xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
1 /*
2  * Copyright 2015 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  *
23  */
24 #include <linux/kthread.h>
25 #include <linux/wait.h>
26 #include <linux/sched.h>
27 #include <drm/drmP.h>
28 #include "amdgpu.h"
29 #include "amdgpu_trace.h"
30 
31 static void amdgpu_job_timedout(struct amd_sched_job *s_job)
32 {
33 	struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
34 
35 	DRM_ERROR("ring %s timeout, last signaled seq=%u, last emitted seq=%u\n",
36 		  job->base.sched->name,
37 		  atomic_read(&job->ring->fence_drv.last_seq),
38 		  job->ring->fence_drv.sync_seq);
39 	amdgpu_gpu_reset(job->adev);
40 }
41 
42 int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
43 		     struct amdgpu_job **job, struct amdgpu_vm *vm)
44 {
45 	size_t size = sizeof(struct amdgpu_job);
46 
47 	if (num_ibs == 0)
48 		return -EINVAL;
49 
50 	size += sizeof(struct amdgpu_ib) * num_ibs;
51 
52 	*job = kzalloc(size, GFP_KERNEL);
53 	if (!*job)
54 		return -ENOMEM;
55 
56 	(*job)->adev = adev;
57 	(*job)->vm = vm;
58 	(*job)->ibs = (void *)&(*job)[1];
59 	(*job)->num_ibs = num_ibs;
60 	(*job)->need_pipeline_sync = false;
61 
62 	amdgpu_sync_create(&(*job)->sync);
63 
64 	return 0;
65 }
66 
67 int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
68 			     struct amdgpu_job **job)
69 {
70 	int r;
71 
72 	r = amdgpu_job_alloc(adev, 1, job, NULL);
73 	if (r)
74 		return r;
75 
76 	r = amdgpu_ib_get(adev, NULL, size, &(*job)->ibs[0]);
77 	if (r)
78 		kfree(*job);
79 
80 	return r;
81 }
82 
83 void amdgpu_job_free_resources(struct amdgpu_job *job)
84 {
85 	struct dma_fence *f;
86 	unsigned i;
87 
88 	/* use sched fence if available */
89 	f = job->base.s_fence ? &job->base.s_fence->finished : job->fence;
90 
91 	for (i = 0; i < job->num_ibs; ++i)
92 		amdgpu_ib_free(job->adev, &job->ibs[i], f);
93 }
94 
95 static void amdgpu_job_free_cb(struct amd_sched_job *s_job)
96 {
97 	struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
98 
99 	dma_fence_put(job->fence);
100 	amdgpu_sync_free(&job->sync);
101 	kfree(job);
102 }
103 
104 void amdgpu_job_free(struct amdgpu_job *job)
105 {
106 	amdgpu_job_free_resources(job);
107 
108 	dma_fence_put(job->fence);
109 	amdgpu_sync_free(&job->sync);
110 	kfree(job);
111 }
112 
113 int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
114 		      struct amd_sched_entity *entity, void *owner,
115 		      struct dma_fence **f)
116 {
117 	int r;
118 	job->ring = ring;
119 
120 	if (!f)
121 		return -EINVAL;
122 
123 	r = amd_sched_job_init(&job->base, &ring->sched, entity, owner);
124 	if (r)
125 		return r;
126 
127 	job->owner = owner;
128 	job->fence_ctx = entity->fence_context;
129 	*f = dma_fence_get(&job->base.s_fence->finished);
130 	amdgpu_job_free_resources(job);
131 	amd_sched_entity_push_job(&job->base);
132 
133 	return 0;
134 }
135 
136 static struct dma_fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
137 {
138 	struct amdgpu_job *job = to_amdgpu_job(sched_job);
139 	struct amdgpu_vm *vm = job->vm;
140 
141 	struct dma_fence *fence = amdgpu_sync_get_fence(&job->sync);
142 
143 	while (fence == NULL && vm && !job->vm_id) {
144 		struct amdgpu_ring *ring = job->ring;
145 		int r;
146 
147 		r = amdgpu_vm_grab_id(vm, ring, &job->sync,
148 				      &job->base.s_fence->finished,
149 				      job);
150 		if (r)
151 			DRM_ERROR("Error getting VM ID (%d)\n", r);
152 
153 		fence = amdgpu_sync_get_fence(&job->sync);
154 	}
155 
156 	if (amd_sched_dependency_optimized(fence, sched_job->s_entity))
157 		job->need_pipeline_sync = true;
158 
159 	return fence;
160 }
161 
162 static struct dma_fence *amdgpu_job_run(struct amd_sched_job *sched_job)
163 {
164 	struct dma_fence *fence = NULL;
165 	struct amdgpu_job *job;
166 	int r;
167 
168 	if (!sched_job) {
169 		DRM_ERROR("job is null\n");
170 		return NULL;
171 	}
172 	job = to_amdgpu_job(sched_job);
173 
174 	BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL));
175 
176 	trace_amdgpu_sched_run_job(job);
177 	r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job, &fence);
178 	if (r)
179 		DRM_ERROR("Error scheduling IBs (%d)\n", r);
180 
181 	/* if gpu reset, hw fence will be replaced here */
182 	dma_fence_put(job->fence);
183 	job->fence = dma_fence_get(fence);
184 	amdgpu_job_free_resources(job);
185 	return fence;
186 }
187 
188 const struct amd_sched_backend_ops amdgpu_sched_ops = {
189 	.dependency = amdgpu_job_dependency,
190 	.run_job = amdgpu_job_run,
191 	.timedout_job = amdgpu_job_timedout,
192 	.free_job = amdgpu_job_free_cb
193 };
194