1 files changed, 121 insertions, 97 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index d4aa843aacfd..722ac1662bdc 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -35,6 +35,7 @@
 #include <linux/pm_runtime.h>
 #include "amdgpu_amdkfd.h"
 #include "amdgpu.h"
+#include "amdgpu_reset.h"
 
 struct mm_struct;
 
@@ -271,11 +272,9 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer)
 	struct kfd_process *proc = NULL;
 	struct kfd_process_device *pdd = NULL;
 	int i;
-	struct kfd_cu_occupancy cu_occupancy[AMDGPU_MAX_QUEUES];
+	struct kfd_cu_occupancy *cu_occupancy;
 	u32 queue_format;
 
-	memset(cu_occupancy, 0x0, sizeof(cu_occupancy));
-
 	pdd = container_of(attr, struct kfd_process_device, attr_cu_occupancy);
 	dev = pdd->dev;
 	if (dev->kfd2kgd->get_cu_occupancy == NULL)
@@ -284,8 +283,8 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer)
 	cu_cnt = 0;
 	proc = pdd->process;
 	if (pdd->qpd.queue_count == 0) {
-		pr_debug("Gpu-Id: %d has no active queues for process %d\n",
-			 dev->id, proc->pasid);
+		pr_debug("Gpu-Id: %d has no active queues for process pid %d\n",
+			 dev->id, (int)proc->lead_thread->pid);
 		return snprintf(buffer, PAGE_SIZE, "%d\n", cu_cnt);
 	}
 
@@ -293,6 +292,10 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer)
 	wave_cnt = 0;
 	max_waves_per_cu = 0;
 
+	cu_occupancy = kcalloc(AMDGPU_MAX_QUEUES, sizeof(*cu_occupancy), GFP_KERNEL);
+	if (!cu_occupancy)
+		return -ENOMEM;
+
 	/*
 	 * For GFX 9.4.3, fetch the CU occupancy from the first XCC in the partition.
 	 * For AQL queues, because of cooperative dispatch we multiply the wave count
@@ -318,18 +321,16 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer)
 
 	/* Translate wave count to number of compute units */
 	cu_cnt = (wave_cnt + (max_waves_per_cu - 1)) / max_waves_per_cu;
+	kfree(cu_occupancy);
 	return snprintf(buffer, PAGE_SIZE, "%d\n", cu_cnt);
 }
 
 static ssize_t kfd_procfs_show(struct kobject *kobj, struct attribute *attr,
 			       char *buffer)
 {
-	if (strcmp(attr->name, "pasid") == 0) {
-		struct kfd_process *p = container_of(attr, struct kfd_process,
-						     attr_pasid);
-
-		return snprintf(buffer, PAGE_SIZE, "%d\n", p->pasid);
-	} else if (strncmp(attr->name, "vram_", 5) == 0) {
+	if (strcmp(attr->name, "pasid") == 0)
+		return snprintf(buffer, PAGE_SIZE, "%d\n", 0);
+	else if (strncmp(attr->name, "vram_", 5) == 0) {
 		struct kfd_process_device *pdd = container_of(attr, struct kfd_process_device,
 							      attr_vram);
 		return snprintf(buffer, PAGE_SIZE, "%llu\n", atomic64_read(&pdd->vram_usage));
@@ -338,8 +339,8 @@ static ssize_t kfd_procfs_show(struct kobject *kobj, struct attribute *attr,
 							      attr_sdma);
 		struct kfd_sdma_activity_handler_workarea sdma_activity_work_handler;
 
-		INIT_WORK(&sdma_activity_work_handler.sdma_activity_work,
-					kfd_sdma_activity_worker);
+		INIT_WORK_ONSTACK(&sdma_activity_work_handler.sdma_activity_work,
+				  kfd_sdma_activity_worker);
 
 		sdma_activity_work_handler.pdd = pdd;
 		sdma_activity_work_handler.sdma_activity_counter = 0;
@@ -347,6 +348,7 @@ static ssize_t kfd_procfs_show(struct kobject *kobj, struct attribute *attr,
 		schedule_work(&sdma_activity_work_handler.sdma_activity_work);
 
 		flush_work(&sdma_activity_work_handler.sdma_activity_work);
+		destroy_work_on_stack(&sdma_activity_work_handler.sdma_activity_work);
 
 		return snprintf(buffer, PAGE_SIZE, "%llu\n",
 				(sdma_activity_work_handler.sdma_activity_counter)/
@@ -837,6 +839,14 @@ struct kfd_process *kfd_create_process(struct task_struct *thread)
 		return ERR_PTR(-EINVAL);
 	}
 
+	/* If the process just called exec(3), it is possible that the
+	 * cleanup of the kfd_process (following the release of the mm
+	 * of the old process image) is still in the cleanup work queue.
+	 * Make sure to drain any job before trying to recreate any
+	 * resource for this process.
+	 */
+	flush_workqueue(kfd_process_wq);
+
 	/*
 	 * take kfd processes mutex before starting of process creation
 	 * so there won't be a case where two threads of the same process
@@ -850,19 +860,13 @@ struct kfd_process *kfd_create_process(struct task_struct *thread)
 		goto out;
 	}
 
-	/* A prior open of /dev/kfd could have already created the process. */
-	process = find_process(thread, false);
+	/* A prior open of /dev/kfd could have already created the process.
+	 * find_process will increase process kref in this case
+	 */
+	process = find_process(thread, true);
 	if (process) {
 		pr_debug("Process already found\n");
 	} else {
-		/* If the process just called exec(3), it is possible that the
-		 * cleanup of the kfd_process (following the release of the mm
-		 * of the old process image) is still in the cleanup work queue.
-		 * Make sure to drain any job before trying to recreate any
-		 * resource for this process.
-		 */
-		flush_workqueue(kfd_process_wq);
-
 		process = create_process(thread);
 		if (IS_ERR(process))
 			goto out;
@@ -896,11 +900,11 @@ struct kfd_process *kfd_create_process(struct task_struct *thread)
 		kfd_procfs_add_sysfs_files(process);
 		kfd_procfs_add_sysfs_counters(process);
 
+		kfd_debugfs_add_process(process);
+
 		init_waitqueue_head(&process->wait_irq_drain);
 	}
 out:
-	if (!IS_ERR(process))
-		kref_get(&process->ref);
 	mutex_unlock(&kfd_processes_mutex);
 	mmput(thread->mm);
 
@@ -1052,17 +1056,15 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)
 	for (i = 0; i < p->n_pdds; i++) {
 		struct kfd_process_device *pdd = p->pdds[i];
 
-		pr_debug("Releasing pdd (topology id %d) for process (pasid 0x%x)\n",
-				pdd->dev->id, p->pasid);
+		kfd_smi_event_process(pdd, false);
 
+		pr_debug("Releasing pdd (topology id %d, for pid %d)\n",
+			pdd->dev->id, p->lead_thread->pid);
 		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->adev, pdd->drm_priv);
+		if (pdd->drm_file)
 			fput(pdd->drm_file);
-		}
 
 		if (pdd->qpd.cwsr_kaddr && !pdd->qpd.cwsr_base)
 			free_pages((unsigned long)pdd->qpd.cwsr_kaddr,
@@ -1072,7 +1074,8 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)
 
 		kfd_free_process_doorbells(pdd->dev->kfd, pdd);
 
-		if (pdd->dev->kfd->shared_resources.enable_mes)
+		if (pdd->dev->kfd->shared_resources.enable_mes &&
+			pdd->proc_ctx_cpu_ptr)
 			amdgpu_amdkfd_free_gtt_mem(pdd->dev->adev,
 						   &pdd->proc_ctx_bo);
 		/*
@@ -1135,6 +1138,17 @@ static void kfd_process_remove_sysfs(struct kfd_process *p)
 	p->kobj = NULL;
 }
 
+/*
+ * If any GPU is ongoing reset, wait for reset complete.
+ */
+static void kfd_process_wait_gpu_reset_complete(struct kfd_process *p)
+{
+	int i;
+
+	for (i = 0; i < p->n_pdds; i++)
+		flush_workqueue(p->pdds[i]->dev->adev->reset_domain->wq);
+}
+
 /* No process locking is needed in this function, because the process
  * is not findable any more. We must assume that no other thread is
  * using it any more, otherwise we couldn't safely free the process
@@ -1149,15 +1163,22 @@ static void kfd_process_wq_release(struct work_struct *work)
 	kfd_process_dequeue_from_all_devices(p);
 	pqm_uninit(&p->pqm);
 
+	/*
+	 * If GPU in reset, user queues may still running, wait for reset complete.
+	 */
+	kfd_process_wait_gpu_reset_complete(p);
+
 	/* Signal the eviction fence after user mode queues are
 	 * destroyed. This allows any BOs to be freed without
 	 * triggering pointless evictions or waiting for fences.
 	 */
 	synchronize_rcu();
 	ef = rcu_access_pointer(p->ef);
-	dma_fence_signal(ef);
+	if (ef)
+		dma_fence_signal(ef);
 
 	kfd_process_remove_sysfs(p);
+	kfd_debugfs_remove_process(p);
 
 	kfd_process_kunmap_signal_bo(p);
 	kfd_process_free_outstanding_kfd_bos(p);
@@ -1168,7 +1189,6 @@ static void kfd_process_wq_release(struct work_struct *work)
 
 	kfd_event_free_process(p);
 
-	kfd_pasid_free(p->pasid);
 	mutex_destroy(&p->mutex);
 
 	put_task_struct(p->lead_thread);
@@ -1186,10 +1206,8 @@ static void kfd_process_ref_release(struct kref *ref)
 
 static struct mmu_notifier *kfd_process_alloc_notifier(struct mm_struct *mm)
 {
-	int idx = srcu_read_lock(&kfd_processes_srcu);
-	struct kfd_process *p = find_process_by_mm(mm);
-
-	srcu_read_unlock(&kfd_processes_srcu, idx);
+	/* This increments p->ref counter if kfd process p exists */
+	struct kfd_process *p = kfd_lookup_process_by_mm(mm);
 
 	return p ? &p->mmu_notifier : ERR_PTR(-ESRCH);
 }
@@ -1521,12 +1539,6 @@ static struct kfd_process *create_process(const struct task_struct *thread)
 	atomic_set(&process->debugged_process_count, 0);
 	sema_init(&process->runtime_enable_sema, 0);
 
-	process->pasid = kfd_pasid_alloc();
-	if (process->pasid == 0) {
-		err = -ENOSPC;
-		goto err_alloc_pasid;
-	}
-
 	err = pqm_init(&process->pqm, process);
 	if (err != 0)
 		goto err_process_pqm_init;
@@ -1580,8 +1592,6 @@ err_init_svm_range_list:
 err_init_apertures:
 	pqm_uninit(&process->pqm);
 err_process_pqm_init:
-	kfd_pasid_free(process->pasid);
-err_alloc_pasid:
 	kfd_event_free_process(process);
 err_event_init:
 	mutex_destroy(&process->mutex);
@@ -1606,7 +1616,6 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev,
 							struct kfd_process *p)
 {
 	struct kfd_process_device *pdd = NULL;
-	int retval = 0;
 
 	if (WARN_ON_ONCE(p->n_pdds >= MAX_GPU_INSTANCE))
 		return NULL;
@@ -1630,21 +1639,6 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev,
 	pdd->user_gpu_id = dev->id;
 	atomic64_set(&pdd->evict_duration_counter, 0);
 
-	if (dev->kfd->shared_resources.enable_mes) {
-		retval = amdgpu_amdkfd_alloc_gtt_mem(dev->adev,
-						AMDGPU_MES_PROC_CTX_SIZE,
-						&pdd->proc_ctx_bo,
-						&pdd->proc_ctx_gpu_addr,
-						&pdd->proc_ctx_cpu_ptr,
-						false);
-		if (retval) {
-			dev_err(dev->adev->dev,
-				"failed to allocate process context bo\n");
-			goto err_free_pdd;
-		}
-		memset(pdd->proc_ctx_cpu_ptr, 0, AMDGPU_MES_PROC_CTX_SIZE);
-	}
-
 	p->pdds[p->n_pdds++] = pdd;
 	if (kfd_dbg_is_per_vmid_supported(pdd->dev))
 		pdd->spi_dbg_override = pdd->dev->kfd2kgd->disable_debug_trap(
@@ -1656,10 +1650,6 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev,
 	idr_init(&pdd->alloc_idr);
 
 	return pdd;
-
-err_free_pdd:
-	kfree(pdd);
-	return NULL;
 }
 
 /**
@@ -1720,15 +1710,21 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,
 	if (ret)
 		goto err_init_cwsr;
 
-	ret = amdgpu_amdkfd_gpuvm_set_vm_pasid(dev->adev, avm, p->pasid);
-	if (ret)
-		goto err_set_pasid;
+	if (unlikely(!avm->pasid)) {
+		dev_warn(pdd->dev->adev->dev, "WARN: vm %p has no pasid associated",
+				 avm);
+		ret = -EINVAL;
+		goto err_get_pasid;
+	}
 
+	pdd->pasid = avm->pasid;
 	pdd->drm_file = drm_file;
 
+	kfd_smi_event_process(pdd, true);
+
 	return 0;
 
-err_set_pasid:
+err_get_pasid:
 	kfd_process_device_destroy_cwsr_dgpu(pdd);
 err_init_cwsr:
 	kfd_process_device_destroy_ib_mem(pdd);
@@ -1814,25 +1810,50 @@ void kfd_process_device_remove_obj_handle(struct kfd_process_device *pdd,
 		idr_remove(&pdd->alloc_idr, handle);
 }
 
-/* This increments the process->ref counter. */
-struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid)
+static struct kfd_process_device *kfd_lookup_process_device_by_pasid(u32 pasid)
 {
-	struct kfd_process *p, *ret_p = NULL;
+	struct kfd_process_device *ret_p = NULL;
+	struct kfd_process *p;
 	unsigned int temp;
-
-	int idx = srcu_read_lock(&kfd_processes_srcu);
+	int i;
 
 	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
-		if (p->pasid == pasid) {
-			kref_get(&p->ref);
-			ret_p = p;
-			break;
+		for (i = 0; i < p->n_pdds; i++) {
+			if (p->pdds[i]->pasid == pasid) {
+				ret_p = p->pdds[i];
+				break;
+			}
 		}
+		if (ret_p)
+			break;
+	}
+	return ret_p;
+}
+
+/* This increments the process->ref counter. */
+struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid,
+						struct kfd_process_device **pdd)
+{
+	struct kfd_process_device *ret_p;
+
+	int idx = srcu_read_lock(&kfd_processes_srcu);
+
+	ret_p = kfd_lookup_process_device_by_pasid(pasid);
+	if (ret_p) {
+		if (pdd)
+			*pdd = ret_p;
+		kref_get(&ret_p->process->ref);
+
+		srcu_read_unlock(&kfd_processes_srcu, idx);
+		return ret_p->process;
 	}
 
 	srcu_read_unlock(&kfd_processes_srcu, idx);
 
-	return ret_p;
+	if (pdd)
+		*pdd = NULL;
+
+	return NULL;
 }
 
 /* This increments the process->ref counter. */
@@ -1988,7 +2009,7 @@ static void evict_process_worker(struct work_struct *work)
 	 */
 	p = container_of(dwork, struct kfd_process, eviction_work);
 
-	pr_debug("Started evicting pasid 0x%x\n", p->pasid);
+	pr_debug("Started evicting process pid %d\n", p->lead_thread->pid);
 	ret = kfd_process_evict_queues(p, KFD_QUEUE_EVICTION_TRIGGER_TTM);
 	if (!ret) {
 		/* If another thread already signaled the eviction fence,
@@ -2000,9 +2021,9 @@ static void evict_process_worker(struct work_struct *work)
 				     msecs_to_jiffies(PROCESS_RESTORE_TIME_MS)))
 			kfd_process_restore_queues(p);
 
-		pr_debug("Finished evicting pasid 0x%x\n", p->pasid);
+		pr_debug("Finished evicting process pid %d\n", p->lead_thread->pid);
 	} else
-		pr_err("Failed to evict queues of pasid 0x%x\n", p->pasid);
+		pr_err("Failed to evict queues of process pid %d\n", p->lead_thread->pid);
 }
 
 static int restore_process_helper(struct kfd_process *p)
@@ -2019,9 +2040,11 @@ static int restore_process_helper(struct kfd_process *p)
 
 	ret = kfd_process_restore_queues(p);
 	if (!ret)
-		pr_debug("Finished restoring pasid 0x%x\n", p->pasid);
+		pr_debug("Finished restoring process pid %d\n",
+			p->lead_thread->pid);
 	else
-		pr_err("Failed to restore queues of pasid 0x%x\n", p->pasid);
+		pr_err("Failed to restore queues of process pid %d\n",
+		      p->lead_thread->pid);
 
 	return ret;
 }
@@ -2038,7 +2061,7 @@ static void restore_process_worker(struct work_struct *work)
 	 * lifetime of this thread, kfd_process p will be valid
 	 */
 	p = container_of(dwork, struct kfd_process, restore_work);
-	pr_debug("Started restoring pasid 0x%x\n", p->pasid);
+	pr_debug("Started restoring process pasid %d\n", (int)p->lead_thread->pid);
 
 	/* Setting last_restore_timestamp before successful restoration.
 	 * Otherwise this would have to be set by KGD (restore_process_bos)
@@ -2054,8 +2077,8 @@ static void restore_process_worker(struct work_struct *work)
 
 	ret = restore_process_helper(p);
 	if (ret) {
-		pr_debug("Failed to restore BOs of pasid 0x%x, retry after %d ms\n",
-			 p->pasid, PROCESS_BACK_OFF_TIME_MS);
+		pr_debug("Failed to restore BOs of process pid %d, retry after %d ms\n",
+			 p->lead_thread->pid, PROCESS_BACK_OFF_TIME_MS);
 		if (mod_delayed_work(kfd_restore_wq, &p->restore_work,
 				     msecs_to_jiffies(PROCESS_RESTORE_TIME_MS)))
 			kfd_process_restore_queues(p);
@@ -2071,7 +2094,7 @@ void kfd_suspend_all_processes(void)
 	WARN(debug_evictions, "Evicting all processes");
 	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
 		if (kfd_process_evict_queues(p, KFD_QUEUE_EVICTION_TRIGGER_SUSPEND))
-			pr_err("Failed to suspend process 0x%x\n", p->pasid);
+			pr_err("Failed to suspend process pid %d\n", p->lead_thread->pid);
 		signal_eviction_fence(p);
 	}
 	srcu_read_unlock(&kfd_processes_srcu, idx);
@@ -2085,8 +2108,8 @@ int kfd_resume_all_processes(void)
 
 	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
 		if (restore_process_helper(p)) {
-			pr_err("Restore process %d failed during resume\n",
-			       p->pasid);
+			pr_err("Restore process pid %d failed during resume\n",
+			      p->lead_thread->pid);
 			ret = -EFAULT;
 		}
 	}
@@ -2141,13 +2164,14 @@ int kfd_process_drain_interrupts(struct kfd_process_device *pdd)
 	memset(irq_drain_fence, 0, sizeof(irq_drain_fence));
 	irq_drain_fence[0] = (KFD_IRQ_FENCE_SOURCEID << 8) |
 							KFD_IRQ_FENCE_CLIENTID;
-	irq_drain_fence[3] = pdd->process->pasid;
+	irq_drain_fence[3] = pdd->pasid;
 
 	/*
-	 * For GFX 9.4.3, send the NodeId also in IH cookie DW[3]
+	 * For GFX 9.4.3/9.5.0, send the NodeId also in IH cookie DW[3]
 	 */
 	if (KFD_GC_VERSION(pdd->dev->kfd) == IP_VERSION(9, 4, 3) ||
-	    KFD_GC_VERSION(pdd->dev->kfd) == IP_VERSION(9, 4, 4)) {
+	    KFD_GC_VERSION(pdd->dev->kfd) == IP_VERSION(9, 4, 4) ||
+	    KFD_GC_VERSION(pdd->dev->kfd) == IP_VERSION(9, 5, 0)) {
 		node_id = ffs(pdd->dev->interrupt_bitmap) - 1;
 		irq_drain_fence[3] |= node_id << 16;
 	}
@@ -2171,7 +2195,7 @@ void kfd_process_close_interrupt_drain(unsigned int pasid)
 {
 	struct kfd_process *p;
 
-	p = kfd_lookup_process_by_pasid(pasid);
+	p = kfd_lookup_process_by_pasid(pasid, NULL);
 
 	if (!p)
 		return;
@@ -2292,8 +2316,8 @@ int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data)
 	int idx = srcu_read_lock(&kfd_processes_srcu);
 
 	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
-		seq_printf(m, "Process %d PASID 0x%x:\n",
-			   p->lead_thread->tgid, p->pasid);
+		seq_printf(m, "Process %d PASID %d:\n",
+			   p->lead_thread->tgid, p->lead_thread->pid);
 
 		mutex_lock(&p->mutex);
 		r = pqm_debugfs_mqds(m, &p->pqm);