1 files changed, 102 insertions, 49 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 21ec8a18cad2..d4c8a6948a9f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -72,6 +72,8 @@ static int kfd_process_init_cwsr_apu(struct kfd_process *p, struct file *filep);
 static void evict_process_worker(struct work_struct *work);
 static void restore_process_worker(struct work_struct *work);
 
+static void kfd_process_device_destroy_cwsr_dgpu(struct kfd_process_device *pdd);
+
 struct kfd_procfs_tree {
 	struct kobject *kobj;
 };
@@ -286,7 +288,7 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer)
 	/* Collect wave count from device if it supports */
 	wave_cnt = 0;
 	max_waves_per_cu = 0;
-	dev->kfd2kgd->get_cu_occupancy(dev->kgd, proc->pasid, &wave_cnt,
+	dev->kfd2kgd->get_cu_occupancy(dev->adev, proc->pasid, &wave_cnt,
 			&max_waves_per_cu);
 
 	/* Translate wave count to number of compute units */
@@ -685,12 +687,17 @@ void kfd_process_destroy_wq(void)
 }
 
 static void kfd_process_free_gpuvm(struct kgd_mem *mem,
-			struct kfd_process_device *pdd)
+			struct kfd_process_device *pdd, void *kptr)
 {
 	struct kfd_dev *dev = pdd->dev;
 
-	amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(dev->kgd, mem, pdd->drm_priv);
-	amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->kgd, mem, pdd->drm_priv,
+	if (kptr) {
+		amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(dev->adev, mem);
+		kptr = NULL;
+	}
+
+	amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(dev->adev, mem, pdd->drm_priv);
+	amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->adev, mem, pdd->drm_priv,
 					       NULL);
 }
 
@@ -702,63 +709,46 @@ static void kfd_process_free_gpuvm(struct kgd_mem *mem,
  */
 static int kfd_process_alloc_gpuvm(struct kfd_process_device *pdd,
 				   uint64_t gpu_va, uint32_t size,
-				   uint32_t flags, void **kptr)
+				   uint32_t flags, struct kgd_mem **mem, void **kptr)
 {
 	struct kfd_dev *kdev = pdd->dev;
-	struct kgd_mem *mem = NULL;
-	int handle;
 	int err;
 
-	err = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(kdev->kgd, gpu_va, size,
-						 pdd->drm_priv, &mem, NULL, flags);
+	err = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(kdev->adev, gpu_va, size,
+						 pdd->drm_priv, mem, NULL, flags);
 	if (err)
 		goto err_alloc_mem;
 
-	err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(kdev->kgd, mem,
+	err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(kdev->adev, *mem,
 			pdd->drm_priv, NULL);
 	if (err)
 		goto err_map_mem;
 
-	err = amdgpu_amdkfd_gpuvm_sync_memory(kdev->kgd, mem, true);
+	err = amdgpu_amdkfd_gpuvm_sync_memory(kdev->adev, *mem, true);
 	if (err) {
 		pr_debug("Sync memory failed, wait interrupted by user signal\n");
 		goto sync_memory_failed;
 	}
 
-	/* Create an obj handle so kfd_process_device_remove_obj_handle
-	 * will take care of the bo removal when the process finishes.
-	 * We do not need to take p->mutex, because the process is just
-	 * created and the ioctls have not had the chance to run.
-	 */
-	handle = kfd_process_device_create_obj_handle(pdd, mem);
-
-	if (handle < 0) {
-		err = handle;
-		goto free_gpuvm;
-	}
-
 	if (kptr) {
-		err = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(kdev->kgd,
-				(struct kgd_mem *)mem, kptr, NULL);
+		err = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(kdev->adev,
+				(struct kgd_mem *)*mem, kptr, NULL);
 		if (err) {
 			pr_debug("Map GTT BO to kernel failed\n");
-			goto free_obj_handle;
+			goto sync_memory_failed;
 		}
 	}
 
 	return err;
 
-free_obj_handle:
-	kfd_process_device_remove_obj_handle(pdd, handle);
-free_gpuvm:
 sync_memory_failed:
-	kfd_process_free_gpuvm(mem, pdd);
-	return err;
+	amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(kdev->adev, *mem, pdd->drm_priv);
 
 err_map_mem:
-	amdgpu_amdkfd_gpuvm_free_memory_of_gpu(kdev->kgd, mem, pdd->drm_priv,
+	amdgpu_amdkfd_gpuvm_free_memory_of_gpu(kdev->adev, *mem, pdd->drm_priv,
 					       NULL);
 err_alloc_mem:
+	*mem = NULL;
 	*kptr = NULL;
 	return err;
 }
@@ -776,6 +766,7 @@ static int kfd_process_device_reserve_ib_mem(struct kfd_process_device *pdd)
 			KFD_IOC_ALLOC_MEM_FLAGS_NO_SUBSTITUTE |
 			KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE |
 			KFD_IOC_ALLOC_MEM_FLAGS_EXECUTABLE;
+	struct kgd_mem *mem;
 	void *kaddr;
 	int ret;
 
@@ -784,15 +775,26 @@ static int kfd_process_device_reserve_ib_mem(struct kfd_process_device *pdd)
 
 	/* ib_base is only set for dGPU */
 	ret = kfd_process_alloc_gpuvm(pdd, qpd->ib_base, PAGE_SIZE, flags,
-				      &kaddr);
+				      &mem, &kaddr);
 	if (ret)
 		return ret;
 
+	qpd->ib_mem = mem;
 	qpd->ib_kaddr = kaddr;
 
 	return 0;
 }
 
+static void kfd_process_device_destroy_ib_mem(struct kfd_process_device *pdd)
+{
+	struct qcm_process_device *qpd = &pdd->qpd;
+
+	if (!qpd->ib_kaddr || !qpd->ib_base)
+		return;
+
+	kfd_process_free_gpuvm(qpd->ib_mem, pdd, qpd->ib_kaddr);
+}
+
 struct kfd_process *kfd_create_process(struct file *filep)
 {
 	struct kfd_process *process;
@@ -938,15 +940,46 @@ static void kfd_process_device_free_bos(struct kfd_process_device *pdd)
 			if (!peer_pdd->drm_priv)
 				continue;
 			amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
-				peer_pdd->dev->kgd, mem, peer_pdd->drm_priv);
+				peer_pdd->dev->adev, mem, peer_pdd->drm_priv);
 		}
 
-		amdgpu_amdkfd_gpuvm_free_memory_of_gpu(pdd->dev->kgd, mem,
+		amdgpu_amdkfd_gpuvm_free_memory_of_gpu(pdd->dev->adev, mem,
 						       pdd->drm_priv, NULL);
 		kfd_process_device_remove_obj_handle(pdd, id);
 	}
 }
 
+/*
+ * Just kunmap and unpin signal BO here. It will be freed in
+ * kfd_process_free_outstanding_kfd_bos()
+ */
+static void kfd_process_kunmap_signal_bo(struct kfd_process *p)
+{
+	struct kfd_process_device *pdd;
+	struct kfd_dev *kdev;
+	void *mem;
+
+	kdev = kfd_device_by_id(GET_GPU_ID(p->signal_handle));
+	if (!kdev)
+		return;
+
+	mutex_lock(&p->mutex);
+
+	pdd = kfd_get_process_device_data(kdev, p);
+	if (!pdd)
+		goto out;
+
+	mem = kfd_process_device_translate_handle(
+		pdd, GET_IDR_HANDLE(p->signal_handle));
+	if (!mem)
+		goto out;
+
+	amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(kdev->adev, mem);
+
+out:
+	mutex_unlock(&p->mutex);
+}
+
 static void kfd_process_free_outstanding_kfd_bos(struct kfd_process *p)
 {
 	int i;
@@ -965,9 +998,12 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)
 		pr_debug("Releasing pdd (topology id %d) for process (pasid 0x%x)\n",
 				pdd->dev->id, p->pasid);
 
+		kfd_process_device_destroy_cwsr_dgpu(pdd);
+		kfd_process_device_destroy_ib_mem(pdd);
+
 		if (pdd->drm_file) {
 			amdgpu_amdkfd_gpuvm_release_process_vm(
-					pdd->dev->kgd, pdd->drm_priv);
+					pdd->dev->adev, pdd->drm_priv);
 			fput(pdd->drm_file);
 		}
 
@@ -1049,9 +1085,11 @@ static void kfd_process_wq_release(struct work_struct *work)
 {
 	struct kfd_process *p = container_of(work, struct kfd_process,
 					     release_work);
+
 	kfd_process_remove_sysfs(p);
 	kfd_iommu_unbind_process(p);
 
+	kfd_process_kunmap_signal_bo(p);
 	kfd_process_free_outstanding_kfd_bos(p);
 	svm_range_list_fini(p);
 
@@ -1198,6 +1236,7 @@ static int kfd_process_device_init_cwsr_dgpu(struct kfd_process_device *pdd)
 	uint32_t flags = KFD_IOC_ALLOC_MEM_FLAGS_GTT
 			| KFD_IOC_ALLOC_MEM_FLAGS_NO_SUBSTITUTE
 			| KFD_IOC_ALLOC_MEM_FLAGS_EXECUTABLE;
+	struct kgd_mem *mem;
 	void *kaddr;
 	int ret;
 
@@ -1206,10 +1245,11 @@ static int kfd_process_device_init_cwsr_dgpu(struct kfd_process_device *pdd)
 
 	/* cwsr_base is only set for dGPU */
 	ret = kfd_process_alloc_gpuvm(pdd, qpd->cwsr_base,
-				      KFD_CWSR_TBA_TMA_SIZE, flags, &kaddr);
+				      KFD_CWSR_TBA_TMA_SIZE, flags, &mem, &kaddr);
 	if (ret)
 		return ret;
 
+	qpd->cwsr_mem = mem;
 	qpd->cwsr_kaddr = kaddr;
 	qpd->tba_addr = qpd->cwsr_base;
 
@@ -1222,6 +1262,17 @@ static int kfd_process_device_init_cwsr_dgpu(struct kfd_process_device *pdd)
 	return 0;
 }
 
+static void kfd_process_device_destroy_cwsr_dgpu(struct kfd_process_device *pdd)
+{
+	struct kfd_dev *dev = pdd->dev;
+	struct qcm_process_device *qpd = &pdd->qpd;
+
+	if (!dev->cwsr_enabled || !qpd->cwsr_kaddr || !qpd->cwsr_base)
+		return;
+
+	kfd_process_free_gpuvm(qpd->cwsr_mem, pdd, qpd->cwsr_kaddr);
+}
+
 void kfd_process_set_trap_handler(struct qcm_process_device *qpd,
 				  uint64_t tba_addr,
 				  uint64_t tma_addr)
@@ -1266,14 +1317,13 @@ bool kfd_process_xnack_mode(struct kfd_process *p, bool supported)
 		 * support the SVM APIs and don't need to be considered
 		 * for the XNACK mode selection.
 		 */
-		if (dev->device_info->asic_family < CHIP_VEGA10)
+		if (!KFD_IS_SOC15(dev))
 			continue;
 		/* Aldebaran can always support XNACK because it can support
 		 * per-process XNACK mode selection. But let the dev->noretry
 		 * setting still influence the default XNACK mode.
 		 */
-		if (supported &&
-		    dev->device_info->asic_family == CHIP_ALDEBARAN)
+		if (supported && KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2))
 			continue;
 
 		/* GFXv10 and later GPUs do not support shader preemption
@@ -1281,7 +1331,7 @@ bool kfd_process_xnack_mode(struct kfd_process *p, bool supported)
 		 * management and memory-manager-related preemptions or
 		 * even deadlocks.
 		 */
-		if (dev->device_info->asic_family >= CHIP_NAVI10)
+		if (KFD_GC_VERSION(dev) >= IP_VERSION(10, 1, 1))
 			return false;
 
 		if (dev->noretry)
@@ -1380,7 +1430,7 @@ static int init_doorbell_bitmap(struct qcm_process_device *qpd,
 	int range_start = dev->shared_resources.non_cp_doorbells_start;
 	int range_end = dev->shared_resources.non_cp_doorbells_end;
 
-	if (!KFD_IS_SOC15(dev->device_info->asic_family))
+	if (!KFD_IS_SOC15(dev))
 		return 0;
 
 	qpd->doorbell_bitmap =
@@ -1496,7 +1546,7 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,
 	dev = pdd->dev;
 
 	ret = amdgpu_amdkfd_gpuvm_acquire_process_vm(
-		dev->kgd, drm_file, p->pasid,
+		dev->adev, drm_file, p->pasid,
 		&p->kgd_process_info, &p->ef);
 	if (ret) {
 		pr_err("Failed to create process VM object\n");
@@ -1664,7 +1714,11 @@ int kfd_process_evict_queues(struct kfd_process *p)
 
 		r = pdd->dev->dqm->ops.evict_process_queues(pdd->dev->dqm,
 							    &pdd->qpd);
-		if (r) {
+		/* evict return -EIO if HWS is hang or asic is resetting, in this case
+		 * we would like to set all the queues to be in evicted state to prevent
+		 * them been add back since they actually not be saved right now.
+		 */
+		if (r && r != -EIO) {
 			pr_err("Failed to evict process queues\n");
 			goto fail;
 		}
@@ -1724,14 +1778,13 @@ int kfd_process_gpuidx_from_gpuid(struct kfd_process *p, uint32_t gpu_id)
 }
 
 int
-kfd_process_gpuid_from_kgd(struct kfd_process *p, struct amdgpu_device *adev,
+kfd_process_gpuid_from_adev(struct kfd_process *p, struct amdgpu_device *adev,
 			   uint32_t *gpuid, uint32_t *gpuidx)
 {
-	struct kgd_dev *kgd = (struct kgd_dev *)adev;
 	int i;
 
 	for (i = 0; i < p->n_pdds; i++)
-		if (p->pdds[i] && p->pdds[i]->dev->kgd == kgd) {
+		if (p->pdds[i] && p->pdds[i]->dev->adev == adev) {
 			*gpuid = p->pdds[i]->dev->id;
 			*gpuidx = i;
 			return 0;
@@ -1896,10 +1949,10 @@ void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type)
 		 * only happens when the first queue is created.
 		 */
 		if (pdd->qpd.vmid)
-			amdgpu_amdkfd_flush_gpu_tlb_vmid(dev->kgd,
+			amdgpu_amdkfd_flush_gpu_tlb_vmid(dev->adev,
 							pdd->qpd.vmid);
 	} else {
-		amdgpu_amdkfd_flush_gpu_tlb_pasid(dev->kgd,
+		amdgpu_amdkfd_flush_gpu_tlb_pasid(dev->adev,
 					pdd->process->pasid, type);
 	}
 }