[mirror_ubuntu-artful-kernel.git] / drivers / gpu / drm / msm / msm_gem_submit.c

/*
 * Copyright (C) 2013 Red Hat
 * Author: Rob Clark <robdclark@gmail.com>
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 as published by
 * the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 *
 * You should have received a copy of the GNU General Public License along with
 * this program.  If not, see <http://www.gnu.org/licenses/>.
 */

#include <linux/sync_file.h>

#include "msm_drv.h"
#include "msm_gpu.h"
#include "msm_gem.h"

/*
 * Cmdstream submission:
 */

/* make sure these don't conflict w/ MSM_SUBMIT_BO_x */
#define BO_VALID    0x8000   /* is current addr in cmdstream correct/valid? */
#define BO_LOCKED   0x4000
#define BO_PINNED   0x2000

static struct msm_gem_submit *submit_create(struct drm_device *dev,
		struct msm_gpu *gpu, int nr_bos, int nr_cmds)
{
	struct msm_gem_submit *submit;
	int sz = sizeof(*submit) + (nr_bos * sizeof(submit->bos[0])) +
			(nr_cmds * sizeof(*submit->cmd));

	submit = kmalloc(sz, GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY);
	if (!submit)
		return NULL;

	submit->dev = dev;
	submit->gpu = gpu;
	submit->fence = NULL;
	submit->pid = get_pid(task_pid(current));
	submit->cmd = (void *)&submit->bos[nr_bos];

	/* initially, until copy_from_user() and bo lookup succeeds: */
	submit->nr_bos = 0;
	submit->nr_cmds = 0;

	INIT_LIST_HEAD(&submit->node);
	INIT_LIST_HEAD(&submit->bo_list);
	ww_acquire_init(&submit->ticket, &reservation_ww_class);

	return submit;
}

void msm_gem_submit_free(struct msm_gem_submit *submit)
{
	dma_fence_put(submit->fence);
	list_del(&submit->node);
	put_pid(submit->pid);
	kfree(submit);
}

static inline unsigned long __must_check
copy_from_user_inatomic(void *to, const void __user *from, unsigned long n)
{
	if (access_ok(VERIFY_READ, from, n))
		return __copy_from_user_inatomic(to, from, n);
	return -EFAULT;
}

static int submit_lookup_objects(struct msm_gem_submit *submit,
		struct drm_msm_gem_submit *args, struct drm_file *file)
{
	unsigned i;
	int ret = 0;

	spin_lock(&file->table_lock);
	pagefault_disable();

	for (i = 0; i < args->nr_bos; i++) {
		struct drm_msm_gem_submit_bo submit_bo;
		struct drm_gem_object *obj;
		struct msm_gem_object *msm_obj;
		void __user *userptr =
			u64_to_user_ptr(args->bos + (i * sizeof(submit_bo)));

		/* make sure we don't have garbage flags, in case we hit
		 * error path before flags is initialized:
		 */
		submit->bos[i].flags = 0;

		ret = copy_from_user_inatomic(&submit_bo, userptr, sizeof(submit_bo));
		if (unlikely(ret)) {
			pagefault_enable();
			spin_unlock(&file->table_lock);
			ret = copy_from_user(&submit_bo, userptr, sizeof(submit_bo));
			if (ret)
				goto out;
			spin_lock(&file->table_lock);
			pagefault_disable();
		}

		if ((submit_bo.flags & ~MSM_SUBMIT_BO_FLAGS) ||
			!(submit_bo.flags & MSM_SUBMIT_BO_FLAGS)) {
			DRM_ERROR("invalid flags: %x\n", submit_bo.flags);
			ret = -EINVAL;
			goto out_unlock;
		}

		submit->bos[i].flags = submit_bo.flags;
		/* in validate_objects() we figure out if this is true: */
		submit->bos[i].iova  = submit_bo.presumed;

		/* normally use drm_gem_object_lookup(), but for bulk lookup
		 * all under single table_lock just hit object_idr directly:
		 */
		obj = idr_find(&file->object_idr, submit_bo.handle);
		if (!obj) {
			DRM_ERROR("invalid handle %u at index %u\n", submit_bo.handle, i);
			ret = -EINVAL;
			goto out_unlock;
		}

		msm_obj = to_msm_bo(obj);

		if (!list_empty(&msm_obj->submit_entry)) {
			DRM_ERROR("handle %u at index %u already on submit list\n",
					submit_bo.handle, i);
			ret = -EINVAL;
			goto out_unlock;
		}

		drm_gem_object_reference(obj);

		submit->bos[i].obj = msm_obj;

		list_add_tail(&msm_obj->submit_entry, &submit->bo_list);
	}

out_unlock:
	pagefault_enable();
	spin_unlock(&file->table_lock);

out:
	submit->nr_bos = i;

	return ret;
}

static void submit_unlock_unpin_bo(struct msm_gem_submit *submit, int i)
{
	struct msm_gem_object *msm_obj = submit->bos[i].obj;

	if (submit->bos[i].flags & BO_PINNED)
		msm_gem_put_iova(&msm_obj->base, submit->gpu->id);

	if (submit->bos[i].flags & BO_LOCKED)
		ww_mutex_unlock(&msm_obj->resv->lock);

	if (!(submit->bos[i].flags & BO_VALID))
		submit->bos[i].iova = 0;

	submit->bos[i].flags &= ~(BO_LOCKED | BO_PINNED);
}

/* This is where we make sure all the bo's are reserved and pin'd: */
static int submit_lock_objects(struct msm_gem_submit *submit)
{
	int contended, slow_locked = -1, i, ret = 0;

retry:
	for (i = 0; i < submit->nr_bos; i++) {
		struct msm_gem_object *msm_obj = submit->bos[i].obj;

		if (slow_locked == i)
			slow_locked = -1;

		contended = i;

		if (!(submit->bos[i].flags & BO_LOCKED)) {
			ret = ww_mutex_lock_interruptible(&msm_obj->resv->lock,
					&submit->ticket);
			if (ret)
				goto fail;
			submit->bos[i].flags |= BO_LOCKED;
		}
	}

	ww_acquire_done(&submit->ticket);

	return 0;

fail:
	for (; i >= 0; i--)
		submit_unlock_unpin_bo(submit, i);

	if (slow_locked > 0)
		submit_unlock_unpin_bo(submit, slow_locked);

	if (ret == -EDEADLK) {
		struct msm_gem_object *msm_obj = submit->bos[contended].obj;
		/* we lost out in a seqno race, lock and retry.. */
		ret = ww_mutex_lock_slow_interruptible(&msm_obj->resv->lock,
				&submit->ticket);
		if (!ret) {
			submit->bos[contended].flags |= BO_LOCKED;
			slow_locked = contended;
			goto retry;
		}
	}

	return ret;
}

static int submit_fence_sync(struct msm_gem_submit *submit)
{
	int i, ret = 0;

	for (i = 0; i < submit->nr_bos; i++) {
		struct msm_gem_object *msm_obj = submit->bos[i].obj;
		bool write = submit->bos[i].flags & MSM_SUBMIT_BO_WRITE;

		ret = msm_gem_sync_object(&msm_obj->base, submit->gpu->fctx, write);
		if (ret)
			break;
	}

	return ret;
}

static int submit_pin_objects(struct msm_gem_submit *submit)
{
	int i, ret = 0;

	submit->valid = true;

	for (i = 0; i < submit->nr_bos; i++) {
		struct msm_gem_object *msm_obj = submit->bos[i].obj;
		uint64_t iova;

		/* if locking succeeded, pin bo: */
		ret = msm_gem_get_iova_locked(&msm_obj->base,
				submit->gpu->id, &iova);

		if (ret)
			break;

		submit->bos[i].flags |= BO_PINNED;

		if (iova == submit->bos[i].iova) {
			submit->bos[i].flags |= BO_VALID;
		} else {
			submit->bos[i].iova = iova;
			/* iova changed, so address in cmdstream is not valid: */
			submit->bos[i].flags &= ~BO_VALID;
			submit->valid = false;
		}
	}

	return ret;
}

static int submit_bo(struct msm_gem_submit *submit, uint32_t idx,
		struct msm_gem_object **obj, uint64_t *iova, bool *valid)
{
	if (idx >= submit->nr_bos) {
		DRM_ERROR("invalid buffer index: %u (out of %u)\n",
				idx, submit->nr_bos);
		return -EINVAL;
	}

	if (obj)
		*obj = submit->bos[idx].obj;
	if (iova)
		*iova = submit->bos[idx].iova;
	if (valid)
		*valid = !!(submit->bos[idx].flags & BO_VALID);

	return 0;
}

/* process the reloc's and patch up the cmdstream as needed: */
static int submit_reloc(struct msm_gem_submit *submit, struct msm_gem_object *obj,
		uint32_t offset, uint32_t nr_relocs, uint64_t relocs)
{
	uint32_t i, last_offset = 0;
	uint32_t *ptr;
	int ret = 0;

	if (offset % 4) {
		DRM_ERROR("non-aligned cmdstream buffer: %u\n", offset);
		return -EINVAL;
	}

	/* For now, just map the entire thing.  Eventually we probably
	 * to do it page-by-page, w/ kmap() if not vmap()d..
	 */
	ptr = msm_gem_get_vaddr_locked(&obj->base);

	if (IS_ERR(ptr)) {
		ret = PTR_ERR(ptr);
		DBG("failed to map: %d", ret);
		return ret;
	}

	for (i = 0; i < nr_relocs; i++) {
		struct drm_msm_gem_submit_reloc submit_reloc;
		void __user *userptr =
			u64_to_user_ptr(relocs + (i * sizeof(submit_reloc)));
		uint32_t off;
		uint64_t iova;
		bool valid;

		ret = copy_from_user(&submit_reloc, userptr, sizeof(submit_reloc));
		if (ret)
			goto out;

		if (submit_reloc.submit_offset % 4) {
			DRM_ERROR("non-aligned reloc offset: %u\n",
					submit_reloc.submit_offset);
			ret = -EINVAL;
			goto out;
		}

		/* offset in dwords: */
		off = submit_reloc.submit_offset / 4;

		if ((off >= (obj->base.size / 4)) ||
				(off < last_offset)) {
			DRM_ERROR("invalid offset %u at reloc %u\n", off, i);
			ret = -EINVAL;
			goto out;
		}

		ret = submit_bo(submit, submit_reloc.reloc_idx, NULL, &iova, &valid);
		if (ret)
			goto out;

		if (valid)
			continue;

		iova += submit_reloc.reloc_offset;

		if (submit_reloc.shift < 0)
			iova >>= -submit_reloc.shift;
		else
			iova <<= submit_reloc.shift;

		ptr[off] = iova | submit_reloc.or;

		last_offset = off;
	}

out:
	msm_gem_put_vaddr_locked(&obj->base);

	return ret;
}

static void submit_cleanup(struct msm_gem_submit *submit)
{
	unsigned i;

	for (i = 0; i < submit->nr_bos; i++) {
		struct msm_gem_object *msm_obj = submit->bos[i].obj;
		submit_unlock_unpin_bo(submit, i);
		list_del_init(&msm_obj->submit_entry);
		drm_gem_object_unreference(&msm_obj->base);
	}

	ww_acquire_fini(&submit->ticket);
}

int msm_ioctl_gem_submit(struct drm_device *dev, void *data,
		struct drm_file *file)
{
	struct msm_drm_private *priv = dev->dev_private;
	struct drm_msm_gem_submit *args = data;
	struct msm_file_private *ctx = file->driver_priv;
	struct msm_gem_submit *submit;
	struct msm_gpu *gpu = priv->gpu;
	struct dma_fence *in_fence = NULL;
	struct sync_file *sync_file = NULL;
	int out_fence_fd = -1;
	unsigned i;
	int ret;

	if (!gpu)
		return -ENXIO;

	/* for now, we just have 3d pipe.. eventually this would need to
	 * be more clever to dispatch to appropriate gpu module:
	 */
	if (MSM_PIPE_ID(args->flags) != MSM_PIPE_3D0)
		return -EINVAL;

	if (MSM_PIPE_FLAGS(args->flags) & ~MSM_SUBMIT_FLAGS)
		return -EINVAL;

	ret = mutex_lock_interruptible(&dev->struct_mutex);
	if (ret)
		return ret;

	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
		out_fence_fd = get_unused_fd_flags(O_CLOEXEC);
		if (out_fence_fd < 0) {
			ret = out_fence_fd;
			goto out_unlock;
		}
	}
	priv->struct_mutex_task = current;

	submit = submit_create(dev, gpu, args->nr_bos, args->nr_cmds);
	if (!submit) {
		ret = -ENOMEM;
		goto out_unlock;
	}

	ret = submit_lookup_objects(submit, args, file);
	if (ret)
		goto out;

	ret = submit_lock_objects(submit);
	if (ret)
		goto out;

	if (args->flags & MSM_SUBMIT_FENCE_FD_IN) {
		in_fence = sync_file_get_fence(args->fence_fd);

		if (!in_fence) {
			ret = -EINVAL;
			goto out;
		}

		/* TODO if we get an array-fence due to userspace merging multiple
		 * fences, we need a way to determine if all the backing fences
		 * are from our own context..
		 */

		if (in_fence->context != gpu->fctx->context) {
			ret = dma_fence_wait(in_fence, true);
			if (ret)
				goto out;
		}

	}

	if (!(args->fence & MSM_SUBMIT_NO_IMPLICIT)) {
		ret = submit_fence_sync(submit);
		if (ret)
			goto out;
	}

	ret = submit_pin_objects(submit);
	if (ret)
		goto out;

	for (i = 0; i < args->nr_cmds; i++) {
		struct drm_msm_gem_submit_cmd submit_cmd;
		void __user *userptr =
			u64_to_user_ptr(args->cmds + (i * sizeof(submit_cmd)));
		struct msm_gem_object *msm_obj;
		uint64_t iova;

		ret = copy_from_user(&submit_cmd, userptr, sizeof(submit_cmd));
		if (ret) {
			ret = -EFAULT;
			goto out;
		}

		/* validate input from userspace: */
		switch (submit_cmd.type) {
		case MSM_SUBMIT_CMD_BUF:
		case MSM_SUBMIT_CMD_IB_TARGET_BUF:
		case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
			break;
		default:
			DRM_ERROR("invalid type: %08x\n", submit_cmd.type);
			ret = -EINVAL;
			goto out;
		}

		ret = submit_bo(submit, submit_cmd.submit_idx,
				&msm_obj, &iova, NULL);
		if (ret)
			goto out;

		if (submit_cmd.size % 4) {
			DRM_ERROR("non-aligned cmdstream buffer size: %u\n",
					submit_cmd.size);
			ret = -EINVAL;
			goto out;
		}

		if ((submit_cmd.size + submit_cmd.submit_offset) >=
				msm_obj->base.size) {
			DRM_ERROR("invalid cmdstream size: %u\n", submit_cmd.size);
			ret = -EINVAL;
			goto out;
		}

		submit->cmd[i].type = submit_cmd.type;
		submit->cmd[i].size = submit_cmd.size / 4;
		submit->cmd[i].iova = iova + submit_cmd.submit_offset;
		submit->cmd[i].idx  = submit_cmd.submit_idx;

		if (submit->valid)
			continue;

		ret = submit_reloc(submit, msm_obj, submit_cmd.submit_offset,
				submit_cmd.nr_relocs, submit_cmd.relocs);
		if (ret)
			goto out;
	}

	submit->nr_cmds = i;

	submit->fence = msm_fence_alloc(gpu->fctx);
	if (IS_ERR(submit->fence)) {
		ret = PTR_ERR(submit->fence);
		submit->fence = NULL;
		goto out;
	}

	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
		sync_file = sync_file_create(submit->fence);
		if (!sync_file) {
			ret = -ENOMEM;
			goto out;
		}
	}

	msm_gpu_submit(gpu, submit, ctx);

	args->fence = submit->fence->seqno;

	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
		fd_install(out_fence_fd, sync_file->file);
		args->fence_fd = out_fence_fd;
	}

out:
	if (in_fence)
		dma_fence_put(in_fence);
	submit_cleanup(submit);
	if (ret)
		msm_gem_submit_free(submit);
out_unlock:
	if (ret && (out_fence_fd >= 0))
		put_unused_fd(out_fence_fd);
	priv->struct_mutex_task = NULL;
	mutex_unlock(&dev->struct_mutex);
	return ret;
}
Commit	Line	Data
7198e6b0 RC	1	/*
	2	* Copyright (C) 2013 Red Hat
	3	* Author: Rob Clark <robdclark@gmail.com>
	4	*
	5	* This program is free software; you can redistribute it and/or modify it
	6	* under the terms of the GNU General Public License version 2 as published by
	7	* the Free Software Foundation.
	8	*
	9	* This program is distributed in the hope that it will be useful, but WITHOUT
	10	* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
	11	* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
	12	* more details.
	13	*
	14	* You should have received a copy of the GNU General Public License along with
	15	* this program. If not, see <http://www.gnu.org/licenses/>.
	16	*/
	17
f0a42bb5 RC	18	#include <linux/sync_file.h>
f0a42bb5 RC	19
7198e6b0 RC	20	#include "msm_drv.h"
	21	#include "msm_gpu.h"
	22	#include "msm_gem.h"
	23
	24	/*
	25	* Cmdstream submission:
	26	*/
	27
7198e6b0	28	/* make sure these don't conflict w/ MSM_SUBMIT_BO_x */
340faef2	29	#define BO_VALID 0x8000 /* is current addr in cmdstream correct/valid? */
7198e6b0 RC	30	#define BO_LOCKED 0x4000
	31	#define BO_PINNED 0x2000
	32
7198e6b0	33	static struct msm_gem_submit submit_create(struct drm_device dev,
6b597ce2	34	struct msm_gpu *gpu, int nr_bos, int nr_cmds)
7198e6b0 RC	35	{
7198e6b0 RC	36	struct msm_gem_submit *submit;
6b597ce2 RC	37	int sz = sizeof(submit) + (nr_bos sizeof(submit->bos[0])) +
6b597ce2 RC	38	(nr_cmds * sizeof(*submit->cmd));
7198e6b0 RC	39
7198e6b0 RC	40	submit = kmalloc(sz, GFP_TEMPORARY \| __GFP_NOWARN \| __GFP_NORETRY);
6860b56c RC	41	if (!submit)
6860b56c RC	42	return NULL;
7198e6b0	43
6860b56c RC	44	submit->dev = dev;
6860b56c RC	45	submit->gpu = gpu;
ba344afd	46	submit->fence = NULL;
4816b626	47	submit->pid = get_pid(task_pid(current));
6b597ce2	48	submit->cmd = (void *)&submit->bos[nr_bos];
7198e6b0	49
6860b56c RC	50	/* initially, until copy_from_user() and bo lookup succeeds: */
	51	submit->nr_bos = 0;
	52	submit->nr_cmds = 0;
	53
ba344afd	54	INIT_LIST_HEAD(&submit->node);
6860b56c RC	55	INIT_LIST_HEAD(&submit->bo_list);
6860b56c RC	56	ww_acquire_init(&submit->ticket, &reservation_ww_class);
7198e6b0 RC	57
	58	return submit;
	59	}
	60
40e6815b RC	61	void msm_gem_submit_free(struct msm_gem_submit *submit)
40e6815b RC	62	{
f54d1867	63	dma_fence_put(submit->fence);
40e6815b	64	list_del(&submit->node);
4816b626	65	put_pid(submit->pid);
40e6815b RC	66	kfree(submit);
	67	}
	68
89f82cbb RC	69	static inline unsigned long __must_check
	70	copy_from_user_inatomic(void to, const void __user from, unsigned long n)
	71	{
	72	if (access_ok(VERIFY_READ, from, n))
	73	return __copy_from_user_inatomic(to, from, n);
	74	return -EFAULT;
	75	}
	76
7198e6b0 RC	77	static int submit_lookup_objects(struct msm_gem_submit *submit,
	78	struct drm_msm_gem_submit args, struct drm_file file)
	79	{
	80	unsigned i;
	81	int ret = 0;
	82
	83	spin_lock(&file->table_lock);
89f82cbb	84	pagefault_disable();
7198e6b0 RC	85
	86	for (i = 0; i < args->nr_bos; i++) {
	87	struct drm_msm_gem_submit_bo submit_bo;
	88	struct drm_gem_object *obj;
	89	struct msm_gem_object *msm_obj;
	90	void __user *userptr =
3ed605bc	91	u64_to_user_ptr(args->bos + (i * sizeof(submit_bo)));
7198e6b0	92
a9e26cab RC	93	/* make sure we don't have garbage flags, in case we hit
	94	* error path before flags is initialized:
	95	*/
	96	submit->bos[i].flags = 0;
	97
89f82cbb RC	98	ret = copy_from_user_inatomic(&submit_bo, userptr, sizeof(submit_bo));
	99	if (unlikely(ret)) {
	100	pagefault_enable();
	101	spin_unlock(&file->table_lock);
	102	ret = copy_from_user(&submit_bo, userptr, sizeof(submit_bo));
	103	if (ret)
	104	goto out;
	105	spin_lock(&file->table_lock);
	106	pagefault_disable();
7198e6b0 RC	107	}
7198e6b0 RC	108
a6cb3b86 JC	109	if ((submit_bo.flags & ~MSM_SUBMIT_BO_FLAGS) \|\|
a6cb3b86 JC	110	!(submit_bo.flags & MSM_SUBMIT_BO_FLAGS)) {
19872533	111	DRM_ERROR("invalid flags: %x\n", submit_bo.flags);
7198e6b0 RC	112	ret = -EINVAL;
	113	goto out_unlock;
	114	}
	115
	116	submit->bos[i].flags = submit_bo.flags;
	117	/* in validate_objects() we figure out if this is true: */
	118	submit->bos[i].iova = submit_bo.presumed;
	119
	120	/* normally use drm_gem_object_lookup(), but for bulk lookup
	121	* all under single table_lock just hit object_idr directly:
	122	*/
	123	obj = idr_find(&file->object_idr, submit_bo.handle);
	124	if (!obj) {
19872533	125	DRM_ERROR("invalid handle %u at index %u\n", submit_bo.handle, i);
7198e6b0 RC	126	ret = -EINVAL;
	127	goto out_unlock;
	128	}
	129
	130	msm_obj = to_msm_bo(obj);
	131
	132	if (!list_empty(&msm_obj->submit_entry)) {
19872533	133	DRM_ERROR("handle %u at index %u already on submit list\n",
7198e6b0 RC	134	submit_bo.handle, i);
	135	ret = -EINVAL;
	136	goto out_unlock;
	137	}
	138
	139	drm_gem_object_reference(obj);
	140
	141	submit->bos[i].obj = msm_obj;
	142
	143	list_add_tail(&msm_obj->submit_entry, &submit->bo_list);
	144	}
	145
	146	out_unlock:
89f82cbb	147	pagefault_enable();
7198e6b0 RC	148	spin_unlock(&file->table_lock);
7198e6b0 RC	149
89f82cbb RC	150	out:
	151	submit->nr_bos = i;
	152
7198e6b0 RC	153	return ret;
	154	}
	155
	156	static void submit_unlock_unpin_bo(struct msm_gem_submit *submit, int i)
	157	{
	158	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	159
	160	if (submit->bos[i].flags & BO_PINNED)
	161	msm_gem_put_iova(&msm_obj->base, submit->gpu->id);
	162
	163	if (submit->bos[i].flags & BO_LOCKED)
	164	ww_mutex_unlock(&msm_obj->resv->lock);
	165
	166	if (!(submit->bos[i].flags & BO_VALID))
	167	submit->bos[i].iova = 0;
	168
	169	submit->bos[i].flags &= ~(BO_LOCKED \| BO_PINNED);
	170	}
	171
	172	/* This is where we make sure all the bo's are reserved and pin'd: */
340faef2	173	static int submit_lock_objects(struct msm_gem_submit *submit)
7198e6b0 RC	174	{
	175	int contended, slow_locked = -1, i, ret = 0;
	176
	177	retry:
7198e6b0 RC	178	for (i = 0; i < submit->nr_bos; i++) {
7198e6b0 RC	179	struct msm_gem_object *msm_obj = submit->bos[i].obj;
7198e6b0 RC	180
	181	if (slow_locked == i)
	182	slow_locked = -1;
	183
	184	contended = i;
	185
	186	if (!(submit->bos[i].flags & BO_LOCKED)) {
	187	ret = ww_mutex_lock_interruptible(&msm_obj->resv->lock,
	188	&submit->ticket);
	189	if (ret)
	190	goto fail;
	191	submit->bos[i].flags \|= BO_LOCKED;
	192	}
7198e6b0 RC	193	}
	194
	195	ww_acquire_done(&submit->ticket);
	196
	197	return 0;
	198
	199	fail:
	200	for (; i >= 0; i--)
	201	submit_unlock_unpin_bo(submit, i);
	202
	203	if (slow_locked > 0)
	204	submit_unlock_unpin_bo(submit, slow_locked);
	205
	206	if (ret == -EDEADLK) {
	207	struct msm_gem_object *msm_obj = submit->bos[contended].obj;
	208	/* we lost out in a seqno race, lock and retry.. */
	209	ret = ww_mutex_lock_slow_interruptible(&msm_obj->resv->lock,
	210	&submit->ticket);
	211	if (!ret) {
	212	submit->bos[contended].flags \|= BO_LOCKED;
	213	slow_locked = contended;
	214	goto retry;
	215	}
	216	}
	217
	218	return ret;
	219	}
	220
b6295f9a RC	221	static int submit_fence_sync(struct msm_gem_submit *submit)
	222	{
	223	int i, ret = 0;
	224
	225	for (i = 0; i < submit->nr_bos; i++) {
	226	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	227	bool write = submit->bos[i].flags & MSM_SUBMIT_BO_WRITE;
	228
	229	ret = msm_gem_sync_object(&msm_obj->base, submit->gpu->fctx, write);
	230	if (ret)
	231	break;
	232	}
	233
	234	return ret;
	235	}
	236
340faef2 RC	237	static int submit_pin_objects(struct msm_gem_submit *submit)
	238	{
	239	int i, ret = 0;
	240
	241	submit->valid = true;
	242
	243	for (i = 0; i < submit->nr_bos; i++) {
	244	struct msm_gem_object *msm_obj = submit->bos[i].obj;
78babc16	245	uint64_t iova;
340faef2 RC	246
	247	/* if locking succeeded, pin bo: */
	248	ret = msm_gem_get_iova_locked(&msm_obj->base,
	249	submit->gpu->id, &iova);
	250
	251	if (ret)
	252	break;
	253
	254	submit->bos[i].flags \|= BO_PINNED;
	255
	256	if (iova == submit->bos[i].iova) {
	257	submit->bos[i].flags \|= BO_VALID;
	258	} else {
	259	submit->bos[i].iova = iova;
	260	/* iova changed, so address in cmdstream is not valid: */
	261	submit->bos[i].flags &= ~BO_VALID;
	262	submit->valid = false;
	263	}
	264	}
	265
	266	return ret;
	267	}
	268
7198e6b0	269	static int submit_bo(struct msm_gem_submit *submit, uint32_t idx,
78babc16	270	struct msm_gem_object *obj, uint64_t iova, bool *valid)
7198e6b0 RC	271	{
7198e6b0 RC	272	if (idx >= submit->nr_bos) {
19872533 RC	273	DRM_ERROR("invalid buffer index: %u (out of %u)\n",
	274	idx, submit->nr_bos);
	275	return -EINVAL;
7198e6b0 RC	276	}
	277
	278	if (obj)
	279	*obj = submit->bos[idx].obj;
	280	if (iova)
	281	*iova = submit->bos[idx].iova;
	282	if (valid)
	283	*valid = !!(submit->bos[idx].flags & BO_VALID);
	284
	285	return 0;
	286	}
	287
	288	/* process the reloc's and patch up the cmdstream as needed: */
	289	static int submit_reloc(struct msm_gem_submit submit, struct msm_gem_object obj,
	290	uint32_t offset, uint32_t nr_relocs, uint64_t relocs)
	291	{
	292	uint32_t i, last_offset = 0;
	293	uint32_t *ptr;
6490abc4	294	int ret = 0;
7198e6b0 RC	295
7198e6b0 RC	296	if (offset % 4) {
19872533	297	DRM_ERROR("non-aligned cmdstream buffer: %u\n", offset);
7198e6b0 RC	298	return -EINVAL;
	299	}
	300
	301	/* For now, just map the entire thing. Eventually we probably
	302	* to do it page-by-page, w/ kmap() if not vmap()d..
	303	*/
18f23049	304	ptr = msm_gem_get_vaddr_locked(&obj->base);
7198e6b0 RC	305
	306	if (IS_ERR(ptr)) {
	307	ret = PTR_ERR(ptr);
	308	DBG("failed to map: %d", ret);
	309	return ret;
	310	}
	311
	312	for (i = 0; i < nr_relocs; i++) {
	313	struct drm_msm_gem_submit_reloc submit_reloc;
	314	void __user *userptr =
3ed605bc	315	u64_to_user_ptr(relocs + (i * sizeof(submit_reloc)));
78babc16 RC	316	uint32_t off;
78babc16 RC	317	uint64_t iova;
7198e6b0 RC	318	bool valid;
	319
	320	ret = copy_from_user(&submit_reloc, userptr, sizeof(submit_reloc));
	321	if (ret)
6490abc4	322	goto out;
7198e6b0 RC	323
7198e6b0 RC	324	if (submit_reloc.submit_offset % 4) {
19872533	325	DRM_ERROR("non-aligned reloc offset: %u\n",
7198e6b0	326	submit_reloc.submit_offset);
6490abc4 JC	327	ret = -EINVAL;
6490abc4 JC	328	goto out;
7198e6b0 RC	329	}
	330
	331	/* offset in dwords: */
	332	off = submit_reloc.submit_offset / 4;
	333
	334	if ((off >= (obj->base.size / 4)) \|\|
	335	(off < last_offset)) {
19872533	336	DRM_ERROR("invalid offset %u at reloc %u\n", off, i);
6490abc4 JC	337	ret = -EINVAL;
6490abc4 JC	338	goto out;
7198e6b0 RC	339	}
	340
	341	ret = submit_bo(submit, submit_reloc.reloc_idx, NULL, &iova, &valid);
	342	if (ret)
6490abc4	343	goto out;
7198e6b0 RC	344
	345	if (valid)
	346	continue;
	347
	348	iova += submit_reloc.reloc_offset;
	349
	350	if (submit_reloc.shift < 0)
	351	iova >>= -submit_reloc.shift;
	352	else
	353	iova <<= submit_reloc.shift;
	354
	355	ptr[off] = iova \| submit_reloc.or;
	356
	357	last_offset = off;
	358	}
	359
6490abc4	360	out:
18f23049 RC	361	msm_gem_put_vaddr_locked(&obj->base);
18f23049 RC	362
6490abc4	363	return ret;
7198e6b0 RC	364	}
7198e6b0 RC	365
40e6815b	366	static void submit_cleanup(struct msm_gem_submit *submit)
7198e6b0 RC	367	{
	368	unsigned i;
	369
7198e6b0 RC	370	for (i = 0; i < submit->nr_bos; i++) {
	371	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	372	submit_unlock_unpin_bo(submit, i);
	373	list_del_init(&msm_obj->submit_entry);
	374	drm_gem_object_unreference(&msm_obj->base);
	375	}
7198e6b0 RC	376
7198e6b0 RC	377	ww_acquire_fini(&submit->ticket);
7198e6b0 RC	378	}
	379
	380	int msm_ioctl_gem_submit(struct drm_device dev, void data,
	381	struct drm_file *file)
	382	{
	383	struct msm_drm_private *priv = dev->dev_private;
	384	struct drm_msm_gem_submit *args = data;
	385	struct msm_file_private *ctx = file->driver_priv;
	386	struct msm_gem_submit *submit;
c01a958e	387	struct msm_gpu *gpu = priv->gpu;
f54d1867	388	struct dma_fence *in_fence = NULL;
4cd09459 RC	389	struct sync_file *sync_file = NULL;
4cd09459 RC	390	int out_fence_fd = -1;
7198e6b0 RC	391	unsigned i;
	392	int ret;
	393
c01a958e RC	394	if (!gpu)
	395	return -ENXIO;
	396
7198e6b0 RC	397	/* for now, we just have 3d pipe.. eventually this would need to
	398	* be more clever to dispatch to appropriate gpu module:
	399	*/
d9c181e2 RC	400	if (MSM_PIPE_ID(args->flags) != MSM_PIPE_3D0)
	401	return -EINVAL;
	402
	403	if (MSM_PIPE_FLAGS(args->flags) & ~MSM_SUBMIT_FLAGS)
7198e6b0 RC	404	return -EINVAL;
7198e6b0 RC	405
b5b4c264 RC	406	ret = mutex_lock_interruptible(&dev->struct_mutex);
	407	if (ret)
	408	return ret;
687f084a	409
4cd09459 RC	410	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	411	out_fence_fd = get_unused_fd_flags(O_CLOEXEC);
	412	if (out_fence_fd < 0) {
	413	ret = out_fence_fd;
	414	goto out_unlock;
	415	}
	416	}
d78d383a	417	priv->struct_mutex_task = current;
4cd09459	418
6b597ce2	419	submit = submit_create(dev, gpu, args->nr_bos, args->nr_cmds);
b5b4c264 RC	420	if (!submit) {
	421	ret = -ENOMEM;
	422	goto out_unlock;
	423	}
7198e6b0 RC	424
	425	ret = submit_lookup_objects(submit, args, file);
	426	if (ret)
	427	goto out;
	428
340faef2 RC	429	ret = submit_lock_objects(submit);
	430	if (ret)
	431	goto out;
	432
f0a42bb5 RC	433	if (args->flags & MSM_SUBMIT_FENCE_FD_IN) {
	434	in_fence = sync_file_get_fence(args->fence_fd);
	435
	436	if (!in_fence) {
	437	ret = -EINVAL;
	438	goto out;
	439	}
	440
	441	/* TODO if we get an array-fence due to userspace merging multiple
	442	* fences, we need a way to determine if all the backing fences
	443	* are from our own context..
	444	*/
	445
	446	if (in_fence->context != gpu->fctx->context) {
f54d1867	447	ret = dma_fence_wait(in_fence, true);
f0a42bb5 RC	448	if (ret)
	449	goto out;
	450	}
	451
	452	}
	453
	454	if (!(args->fence & MSM_SUBMIT_NO_IMPLICIT)) {
	455	ret = submit_fence_sync(submit);
	456	if (ret)
	457	goto out;
	458	}
b6295f9a	459
340faef2	460	ret = submit_pin_objects(submit);
7198e6b0 RC	461	if (ret)
	462	goto out;
	463
	464	for (i = 0; i < args->nr_cmds; i++) {
	465	struct drm_msm_gem_submit_cmd submit_cmd;
	466	void __user *userptr =
3ed605bc	467	u64_to_user_ptr(args->cmds + (i * sizeof(submit_cmd)));
7198e6b0	468	struct msm_gem_object *msm_obj;
78babc16	469	uint64_t iova;
7198e6b0 RC	470
	471	ret = copy_from_user(&submit_cmd, userptr, sizeof(submit_cmd));
	472	if (ret) {
	473	ret = -EFAULT;
	474	goto out;
	475	}
	476
93ddb0d3 RC	477	/* validate input from userspace: */
	478	switch (submit_cmd.type) {
	479	case MSM_SUBMIT_CMD_BUF:
	480	case MSM_SUBMIT_CMD_IB_TARGET_BUF:
	481	case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
	482	break;
	483	default:
	484	DRM_ERROR("invalid type: %08x\n", submit_cmd.type);
	485	ret = -EINVAL;
	486	goto out;
	487	}
	488
7198e6b0 RC	489	ret = submit_bo(submit, submit_cmd.submit_idx,
	490	&msm_obj, &iova, NULL);
	491	if (ret)
	492	goto out;
	493
	494	if (submit_cmd.size % 4) {
19872533	495	DRM_ERROR("non-aligned cmdstream buffer size: %u\n",
7198e6b0 RC	496	submit_cmd.size);
	497	ret = -EINVAL;
	498	goto out;
	499	}
	500
19872533 RC	501	if ((submit_cmd.size + submit_cmd.submit_offset) >=
	502	msm_obj->base.size) {
	503	DRM_ERROR("invalid cmdstream size: %u\n", submit_cmd.size);
7198e6b0 RC	504	ret = -EINVAL;
	505	goto out;
	506	}
	507
	508	submit->cmd[i].type = submit_cmd.type;
	509	submit->cmd[i].size = submit_cmd.size / 4;
	510	submit->cmd[i].iova = iova + submit_cmd.submit_offset;
a7d3c950	511	submit->cmd[i].idx = submit_cmd.submit_idx;
7198e6b0 RC	512
	513	if (submit->valid)
	514	continue;
	515
	516	ret = submit_reloc(submit, msm_obj, submit_cmd.submit_offset,
	517	submit_cmd.nr_relocs, submit_cmd.relocs);
	518	if (ret)
	519	goto out;
	520	}
	521
	522	submit->nr_cmds = i;
	523
f44d32c7 RC	524	submit->fence = msm_fence_alloc(gpu->fctx);
	525	if (IS_ERR(submit->fence)) {
	526	ret = PTR_ERR(submit->fence);
	527	submit->fence = NULL;
	528	goto out;
	529	}
	530
4cd09459 RC	531	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	532	sync_file = sync_file_create(submit->fence);
	533	if (!sync_file) {
	534	ret = -ENOMEM;
	535	goto out;
	536	}
	537	}
	538
f44d32c7	539	msm_gpu_submit(gpu, submit, ctx);
7198e6b0	540
b6295f9a	541	args->fence = submit->fence->seqno;
7198e6b0	542
4cd09459 RC	543	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	544	fd_install(out_fence_fd, sync_file->file);
	545	args->fence_fd = out_fence_fd;
	546	}
	547
7198e6b0	548	out:
f0a42bb5	549	if (in_fence)
f54d1867	550	dma_fence_put(in_fence);
40e6815b RC	551	submit_cleanup(submit);
	552	if (ret)
	553	msm_gem_submit_free(submit);
b5b4c264	554	out_unlock:
4cd09459 RC	555	if (ret && (out_fence_fd >= 0))
4cd09459 RC	556	put_unused_fd(out_fence_fd);
d78d383a	557	priv->struct_mutex_task = NULL;
c2703b13	558	mutex_unlock(&dev->struct_mutex);
7198e6b0 RC	559	return ret;
7198e6b0 RC	560	}