amdgpu_job.c 4.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190
  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. static void amdgpu_job_timedout(struct amd_sched_job *s_job)
  31. {
  32. struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
  33. DRM_ERROR("ring %s timeout, last signaled seq=%u, last emitted seq=%u\n",
  34. job->base.sched->name,
  35. atomic_read(&job->ring->fence_drv.last_seq),
  36. job->ring->fence_drv.sync_seq);
  37. amdgpu_gpu_reset(job->adev);
  38. }
  39. int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
  40. struct amdgpu_job **job, struct amdgpu_vm *vm)
  41. {
  42. size_t size = sizeof(struct amdgpu_job);
  43. if (num_ibs == 0)
  44. return -EINVAL;
  45. size += sizeof(struct amdgpu_ib) * num_ibs;
  46. *job = kzalloc(size, GFP_KERNEL);
  47. if (!*job)
  48. return -ENOMEM;
  49. (*job)->adev = adev;
  50. (*job)->vm = vm;
  51. (*job)->ibs = (void *)&(*job)[1];
  52. (*job)->num_ibs = num_ibs;
  53. amdgpu_sync_create(&(*job)->sync);
  54. return 0;
  55. }
  56. int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
  57. struct amdgpu_job **job)
  58. {
  59. int r;
  60. r = amdgpu_job_alloc(adev, 1, job, NULL);
  61. if (r)
  62. return r;
  63. r = amdgpu_ib_get(adev, NULL, size, &(*job)->ibs[0]);
  64. if (r)
  65. kfree(*job);
  66. return r;
  67. }
  68. void amdgpu_job_free_resources(struct amdgpu_job *job)
  69. {
  70. struct fence *f;
  71. unsigned i;
  72. /* use sched fence if available */
  73. f = job->base.s_fence ? &job->base.s_fence->finished : job->fence;
  74. for (i = 0; i < job->num_ibs; ++i)
  75. amdgpu_ib_free(job->adev, &job->ibs[i], f);
  76. }
  77. static void amdgpu_job_free_cb(struct amd_sched_job *s_job)
  78. {
  79. struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
  80. fence_put(job->fence);
  81. amdgpu_sync_free(&job->sync);
  82. kfree(job);
  83. }
  84. void amdgpu_job_free(struct amdgpu_job *job)
  85. {
  86. amdgpu_job_free_resources(job);
  87. fence_put(job->fence);
  88. amdgpu_sync_free(&job->sync);
  89. kfree(job);
  90. }
  91. int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
  92. struct amd_sched_entity *entity, void *owner,
  93. struct fence **f)
  94. {
  95. int r;
  96. job->ring = ring;
  97. if (!f)
  98. return -EINVAL;
  99. r = amd_sched_job_init(&job->base, &ring->sched, entity, owner);
  100. if (r)
  101. return r;
  102. job->owner = owner;
  103. job->fence_ctx = entity->fence_context;
  104. *f = fence_get(&job->base.s_fence->finished);
  105. amdgpu_job_free_resources(job);
  106. amd_sched_entity_push_job(&job->base);
  107. return 0;
  108. }
  109. static struct fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
  110. {
  111. struct amdgpu_job *job = to_amdgpu_job(sched_job);
  112. struct amdgpu_vm *vm = job->vm;
  113. struct fence *fence = amdgpu_sync_get_fence(&job->sync);
  114. if (fence == NULL && vm && !job->vm_id) {
  115. struct amdgpu_ring *ring = job->ring;
  116. int r;
  117. r = amdgpu_vm_grab_id(vm, ring, &job->sync,
  118. &job->base.s_fence->finished,
  119. job);
  120. if (r)
  121. DRM_ERROR("Error getting VM ID (%d)\n", r);
  122. fence = amdgpu_sync_get_fence(&job->sync);
  123. }
  124. return fence;
  125. }
  126. static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job)
  127. {
  128. struct fence *fence = NULL;
  129. struct amdgpu_job *job;
  130. int r;
  131. if (!sched_job) {
  132. DRM_ERROR("job is null\n");
  133. return NULL;
  134. }
  135. job = to_amdgpu_job(sched_job);
  136. BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL));
  137. trace_amdgpu_sched_run_job(job);
  138. r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs,
  139. job->sync.last_vm_update, job, &fence);
  140. if (r)
  141. DRM_ERROR("Error scheduling IBs (%d)\n", r);
  142. /* if gpu reset, hw fence will be replaced here */
  143. fence_put(job->fence);
  144. job->fence = fence_get(fence);
  145. amdgpu_job_free_resources(job);
  146. return fence;
  147. }
  148. const struct amd_sched_backend_ops amdgpu_sched_ops = {
  149. .dependency = amdgpu_job_dependency,
  150. .run_job = amdgpu_job_run,
  151. .timedout_job = amdgpu_job_timedout,
  152. .free_job = amdgpu_job_free_cb
  153. };