[linux-block.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_gem_address_space *aspace,
		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->aspace = aspace;
	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->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 = dma_resv_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->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, ctx->aspace, 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,
295b22ae JC	24	struct msm_gpu gpu, struct msm_gem_address_space aspace,
	25	struct msm_gpu_submitqueue *queue, uint32_t nr_bos,
	26	uint32_t nr_cmds)
7198e6b0 RC	27	{
7198e6b0 RC	28	struct msm_gem_submit *submit;
65e93108 DC	29	uint64_t sz = sizeof(submit) + ((u64)nr_bos sizeof(submit->bos[0])) +
65e93108 DC	30	((u64)nr_cmds * sizeof(submit->cmd[0]));
4a630fad KL	31
	32	if (sz > SIZE_MAX)
	33	return NULL;
7198e6b0	34
0ee931c4	35	submit = kmalloc(sz, GFP_KERNEL \| __GFP_NOWARN \| __GFP_NORETRY);
6860b56c RC	36	if (!submit)
6860b56c RC	37	return NULL;
7198e6b0	38
6860b56c	39	submit->dev = dev;
295b22ae	40	submit->aspace = aspace;
6860b56c	41	submit->gpu = gpu;
ba344afd	42	submit->fence = NULL;
6b597ce2	43	submit->cmd = (void *)&submit->bos[nr_bos];
f7de1545	44	submit->queue = queue;
f97decac	45	submit->ring = gpu->rb[queue->prio];
7198e6b0	46
6860b56c RC	47	/* initially, until copy_from_user() and bo lookup succeeds: */
	48	submit->nr_bos = 0;
	49	submit->nr_cmds = 0;
	50
ba344afd	51	INIT_LIST_HEAD(&submit->node);
6860b56c RC	52	INIT_LIST_HEAD(&submit->bo_list);
6860b56c RC	53	ww_acquire_init(&submit->ticket, &reservation_ww_class);
7198e6b0 RC	54
	55	return submit;
	56	}
	57
40e6815b RC	58	void msm_gem_submit_free(struct msm_gem_submit *submit)
40e6815b RC	59	{
f54d1867	60	dma_fence_put(submit->fence);
40e6815b	61	list_del(&submit->node);
4816b626	62	put_pid(submit->pid);
f7de1545 JC	63	msm_submitqueue_put(submit->queue);
f7de1545 JC	64
40e6815b RC	65	kfree(submit);
	66	}
	67
7198e6b0 RC	68	static int submit_lookup_objects(struct msm_gem_submit *submit,
	69	struct drm_msm_gem_submit args, struct drm_file file)
	70	{
	71	unsigned i;
	72	int ret = 0;
	73
7198e6b0 RC	74	for (i = 0; i < args->nr_bos; i++) {
7198e6b0 RC	75	struct drm_msm_gem_submit_bo submit_bo;
7198e6b0	76	void __user *userptr =
3ed605bc	77	u64_to_user_ptr(args->bos + (i * sizeof(submit_bo)));
7198e6b0	78
a9e26cab RC	79	/* make sure we don't have garbage flags, in case we hit
	80	* error path before flags is initialized:
	81	*/
	82	submit->bos[i].flags = 0;
	83
b673499a KK	84	if (copy_from_user(&submit_bo, userptr, sizeof(submit_bo))) {
	85	ret = -EFAULT;
	86	i = 0;
	87	goto out;
7198e6b0 RC	88	}
7198e6b0 RC	89
7a93d5c3 RC	90	/* at least one of READ and/or WRITE flags should be set: */
	91	#define MANDATORY_FLAGS (MSM_SUBMIT_BO_READ \| MSM_SUBMIT_BO_WRITE)
	92
a6cb3b86	93	if ((submit_bo.flags & ~MSM_SUBMIT_BO_FLAGS) \|\|
7a93d5c3	94	!(submit_bo.flags & MANDATORY_FLAGS)) {
19872533	95	DRM_ERROR("invalid flags: %x\n", submit_bo.flags);
7198e6b0	96	ret = -EINVAL;
b673499a KK	97	i = 0;
b673499a KK	98	goto out;
7198e6b0 RC	99	}
7198e6b0 RC	100
b673499a	101	submit->bos[i].handle = submit_bo.handle;
7198e6b0 RC	102	submit->bos[i].flags = submit_bo.flags;
	103	/* in validate_objects() we figure out if this is true: */
	104	submit->bos[i].iova = submit_bo.presumed;
b673499a KK	105	}
	106
	107	spin_lock(&file->table_lock);
	108
	109	for (i = 0; i < args->nr_bos; i++) {
	110	struct drm_gem_object *obj;
	111	struct msm_gem_object *msm_obj;
7198e6b0 RC	112
	113	/* normally use drm_gem_object_lookup(), but for bulk lookup
	114	* all under single table_lock just hit object_idr directly:
	115	*/
b673499a	116	obj = idr_find(&file->object_idr, submit->bos[i].handle);
7198e6b0	117	if (!obj) {
b673499a	118	DRM_ERROR("invalid handle %u at index %u\n", submit->bos[i].handle, i);
7198e6b0 RC	119	ret = -EINVAL;
	120	goto out_unlock;
	121	}
	122
	123	msm_obj = to_msm_bo(obj);
	124
	125	if (!list_empty(&msm_obj->submit_entry)) {
19872533	126	DRM_ERROR("handle %u at index %u already on submit list\n",
b673499a	127	submit->bos[i].handle, i);
7198e6b0 RC	128	ret = -EINVAL;
	129	goto out_unlock;
	130	}
	131
64686886	132	drm_gem_object_get(obj);
7198e6b0 RC	133
	134	submit->bos[i].obj = msm_obj;
	135
	136	list_add_tail(&msm_obj->submit_entry, &submit->bo_list);
	137	}
	138
	139	out_unlock:
7198e6b0 RC	140	spin_unlock(&file->table_lock);
7198e6b0 RC	141
89f82cbb RC	142	out:
	143	submit->nr_bos = i;
	144
7198e6b0 RC	145	return ret;
	146	}
	147
544a32db RC	148	static void submit_unlock_unpin_bo(struct msm_gem_submit *submit,
544a32db RC	149	int i, bool backoff)
7198e6b0 RC	150	{
	151	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	152
	153	if (submit->bos[i].flags & BO_PINNED)
295b22ae	154	msm_gem_unpin_iova(&msm_obj->base, submit->aspace);
7198e6b0 RC	155
7198e6b0 RC	156	if (submit->bos[i].flags & BO_LOCKED)
dd55cf69	157	ww_mutex_unlock(&msm_obj->base.resv->lock);
7198e6b0	158
544a32db	159	if (backoff && !(submit->bos[i].flags & BO_VALID))
7198e6b0 RC	160	submit->bos[i].iova = 0;
	161
	162	submit->bos[i].flags &= ~(BO_LOCKED \| BO_PINNED);
	163	}
	164
	165	/* This is where we make sure all the bo's are reserved and pin'd: */
340faef2	166	static int submit_lock_objects(struct msm_gem_submit *submit)
7198e6b0 RC	167	{
	168	int contended, slow_locked = -1, i, ret = 0;
	169
	170	retry:
7198e6b0 RC	171	for (i = 0; i < submit->nr_bos; i++) {
7198e6b0 RC	172	struct msm_gem_object *msm_obj = submit->bos[i].obj;
7198e6b0 RC	173
	174	if (slow_locked == i)
	175	slow_locked = -1;
	176
	177	contended = i;
	178
	179	if (!(submit->bos[i].flags & BO_LOCKED)) {
dd55cf69	180	ret = ww_mutex_lock_interruptible(&msm_obj->base.resv->lock,
7198e6b0 RC	181	&submit->ticket);
	182	if (ret)
	183	goto fail;
	184	submit->bos[i].flags \|= BO_LOCKED;
	185	}
7198e6b0 RC	186	}
	187
	188	ww_acquire_done(&submit->ticket);
	189
	190	return 0;
	191
	192	fail:
	193	for (; i >= 0; i--)
544a32db	194	submit_unlock_unpin_bo(submit, i, true);
7198e6b0 RC	195
7198e6b0 RC	196	if (slow_locked > 0)
544a32db	197	submit_unlock_unpin_bo(submit, slow_locked, true);
7198e6b0 RC	198
	199	if (ret == -EDEADLK) {
	200	struct msm_gem_object *msm_obj = submit->bos[contended].obj;
	201	/* we lost out in a seqno race, lock and retry.. */
dd55cf69	202	ret = ww_mutex_lock_slow_interruptible(&msm_obj->base.resv->lock,
7198e6b0 RC	203	&submit->ticket);
	204	if (!ret) {
	205	submit->bos[contended].flags \|= BO_LOCKED;
	206	slow_locked = contended;
	207	goto retry;
	208	}
	209	}
	210
	211	return ret;
	212	}
	213
ad5149c4	214	static int submit_fence_sync(struct msm_gem_submit *submit, bool no_implicit)
b6295f9a RC	215	{
	216	int i, ret = 0;
	217
	218	for (i = 0; i < submit->nr_bos; i++) {
	219	struct msm_gem_object *msm_obj = submit->bos[i].obj;
	220	bool write = submit->bos[i].flags & MSM_SUBMIT_BO_WRITE;
	221
ad5149c4 RC	222	if (!write) {
	223	/* NOTE: _reserve_shared() must happen before
	224	* _add_shared_fence(), which makes this a slightly
	225	* strange place to call it. OTOH this is a
	226	* convenient can-fail point to hook it in.
	227	*/
52791eee	228	ret = dma_resv_reserve_shared(msm_obj->base.resv,
ca05359f	229	1);
ad5149c4 RC	230	if (ret)
	231	return ret;
	232	}
	233
	234	if (no_implicit)
	235	continue;
	236
f97decac JC	237	ret = msm_gem_sync_object(&msm_obj->base, submit->ring->fctx,
f97decac JC	238	write);
b6295f9a RC	239	if (ret)
	240	break;
	241	}
	242
	243	return ret;
	244	}
	245
340faef2 RC	246	static int submit_pin_objects(struct msm_gem_submit *submit)
	247	{
	248	int i, ret = 0;
	249
	250	submit->valid = true;
	251
	252	for (i = 0; i < submit->nr_bos; i++) {
	253	struct msm_gem_object *msm_obj = submit->bos[i].obj;
78babc16	254	uint64_t iova;
340faef2 RC	255
340faef2 RC	256	/* if locking succeeded, pin bo: */
9fe041f6	257	ret = msm_gem_get_and_pin_iova(&msm_obj->base,
295b22ae	258	submit->aspace, &iova);
340faef2 RC	259
	260	if (ret)
	261	break;
	262
	263	submit->bos[i].flags \|= BO_PINNED;
	264
	265	if (iova == submit->bos[i].iova) {
	266	submit->bos[i].flags \|= BO_VALID;
	267	} else {
	268	submit->bos[i].iova = iova;
	269	/* iova changed, so address in cmdstream is not valid: */
	270	submit->bos[i].flags &= ~BO_VALID;
	271	submit->valid = false;
	272	}
	273	}
	274
	275	return ret;
	276	}
	277
7198e6b0	278	static int submit_bo(struct msm_gem_submit *submit, uint32_t idx,
78babc16	279	struct msm_gem_object *obj, uint64_t iova, bool *valid)
7198e6b0 RC	280	{
7198e6b0 RC	281	if (idx >= submit->nr_bos) {
19872533 RC	282	DRM_ERROR("invalid buffer index: %u (out of %u)\n",
	283	idx, submit->nr_bos);
	284	return -EINVAL;
7198e6b0 RC	285	}
	286
	287	if (obj)
	288	*obj = submit->bos[idx].obj;
	289	if (iova)
	290	*iova = submit->bos[idx].iova;
	291	if (valid)
	292	*valid = !!(submit->bos[idx].flags & BO_VALID);
	293
	294	return 0;
	295	}
	296
	297	/* process the reloc's and patch up the cmdstream as needed: */
	298	static int submit_reloc(struct msm_gem_submit submit, struct msm_gem_object obj,
	299	uint32_t offset, uint32_t nr_relocs, uint64_t relocs)
	300	{
	301	uint32_t i, last_offset = 0;
	302	uint32_t *ptr;
6490abc4	303	int ret = 0;
7198e6b0	304
d5290043 JC	305	if (!nr_relocs)
	306	return 0;
	307
7198e6b0	308	if (offset % 4) {
19872533	309	DRM_ERROR("non-aligned cmdstream buffer: %u\n", offset);
7198e6b0 RC	310	return -EINVAL;
	311	}
	312
	313	/* For now, just map the entire thing. Eventually we probably
	314	* to do it page-by-page, w/ kmap() if not vmap()d..
	315	*/
0e08270a	316	ptr = msm_gem_get_vaddr(&obj->base);
7198e6b0 RC	317
	318	if (IS_ERR(ptr)) {
	319	ret = PTR_ERR(ptr);
	320	DBG("failed to map: %d", ret);
	321	return ret;
	322	}
	323
	324	for (i = 0; i < nr_relocs; i++) {
	325	struct drm_msm_gem_submit_reloc submit_reloc;
	326	void __user *userptr =
3ed605bc	327	u64_to_user_ptr(relocs + (i * sizeof(submit_reloc)));
78babc16 RC	328	uint32_t off;
78babc16 RC	329	uint64_t iova;
7198e6b0 RC	330	bool valid;
7198e6b0 RC	331
21c42da1 DC	332	if (copy_from_user(&submit_reloc, userptr, sizeof(submit_reloc))) {
21c42da1 DC	333	ret = -EFAULT;
6490abc4	334	goto out;
21c42da1	335	}
7198e6b0 RC	336
7198e6b0 RC	337	if (submit_reloc.submit_offset % 4) {
19872533	338	DRM_ERROR("non-aligned reloc offset: %u\n",
7198e6b0	339	submit_reloc.submit_offset);
6490abc4 JC	340	ret = -EINVAL;
6490abc4 JC	341	goto out;
7198e6b0 RC	342	}
	343
	344	/* offset in dwords: */
	345	off = submit_reloc.submit_offset / 4;
	346
	347	if ((off >= (obj->base.size / 4)) \|\|
	348	(off < last_offset)) {
19872533	349	DRM_ERROR("invalid offset %u at reloc %u\n", off, i);
6490abc4 JC	350	ret = -EINVAL;
6490abc4 JC	351	goto out;
7198e6b0 RC	352	}
	353
	354	ret = submit_bo(submit, submit_reloc.reloc_idx, NULL, &iova, &valid);
	355	if (ret)
6490abc4	356	goto out;
7198e6b0 RC	357
	358	if (valid)
	359	continue;
	360
	361	iova += submit_reloc.reloc_offset;
	362
	363	if (submit_reloc.shift < 0)
	364	iova >>= -submit_reloc.shift;
	365	else
	366	iova <<= submit_reloc.shift;
	367
	368	ptr[off] = iova \| submit_reloc.or;
	369
	370	last_offset = off;
	371	}
	372
6490abc4	373	out:
0e08270a	374	msm_gem_put_vaddr(&obj->base);
18f23049	375
6490abc4	376	return ret;
7198e6b0 RC	377	}
7198e6b0 RC	378
40e6815b	379	static void submit_cleanup(struct msm_gem_submit *submit)
7198e6b0 RC	380	{
	381	unsigned i;
	382
7198e6b0 RC	383	for (i = 0; i < submit->nr_bos; i++) {
7198e6b0 RC	384	struct msm_gem_object *msm_obj = submit->bos[i].obj;
544a32db	385	submit_unlock_unpin_bo(submit, i, false);
7198e6b0	386	list_del_init(&msm_obj->submit_entry);
64686886	387	drm_gem_object_put(&msm_obj->base);
7198e6b0	388	}
7198e6b0 RC	389
7198e6b0 RC	390	ww_acquire_fini(&submit->ticket);
7198e6b0 RC	391	}
	392
	393	int msm_ioctl_gem_submit(struct drm_device dev, void data,
	394	struct drm_file *file)
	395	{
4241db42	396	static atomic_t ident = ATOMIC_INIT(0);
7198e6b0 RC	397	struct msm_drm_private *priv = dev->dev_private;
	398	struct drm_msm_gem_submit *args = data;
	399	struct msm_file_private *ctx = file->driver_priv;
	400	struct msm_gem_submit *submit;
c01a958e	401	struct msm_gpu *gpu = priv->gpu;
4cd09459	402	struct sync_file *sync_file = NULL;
f7de1545	403	struct msm_gpu_submitqueue *queue;
f97decac	404	struct msm_ringbuffer *ring;
4cd09459	405	int out_fence_fd = -1;
4241db42	406	struct pid *pid = get_pid(task_pid(current));
7198e6b0	407	unsigned i;
4241db42	408	int ret, submitid;
c01a958e RC	409	if (!gpu)
	410	return -ENXIO;
	411
7198e6b0 RC	412	/* for now, we just have 3d pipe.. eventually this would need to
	413	* be more clever to dispatch to appropriate gpu module:
	414	*/
d9c181e2 RC	415	if (MSM_PIPE_ID(args->flags) != MSM_PIPE_3D0)
	416	return -EINVAL;
	417
	418	if (MSM_PIPE_FLAGS(args->flags) & ~MSM_SUBMIT_FLAGS)
7198e6b0 RC	419	return -EINVAL;
7198e6b0 RC	420
6a8bd08d RC	421	if (args->flags & MSM_SUBMIT_SUDO) {
	422	if (!IS_ENABLED(CONFIG_DRM_MSM_GPU_SUDO) \|\|
	423	!capable(CAP_SYS_RAWIO))
	424	return -EINVAL;
	425	}
	426
f7de1545 JC	427	queue = msm_submitqueue_get(ctx, args->queueid);
	428	if (!queue)
	429	return -ENOENT;
	430
4241db42 JC	431	/* Get a unique identifier for the submission for logging purposes */
	432	submitid = atomic_inc_return(&ident) - 1;
	433
f97decac	434	ring = gpu->rb[queue->prio];
4241db42 JC	435	trace_msm_gpu_submit(pid_nr(pid), ring->id, submitid,
4241db42 JC	436	args->nr_bos, args->nr_cmds);
f97decac	437
48f243c9	438	if (args->flags & MSM_SUBMIT_FENCE_FD_IN) {
2189463d RF	439	struct dma_fence *in_fence;
2189463d RF	440
48f243c9 RC	441	in_fence = sync_file_get_fence(args->fence_fd);
	442
	443	if (!in_fence)
	444	return -EINVAL;
	445
3cfac69c PZ	446	/*
	447	* Wait if the fence is from a foreign context, or if the fence
	448	* array contains any fence from a foreign context.
48f243c9	449	*/
2189463d RF	450	ret = 0;
2189463d RF	451	if (!dma_fence_match_context(in_fence, ring->fctx->context))
48f243c9	452	ret = dma_fence_wait(in_fence, true);
2189463d RF	453
	454	dma_fence_put(in_fence);
	455	if (ret)
	456	return ret;
48f243c9 RC	457	}
48f243c9 RC	458
b5b4c264 RC	459	ret = mutex_lock_interruptible(&dev->struct_mutex);
	460	if (ret)
	461	return ret;
687f084a	462
4cd09459 RC	463	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	464	out_fence_fd = get_unused_fd_flags(O_CLOEXEC);
	465	if (out_fence_fd < 0) {
	466	ret = out_fence_fd;
	467	goto out_unlock;
	468	}
	469	}
	470
295b22ae JC	471	submit = submit_create(dev, gpu, ctx->aspace, queue, args->nr_bos,
295b22ae JC	472	args->nr_cmds);
b5b4c264 RC	473	if (!submit) {
	474	ret = -ENOMEM;
	475	goto out_unlock;
	476	}
7198e6b0	477
4241db42 JC	478	submit->pid = pid;
	479	submit->ident = submitid;
	480
6a8bd08d RC	481	if (args->flags & MSM_SUBMIT_SUDO)
	482	submit->in_rb = true;
	483
7198e6b0 RC	484	ret = submit_lookup_objects(submit, args, file);
	485	if (ret)
	486	goto out;
	487
340faef2 RC	488	ret = submit_lock_objects(submit);
	489	if (ret)
	490	goto out;
	491
ad5149c4 RC	492	ret = submit_fence_sync(submit, !!(args->flags & MSM_SUBMIT_NO_IMPLICIT));
	493	if (ret)
	494	goto out;
b6295f9a	495
340faef2	496	ret = submit_pin_objects(submit);
7198e6b0 RC	497	if (ret)
	498	goto out;
	499
	500	for (i = 0; i < args->nr_cmds; i++) {
	501	struct drm_msm_gem_submit_cmd submit_cmd;
	502	void __user *userptr =
3ed605bc	503	u64_to_user_ptr(args->cmds + (i * sizeof(submit_cmd)));
7198e6b0	504	struct msm_gem_object *msm_obj;
78babc16	505	uint64_t iova;
7198e6b0 RC	506
	507	ret = copy_from_user(&submit_cmd, userptr, sizeof(submit_cmd));
	508	if (ret) {
	509	ret = -EFAULT;
	510	goto out;
	511	}
	512
93ddb0d3 RC	513	/* validate input from userspace: */
	514	switch (submit_cmd.type) {
	515	case MSM_SUBMIT_CMD_BUF:
	516	case MSM_SUBMIT_CMD_IB_TARGET_BUF:
	517	case MSM_SUBMIT_CMD_CTX_RESTORE_BUF:
	518	break;
	519	default:
	520	DRM_ERROR("invalid type: %08x\n", submit_cmd.type);
	521	ret = -EINVAL;
	522	goto out;
	523	}
	524
7198e6b0 RC	525	ret = submit_bo(submit, submit_cmd.submit_idx,
	526	&msm_obj, &iova, NULL);
	527	if (ret)
	528	goto out;
	529
	530	if (submit_cmd.size % 4) {
19872533	531	DRM_ERROR("non-aligned cmdstream buffer size: %u\n",
7198e6b0 RC	532	submit_cmd.size);
	533	ret = -EINVAL;
	534	goto out;
	535	}
	536
d72fea53 JC	537	if (!submit_cmd.size \|\|
	538	((submit_cmd.size + submit_cmd.submit_offset) >
	539	msm_obj->base.size)) {
19872533	540	DRM_ERROR("invalid cmdstream size: %u\n", submit_cmd.size);
7198e6b0 RC	541	ret = -EINVAL;
	542	goto out;
	543	}
	544
	545	submit->cmd[i].type = submit_cmd.type;
	546	submit->cmd[i].size = submit_cmd.size / 4;
	547	submit->cmd[i].iova = iova + submit_cmd.submit_offset;
a7d3c950	548	submit->cmd[i].idx = submit_cmd.submit_idx;
7198e6b0 RC	549
	550	if (submit->valid)
	551	continue;
	552
	553	ret = submit_reloc(submit, msm_obj, submit_cmd.submit_offset,
	554	submit_cmd.nr_relocs, submit_cmd.relocs);
	555	if (ret)
	556	goto out;
	557	}
	558
	559	submit->nr_cmds = i;
	560
f97decac	561	submit->fence = msm_fence_alloc(ring->fctx);
f44d32c7 RC	562	if (IS_ERR(submit->fence)) {
	563	ret = PTR_ERR(submit->fence);
	564	submit->fence = NULL;
	565	goto out;
	566	}
	567
4cd09459 RC	568	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	569	sync_file = sync_file_create(submit->fence);
	570	if (!sync_file) {
	571	ret = -ENOMEM;
	572	goto out;
	573	}
	574	}
	575
f44d32c7	576	msm_gpu_submit(gpu, submit, ctx);
7198e6b0	577
b6295f9a	578	args->fence = submit->fence->seqno;
7198e6b0	579
4cd09459 RC	580	if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) {
	581	fd_install(out_fence_fd, sync_file->file);
	582	args->fence_fd = out_fence_fd;
	583	}
	584
7198e6b0	585	out:
40e6815b RC	586	submit_cleanup(submit);
	587	if (ret)
	588	msm_gem_submit_free(submit);
b5b4c264	589	out_unlock:
4cd09459 RC	590	if (ret && (out_fence_fd >= 0))
4cd09459 RC	591	put_unused_fd(out_fence_fd);
c2703b13	592	mutex_unlock(&dev->struct_mutex);
7198e6b0 RC	593	return ret;
7198e6b0 RC	594	}