *pipe;
+	struct timing_generator *tg;
 
 	if (dangling_context == NULL)
 		return;
@@ -1122,6 +1123,7 @@ static void disable_dangling_plane(struct dc *dc, struct dc_state *context)
 
 		if (should_disable && old_stream) {
 			pipe = &dc->current_state->res_ctx.pipe_ctx[i];
+			tg = pipe->stream_res.tg;
 			/* When disabling plane for a phantom pipe, we must turn on the
 			 * phantom OTG so the disable programming gets the double buffer
 			 * update. Otherwise the pipe will be left in a partially disabled
@@ -1129,7 +1131,8 @@ static void disable_dangling_plane(struct dc *dc, struct dc_state *context)
 			 * again for different use.
 			 */
 			if (old_stream->mall_stream_config.type == SUBVP_PHANTOM) {
-				pipe->stream_res.tg->funcs->enable_crtc(pipe->stream_res.tg);
+				if (tg->funcs->enable_crtc)
+					tg->funcs->enable_crtc(tg);
 			}
 			dc_rem_all_planes_for_stream(dc, old_stream, dangling_context);
 			disable_all_writeback_pipes_for_stream(dc, old_stream, dangling_context);
@@ -1146,6 +1149,15 @@ static void disable_dangling_plane(struct dc *dc, struct dc_state *context)
 				dc->hwss.interdependent_update_lock(dc, dc->current_state, false);
 				dc->hwss.post_unlock_program_front_end(dc, dangling_context);
 			}
+			/* We need to put the phantom OTG back into it's default (disabled) state or we
+			 * can get corruption when transition from one SubVP config to a different one.
+			 * The OTG is set to disable on falling edge of VUPDATE so the plane disable
+			 * will still get it's double buffer update.
+			 */
+			if (old_stream->mall_stream_config.type == SUBVP_PHANTOM) {
+				if (tg->funcs->disable_phantom_crtc)
+					tg->funcs->disable_phantom_crtc(tg);
+			}
 		}
 	}
 
@@ -1942,6 +1954,9 @@ enum dc_status dc_commit_streams(struct dc *dc,
 	struct pipe_ctx *pipe;
 	bool handle_exit_odm2to1 = false;
 
+	if (dc->ctx->dce_environment == DCE_ENV_VIRTUAL_HW)
+		return res;
+
 	if (!streams_changed(dc, streams, stream_count))
 		return res;
 
@@ -1984,21 +1999,33 @@ enum dc_status dc_commit_streams(struct dc *dc,
 
 	dc_resource_state_copy_construct_current(dc, context);
 
-	/*
-	 * Previous validation was perfomred with fast_validation = true and
-	 * the full DML state required for hardware programming was skipped.
-	 *
-	 * Re-validate here to calculate these parameters / watermarks.
-	 */
-	res = dc_validate_global_state(dc, context, false);
+	res = dc_validate_with_context(dc, set, stream_count, context, false);
 	if (res != DC_OK) {
-		DC_LOG_ERROR("DC commit global validation failure: %s (%d)",
-			     dc_status_to_str(res), res);
-		return res;
+		BREAK_TO_DEBUGGER();
+		goto fail;
 	}
 
 	res = dc_commit_state_no_check(dc, context);
 
+	for (i = 0; i < stream_count; i++) {
+		for (j = 0; j < context->stream_count; j++) {
+			if (streams[i]->stream_id == context->streams[j]->stream_id)
+				streams[i]->out.otg_offset = context->stream_status[j].primary_otg_inst;
+
+			if (dc_is_embedded_signal(streams[i]->signal)) {
+				struct dc_stream_status *status = dc_stream_get_status_from_state(context, streams[i]);
+
+				if (dc->hwss.is_abm_supported)
+					status->is_abm_supported = dc->hwss.is_abm_supported(dc, context, streams[i]);
+				else
+					status->is_abm_supported = true;
+			}
+		}
+	}
+
+fail:
+	dc_release_state(context);
+
 context_alloc_fail:
 
 	DC_LOG_DC("%s Finished.\n", __func__);
@@ -3122,6 +3149,19 @@ static bool update_planes_and_stream_state(struct dc *dc,
 
 	if (update_type == UPDATE_TYPE_FULL) {
 		if (!dc->res_pool->funcs->validate_bandwidth(dc, context, false)) {
+			/* For phantom pipes we remove and create a new set of phantom pipes
+			 * for each full update (because we don't know if we'll need phantom
+			 * pipes until after the first round of validation). However, if validation
+			 * fails we need to keep the existing phantom pipes (because we don't update
+			 * the dc->current_state).
+			 *
+			 * The phantom stream/plane refcount is decremented for validation because
+			 * we assume it'll be removed (the free comes when the dc_state is freed),
+			 * but if validation fails we have to increment back the refcount so it's
+			 * consistent.
+			 */
+			if (dc->res_pool->funcs->retain_phantom_pipes)
+				dc->res_pool->funcs->retain_phantom_pipes(dc, dc->current_state);
 			BREAK_TO_DEBUGGER();
 			goto fail;
 		}
@@ -3987,6 +4027,18 @@ void dc_commit_updates_for_stream(struct dc *dc,
 	struct dc_context *dc_ctx = dc->ctx;
 	int i, j;
 
+	/* TODO: Since change commit sequence can have a huge impact,
+	 * we decided to only enable it for DCN3x. However, as soon as
+	 * we get more confident about this change we'll need to enable
+	 * the new sequence for all ASICs.
+	 */
+	if (dc->ctx->dce_version >= DCN_VERSION_3_2) {
+		dc_update_planes_and_stream(dc, srf_updates,
+					    surface_count, stream,
+					    stream_update);
+		return;
+	}
+
 	stream_status = dc_stream_get_status(stream);
 	context = dc->current_state;
 

sys/dev/pci/drm/amd/display/dc/core/dc_link.c

@@ -1141,6 +1141,11 @@ static bool detect_link_and_local_sink(struct dc_link *link,
 					(link->dpcd_caps.dongle_type !=
 							DISPLAY_DONGLE_DP_HDMI_CONVERTER))
 				converter_disable_audio = true;
+
+			/* limited link rate to HBR3 for DPIA until we implement USB4 V2 */
+			if (link->ep_type == DISPLAY_ENDPOINT_USB4_DPIA &&
+					link->reported_link_cap.link_rate > LINK_RATE_HIGH3)
+				link->reported_link_cap.link_rate = LINK_RATE_HIGH3;
 			break;
 		}
 
@@ -2092,6 +2097,7 @@ static enum dc_status enable_link_dp_mst(
 		struct pipe_ctx *pipe_ctx)
 {
 	struct dc_link *link = pipe_ctx->stream->link;
+	unsigned char mstm_cntl;
 
 	/* sink signal type after MST branch is MST. Multiple MST sinks
 	 * share one link. Link DP PHY is enable or training only once.
@@ -2100,7 +2106,9 @@ static enum dc_status enable_link_dp_mst(
 		return DC_OK;
 
 	/* clear payload table */
-	dm_helpers_dp_mst_clear_payload_allocation_table(link->ctx, link);
+	core_link_read_dpcd(link, DP_MSTM_CTRL, &mstm_cntl, 1);
+	if (mstm_cntl & DP_MST_EN)
+		dm_helpers_dp_mst_clear_payload_allocation_table(link->ctx, link);
 
 	/* to make sure the pending down rep can be processed
 	 * before enabling the link

sys/dev/pci/drm/amd/display/dc/core/dc_resource.c

@@ -2616,15 +2616,241 @@ bool dc_resource_is_dsc_encoding_supported(const struct dc *dc)
 	return dc->res_pool->res_cap->num_dsc > 0;
 }
 
+static bool planes_changed_for_existing_stream(struct dc_state *context,
+					       struct dc_stream_state *stream,
+					       const struct dc_validation_set set[],
+					       int set_count)
+{
+	int i, j;
+	struct dc_stream_status *stream_status = NULL;
+
+	for (i = 0; i < context->stream_count; i++) {
+		if (context->streams[i] == stream) {
+			stream_status = &context->stream_status[i];
+			break;
+		}
+	}
+
+	if (!stream_status)
+		ASSERT(0);
+
+	for (i = 0; i < set_count; i++)
+		if (set[i].stream == stream)
+			break;
+
+	if (i == set_count)
+		ASSERT(0);
+
+	if (set[i].plane_count != stream_status->plane_count)
+		return true;
+
+	for (j = 0; j < set[i].plane_count; j++)
+		if (set[i].plane_states[j] != stream_status->plane_states[j])
+			return true;
+
+	return false;
+}
 
 /**
- * dc_validate_global_state() - Determine if HW can support a given state
- * Checks HW resource availability and bandwidth requirement.
+ * dc_validate_with_context - Validate and update the potential new stream in the context object
+ *
+ * @dc: Used to get the current state status
+ * @set: An array of dc_validation_set with all the current streams reference
+ * @set_count: Total of streams
+ * @context: New context
+ * @fast_validate: Enable or disable fast validation
+ *
+ * This function updates the potential new stream in the context object. It
+ * creates multiple lists for the add, remove, and unchanged streams. In
+ * particular, if the unchanged streams have a plane that changed, it is
+ * necessary to remove all planes from the unchanged streams. In summary, this
+ * function is responsible for validating the new context.
+ *
+ * Return:
+ * In case of success, return DC_OK (1), otherwise, return a DC error.
+ */
+enum dc_status dc_validate_with_context(struct dc *dc,
+					const struct dc_validation_set set[],
+					int set_count,
+					struct dc_state *context,
+					bool fast_validate)
+{
+	struct dc_stream_state *unchanged_streams[MAX_PIPES] = { 0 };
+	struct dc_stream_state *del_streams[MAX_PIPES] = { 0 };
+	struct dc_stream_state *add_streams[MAX_PIPES] = { 0 };
+	int old_stream_count = context->stream_count;
+	enum dc_status res = DC_ERROR_UNEXPECTED;
+	int unchanged_streams_count = 0;
+	int del_streams_count = 0;
+	int add_streams_count = 0;
+	bool found = false;
+	int i, j, k;
+
+	DC_LOGGER_INIT(dc->ctx->logger);
+
+	/* First build a list of streams to be remove from current context */
+	for (i = 0; i < old_stream_count; i++) {
+		struct dc_stream_state *stream = context->streams[i];
+
+		for (j = 0; j < set_count; j++) {
+			if (stream == set[j].stream) {
+				found = true;
+				break;
+			}
+		}
+
+		if (!found)
+			del_streams[del_streams_count++] = stream;
+
+		found = false;
+	}
+
+	/* Second, build a list of new streams */
+	for (i = 0; i < set_count; i++) {
+		struct dc_stream_state *stream = set[i].stream;
+
+		for (j = 0; j < old_stream_count; j++) {
+			if (stream == context->streams[j]) {
+				found = true;
+				break;
+			}
+		}
+
+		if (!found)
+			add_streams[add_streams_count++] = stream;
+
+		found = false;
+	}
+
+	/* Build a list of unchanged streams which is necessary for handling
+	 * planes change such as added, removed, and updated.
+	 */
+	for (i = 0; i < set_count; i++) {
+		/* Check if stream is part of the delete list */
+		for (j = 0; j < del_streams_count; j++) {
+			if (set[i].stream == del_streams[j]) {
+				found = true;
+				break;
+			}
+		}
+
+		if (!found) {
+			/* Check if stream is part of the add list */
+			for (j = 0; j < add_streams_count; j++) {
+				if (set[i].stream == add_streams[j]) {
+					found = true;
+					break;
+				}
+			}
+		}
+
+		if (!found)
+			unchanged_streams[unchanged_streams_count++] = set[i].stream;
+
+		found = false;
+	}
+
+	/* Remove all planes for unchanged streams if planes changed */
+	for (i = 0; i < unchanged_streams_count; i++) {
+		if (planes_changed_for_existing_stream(context,
+						       unchanged_streams[i],
+						       set,
+						       set_count)) {
+			if (!dc_rem_all_planes_for_stream(dc,
+							  unchanged_streams[i],
+							  context)) {
+				res = DC_FAIL_DETACH_SURFACES;
+				goto fail;
+			}
+		}
+	}
+
+	/* Remove all planes for removed streams and then remove the streams */
+	for (i = 0; i < del_streams_count; i++) {
+		/* Need to cpy the dwb data from the old stream in order to efc to work */
+		if (del_streams[i]->num_wb_info > 0) {
+			for (j = 0; j < add_streams_count; j++) {
+				if (del_streams[i]->sink == add_streams[j]->sink) {
+					add_streams[j]->num_wb_info = del_streams[i]->num_wb_info;
+					for (k = 0; k < del_streams[i]->num_wb_info; k++)
+						add_streams[j]->writeback_info[k] = del_streams[i]->writeback_info[k];
+				}
+			}
+		}
+
+		if (!dc_rem_all_planes_for_stream(dc, del_streams[i], context)) {
+			res = DC_FAIL_DETACH_SURFACES;
+			goto fail;
+		}
+
+		res = dc_remove_stream_from_ctx(dc, context, del_streams[i]);
+		if (res != DC_OK)
+			goto fail;
+	}
+
+	/* Swap seamless boot stream to pipe 0 (if needed) to ensure pipe_ctx
+	 * matches. This may change in the future if seamless_boot_stream can be
+	 * multiple.
+	 */
+	for (i = 0; i < add_streams_count; i++) {
+		mark_seamless_boot_stream(dc, add_streams[i]);
+		if (add_streams[i]->apply_seamless_boot_optimization && i != 0) {
+			struct dc_stream_state *temp = add_streams[0];
+
+			add_streams[0] = add_streams[i];
+			add_streams[i] = temp;
+			break;
+		}
+	}
+
+	/* Add new streams and then add all planes for the new stream */
+	for (i = 0; i < add_streams_count; i++) {
+		calculate_phy_pix_clks(add_streams[i]);
+		res = dc_add_stream_to_ctx(dc, context, add_streams[i]);
+		if (res != DC_OK)
+			goto fail;
+
+		if (!add_all_planes_for_stream(dc, add_streams[i], set, set_count, context)) {
+			res = DC_FAIL_ATTACH_SURFACES;
+			goto fail;
+		}
+	}
+
+	/* Add all planes for unchanged streams if planes changed */
+	for (i = 0; i < unchanged_streams_count; i++) {
+		if (planes_changed_for_existing_stream(context,
+						       unchanged_streams[i],
+						       set,
+						       set_count)) {
+			if (!add_all_planes_for_stream(dc, unchanged_streams[i], set, set_count, context)) {
+				res = DC_FAIL_ATTACH_SURFACES;
+				goto fail;
+			}
+		}
+	}
+
+	res = dc_validate_global_state(dc, context, fast_validate);
+
+fail:
+	if (res != DC_OK)
+		DC_LOG_WARNING("%s:resource validation failed, dc_status:%d\n",
+			       __func__,
+			       res);
+
+	return res;
+}
+
+/**
+ * dc_validate_global_state() - Determine if hardware can support a given state
+ *
  * @dc: dc struct for this driver
  * @new_ctx: state to be validated
  * @fast_validate: set to true if only yes/no to support matters
  *
- * Return: DC_OK if the result can be programmed.  Otherwise, an error code.
+ * Checks hardware resource availability and bandwidth requirement.
+ *
+ * Return:
+ * DC_OK if the result can be programmed. Otherwise, an error code.
  */
 enum dc_status dc_validate_global_state(
 		struct dc *dc,
@@ -3757,4 +3983,4 @@ bool dc_resource_acquire_secondary_pipe_for_mpc_odm(
 	}
 
 	return true;
-}
+}

sys/dev/pci/drm/amd/display/dc/dc.h

@@ -1298,6 +1298,12 @@ enum dc_status dc_validate_plane(struct dc *dc, const struct dc_plane_state *pla
 
 void get_clock_requirements_for_state(struct dc_state *state, struct AsicStateEx *info);
 
+enum dc_status dc_validate_with_context(struct dc *dc,
+					const struct dc_validation_set set[],
+					int set_count,
+					struct dc_state *context,
+					bool fast_validate);
+
 bool dc_set_generic_gpio_for_stereo(bool enable,
 		struct gpio_service *gpio_service);
 

sys/dev/pci/drm/amd/display/dc/dcn10/dcn10_hw_sequencer.c

@@ -2284,6 +2284,12 @@ void dcn10_enable_timing_synchronization(
 		opp = grouped_pipes[i]->stream_res.opp;
 		tg = grouped_pipes[i]->stream_res.tg;
 		tg->funcs->get_otg_active_size(tg, &width, &height);
+
+		if (!tg->funcs->is_tg_enabled(tg)) {
+			DC_SYNC_INFO("Skipping timing sync on disabled OTG\n");
+			return;
+		}
+
 		if (opp->funcs->opp_program_dpg_dimensions)
 			opp->funcs->opp_program_dpg_dimensions(opp, width, 2*(height) + 1);
 	}

sys/dev/pci/drm/amd/display/dc/dcn30/dcn30_dpp.c

@@ -357,8 +357,11 @@ void dpp3_set_cursor_attributes(
 	int cur_rom_en = 0;
 
 	if (color_format == CURSOR_MODE_COLOR_PRE_MULTIPLIED_ALPHA ||
-		color_format == CURSOR_MODE_COLOR_UN_PRE_MULTIPLIED_ALPHA)
-		cur_rom_en = 1;
+		color_format == CURSOR_MODE_COLOR_UN_PRE_MULTIPLIED_ALPHA) {
+		if (cursor_attributes->attribute_flags.bits.ENABLE_CURSOR_DEGAMMA) {
+			cur_rom_en = 1;
+		}
+	}
 
 	REG_UPDATE_3(CURSOR0_CONTROL,
 			CUR0_MODE, color_format,

sys/dev/pci/drm/amd/display/dc/dcn32/dcn32_optc.c

@@ -167,6 +167,13 @@ static void optc32_phantom_crtc_post_enable(struct timing_generator *optc)
 	REG_WAIT(OTG_CLOCK_CONTROL, OTG_BUSY, 0, 1, 100000);
 }
 
+static void optc32_disable_phantom_otg(struct timing_generator *optc)
+{
+	struct optc *optc1 = DCN10TG_FROM_TG(optc);
+
+	REG_UPDATE(OTG_CONTROL, OTG_MASTER_EN, 0);
+}
+
 static void optc32_set_odm_bypass(struct timing_generator *optc,
 		const struct dc_crtc_timing *dc_crtc_timing)
 {
@@ -260,6 +267,7 @@ static struct timing_generator_funcs dcn32_tg_funcs = {
 		.enable_crtc = optc32_enable_crtc,
 		.disable_crtc = optc32_disable_crtc,
 		.phantom_crtc_post_enable = optc32_phantom_crtc_post_enable,
+		.disable_phantom_crtc = optc32_disable_phantom_otg,
 		/* used by enable_timing_synchronization. Not need for FPGA */
 		.is_counter_moving = optc1_is_counter_moving,
 		.get_position = optc1_get_position,

sys/dev/pci/drm/amd/display/dc/dcn32/dcn32_resource.c

@@ -1719,6 +1719,27 @@ static struct dc_stream_state *dcn32_enable_phantom_stream(struct dc *dc,
 	return phantom_stream;
 }
 
+void dcn32_retain_phantom_pipes(struct dc *dc, struct dc_state *context)
+{
+	int i;
+	struct dc_plane_state *phantom_plane = NULL;
+	struct dc_stream_state *phantom_stream = NULL;
+
+	for (i = 0; i < dc->res_pool->pipe_count; i++) {
+		struct pipe_ctx *pipe = &context->res_ctx.pipe_ctx[i];
+
+		if (!pipe->top_pipe && !pipe->prev_odm_pipe &&
+				pipe->plane_state && pipe->stream &&
+				pipe->stream->mall_stream_config.type == SUBVP_PHANTOM) {
+			phantom_plane = pipe->plane_state;
+			phantom_stream = pipe->stream;
+
+			dc_plane_state_retain(phantom_plane);
+			dc_stream_retain(phantom_stream);
+		}
+	}
+}
+
 // return true if removed piped from ctx, false otherwise
 bool dcn32_remove_phantom_pipes(struct dc *dc, struct dc_state *context)
 {
@@ -2035,6 +2056,7 @@ static struct resource_funcs dcn32_res_pool_funcs = {
 	.update_soc_for_wm_a = dcn30_update_soc_for_wm_a,
 	.add_phantom_pipes = dcn32_add_phantom_pipes,
 	.remove_phantom_pipes = dcn32_remove_phantom_pipes,
+	.retain_phantom_pipes = dcn32_retain_phantom_pipes,
 };
 
 static uint32_t read_pipe_fuses(struct dc_context *ctx)

sys/dev/pci/drm/amd/display/dc/dcn32/dcn32_resource.h

@@ -83,6 +83,9 @@ bool dcn32_release_post_bldn_3dlut(
 bool dcn32_remove_phantom_pipes(struct dc *dc,
 		struct dc_state *context);
 
+void dcn32_retain_phantom_pipes(struct dc *dc,
+		struct dc_state *context);
+
 void dcn32_add_phantom_pipes(struct dc *dc,
 		struct dc_state *context,
 		display_e2e_pipe_params_st *pipes,

sys/dev/pci/drm/amd/display/dc/dcn321/dcn321_resource.c

@@ -1619,6 +1619,7 @@ static struct resource_funcs dcn321_res_pool_funcs = {
 	.update_soc_for_wm_a = dcn30_update_soc_for_wm_a,
 	.add_phantom_pipes = dcn32_add_phantom_pipes,
 	.remove_phantom_pipes = dcn32_remove_phantom_pipes,
+	.retain_phantom_pipes = dcn32_retain_phantom_pipes,
 };
 
 static uint32_t read_pipe_fuses(struct dc_context *ctx)

sys/dev/pci/drm/amd/display/dc/inc/core_types.h

@@ -234,6 +234,7 @@ struct resource_funcs {
             unsigned int index);
 
 	bool (*remove_phantom_pipes)(struct dc *dc, struct dc_state *context);
+	void (*retain_phantom_pipes)(struct dc *dc, struct dc_state *context);
 	void (*get_panel_config_defaults)(struct dc_panel_config *panel_config);
 };
 

sys/dev/pci/drm/amd/display/dc/inc/hw/timing_generator.h

@@ -185,6 +185,7 @@ struct timing_generator_funcs {
 #ifdef CONFIG_DRM_AMD_DC_DCN
 	void (*phantom_crtc_post_enable)(struct timing_generator *tg);
 #endif
+	void (*disable_phantom_crtc)(struct timing_generator *tg);
 	bool (*immediate_disable_crtc)(struct timing_generator *tg);
 	bool (*is_counter_moving)(struct timing_generator *tg);
 	void (*get_position)(struct timing_generator *tg,

sys/dev/pci/drm/amd/include/atomfirmware.h

@@ -705,20 +705,65 @@ struct atom_gpio_pin_lut_v2_1
 };
 
 
-/* 
-  ***************************************************************************
-    Data Table vram_usagebyfirmware  structure
-  ***************************************************************************
-*/
+/*
+ * VBIOS/PRE-OS always reserve a FB region at the top of frame buffer. driver should not write
+ * access that region. driver can allocate their own reservation region as long as it does not
+ * overlap firwmare's reservation region.
+ * if (pre-NV1X) atom data table firmwareInfoTable version < 3.3:
+ * in this case, atom data table vram_usagebyfirmwareTable version always <= 2.1
+ *   if VBIOS/UEFI GOP is posted:
+ *     VBIOS/UEFIGOP update used_by_firmware_in_kb = total reserved size by VBIOS
+ *     update start_address_in_kb = total_mem_size_in_kb - used_by_firmware_in_kb;
+ *     ( total_mem_size_in_kb = reg(CONFIG_MEMSIZE)<<10)
+ *     driver can allocate driver reservation region under firmware reservation,
+ *     used_by_driver_in_kb = driver reservation size
+ *     driver reservation start address =  (start_address_in_kb - used_by_driver_in_kb)
+ *     Comment1[hchan]: There is only one reservation at the beginning of the FB reserved by
+ *     host driver. Host driver would overwrite the table with the following
+ *     used_by_firmware_in_kb = total reserved size for pf-vf info exchange and
+ *     set SRIOV_MSG_SHARE_RESERVATION mask start_address_in_kb = 0
+ *   else there is no VBIOS reservation region:
+ *     driver must allocate driver reservation region at top of FB.
+ *     driver set used_by_driver_in_kb = driver reservation size
+ *     driver reservation start address =  (total_mem_size_in_kb - used_by_driver_in_kb)
+ *     same as Comment1
+ * else (NV1X and after):
+ *   if VBIOS/UEFI GOP is posted:
+ *     VBIOS/UEFIGOP update:
+ *       used_by_firmware_in_kb = atom_firmware_Info_v3_3.fw_reserved_size_in_kb;
+ *       start_address_in_kb = total_mem_size_in_kb - used_by_firmware_in_kb;
+ *       (total_mem_size_in_kb = reg(CONFIG_MEMSIZE)<<10)
+ *   if vram_usagebyfirmwareTable version <= 2.1:
+ *     driver can allocate driver reservation region under firmware reservation,
+ *     driver set used_by_driver_in_kb = driver reservation size
+ *     driver reservation start address = start_address_in_kb - used_by_driver_in_kb
+ *     same as Comment1
+ *   else driver can:
+ *     allocate it reservation any place as long as it does overlap pre-OS FW reservation area
+ *     set used_by_driver_region0_in_kb = driver reservation size
+ *     set driver_region0_start_address_in_kb =  driver reservation region start address
+ *     Comment2[hchan]: Host driver can set used_by_firmware_in_kb and start_address_in_kb to
+ *     zero as the reservation for VF as it doesn’t exist.  And Host driver should also
+ *     update atom_firmware_Info table to remove the same VBIOS reservation as well.
+ */
 
 struct vram_usagebyfirmware_v2_1
 {
-  struct  atom_common_table_header  table_header;
-  uint32_t  start_address_in_kb;
-  uint16_t  used_by_firmware_in_kb;
-  uint16_t  used_by_driver_in_kb; 
+	struct  atom_common_table_header  table_header;
+	uint32_t  start_address_in_kb;
+	uint16_t  used_by_firmware_in_kb;
+	uint16_t  used_by_driver_in_kb;
 };
 
+struct vram_usagebyfirmware_v2_2 {
+	struct  atom_common_table_header  table_header;
+	uint32_t  fw_region_start_address_in_kb;
+	uint16_t  used_by_firmware_in_kb;
+	uint16_t  reserved;
+	uint32_t  driver_region0_start_address_in_kb;
+	uint32_t  used_by_driver_region0_in_kb;
+	uint32_t  reserved32[7];
+};
 
 /* 
   ***************************************************************************

sys/dev/pci/drm/amd/include/kgd_pp_interface.h

@@ -139,6 +139,8 @@ enum amd_pp_sensors {
 	AMDGPU_PP_SENSOR_MIN_FAN_RPM,
 	AMDGPU_PP_SENSOR_MAX_FAN_RPM,
 	AMDGPU_PP_SENSOR_VCN_POWER_STATE,
+	AMDGPU_PP_SENSOR_PEAK_PSTATE_SCLK,
+	AMDGPU_PP_SENSOR_PEAK_PSTATE_MCLK,
 };
 
 enum amd_pp_task {

sys/dev/pci/drm/amd/pm/inc/amdgpu_dpm.h

@@ -89,6 +89,8 @@ struct amdgpu_dpm_thermal {
 	int                max_mem_crit_temp;
 	/* memory max emergency(shutdown) temp */
 	int                max_mem_emergency_temp;
+	/* SWCTF threshold */
+	int                sw_ctf_threshold;
 	/* was last interrupt low to high or high to low */
 	bool               high_to_low;
 	/* interrupt source */

sys/dev/pci/drm/amd/pm/powerplay/amd_powerplay.c

@@ -26,6 +26,7 @@
 #include <linux/gfp.h>
 #include <linux/slab.h>
 #include <linux/firmware.h>
+#include <linux/reboot.h>
 #include "amd_shared.h"
 #include "amd_powerplay.h"
 #include "power_state.h"
@@ -91,6 +92,45 @@ static int pp_early_init(void *handle)
 	return 0;
 }
 
+static void pp_swctf_delayed_work_handler(struct work_struct *work)
+{
+	struct pp_hwmgr *hwmgr =
+		container_of(work, struct pp_hwmgr, swctf_delayed_work.work);
+	struct amdgpu_device *adev = hwmgr->adev;
+	struct amdgpu_dpm_thermal *range =
+				&adev->pm.dpm.thermal;
+	uint32_t gpu_temperature, size;
+	int ret;
+
+	/*
+	 * If the hotspot/edge temperature is confirmed as below SW CTF setting point
+	 * after the delay enforced, nothing will be done.
+	 * Otherwise, a graceful shutdown will be performed to prevent further damage.
+	 */
+	if (range->sw_ctf_threshold &&
+	    hwmgr->hwmgr_func->read_sensor) {
+		ret = hwmgr->hwmgr_func->read_sensor(hwmgr,
+						     AMDGPU_PP_SENSOR_HOTSPOT_TEMP,
+						     &gpu_temperature,
+						     &size);
+		/*
+		 * For some legacy ASICs, hotspot temperature retrieving might be not
+		 * supported. Check the edge temperature instead then.
+		 */
+		if (ret == -EOPNOTSUPP)
+			ret = hwmgr->hwmgr_func->read_sensor(hwmgr,
+							     AMDGPU_PP_SENSOR_EDGE_TEMP,
+							     &gpu_temperature,
+							     &size);
+		if (!ret && gpu_temperature / 1000 < range->sw_ctf_threshold)
+			return;
+	}
+
+	dev_emerg(adev->dev, "ERROR: GPU over temperature range(SW CTF) detected!\n");
+	dev_emerg(adev->dev, "ERROR: System is going to shutdown due to GPU SW CTF!\n");
+	orderly_poweroff(true);
+}
+
 static int pp_sw_init(void *handle)
 {
 	struct amdgpu_device *adev = handle;
@@ -101,6 +141,10 @@ static int pp_sw_init(void *handle)
 
 	pr_debug("powerplay sw init %s\n", ret ? "failed" : "successfully");
 
+	if (!ret)
+		INIT_DELAYED_WORK(&hwmgr->swctf_delayed_work,
+				  pp_swctf_delayed_work_handler);
+
 	return ret;
 }
 
@@ -136,6 +180,8 @@ static int pp_hw_fini(void *handle)
 	struct amdgpu_device *adev = handle;
 	struct pp_hwmgr *hwmgr = adev->powerplay.pp_handle;
 
+	cancel_delayed_work_sync(&hwmgr->swctf_delayed_work);
+
 	hwmgr_hw_fini(hwmgr);
 
 	return 0;
@@ -222,6 +268,8 @@ static int pp_suspend(void *handle)
 	struct amdgpu_device *adev = handle;
 	struct pp_hwmgr *hwmgr = adev->powerplay.pp_handle;
 
+	cancel_delayed_work_sync(&hwmgr->swctf_delayed_work);
+
 	return hwmgr_suspend(hwmgr);
 }
 
@@ -769,10 +817,16 @@ static int pp_dpm_read_sensor(void *handle, int idx,
 
 	switch (idx) {
 	case AMDGPU_PP_SENSOR_STABLE_PSTATE_SCLK:
-		*((uint32_t *)value) = hwmgr->pstate_sclk;
+		*((uint32_t *)value) = hwmgr->pstate_sclk * 100;
 		return 0;
 	case AMDGPU_PP_SENSOR_STABLE_PSTATE_MCLK:
-		*((uint32_t *)value) = hwmgr->pstate_mclk;
+		*((uint32_t *)value) = hwmgr->pstate_mclk * 100;
+		return 0;
+	case AMDGPU_PP_SENSOR_PEAK_PSTATE_SCLK:
+		*((uint32_t *)value) = hwmgr->pstate_sclk_peak * 100;
+		return 0;
+	case AMDGPU_PP_SENSOR_PEAK_PSTATE_MCLK:
+		*((uint32_t *)value) = hwmgr->pstate_mclk_peak * 100;
 		return 0;
 	case AMDGPU_PP_SENSOR_MIN_FAN_RPM:
 		*((uint32_t *)value) = hwmgr->thermal_controller.fanInfo.ulMinRPM;

sys/dev/pci/drm/amd/pm/powerplay/hwmgr/hardwaremanager.c

@@ -241,7 +241,8 @@ int phm_start_thermal_controller(struct pp_hwmgr *hwmgr)
 		TEMP_RANGE_MAX,
 		TEMP_RANGE_MIN,
 		TEMP_RANGE_MAX,
-		TEMP_RANGE_MAX};
+		TEMP_RANGE_MAX,
+		0};
 	struct amdgpu_device *adev = hwmgr->adev;
 
 	if (!hwmgr->not_vf)
@@ -265,6 +266,7 @@ int phm_start_thermal_controller(struct pp_hwmgr *hwmgr)
 	adev->pm.dpm.thermal.min_mem_temp = range.mem_min;
 	adev->pm.dpm.thermal.max_mem_crit_temp = range.mem_crit_max;
 	adev->pm.dpm.thermal.max_mem_emergency_temp = range.mem_emergency_max;
+	adev->pm.dpm.thermal.sw_ctf_threshold = range.sw_ctf_threshold;
 
 	return ret;
 }

sys/dev/pci/drm/amd/pm/powerplay/hwmgr/smu10_hwmgr.c

@@ -375,6 +375,17 @@ static int smu10_enable_gfx_off(struct pp_hwmgr *hwmgr)
 	return 0;
 }
 
+static void smu10_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
+{
+	hwmgr->pstate_sclk = SMU10_UMD_PSTATE_GFXCLK;
+	hwmgr->pstate_mclk = SMU10_UMD_PSTATE_FCLK;
+
+	smum_send_msg_to_smc(hwmgr,
+			     PPSMC_MSG_GetMaxGfxclkFrequency,
+			     &hwmgr->pstate_sclk_peak);
+	hwmgr->pstate_mclk_peak = SMU10_UMD_PSTATE_PEAK_FCLK;
+}
+
 static int smu10_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 {
 	struct amdgpu_device *adev = hwmgr->adev;
@@ -398,6 +409,8 @@ static int smu10_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 			return ret;
 	}
 
+	smu10_populate_umdpstate_clocks(hwmgr);
+
 	return 0;
 }
 
@@ -574,9 +587,6 @@ static int smu10_hwmgr_backend_init(struct pp_hwmgr *hwmgr)
 
 	hwmgr->platform_descriptor.minimumClocksReductionPercentage = 50;
 
-	hwmgr->pstate_sclk = SMU10_UMD_PSTATE_GFXCLK * 100;
-	hwmgr->pstate_mclk = SMU10_UMD_PSTATE_FCLK * 100;
-
 	/* enable the pp_od_clk_voltage sysfs file */
 	hwmgr->od_enabled = 1;
 	/* disabled fine grain tuning function by default */

sys/dev/pci/drm/amd/pm/powerplay/hwmgr/smu7_hwmgr.c

@@ -1501,6 +1501,67 @@ static int smu7_populate_edc_leakage_registers(struct pp_hwmgr *hwmgr)
 	return ret;
 }
 
+static void smu7_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
+{
+	struct smu7_hwmgr *data = (struct smu7_hwmgr *)(hwmgr->backend);
+	struct smu7_dpm_table *golden_dpm_table = &data->golden_dpm_table;
+	int32_t tmp_sclk, count, percentage;
+
+	if (golden_dpm_table->mclk_table.count == 1) {
+		percentage = 70;
+		hwmgr->pstate_mclk = golden_dpm_table->mclk_table.dpm_levels[0].value;
+	} else {
+		percentage = 100 * golden_dpm_table->sclk_table.dpm_levels[golden_dpm_table->sclk_table.count - 1].value /
+				golden_dpm_table->mclk_table.dpm_levels[golden_dpm_table->mclk_table.count - 1].value;
+		hwmgr->pstate_mclk = golden_dpm_table->mclk_table.dpm_levels[golden_dpm_table->mclk_table.count - 2].value;
+	}
+
+	tmp_sclk = hwmgr->pstate_mclk * percentage / 100;
+
+	if (hwmgr->pp_table_version == PP_TABLE_V0) {
+		struct phm_clock_voltage_dependency_table *vddc_dependency_on_sclk =
+			hwmgr->dyn_state.vddc_dependency_on_sclk;
+
+		for (count = vddc_dependency_on_sclk->count - 1; count >= 0; count--) {
+			if (tmp_sclk >= vddc_dependency_on_sclk->entries[count].clk) {
+				hwmgr->pstate_sclk = vddc_dependency_on_sclk->entries[count].clk;
+				break;
+			}
+		}
+		if (count < 0)
+			hwmgr->pstate_sclk = vddc_dependency_on_sclk->entries[0].clk;
+
+		hwmgr->pstate_sclk_peak =
+			vddc_dependency_on_sclk->entries[vddc_dependency_on_sclk->count - 1].clk;
+	} else if (hwmgr->pp_table_version == PP_TABLE_V1) {
+		struct phm_ppt_v1_information *table_info =
+			(struct phm_ppt_v1_information *)(hwmgr->pptable);
+		struct phm_ppt_v1_clock_voltage_dependency_table *vdd_dep_on_sclk =
+			table_info->vdd_dep_on_sclk;
+
+		for (count = vdd_dep_on_sclk->count - 1; count >= 0; count--) {
+			if (tmp_sclk >= vdd_dep_on_sclk->entries[count].clk) {
+				hwmgr->pstate_sclk = vdd_dep_on_sclk->entries[count].clk;
+				break;
+			}
+		}
+		if (count < 0)
+			hwmgr->pstate_sclk = vdd_dep_on_sclk->entries[0].clk;
+
+		hwmgr->pstate_sclk_peak =
+			vdd_dep_on_sclk->entries[vdd_dep_on_sclk->count - 1].clk;
+	}
+
+	hwmgr->pstate_mclk_peak =
+		golden_dpm_table->mclk_table.dpm_levels[golden_dpm_table->mclk_table.count - 1].value;
+
+	/* make sure the output is in Mhz */
+	hwmgr->pstate_sclk /= 100;
+	hwmgr->pstate_mclk /= 100;
+	hwmgr->pstate_sclk_peak /= 100;
+	hwmgr->pstate_mclk_peak /= 100;
+}
+
 static int smu7_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 {
 	int tmp_result = 0;
@@ -1625,6 +1686,8 @@ static int smu7_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 	PP_ASSERT_WITH_CODE((0 == tmp_result),
 			"pcie performance request failed!", result = tmp_result);
 
+	smu7_populate_umdpstate_clocks(hwmgr);
+
 	return 0;
 }
 
@@ -3143,15 +3206,12 @@ static int smu7_get_profiling_clk(struct pp_hwmgr *hwmgr, enum amd_dpm_forced_le
 		for (count = hwmgr->dyn_state.vddc_dependency_on_sclk->count-1;
 			count >= 0; count--) {
 			if (tmp_sclk >= hwmgr->dyn_state.vddc_dependency_on_sclk->entries[count].clk) {
-				tmp_sclk = hwmgr->dyn_state.vddc_dependency_on_sclk->entries[count].clk;
 				*sclk_mask = count;
 				break;
 			}
 		}
-		if (count < 0 || level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK) {
+		if (count < 0 || level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK)
 			*sclk_mask = 0;
-			tmp_sclk = hwmgr->dyn_state.vddc_dependency_on_sclk->entries[0].clk;
-		}
 
 		if (level == AMD_DPM_FORCED_LEVEL_PROFILE_PEAK)
 			*sclk_mask = hwmgr->dyn_state.vddc_dependency_on_sclk->count-1;
@@ -3161,15 +3221,12 @@ static int smu7_get_profiling_clk(struct pp_hwmgr *hwmgr, enum amd_dpm_forced_le
 
 		for (count = table_info->vdd_dep_on_sclk->count-1; count >= 0; count--) {
 			if (tmp_sclk >= table_info->vdd_dep_on_sclk->entries[count].clk) {
-				tmp_sclk = table_info->vdd_dep_on_sclk->entries[count].clk;
 				*sclk_mask = count;
 				break;
 			}
 		}
-		if (count < 0 || level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK) {
+		if (count < 0 || level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK)
 			*sclk_mask = 0;
-			tmp_sclk =  table_info->vdd_dep_on_sclk->entries[0].clk;
-		}
 
 		if (level == AMD_DPM_FORCED_LEVEL_PROFILE_PEAK)
 			*sclk_mask = table_info->vdd_dep_on_sclk->count - 1;
@@ -3181,8 +3238,6 @@ static int smu7_get_profiling_clk(struct pp_hwmgr *hwmgr, enum amd_dpm_forced_le
 		*mclk_mask = golden_dpm_table->mclk_table.count - 1;
 
 	*pcie_mask = data->dpm_table.pcie_speed_table.count - 1;
-	hwmgr->pstate_sclk = tmp_sclk;
-	hwmgr->pstate_mclk = tmp_mclk;
 
 	return 0;
 }
@@ -3195,9 +3250,6 @@ static int smu7_force_dpm_level(struct pp_hwmgr *hwmgr,
 	uint32_t mclk_mask = 0;
 	uint32_t pcie_mask = 0;
 
-	if (hwmgr->pstate_sclk == 0)
-		smu7_get_profiling_clk(hwmgr, level, &sclk_mask, &mclk_mask, &pcie_mask);
-
 	switch (level) {
 	case AMD_DPM_FORCED_LEVEL_HIGH:
 		ret = smu7_force_dpm_highest(hwmgr);
@@ -5381,6 +5433,8 @@ static int smu7_get_thermal_temperature_range(struct pp_hwmgr *hwmgr,
 		thermal_data->max = data->thermal_temp_setting.temperature_shutdown *
 			PP_TEMPERATURE_UNITS_PER_CENTIGRADES;
 
+	thermal_data->sw_ctf_threshold = thermal_data->max;
+
 	return 0;
 }
 

sys/dev/pci/drm/amd/pm/powerplay/hwmgr/smu8_hwmgr.c

@@ -1016,6 +1016,18 @@ static void smu8_reset_acp_boot_level(struct pp_hwmgr *hwmgr)
 	data->acp_boot_level = 0xff;
 }
 
+static void smu8_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
+{
+	struct phm_clock_voltage_dependency_table *table =
+				hwmgr->dyn_state.vddc_dependency_on_sclk;
+
+	hwmgr->pstate_sclk = table->entries[0].clk / 100;
+	hwmgr->pstate_mclk = 0;
+
+	hwmgr->pstate_sclk_peak = table->entries[table->count - 1].clk / 100;
+	hwmgr->pstate_mclk_peak = 0;
+}
+
 static int smu8_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 {
 	smu8_program_voting_clients(hwmgr);
@@ -1024,6 +1036,8 @@ static int smu8_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 	smu8_program_bootup_state(hwmgr);
 	smu8_reset_acp_boot_level(hwmgr);
 
+	smu8_populate_umdpstate_clocks(hwmgr);
+
 	return 0;
 }
 
@@ -1167,8 +1181,6 @@ static int smu8_phm_unforce_dpm_levels(struct pp_hwmgr *hwmgr)
 
 	data->sclk_dpm.soft_min_clk = table->entries[0].clk;
 	data->sclk_dpm.hard_min_clk = table->entries[0].clk;
-	hwmgr->pstate_sclk = table->entries[0].clk;
-	hwmgr->pstate_mclk = 0;
 
 	level = smu8_get_max_sclk_level(hwmgr) - 1;
 

sys/dev/pci/drm/amd/pm/powerplay/hwmgr/smu_helper.c

@@ -603,21 +603,17 @@ int phm_irq_process(struct amdgpu_device *adev,
 			   struct amdgpu_irq_src *source,
 			   struct amdgpu_iv_entry *entry)
 {
+	struct pp_hwmgr *hwmgr = adev->powerplay.pp_handle;
 	uint32_t client_id = entry->client_id;
 	uint32_t src_id = entry->src_id;
 
 	if (client_id == AMDGPU_IRQ_CLIENTID_LEGACY) {
 		if (src_id == VISLANDS30_IV_SRCID_CG_TSS_THERMAL_LOW_TO_HIGH) {
-			dev_emerg(adev->dev, "ERROR: GPU over temperature range(SW CTF) detected!\n");
-			/*
-			 * SW CTF just occurred.
-			 * Try to do a graceful shutdown to prevent further damage.
-			 */
-			dev_emerg(adev->dev, "ERROR: System is going to shutdown due to GPU SW CTF!\n");
-			orderly_poweroff(true);
-		} else if (src_id == VISLANDS30_IV_SRCID_CG_TSS_THERMAL_HIGH_TO_LOW)
+			schedule_delayed_work(&hwmgr->swctf_delayed_work,
+					      msecs_to_jiffies(AMDGPU_SWCTF_EXTRA_DELAY));
+		} else if (src_id == VISLANDS30_IV_SRCID_CG_TSS_THERMAL_HIGH_TO_LOW) {
 			dev_emerg(adev->dev, "ERROR: GPU under temperature range detected!\n");
-		else if (src_id == VISLANDS30_IV_SRCID_GPIO_19) {
+		} else if (src_id == VISLANDS30_IV_SRCID_GPIO_19) {
 			dev_emerg(adev->dev, "ERROR: GPU HW Critical Temperature Fault(aka CTF) detected!\n");
 			/*
 			 * HW CTF just occurred. Shutdown to prevent further damage.
@@ -626,15 +622,10 @@ int phm_irq_process(struct amdgpu_device *adev,
 			orderly_poweroff(true);
 		}
 	} else if (client_id == SOC15_IH_CLIENTID_THM) {
-		if (src_id == 0) {
-			dev_emerg(adev->dev, "ERROR: GPU over temperature range(SW CTF) detected!\n");
-			/*
-			 * SW CTF just occurred.
-			 * Try to do a graceful shutdown to prevent further damage.
-			 */
-			dev_emerg(adev->dev, "ERROR: System is going to shutdown due to GPU SW CTF!\n");
-			orderly_poweroff(true);
-		} else
+		if (src_id == 0)
+			schedule_delayed_work(&hwmgr->swctf_delayed_work,
+					      msecs_to_jiffies(AMDGPU_SWCTF_EXTRA_DELAY));
+		else
 			dev_emerg(adev->dev, "ERROR: GPU under temperature range detected!\n");
 	} else if (client_id == SOC15_IH_CLIENTID_ROM_SMUIO) {
 		dev_emerg(adev->dev, "ERROR: GPU HW Critical Temperature Fault(aka CTF) detected!\n");

sys/dev/pci/drm/amd/pm/powerplay/hwmgr/vega10_hwmgr.c

@@ -3008,6 +3008,30 @@ static int vega10_enable_disable_PCC_limit_feature(struct pp_hwmgr *hwmgr, bool
 	return 0;
 }
 
+static void vega10_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
+{
+	struct phm_ppt_v2_information *table_info =
+			(struct phm_ppt_v2_information *)(hwmgr->pptable);
+
+	if (table_info->vdd_dep_on_sclk->count > VEGA10_UMD_PSTATE_GFXCLK_LEVEL &&
+	    table_info->vdd_dep_on_mclk->count > VEGA10_UMD_PSTATE_MCLK_LEVEL) {
+		hwmgr->pstate_sclk = table_info->vdd_dep_on_sclk->entries[VEGA10_UMD_PSTATE_GFXCLK_LEVEL].clk;
+		hwmgr->pstate_mclk = table_info->vdd_dep_on_mclk->entries[VEGA10_UMD_PSTATE_MCLK_LEVEL].clk;
+	} else {
+		hwmgr->pstate_sclk = table_info->vdd_dep_on_sclk->entries[0].clk;
+		hwmgr->pstate_mclk = table_info->vdd_dep_on_mclk->entries[0].clk;
+	}
+
+	hwmgr->pstate_sclk_peak = table_info->vdd_dep_on_sclk->entries[table_info->vdd_dep_on_sclk->count - 1].clk;
+	hwmgr->pstate_mclk_peak = table_info->vdd_dep_on_mclk->entries[table_info->vdd_dep_on_mclk->count - 1].clk;
+
+	/* make sure the output is in Mhz */
+	hwmgr->pstate_sclk /= 100;
+	hwmgr->pstate_mclk /= 100;
+	hwmgr->pstate_sclk_peak /= 100;
+	hwmgr->pstate_mclk_peak /= 100;
+}
+
 static int vega10_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 {
 	struct vega10_hwmgr *data = hwmgr->backend;
@@ -3082,6 +3106,8 @@ static int vega10_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 				    result = tmp_result);
 	}
 
+	vega10_populate_umdpstate_clocks(hwmgr);
+
 	return result;
 }
 
@@ -4169,8 +4195,6 @@ static int vega10_get_profiling_clk_mask(struct pp_hwmgr *hwmgr, enum amd_dpm_fo
 		*sclk_mask = VEGA10_UMD_PSTATE_GFXCLK_LEVEL;
 		*soc_mask = VEGA10_UMD_PSTATE_SOCCLK_LEVEL;
 		*mclk_mask = VEGA10_UMD_PSTATE_MCLK_LEVEL;
-		hwmgr->pstate_sclk = table_info->vdd_dep_on_sclk->entries[VEGA10_UMD_PSTATE_GFXCLK_LEVEL].clk;
-		hwmgr->pstate_mclk = table_info->vdd_dep_on_mclk->entries[VEGA10_UMD_PSTATE_MCLK_LEVEL].clk;
 	}
 
 	if (level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK) {
@@ -4281,9 +4305,6 @@ static int vega10_dpm_force_dpm_level(struct pp_hwmgr *hwmgr,
 	uint32_t mclk_mask = 0;
 	uint32_t soc_mask = 0;
 
-	if (hwmgr->pstate_sclk == 0)
-		vega10_get_profiling_clk_mask(hwmgr, level, &sclk_mask, &mclk_mask, &soc_mask);
-
 	switch (level) {
 	case AMD_DPM_FORCED_LEVEL_HIGH:
 		ret = vega10_force_dpm_highest(hwmgr);
@@ -5221,6 +5242,9 @@ static int vega10_get_thermal_temperature_range(struct pp_hwmgr *hwmgr,
 {
 	struct vega10_hwmgr *data = hwmgr->backend;
 	PPTable_t *pp_table = &(data->smc_state_table.pp_table);
+	struct phm_ppt_v2_information *pp_table_info =
+		(struct phm_ppt_v2_information *)(hwmgr->pptable);
+	struct phm_tdp_table *tdp_table = pp_table_info->tdp_table;
 
 	memcpy(thermal_data, &SMU7ThermalWithDelayPolicy[0], sizeof(struct PP_TemperatureRange));
 
@@ -5237,6 +5261,13 @@ static int vega10_get_thermal_temperature_range(struct pp_hwmgr *hwmgr,
 	thermal_data->mem_emergency_max = (pp_table->ThbmLimit + CTF_OFFSET_HBM)*
 		PP_TEMPERATURE_UNITS_PER_CENTIGRADES;
 
+	if (tdp_table->usSoftwareShutdownTemp > pp_table->ThotspotLimit &&
+	    tdp_table->usSoftwareShutdownTemp < VEGA10_THERMAL_MAXIMUM_ALERT_TEMP)
+		thermal_data->sw_ctf_threshold = tdp_table->usSoftwareShutdownTemp;
+	else
+		thermal_data->sw_ctf_threshold = VEGA10_THERMAL_MAXIMUM_ALERT_TEMP;
+	thermal_data->sw_ctf_threshold *= PP_TEMPERATURE_UNITS_PER_CENTIGRADES;
+
 	return 0;
 }
 

sys/dev/pci/drm/amd/pm/powerplay/hwmgr/vega12_hwmgr.c

@@ -1026,6 +1026,25 @@ static int vega12_get_all_clock_ranges(struct pp_hwmgr *hwmgr)
 	return 0;
 }
 
+static void vega12_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
+{
+	struct vega12_hwmgr *data = (struct vega12_hwmgr *)(hwmgr->backend);
+	struct vega12_single_dpm_table *gfx_dpm_table = &(data->dpm_table.gfx_table);
+	struct vega12_single_dpm_table *mem_dpm_table = &(data->dpm_table.mem_table);
+
+	if (gfx_dpm_table->count > VEGA12_UMD_PSTATE_GFXCLK_LEVEL &&
+	    mem_dpm_table->count > VEGA12_UMD_PSTATE_MCLK_LEVEL) {
+		hwmgr->pstate_sclk = gfx_dpm_table->dpm_levels[VEGA12_UMD_PSTATE_GFXCLK_LEVEL].value;
+		hwmgr->pstate_mclk = mem_dpm_table->dpm_levels[VEGA12_UMD_PSTATE_MCLK_LEVEL].value;
+	} else {
+		hwmgr->pstate_sclk = gfx_dpm_table->dpm_levels[0].value;
+		hwmgr->pstate_mclk = mem_dpm_table->dpm_levels[0].value;
+	}
+
+	hwmgr->pstate_sclk_peak = gfx_dpm_table->dpm_levels[gfx_dpm_table->count].value;
+	hwmgr->pstate_mclk_peak = mem_dpm_table->dpm_levels[mem_dpm_table->count].value;
+}
+
 static int vega12_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 {
 	int tmp_result, result = 0;
@@ -1077,6 +1096,9 @@ static int vega12_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 	PP_ASSERT_WITH_CODE(!result,
 			"Failed to setup default DPM tables!",
 			return result);
+
+	vega12_populate_umdpstate_clocks(hwmgr);
+
 	return result;
 }
 
@@ -2742,6 +2764,8 @@ static int vega12_notify_cac_buffer_info(struct pp_hwmgr *hwmgr,
 static int vega12_get_thermal_temperature_range(struct pp_hwmgr *hwmgr,
 		struct PP_TemperatureRange *thermal_data)
 {
+	struct phm_ppt_v3_information *pptable_information =
+		(struct phm_ppt_v3_information *)hwmgr->pptable;
 	struct vega12_hwmgr *data =
 			(struct vega12_hwmgr *)(hwmgr->backend);
 	PPTable_t *pp_table = &(data->smc_state_table.pp_table);
@@ -2760,6 +2784,8 @@ static int vega12_get_thermal_temperature_range(struct pp_hwmgr *hwmgr,
 		PP_TEMPERATURE_UNITS_PER_CENTIGRADES;
 	thermal_data->mem_emergency_max = (pp_table->ThbmLimit + CTF_OFFSET_HBM)*
 		PP_TEMPERATURE_UNITS_PER_CENTIGRADES;
+	thermal_data->sw_ctf_threshold = pptable_information->us_software_shutdown_temp *
+		PP_TEMPERATURE_UNITS_PER_CENTIGRADES;
 
 	return 0;
 }

sys/dev/pci/drm/amd/pm/powerplay/hwmgr/vega20_hwmgr.c

@@ -1555,26 +1555,23 @@ static int vega20_set_mclk_od(
 	return 0;
 }
 
-static int vega20_populate_umdpstate_clocks(
-		struct pp_hwmgr *hwmgr)
+static void vega20_populate_umdpstate_clocks(struct pp_hwmgr *hwmgr)
 {
 	struct vega20_hwmgr *data = (struct vega20_hwmgr *)(hwmgr->backend);
 	struct vega20_single_dpm_table *gfx_table = &(data->dpm_table.gfx_table);
 	struct vega20_single_dpm_table *mem_table = &(data->dpm_table.mem_table);
 
-	hwmgr->pstate_sclk = gfx_table->dpm_levels[0].value;
-	hwmgr->pstate_mclk = mem_table->dpm_levels[0].value;
-
 	if (gfx_table->count > VEGA20_UMD_PSTATE_GFXCLK_LEVEL &&
 	    mem_table->count > VEGA20_UMD_PSTATE_MCLK_LEVEL) {
 		hwmgr->pstate_sclk = gfx_table->dpm_levels[VEGA20_UMD_PSTATE_GFXCLK_LEVEL].value;
 		hwmgr->pstate_mclk = mem_table->dpm_levels[VEGA20_UMD_PSTATE_MCLK_LEVEL].value;
+	} else {
+		hwmgr->pstate_sclk = gfx_table->dpm_levels[0].value;
+		hwmgr->pstate_mclk = mem_table->dpm_levels[0].value;
 	}
 
-	hwmgr->pstate_sclk = hwmgr->pstate_sclk * 100;
-	hwmgr->pstate_mclk = hwmgr->pstate_mclk * 100;
-
-	return 0;
+	hwmgr->pstate_sclk_peak = gfx_table->dpm_levels[gfx_table->count - 1].value;
+	hwmgr->pstate_mclk_peak = mem_table->dpm_levels[mem_table->count - 1].value;
 }
 
 static int vega20_get_max_sustainable_clock(struct pp_hwmgr *hwmgr,
@@ -1753,10 +1750,7 @@ static int vega20_enable_dpm_tasks(struct pp_hwmgr *hwmgr)
 			"[EnableDPMTasks] Failed to initialize odn settings!",
 			return result);
 
-	result = vega20_populate_umdpstate_clocks(hwmgr);
-	PP_ASSERT_WITH_CODE(!result,
-			"[EnableDPMTasks] Failed to populate umdpstate clocks!",
-			return result);
+	vega20_populate_umdpstate_clocks(hwmgr);
 
 	result = smum_send_msg_to_smc_with_parameter(hwmgr, PPSMC_MSG_GetPptLimit,
 			POWER_SOURCE_AC << 16, &hwmgr->default_power_limit);
@@ -4213,6 +4207,8 @@ static int vega20_notify_cac_buffer_info(struct pp_hwmgr *hwmgr,
 static int vega20_get_thermal_temperature_range(struct pp_hwmgr *hwmgr,
 		struct PP_TemperatureRange *thermal_data)
 {
+	struct phm_ppt_v3_information *pptable_information =
+		(struct phm_ppt_v3_information *)hwmgr->pptable;
 	struct vega20_hwmgr *data =
 			(struct vega20_hwmgr *)(hwmgr->backend);
 	PPTable_t *pp_table = &(data->smc_state_table.pp_table);
@@ -4231,6 +4227,8 @@ static int vega20_get_thermal_temperature_range(struct pp_hwmgr *hwmgr,
 		PP_TEMPERATURE_UNITS_PER_CENTIGRADES;
 	thermal_data->mem_emergency_max = (pp_table->ThbmLimit + CTF_OFFSET_HBM)*
 		PP_TEMPERATURE_UNITS_PER_CENTIGRADES;
+	thermal_data->sw_ctf_threshold = pptable_information->us_software_shutdown_temp *
+		PP_TEMPERATURE_UNITS_PER_CENTIGRADES;
 
 	return 0;
 }

sys/dev/pci/drm/amd/pm/powerplay/inc/hwmgr.h

@@ -809,6 +809,10 @@ struct pp_hwmgr {
 	uint32_t workload_prority[Workload_Policy_Max];
 	uint32_t workload_setting[Workload_Policy_Max];
 	bool gfxoff_state_changed_by_workload;
+	uint32_t pstate_sclk_peak;
+	uint32_t pstate_mclk_peak;
+
+	struct delayed_work swctf_delayed_work;
 };
 
 int hwmgr_early_init(struct pp_hwmgr *hwmgr);

sys/dev/pci/drm/amd/pm/powerplay/inc/power_state.h

@@ -131,6 +131,7 @@ struct PP_TemperatureRange {
 	int mem_min;
 	int mem_crit_max;
 	int mem_emergency_max;
+	int sw_ctf_threshold;
 };
 
 struct PP_StateValidationBlock {

sys/dev/pci/drm/amd/pm/swsmu/amdgpu_smu.c

@@ -24,6 +24,7 @@
 
 #include <linux/firmware.h>
 #include <linux/pci.h>
+#include <linux/reboot.h>
 
 #include "amdgpu.h"
 #include "amdgpu_smu.h"
@@ -1061,6 +1062,34 @@ static void smu_interrupt_work_fn(struct work_struct *work)
 		smu->ppt_funcs->interrupt_work(smu);
 }
 
+static void smu_swctf_delayed_work_handler(struct work_struct *work)
+{
+	struct smu_context *smu =
+		container_of(work, struct smu_context, swctf_delayed_work.work);
+	struct smu_temperature_range *range =
+				&smu->thermal_range;
+	struct amdgpu_device *adev = smu->adev;
+	uint32_t hotspot_tmp, size;
+
+	/*
+	 * If the hotspot temperature is confirmed as below SW CTF setting point
+	 * after the delay enforced, nothing will be done.
+	 * Otherwise, a graceful shutdown will be performed to prevent further damage.
+	 */
+	if (range->software_shutdown_temp &&
+	    smu->ppt_funcs->read_sensor &&
+	    !smu->ppt_funcs->read_sensor(smu,
+					 AMDGPU_PP_SENSOR_HOTSPOT_TEMP,
+					 &hotspot_tmp,
+					 &size) &&
+	    hotspot_tmp / 1000 < range->software_shutdown_temp)
+		return;
+
+	dev_emerg(adev->dev, "ERROR: GPU over temperature range(SW CTF) detected!\n");
+	dev_emerg(adev->dev, "ERROR: System is going to shutdown due to GPU SW CTF!\n");
+	orderly_poweroff(true);
+}
+
 static int smu_sw_init(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -1109,6 +1138,9 @@ static int smu_sw_init(void *handle)
 		return ret;
 	}
 
+	INIT_DELAYED_WORK(&smu->swctf_delayed_work,
+			  smu_swctf_delayed_work_handler);
+
 	ret = smu_smc_table_sw_init(smu);
 	if (ret) {
 		dev_err(adev->dev, "Failed to sw init smc table!\n");
@@ -1581,6 +1613,8 @@ static int smu_smc_hw_cleanup(struct smu_context *smu)
 		return ret;
 	}
 
+	cancel_delayed_work_sync(&smu->swctf_delayed_work);
+
 	ret = smu_disable_dpms(smu);
 	if (ret) {
 		dev_err(adev->dev, "Fail to disable dpm features!\n");
@@ -2520,6 +2554,14 @@ static int smu_read_sensor(void *handle,
 		*((uint32_t *)data) = pstate_table->uclk_pstate.standard * 100;
 		*size = 4;
 		break;
+	case AMDGPU_PP_SENSOR_PEAK_PSTATE_SCLK:
+		*((uint32_t *)data) = pstate_table->gfxclk_pstate.peak * 100;
+		*size = 4;
+		break;
+	case AMDGPU_PP_SENSOR_PEAK_PSTATE_MCLK:
+		*((uint32_t *)data) = pstate_table->uclk_pstate.peak * 100;
+		*size = 4;
+		break;
 	case AMDGPU_PP_SENSOR_ENABLED_SMC_FEATURES_MASK:
 		ret = smu_feature_get_enabled_mask(smu, (uint64_t *)data);
 		*size = 8;

sys/dev/pci/drm/amd/pm/swsmu/inc/amdgpu_smu.h

@@ -573,6 +573,8 @@ struct smu_context
 	u32 debug_param_reg;
 	u32 debug_msg_reg;
 	u32 debug_resp_reg;
+
+	struct delayed_work		swctf_delayed_work;
 };
 
 struct i2c_adapter;

sys/dev/pci/drm/amd/pm/swsmu/smu11/smu_v11_0.c

@@ -1438,13 +1438,8 @@ static int smu_v11_0_irq_process(struct amdgpu_device *adev,
 	if (client_id == SOC15_IH_CLIENTID_THM) {
 		switch (src_id) {
 		case THM_11_0__SRCID__THM_DIG_THERM_L2H:
-			dev_emerg(adev->dev, "ERROR: GPU over temperature range(SW CTF) detected!\n");
-			/*
-			 * SW CTF just occurred.
-			 * Try to do a graceful shutdown to prevent further damage.
-			 */
-			dev_emerg(adev->dev, "ERROR: System is going to shutdown due to GPU SW CTF!\n");
-			orderly_poweroff(true);
+			schedule_delayed_work(&smu->swctf_delayed_work,
+					      msecs_to_jiffies(AMDGPU_SWCTF_EXTRA_DELAY));
 		break;
 		case THM_11_0__SRCID__THM_DIG_THERM_H2L:
 			dev_emerg(adev->dev, "ERROR: GPU under temperature range detected\n");

sys/dev/pci/drm/amd/pm/swsmu/smu13/smu_v13_0.c

@@ -1386,13 +1386,8 @@ static int smu_v13_0_irq_process(struct amdgpu_device *adev,
 	if (client_id == SOC15_IH_CLIENTID_THM) {
 		switch (src_id) {
 		case THM_11_0__SRCID__THM_DIG_THERM_L2H:
-			dev_emerg(adev->dev, "ERROR: GPU over temperature range(SW CTF) detected!\n");
-			/*
-			 * SW CTF just occurred.
-			 * Try to do a graceful shutdown to prevent further damage.
-			 */
-			dev_emerg(adev->dev, "ERROR: System is going to shutdown due to GPU SW CTF!\n");
-			orderly_poweroff(true);
+			schedule_delayed_work(&smu->swctf_delayed_work,
+					      msecs_to_jiffies(AMDGPU_SWCTF_EXTRA_DELAY));
 			break;
 		case THM_11_0__SRCID__THM_DIG_THERM_H2L:
 			dev_emerg(adev->dev, "ERROR: GPU under temperature range detected\n");

sys/dev/pci/drm/dma-resv.c

@@ -588,13 +588,12 @@ int dma_resv_get_fences(struct dma_resv *obj, enum dma_resv_usage usage,
 #else
 			nfences = kmalloc(count * sizeof(void *),
 						 GFP_KERNEL);
-			if (nfences != NULL && *fences != NULL)
+			if (nfences != NULL && *fences != NULL) {
 				memcpy(nfences, *fences,
 				    (count - 1) *  sizeof(void *));
-			if (nfences) {
 				kfree(*fences);
-				new_fences = nfences;
 			}
+			new_fences = nfences;
 #endif
 			if (count && !new_fences) {
 				kfree(*fences);

sys/dev/pci/drm/i915/display/intel_display.c

@@ -7123,8 +7123,6 @@ static void intel_update_crtc(struct intel_atomic_state *state,
 
 	intel_fbc_update(state, crtc);
 
-	drm_WARN_ON(&i915->drm, !intel_display_power_is_enabled(i915, POWER_DOMAIN_DC_OFF));
-
 	if (!modeset &&
 	    (new_crtc_state->uapi.color_mgmt_changed ||
 	     new_crtc_state->update_pipe))
@@ -7501,28 +7499,8 @@ static void intel_atomic_commit_tail(struct intel_atomic_state *state)
 	drm_atomic_helper_wait_for_dependencies(&state->base);
 	drm_dp_mst_atomic_wait_for_dependencies(&state->base);
 
-	/*
-	 * During full modesets we write a lot of registers, wait
-	 * for PLLs, etc. Doing that while DC states are enabled
-	 * is not a good idea.
-	 *
-	 * During fastsets and other updates we also need to
-	 * disable DC states due to the following scenario:
-	 * 1. DC5 exit and PSR exit happen
-	 * 2. Some or all _noarm() registers are written
-	 * 3. Due to some long delay PSR is re-entered
-	 * 4. DC5 entry -> DMC saves the already written new
-	 *    _noarm() registers and the old not yet written
-	 *    _arm() registers
-	 * 5. DC5 exit -> DMC restores a mixture of old and
-	 *    new register values and arms the update
-	 * 6. PSR exit -> hardware latches a mixture of old and
-	 *    new register values -> corrupted frame, or worse
-	 * 7. New _arm() registers are finally written
-	 * 8. Hardware finally latches a complete set of new
-	 *    register values, and subsequent frames will be OK again
-	 */
-	wakeref = intel_display_power_get(dev_priv, POWER_DOMAIN_DC_OFF);
+	if (state->modeset)
+		wakeref = intel_display_power_get(dev_priv, POWER_DOMAIN_MODESET);
 
 	intel_atomic_prepare_plane_clear_colors(state);
 
@@ -7661,8 +7639,8 @@ static void intel_atomic_commit_tail(struct intel_atomic_state *state)
 		 * the culprit.
 		 */
 		intel_uncore_arm_unclaimed_mmio_detection(&dev_priv->uncore);
+		intel_display_power_put(dev_priv, POWER_DOMAIN_MODESET, wakeref);
 	}
-	intel_display_power_put(dev_priv, POWER_DOMAIN_DC_OFF, wakeref);
 	intel_runtime_pm_put(&dev_priv->runtime_pm, state->wakeref);
 
 	/*

sys/dev/pci/drm/i915/gt/gen8_engine_cs.c

@@ -256,8 +256,8 @@ int gen12_emit_flush_rcs(struct i915_request *rq, u32 mode)
 
 		if (!HAS_FLAT_CCS(rq->engine->i915)) {
 			/* hsdes: 1809175790 */
-			cs = gen12_emit_aux_table_inv(rq->engine->gt,
-						      cs, GEN12_GFX_CCS_AUX_NV);
+			cs = gen12_emit_aux_table_inv(rq->engine->gt, cs,
+						      GEN12_CCS_AUX_INV);
 		}
 
 		*cs++ = preparser_disable(false);
@@ -317,10 +317,10 @@ int gen12_emit_flush_xcs(struct i915_request *rq, u32 mode)
 	if (aux_inv) { /* hsdes: 1809175790 */
 		if (rq->engine->class == VIDEO_DECODE_CLASS)
 			cs = gen12_emit_aux_table_inv(rq->engine->gt,
-						      cs, GEN12_VD0_AUX_NV);
+						      cs, GEN12_VD0_AUX_INV);
 		else
 			cs = gen12_emit_aux_table_inv(rq->engine->gt,
-						      cs, GEN12_VE0_AUX_NV);
+						      cs, GEN12_VE0_AUX_INV);
 	}
 
 	if (mode & EMIT_INVALIDATE)

sys/dev/pci/drm/i915/gt/intel_gt_regs.h

@@ -301,9 +301,11 @@
 #define GEN8_PRIVATE_PAT_HI			_MMIO(0x40e0 + 4)
 #define GEN10_PAT_INDEX(index)			_MMIO(0x40e0 + (index) * 4)
 #define BSD_HWS_PGA_GEN7			_MMIO(0x4180)
-#define GEN12_GFX_CCS_AUX_NV			_MMIO(0x4208)
-#define GEN12_VD0_AUX_NV			_MMIO(0x4218)
-#define GEN12_VD1_AUX_NV			_MMIO(0x4228)
+
+#define GEN12_CCS_AUX_INV			_MMIO(0x4208)
+#define GEN12_VD0_AUX_INV			_MMIO(0x4218)
+#define GEN12_VE0_AUX_INV			_MMIO(0x4238)
+#define GEN12_BCS0_AUX_INV			_MMIO(0x4248)
 
 #define GEN8_RTCR				_MMIO(0x4260)
 #define GEN8_M1TCR				_MMIO(0x4264)
@@ -311,14 +313,12 @@
 #define GEN8_BTCR				_MMIO(0x426c)
 #define GEN8_VTCR				_MMIO(0x4270)
 
-#define GEN12_VD2_AUX_NV			_MMIO(0x4298)
-#define GEN12_VD3_AUX_NV			_MMIO(0x42a8)
-#define GEN12_VE0_AUX_NV			_MMIO(0x4238)
-
 #define BLT_HWS_PGA_GEN7			_MMIO(0x4280)
 
-#define GEN12_VE1_AUX_NV			_MMIO(0x42b8)
+#define GEN12_VD2_AUX_INV			_MMIO(0x4298)
+#define GEN12_CCS0_AUX_INV			_MMIO(0x42c8)
 #define   AUX_INV				REG_BIT(0)
+
 #define VEBOX_HWS_PGA_GEN7			_MMIO(0x4380)
 
 #define GEN12_AUX_ERR_DBG			_MMIO(0x43f4)

sys/dev/pci/drm/i915/gt/intel_lrc.c

@@ -1304,7 +1304,7 @@ gen12_emit_indirect_ctx_rcs(const struct intel_context *ce, u32 *cs)
 	/* hsdes: 1809175790 */
 	if (!HAS_FLAT_CCS(ce->engine->i915))
 		cs = gen12_emit_aux_table_inv(ce->engine->gt,
-					      cs, GEN12_GFX_CCS_AUX_NV);
+					      cs, GEN12_CCS_AUX_INV);
 
 	/* Wa_16014892111 */
 	if (IS_DG2(ce->engine->i915))
@@ -1331,10 +1331,10 @@ gen12_emit_indirect_ctx_xcs(const struct intel_context *ce, u32 *cs)
 	if (!HAS_FLAT_CCS(ce->engine->i915)) {
 		if (ce->engine->class == VIDEO_DECODE_CLASS)
 			cs = gen12_emit_aux_table_inv(ce->engine->gt,
-						      cs, GEN12_VD0_AUX_NV);
+						      cs, GEN12_VD0_AUX_INV);
 		else if (ce->engine->class == VIDEO_ENHANCEMENT_CLASS)
 			cs = gen12_emit_aux_table_inv(ce->engine->gt,
-						      cs, GEN12_VE0_AUX_NV);
+						      cs, GEN12_VE0_AUX_INV);
 	}
 
 	return cs;

sys/dev/pci/drm/i915/i915_active.c

@@ -461,8 +461,11 @@ int i915_active_add_request(struct i915_active *ref, struct i915_request *rq)
 		}
 	} while (unlikely(is_barrier(active)));
 
-	if (!__i915_active_fence_set(active, fence))
+	fence = __i915_active_fence_set(active, fence);
+	if (!fence)
 		__i915_active_acquire(ref);
+	else
+		dma_fence_put(fence);
 
 out:
 	i915_active_release(ref);
@@ -481,13 +484,9 @@ __i915_active_set_fence(struct i915_active *ref,
 		return NULL;
 	}
 
-	rcu_read_lock();
 	prev = __i915_active_fence_set(active, fence);
-	if (prev)
-		prev = dma_fence_get_rcu(prev);
-	else
+	if (!prev)
 		__i915_active_acquire(ref);
-	rcu_read_unlock();
 
 	return prev;
 }
@@ -1043,10 +1042,11 @@ void i915_request_add_active_barriers(struct i915_request *rq)
  *
  * Records the new @fence as the last active fence along its timeline in
  * this active tracker, moving the tracking callbacks from the previous
- * fence onto this one. Returns the previous fence (if not already completed),
- * which the caller must ensure is executed before the new fence. To ensure
- * that the order of fences within the timeline of the i915_active_fence is
- * understood, it should be locked by the caller.
+ * fence onto this one. Gets and returns a reference to the previous fence
+ * (if not already completed), which the caller must put after making sure
+ * that it is executed before the new fence. To ensure that the order of
+ * fences within the timeline of the i915_active_fence is understood, it
+ * should be locked by the caller.
  */
 struct dma_fence *
 __i915_active_fence_set(struct i915_active_fence *active,
@@ -1055,7 +1055,23 @@ __i915_active_fence_set(struct i915_active_fence *active,
 	struct dma_fence *prev;
 	unsigned long flags;
 
-	if (fence == rcu_access_pointer(active->fence))
+	/*
+	 * In case of fences embedded in i915_requests, their memory is
+	 * SLAB_FAILSAFE_BY_RCU, then it can be reused right after release
+	 * by new requests.  Then, there is a risk of passing back a pointer
+	 * to a new, completely unrelated fence that reuses the same memory
+	 * while tracked under a different active tracker.  Combined with i915
+	 * perf open/close operations that build await dependencies between
+	 * engine kernel context requests and user requests from different
+	 * timelines, this can lead to dependency loops and infinite waits.
+	 *
+	 * As a countermeasure, we try to get a reference to the active->fence
+	 * first, so if we succeed and pass it back to our user then it is not
+	 * released and potentially reused by an unrelated request before the
+	 * user has a chance to set up an await dependency on it.
+	 */
+	prev = i915_active_fence_get(active);
+	if (fence == prev)
 		return fence;
 
 	GEM_BUG_ON(test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags));
@@ -1064,27 +1080,56 @@ __i915_active_fence_set(struct i915_active_fence *active,
 	 * Consider that we have two threads arriving (A and B), with
 	 * C already resident as the active->fence.
 	 *
-	 * A does the xchg first, and so it sees C or NULL depending
-	 * on the timing of the interrupt handler. If it is NULL, the
-	 * previous fence must have been signaled and we know that
-	 * we are first on the timeline. If it is still present,
-	 * we acquire the lock on that fence and serialise with the interrupt
-	 * handler, in the process removing it from any future interrupt
-	 * callback. A will then wait on C before executing (if present).
-	 *
-	 * As B is second, it sees A as the previous fence and so waits for
-	 * it to complete its transition and takes over the occupancy for
-	 * itself -- remembering that it needs to wait on A before executing.
+	 * Both A and B have got a reference to C or NULL, depending on the
+	 * timing of the interrupt handler.  Let's assume that if A has got C
+	 * then it has locked C first (before B).
 	 *
 	 * Note the strong ordering of the timeline also provides consistent
 	 * nesting rules for the fence->lock; the inner lock is always the
 	 * older lock.
 	 */
 	spin_lock_irqsave(fence->lock, flags);
-	prev = xchg(__active_fence_slot(active), fence);
-	if (prev) {
-		GEM_BUG_ON(prev == fence);
+	if (prev)
 		spin_lock_nested(prev->lock, SINGLE_DEPTH_NESTING);
+
+	/*
+	 * A does the cmpxchg first, and so it sees C or NULL, as before, or
+	 * something else, depending on the timing of other threads and/or
+	 * interrupt handler.  If not the same as before then A unlocks C if
+	 * applicable and retries, starting from an attempt to get a new
+	 * active->fence.  Meanwhile, B follows the same path as A.
+	 * Once A succeeds with cmpxch, B fails again, retires, gets A from
+	 * active->fence, locks it as soon as A completes, and possibly
+	 * succeeds with cmpxchg.
+	 */
+	while (cmpxchg(__active_fence_slot(active), prev, fence) != prev) {
+		if (prev) {
+			spin_unlock(prev->lock);
+			dma_fence_put(prev);
+		}
+		spin_unlock_irqrestore(fence->lock, flags);
+
+		prev = i915_active_fence_get(active);
+		GEM_BUG_ON(prev == fence);
+
+		spin_lock_irqsave(fence->lock, flags);
+		if (prev)
+			spin_lock_nested(prev->lock, SINGLE_DEPTH_NESTING);
+	}
+
+	/*
+	 * If prev is NULL then the previous fence must have been signaled
+	 * and we know that we are first on the timeline.  If it is still
+	 * present then, having the lock on that fence already acquired, we
+	 * serialise with the interrupt handler, in the process of removing it
+	 * from any future interrupt callback.  A will then wait on C before
+	 * executing (if present).
+	 *
+	 * As B is second, it sees A as the previous fence and so waits for
+	 * it to complete its transition and takes over the occupancy for
+	 * itself -- remembering that it needs to wait on A before executing.
+	 */
+	if (prev) {
 		__list_del_entry(&active->cb.node);
 		spin_unlock(prev->lock); /* serialise with prev->cb_list */
 	}
@@ -1101,11 +1146,7 @@ int i915_active_fence_set(struct i915_active_fence *active,
 	int err = 0;
 
 	/* Must maintain timeline ordering wrt previous active requests */
-	rcu_read_lock();
 	fence = __i915_active_fence_set(active, &rq->fence);
-	if (fence) /* but the previous fence may not belong to that timeline! */
-		fence = dma_fence_get_rcu(fence);
-	rcu_read_unlock();
 	if (fence) {
 		err = i915_request_await_dma_fence(rq, fence);
 		dma_fence_put(fence);

sys/dev/pci/drm/i915/i915_request.c

@@ -1728,6 +1728,11 @@ __i915_request_ensure_parallel_ordering(struct i915_request *rq,
 
 	request_to_parent(rq)->parallel.last_rq = i915_request_get(rq);
 
+	/*
+	 * Users have to put a reference potentially got by
+	 * __i915_active_fence_set() to the returned request
+	 * when no longer needed
+	 */
 	return to_request(__i915_active_fence_set(&timeline->last_request,
 						  &rq->fence));
 }
@@ -1774,6 +1779,10 @@ __i915_request_ensure_ordering(struct i915_request *rq,
 							 0);
 	}
 
+	/*
+	 * Users have to put the reference to prev potentially got
+	 * by __i915_active_fence_set() when no longer needed
+	 */
 	return prev;
 }
 
@@ -1817,6 +1826,8 @@ __i915_request_add_to_timeline(struct i915_request *rq)
 		prev = __i915_request_ensure_ordering(rq, timeline);
 	else
 		prev = __i915_request_ensure_parallel_ordering(rq, timeline);
+	if (prev)
+		i915_request_put(prev);
 
 	/*
 	 * Make sure that no request gazumped us - if it was allocated after

sys/dev/pci/drm/ttm/ttm_bo.c

@@ -552,7 +552,8 @@ static bool ttm_bo_evict_swapout_allowable(struct ttm_buffer_object *bo,
 
 	if (bo->pin_count) {
 		*locked = false;
-		*busy = false;
+		if (busy)
+			*busy = false;
 		return false;
 	}
 

sys/dev/pci/if_aq_pci.c

@@ -1,4 +1,4 @@
-/* $OpenBSD: if_aq_pci.c,v 1.22 2023/05/02 12:32:22 kettenis Exp $ */
+/* $OpenBSD: if_aq_pci.c,v 1.23 2023/08/15 08:27:30 miod Exp $ */
 /*	$NetBSD: if_aq.c,v 1.27 2021/06/16 00:21:18 riastradh Exp $	*/
 
 /*
@@ -175,7 +175,7 @@
 #define AQ_INTR_CTRL_IRQMODE_MSIX		2
 #define  AQ_INTR_CTRL_MULTIVEC			(1 << 2)
 #define  AQ_INTR_CTRL_RESET_DIS			(1 << 29)
-#define  AQ_INTR_CTRL_RESET_IRQ			(1 << 31)
+#define  AQ_INTR_CTRL_RESET_IRQ			(1U << 31)
 #define AQ_MBOXIF_POWER_GATING_CONTROL_REG	0x32a8
 
 #define FW_MPI_MBOX_ADDR_REG			0x0360
@@ -220,12 +220,12 @@
 #define  RPF_L2UC_MSW_MACADDR_HI		0xFFFF
 #define  RPF_L2UC_MSW_ACTION			0x70000
 #define  RPF_L2UC_MSW_TAG			0x03c00000
-#define  RPF_L2UC_MSW_EN			(1 << 31)
+#define  RPF_L2UC_MSW_EN			(1U << 31)
 #define AQ_HW_MAC_NUM				34
 
 /* RPF_MCAST_FILTER_REG[8] 0x5250-0x5270 */
 #define RPF_MCAST_FILTER_REG(i)			(0x5250 + (i) * 4)
-#define  RPF_MCAST_FILTER_EN			(1 << 31)
+#define  RPF_MCAST_FILTER_EN			(1U << 31)
 #define RPF_MCAST_FILTER_MASK_REG		0x5270
 #define  RPF_MCAST_FILTER_MASK_ALLMULTI		(1 << 14)
 
@@ -240,14 +240,14 @@
 
 /* RPF_ETHERTYPE_FILTER_REG[AQ_RINGS_NUM] 0x5300-0x5380 */
 #define RPF_ETHERTYPE_FILTER_REG(i)		(0x5300 + (i) * 4)
-#define  RPF_ETHERTYPE_FILTER_EN		(1 << 31)
+#define  RPF_ETHERTYPE_FILTER_EN		(1U << 31)
 
 /* RPF_L3_FILTER_REG[8] 0x5380-0x53a0 */
 #define RPF_L3_FILTER_REG(i)			(0x5380 + (i) * 4)
-#define  RPF_L3_FILTER_L4_EN			(1 << 31)
+#define  RPF_L3_FILTER_L4_EN			(1U << 31)
 
 #define RX_FLR_RSS_CONTROL1_REG			0x54c0
-#define  RX_FLR_RSS_CONTROL1_EN			(1 << 31)
+#define  RX_FLR_RSS_CONTROL1_EN			(1U << 31)
 
 #define RPF_RPB_RX_TC_UPT_REG                   0x54c4
 #define  RPF_RPB_RX_TC_UPT_MASK(i)              (0x00000007 << ((i) * 4))
@@ -278,7 +278,7 @@
 #define RPB_RXB_BUFSIZE_REG(i)			(0x5710 + (i) * 0x10)
 #define  RPB_RXB_BUFSIZE			0x1FF
 #define RPB_RXB_XOFF_REG(i)			(0x5714 + (i) * 0x10)
-#define  RPB_RXB_XOFF_EN			(1 << 31)
+#define  RPB_RXB_XOFF_EN			(1U << 31)
 #define  RPB_RXB_XOFF_THRESH_HI                 0x3FFF0000
 #define  RPB_RXB_XOFF_THRESH_LO                 0x3FFF
 
@@ -301,7 +301,7 @@
 #define  RX_DMA_DESC_RESET			(1 << 25)
 #define  RX_DMA_DESC_HEADER_SPLIT		(1 << 28)
 #define  RX_DMA_DESC_VLAN_STRIP			(1 << 29)
-#define  RX_DMA_DESC_EN				(1 << 31)
+#define  RX_DMA_DESC_EN				(1U << 31)
 #define RX_DMA_DESC_HEAD_PTR_REG(i)		(0x5b0c + (i) * 0x20)
 #define  RX_DMA_DESC_HEAD_PTR			0xFFF
 #define RX_DMA_DESC_TAIL_PTR_REG(i)		(0x5b10 + (i) * 0x20)
@@ -313,10 +313,10 @@
 #define  RX_DMA_DCAD_CPUID			0xFF
 #define  RX_DMA_DCAD_PAYLOAD_EN			(1 << 29)
 #define  RX_DMA_DCAD_HEADER_EN			(1 << 30)
-#define  RX_DMA_DCAD_DESC_EN			(1 << 31)
+#define  RX_DMA_DCAD_DESC_EN			(1U << 31)
 
 #define RX_DMA_DCA_REG				0x6180
-#define  RX_DMA_DCA_EN				(1 << 31)
+#define  RX_DMA_DCA_EN				(1U << 31)
 #define  RX_DMA_DCA_MODE			0xF
 
 #define TX_SYSCONTROL_REG			0x7000
@@ -328,7 +328,7 @@
 #define TPS_DESC_VM_ARB_MODE_REG		0x7300
 #define  TPS_DESC_VM_ARB_MODE			(1 << 0)
 #define TPS_DESC_RATE_REG			0x7310
-#define  TPS_DESC_RATE_TA_RST			(1 << 31)
+#define  TPS_DESC_RATE_TA_RST			(1U << 31)
 #define  TPS_DESC_RATE_LIM			0x7FF
 #define TPS_DESC_TC_ARB_MODE_REG		0x7200
 #define  TPS_DESC_TC_ARB_MODE			0x3
@@ -393,7 +393,7 @@
 #define  TDM_DCAD_CPUID_EN			0x80000000
 
 #define TDM_DCA_REG				0x8480
-#define  TDM_DCA_EN				(1 << 31)
+#define  TDM_DCA_EN				(1U << 31)
 #define  TDM_DCA_MODE				0xF
 
 #define TX_INTR_MODERATION_CTL_REG(i)		(0x8980 + (i) * 4)
@@ -418,7 +418,7 @@
 #define  AQ2_MIF_BOOT_CRASH_INIT		(1 << 27)
 #define  AQ2_MIF_BOOT_BOOT_CODE_FAILED		(1 << 28)
 #define  AQ2_MIF_BOOT_FW_INIT_FAILED		(1 << 29)
-#define  AQ2_MIF_BOOT_FW_INIT_COMP_SUCCESS	(1 << 31)
+#define  AQ2_MIF_BOOT_FW_INIT_COMP_SUCCESS	(1U << 31)
 
 /* AQ2 action resolver table */
 #define AQ2_ART_ACTION_ACT_SHIFT		8

sys/dev/pci/if_mcx.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_mcx.c,v 1.107 2023/06/06 01:40:04 dlg Exp $ */
+/*	$OpenBSD: if_mcx.c,v 1.108 2023/08/15 08:27:30 miod Exp $ */
 
 /*
  * Copyright (c) 2017 David Gwynne <dlg@openbsd.org>
@@ -149,8 +149,8 @@ CTASSERT(MCX_MAX_QUEUES * MCX_WQ_DOORBELL_STRIDE <
 #define MCX_CMDQ_DOORBELL		0x0018
 
 #define MCX_STATE			0x01fc
-#define MCX_STATE_MASK				(1 << 31)
-#define MCX_STATE_INITIALIZING			(1 << 31)
+#define MCX_STATE_MASK				(1U << 31)
+#define MCX_STATE_INITIALIZING			(1U << 31)
 #define MCX_STATE_READY				(0 << 31)
 #define MCX_STATE_INTERFACE_MASK		(0x3 << 24)
 #define MCX_STATE_INTERFACE_FULL_DRIVER		(0x0 << 24)
@@ -1405,7 +1405,7 @@ struct mcx_cmd_create_tir_mb_in {
 #define MCX_TIR_CTX_HASH_SEL_SPORT	(1 << 2)
 #define MCX_TIR_CTX_HASH_SEL_DPORT	(1 << 3)
 #define MCX_TIR_CTX_HASH_SEL_IPV4	(0 << 31)
-#define MCX_TIR_CTX_HASH_SEL_IPV6	(1 << 31)
+#define MCX_TIR_CTX_HASH_SEL_IPV6	(1U << 31)
 #define MCX_TIR_CTX_HASH_SEL_TCP	(0 << 30)
 #define MCX_TIR_CTX_HASH_SEL_UDP	(1 << 30)
 	uint32_t		cmd_rx_hash_sel_inner;
@@ -1675,7 +1675,7 @@ CTASSERT(sizeof(struct mcx_wq_ctx) == 0xC0);
 
 struct mcx_sq_ctx {
 	uint32_t		sq_flags;
-#define MCX_SQ_CTX_RLKEY			(1 << 31)
+#define MCX_SQ_CTX_RLKEY			(1U << 31)
 #define MCX_SQ_CTX_FRE_SHIFT			(1 << 29)
 #define MCX_SQ_CTX_FLUSH_IN_ERROR		(1 << 28)
 #define MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT		24
@@ -1722,7 +1722,7 @@ struct mcx_sq_entry {
 	/* ethernet segment */
 	uint32_t		sqe_reserved1;
 	uint32_t		sqe_mss_csum;
-#define MCX_SQE_L4_CSUM				(1 << 31)
+#define MCX_SQE_L4_CSUM				(1U << 31)
 #define MCX_SQE_L3_CSUM				(1 << 30)
 	uint32_t		sqe_reserved2;
 	uint16_t		sqe_inline_header_size;
@@ -1789,7 +1789,7 @@ struct mcx_cmd_destroy_sq_out {
 
 struct mcx_rq_ctx {
 	uint32_t		rq_flags;
-#define MCX_RQ_CTX_RLKEY			(1 << 31)
+#define MCX_RQ_CTX_RLKEY			(1U << 31)
 #define MCX_RQ_CTX_VLAN_STRIP_DIS		(1 << 28)
 #define MCX_RQ_CTX_MEM_RQ_TYPE_SHIFT		24
 #define MCX_RQ_CTX_STATE_SHIFT			20

sys/dev/pci/igc_regs.h

@@ -1,4 +1,4 @@
-/*	$OpenBSD: igc_regs.h,v 1.1 2021/10/31 14:52:57 patrick Exp $	*/
+/*	$OpenBSD: igc_regs.h,v 1.2 2023/08/15 08:27:30 miod Exp $	*/
 /*-
  * Copyright 2021 Intel Corp
  * Copyright 2021 Rubicon Communications, LLC (Netgate)
@@ -333,7 +333,7 @@
 /* ETQF register bit definitions */
 #define IGC_ETQF_FILTER_ENABLE	(1 << 26)
 #define IGC_ETQF_IMM_INT	(1 << 29)
-#define IGC_ETQF_QUEUE_ENABLE	(1 << 31)
+#define IGC_ETQF_QUEUE_ENABLE	(1U << 31)
 #define IGC_ETQF_QUEUE_SHIFT	16
 #define IGC_ETQF_QUEUE_MASK	0x00070000
 #define IGC_ETQF_ETYPE_MASK	0x0000FFFF

sys/dev/pci/ixgbe_type.h

@@ -1,4 +1,4 @@
-/*	$OpenBSD: ixgbe_type.h,v 1.37 2023/05/18 08:22:37 jan Exp $	*/
+/*	$OpenBSD: ixgbe_type.h,v 1.38 2023/08/15 08:27:30 miod Exp $	*/
 
 /******************************************************************************
   SPDX-License-Identifier: BSD-3-Clause
@@ -4454,7 +4454,7 @@ struct ixgbe_bypass_eeprom {
 #define IXGBE_KRM_LINK_CTRL_1_TETH_EEE_CAP_KR		(1 << 26)
 #define IXGBE_KRM_LINK_S1_MAC_AN_COMPLETE		(1 << 28)
 #define IXGBE_KRM_LINK_CTRL_1_TETH_AN_ENABLE		(1 << 29)
-#define IXGBE_KRM_LINK_CTRL_1_TETH_AN_RESTART		(1 << 31)
+#define IXGBE_KRM_LINK_CTRL_1_TETH_AN_RESTART		(1U << 31)
 
 #define IXGBE_KRM_AN_CNTL_1_SYM_PAUSE			(1 << 28)
 #define IXGBE_KRM_AN_CNTL_1_ASM_PAUSE			(1 << 29)
@@ -4484,7 +4484,7 @@ struct ixgbe_bypass_eeprom {
 #define IXGBE_KRM_TX_COEFF_CTRL_1_CMINUS1_OVRRD_EN	(1 << 1)
 #define IXGBE_KRM_TX_COEFF_CTRL_1_CPLUS1_OVRRD_EN	(1 << 2)
 #define IXGBE_KRM_TX_COEFF_CTRL_1_CZERO_EN		(1 << 3)
-#define IXGBE_KRM_TX_COEFF_CTRL_1_OVRRD_EN		(1 << 31)
+#define IXGBE_KRM_TX_COEFF_CTRL_1_OVRRD_EN		(1U << 31)
 
 #define IXGBE_SB_IOSF_INDIRECT_CTRL	0x00011144
 #define IXGBE_SB_IOSF_INDIRECT_DATA	0x00011148