aboutsummaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_job.c')
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_job.c120
1 files changed, 90 insertions, 30 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 759b34799221..46c99331d7f1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -25,8 +25,11 @@
#include <linux/wait.h>
#include <linux/sched.h>
+#include <drm/drm_drv.h>
+
#include "amdgpu.h"
#include "amdgpu_trace.h"
+#include "amdgpu_reset.h"
static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
{
@@ -34,14 +37,25 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
struct amdgpu_job *job = to_amdgpu_job(s_job);
struct amdgpu_task_info ti;
struct amdgpu_device *adev = ring->adev;
+ int idx;
+ int r;
+
+ if (!drm_dev_enter(adev_to_drm(adev), &idx)) {
+ DRM_INFO("%s - device unplugged skipping recovery on scheduler:%s",
+ __func__, s_job->sched->name);
+
+ /* Effectively the job is aborted as the device is gone */
+ return DRM_GPU_SCHED_STAT_ENODEV;
+ }
memset(&ti, 0, sizeof(struct amdgpu_task_info));
+ adev->job_hang = true;
if (amdgpu_gpu_recovery &&
amdgpu_ring_soft_recovery(ring, job->vmid, s_job->s_fence->parent)) {
DRM_ERROR("ring %s timeout, but soft recovered\n",
s_job->sched->name);
- return DRM_GPU_SCHED_STAT_NOMINAL;
+ goto exit;
}
amdgpu_vm_get_task_info(ring->adev, job->pasid, &ti);
@@ -52,27 +66,36 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
ti.process_name, ti.tgid, ti.task_name, ti.pid);
if (amdgpu_device_should_recover_gpu(ring->adev)) {
- amdgpu_device_gpu_recover(ring->adev, job);
- return DRM_GPU_SCHED_STAT_NOMINAL;
+ struct amdgpu_reset_context reset_context;
+ memset(&reset_context, 0, sizeof(reset_context));
+
+ reset_context.method = AMD_RESET_METHOD_NONE;
+ reset_context.reset_req_dev = adev;
+ clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
+ clear_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context.flags);
+
+ r = amdgpu_device_gpu_recover(ring->adev, job, &reset_context);
+ if (r)
+ DRM_ERROR("GPU Recovery Failed: %d\n", r);
} else {
drm_sched_suspend_timeout(&ring->sched);
if (amdgpu_sriov_vf(adev))
adev->virt.tdr_debug = true;
- return DRM_GPU_SCHED_STAT_NOMINAL;
}
+
+exit:
+ adev->job_hang = false;
+ drm_dev_exit(idx);
+ return DRM_GPU_SCHED_STAT_NOMINAL;
}
int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
struct amdgpu_job **job, struct amdgpu_vm *vm)
{
- 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);
+ *job = kzalloc(struct_size(*job, ibs, num_ibs), GFP_KERNEL);
if (!*job)
return -ENOMEM;
@@ -82,8 +105,6 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs,
*/
(*job)->base.sched = &adev->rings[0]->sched;
(*job)->vm = vm;
- (*job)->ibs = (void *)&(*job)[1];
- (*job)->num_ibs = num_ibs;
amdgpu_sync_create(&(*job)->sync);
amdgpu_sync_create(&(*job)->sched_sync);
@@ -103,6 +124,7 @@ int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
if (r)
return r;
+ (*job)->num_ibs = 1;
r = amdgpu_ib_get(adev, NULL, size, pool_type, &(*job)->ibs[0]);
if (r)
kfree(*job);
@@ -110,6 +132,23 @@ int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size,
return r;
}
+void amdgpu_job_set_resources(struct amdgpu_job *job, struct amdgpu_bo *gds,
+ struct amdgpu_bo *gws, struct amdgpu_bo *oa)
+{
+ if (gds) {
+ job->gds_base = amdgpu_bo_gpu_offset(gds) >> PAGE_SHIFT;
+ job->gds_size = amdgpu_bo_size(gds) >> PAGE_SHIFT;
+ }
+ if (gws) {
+ job->gws_base = amdgpu_bo_gpu_offset(gws) >> PAGE_SHIFT;
+ job->gws_size = amdgpu_bo_size(gws) >> PAGE_SHIFT;
+ }
+ if (oa) {
+ job->oa_base = amdgpu_bo_gpu_offset(oa) >> PAGE_SHIFT;
+ job->oa_size = amdgpu_bo_size(oa) >> PAGE_SHIFT;
+ }
+}
+
void amdgpu_job_free_resources(struct amdgpu_job *job)
{
struct amdgpu_ring *ring = to_amdgpu_ring(job->base.sched);
@@ -117,8 +156,7 @@ void amdgpu_job_free_resources(struct amdgpu_job *job)
unsigned i;
/* use sched fence if available */
- f = job->base.s_fence ? &job->base.s_fence->finished : job->fence;
-
+ f = job->base.s_fence ? &job->base.s_fence->finished : &job->hw_fence;
for (i = 0; i < job->num_ibs; ++i)
amdgpu_ib_free(ring->adev, &job->ibs[i], f);
}
@@ -129,20 +167,40 @@ static void amdgpu_job_free_cb(struct drm_sched_job *s_job)
drm_sched_job_cleanup(s_job);
- dma_fence_put(job->fence);
amdgpu_sync_free(&job->sync);
amdgpu_sync_free(&job->sched_sync);
- kfree(job);
+
+ dma_fence_put(&job->hw_fence);
+}
+
+void amdgpu_job_set_gang_leader(struct amdgpu_job *job,
+ struct amdgpu_job *leader)
+{
+ struct dma_fence *fence = &leader->base.s_fence->scheduled;
+
+ WARN_ON(job->gang_submit);
+
+ /*
+ * Don't add a reference when we are the gang leader to avoid circle
+ * dependency.
+ */
+ if (job != leader)
+ dma_fence_get(fence);
+ job->gang_submit = fence;
}
void amdgpu_job_free(struct amdgpu_job *job)
{
amdgpu_job_free_resources(job);
-
- dma_fence_put(job->fence);
amdgpu_sync_free(&job->sync);
amdgpu_sync_free(&job->sched_sync);
- kfree(job);
+ if (job->gang_submit != &job->base.s_fence->scheduled)
+ dma_fence_put(job->gang_submit);
+
+ if (!job->hw_fence.ops)
+ kfree(job);
+ else
+ dma_fence_put(&job->hw_fence);
}
int amdgpu_job_submit(struct amdgpu_job *job, struct drm_sched_entity *entity,
@@ -157,9 +215,11 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct drm_sched_entity *entity,
if (r)
return r;
+ drm_sched_job_arm(&job->base);
+
*f = dma_fence_get(&job->base.s_fence->finished);
amdgpu_job_free_resources(job);
- drm_sched_entity_push_job(&job->base, entity);
+ drm_sched_entity_push_job(&job->base);
return 0;
}
@@ -170,8 +230,8 @@ int amdgpu_job_submit_direct(struct amdgpu_job *job, struct amdgpu_ring *ring,
int r;
job->base.sched = &ring->sched;
- r = amdgpu_ib_schedule(ring, job->num_ibs, job->ibs, NULL, fence);
- job->fence = dma_fence_get(*fence);
+ r = amdgpu_ib_schedule(ring, job->num_ibs, job->ibs, job, fence);
+
if (r)
return r;
@@ -205,12 +265,16 @@ static struct dma_fence *amdgpu_job_dependency(struct drm_sched_job *sched_job,
fence = amdgpu_sync_get_fence(&job->sync);
}
+ if (!fence && job->gang_submit)
+ fence = amdgpu_device_switch_gang(ring->adev, job->gang_submit);
+
return fence;
}
static struct dma_fence *amdgpu_job_run(struct drm_sched_job *sched_job)
{
struct amdgpu_ring *ring = to_amdgpu_ring(sched_job->sched);
+ struct amdgpu_device *adev = ring->adev;
struct dma_fence *fence = NULL, *finished;
struct amdgpu_job *job;
int r = 0;
@@ -222,8 +286,10 @@ static struct dma_fence *amdgpu_job_run(struct drm_sched_job *sched_job)
trace_amdgpu_sched_run_job(job);
- if (job->vram_lost_counter != atomic_read(&ring->adev->vram_lost_counter))
- dma_fence_set_error(finished, -ECANCELED);/* skip IB as well if VRAM lost */
+ /* Skip job if VRAM is lost and never resubmit gangs */
+ if (job->vram_lost_counter != atomic_read(&adev->vram_lost_counter) ||
+ (job->job_run_counter && job->gang_submit))
+ dma_fence_set_error(finished, -ECANCELED);
if (finished->error < 0) {
DRM_INFO("Skip scheduling IBs!\n");
@@ -233,10 +299,8 @@ static struct dma_fence *amdgpu_job_run(struct drm_sched_job *sched_job)
if (r)
DRM_ERROR("Error scheduling IBs (%d)\n", r);
}
- /* if gpu reset, hw fence will be replaced here */
- dma_fence_put(job->fence);
- job->fence = dma_fence_get(fence);
+ job->job_run_counter++;
amdgpu_job_free_resources(job);
fence = r ? ERR_PTR(r) : fence;
@@ -255,10 +319,6 @@ void amdgpu_job_stop_all_jobs_on_sched(struct drm_gpu_scheduler *sched)
/* Signal all jobs not yet scheduled */
for (i = DRM_SCHED_PRIORITY_COUNT - 1; i >= DRM_SCHED_PRIORITY_MIN; i--) {
struct drm_sched_rq *rq = &sched->sched_rq[i];
-
- if (!rq)
- continue;
-
spin_lock(&rq->lock);
list_for_each_entry(s_entity, &rq->entities, list) {
while ((s_job = to_drm_sched_job(spsc_queue_pop(&s_entity->job_queue)))) {
Powered by cgit, Gruvboxed by Blaster4385.