12 files changed, 97 insertions, 27 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index b6ba0697c531..3f0300e53727 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -42,6 +42,7 @@
 
 static long kfd_ioctl(struct file *, unsigned int, unsigned long);
 static int kfd_open(struct inode *, struct file *);
+static int kfd_release(struct inode *, struct file *);
 static int kfd_mmap(struct file *, struct vm_area_struct *);
 
 static const char kfd_dev_name[] = "kfd";
@@ -51,6 +52,7 @@ static const struct file_operations kfd_fops = {
 	.unlocked_ioctl = kfd_ioctl,
 	.compat_ioctl = compat_ptr_ioctl,
 	.open = kfd_open,
+	.release = kfd_release,
 	.mmap = kfd_mmap,
 };
 
@@ -124,8 +126,13 @@ static int kfd_open(struct inode *inode, struct file *filep)
 	if (IS_ERR(process))
 		return PTR_ERR(process);
 
-	if (kfd_is_locked())
+	if (kfd_is_locked()) {
+		kfd_unref_process(process);
 		return -EAGAIN;
+	}
+
+	/* filep now owns the reference returned by kfd_create_process */
+	filep->private_data = process;
 
 	dev_dbg(kfd_device, "process %d opened, compat mode (32 bit) - %d\n",
 		process->pasid, process->is_32bit_user_mode);
@@ -133,6 +140,16 @@ static int kfd_open(struct inode *inode, struct file *filep)
 	return 0;
 }
 
+static int kfd_release(struct inode *inode, struct file *filep)
+{
+	struct kfd_process *process = filep->private_data;
+
+	if (process)
+		kfd_unref_process(process);
+
+	return 0;
+}
+
 static int kfd_ioctl_get_version(struct file *filep, struct kfd_process *p,
 					void *data)
 {
@@ -1801,9 +1818,14 @@ static long kfd_ioctl(struct file *filep, unsigned int cmd, unsigned long arg)
 
 	dev_dbg(kfd_device, "ioctl cmd 0x%x (#0x%x), arg 0x%lx\n", cmd, nr, arg);
 
-	process = kfd_get_process(current);
-	if (IS_ERR(process)) {
-		dev_dbg(kfd_device, "no process\n");
+	/* Get the process struct from the filep. Only the process
+	 * that opened /dev/kfd can use the file descriptor. Child
+	 * processes need to create their own KFD device context.
+	 */
+	process = filep->private_data;
+	if (process->lead_thread != current->group_leader) {
+		dev_dbg(kfd_device, "Using KFD FD in wrong process\n");
+		retcode = -EBADF;
 		goto err_i1;
 	}
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c b/drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c
index 15c523027285..511712c2e382 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c
@@ -93,7 +93,7 @@ void kfd_debugfs_init(void)
 			    kfd_debugfs_hqds_by_device, &kfd_debugfs_fops);
 	debugfs_create_file("rls", S_IFREG | 0444, debugfs_root,
 			    kfd_debugfs_rls_by_device, &kfd_debugfs_fops);
-	debugfs_create_file("hang_hws", S_IFREG | 0644, debugfs_root,
+	debugfs_create_file("hang_hws", S_IFREG | 0200, debugfs_root,
 			    NULL, &kfd_debugfs_hang_hws_fops);
 }
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 209bfc849352..2a9e40131735 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -728,6 +728,9 @@ int kgd2kfd_pre_reset(struct kfd_dev *kfd)
 {
 	if (!kfd->init_complete)
 		return 0;
+
+	kfd->dqm->ops.pre_reset(kfd->dqm);
+
 	kgd2kfd_suspend(kfd);
 
 	kfd_signal_reset_event(kfd);
@@ -822,6 +825,21 @@ dqm_start_error:
 	return err;
 }
 
+static inline void kfd_queue_work(struct workqueue_struct *wq,
+				  struct work_struct *work)
+{
+	int cpu, new_cpu;
+
+	cpu = new_cpu = smp_processor_id();
+	do {
+		new_cpu = cpumask_next(new_cpu, cpu_online_mask) % nr_cpu_ids;
+		if (cpu_to_node(new_cpu) == numa_node_id())
+			break;
+	} while (cpu != new_cpu);
+
+	queue_work_on(new_cpu, wq, work);
+}
+
 /* This is called directly from KGD at ISR. */
 void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
 {
@@ -844,7 +862,7 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
 				   patched_ihre, &is_patched)
 	    && enqueue_ih_ring_entry(kfd,
 				     is_patched ? patched_ihre : ih_ring_entry))
-		queue_work(kfd->ih_wq, &kfd->interrupt_work);
+		kfd_queue_work(kfd->ih_wq, &kfd->interrupt_work);
 
 	spin_unlock_irqrestore(&kfd->interrupt_lock, flags);
 }
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index f7f6df40875e..d7eb6ac37f62 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -930,7 +930,6 @@ static void uninitialize(struct device_queue_manager *dqm)
 	for (i = 0 ; i < KFD_MQD_TYPE_MAX ; i++)
 		kfree(dqm->mqd_mgrs[i]);
 	mutex_destroy(&dqm->lock_hidden);
-	kfd_gtt_sa_free(dqm->dev, dqm->pipeline_mem);
 }
 
 static int start_nocpsch(struct device_queue_manager *dqm)
@@ -947,12 +946,19 @@ static int start_nocpsch(struct device_queue_manager *dqm)
 static int stop_nocpsch(struct device_queue_manager *dqm)
 {
 	if (dqm->dev->device_info->asic_family == CHIP_HAWAII)
-		pm_uninit(&dqm->packets);
+		pm_uninit(&dqm->packets, false);
 	dqm->sched_running = false;
 
 	return 0;
 }
 
+static void pre_reset(struct device_queue_manager *dqm)
+{
+	dqm_lock(dqm);
+	dqm->is_resetting = true;
+	dqm_unlock(dqm);
+}
+
 static int allocate_sdma_queue(struct device_queue_manager *dqm,
 				struct queue *q)
 {
@@ -1100,6 +1106,7 @@ static int start_cpsch(struct device_queue_manager *dqm)
 	dqm_lock(dqm);
 	/* clear hang status when driver try to start the hw scheduler */
 	dqm->is_hws_hang = false;
+	dqm->is_resetting = false;
 	dqm->sched_running = true;
 	execute_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 0);
 	dqm_unlock(dqm);
@@ -1107,20 +1114,24 @@ static int start_cpsch(struct device_queue_manager *dqm)
 	return 0;
 fail_allocate_vidmem:
 fail_set_sched_resources:
-	pm_uninit(&dqm->packets);
+	pm_uninit(&dqm->packets, false);
 fail_packet_manager_init:
 	return retval;
 }
 
 static int stop_cpsch(struct device_queue_manager *dqm)
 {
+	bool hanging;
+
 	dqm_lock(dqm);
-	unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0);
+	if (!dqm->is_hws_hang)
+		unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0);
+	hanging = dqm->is_hws_hang || dqm->is_resetting;
 	dqm->sched_running = false;
 	dqm_unlock(dqm);
 
 	kfd_gtt_sa_free(dqm->dev, dqm->fence_mem);
-	pm_uninit(&dqm->packets);
+	pm_uninit(&dqm->packets, hanging);
 
 	return 0;
 }
@@ -1352,8 +1363,17 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm,
 	/* should be timed out */
 	retval = amdkfd_fence_wait_timeout(dqm->fence_addr, KFD_FENCE_COMPLETED,
 				queue_preemption_timeout_ms);
-	if (retval)
+	if (retval) {
+		pr_err("The cp might be in an unrecoverable state due to an unsuccessful queues preemption\n");
+		dqm->is_hws_hang = true;
+		/* It's possible we're detecting a HWS hang in the
+		 * middle of a GPU reset. No need to schedule another
+		 * reset in this case.
+		 */
+		if (!dqm->is_resetting)
+			schedule_work(&dqm->hw_exception_work);
 		return retval;
+	}
 
 	pm_release_ib(&dqm->packets);
 	dqm->active_runlist = false;
@@ -1371,12 +1391,8 @@ static int execute_queues_cpsch(struct device_queue_manager *dqm,
 	if (dqm->is_hws_hang)
 		return -EIO;
 	retval = unmap_queues_cpsch(dqm, filter, filter_param);
-	if (retval) {
-		pr_err("The cp might be in an unrecoverable state due to an unsuccessful queues preemption\n");
-		dqm->is_hws_hang = true;
-		schedule_work(&dqm->hw_exception_work);
+	if (retval)
 		return retval;
-	}
 
 	return map_queues_cpsch(dqm);
 }
@@ -1770,6 +1786,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 		dqm->ops.initialize = initialize_cpsch;
 		dqm->ops.start = start_cpsch;
 		dqm->ops.stop = stop_cpsch;
+		dqm->ops.pre_reset = pre_reset;
 		dqm->ops.destroy_queue = destroy_queue_cpsch;
 		dqm->ops.update_queue = update_queue;
 		dqm->ops.register_process = register_process;
@@ -1788,6 +1805,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 		/* initialize dqm for no cp scheduling */
 		dqm->ops.start = start_nocpsch;
 		dqm->ops.stop = stop_nocpsch;
+		dqm->ops.pre_reset = pre_reset;
 		dqm->ops.create_queue = create_queue_nocpsch;
 		dqm->ops.destroy_queue = destroy_queue_nocpsch;
 		dqm->ops.update_queue = update_queue;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index a8c37e6da027..871d3b628d2d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -104,6 +104,7 @@ struct device_queue_manager_ops {
 	int	(*initialize)(struct device_queue_manager *dqm);
 	int	(*start)(struct device_queue_manager *dqm);
 	int	(*stop)(struct device_queue_manager *dqm);
+	void	(*pre_reset)(struct device_queue_manager *dqm);
 	void	(*uninitialize)(struct device_queue_manager *dqm);
 	int	(*create_kernel_queue)(struct device_queue_manager *dqm,
 					struct kernel_queue *kq,
@@ -190,7 +191,6 @@ struct device_queue_manager {
 	/* the pasid mapping for each kfd vmid */
 	uint16_t		vmid_pasid[VMID_NUM];
 	uint64_t		pipelines_addr;
-	struct kfd_mem_obj	*pipeline_mem;
 	uint64_t		fence_gpu_addr;
 	unsigned int		*fence_addr;
 	struct kfd_mem_obj	*fence_mem;
@@ -199,6 +199,7 @@ struct device_queue_manager {
 
 	/* hw exception  */
 	bool			is_hws_hang;
+	bool			is_resetting;
 	struct work_struct	hw_exception_work;
 	struct kfd_mem_obj	hiq_sdma_mqd;
 	bool			sched_running;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
index 2d56dc534459..bae706462f96 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
@@ -195,9 +195,9 @@ err_get_kernel_doorbell:
 }
 
 /* Uninitialize a kernel queue and free all its memory usages. */
-static void kq_uninitialize(struct kernel_queue *kq)
+static void kq_uninitialize(struct kernel_queue *kq, bool hanging)
 {
-	if (kq->queue->properties.type == KFD_QUEUE_TYPE_HIQ)
+	if (kq->queue->properties.type == KFD_QUEUE_TYPE_HIQ && !hanging)
 		kq->mqd_mgr->destroy_mqd(kq->mqd_mgr,
 					kq->queue->mqd,
 					KFD_PREEMPT_TYPE_WAVEFRONT_RESET,
@@ -337,9 +337,9 @@ struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
 	return NULL;
 }
 
-void kernel_queue_uninit(struct kernel_queue *kq)
+void kernel_queue_uninit(struct kernel_queue *kq, bool hanging)
 {
-	kq_uninitialize(kq);
+	kq_uninitialize(kq, hanging);
 	kfree(kq);
 }
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
index 6cabed06ef5d..dc406e6dee23 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
@@ -264,10 +264,10 @@ int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm)
 	return 0;
 }
 
-void pm_uninit(struct packet_manager *pm)
+void pm_uninit(struct packet_manager *pm, bool hanging)
 {
 	mutex_destroy(&pm->lock);
-	kernel_queue_uninit(pm->priv_queue);
+	kernel_queue_uninit(pm->priv_queue, hanging);
 }
 
 int pm_send_set_resources(struct packet_manager *pm,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index fc61b5ec068e..6af1b5881f43 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -883,7 +883,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev);
 void device_queue_manager_uninit(struct device_queue_manager *dqm);
 struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
 					enum kfd_queue_type type);
-void kernel_queue_uninit(struct kernel_queue *kq);
+void kernel_queue_uninit(struct kernel_queue *kq, bool hanging);
 int kfd_process_vm_fault(struct device_queue_manager *dqm, unsigned int pasid);
 
 /* Process Queue Manager */
@@ -972,7 +972,7 @@ extern const struct packet_manager_funcs kfd_vi_pm_funcs;
 extern const struct packet_manager_funcs kfd_v9_pm_funcs;
 
 int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm);
-void pm_uninit(struct packet_manager *pm);
+void pm_uninit(struct packet_manager *pm, bool hanging);
 int pm_send_set_resources(struct packet_manager *pm,
 				struct scheduling_resources *res);
 int pm_send_runlist(struct packet_manager *pm, struct list_head *dqm_queues);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 8276601a122f..536a153ac9a4 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -324,6 +324,8 @@ struct kfd_process *kfd_create_process(struct file *filep)
 					(int)process->lead_thread->pid);
 	}
 out:
+	if (!IS_ERR(process))
+		kref_get(&process->ref);
 	mutex_unlock(&kfd_processes_mutex);
 
 	return process;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
index 1152490bbf53..31fcd1b51f00 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -374,7 +374,7 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid)
 		/* destroy kernel queue (DIQ) */
 		dqm = pqn->kq->dev->dqm;
 		dqm->ops.destroy_kernel_queue(dqm, pqn->kq, &pdd->qpd);
-		kernel_queue_uninit(pqn->kq);
+		kernel_queue_uninit(pqn->kq, false);
 	}
 
 	if (pqn->q) {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index 69bd0628fdc6..203c823d65f1 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -486,6 +486,10 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
 			dev->node_props.num_sdma_engines);
 	sysfs_show_32bit_prop(buffer, "num_sdma_xgmi_engines",
 			dev->node_props.num_sdma_xgmi_engines);
+	sysfs_show_32bit_prop(buffer, "num_sdma_queues_per_engine",
+			dev->node_props.num_sdma_queues_per_engine);
+	sysfs_show_32bit_prop(buffer, "num_cp_queues",
+			dev->node_props.num_cp_queues);
 
 	if (dev->gpu) {
 		log_max_watch_addr =
@@ -1309,9 +1313,12 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 	dev->node_props.num_sdma_engines = gpu->device_info->num_sdma_engines;
 	dev->node_props.num_sdma_xgmi_engines =
 				gpu->device_info->num_xgmi_sdma_engines;
+	dev->node_props.num_sdma_queues_per_engine =
+				gpu->device_info->num_sdma_queues_per_engine;
 	dev->node_props.num_gws = (hws_gws_support &&
 		dev->gpu->dqm->sched_policy != KFD_SCHED_POLICY_NO_HWS) ?
 		amdgpu_amdkfd_get_num_gws(dev->gpu->kgd) : 0;
+	dev->node_props.num_cp_queues = get_queues_num(dev->gpu->dqm);
 
 	kfd_fill_mem_clk_max_info(dev);
 	kfd_fill_iolink_non_crat_info(dev);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
index 15843e0fc756..74e9b1682af8 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
@@ -81,6 +81,8 @@ struct kfd_node_properties {
 	int32_t  drm_render_minor;
 	uint32_t num_sdma_engines;
 	uint32_t num_sdma_xgmi_engines;
+	uint32_t num_sdma_queues_per_engine;
+	uint32_t num_cp_queues;
 	char name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE];
 };