amdgpu_job.c 4.9 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
/*
 * Copyright 2015 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 *
 *
 */
#include <linux/kthread.h>
#include <linux/wait.h>
#include <linux/sched.h>
#include <drm/drmP.h>
#include "amdgpu.h"
29
#include "amdgpu_trace.h"
30

31 32 33 34 35 36
static void amdgpu_job_free_handler(struct work_struct *ws)
{
	struct amdgpu_job *job = container_of(ws, struct amdgpu_job, base.work_free_job);
	kfree(job);
}

37 38 39 40 41 42 43 44 45
void amdgpu_job_timeout_func(struct work_struct *work)
{
	struct amdgpu_job *job = container_of(work, struct amdgpu_job, base.work_tdr.work);
	DRM_ERROR("ring %s timeout, last signaled seq=%u, last emitted seq=%u\n",
				job->base.sched->name,
				(uint32_t)atomic_read(&job->ring->fence_drv.last_seq),
				job->ring->fence_drv.sync_seq);
}

46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62
int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
		     struct amdgpu_job **job)
{
	size_t size = sizeof(struct amdgpu_job);

	if (num_ibs == 0)
		return -EINVAL;

	size += sizeof(struct amdgpu_ib) * num_ibs;

	*job = kzalloc(size, GFP_KERNEL);
	if (!*job)
		return -ENOMEM;

	(*job)->adev = adev;
	(*job)->ibs = (void *)&(*job)[1];
	(*job)->num_ibs = num_ibs;
63
	INIT_WORK(&(*job)->base.work_free_job, amdgpu_job_free_handler);
64

65 66
	amdgpu_sync_create(&(*job)->sync);

67 68 69
	return 0;
}

70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85
int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
			     struct amdgpu_job **job)
{
	int r;

	r = amdgpu_job_alloc(adev, 1, job);
	if (r)
		return r;

	r = amdgpu_ib_get(adev, NULL, size, &(*job)->ibs[0]);
	if (r)
		kfree(*job);

	return r;
}

86 87 88
void amdgpu_job_free(struct amdgpu_job *job)
{
	unsigned i;
89 90 91
	struct fence *f;
	/* use sched fence if available */
	f = (job->base.s_fence)? &job->base.s_fence->base : job->fence;
92 93

	for (i = 0; i < job->num_ibs; ++i)
94
		amdgpu_sa_bo_free(job->adev, &job->ibs[i].sa_bo, f);
95
	fence_put(job->fence);
96 97

	amdgpu_bo_unref(&job->uf.bo);
98
	amdgpu_sync_free(&job->sync);
99 100 101

	if (!job->base.use_sched)
		kfree(job);
102 103 104
}

int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
105 106
		      struct amd_sched_entity *entity, void *owner,
		      struct fence **f)
107
{
108 109
	struct fence *fence;
	int r;
110 111
	job->ring = ring;

112 113 114
	if (!f)
		return -EINVAL;

115 116 117 118
	r = amd_sched_job_init(&job->base, &ring->sched,
							entity, owner,
							amdgpu_job_timeout_func,
							&fence);
119 120
	if (r)
		return r;
121 122

	job->owner = owner;
123
	*f = fence_get(fence);
124 125 126
	amd_sched_entity_push_job(&job->base);

	return 0;
127 128
}

129
static struct fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
130
{
131
	struct amdgpu_job *job = to_amdgpu_job(sched_job);
132 133
	struct amdgpu_vm *vm = job->ibs->vm;

134
	struct fence *fence = amdgpu_sync_get_fence(&job->sync);
135

136
	if (fence == NULL && vm && !job->ibs->vm_id) {
137
		struct amdgpu_ring *ring = job->ring;
138 139
		unsigned i, vm_id;
		uint64_t vm_pd_addr;
140 141
		int r;

142
		r = amdgpu_vm_grab_id(vm, ring, &job->sync,
143 144
				      &job->base.s_fence->base,
				      &vm_id, &vm_pd_addr);
145
		if (r)
146
			DRM_ERROR("Error getting VM ID (%d)\n", r);
147 148 149 150 151 152
		else {
			for (i = 0; i < job->num_ibs; ++i) {
				job->ibs[i].vm_id = vm_id;
				job->ibs[i].vm_pd_addr = vm_pd_addr;
			}
		}
153

154
		fence = amdgpu_sync_get_fence(&job->sync);
155 156 157
	}

	return fence;
158 159
}

160
static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job)
161
{
162
	struct fence *fence = NULL;
163
	struct amdgpu_job *job;
164
	int r;
165

166
	if (!sched_job) {
167
		DRM_ERROR("job is null\n");
168
		return NULL;
169
	}
170
	job = to_amdgpu_job(sched_job);
171 172 173 174 175 176 177

	r = amdgpu_sync_wait(&job->sync);
	if (r) {
		DRM_ERROR("failed to sync wait (%d)\n", r);
		return NULL;
	}

178
	trace_amdgpu_sched_run_job(job);
179
	r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs,
180
			       job->sync.last_vm_update, &fence);
181 182
	if (r) {
		DRM_ERROR("Error scheduling IBs (%d)\n", r);
183
		goto err;
184 185 186
	}

err:
187
	job->fence = fence;
188
	amdgpu_job_free(job);
189
	return fence;
190 191 192
}

struct amd_sched_backend_ops amdgpu_sched_ops = {
193 194
	.dependency = amdgpu_job_dependency,
	.run_job = amdgpu_job_run,
195 196
	.begin_job = amd_sched_job_begin,
	.finish_job = amd_sched_job_finish,
197
};