[deliverable/linux.git] / drivers / gpu / drm / amd / amdkfd / 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/slab.h>
#include <linux/amd-iommu.h>
#include <linux/notifier.h>
#include <linux/compat.h>

struct mm_struct;

#include "kfd_priv.h"

/*
 * Initial size for the array of queues.
 * The allocated size is doubled each time
 * it is exceeded up to MAX_PROCESS_QUEUES.
 */
#define INITIAL_QUEUE_ARRAY_SIZE 16

/*
 * List of struct kfd_process (field kfd_process).
 * Unique/indexed by mm_struct*
 */
#define KFD_PROCESS_TABLE_SIZE 5 /* bits: 32 entries */
static DEFINE_HASHTABLE(kfd_processes_table, KFD_PROCESS_TABLE_SIZE);
static DEFINE_MUTEX(kfd_processes_mutex);

DEFINE_STATIC_SRCU(kfd_processes_srcu);

static struct workqueue_struct *kfd_process_wq;

struct kfd_process_release_work {
	struct work_struct kfd_work;
	struct kfd_process *p;
};

static struct kfd_process *find_process(const struct task_struct *thread);
static struct kfd_process *create_process(const struct task_struct *thread);

void kfd_process_create_wq(void)
{
	if (!kfd_process_wq)
		kfd_process_wq = create_workqueue("kfd_process_wq");
}

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

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

	BUG_ON(!kfd_process_wq);

	if (thread->mm == NULL)
		return ERR_PTR(-EINVAL);

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

	/* Take mmap_sem because we call __mmu_notifier_register inside */
	down_write(&thread->mm->mmap_sem);

	/*
	 * 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("kfd: process already found\n");

	if (!process)
		process = create_process(thread);

	mutex_unlock(&kfd_processes_mutex);

	up_write(&thread->mm->mmap_sem);

	return process;
}

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

	if (thread->mm == NULL)
		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);

	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;
}

static void kfd_process_wq_release(struct work_struct *work)
{
	struct kfd_process_release_work *my_work;
	struct kfd_process_device *pdd, *temp;
	struct kfd_process *p;

	my_work = (struct kfd_process_release_work *) work;

	p = my_work->p;

	pr_debug("Releasing process (pasid %d) in workqueue\n",
			p->pasid);

	mutex_lock(&p->mutex);

	list_for_each_entry_safe(pdd, temp, &p->per_device_data,
							per_device_list) {
		pr_debug("Releasing pdd (topology id %d) for process (pasid %d) in workqueue\n",
				pdd->dev->id, p->pasid);

		amd_iommu_unbind_pasid(pdd->dev->pdev, p->pasid);
		list_del(&pdd->per_device_list);

		kfree(pdd);
	}

	kfd_event_free_process(p);

	kfd_pasid_free(p->pasid);

	mutex_unlock(&p->mutex);

	mutex_destroy(&p->mutex);

	kfree(p->queues);

	kfree(p);

	kfree((void *)work);
}

static void kfd_process_destroy_delayed(struct rcu_head *rcu)
{
	struct kfd_process_release_work *work;
	struct kfd_process *p;

	BUG_ON(!kfd_process_wq);

	p = container_of(rcu, struct kfd_process, rcu);
	BUG_ON(atomic_read(&p->mm->mm_count) <= 0);

	mmdrop(p->mm);

	work = kmalloc(sizeof(struct kfd_process_release_work), GFP_ATOMIC);

	if (work) {
		INIT_WORK((struct work_struct *) work, kfd_process_wq_release);
		work->p = p;
		queue_work(kfd_process_wq, (struct work_struct *) work);
	}
}

static void kfd_process_notifier_release(struct mmu_notifier *mn,
					struct mm_struct *mm)
{
	struct kfd_process *p;

	/*
	 * The kfd_process structure can not be free because the
	 * mmu_notifier srcu is read locked
	 */
	p = container_of(mn, struct kfd_process, mmu_notifier);
	BUG_ON(p->mm != mm);

	mutex_lock(&kfd_processes_mutex);
	hash_del_rcu(&p->kfd_processes);
	mutex_unlock(&kfd_processes_mutex);
	synchronize_srcu(&kfd_processes_srcu);

	mutex_lock(&p->mutex);

	/* In case our notifier is called before IOMMU notifier */
	pqm_uninit(&p->pqm);

	mutex_unlock(&p->mutex);

	/*
	 * Because we drop mm_count inside kfd_process_destroy_delayed
	 * and because the mmu_notifier_unregister function also drop
	 * mm_count we need to take an extra count here.
	 */
	atomic_inc(&p->mm->mm_count);
	mmu_notifier_unregister_no_release(&p->mmu_notifier, p->mm);
	mmu_notifier_call_srcu(&p->rcu, &kfd_process_destroy_delayed);
}

static const struct mmu_notifier_ops kfd_process_mmu_notifier_ops = {
	.release = kfd_process_notifier_release,
};

static struct kfd_process *create_process(const struct task_struct *thread)
{
	struct kfd_process *process;
	int err = -ENOMEM;

	process = kzalloc(sizeof(*process), GFP_KERNEL);

	if (!process)
		goto err_alloc_process;

	process->queues = kmalloc_array(INITIAL_QUEUE_ARRAY_SIZE,
					sizeof(process->queues[0]), GFP_KERNEL);
	if (!process->queues)
		goto err_alloc_queues;

	process->pasid = kfd_pasid_alloc();
	if (process->pasid == 0)
		goto err_alloc_pasid;

	mutex_init(&process->mutex);

	process->mm = thread->mm;

	/* register notifier */
	process->mmu_notifier.ops = &kfd_process_mmu_notifier_ops;
	err = __mmu_notifier_register(&process->mmu_notifier, process->mm);
	if (err)
		goto err_mmu_notifier;

	hash_add_rcu(kfd_processes_table, &process->kfd_processes,
			(uintptr_t)process->mm);

	process->lead_thread = thread->group_leader;

	process->queue_array_size = INITIAL_QUEUE_ARRAY_SIZE;

	INIT_LIST_HEAD(&process->per_device_data);

	kfd_event_init_process(process);

	err = pqm_init(&process->pqm, process);
	if (err != 0)
		goto err_process_pqm_init;

	/* init process apertures*/
	process->is_32bit_user_mode = is_compat_task();
	if (kfd_init_apertures(process) != 0)
		goto err_init_apretures;

	return process;

err_init_apretures:
	pqm_uninit(&process->pqm);
err_process_pqm_init:
	hash_del_rcu(&process->kfd_processes);
	synchronize_rcu();
	mmu_notifier_unregister_no_release(&process->mmu_notifier, process->mm);
err_mmu_notifier:
	kfd_pasid_free(process->pasid);
err_alloc_pasid:
	kfree(process->queues);
err_alloc_queues:
	kfree(process);
err_alloc_process:
	return ERR_PTR(err);
}

struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
							struct kfd_process *p)
{
	struct kfd_process_device *pdd = NULL;

	list_for_each_entry(pdd, &p->per_device_data, per_device_list)
		if (pdd->dev == dev)
			break;

	return pdd;
}

struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
							struct kfd_process *p)
{
	struct kfd_process_device *pdd = NULL;

	pdd = kzalloc(sizeof(*pdd), GFP_KERNEL);
	if (pdd != NULL) {
		pdd->dev = dev;
		INIT_LIST_HEAD(&pdd->qpd.queues_list);
		INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
		pdd->qpd.dqm = dev->dqm;
		list_add(&pdd->per_device_list, &p->per_device_data);
	}

	return pdd;
}

/*
 * Direct the IOMMU to bind the process (specifically the pasid->mm)
 * to the device.
 * Unbinding occurs when the process dies or the device is removed.
 *
 * Assumes that the process lock is held.
 */
struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev,
							struct kfd_process *p)
{
	struct kfd_process_device *pdd;
	int err;

	pdd = kfd_get_process_device_data(dev, p);
	if (!pdd) {
		pr_err("Process device data doesn't exist\n");
		return ERR_PTR(-ENOMEM);
	}

	if (pdd->bound)
		return pdd;

	err = amd_iommu_bind_pasid(dev->pdev, p->pasid, p->lead_thread);
	if (err < 0)
		return ERR_PTR(err);

	pdd->bound = true;

	return pdd;
}

void kfd_unbind_process_from_device(struct kfd_dev *dev, unsigned int pasid)
{
	struct kfd_process *p;
	struct kfd_process_device *pdd;
	int idx, i;

	BUG_ON(dev == NULL);

	idx = srcu_read_lock(&kfd_processes_srcu);

	hash_for_each_rcu(kfd_processes_table, i, p, kfd_processes)
		if (p->pasid == pasid)
			break;

	srcu_read_unlock(&kfd_processes_srcu, idx);

	BUG_ON(p->pasid != pasid);

	mutex_lock(&p->mutex);

	pqm_uninit(&p->pqm);

	pdd = kfd_get_process_device_data(dev, p);

	/*
	 * Just mark pdd as unbound, because we still need it to call
	 * amd_iommu_unbind_pasid() in when the process exits.
	 * We don't call amd_iommu_unbind_pasid() here
	 * because the IOMMU called us.
	 */
	if (pdd)
		pdd->bound = false;

	mutex_unlock(&p->mutex);
}

struct kfd_process_device *kfd_get_first_process_device_data(struct kfd_process *p)
{
	return list_first_entry(&p->per_device_data,
				struct kfd_process_device,
				per_device_list);
}

struct kfd_process_device *kfd_get_next_process_device_data(struct kfd_process *p,
						struct kfd_process_device *pdd)
{
	if (list_is_last(&pdd->per_device_list, &p->per_device_data))
		return NULL;
	return list_next_entry(pdd, per_device_list);
}

bool kfd_has_process_device_data(struct kfd_process *p)
{
	return !(list_empty(&p->per_device_data));
}

/* This returns with process->mutex locked. */
struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid)
{
	struct kfd_process *p;
	unsigned int temp;

	int idx = srcu_read_lock(&kfd_processes_srcu);

	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
		if (p->pasid == pasid) {
			mutex_lock(&p->mutex);
			break;
		}
	}

	srcu_read_unlock(&kfd_processes_srcu, idx);

	return p;
}
Commit	Line	Data
19f6d2a6 OG	1	/*
	2	* Copyright 2014 Advanced Micro Devices, Inc.
	3	*
	4	* Permission is hereby granted, free of charge, to any person obtaining a
	5	* copy of this software and associated documentation files (the "Software"),
	6	* to deal in the Software without restriction, including without limitation
	7	* the rights to use, copy, modify, merge, publish, distribute, sublicense,
	8	* and/or sell copies of the Software, and to permit persons to whom the
	9	* Software is furnished to do so, subject to the following conditions:
	10	*
	11	* The above copyright notice and this permission notice shall be included in
	12	* all copies or substantial portions of the Software.
	13	*
	14	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
	15	* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
	16	* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
	17	* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
	18	* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
	19	* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
	20	* OTHER DEALINGS IN THE SOFTWARE.
	21	*/
	22
	23	#include <linux/mutex.h>
	24	#include <linux/log2.h>
	25	#include <linux/sched.h>
	26	#include <linux/slab.h>
b17f068a	27	#include <linux/amd-iommu.h>
19f6d2a6	28	#include <linux/notifier.h>
dd59239a AS	29	#include <linux/compat.h>
dd59239a AS	30
19f6d2a6 OG	31	struct mm_struct;
	32
	33	#include "kfd_priv.h"
	34
	35	/*
	36	* Initial size for the array of queues.
	37	* The allocated size is doubled each time
	38	* it is exceeded up to MAX_PROCESS_QUEUES.
	39	*/
	40	#define INITIAL_QUEUE_ARRAY_SIZE 16
	41
	42	/*
	43	* List of struct kfd_process (field kfd_process).
	44	* Unique/indexed by mm_struct*
	45	*/
	46	#define KFD_PROCESS_TABLE_SIZE 5 /* bits: 32 entries */
	47	static DEFINE_HASHTABLE(kfd_processes_table, KFD_PROCESS_TABLE_SIZE);
	48	static DEFINE_MUTEX(kfd_processes_mutex);
	49
	50	DEFINE_STATIC_SRCU(kfd_processes_srcu);
	51
	52	static struct workqueue_struct *kfd_process_wq;
	53
	54	struct kfd_process_release_work {
	55	struct work_struct kfd_work;
	56	struct kfd_process *p;
	57	};
	58
	59	static struct kfd_process find_process(const struct task_struct thread);
	60	static struct kfd_process create_process(const struct task_struct thread);
	61
	62	void kfd_process_create_wq(void)
	63	{
	64	if (!kfd_process_wq)
	65	kfd_process_wq = create_workqueue("kfd_process_wq");
	66	}
	67
	68	void kfd_process_destroy_wq(void)
	69	{
	70	if (kfd_process_wq) {
	71	flush_workqueue(kfd_process_wq);
	72	destroy_workqueue(kfd_process_wq);
	73	kfd_process_wq = NULL;
	74	}
	75	}
	76
	77	struct kfd_process kfd_create_process(const struct task_struct thread)
	78	{
	79	struct kfd_process *process;
	80
	81	BUG_ON(!kfd_process_wq);
	82
	83	if (thread->mm == NULL)
	84	return ERR_PTR(-EINVAL);
	85
	86	/* Only the pthreads threading model is supported. */
	87	if (thread->group_leader->mm != thread->mm)
	88	return ERR_PTR(-EINVAL);
	89
	90	/* Take mmap_sem because we call __mmu_notifier_register inside */
	91	down_write(&thread->mm->mmap_sem);
	92
	93	/*
	94	* take kfd processes mutex before starting of process creation
95	* so there won't be a case where two threads of the same process
96	* create two kfd_process structures
97	*/
98	mutex_lock(&kfd_processes_mutex);
99
100	/* A prior open of /dev/kfd could have already created the process. */
101	process = find_process(thread);
102	if (process)
103	pr_debug("kfd: process already found\n");
104
105	if (!process)
106	process = create_process(thread);
107
108	mutex_unlock(&kfd_processes_mutex);
109
110	up_write(&thread->mm->mmap_sem);
111
112	return process;
113	}
114
115	struct kfd_process kfd_get_process(const struct task_struct thread)
116	{
117	struct kfd_process *process;
118
119	if (thread->mm == NULL)
120	return ERR_PTR(-EINVAL);
121
122	/* Only the pthreads threading model is supported. */
123	if (thread->group_leader->mm != thread->mm)
124	return ERR_PTR(-EINVAL);
125
126	process = find_process(thread);
127
128	return process;
129	}
130
131	static struct kfd_process find_process_by_mm(const struct mm_struct mm)
132	{
133	struct kfd_process *process;
134
135	hash_for_each_possible_rcu(kfd_processes_table, process,
136	kfd_processes, (uintptr_t)mm)
137	if (process->mm == mm)
138	return process;
139
140	return NULL;
141	}
142
143	static struct kfd_process find_process(const struct task_struct thread)
144	{
145	struct kfd_process *p;
146	int idx;
147
148	idx = srcu_read_lock(&kfd_processes_srcu);
149	p = find_process_by_mm(thread->mm);
150	srcu_read_unlock(&kfd_processes_srcu, idx);
151
152	return p;
153	}
154
155	static void kfd_process_wq_release(struct work_struct *work)
156	{
157	struct kfd_process_release_work *my_work;
158	struct kfd_process_device pdd, temp;
159	struct kfd_process *p;
160
161	my_work = (struct kfd_process_release_work *) work;
162
163	p = my_work->p;
164
94a1ee09 OG	165	pr_debug("Releasing process (pasid %d) in workqueue\n",
	166	p->pasid);
	167
19f6d2a6 OG	168	mutex_lock(&p->mutex);
	169
	170	list_for_each_entry_safe(pdd, temp, &p->per_device_data,
	171	per_device_list) {
94a1ee09 OG	172	pr_debug("Releasing pdd (topology id %d) for process (pasid %d) in workqueue\n",
	173	pdd->dev->id, p->pasid);
	174
b17f068a	175	amd_iommu_unbind_pasid(pdd->dev->pdev, p->pasid);
19f6d2a6 OG	176	list_del(&pdd->per_device_list);
	177
	178	kfree(pdd);
	179	}
	180
f3a39818 AL	181	kfd_event_free_process(p);
f3a39818 AL	182
19f6d2a6 OG	183	kfd_pasid_free(p->pasid);
	184
	185	mutex_unlock(&p->mutex);
	186
	187	mutex_destroy(&p->mutex);
	188
	189	kfree(p->queues);
	190
	191	kfree(p);
	192
	193	kfree((void *)work);
	194	}
	195
	196	static void kfd_process_destroy_delayed(struct rcu_head *rcu)
	197	{
	198	struct kfd_process_release_work *work;
	199	struct kfd_process *p;
	200
	201	BUG_ON(!kfd_process_wq);
	202
	203	p = container_of(rcu, struct kfd_process, rcu);
	204	BUG_ON(atomic_read(&p->mm->mm_count) <= 0);
	205
	206	mmdrop(p->mm);
	207
1549fcd1	208	work = kmalloc(sizeof(struct kfd_process_release_work), GFP_ATOMIC);
19f6d2a6 OG	209
	210	if (work) {
	211	INIT_WORK((struct work_struct *) work, kfd_process_wq_release);
	212	work->p = p;
	213	queue_work(kfd_process_wq, (struct work_struct *) work);
	214	}
	215	}
	216
	217	static void kfd_process_notifier_release(struct mmu_notifier *mn,
	218	struct mm_struct *mm)
	219	{
	220	struct kfd_process *p;
	221
	222	/*
	223	* The kfd_process structure can not be free because the
	224	* mmu_notifier srcu is read locked
	225	*/
	226	p = container_of(mn, struct kfd_process, mmu_notifier);
	227	BUG_ON(p->mm != mm);
	228
	229	mutex_lock(&kfd_processes_mutex);
	230	hash_del_rcu(&p->kfd_processes);
	231	mutex_unlock(&kfd_processes_mutex);
	232	synchronize_srcu(&kfd_processes_srcu);
	233
45102048 BG	234	mutex_lock(&p->mutex);
	235
	236	/* In case our notifier is called before IOMMU notifier */
	237	pqm_uninit(&p->pqm);
	238
	239	mutex_unlock(&p->mutex);
	240
19f6d2a6 OG	241	/*
	242	* Because we drop mm_count inside kfd_process_destroy_delayed
	243	* and because the mmu_notifier_unregister function also drop
	244	* mm_count we need to take an extra count here.
	245	*/
	246	atomic_inc(&p->mm->mm_count);
	247	mmu_notifier_unregister_no_release(&p->mmu_notifier, p->mm);
	248	mmu_notifier_call_srcu(&p->rcu, &kfd_process_destroy_delayed);
	249	}
	250
	251	static const struct mmu_notifier_ops kfd_process_mmu_notifier_ops = {
	252	.release = kfd_process_notifier_release,
	253	};
	254
	255	static struct kfd_process create_process(const struct task_struct thread)
	256	{
	257	struct kfd_process *process;
	258	int err = -ENOMEM;
	259
	260	process = kzalloc(sizeof(*process), GFP_KERNEL);
	261
	262	if (!process)
	263	goto err_alloc_process;
	264
	265	process->queues = kmalloc_array(INITIAL_QUEUE_ARRAY_SIZE,
	266	sizeof(process->queues[0]), GFP_KERNEL);
	267	if (!process->queues)
	268	goto err_alloc_queues;
	269
	270	process->pasid = kfd_pasid_alloc();
	271	if (process->pasid == 0)
	272	goto err_alloc_pasid;
	273
	274	mutex_init(&process->mutex);
	275
	276	process->mm = thread->mm;
	277
	278	/* register notifier */
	279	process->mmu_notifier.ops = &kfd_process_mmu_notifier_ops;
	280	err = __mmu_notifier_register(&process->mmu_notifier, process->mm);
	281	if (err)
	282	goto err_mmu_notifier;
	283
	284	hash_add_rcu(kfd_processes_table, &process->kfd_processes,
	285	(uintptr_t)process->mm);
	286
	287	process->lead_thread = thread->group_leader;
	288
	289	process->queue_array_size = INITIAL_QUEUE_ARRAY_SIZE;
	290
	291	INIT_LIST_HEAD(&process->per_device_data);
	292
f3a39818 AL	293	kfd_event_init_process(process);
f3a39818 AL	294
45102048 BG	295	err = pqm_init(&process->pqm, process);
	296	if (err != 0)
	297	goto err_process_pqm_init;
	298
dd59239a AS	299	/* init process apertures*/
	300	process->is_32bit_user_mode = is_compat_task();
	301	if (kfd_init_apertures(process) != 0)
	302	goto err_init_apretures;
	303
19f6d2a6 OG	304	return process;
19f6d2a6 OG	305
dd59239a AS	306	err_init_apretures:
dd59239a AS	307	pqm_uninit(&process->pqm);
45102048 BG	308	err_process_pqm_init:
	309	hash_del_rcu(&process->kfd_processes);
	310	synchronize_rcu();
	311	mmu_notifier_unregister_no_release(&process->mmu_notifier, process->mm);
19f6d2a6 OG	312	err_mmu_notifier:
	313	kfd_pasid_free(process->pasid);
	314	err_alloc_pasid:
	315	kfree(process->queues);
	316	err_alloc_queues:
	317	kfree(process);
	318	err_alloc_process:
	319	return ERR_PTR(err);
	320	}
	321
	322	struct kfd_process_device kfd_get_process_device_data(struct kfd_dev dev,
093c7d8c	323	struct kfd_process *p)
19f6d2a6 OG	324	{
	325	struct kfd_process_device *pdd = NULL;
	326
	327	list_for_each_entry(pdd, &p->per_device_data, per_device_list)
	328	if (pdd->dev == dev)
093c7d8c AS	329	break;
	330
	331	return pdd;
	332	}
	333
	334	struct kfd_process_device kfd_create_process_device_data(struct kfd_dev dev,
	335	struct kfd_process *p)
	336	{
	337	struct kfd_process_device *pdd = NULL;
	338
	339	pdd = kzalloc(sizeof(*pdd), GFP_KERNEL);
	340	if (pdd != NULL) {
	341	pdd->dev = dev;
	342	INIT_LIST_HEAD(&pdd->qpd.queues_list);
	343	INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
	344	pdd->qpd.dqm = dev->dqm;
	345	list_add(&pdd->per_device_list, &p->per_device_data);
19f6d2a6 OG	346	}
	347
	348	return pdd;
	349	}
	350
	351	/*
	352	* Direct the IOMMU to bind the process (specifically the pasid->mm)
	353	* to the device.
	354	* Unbinding occurs when the process dies or the device is removed.
	355	*
	356	* Assumes that the process lock is held.
	357	*/
	358	struct kfd_process_device kfd_bind_process_to_device(struct kfd_dev dev,
	359	struct kfd_process *p)
	360	{
093c7d8c	361	struct kfd_process_device *pdd;
b17f068a	362	int err;
19f6d2a6	363
093c7d8c AS	364	pdd = kfd_get_process_device_data(dev, p);
	365	if (!pdd) {
	366	pr_err("Process device data doesn't exist\n");
19f6d2a6	367	return ERR_PTR(-ENOMEM);
093c7d8c	368	}
19f6d2a6 OG	369
	370	if (pdd->bound)
	371	return pdd;
	372
b17f068a OG	373	err = amd_iommu_bind_pasid(dev->pdev, p->pasid, p->lead_thread);
	374	if (err < 0)
	375	return ERR_PTR(err);
	376
19f6d2a6 OG	377	pdd->bound = true;
	378
	379	return pdd;
	380	}
	381
	382	void kfd_unbind_process_from_device(struct kfd_dev *dev, unsigned int pasid)
	383	{
	384	struct kfd_process *p;
	385	struct kfd_process_device *pdd;
	386	int idx, i;
	387
	388	BUG_ON(dev == NULL);
	389
	390	idx = srcu_read_lock(&kfd_processes_srcu);
	391
	392	hash_for_each_rcu(kfd_processes_table, i, p, kfd_processes)
	393	if (p->pasid == pasid)
	394	break;
	395
	396	srcu_read_unlock(&kfd_processes_srcu, idx);
	397
	398	BUG_ON(p->pasid != pasid);
	399
	400	mutex_lock(&p->mutex);
	401
45102048 BG	402	pqm_uninit(&p->pqm);
45102048 BG	403
093c7d8c	404	pdd = kfd_get_process_device_data(dev, p);
19f6d2a6 OG	405
	406	/*
	407	* Just mark pdd as unbound, because we still need it to call
	408	* amd_iommu_unbind_pasid() in when the process exits.
	409	* We don't call amd_iommu_unbind_pasid() here
	410	* because the IOMMU called us.
	411	*/
	412	if (pdd)
	413	pdd->bound = false;
	414
	415	mutex_unlock(&p->mutex);
	416	}
	417
	418	struct kfd_process_device kfd_get_first_process_device_data(struct kfd_process p)
	419	{
	420	return list_first_entry(&p->per_device_data,
	421	struct kfd_process_device,
	422	per_device_list);
	423	}
	424
	425	struct kfd_process_device kfd_get_next_process_device_data(struct kfd_process p,
	426	struct kfd_process_device *pdd)
	427	{
	428	if (list_is_last(&pdd->per_device_list, &p->per_device_data))
	429	return NULL;
	430	return list_next_entry(pdd, per_device_list);
	431	}
	432
	433	bool kfd_has_process_device_data(struct kfd_process *p)
	434	{
	435	return !(list_empty(&p->per_device_data));
	436	}
f3a39818 AL	437
	438	/* This returns with process->mutex locked. */
	439	struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid)
	440	{
	441	struct kfd_process *p;
	442	unsigned int temp;
	443
	444	int idx = srcu_read_lock(&kfd_processes_srcu);
	445
	446	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
	447	if (p->pasid == pasid) {
	448	mutex_lock(&p->mutex);
	449	break;
	450	}
	451	}
	452
	453	srcu_read_unlock(&kfd_processes_srcu, idx);
	454
	455	return p;
	456	}