drm/amdgpu: stop reserving the BO in the MMU callback v3

Instead take the callback lock during the final parts of CS.

This should solve the last remaining locking order problems with BO reservations.

v2: rebase, make dummy functions static inline
v3: add one more missing inline and comments

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 24 insertions, 6 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 4877df83b801..c2310d4eebc8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -513,8 +513,11 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
         INIT_LIST_HEAD(&p->validated);
 
         p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);
-        if (p->bo_list)
+        if (p->bo_list) {
                 amdgpu_bo_list_get_list(p->bo_list, &p->validated);
+                if (p->bo_list->first_userptr != p->bo_list->num_entries)
+                        p->mn = amdgpu_mn_get(p->adev);
+        }
 
         INIT_LIST_HEAD(&duplicates);
         amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd);
@@ -722,11 +725,7 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error,
 {
         unsigned i;
 
-        if (!error)
-                ttm_eu_fence_buffer_objects(&parser->ticket,
-                                            &parser->validated,
-                                            parser->fence);
-        else if (backoff)
+        if (error && backoff)
                 ttm_eu_backoff_reservation(&parser->ticket,
                                            &parser->validated);
 
@@ -1127,14 +1126,29 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
         struct amdgpu_ring *ring = p->job->ring;
         struct amd_sched_entity *entity = &p->ctx->rings[ring->idx].entity;
         struct amdgpu_job *job;
+        unsigned i;
         int r;
 
+        amdgpu_mn_lock(p->mn);
+        if (p->bo_list) {
+                for (i = p->bo_list->first_userptr;
+                     i < p->bo_list->num_entries; ++i) {
+                        struct amdgpu_bo *bo = p->bo_list->array[i].robj;
+
+                        if (amdgpu_ttm_tt_userptr_needs_pages(bo->tbo.ttm)) {
+                                amdgpu_mn_unlock(p->mn);
+                                return -ERESTARTSYS;
+                        }
+                }
+        }
+
         job = p->job;
         p->job = NULL;
 
         r = amd_sched_job_init(&job->base, &ring->sched, entity, p->filp);
         if (r) {
                 amdgpu_job_free(job);
+                amdgpu_mn_unlock(p->mn);
                 return r;
         }
 
@@ -1150,6 +1164,10 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 
         trace_amdgpu_cs_ioctl(job);
         amd_sched_entity_push_job(&job->base);
+
+        ttm_eu_fence_buffer_objects(&p->ticket, &p->validated, p->fence);
+        amdgpu_mn_unlock(p->mn);
+
         return 0;
 }