[thirdparty/linux.git] / drivers / gpu / drm / msm / msm_gem_submit.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * Copyright (C) 2013 Red Hat
 * Author: Rob Clark <robdclark@gmail.com>
 */

#include <linux/sync_file.h>

#include "msm_drv.h"
#include "msm_gpu.h"
#include "msm_gem.h"
#include "msm_gpu_trace.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, struct msm_gpu_submitqueue *queue,
		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->cmd = (void *)&submit->bos[nr_bos];
	submit->queue = queue;
	submit->ring = gpu->rb[queue->prio];

	/* 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);
	msm_submitqueue_put(submit->queue);

	kfree(submit);
}

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;

	for (i = 0; i < args->nr_bos; i++) {
		struct drm_msm_gem_submit_bo submit_bo;
		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(&submit_bo, userptr, sizeof(submit_bo))) {
			ret = -EFAULT;
			i = 0;
			goto out;
		}

/* at least one of READ and/or WRITE flags should be set: */
#define MANDATORY_FLAGS (MSM_SUBMIT_BO_READ | MSM_SUBMIT_BO_WRITE)

		if ((submit_bo.flags & ~MSM_SUBMIT_BO_FLAGS) ||
			!(submit_bo.flags & MANDATORY_FLAGS)) {
			DRM_ERROR("invalid flags: %x\n", submit_bo.flags);
			ret = -EINVAL;
			i = 0;
			goto out;
		}

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

	spin_lock(&file->table_lock);

	for (i = 0; i < args->nr_bos; i++) {
		struct drm_gem_object *obj;
		struct msm_gem_object *msm_obj;

		/* 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->bos[i].handle);
		if (!obj) {
			DRM_ERROR("invalid handle %u at index %u\n", submit->bos[i].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->bos[i].handle, i);
			ret = -EINVAL;
			goto out_unlock;
		}

		drm_gem_object_get(obj);

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

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

out_unlock:
	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, bool backoff)
{
	struct msm_gem_object *msm_obj = submit->bos[i].obj;

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

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

	if (backoff && !(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->base.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, true);

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

	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->base.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->base.resv,
								1);
			if (ret)
				return ret;
		}

		if (no_implicit)
			continue;

		ret = msm_gem_sync_object(&msm_obj->base, submit->ring->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_and_pin_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 (!nr_relocs)
		return 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, false);
		list_del_init(&msm_obj->submit_entry);
		drm_gem_object_put(&msm_obj->base);
	}

	ww_acquire_fini(&submit->ticket);
}

int msm_ioctl_gem_submit(struct drm_device *dev, void *data,
		struct drm_file *file)
{
	static atomic_t ident = ATOMIC_INIT(0);
	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 sync_file *sync_file = NULL;
	struct msm_gpu_submitqueue *queue;
	struct msm_ringbuffer *ring;
	int out_fence_fd = -1;
	struct pid *pid = get_pid(task_pid(current));
	unsigned i;
	int ret, submitid;
	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_SUDO) {
		if (!IS_ENABLED(CONFIG_DRM_MSM_GPU_SUDO) ||
		    !capable(CAP_SYS_RAWIO))
			return -EINVAL;
	}

	queue = msm_submitqueue_get(ctx, args->queueid);
	if (!queue)
		return -ENOENT;

	/* Get a unique identifier for the submission for logging purposes */
	submitid = atomic_inc_return(&ident) - 1;

	ring = gpu->rb[queue->prio];
	trace_msm_gpu_submit(pid_nr(pid), ring->id, submitid,
		args->nr_bos, args->nr_cmds);

	if (args->flags & MSM_SUBMIT_FENCE_FD_IN) {
		struct dma_fence *in_fence;

		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.
		 */
		ret = 0;
		if (!dma_fence_match_context(in_fence, ring->fctx->context))
			ret = dma_fence_wait(in_fence, true);

		dma_fence_put(in_fence);
		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;
		}
	}

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

	submit->pid = pid;
	submit->ident = submitid;

	if (args->flags & MSM_SUBMIT_SUDO)
		submit->in_rb = true;

	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(ring->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:
	submit_cleanup(submit);
	if (ret)
		msm_gem_submit_free(submit);
out_unlock:
	if (ret && (out_fence_fd >= 0))
		put_unused_fd(out_fence_fd);
	mutex_unlock(&dev->struct_mutex);
	return ret;
}
Commit	Line	Data
caab277b	1	// SPDX-License-Identifier: GPL-2.0-only
7198e6b0 RC	2	/*
	3	* Copyright (C) 2013 Red Hat
	4	* Author: Rob Clark <robdclark@gmail.com>
7198e6b0 RC	5	*/
7198e6b0 RC	6
f0a42bb5 RC	7	#include <linux/sync_file.h>
f0a42bb5 RC	8
7198e6b0 RC	9	#include "msm_drv.h"
	10	#include "msm_gpu.h"
	11	#include "msm_gem.h"
4241db42	12	#include "msm_gpu_trace.h"
7198e6b0 RC	13
	14	/*
	15	* Cmdstream submission:
	16	*/
	17
7198e6b0	18	/* make sure these don't conflict w/ MSM_SUBMIT_BO_x */
340faef2	19	#define BO_VALID 0x8000 /* is current addr in cmdstream correct/valid? */
7198e6b0 RC	20	#define BO_LOCKED 0x4000
	21	#define BO_PINNED 0x2000
	22
7198e6b0	23	static struct msm_gem_submit submit_create(struct drm_device dev,
f7de1545 JC	24	struct msm_gpu gpu, struct msm_gpu_submitqueue queue,
f7de1545 JC	25	uint32_t nr_bos, uint32_t nr_cmds)
7198e6b0 RC	26	{
7198e6b0 RC	27	struct msm_gem_submit *submit;
65e93108 DC	28	uint64_t sz = sizeof(submit) + ((u64)nr_bos sizeof(submit->bos[0])) +
65e93108 DC	29	((u64)nr_cmds * sizeof(submit->cmd[0]));
4a630fad KL	30
	31	if (sz > SIZE_MAX)
	32	return NULL;
7198e6b0	33
0ee931c4	34	submit = kmalloc(sz, GFP_KERNEL \| __GFP_NOWARN \| __GFP_NORETRY);
6860b56c RC	35	if (!submit)
6860b56c RC	36	return NULL;
7198e6b0	37
6860b56c RC	38	submit->dev = dev;
6860b56c RC	39	submit->gpu = gpu;
ba344afd	40	submit->fence = NULL;
6b597ce2	41	submit->cmd = (void *)&submit->bos[nr_bos];
f7de1545	42	submit->queue = queue;
f97decac	43	submit->ring = gpu->rb[queue->prio];
7198e6b0	44
6860b56c RC	45	/* initially, until copy_from_user() and bo lookup succeeds: */
	46	submit->nr_bos = 0;
	47	submit->nr_cmds = 0;
	48
ba344afd	49	INIT_LIST_HEAD(&submit->node);
6860b56c RC	50	INIT_LIST_HEAD(&submit->bo_list);
6860b56c RC	51	ww_acquire_init(&submit->ticket, &reservation_ww_class);
7198e6b0 RC	52
	53	return submit;
	54	}
	55
40e6815b RC	56	void msm_gem_submit_free(struct msm_gem_submit *submit)
40e6815b RC	57	{
f54d1867	58	dma_fence_put(submit->fence);
40e6815b	59	list_del(&submit->node);
4816b626	60	put_pid(submit->pid);
f7de1545 JC	61	msm_submitqueue_put(submit->queue);
f7de1545 JC	62
40e6815b RC	63	kfree(submit);
	64	}
	65
7198e6b0 RC	66	static int submit_lookup_objects(struct msm_gem_submit *submit,
	67	struct drm_msm_gem_submit args, struct drm_file file)
	68	{
	69	unsigned i;
	70	int ret = 0;
	71
7198e6b0 RC	72	for (i = 0; i < args->nr_bos; i++) {
7198e6b0 RC	73	struct drm_msm_gem_submit_bo submit_bo;
7198e6b0	74	void __user *userptr =
3ed605bc	75	u64_to_user_ptr(args->bos + (i * sizeof(submit_bo)));
7198e6b0	76
a9e26cab RC	77	/* make sure we don't have garbage flags, in case we hit
	78	* error path before flags is initialized:
	79	*/
	80	submit->bos[i].flags = 0;
	81
b673499a KK	82	if (copy_from_user(&submit_bo, userptr, sizeof(submit_bo))) {
	83	ret = -EFAULT;
	84	i = 0;
	85	goto out;
7198e6b0 RC	86	}
7198e6b0 RC	87
7a93d5c3 RC	88	/* at least one of READ and/or WRITE flags should be set: */
	89	#define MANDATORY_FLAGS (MSM_SUBMIT_BO_READ \| MSM_SUBMIT_BO_WRITE)
	90
a6cb3b86	91	if ((submit_bo.flags & ~MSM_SUBMIT_BO_FLAGS) \|\|
7a93d5c3	92	!(submit_bo.flags & MANDATORY_FLAGS)) {
19872533	93	DRM_ERROR("invalid flags: %x\n", submit_bo.flags);
7198e6b0	94	ret = -EINVAL;
b673499a KK	95	i = 0;
b673499a KK	96	goto out;
7198e6b0 RC	97	}
7198e6b0 RC	98
b673499a	99	submit->bos[i].handle = submit_bo.handle;
7198e6b0 RC	100	submit->bos[i].flags = submit_bo.flags;
	101	/* in validate_objects() we figure out if this is true: */
	102	submit->bos[i].iova = submit_bo.presumed;
b673499a KK	103	}
	104
	105	spin_lock(&file->table_lock);
	106
	107	for (i = 0; i < args->nr_bos; i++) {
	108	struct drm_gem_object *obj;
	109	struct msm_gem_object *msm_obj;
7198e6b0 RC	110
	111	/* normally use drm_gem_object_lookup(), but for bulk lookup
	112	* all under single table_lock just hit object_idr directly:
	113	*/
b673499a	114	obj = idr_find(&file->object_idr, submit->bos[i].handle);
7198e6b0	115	if (!obj) {
b673499a	116	DRM_ERROR("invalid handle %u at index %u\n", submit->bos[i].handle, i);
7198e6b0 RC	117	ret = -EINVAL;
	118	goto out_unlock;
	119	}
	120
	121	msm_obj = to_msm_bo(obj);
	122
	123	if (!list_empty(&msm_obj->submit_entry)) {
19872533	124	DRM_ERROR("handle %u at index %u already on submit list\n",
b673499a	125	submit->bos[i].handle, i);
7198e6b0 RC	126	ret = -EINVAL;
	127	goto out_unlock;
	128	}
	129
64686886	130	drm_gem_object_get(obj);
7198e6b0 RC	131
	132	submit->bos[i].obj = msm_obj;
	133
	134	list_add_tail(&msm_obj->submit_entry, &submit->bo_list);
	135	}
	136
	137	out_unlock:
7198e6b0 RC	138	spin_unlock(&file->table_lock);
7198e6b0 RC	139
89f82cbb RC	140	out:
	141	submit->nr_bos = i;
	142
7198e6b0 RC	143	return ret;
	144	}
	145
544a32db RC	146	static void submit_unlock_unpin_bo(struct msm_gem_submit *submit,
544a32db RC	147	int i, bool backoff)
7198e6b0 RC	148	{
	149	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	150
	151	if (submit->bos[i].flags & BO_PINNED)
7ad0e8cf	152	msm_gem_unpin_iova(&msm_obj->base, submit->gpu->aspace);
7198e6b0 RC	153
7198e6b0 RC	154	if (submit->bos[i].flags & BO_LOCKED)
dd55cf69	155	ww_mutex_unlock(&msm_obj->base.resv->lock);
7198e6b0	156
544a32db	157	if (backoff && !(submit->bos[i].flags & BO_VALID))
7198e6b0 RC	158	submit->bos[i].iova = 0;
	159
	160	submit->bos[i].flags &= ~(BO_LOCKED \| BO_PINNED);
	161	}
	162
	163	/* This is where we make sure all the bo's are reserved and pin'd: */
340faef2	164	static int submit_lock_objects(struct msm_gem_submit *submit)
7198e6b0 RC	165	{
	166	int contended, slow_locked = -1, i, ret = 0;
	167
	168	retry:
7198e6b0 RC	169	for (i = 0; i < submit->nr_bos; i++) {
7198e6b0 RC	170	struct msm_gem_object *msm_obj = submit->bos[i].obj;
7198e6b0 RC	171
	172	if (slow_locked == i)
	173	slow_locked = -1;
	174
	175	contended = i;
	176
	177	if (!(submit->bos[i].flags & BO_LOCKED)) {
dd55cf69	178	ret = ww_mutex_lock_interruptible(&msm_obj->base.resv->lock,
7198e6b0 RC	179	&submit->ticket);
	180	if (ret)
	181	goto fail;
	182	submit->bos[i].flags \|= BO_LOCKED;
	183	}
7198e6b0 RC	184	}
	185
	186	ww_acquire_done(&submit->ticket);
	187
	188	return 0;
	189
	190	fail:
	191	for (; i >= 0; i--)
544a32db	192	submit_unlock_unpin_bo(submit, i, true);
7198e6b0 RC	193
7198e6b0 RC	194	if (slow_locked > 0)
544a32db	195	submit_unlock_unpin_bo(submit, slow_locked, true);
7198e6b0 RC	196
	197	if (ret == -EDEADLK) {
	198	struct msm_gem_object *msm_obj = submit->bos[contended].obj;
	199	/* we lost out in a seqno race, lock and retry.. */
dd55cf69	200	ret = ww_mutex_lock_slow_interruptible(&msm_obj->base.resv->lock,
7198e6b0 RC	201	&submit->ticket);
	202	if (!ret) {
	203	submit->bos[contended].flags \|= BO_LOCKED;
	204	slow_locked = contended;
	205	goto retry;
	206	}
	207	}
	208
	209	return ret;
	210	}
	211
ad5149c4	212	static int submit_fence_sync(struct msm_gem_submit *submit, bool no_implicit)
b6295f9a RC	213	{
	214	int i, ret = 0;
	215
	216	for (i = 0; i < submit->nr_bos; i++) {
	217	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	218	bool write = submit->bos[i].flags & MSM_SUBMIT_BO_WRITE;
	219
ad5149c4 RC	220	if (!write) {
	221	/* NOTE: _reserve_shared() must happen before
	222	* _add_shared_fence(), which makes this a slightly
	223	* strange place to call it. OTOH this is a
	224	* convenient can-fail point to hook it in.
	225	*/
dd55cf69	226	ret = reservation_object_reserve_shared(msm_obj->base.resv,
ca05359f	227	1);
ad5149c4 RC	228	if (ret)
	229	return ret;
	230	}
	231
	232	if (no_implicit)
	233	continue;
	234
f97decac JC	235	ret = msm_gem_sync_object(&msm_obj->base, submit->ring->fctx,
f97decac JC	236	write);
b6295f9a RC	237	if (ret)
	238	break;
	239	}
	240
	241	return ret;
	242	}
	243
340faef2 RC	244	static int submit_pin_objects(struct msm_gem_submit *submit)
	245	{
	246	int i, ret = 0;
	247
	248	submit->valid = true;
	249
	250	for (i = 0; i < submit->nr_bos; i++) {
	251	struct msm_gem_object *msm_obj = submit->bos[i].obj;
78babc16	252	uint64_t iova;
340faef2 RC	253
340faef2 RC	254	/* if locking succeeded, pin bo: */
9fe041f6	255	ret = msm_gem_get_and_pin_iova(&msm_obj->base,
8bdcd949	256	submit->gpu->aspace, &iova);
340faef2 RC	257
	258	if (ret)
	259	break;
	260
	261	submit->bos[i].flags \|= BO_PINNED;
	262
	263	if (iova == submit->bos[i].iova) {
	264	submit->bos[i].flags \|= BO_VALID;
	265	} else {
	266	submit->bos[i].iova = iova;
	267	/* iova changed, so address in cmdstream is not valid: */
	268	submit->bos[i].flags &= ~BO_VALID;
	269	submit->valid = false;
	270	}
	271	}
	272
	273	return ret;
	274	}
	275
7198e6b0	276	static int submit_bo(struct msm_gem_submit *submit, uint32_t idx,
78babc16	277	struct msm_gem_object *obj, uint64_t iova, bool *valid)
7198e6b0 RC	278	{
7198e6b0 RC	279	if (idx >= submit->nr_bos) {
19872533 RC	280	DRM_ERROR("invalid buffer index: %u (out of %u)\n",
	281	idx, submit->nr_bos);
	282	return -EINVAL;
7198e6b0 RC	283	}
	284
	285	if (obj)
	286	*obj = submit->bos[idx].obj;
	287	if (iova)
	288	*iova = submit->bos[idx].iova;
	289	if (valid)
	290	*valid = !!(submit->bos[idx].flags & BO_VALID);
	291
	292	return 0;
	293	}
	294
	295	/* process the reloc's and patch up the cmdstream as needed: */
	296	static int submit_reloc(struct msm_gem_submit submit, struct msm_gem_object obj,
	297	uint32_t offset, uint32_t nr_relocs, uint64_t relocs)
	298	{
	299	uint32_t i, last_offset = 0;
	300	uint32_t *ptr;
6490abc4	301	int ret = 0;
7198e6b0	302
d5290043 JC	303	if (!nr_relocs)
	304	return 0;
	305
7198e6b0	306	if (offset % 4) {
19872533	307	DRM_ERROR("non-aligned cmdstream buffer: %u\n", offset);
7198e6b0 RC	308	return -EINVAL;
	309	}
	310
	311	/* For now, just map the entire thing. Eventually we probably
	312	* to do it page-by-page, w/ kmap() if not vmap()d..
	313	*/
0e08270a	314	ptr = msm_gem_get_vaddr(&obj->base);
7198e6b0 RC	315
	316	if (IS_ERR(ptr)) {
	317	ret = PTR_ERR(ptr);
	318	DBG("failed to map: %d", ret);
	319	return ret;
	320	}
	321
	322	for (i = 0; i < nr_relocs; i++) {
	323	struct drm_msm_gem_submit_reloc submit_reloc;
	324	void __user *userptr =
3ed605bc	325	u64_to_user_ptr(relocs + (i * sizeof(submit_reloc)));
78babc16 RC	326	uint32_t off;
78babc16 RC	327	uint64_t iova;
7198e6b0 RC	328	bool valid;
7198e6b0 RC	329
21c42da1 DC	330	if (copy_from_user(&submit_reloc, userptr, sizeof(submit_reloc))) {
21c42da1 DC	331	ret = -EFAULT;
6490abc4	332	goto out;
21c42da1	333	}
7198e6b0 RC	334
7198e6b0 RC	335	if (submit_reloc.submit_offset % 4) {
19872533	336	DRM_ERROR("non-aligned reloc offset: %u\n",
7198e6b0	337	submit_reloc.submit_offset);
6490abc4 JC	338	ret = -EINVAL;
6490abc4 JC	339	goto out;
7198e6b0 RC	340	}
	341
	342	/* offset in dwords: */
	343	off = submit_reloc.submit_offset / 4;
	344
	345	if ((off >= (obj->base.size / 4)) \|\|
	346	(off < last_offset)) {
19872533	347	DRM_ERROR("invalid offset %u at reloc %u\n", off, i);
6490abc4 JC	348	ret = -EINVAL;
6490abc4 JC	349	goto out;
7198e6b0 RC	350	}
	351
	352	ret = submit_bo(submit, submit_reloc.reloc_idx, NULL, &iova, &valid);
	353	if (ret)
6490abc4	354	goto out;
7198e6b0 RC	355
	356	if (valid)
	357	continue;
	358
	359	iova += submit_reloc.reloc_offset;
	360
	361	if (submit_reloc.shift < 0)
	362	iova >>= -submit_reloc.shift;
	363	else
	364	iova <<= submit_reloc.shift;
	365
	366	ptr[off] = iova \| submit_reloc.or;
	367
	368	last_offset = off;
	369	}
	370
6490abc4	371	out:
0e08270a	372	msm_gem_put_vaddr(&obj->base);
18f23049	373
6490abc4	374	return ret;
7198e6b0 RC	375	}
7198e6b0 RC	376
40e6815b	377	static void submit_cleanup(struct msm_gem_submit *submit)
7198e6b0 RC	378	{
	379	unsigned i;
	380
7198e6b0 RC	381	for (i = 0; i < submit->nr_bos; i++) {
7198e6b0 RC	382	struct msm_gem_object *msm_obj = submit->bos[i].obj;
544a32db	383	submit_unlock_unpin_bo(submit, i, false);
7198e6b0	384	list_del_init(&msm_obj->submit_entry);
64686886	385	drm_gem_object_put(&msm_obj->base);
7198e6b0	386	}
7198e6b0 RC	387
7198e6b0 RC	388	ww_acquire_fini(&submit->ticket);
7198e6b0 RC	389	}
	390
	391	int msm_ioctl_gem_submit(struct drm_device dev, void data,
	392	struct drm_file *file)
	393	{
4241db42	394	static atomic_t ident = ATOMIC_INIT(0);
7198e6b0 RC	395	struct msm_drm_private *priv = dev->dev_private;
	396	struct drm_msm_gem_submit *args = data;
	397	struct msm_file_private *ctx = file->driver_priv;
	398	struct msm_gem_submit *submit;
c01a958e	399	struct msm_gpu *gpu = priv->gpu;
4cd09459	400	struct sync_file *sync_file = NULL;
f7de1545	401	struct msm_gpu_submitqueue *queue;
f97decac	402	struct msm_ringbuffer *ring;
4cd09459	403	int out_fence_fd = -1;
4241db42	404	struct pid *pid = get_pid(task_pid(current));
7198e6b0	405	unsigned i;
4241db42	406	int ret, submitid;
c01a958e RC	407	if (!gpu)
	408	return -ENXIO;
	409
7198e6b0 RC	410	/* for now, we just have 3d pipe.. eventually this would need to
	411	* be more clever to dispatch to appropriate gpu module:
	412	*/
d9c181e2 RC	413	if (MSM_PIPE_ID(args->flags) != MSM_PIPE_3D0)
	414	return -EINVAL;
	415
	416	if (MSM_PIPE_FLAGS(args->flags) & ~MSM_SUBMIT_FLAGS)
7198e6b0 RC	417	return -EINVAL;
7198e6b0 RC	418
6a8bd08d RC	419	if (args->flags & MSM_SUBMIT_SUDO) {
	420	if (!IS_ENABLED(CONFIG_DRM_MSM_GPU_SUDO) \|\|
	421	!capable(CAP_SYS_RAWIO))
	422	return -EINVAL;
	423	}
	424
f7de1545 JC	425	queue = msm_submitqueue_get(ctx, args->queueid);
	426	if (!queue)
	427	return -ENOENT;
	428
4241db42 JC	429	/* Get a unique identifier for the submission for logging purposes */
	430	submitid = atomic_inc_return(&ident) - 1;
	431
f97decac	432	ring = gpu->rb[queue->prio];
4241db42 JC	433	trace_msm_gpu_submit(pid_nr(pid), ring->id, submitid,
4241db42 JC	434	args->nr_bos, args->nr_cmds);
f97decac	435
48f243c9	436	if (args->flags & MSM_SUBMIT_FENCE_FD_IN) {
2189463d RF	437	struct dma_fence *in_fence;
2189463d RF	438
48f243c9 RC	439	in_fence = sync_file_get_fence(args->fence_fd);
	440
	441	if (!in_fence)
	442	return -EINVAL;
	443
3cfac69c PZ	444	/*
	445	* Wait if the fence is from a foreign context, or if the fence
	446	* array contains any fence from a foreign context.
48f243c9	447	*/
2189463d RF	448	ret = 0;
2189463d RF	449	if (!dma_fence_match_context(in_fence, ring->fctx->context))
48f243c9	450	ret = dma_fence_wait(in_fence, true);
2189463d RF	451
	452	dma_fence_put(in_fence);
	453	if (ret)
	454	return ret;
48f243c9 RC	455	}
48f243c9 RC	456
b5b4c264 RC	457	ret = mutex_lock_interruptible(&dev->struct_mutex);
	458	if (ret)
	459	return ret;
687f084a	460
4cd09459 RC	461	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	462	out_fence_fd = get_unused_fd_flags(O_CLOEXEC);
	463	if (out_fence_fd < 0) {
	464	ret = out_fence_fd;
	465	goto out_unlock;
	466	}
	467	}
	468
f7de1545	469	submit = submit_create(dev, gpu, queue, args->nr_bos, args->nr_cmds);
b5b4c264 RC	470	if (!submit) {
	471	ret = -ENOMEM;
	472	goto out_unlock;
	473	}
7198e6b0	474
4241db42 JC	475	submit->pid = pid;
	476	submit->ident = submitid;
	477
6a8bd08d RC	478	if (args->flags & MSM_SUBMIT_SUDO)
	479	submit->in_rb = true;
	480
7198e6b0 RC	481	ret = submit_lookup_objects(submit, args, file);
	482	if (ret)
	483	goto out;
	484
340faef2 RC	485	ret = submit_lock_objects(submit);
	486	if (ret)
	487	goto out;
	488
ad5149c4 RC	489	ret = submit_fence_sync(submit, !!(args->flags & MSM_SUBMIT_NO_IMPLICIT));
	490	if (ret)
	491	goto out;
b6295f9a	492
340faef2	493	ret = submit_pin_objects(submit);
7198e6b0 RC	494	if (ret)
	495	goto out;
	496
	497	for (i = 0; i < args->nr_cmds; i++) {
	498	struct drm_msm_gem_submit_cmd submit_cmd;
	499	void __user *userptr =
3ed605bc	500	u64_to_user_ptr(args->cmds + (i * sizeof(submit_cmd)));
7198e6b0	501	struct msm_gem_object *msm_obj;
78babc16	502	uint64_t iova;
7198e6b0 RC	503
	504	ret = copy_from_user(&submit_cmd, userptr, sizeof(submit_cmd));
	505	if (ret) {
	506	ret = -EFAULT;
	507	goto out;
	508	}
	509
93ddb0d3 RC	510	/* validate input from userspace: */
	511	switch (submit_cmd.type) {
	512	case MSM_SUBMIT_CMD_BUF:
	513	case MSM_SUBMIT_CMD_IB_TARGET_BUF:
	514	case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
	515	break;
	516	default:
	517	DRM_ERROR("invalid type: %08x\n", submit_cmd.type);
	518	ret = -EINVAL;
	519	goto out;
	520	}
	521
7198e6b0 RC	522	ret = submit_bo(submit, submit_cmd.submit_idx,
	523	&msm_obj, &iova, NULL);
	524	if (ret)
	525	goto out;
	526
	527	if (submit_cmd.size % 4) {
19872533	528	DRM_ERROR("non-aligned cmdstream buffer size: %u\n",
7198e6b0 RC	529	submit_cmd.size);
	530	ret = -EINVAL;
	531	goto out;
	532	}
	533
d72fea53 JC	534	if (!submit_cmd.size \|\|
	535	((submit_cmd.size + submit_cmd.submit_offset) >
	536	msm_obj->base.size)) {
19872533	537	DRM_ERROR("invalid cmdstream size: %u\n", submit_cmd.size);
7198e6b0 RC	538	ret = -EINVAL;
	539	goto out;
	540	}
	541
	542	submit->cmd[i].type = submit_cmd.type;
	543	submit->cmd[i].size = submit_cmd.size / 4;
	544	submit->cmd[i].iova = iova + submit_cmd.submit_offset;
a7d3c950	545	submit->cmd[i].idx = submit_cmd.submit_idx;
7198e6b0 RC	546
	547	if (submit->valid)
	548	continue;
	549
	550	ret = submit_reloc(submit, msm_obj, submit_cmd.submit_offset,
	551	submit_cmd.nr_relocs, submit_cmd.relocs);
	552	if (ret)
	553	goto out;
	554	}
	555
	556	submit->nr_cmds = i;
	557
f97decac	558	submit->fence = msm_fence_alloc(ring->fctx);
f44d32c7 RC	559	if (IS_ERR(submit->fence)) {
	560	ret = PTR_ERR(submit->fence);
	561	submit->fence = NULL;
	562	goto out;
	563	}
	564
4cd09459 RC	565	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	566	sync_file = sync_file_create(submit->fence);
	567	if (!sync_file) {
	568	ret = -ENOMEM;
	569	goto out;
	570	}
	571	}
	572
f44d32c7	573	msm_gpu_submit(gpu, submit, ctx);
7198e6b0	574
b6295f9a	575	args->fence = submit->fence->seqno;
7198e6b0	576
4cd09459 RC	577	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	578	fd_install(out_fence_fd, sync_file->file);
	579	args->fence_fd = out_fence_fd;
	580	}
	581
7198e6b0	582	out:
40e6815b RC	583	submit_cleanup(submit);
	584	if (ret)
	585	msm_gem_submit_free(submit);
b5b4c264	586	out_unlock:
4cd09459 RC	587	if (ret && (out_fence_fd >= 0))
4cd09459 RC	588	put_unused_fd(out_fence_fd);
c2703b13	589	mutex_unlock(&dev->struct_mutex);
7198e6b0 RC	590	return ret;
7198e6b0 RC	591	}