kfd_process.c

/*
 * Copyright 2014 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 */

#include <linux/mutex.h>
#include <linux/log2.h>
#include <linux/sched.h>
#include <linux/sched/mm.h>
#include <linux/sched/task.h>
#include <linux/mmu_context.h>
#include <linux/slab.h>
#include <linux/amd-iommu.h>
#include <linux/notifier.h>
#include <linux/compat.h>
#include <linux/mman.h>
#include <linux/file.h>
#include <linux/pm_runtime.h>
#include "amdgpu_amdkfd.h"
#include "amdgpu.h"

struct mm_struct;

#include "kfd_priv.h"
#include "kfd_device_queue_manager.h"
#include "kfd_dbgmgr.h"
#include "kfd_iommu.h"
#include "kfd_svm.h"

/*
 * List of struct kfd_process (field kfd_process).
 * Unique/indexed by mm_struct*
 */
DEFINE_HASHTABLE(kfd_processes_table, KFD_PROCESS_TABLE_SIZE);
static DEFINE_MUTEX(kfd_processes_mutex);

DEFINE_SRCU(kfd_processes_srcu);

/* For process termination handling */
static struct workqueue_struct *kfd_process_wq;

/* Ordered, single-threaded workqueue for restoring evicted
 * processes. Restoring multiple processes concurrently under memory
 * pressure can lead to processes blocking each other from validating
 * their BOs and result in a live-lock situation where processes
 * remain evicted indefinitely.
 */
static struct workqueue_struct *kfd_restore_wq;

static struct kfd_process *find_process(const struct task_struct *thread);
static void kfd_process_ref_release(struct kref *ref);
static struct kfd_process *create_process(const struct task_struct *thread);
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);

struct kfd_procfs_tree {
	struct kobject *kobj;
};

static struct kfd_procfs_tree procfs;

/*
 * Structure for SDMA activity tracking
 */
struct kfd_sdma_activity_handler_workarea {
	struct work_struct sdma_activity_work;
	struct kfd_process_device *pdd;
	uint64_t sdma_activity_counter;
};

struct temp_sdma_queue_list {
	uint64_t __user *rptr;
	uint64_t sdma_val;
	unsigned int queue_id;
	struct list_head list;
};

static void kfd_sdma_activity_worker(struct work_struct *work)
{
	struct kfd_sdma_activity_handler_workarea *workarea;
	struct kfd_process_device *pdd;
	uint64_t val;
	struct mm_struct *mm;
	struct queue *q;
	struct qcm_process_device *qpd;
	struct device_queue_manager *dqm;
	int ret = 0;
	struct temp_sdma_queue_list sdma_q_list;
	struct temp_sdma_queue_list *sdma_q, *next;

	workarea = container_of(work, struct kfd_sdma_activity_handler_workarea,
				sdma_activity_work);

	pdd = workarea->pdd;
	if (!pdd)
		return;
	dqm = pdd->dev->dqm;
	qpd = &pdd->qpd;
	if (!dqm || !qpd)
		return;
	/*
	 * Total SDMA activity is current SDMA activity + past SDMA activity
	 * Past SDMA count is stored in pdd.
	 * To get the current activity counters for all active SDMA queues,
	 * we loop over all SDMA queues and get their counts from user-space.
	 *
	 * We cannot call get_user() with dqm_lock held as it can cause
	 * a circular lock dependency situation. To read the SDMA stats,
	 * we need to do the following:
	 *
	 * 1. Create a temporary list of SDMA queue nodes from the qpd->queues_list,
	 *    with dqm_lock/dqm_unlock().
	 * 2. Call get_user() for each node in temporary list without dqm_lock.
	 *    Save the SDMA count for each node and also add the count to the total
	 *    SDMA count counter.
	 *    Its possible, during this step, a few SDMA queue nodes got deleted
	 *    from the qpd->queues_list.
	 * 3. Do a second pass over qpd->queues_list to check if any nodes got deleted.
	 *    If any node got deleted, its SDMA count would be captured in the sdma
	 *    past activity counter. So subtract the SDMA counter stored in step 2
	 *    for this node from the total SDMA count.
	 */
	INIT_LIST_HEAD(&sdma_q_list.list);

	/*
	 * Create the temp list of all SDMA queues
	 */
	dqm_lock(dqm);

	list_for_each_entry(q, &qpd->queues_list, list) {
		if ((q->properties.type != KFD_QUEUE_TYPE_SDMA) &&
		    (q->properties.type != KFD_QUEUE_TYPE_SDMA_XGMI))
			continue;

		sdma_q = kzalloc(sizeof(struct temp_sdma_queue_list), GFP_KERNEL);
		if (!sdma_q) {
			dqm_unlock(dqm);
			goto cleanup;
		}

		INIT_LIST_HEAD(&sdma_q->list);
		sdma_q->rptr = (uint64_t __user *)q->properties.read_ptr;
		sdma_q->queue_id = q->properties.queue_id;
		list_add_tail(&sdma_q->list, &sdma_q_list.list);
	}

	/*
	 * If the temp list is empty, then no SDMA queues nodes were found in
	 * qpd->queues_list. Return the past activity count as the total sdma
	 * count
	 */
	if (list_empty(&sdma_q_list.list)) {
		workarea->sdma_activity_counter = pdd->sdma_past_activity_counter;
		dqm_unlock(dqm);
		return;
	}

	dqm_unlock(dqm);

	/*
	 * Get the usage count for each SDMA queue in temp_list.
	 */
	mm = get_task_mm(pdd->process->lead_thread);
	if (!mm)
		goto cleanup;

	kthread_use_mm(mm);

	list_for_each_entry(sdma_q, &sdma_q_list.list, list) {
		val = 0;
		ret = read_sdma_queue_counter(sdma_q->rptr, &val);
		if (ret) {
			pr_debug("Failed to read SDMA queue active counter for queue id: %d",
				 sdma_q->queue_id);
		} else {
			sdma_q->sdma_val = val;
			workarea->sdma_activity_counter += val;
		}
	}

	kthread_unuse_mm(mm);
	mmput(mm);

	/*
	 * Do a second iteration over qpd_queues_list to check if any SDMA
	 * nodes got deleted while fetching SDMA counter.
	 */
	dqm_lock(dqm);

	workarea->sdma_activity_counter += pdd->sdma_past_activity_counter;

	list_for_each_entry(q, &qpd->queues_list, list) {
		if (list_empty(&sdma_q_list.list))
			break;

		if ((q->properties.type != KFD_QUEUE_TYPE_SDMA) &&
		    (q->properties.type != KFD_QUEUE_TYPE_SDMA_XGMI))
			continue;

		list_for_each_entry_safe(sdma_q, next, &sdma_q_list.list, list) {
			if (((uint64_t __user *)q->properties.read_ptr == sdma_q->rptr) &&
			     (sdma_q->queue_id == q->properties.queue_id)) {
				list_del(&sdma_q->list);
				kfree(sdma_q);
				break;
			}
		}
	}

	dqm_unlock(dqm);

	/*
	 * If temp list is not empty, it implies some queues got deleted
	 * from qpd->queues_list during SDMA usage read. Subtract the SDMA
	 * count for each node from the total SDMA count.
	 */
	list_for_each_entry_safe(sdma_q, next, &sdma_q_list.list, list) {
		workarea->sdma_activity_counter -= sdma_q->sdma_val;
		list_del(&sdma_q->list);
		kfree(sdma_q);
	}

	return;

cleanup:
	list_for_each_entry_safe(sdma_q, next, &sdma_q_list.list, list) {
		list_del(&sdma_q->list);
		kfree(sdma_q);
	}
}

/**
 * @kfd_get_cu_occupancy - Collect number of waves in-flight on this device
 * by current process. Translates acquired wave count into number of compute units
 * that are occupied.
 *
 * @atr: Handle of attribute that allows reporting of wave count. The attribute
 * handle encapsulates GPU device it is associated with, thereby allowing collection
 * of waves in flight, etc
 *
 * @buffer: Handle of user provided buffer updated with wave count
 *
 * Return: Number of bytes written to user buffer or an error value
 */
static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer)
{
	int cu_cnt;
	int wave_cnt;
	int max_waves_per_cu;
	struct kfd_dev *dev = NULL;
	struct kfd_process *proc = NULL;
	struct kfd_process_device *pdd = NULL;

	pdd = container_of(attr, struct kfd_process_device, attr_cu_occupancy);
	dev = pdd->dev;
	if (dev->kfd2kgd->get_cu_occupancy == NULL)
		return -EINVAL;

	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);
		return snprintf(buffer, PAGE_SIZE, "%d\n", cu_cnt);
	}

	/* 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,
			&max_waves_per_cu);

	/* Translate wave count to number of compute units */
	cu_cnt = (wave_cnt + (max_waves_per_cu - 1)) / max_waves_per_cu;
	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) {
		struct kfd_process_device *pdd = container_of(attr, struct kfd_process_device,
							      attr_vram);
		return snprintf(buffer, PAGE_SIZE, "%llu\n", READ_ONCE(pdd->vram_usage));
	} else if (strncmp(attr->name, "sdma_", 5) == 0) {
		struct kfd_process_device *pdd = container_of(attr, struct kfd_process_device,
							      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);

		sdma_activity_work_handler.pdd = pdd;
		sdma_activity_work_handler.sdma_activity_counter = 0;

		schedule_work(&sdma_activity_work_handler.sdma_activity_work);

		flush_work(&sdma_activity_work_handler.sdma_activity_work);

		return snprintf(buffer, PAGE_SIZE, "%llu\n",
				(sdma_activity_work_handler.sdma_activity_counter)/
				 SDMA_ACTIVITY_DIVISOR);
	} else {
		pr_err("Invalid attribute");
		return -EINVAL;
	}

	return 0;
}

static void kfd_procfs_kobj_release(struct kobject *kobj)
{
	kfree(kobj);
}

static const struct sysfs_ops kfd_procfs_ops = {
	.show = kfd_procfs_show,
};

static struct kobj_type procfs_type = {
	.release = kfd_procfs_kobj_release,
	.sysfs_ops = &kfd_procfs_ops,
};

void kfd_procfs_init(void)
{
	int ret = 0;

	procfs.kobj = kfd_alloc_struct(procfs.kobj);
	if (!procfs.kobj)
		return;

	ret = kobject_init_and_add(procfs.kobj, &procfs_type,
				   &kfd_device->kobj, "proc");
	if (ret) {
		pr_warn("Could not create procfs proc folder");
		/* If we fail to create the procfs, clean up */
		kfd_procfs_shutdown();
	}
}

void kfd_procfs_shutdown(void)
{
	if (procfs.kobj) {
		kobject_del(procfs.kobj);
		kobject_put(procfs.kobj);
		procfs.kobj = NULL;
	}
}

static ssize_t kfd_procfs_queue_show(struct kobject *kobj,
				     struct attribute *attr, char *buffer)
{
	struct queue *q = container_of(kobj, struct queue, kobj);

	if (!strcmp(attr->name, "size"))
		return snprintf(buffer, PAGE_SIZE, "%llu",
				q->properties.queue_size);
	else if (!strcmp(attr->name, "type"))
		return snprintf(buffer, PAGE_SIZE, "%d", q->properties.type);
	else if (!strcmp(attr->name, "gpuid"))
		return snprintf(buffer, PAGE_SIZE, "%u", q->device->id);
	else
		pr_err("Invalid attribute");

	return 0;
}

static ssize_t kfd_procfs_stats_show(struct kobject *kobj,
				     struct attribute *attr, char *buffer)
{
	if (strcmp(attr->name, "evicted_ms") == 0) {
		struct kfd_process_device *pdd = container_of(attr,
				struct kfd_process_device,
				attr_evict);
		uint64_t evict_jiffies;

		evict_jiffies = atomic64_read(&pdd->evict_duration_counter);

		return snprintf(buffer,
				PAGE_SIZE,
				"%llu\n",
				jiffies64_to_msecs(evict_jiffies));

	/* Sysfs handle that gets CU occupancy is per device */
	} else if (strcmp(attr->name, "cu_occupancy") == 0) {
		return kfd_get_cu_occupancy(attr, buffer);
	} else {
		pr_err("Invalid attribute");
	}

	return 0;
}

static ssize_t kfd_sysfs_counters_show(struct kobject *kobj,
				       struct attribute *attr, char *buf)
{
	struct kfd_process_device *pdd;

	if (!strcmp(attr->name, "faults")) {
		pdd = container_of(attr, struct kfd_process_device,
				   attr_faults);
		return sysfs_emit(buf, "%llu\n", READ_ONCE(pdd->faults));
	}
	if (!strcmp(attr->name, "page_in")) {
		pdd = container_of(attr, struct kfd_process_device,
				   attr_page_in);
		return sysfs_emit(buf, "%llu\n", READ_ONCE(pdd->page_in));
	}
	if (!strcmp(attr->name, "page_out")) {
		pdd = container_of(attr, struct kfd_process_device,
				   attr_page_out);
		return sysfs_emit(buf, "%llu\n", READ_ONCE(pdd->page_out));
	}
	return 0;
}

static struct attribute attr_queue_size = {
	.name = "size",
	.mode = KFD_SYSFS_FILE_MODE
};

static struct attribute attr_queue_type = {
	.name = "type",
	.mode = KFD_SYSFS_FILE_MODE
};

static struct attribute attr_queue_gpuid = {
	.name = "gpuid",
	.mode = KFD_SYSFS_FILE_MODE
};

static struct attribute *procfs_queue_attrs[] = {
	&attr_queue_size,
	&attr_queue_type,
	&attr_queue_gpuid,
	NULL
};

static const struct sysfs_ops procfs_queue_ops = {
	.show = kfd_procfs_queue_show,
};

static struct kobj_type procfs_queue_type = {
	.sysfs_ops = &procfs_queue_ops,
	.default_attrs = procfs_queue_attrs,
};

static const struct sysfs_ops procfs_stats_ops = {
	.show = kfd_procfs_stats_show,
};

static struct kobj_type procfs_stats_type = {
	.sysfs_ops = &procfs_stats_ops,
	.release = kfd_procfs_kobj_release,
};

static const struct sysfs_ops sysfs_counters_ops = {
	.show = kfd_sysfs_counters_show,
};

static struct kobj_type sysfs_counters_type = {
	.sysfs_ops = &sysfs_counters_ops,
	.release = kfd_procfs_kobj_release,
};

int kfd_procfs_add_queue(struct queue *q)
{
	struct kfd_process *proc;
	int ret;

	if (!q || !q->process)
		return -EINVAL;
	proc = q->process;

	/* Create proc/<pid>/queues/<queue id> folder */
	if (!proc->kobj_queues)
		return -EFAULT;
	ret = kobject_init_and_add(&q->kobj, &procfs_queue_type,
			proc->kobj_queues, "%u", q->properties.queue_id);
	if (ret < 0) {
		pr_warn("Creating proc/<pid>/queues/%u failed",
			q->properties.queue_id);
		kobject_put(&q->kobj);
		return ret;
	}

	return 0;
}

static void kfd_sysfs_create_file(struct kobject *kobj, struct attribute *attr,
				 char *name)
{
	int ret;

	if (!kobj || !attr || !name)
		return;

	attr->name = name;
	attr->mode = KFD_SYSFS_FILE_MODE;
	sysfs_attr_init(attr);

	ret = sysfs_create_file(kobj, attr);
	if (ret)
		pr_warn("Create sysfs %s/%s failed %d", kobj->name, name, ret);
}

static void kfd_procfs_add_sysfs_stats(struct kfd_process *p)
{
	int ret;
	int i;
	char stats_dir_filename[MAX_SYSFS_FILENAME_LEN];

	if (!p || !p->kobj)
		return;

	/*
	 * Create sysfs files for each GPU:
	 * - proc/<pid>/stats_<gpuid>/
	 * - proc/<pid>/stats_<gpuid>/evicted_ms
	 * - proc/<pid>/stats_<gpuid>/cu_occupancy
	 */
	for (i = 0; i < p->n_pdds; i++) {
		struct kfd_process_device *pdd = p->pdds[i];

		snprintf(stats_dir_filename, MAX_SYSFS_FILENAME_LEN,
				"stats_%u", pdd->dev->id);
		pdd->kobj_stats = kfd_alloc_struct(pdd->kobj_stats);
		if (!pdd->kobj_stats)
			return;

		ret = kobject_init_and_add(pdd->kobj_stats,
					   &procfs_stats_type,
					   p->kobj,
					   stats_dir_filename);

		if (ret) {
			pr_warn("Creating KFD proc/stats_%s folder failed",
				stats_dir_filename);
			kobject_put(pdd->kobj_stats);
			pdd->kobj_stats = NULL;
			return;
		}

		kfd_sysfs_create_file(pdd->kobj_stats, &pdd->attr_evict,
				      "evicted_ms");
		/* Add sysfs file to report compute unit occupancy */
		if (pdd->dev->kfd2kgd->get_cu_occupancy)
			kfd_sysfs_create_file(pdd->kobj_stats,
					      &pdd->attr_cu_occupancy,
					      "cu_occupancy");
	}
}

static void kfd_procfs_add_sysfs_counters(struct kfd_process *p)
{
	int ret = 0;
	int i;
	char counters_dir_filename[MAX_SYSFS_FILENAME_LEN];

	if (!p || !p->kobj)
		return;

	/*
	 * Create sysfs files for each GPU which supports SVM
	 * - proc/<pid>/counters_<gpuid>/
	 * - proc/<pid>/counters_<gpuid>/faults
	 * - proc/<pid>/counters_<gpuid>/page_in
	 * - proc/<pid>/counters_<gpuid>/page_out
	 */
	for_each_set_bit(i, p->svms.bitmap_supported, p->n_pdds) {
		struct kfd_process_device *pdd = p->pdds[i];
		struct kobject *kobj_counters;

		snprintf(counters_dir_filename, MAX_SYSFS_FILENAME_LEN,
			"counters_%u", pdd->dev->id);
		kobj_counters = kfd_alloc_struct(kobj_counters);
		if (!kobj_counters)
			return;

		ret = kobject_init_and_add(kobj_counters, &sysfs_counters_type,
					   p->kobj, counters_dir_filename);
		if (ret) {
			pr_warn("Creating KFD proc/%s folder failed",
				counters_dir_filename);
			kobject_put(kobj_counters);
			return;
		}

		pdd->kobj_counters = kobj_counters;
		kfd_sysfs_create_file(kobj_counters, &pdd->attr_faults,
				      "faults");
		kfd_sysfs_create_file(kobj_counters, &pdd->attr_page_in,
				      "page_in");
		kfd_sysfs_create_file(kobj_counters, &pdd->attr_page_out,
				      "page_out");
	}
}

static void kfd_procfs_add_sysfs_files(struct kfd_process *p)
{
	int i;

	if (!p || !p->kobj)
		return;

	/*
	 * Create sysfs files for each GPU:
	 * - proc/<pid>/vram_<gpuid>
	 * - proc/<pid>/sdma_<gpuid>
	 */
	for (i = 0; i < p->n_pdds; i++) {
		struct kfd_process_device *pdd = p->pdds[i];

		snprintf(pdd->vram_filename, MAX_SYSFS_FILENAME_LEN, "vram_%u",
			 pdd->dev->id);
		kfd_sysfs_create_file(p->kobj, &pdd->attr_vram,
				      pdd->vram_filename);

		snprintf(pdd->sdma_filename, MAX_SYSFS_FILENAME_LEN, "sdma_%u",
			 pdd->dev->id);
		kfd_sysfs_create_file(p->kobj, &pdd->attr_sdma,
					    pdd->sdma_filename);
	}
}

void kfd_procfs_del_queue(struct queue *q)
{
	if (!q)
		return;

	kobject_del(&q->kobj);
	kobject_put(&q->kobj);
}

int kfd_process_create_wq(void)
{
	if (!kfd_process_wq)
		kfd_process_wq = alloc_workqueue("kfd_process_wq", 0, 0);
	if (!kfd_restore_wq)
		kfd_restore_wq = alloc_ordered_workqueue("kfd_restore_wq", 0);

	if (!kfd_process_wq || !kfd_restore_wq) {
		kfd_process_destroy_wq();
		return -ENOMEM;
	}

	return 0;
}

void kfd_process_destroy_wq(void)
{
	if (kfd_process_wq) {
		destroy_workqueue(kfd_process_wq);
		kfd_process_wq = NULL;
	}
	if (kfd_restore_wq) {
		destroy_workqueue(kfd_restore_wq);
		kfd_restore_wq = NULL;
	}
}

static void kfd_process_free_gpuvm(struct kgd_mem *mem,
			struct kfd_process_device *pdd)
{
	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,
					       NULL);
}

/* kfd_process_alloc_gpuvm - Allocate GPU VM for the KFD process
 *	This function should be only called right after the process
 *	is created and when kfd_processes_mutex is still being held
 *	to avoid concurrency. Because of that exclusiveness, we do
 *	not need to take p->mutex.
 */
static int kfd_process_alloc_gpuvm(struct kfd_process_device *pdd,
				   uint64_t gpu_va, uint32_t size,
				   uint32_t flags, 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);
	if (err)
		goto err_alloc_mem;

	err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(kdev->kgd, mem, pdd->drm_priv);
	if (err)
		goto err_map_mem;

	err = amdgpu_amdkfd_gpuvm_sync_memory(kdev->kgd, 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);
		if (err) {
			pr_debug("Map GTT BO to kernel failed\n");
			goto free_obj_handle;
		}
	}

	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;

err_map_mem:
	amdgpu_amdkfd_gpuvm_free_memory_of_gpu(kdev->kgd, mem, pdd->drm_priv,
					       NULL);
err_alloc_mem:
	*kptr = NULL;
	return err;
}

/* kfd_process_device_reserve_ib_mem - Reserve memory inside the
 *	process for IB usage The memory reserved is for KFD to submit
 *	IB to AMDGPU from kernel.  If the memory is reserved
 *	successfully, ib_kaddr will have the CPU/kernel
 *	address. Check ib_kaddr before accessing the memory.
 */
static int kfd_process_device_reserve_ib_mem(struct kfd_process_device *pdd)
{
	struct qcm_process_device *qpd = &pdd->qpd;
	uint32_t flags = KFD_IOC_ALLOC_MEM_FLAGS_GTT |
			KFD_IOC_ALLOC_MEM_FLAGS_NO_SUBSTITUTE |
			KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE |
			KFD_IOC_ALLOC_MEM_FLAGS_EXECUTABLE;
	void *kaddr;
	int ret;

	if (qpd->ib_kaddr || !qpd->ib_base)
		return 0;

	/* ib_base is only set for dGPU */
	ret = kfd_process_alloc_gpuvm(pdd, qpd->ib_base, PAGE_SIZE, flags,
				      &kaddr);
	if (ret)
		return ret;

	qpd->ib_kaddr = kaddr;

	return 0;
}

struct kfd_process *kfd_create_process(struct file *filep)
{
	struct kfd_process *process;
	struct task_struct *thread = current;
	int ret;

	if (!thread->mm)
		return ERR_PTR(-EINVAL);

	/* Only the pthreads threading model is supported. */
	if (thread->group_leader->mm != thread->mm)
		return ERR_PTR(-EINVAL);

	/*
	 * take kfd processes mutex before starting of process creation
	 * so there won't be a case where two threads of the same process
	 * create two kfd_process structures
	 */
	mutex_lock(&kfd_processes_mutex);

	/* A prior open of /dev/kfd could have already created the process. */
	process = find_process(thread);
	if (process) {
		pr_debug("Process already found\n");
	} else {
		process = create_process(thread);
		if (IS_ERR(process))
			goto out;

		ret = kfd_process_init_cwsr_apu(process, filep);
		if (ret)
			goto out_destroy;

		if (!procfs.kobj)
			goto out;

		process->kobj = kfd_alloc_struct(process->kobj);
		if (!process->kobj) {
			pr_warn("Creating procfs kobject failed");
			goto out;
		}
		ret = kobject_init_and_add(process->kobj, &procfs_type,
					   procfs.kobj, "%d",
					   (int)process->lead_thread->pid);
		if (ret) {
			pr_warn("Creating procfs pid directory failed");
			kobject_put(process->kobj);
			goto out;
		}

		kfd_sysfs_create_file(process->kobj, &process->attr_pasid,
				      "pasid");

		process->kobj_queues = kobject_create_and_add("queues",
							process->kobj);
		if (!process->kobj_queues)
			pr_warn("Creating KFD proc/queues folder failed");

		kfd_procfs_add_sysfs_stats(process);
		kfd_procfs_add_sysfs_files(process);
		kfd_procfs_add_sysfs_counters(process);
	}
out:
	if (!IS_ERR(process))
		kref_get(&process->ref);
	mutex_unlock(&kfd_processes_mutex);

	return process;

out_destroy:
	hash_del_rcu(&process->kfd_processes);
	mutex_unlock(&kfd_processes_mutex);
	synchronize_srcu(&kfd_processes_srcu);
	/* kfd_process_free_notifier will trigger the cleanup */
	mmu_notifier_put(&process->mmu_notifier);
	return ERR_PTR(ret);
}

struct kfd_process *kfd_get_process(const struct task_struct *thread)
{
	struct kfd_process *process;

	if (!thread->mm)
		return ERR_PTR(-EINVAL);

	/* Only the pthreads threading model is supported. */
	if (thread->group_leader->mm != thread->mm)
		return ERR_PTR(-EINVAL);

	process = find_process(thread);
	if (!process)
		return ERR_PTR(-EINVAL);

	return process;
}

static struct kfd_process *find_process_by_mm(const struct mm_struct *mm)
{
	struct kfd_process *process;

	hash_for_each_possible_rcu(kfd_processes_table, process,
					kfd_processes, (uintptr_t)mm)
		if (process->mm == mm)
			return process;

	return NULL;
}

static struct kfd_process *find_process(const struct task_struct *thread)
{
	struct kfd_process *p;
	int idx;

	idx = srcu_read_lock(&kfd_processes_srcu);
	p = find_process_by_mm(thread->mm);
	srcu_read_unlock(&kfd_processes_srcu, idx);

	return p;
}

void kfd_unref_process(struct kfd_process *p)
{
	kref_put(&p->ref, kfd_process_ref_release);
}


static void kfd_process_device_free_bos(struct kfd_process_device *pdd)
{
	struct kfd_process *p = pdd->process;
	void *mem;
	int id;
	int i;

	/*
	 * Remove all handles from idr and release appropriate
	 * local memory object
	 */
	idr_for_each_entry(&pdd->alloc_idr, mem, id) {

		for (i = 0; i < p->n_pdds; i++) {
			struct kfd_process_device *peer_pdd = p->pdds[i];

			if (!peer_pdd->drm_priv)
				continue;
			amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
				peer_pdd->dev->kgd, mem, peer_pdd->drm_priv);
		}

		amdgpu_amdkfd_gpuvm_free_memory_of_gpu(pdd->dev->kgd, mem,
						       pdd->drm_priv, NULL);
		kfd_process_device_remove_obj_handle(pdd, id);
	}
}

static void kfd_process_free_outstanding_kfd_bos(struct kfd_process *p)
{
	int i;

	for (i = 0; i < p->n_pdds; i++)
		kfd_process_device_free_bos(p->pdds[i]);
}

static void kfd_process_destroy_pdds(struct kfd_process *p)
{
	int i;

	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);

		if (pdd->drm_file) {
			amdgpu_amdkfd_gpuvm_release_process_vm(
					pdd->dev->kgd, pdd->drm_priv);
			fput(pdd->drm_file);
		}

		if (pdd->qpd.cwsr_kaddr && !pdd->qpd.cwsr_base)
			free_pages((unsigned long)pdd->qpd.cwsr_kaddr,
				get_order(KFD_CWSR_TBA_TMA_SIZE));

		kfree(pdd->qpd.doorbell_bitmap);
		idr_destroy(&pdd->alloc_idr);

		kfd_free_process_doorbells(pdd->dev, pdd->doorbell_index);

		/*
		 * before destroying pdd, make sure to report availability
		 * for auto suspend
		 */
		if (pdd->runtime_inuse) {
			pm_runtime_mark_last_busy(pdd->dev->ddev->dev);
			pm_runtime_put_autosuspend(pdd->dev->ddev->dev);
			pdd->runtime_inuse = false;
		}

		kfree(pdd);
		p->pdds[i] = NULL;
	}
	p->n_pdds = 0;
}

static void kfd_process_remove_sysfs(struct kfd_process *p)
{
	struct kfd_process_device *pdd;