xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c (revision 7181e5590e5ba898804aef3ee6be7f27606e6f8b)
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 
61 	amdgpu_sync_create(&(*job)->sync);
62 
63 	return 0;
64 }
65 
66 int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
67 			     struct amdgpu_job **job)
68 {
69 	int r;
70 
71 	r = amdgpu_job_alloc(adev, 1, job, NULL);
72 	if (r)
73 		return r;
74 
75 	r = amdgpu_ib_get(adev, NULL, size, &(*job)->ibs[0]);
76 	if (r)
77 		kfree(*job);
78 
79 	return r;
80 }
81 
82 void amdgpu_job_free_resources(struct amdgpu_job *job)
83 {
84 	struct fence *f;
85 	unsigned i;
86 
87 	/* use sched fence if available */
88 	f = job->base.s_fence ? &job->base.s_fence->finished : job->fence;
89 
90 	for (i = 0; i < job->num_ibs; ++i)
91 		amdgpu_ib_free(job->adev, &job->ibs[i], f);
92 }
93 
94 static void amdgpu_job_free_cb(struct amd_sched_job *s_job)
95 {
96 	struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
97 
98 	fence_put(job->fence);
99 	amdgpu_sync_free(&job->sync);
100 	kfree(job);
101 }
102 
103 void amdgpu_job_free(struct amdgpu_job *job)
104 {
105 	amdgpu_job_free_resources(job);
106 
107 	fence_put(job->fence);
108 	amdgpu_sync_free(&job->sync);
109 	kfree(job);
110 }
111 
112 int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
113 		      struct amd_sched_entity *entity, void *owner,
114 		      struct fence **f)
115 {
116 	int r;
117 	job->ring = ring;
118 
119 	if (!f)
120 		return -EINVAL;
121 
122 	r = amd_sched_job_init(&job->base, &ring->sched, entity, owner);
123 	if (r)
124 		return r;
125 
126 	job->owner = owner;
127 	job->fence_ctx = entity->fence_context;
128 	*f = fence_get(&job->base.s_fence->finished);
129 	amdgpu_job_free_resources(job);
130 	amd_sched_entity_push_job(&job->base);
131 
132 	return 0;
133 }
134 
135 static struct fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
136 {
137 	struct amdgpu_job *job = to_amdgpu_job(sched_job);
138 	struct amdgpu_vm *vm = job->vm;
139 
140 	struct fence *fence = amdgpu_sync_get_fence(&job->sync);
141 
142 	if (fence == NULL && vm && !job->vm_id) {
143 		struct amdgpu_ring *ring = job->ring;
144 		int r;
145 
146 		r = amdgpu_vm_grab_id(vm, ring, &job->sync,
147 				      &job->base.s_fence->finished,
148 				      job);
149 		if (r)
150 			DRM_ERROR("Error getting VM ID (%d)\n", r);
151 
152 		fence = amdgpu_sync_get_fence(&job->sync);
153 	}
154 
155 	return fence;
156 }
157 
158 static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job)
159 {
160 	struct fence *fence = NULL;
161 	struct amdgpu_job *job;
162 	int r;
163 
164 	if (!sched_job) {
165 		DRM_ERROR("job is null\n");
166 		return NULL;
167 	}
168 	job = to_amdgpu_job(sched_job);
169 
170 	BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL));
171 
172 	trace_amdgpu_sched_run_job(job);
173 	r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs,
174 			       job->sync.last_vm_update, job, &fence);
175 	if (r)
176 		DRM_ERROR("Error scheduling IBs (%d)\n", r);
177 
178 	/* if gpu reset, hw fence will be replaced here */
179 	fence_put(job->fence);
180 	job->fence = fence_get(fence);
181 	amdgpu_job_free_resources(job);
182 	return fence;
183 }
184 
185 const struct amd_sched_backend_ops amdgpu_sched_ops = {
186 	.dependency = amdgpu_job_dependency,
187 	.run_job = amdgpu_job_run,
188 	.timedout_job = amdgpu_job_timedout,
189 	.free_job = amdgpu_job_free_cb
190 };
191