[mirror_ubuntu-focal-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, uint32_t nr_bos, uint32_t nr_cmds)
{
	struct msm_gem_submit *submit;
	uint64_t sz = sizeof(*submit) + ((u64)nr_bos * sizeof(submit->bos[0])) +
		((u64)nr_cmds * sizeof(submit->cmd[0]));

	if (sz > SIZE_MAX)
		return NULL;

	submit = kmalloc(sz, GFP_KERNEL | __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;

		if (copy_from_user_inatomic(&submit_bo, userptr, sizeof(submit_bo))) {
			pagefault_enable();
			spin_unlock(&file->table_lock);
			if (copy_from_user(&submit_bo, userptr, sizeof(submit_bo))) {
				ret = -EFAULT;
				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->aspace);

	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, bool no_implicit)
{
	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;

		if (!write) {
			/* NOTE: _reserve_shared() must happen before
			 * _add_shared_fence(), which makes this a slightly
			 * strange place to call it.  OTOH this is a
			 * convenient can-fail point to hook it in.
			 */
			ret = reservation_object_reserve_shared(msm_obj->resv);
			if (ret)
				return ret;
		}

		if (no_implicit)
			continue;

		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(&msm_obj->base,
				submit->gpu->aspace, &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(&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;

		if (copy_from_user(&submit_reloc, userptr, sizeof(submit_reloc))) {
			ret = -EFAULT;
			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(&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;

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

		if (!in_fence)
			return -EINVAL;

		/*
		 * Wait if the fence is from a foreign context, or if the fence
		 * array contains any fence from a foreign context.
		 */
		if (!dma_fence_match_context(in_fence, gpu->fctx->context)) {
			ret = dma_fence_wait(in_fence, true);
			if (ret)
				return ret;
		}
	}

	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;

	ret = submit_fence_sync(submit, !!(args->flags & MSM_SUBMIT_NO_IMPLICIT));
	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.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,
4a630fad	34	struct msm_gpu *gpu, uint32_t nr_bos, uint32_t nr_cmds)
7198e6b0 RC	35	{
7198e6b0 RC	36	struct msm_gem_submit *submit;
65e93108 DC	37	uint64_t sz = sizeof(submit) + ((u64)nr_bos sizeof(submit->bos[0])) +
65e93108 DC	38	((u64)nr_cmds * sizeof(submit->cmd[0]));
4a630fad KL	39
	40	if (sz > SIZE_MAX)
	41	return NULL;
7198e6b0	42
0ee931c4	43	submit = kmalloc(sz, GFP_KERNEL \| __GFP_NOWARN \| __GFP_NORETRY);
6860b56c RC	44	if (!submit)
6860b56c RC	45	return NULL;
7198e6b0	46
6860b56c RC	47	submit->dev = dev;
6860b56c RC	48	submit->gpu = gpu;
ba344afd	49	submit->fence = NULL;
4816b626	50	submit->pid = get_pid(task_pid(current));
6b597ce2	51	submit->cmd = (void *)&submit->bos[nr_bos];
7198e6b0	52
6860b56c RC	53	/* initially, until copy_from_user() and bo lookup succeeds: */
	54	submit->nr_bos = 0;
	55	submit->nr_cmds = 0;
	56
ba344afd	57	INIT_LIST_HEAD(&submit->node);
6860b56c RC	58	INIT_LIST_HEAD(&submit->bo_list);
6860b56c RC	59	ww_acquire_init(&submit->ticket, &reservation_ww_class);
7198e6b0 RC	60
	61	return submit;
	62	}
	63
40e6815b RC	64	void msm_gem_submit_free(struct msm_gem_submit *submit)
40e6815b RC	65	{
f54d1867	66	dma_fence_put(submit->fence);
40e6815b	67	list_del(&submit->node);
4816b626	68	put_pid(submit->pid);
40e6815b RC	69	kfree(submit);
	70	}
	71
89f82cbb RC	72	static inline unsigned long __must_check
	73	copy_from_user_inatomic(void to, const void __user from, unsigned long n)
	74	{
	75	if (access_ok(VERIFY_READ, from, n))
	76	return __copy_from_user_inatomic(to, from, n);
	77	return -EFAULT;
	78	}
	79
7198e6b0 RC	80	static int submit_lookup_objects(struct msm_gem_submit *submit,
	81	struct drm_msm_gem_submit args, struct drm_file file)
	82	{
	83	unsigned i;
	84	int ret = 0;
	85
	86	spin_lock(&file->table_lock);
89f82cbb	87	pagefault_disable();
7198e6b0 RC	88
	89	for (i = 0; i < args->nr_bos; i++) {
	90	struct drm_msm_gem_submit_bo submit_bo;
	91	struct drm_gem_object *obj;
	92	struct msm_gem_object *msm_obj;
	93	void __user *userptr =
3ed605bc	94	u64_to_user_ptr(args->bos + (i * sizeof(submit_bo)));
7198e6b0	95
a9e26cab RC	96	/* make sure we don't have garbage flags, in case we hit
	97	* error path before flags is initialized:
	98	*/
	99	submit->bos[i].flags = 0;
	100
21c42da1	101	if (copy_from_user_inatomic(&submit_bo, userptr, sizeof(submit_bo))) {
89f82cbb RC	102	pagefault_enable();
89f82cbb RC	103	spin_unlock(&file->table_lock);
21c42da1 DC	104	if (copy_from_user(&submit_bo, userptr, sizeof(submit_bo))) {
21c42da1 DC	105	ret = -EFAULT;
89f82cbb	106	goto out;
21c42da1	107	}
89f82cbb RC	108	spin_lock(&file->table_lock);
89f82cbb RC	109	pagefault_disable();
7198e6b0 RC	110	}
7198e6b0 RC	111
a6cb3b86 JC	112	if ((submit_bo.flags & ~MSM_SUBMIT_BO_FLAGS) \|\|
a6cb3b86 JC	113	!(submit_bo.flags & MSM_SUBMIT_BO_FLAGS)) {
19872533	114	DRM_ERROR("invalid flags: %x\n", submit_bo.flags);
7198e6b0 RC	115	ret = -EINVAL;
	116	goto out_unlock;
	117	}
	118
	119	submit->bos[i].flags = submit_bo.flags;
	120	/* in validate_objects() we figure out if this is true: */
	121	submit->bos[i].iova = submit_bo.presumed;
	122
	123	/* normally use drm_gem_object_lookup(), but for bulk lookup
	124	* all under single table_lock just hit object_idr directly:
	125	*/
	126	obj = idr_find(&file->object_idr, submit_bo.handle);
	127	if (!obj) {
19872533	128	DRM_ERROR("invalid handle %u at index %u\n", submit_bo.handle, i);
7198e6b0 RC	129	ret = -EINVAL;
	130	goto out_unlock;
	131	}
	132
	133	msm_obj = to_msm_bo(obj);
	134
	135	if (!list_empty(&msm_obj->submit_entry)) {
19872533	136	DRM_ERROR("handle %u at index %u already on submit list\n",
7198e6b0 RC	137	submit_bo.handle, i);
	138	ret = -EINVAL;
	139	goto out_unlock;
	140	}
	141
	142	drm_gem_object_reference(obj);
	143
	144	submit->bos[i].obj = msm_obj;
	145
	146	list_add_tail(&msm_obj->submit_entry, &submit->bo_list);
	147	}
	148
	149	out_unlock:
89f82cbb	150	pagefault_enable();
7198e6b0 RC	151	spin_unlock(&file->table_lock);
7198e6b0 RC	152
89f82cbb RC	153	out:
	154	submit->nr_bos = i;
	155
7198e6b0 RC	156	return ret;
	157	}
	158
	159	static void submit_unlock_unpin_bo(struct msm_gem_submit *submit, int i)
	160	{
	161	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	162
	163	if (submit->bos[i].flags & BO_PINNED)
8bdcd949	164	msm_gem_put_iova(&msm_obj->base, submit->gpu->aspace);
7198e6b0 RC	165
	166	if (submit->bos[i].flags & BO_LOCKED)
	167	ww_mutex_unlock(&msm_obj->resv->lock);
	168
	169	if (!(submit->bos[i].flags & BO_VALID))
	170	submit->bos[i].iova = 0;
	171
	172	submit->bos[i].flags &= ~(BO_LOCKED \| BO_PINNED);
	173	}
	174
	175	/* This is where we make sure all the bo's are reserved and pin'd: */
340faef2	176	static int submit_lock_objects(struct msm_gem_submit *submit)
7198e6b0 RC	177	{
	178	int contended, slow_locked = -1, i, ret = 0;
	179
	180	retry:
7198e6b0 RC	181	for (i = 0; i < submit->nr_bos; i++) {
7198e6b0 RC	182	struct msm_gem_object *msm_obj = submit->bos[i].obj;
7198e6b0 RC	183
	184	if (slow_locked == i)
	185	slow_locked = -1;
	186
	187	contended = i;
	188
	189	if (!(submit->bos[i].flags & BO_LOCKED)) {
	190	ret = ww_mutex_lock_interruptible(&msm_obj->resv->lock,
	191	&submit->ticket);
	192	if (ret)
	193	goto fail;
	194	submit->bos[i].flags \|= BO_LOCKED;
	195	}
7198e6b0 RC	196	}
	197
	198	ww_acquire_done(&submit->ticket);
	199
	200	return 0;
	201
	202	fail:
	203	for (; i >= 0; i--)
	204	submit_unlock_unpin_bo(submit, i);
	205
	206	if (slow_locked > 0)
	207	submit_unlock_unpin_bo(submit, slow_locked);
	208
	209	if (ret == -EDEADLK) {
	210	struct msm_gem_object *msm_obj = submit->bos[contended].obj;
	211	/* we lost out in a seqno race, lock and retry.. */
	212	ret = ww_mutex_lock_slow_interruptible(&msm_obj->resv->lock,
	213	&submit->ticket);
	214	if (!ret) {
	215	submit->bos[contended].flags \|= BO_LOCKED;
	216	slow_locked = contended;
	217	goto retry;
	218	}
	219	}
	220
	221	return ret;
	222	}
	223
06451a3d	224	static int submit_fence_sync(struct msm_gem_submit *submit, bool no_implicit)
b6295f9a RC	225	{
	226	int i, ret = 0;
	227
	228	for (i = 0; i < submit->nr_bos; i++) {
	229	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	230	bool write = submit->bos[i].flags & MSM_SUBMIT_BO_WRITE;
	231
06451a3d RC	232	if (!write) {
	233	/* NOTE: _reserve_shared() must happen before
	234	* _add_shared_fence(), which makes this a slightly
	235	* strange place to call it. OTOH this is a
	236	* convenient can-fail point to hook it in.
	237	*/
	238	ret = reservation_object_reserve_shared(msm_obj->resv);
	239	if (ret)
	240	return ret;
	241	}
	242
	243	if (no_implicit)
	244	continue;
	245
b6295f9a RC	246	ret = msm_gem_sync_object(&msm_obj->base, submit->gpu->fctx, write);
	247	if (ret)
	248	break;
	249	}
	250
	251	return ret;
	252	}
	253
340faef2 RC	254	static int submit_pin_objects(struct msm_gem_submit *submit)
	255	{
	256	int i, ret = 0;
	257
	258	submit->valid = true;
	259
	260	for (i = 0; i < submit->nr_bos; i++) {
	261	struct msm_gem_object *msm_obj = submit->bos[i].obj;
78babc16	262	uint64_t iova;
340faef2 RC	263
340faef2 RC	264	/* if locking succeeded, pin bo: */
0e08270a	265	ret = msm_gem_get_iova(&msm_obj->base,
8bdcd949	266	submit->gpu->aspace, &iova);
340faef2 RC	267
	268	if (ret)
	269	break;
	270
	271	submit->bos[i].flags \|= BO_PINNED;
	272
	273	if (iova == submit->bos[i].iova) {
	274	submit->bos[i].flags \|= BO_VALID;
	275	} else {
	276	submit->bos[i].iova = iova;
	277	/* iova changed, so address in cmdstream is not valid: */
	278	submit->bos[i].flags &= ~BO_VALID;
	279	submit->valid = false;
	280	}
	281	}
	282
	283	return ret;
	284	}
	285
7198e6b0	286	static int submit_bo(struct msm_gem_submit *submit, uint32_t idx,
78babc16	287	struct msm_gem_object *obj, uint64_t iova, bool *valid)
7198e6b0 RC	288	{
7198e6b0 RC	289	if (idx >= submit->nr_bos) {
19872533 RC	290	DRM_ERROR("invalid buffer index: %u (out of %u)\n",
	291	idx, submit->nr_bos);
	292	return -EINVAL;
7198e6b0 RC	293	}
	294
	295	if (obj)
	296	*obj = submit->bos[idx].obj;
	297	if (iova)
	298	*iova = submit->bos[idx].iova;
	299	if (valid)
	300	*valid = !!(submit->bos[idx].flags & BO_VALID);
	301
	302	return 0;
	303	}
	304
	305	/* process the reloc's and patch up the cmdstream as needed: */
	306	static int submit_reloc(struct msm_gem_submit submit, struct msm_gem_object obj,
	307	uint32_t offset, uint32_t nr_relocs, uint64_t relocs)
	308	{
	309	uint32_t i, last_offset = 0;
	310	uint32_t *ptr;
6490abc4	311	int ret = 0;
7198e6b0 RC	312
7198e6b0 RC	313	if (offset % 4) {
19872533	314	DRM_ERROR("non-aligned cmdstream buffer: %u\n", offset);
7198e6b0 RC	315	return -EINVAL;
	316	}
	317
	318	/* For now, just map the entire thing. Eventually we probably
	319	* to do it page-by-page, w/ kmap() if not vmap()d..
	320	*/
0e08270a	321	ptr = msm_gem_get_vaddr(&obj->base);
7198e6b0 RC	322
	323	if (IS_ERR(ptr)) {
	324	ret = PTR_ERR(ptr);
	325	DBG("failed to map: %d", ret);
	326	return ret;
	327	}
	328
	329	for (i = 0; i < nr_relocs; i++) {
	330	struct drm_msm_gem_submit_reloc submit_reloc;
	331	void __user *userptr =
3ed605bc	332	u64_to_user_ptr(relocs + (i * sizeof(submit_reloc)));
78babc16 RC	333	uint32_t off;
78babc16 RC	334	uint64_t iova;
7198e6b0 RC	335	bool valid;
7198e6b0 RC	336
21c42da1 DC	337	if (copy_from_user(&submit_reloc, userptr, sizeof(submit_reloc))) {
21c42da1 DC	338	ret = -EFAULT;
6490abc4	339	goto out;
21c42da1	340	}
7198e6b0 RC	341
7198e6b0 RC	342	if (submit_reloc.submit_offset % 4) {
19872533	343	DRM_ERROR("non-aligned reloc offset: %u\n",
7198e6b0	344	submit_reloc.submit_offset);
6490abc4 JC	345	ret = -EINVAL;
6490abc4 JC	346	goto out;
7198e6b0 RC	347	}
	348
	349	/* offset in dwords: */
	350	off = submit_reloc.submit_offset / 4;
	351
	352	if ((off >= (obj->base.size / 4)) \|\|
	353	(off < last_offset)) {
19872533	354	DRM_ERROR("invalid offset %u at reloc %u\n", off, i);
6490abc4 JC	355	ret = -EINVAL;
6490abc4 JC	356	goto out;
7198e6b0 RC	357	}
	358
	359	ret = submit_bo(submit, submit_reloc.reloc_idx, NULL, &iova, &valid);
	360	if (ret)
6490abc4	361	goto out;
7198e6b0 RC	362
	363	if (valid)
	364	continue;
	365
	366	iova += submit_reloc.reloc_offset;
	367
	368	if (submit_reloc.shift < 0)
	369	iova >>= -submit_reloc.shift;
	370	else
	371	iova <<= submit_reloc.shift;
	372
	373	ptr[off] = iova \| submit_reloc.or;
	374
	375	last_offset = off;
	376	}
	377
6490abc4	378	out:
0e08270a	379	msm_gem_put_vaddr(&obj->base);
18f23049	380
6490abc4	381	return ret;
7198e6b0 RC	382	}
7198e6b0 RC	383
40e6815b	384	static void submit_cleanup(struct msm_gem_submit *submit)
7198e6b0 RC	385	{
	386	unsigned i;
	387
7198e6b0 RC	388	for (i = 0; i < submit->nr_bos; i++) {
	389	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	390	submit_unlock_unpin_bo(submit, i);
	391	list_del_init(&msm_obj->submit_entry);
	392	drm_gem_object_unreference(&msm_obj->base);
	393	}
7198e6b0 RC	394
7198e6b0 RC	395	ww_acquire_fini(&submit->ticket);
7198e6b0 RC	396	}
	397
	398	int msm_ioctl_gem_submit(struct drm_device dev, void data,
	399	struct drm_file *file)
	400	{
	401	struct msm_drm_private *priv = dev->dev_private;
	402	struct drm_msm_gem_submit *args = data;
	403	struct msm_file_private *ctx = file->driver_priv;
	404	struct msm_gem_submit *submit;
c01a958e	405	struct msm_gpu *gpu = priv->gpu;
f54d1867	406	struct dma_fence *in_fence = NULL;
4cd09459 RC	407	struct sync_file *sync_file = NULL;
4cd09459 RC	408	int out_fence_fd = -1;
7198e6b0 RC	409	unsigned i;
	410	int ret;
	411
c01a958e RC	412	if (!gpu)
	413	return -ENXIO;
	414
7198e6b0 RC	415	/* for now, we just have 3d pipe.. eventually this would need to
	416	* be more clever to dispatch to appropriate gpu module:
	417	*/
d9c181e2 RC	418	if (MSM_PIPE_ID(args->flags) != MSM_PIPE_3D0)
	419	return -EINVAL;
	420
	421	if (MSM_PIPE_FLAGS(args->flags) & ~MSM_SUBMIT_FLAGS)
7198e6b0 RC	422	return -EINVAL;
7198e6b0 RC	423
48f243c9 RC	424	if (args->flags & MSM_SUBMIT_FENCE_FD_IN) {
	425	in_fence = sync_file_get_fence(args->fence_fd);
	426
	427	if (!in_fence)
	428	return -EINVAL;
	429
3cfac69c PZ	430	/*
	431	* Wait if the fence is from a foreign context, or if the fence
	432	* array contains any fence from a foreign context.
48f243c9	433	*/
3cfac69c	434	if (!dma_fence_match_context(in_fence, gpu->fctx->context)) {
48f243c9 RC	435	ret = dma_fence_wait(in_fence, true);
	436	if (ret)
	437	return ret;
	438	}
	439	}
	440
b5b4c264 RC	441	ret = mutex_lock_interruptible(&dev->struct_mutex);
	442	if (ret)
	443	return ret;
687f084a	444
4cd09459 RC	445	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	446	out_fence_fd = get_unused_fd_flags(O_CLOEXEC);
	447	if (out_fence_fd < 0) {
	448	ret = out_fence_fd;
	449	goto out_unlock;
	450	}
	451	}
d78d383a	452	priv->struct_mutex_task = current;
4cd09459	453
6b597ce2	454	submit = submit_create(dev, gpu, args->nr_bos, args->nr_cmds);
b5b4c264 RC	455	if (!submit) {
	456	ret = -ENOMEM;
	457	goto out_unlock;
	458	}
7198e6b0 RC	459
	460	ret = submit_lookup_objects(submit, args, file);
	461	if (ret)
	462	goto out;
	463
340faef2 RC	464	ret = submit_lock_objects(submit);
	465	if (ret)
	466	goto out;
	467
06451a3d RC	468	ret = submit_fence_sync(submit, !!(args->flags & MSM_SUBMIT_NO_IMPLICIT));
	469	if (ret)
	470	goto out;
b6295f9a	471
340faef2	472	ret = submit_pin_objects(submit);
7198e6b0 RC	473	if (ret)
	474	goto out;
	475
	476	for (i = 0; i < args->nr_cmds; i++) {
	477	struct drm_msm_gem_submit_cmd submit_cmd;
	478	void __user *userptr =
3ed605bc	479	u64_to_user_ptr(args->cmds + (i * sizeof(submit_cmd)));
7198e6b0	480	struct msm_gem_object *msm_obj;
78babc16	481	uint64_t iova;
7198e6b0 RC	482
	483	ret = copy_from_user(&submit_cmd, userptr, sizeof(submit_cmd));
	484	if (ret) {
	485	ret = -EFAULT;
	486	goto out;
	487	}
	488
93ddb0d3 RC	489	/* validate input from userspace: */
	490	switch (submit_cmd.type) {
	491	case MSM_SUBMIT_CMD_BUF:
	492	case MSM_SUBMIT_CMD_IB_TARGET_BUF:
	493	case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
	494	break;
	495	default:
	496	DRM_ERROR("invalid type: %08x\n", submit_cmd.type);
	497	ret = -EINVAL;
	498	goto out;
	499	}
	500
7198e6b0 RC	501	ret = submit_bo(submit, submit_cmd.submit_idx,
	502	&msm_obj, &iova, NULL);
	503	if (ret)
	504	goto out;
	505
	506	if (submit_cmd.size % 4) {
19872533	507	DRM_ERROR("non-aligned cmdstream buffer size: %u\n",
7198e6b0 RC	508	submit_cmd.size);
	509	ret = -EINVAL;
	510	goto out;
	511	}
	512
d72fea53 JC	513	if (!submit_cmd.size \|\|
	514	((submit_cmd.size + submit_cmd.submit_offset) >
	515	msm_obj->base.size)) {
19872533	516	DRM_ERROR("invalid cmdstream size: %u\n", submit_cmd.size);
7198e6b0 RC	517	ret = -EINVAL;
	518	goto out;
	519	}
	520
	521	submit->cmd[i].type = submit_cmd.type;
	522	submit->cmd[i].size = submit_cmd.size / 4;
	523	submit->cmd[i].iova = iova + submit_cmd.submit_offset;
a7d3c950	524	submit->cmd[i].idx = submit_cmd.submit_idx;
7198e6b0 RC	525
	526	if (submit->valid)
	527	continue;
	528
	529	ret = submit_reloc(submit, msm_obj, submit_cmd.submit_offset,
	530	submit_cmd.nr_relocs, submit_cmd.relocs);
	531	if (ret)
	532	goto out;
	533	}
	534
	535	submit->nr_cmds = i;
	536
f44d32c7 RC	537	submit->fence = msm_fence_alloc(gpu->fctx);
	538	if (IS_ERR(submit->fence)) {
	539	ret = PTR_ERR(submit->fence);
	540	submit->fence = NULL;
	541	goto out;
	542	}
	543
4cd09459 RC	544	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	545	sync_file = sync_file_create(submit->fence);
	546	if (!sync_file) {
	547	ret = -ENOMEM;
	548	goto out;
	549	}
	550	}
	551
f44d32c7	552	msm_gpu_submit(gpu, submit, ctx);
7198e6b0	553
b6295f9a	554	args->fence = submit->fence->seqno;
7198e6b0	555
4cd09459 RC	556	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	557	fd_install(out_fence_fd, sync_file->file);
	558	args->fence_fd = out_fence_fd;
	559	}
	560
7198e6b0	561	out:
f0a42bb5	562	if (in_fence)
f54d1867	563	dma_fence_put(in_fence);
40e6815b RC	564	submit_cleanup(submit);
	565	if (ret)
	566	msm_gem_submit_free(submit);
b5b4c264	567	out_unlock:
4cd09459 RC	568	if (ret && (out_fence_fd >= 0))
4cd09459 RC	569	put_unused_fd(out_fence_fd);
d78d383a	570	priv->struct_mutex_task = NULL;
c2703b13	571	mutex_unlock(&dev->struct_mutex);
7198e6b0 RC	572	return ret;
7198e6b0 RC	573	}