Merge tag 'amd-drm-next-5.19-2022-04-29' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-5.19-2022-04-29:

amdgpu
- RAS updates
- SI dpm deadlock fix
- Misc code cleanups
- HDCP fixes
- PSR fixes
- DSC fixes
- SDMA doorbell cleanups
- S0ix fix
- DC FP fix
- Zen dom0 regression fix for APUs
- IP discovery updates
- Initial SoC21 support
- Support for new vbios tables
- Runtime PM fixes
- Add PSP TA debugfs interface

amdkfd:
- Misc code cleanups
- Ignore bogus MEC signals more efficiently
- SVM fixes
- Use bitmap helpers

radeon:
- Misc code cleanups
- Spelling/grammer fixes

From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220429144853.5742-1-alexander.deucher@amd.com

11 files changed, 138 insertions, 68 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index ee8b288dd8cc..1c7016958d6d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -944,8 +944,6 @@ err_drm_file:
 
 bool kfd_dev_is_large_bar(struct kfd_dev *dev)
 {
-	struct kfd_local_mem_info mem_info;
-
 	if (debug_largebar) {
 		pr_debug("Simulate large-bar allocation on non large-bar machine\n");
 		return true;
@@ -954,9 +952,8 @@ bool kfd_dev_is_large_bar(struct kfd_dev *dev)
 	if (dev->use_iommu_v2)
 		return false;
 
-	amdgpu_amdkfd_get_local_mem_info(dev->adev, &mem_info);
-	if (mem_info.local_mem_size_private == 0 &&
-			mem_info.local_mem_size_public > 0)
+	if (dev->local_mem_info.local_mem_size_private == 0 &&
+			dev->local_mem_info.local_mem_size_public > 0)
 		return true;
 	return false;
 }
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
index afc8a7fcdad8..90c70adc946c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
@@ -2152,7 +2152,7 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image,
 	 * report the total FB size (public+private) as a single
 	 * private heap.
 	 */
-	amdgpu_amdkfd_get_local_mem_info(kdev->adev, &local_mem_info);
+	local_mem_info = kdev->local_mem_info;
 	sub_type_hdr = (typeof(sub_type_hdr))((char *)sub_type_hdr +
 			sub_type_hdr->length);
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 62aa6c9d5123..ed33e95c03e6 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -575,6 +575,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	if (kfd_resume(kfd))
 		goto kfd_resume_error;
 
+	amdgpu_amdkfd_get_local_mem_info(kfd->adev, &kfd->local_mem_info);
+
 	if (kfd_topology_add_device(kfd)) {
 		dev_err(kfd_device, "Error adding device to topology\n");
 		goto kfd_topology_add_device_error;
@@ -873,8 +875,6 @@ out:
 static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
 				unsigned int chunk_size)
 {
-	unsigned int num_of_longs;
-
 	if (WARN_ON(buf_size < chunk_size))
 		return -EINVAL;
 	if (WARN_ON(buf_size == 0))
@@ -885,11 +885,8 @@ static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
 	kfd->gtt_sa_chunk_size = chunk_size;
 	kfd->gtt_sa_num_of_chunks = buf_size / chunk_size;
 
-	num_of_longs = (kfd->gtt_sa_num_of_chunks + BITS_PER_LONG - 1) /
-		BITS_PER_LONG;
-
-	kfd->gtt_sa_bitmap = kcalloc(num_of_longs, sizeof(long), GFP_KERNEL);
-
+	kfd->gtt_sa_bitmap = bitmap_zalloc(kfd->gtt_sa_num_of_chunks,
+					   GFP_KERNEL);
 	if (!kfd->gtt_sa_bitmap)
 		return -ENOMEM;
 
@@ -899,13 +896,12 @@ static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
 	mutex_init(&kfd->gtt_sa_lock);
 
 	return 0;
-
 }
 
 static void kfd_gtt_sa_fini(struct kfd_dev *kfd)
 {
 	mutex_destroy(&kfd->gtt_sa_lock);
-	kfree(kfd->gtt_sa_bitmap);
+	bitmap_free(kfd->gtt_sa_bitmap);
 }
 
 static inline uint64_t kfd_gtt_sa_calc_gpu_addr(uint64_t start_addr,
@@ -973,7 +969,7 @@ kfd_gtt_restart_search:
 	/* If we need only one chunk, mark it as allocated and get out */
 	if (size <= kfd->gtt_sa_chunk_size) {
 		pr_debug("Single bit\n");
-		set_bit(found, kfd->gtt_sa_bitmap);
+		__set_bit(found, kfd->gtt_sa_bitmap);
 		goto kfd_gtt_out;
 	}
 
@@ -1011,10 +1007,8 @@ kfd_gtt_restart_search:
 		(*mem_obj)->range_start, (*mem_obj)->range_end);
 
 	/* Mark the chunks as allocated */
-	for (found = (*mem_obj)->range_start;
-		found <= (*mem_obj)->range_end;
-		found++)
-		set_bit(found, kfd->gtt_sa_bitmap);
+	bitmap_set(kfd->gtt_sa_bitmap, (*mem_obj)->range_start,
+		   (*mem_obj)->range_end - (*mem_obj)->range_start + 1);
 
 kfd_gtt_out:
 	mutex_unlock(&kfd->gtt_sa_lock);
@@ -1029,8 +1023,6 @@ kfd_gtt_no_free_chunk:
 
 int kfd_gtt_sa_free(struct kfd_dev *kfd, struct kfd_mem_obj *mem_obj)
 {
-	unsigned int bit;
-
 	/* Act like kfree when trying to free a NULL object */
 	if (!mem_obj)
 		return 0;
@@ -1041,10 +1033,8 @@ int kfd_gtt_sa_free(struct kfd_dev *kfd, struct kfd_mem_obj *mem_obj)
 	mutex_lock(&kfd->gtt_sa_lock);
 
 	/* Mark the chunks as free */
-	for (bit = mem_obj->range_start;
-		bit <= mem_obj->range_end;
-		bit++)
-		clear_bit(bit, kfd->gtt_sa_bitmap);
+	bitmap_clear(kfd->gtt_sa_bitmap, mem_obj->range_start,
+		     mem_obj->range_end - mem_obj->range_start + 1);
 
 	mutex_unlock(&kfd->gtt_sa_lock);
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
index 6e5e8d637f48..4df9c36146ba 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
@@ -238,12 +238,24 @@ static int create_other_event(struct kfd_process *p, struct kfd_event *ev, const
 	return 0;
 }
 
-void kfd_event_init_process(struct kfd_process *p)
+int kfd_event_init_process(struct kfd_process *p)
 {
+	int id;
+
 	mutex_init(&p->event_mutex);
 	idr_init(&p->event_idr);
 	p->signal_page = NULL;
-	p->signal_event_count = 0;
+	p->signal_event_count = 1;
+	/* Allocate event ID 0. It is used for a fast path to ignore bogus events
+	 * that are sent by the CP without a context ID
+	 */
+	id = idr_alloc(&p->event_idr, NULL, 0, 1, GFP_KERNEL);
+	if (id < 0) {
+		idr_destroy(&p->event_idr);
+		mutex_destroy(&p->event_mutex);
+		return id;
+	}
+	return 0;
 }
 
 static void destroy_event(struct kfd_process *p, struct kfd_event *ev)
@@ -271,8 +283,10 @@ static void destroy_events(struct kfd_process *p)
 	uint32_t id;
 
 	idr_for_each_entry(&p->event_idr, ev, id)
-		destroy_event(p, ev);
+		if (ev)
+			destroy_event(p, ev);
 	idr_destroy(&p->event_idr);
+	mutex_destroy(&p->event_mutex);
 }
 
 /*
@@ -749,7 +763,7 @@ void kfd_signal_event_interrupt(u32 pasid, uint32_t partial_id,
 			 * iterate over the signal slots and lookup
 			 * only signaled events from the IDR.
 			 */
-			for (id = 0; id < KFD_SIGNAL_EVENT_LIMIT; id++)
+			for (id = 1; id < KFD_SIGNAL_EVENT_LIMIT; id++)
 				if (READ_ONCE(slots[id]) != UNSIGNALED_EVENT_SLOT) {
 					ev = lookup_event_by_id(p, id);
 					set_event_from_interrupt(p, ev);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
index 03c29bdd89a1..f27fe022ef6f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
@@ -141,6 +141,25 @@ static void event_interrupt_poison_consumption(struct kfd_dev *dev,
 	}
 }
 
+static bool context_id_expected(struct kfd_dev *dev)
+{
+	switch (KFD_GC_VERSION(dev)) {
+	case IP_VERSION(9, 0, 1):
+		return dev->mec_fw_version >= 0x817a;
+	case IP_VERSION(9, 1, 0):
+	case IP_VERSION(9, 2, 1):
+	case IP_VERSION(9, 2, 2):
+	case IP_VERSION(9, 3, 0):
+	case IP_VERSION(9, 4, 0):
+		return dev->mec_fw_version >= 0x17a;
+	default:
+		/* Other GFXv9 and later GPUs always sent valid context IDs
+		 * on legitimate events
+		 */
+		return KFD_GC_VERSION(dev) >= IP_VERSION(9, 4, 1);
+	}
+}
+
 static bool event_interrupt_isr_v9(struct kfd_dev *dev,
 					const uint32_t *ih_ring_entry,
 					uint32_t *patched_ihre,
@@ -206,6 +225,20 @@ static bool event_interrupt_isr_v9(struct kfd_dev *dev,
 	if (WARN_ONCE(pasid == 0, "Bug: No PASID in KFD interrupt"))
 		return false;
 
+	/* Workaround CP firmware sending bogus signals with 0 context_id.
+	 * Those can be safely ignored on hardware and firmware versions that
+	 * include a valid context_id on legitimate signals. This avoids the
+	 * slow path in kfd_signal_event_interrupt that scans all event slots
+	 * for signaled events.
+	 */
+	if (source_id == SOC15_INTSRC_CP_END_OF_PIPE) {
+		uint32_t context_id =
+			SOC15_CONTEXT_ID0_FROM_IH_ENTRY(ih_ring_entry);
+
+		if (context_id == 0 && context_id_expected(dev))
+			return false;
+	}
+
 	/* Interrupt types we care about: various signals and faults.
 	 * They will be forwarded to a work queue (see below).
 	 */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
index 7e3a7fcb9fe6..997650d597ec 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
@@ -410,7 +410,6 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
 	struct migrate_vma migrate;
 	unsigned long cpages = 0;
 	dma_addr_t *scratch;
-	size_t size;
 	void *buf;
 	int r = -ENOMEM;
 
@@ -421,9 +420,9 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
 	migrate.flags = MIGRATE_VMA_SELECT_SYSTEM;
 	migrate.pgmap_owner = SVM_ADEV_PGMAP_OWNER(adev);
 
-	size = 2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t);
-	size *= npages;
-	buf = kvmalloc(size, GFP_KERNEL | __GFP_ZERO);
+	buf = kvcalloc(npages,
+		       2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t),
+		       GFP_KERNEL);
 	if (!buf)
 		goto out;
 
@@ -665,7 +664,6 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
 	struct dma_fence *mfence = NULL;
 	struct migrate_vma migrate;
 	dma_addr_t *scratch;
-	size_t size;
 	void *buf;
 	int r = -ENOMEM;
 
@@ -676,9 +674,10 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
 	migrate.flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE;
 	migrate.pgmap_owner = SVM_ADEV_PGMAP_OWNER(adev);
 
-	size = 2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t);
-	size *= npages;
-	buf = kvmalloc(size, GFP_KERNEL | __GFP_ZERO);
+	buf = kvcalloc(npages,
+		       2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t),
+		       GFP_KERNEL);
+
 	if (!buf)
 		goto out;
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index bfb3b5c288cb..10bb3bb46246 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -272,6 +272,7 @@ struct kfd_dev {
 
 	struct kgd2kfd_shared_resources shared_resources;
 	struct kfd_vmid_info vm_info;
+	struct kfd_local_mem_info local_mem_info;
 
 	const struct kfd2kgd_calls *kfd2kgd;
 	struct mutex doorbell_mutex;
@@ -1295,7 +1296,7 @@ extern const struct kfd_event_interrupt_class event_interrupt_class_v9;
 
 extern const struct kfd_device_global_init_class device_global_init_class_cik;
 
-void kfd_event_init_process(struct kfd_process *p);
+int kfd_event_init_process(struct kfd_process *p);
 void kfd_event_free_process(struct kfd_process *p);
 int kfd_event_mmap(struct kfd_process *process, struct vm_area_struct *vma);
 int kfd_wait_on_events(struct kfd_process *p,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 9e82d7aa67fa..cb8f4a459add 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -1370,12 +1370,16 @@ static struct kfd_process *create_process(const struct task_struct *thread)
 	INIT_DELAYED_WORK(&process->eviction_work, evict_process_worker);
 	INIT_DELAYED_WORK(&process->restore_work, restore_process_worker);
 	process->last_restore_timestamp = get_jiffies_64();
-	kfd_event_init_process(process);
+	err = kfd_event_init_process(process);
+	if (err)
+		goto err_event_init;
 	process->is_32bit_user_mode = in_compat_syscall();
 
 	process->pasid = kfd_pasid_alloc();
-	if (process->pasid == 0)
+	if (process->pasid == 0) {
+		err = -ENOSPC;
 		goto err_alloc_pasid;
+	}
 
 	err = pqm_init(&process->pqm, process);
 	if (err != 0)
@@ -1424,6 +1428,8 @@ err_init_apertures:
 err_process_pqm_init:
 	kfd_pasid_free(process->pasid);
 err_alloc_pasid:
+	kfd_event_free_process(process);
+err_event_init:
 	mutex_destroy(&process->mutex);
 	kfree(process);
 err_alloc_process:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 11b395b90a3d..29e9ebf6d8d5 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -149,8 +149,7 @@ svm_range_dma_map_dev(struct amdgpu_device *adev, struct svm_range *prange,
 	int i, r;
 
 	if (!addr) {
-		addr = kvmalloc_array(prange->npages, sizeof(*addr),
-				      GFP_KERNEL | __GFP_ZERO);
+		addr = kvcalloc(prange->npages, sizeof(*addr), GFP_KERNEL);
 		if (!addr)
 			return -ENOMEM;
 		prange->dma_addr[gpuidx] = addr;
@@ -686,7 +685,8 @@ svm_range_check_attr(struct kfd_process *p,
 
 static void
 svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
-		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs)
+		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs,
+		      bool *update_mapping)
 {
 	uint32_t i;
 	int gpuidx;
@@ -702,6 +702,7 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
 		case KFD_IOCTL_SVM_ATTR_ACCESS:
 		case KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE:
 		case KFD_IOCTL_SVM_ATTR_NO_ACCESS:
+			*update_mapping = true;
 			gpuidx = kfd_process_gpuidx_from_gpuid(p,
 							       attrs[i].value);
 			if (attrs[i].type == KFD_IOCTL_SVM_ATTR_NO_ACCESS) {
@@ -716,9 +717,11 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
 			}
 			break;
 		case KFD_IOCTL_SVM_ATTR_SET_FLAGS:
+			*update_mapping = true;
 			prange->flags |= attrs[i].value;
 			break;
 		case KFD_IOCTL_SVM_ATTR_CLR_FLAGS:
+			*update_mapping = true;
 			prange->flags &= ~attrs[i].value;
 			break;
 		case KFD_IOCTL_SVM_ATTR_GRANULARITY:
@@ -951,6 +954,7 @@ svm_range_split_adjust(struct svm_range *new, struct svm_range *old,
 	new->prefetch_loc = old->prefetch_loc;
 	new->actual_loc = old->actual_loc;
 	new->granularity = old->granularity;
+	new->mapped_to_gpu = old->mapped_to_gpu;
 	bitmap_copy(new->bitmap_access, old->bitmap_access, MAX_GPU_INSTANCE);
 	bitmap_copy(new->bitmap_aip, old->bitmap_aip, MAX_GPU_INSTANCE);
 
@@ -1204,6 +1208,17 @@ svm_range_unmap_from_gpus(struct svm_range *prange, unsigned long start,
 	uint32_t gpuidx;
 	int r = 0;
 
+	if (!prange->mapped_to_gpu) {
+		pr_debug("prange 0x%p [0x%lx 0x%lx] not mapped to GPU\n",
+			 prange, prange->start, prange->last);
+		return 0;
+	}
+
+	if (prange->start == start && prange->last == last) {
+		pr_debug("unmap svms 0x%p prange 0x%p\n", prange->svms, prange);
+		prange->mapped_to_gpu = false;
+	}
+
 	bitmap_or(bitmap, prange->bitmap_access, prange->bitmap_aip,
 		  MAX_GPU_INSTANCE);
 	p = container_of(prange->svms, struct kfd_process, svms);
@@ -1239,7 +1254,7 @@ static int
 svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
 		     unsigned long offset, unsigned long npages, bool readonly,
 		     dma_addr_t *dma_addr, struct amdgpu_device *bo_adev,
-		     struct dma_fence **fence)
+		     struct dma_fence **fence, bool flush_tlb)
 {
 	struct amdgpu_device *adev = pdd->dev->adev;
 	struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
@@ -1277,7 +1292,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange,
 			 (last_domain == SVM_RANGE_VRAM_DOMAIN) ? 1 : 0,
 			 pte_flags);
 
-		r = amdgpu_vm_update_range(adev, vm, false, false, false, NULL,
+		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb, NULL,
 					   last_start, prange->start + i,
 					   pte_flags,
 					   last_start - prange->start,
@@ -1311,7 +1326,7 @@ out:
 static int
 svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
 		      unsigned long npages, bool readonly,
-		      unsigned long *bitmap, bool wait)
+		      unsigned long *bitmap, bool wait, bool flush_tlb)
 {
 	struct kfd_process_device *pdd;
 	struct amdgpu_device *bo_adev;
@@ -1346,7 +1361,8 @@ svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset,
 
 		r = svm_range_map_to_gpu(pdd, prange, offset, npages, readonly,
 					 prange->dma_addr[gpuidx],
-					 bo_adev, wait ? &fence : NULL);
+					 bo_adev, wait ? &fence : NULL,
+					 flush_tlb);
 		if (r)
 			break;
 
@@ -1467,8 +1483,8 @@ static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx)
  * 5. Release page table (and SVM BO) reservation
  */
 static int svm_range_validate_and_map(struct mm_struct *mm,
-				      struct svm_range *prange,
-				      int32_t gpuidx, bool intr, bool wait)
+				      struct svm_range *prange, int32_t gpuidx,
+				      bool intr, bool wait, bool flush_tlb)
 {
 	struct svm_validate_context ctx;
 	unsigned long start, end, addr;
@@ -1507,8 +1523,12 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 			  prange->bitmap_aip, MAX_GPU_INSTANCE);
 	}
 
-	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE))
-		return 0;
+	if (bitmap_empty(ctx.bitmap, MAX_GPU_INSTANCE)) {
+		if (!prange->mapped_to_gpu)
+			return 0;
+
+		bitmap_copy(ctx.bitmap, prange->bitmap_access, MAX_GPU_INSTANCE);
+	}
 
 	if (prange->actual_loc && !prange->ttm_res) {
 		/* This should never happen. actual_loc gets set by
@@ -1580,7 +1600,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm,
 		}
 
 		r = svm_range_map_to_gpus(prange, offset, npages, readonly,
-					  ctx.bitmap, wait);
+					  ctx.bitmap, wait, flush_tlb);
 
 unlock_out:
 		svm_range_unlock(prange);
@@ -1588,8 +1608,10 @@ unlock_out:
 		addr = next;
 	}
 
-	if (addr == end)
+	if (addr == end) {
 		prange->validated_once = true;
+		prange->mapped_to_gpu = true;
+	}
 
 unreserve_out:
 	svm_range_unreserve_bos(&ctx);
@@ -1674,7 +1696,7 @@ static void svm_range_restore_work(struct work_struct *work)
 		mutex_lock(&prange->migrate_mutex);
 
 		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       false, true);
+					       false, true, false);
 		if (r)
 			pr_debug("failed %d to map 0x%lx to gpus\n", r,
 				 prange->start);
@@ -1820,6 +1842,7 @@ static struct svm_range *svm_range_clone(struct svm_range *old)
 	new->prefetch_loc = old->prefetch_loc;
 	new->actual_loc = old->actual_loc;
 	new->granularity = old->granularity;
+	new->mapped_to_gpu = old->mapped_to_gpu;
 	bitmap_copy(new->bitmap_access, old->bitmap_access, MAX_GPU_INSTANCE);
 	bitmap_copy(new->bitmap_aip, old->bitmap_aip, MAX_GPU_INSTANCE);
 
@@ -2811,7 +2834,7 @@ retry_write_locked:
 		}
 	}
 
-	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false);
+	r = svm_range_validate_and_map(mm, prange, gpuidx, false, false, false);
 	if (r)
 		pr_debug("failed %d to map svms 0x%p [0x%lx 0x%lx] to gpus\n",
 			 r, svms, prange->start, prange->last);
@@ -3224,6 +3247,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
 	struct svm_range_list *svms;
 	struct svm_range *prange;
 	struct svm_range *next;
+	bool update_mapping = false;
+	bool flush_tlb;
 	int r = 0;
 
 	pr_debug("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n",
@@ -3262,7 +3287,7 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
 		svm_range_add_notifier_locked(mm, prange);
 	}
 	list_for_each_entry(prange, &update_list, update_list) {
-		svm_range_apply_attrs(p, prange, nattr, attrs);
+		svm_range_apply_attrs(p, prange, nattr, attrs, &update_mapping);
 		/* TODO: unmap ranges from GPU that lost access */
 	}
 	list_for_each_entry_safe(prange, next, &remove_list, update_list) {
@@ -3295,8 +3320,15 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,
 			continue;
 		}
 
+		if (!migrated && !update_mapping) {
+			mutex_unlock(&prange->migrate_mutex);
+			continue;
+		}
+
+		flush_tlb = !migrated && update_mapping && prange->mapped_to_gpu;
+
 		r = svm_range_validate_and_map(mm, prange, MAX_GPU_INSTANCE,
-					       true, true);
+					       true, true, flush_tlb);
 		if (r)
 			pr_debug("failed %d to map svm range\n", r);
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
index 66c77f00ac3e..2d54147b4dda 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
@@ -133,6 +133,7 @@ struct svm_range {
 	DECLARE_BITMAP(bitmap_access, MAX_GPU_INSTANCE);
 	DECLARE_BITMAP(bitmap_aip, MAX_GPU_INSTANCE);
 	bool				validated_once;
+	bool				mapped_to_gpu;
 };
 
 static inline void svm_range_lock(struct svm_range *prange)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index 8b7710b4d3ed..05089f1de4e9 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -1112,15 +1112,12 @@ static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu)
 	uint32_t buf[7];
 	uint64_t local_mem_size;
 	int i;
-	struct kfd_local_mem_info local_mem_info;
 
 	if (!gpu)
 		return 0;
 
-	amdgpu_amdkfd_get_local_mem_info(gpu->adev, &local_mem_info);
-
-	local_mem_size = local_mem_info.local_mem_size_private +
-			local_mem_info.local_mem_size_public;
+	local_mem_size = gpu->local_mem_info.local_mem_size_private +
+			gpu->local_mem_info.local_mem_size_public;
 
 	buf[0] = gpu->pdev->devfn;
 	buf[1] = gpu->pdev->subsystem_vendor |
@@ -1534,13 +1531,13 @@ static void kfd_topology_update_io_links(int proximity_domain)
 				list_del(&iolink->list);
 				dev->io_link_count--;
 				dev->node_props.io_links_count--;
-			} else if (iolink->node_from > proximity_domain) {
-				iolink->node_from--;
-			} else if (iolink->node_to > proximity_domain) {
-				iolink->node_to--;
+			} else {
+				if (iolink->node_from > proximity_domain)
+					iolink->node_from--;
+				if (iolink->node_to > proximity_domain)
+					iolink->node_to--;
 			}
 		}
-
 	}
 }