1 files changed, 90 insertions, 45 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 8516c814bc9b..040f4cb6ab2d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -32,6 +32,8 @@
 
 #include <drm/amdgpu_drm.h>
 #include <drm/drm_syncobj.h>
+#include <drm/ttm/ttm_tt.h>
+
 #include "amdgpu_cs.h"
 #include "amdgpu.h"
 #include "amdgpu_trace.h"
@@ -61,6 +63,8 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p,
 		amdgpu_ctx_put(p->ctx);
 		return -ECANCELED;
 	}
+
+	amdgpu_sync_create(&p->sync);
 	return 0;
 }
 
@@ -108,6 +112,9 @@ static int amdgpu_cs_p1_ib(struct amdgpu_cs_parser *p,
 	if (r < 0)
 		return r;
 
+	if (num_ibs[r] >= amdgpu_ring_max_ibs(chunk_ib->ip_type))
+		return -EINVAL;
+
 	++(num_ibs[r]);
 	p->gang_leader_idx = r;
 	return 0;
@@ -129,9 +136,6 @@ static int amdgpu_cs_p1_user_fence(struct amdgpu_cs_parser *p,
 	bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
 	p->uf_entry.priority = 0;
 	p->uf_entry.tv.bo = &bo->tbo;
-	/* One for TTM and two for the CS job */
-	p->uf_entry.tv.num_shared = 3;
-
 	drm_gem_object_put(gobj);
 
 	size = amdgpu_bo_size(bo);
@@ -188,7 +192,7 @@ static int amdgpu_cs_pass1(struct amdgpu_cs_parser *p,
 	uint64_t *chunk_array_user;
 	uint64_t *chunk_array;
 	uint32_t uf_offset = 0;
-	unsigned int size;
+	size_t size;
 	int ret;
 	int i;
 
@@ -281,6 +285,7 @@ static int amdgpu_cs_pass1(struct amdgpu_cs_parser *p,
 		case AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES:
 		case AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_WAIT:
 		case AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_SIGNAL:
+		case AMDGPU_CHUNK_ID_CP_GFX_SHADOW:
 			break;
 
 		default:
@@ -301,7 +306,7 @@ static int amdgpu_cs_pass1(struct amdgpu_cs_parser *p,
 	}
 	p->gang_leader = p->jobs[p->gang_leader_idx];
 
-	if (p->ctx->vram_lost_counter != p->gang_leader->vram_lost_counter) {
+	if (p->ctx->generation != p->gang_leader->generation) {
 		ret = -ECANCELED;
 		goto free_all_kdata;
 	}
@@ -389,7 +394,7 @@ static int amdgpu_cs_p2_dependencies(struct amdgpu_cs_parser *p,
 {
 	struct drm_amdgpu_cs_chunk_dep *deps = chunk->kdata;
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
-	unsigned num_deps;
+	unsigned int num_deps;
 	int i, r;
 
 	num_deps = chunk->length_dw * 4 /
@@ -452,18 +457,6 @@ static int amdgpu_syncobj_lookup_and_add(struct amdgpu_cs_parser *p,
 	}
 
 	r = amdgpu_sync_fence(&p->sync, fence);
-	if (r)
-		goto error;
-
-	/*
-	 * When we have an explicit dependency it might be necessary to insert a
-	 * pipeline sync to make sure that all caches etc are flushed and the
-	 * next job actually sees the results from the previous one.
-	 */
-	if (fence->context == p->gang_leader->base.entity->fence_context)
-		r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence);
-
-error:
 	dma_fence_put(fence);
 	return r;
 }
@@ -472,7 +465,7 @@ static int amdgpu_cs_p2_syncobj_in(struct amdgpu_cs_parser *p,
 				   struct amdgpu_cs_chunk *chunk)
 {
 	struct drm_amdgpu_cs_chunk_sem *deps = chunk->kdata;
-	unsigned num_deps;
+	unsigned int num_deps;
 	int i, r;
 
 	num_deps = chunk->length_dw * 4 /
@@ -490,7 +483,7 @@ static int amdgpu_cs_p2_syncobj_timeline_wait(struct amdgpu_cs_parser *p,
 					      struct amdgpu_cs_chunk *chunk)
 {
 	struct drm_amdgpu_cs_chunk_syncobj *syncobj_deps = chunk->kdata;
-	unsigned num_deps;
+	unsigned int num_deps;
 	int i, r;
 
 	num_deps = chunk->length_dw * 4 /
@@ -510,7 +503,7 @@ static int amdgpu_cs_p2_syncobj_out(struct amdgpu_cs_parser *p,
 				    struct amdgpu_cs_chunk *chunk)
 {
 	struct drm_amdgpu_cs_chunk_sem *deps = chunk->kdata;
-	unsigned num_deps;
+	unsigned int num_deps;
 	int i;
 
 	num_deps = chunk->length_dw * 4 /
@@ -544,7 +537,7 @@ static int amdgpu_cs_p2_syncobj_timeline_signal(struct amdgpu_cs_parser *p,
 						struct amdgpu_cs_chunk *chunk)
 {
 	struct drm_amdgpu_cs_chunk_syncobj *syncobj_deps = chunk->kdata;
-	unsigned num_deps;
+	unsigned int num_deps;
 	int i;
 
 	num_deps = chunk->length_dw * 4 /
@@ -583,6 +576,26 @@ static int amdgpu_cs_p2_syncobj_timeline_signal(struct amdgpu_cs_parser *p,
 	return 0;
 }
 
+static int amdgpu_cs_p2_shadow(struct amdgpu_cs_parser *p,
+			       struct amdgpu_cs_chunk *chunk)
+{
+	struct drm_amdgpu_cs_chunk_cp_gfx_shadow *shadow = chunk->kdata;
+	int i;
+
+	if (shadow->flags & ~AMDGPU_CS_CHUNK_CP_GFX_SHADOW_FLAGS_INIT_SHADOW)
+		return -EINVAL;
+
+	for (i = 0; i < p->gang_size; ++i) {
+		p->jobs[i]->shadow_va = shadow->shadow_va;
+		p->jobs[i]->csa_va = shadow->csa_va;
+		p->jobs[i]->gds_va = shadow->gds_va;
+		p->jobs[i]->init_shadow =
+			shadow->flags & AMDGPU_CS_CHUNK_CP_GFX_SHADOW_FLAGS_INIT_SHADOW;
+	}
+
+	return 0;
+}
+
 static int amdgpu_cs_pass2(struct amdgpu_cs_parser *p)
 {
 	unsigned int ce_preempt = 0, de_preempt = 0;
@@ -625,6 +638,11 @@ static int amdgpu_cs_pass2(struct amdgpu_cs_parser *p)
 			if (r)
 				return r;
 			break;
+		case AMDGPU_CHUNK_ID_CP_GFX_SHADOW:
+			r = amdgpu_cs_p2_shadow(p, chunk);
+			if (r)
+				return r;
+			break;
 		}
 	}
 
@@ -737,6 +755,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev,
 
 		if (used_vis_vram < total_vis_vram) {
 			u64 free_vis_vram = total_vis_vram - used_vis_vram;
+
 			adev->mm_stats.accum_us_vis = min(adev->mm_stats.accum_us_vis +
 							  increment_us, us_upper_bound);
 
@@ -890,15 +909,19 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 
 	mutex_lock(&p->bo_list->bo_list_mutex);
 
-	/* One for TTM and one for the CS job */
+	/* One for TTM and one for each CS job */
 	amdgpu_bo_list_for_each_entry(e, p->bo_list)
-		e->tv.num_shared = 2;
+		e->tv.num_shared = 1 + p->gang_size;
+	p->uf_entry.tv.num_shared = 1 + p->gang_size;
 
 	amdgpu_bo_list_get_list(p->bo_list, &p->validated);
 
 	INIT_LIST_HEAD(&duplicates);
 	amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd);
 
+	/* Two for VM updates, one for TTM and one for each CS job */
+	p->vm_pd.tv.num_shared = 3 + p->gang_size;
+
 	if (p->uf_entry.tv.bo && !ttm_to_amdgpu_bo(p->uf_entry.tv.bo)->parent)
 		list_add(&p->uf_entry.tv.head, &p->validated);
 
@@ -1055,9 +1078,8 @@ static int amdgpu_cs_patch_ibs(struct amdgpu_cs_parser *p,
 
 		/* the IB should be reserved at this point */
 		r = amdgpu_bo_kmap(aobj, (void **)&kptr);
-		if (r) {
+		if (r)
 			return r;
-		}
 
 		kptr += va_start - (m->start * AMDGPU_GPU_PAGE_SIZE);
 
@@ -1188,10 +1210,19 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 {
 	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
+	struct drm_gpu_scheduler *sched;
 	struct amdgpu_bo_list_entry *e;
+	struct dma_fence *fence;
 	unsigned int i;
 	int r;
 
+	r = amdgpu_ctx_wait_prev_fence(p->ctx, p->entities[p->gang_leader_idx]);
+	if (r) {
+		if (r != -ERESTARTSYS)
+			DRM_ERROR("amdgpu_ctx_wait_prev_fence failed.\n");
+		return r;
+	}
+
 	list_for_each_entry(e, &p->validated, tv.head) {
 		struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
 		struct dma_resv *resv = bo->tbo.base.resv;
@@ -1211,10 +1242,27 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 			return r;
 	}
 
-	r = amdgpu_ctx_wait_prev_fence(p->ctx, p->entities[p->gang_leader_idx]);
-	if (r && r != -ERESTARTSYS)
-		DRM_ERROR("amdgpu_ctx_wait_prev_fence failed.\n");
-	return r;
+	sched = p->gang_leader->base.entity->rq->sched;
+	while ((fence = amdgpu_sync_get_fence(&p->sync))) {
+		struct drm_sched_fence *s_fence = to_drm_sched_fence(fence);
+
+		/*
+		 * When we have an dependency it might be necessary to insert a
+		 * pipeline sync to make sure that all caches etc are flushed and the
+		 * next job actually sees the results from the previous one
+		 * before we start executing on the same scheduler ring.
+		 */
+		if (!s_fence || s_fence->sched != sched) {
+			dma_fence_put(fence);
+			continue;
+		}
+
+		r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence);
+		dma_fence_put(fence);
+		if (r)
+			return r;
+	}
+	return 0;
 }
 
 static void amdgpu_cs_post_dependencies(struct amdgpu_cs_parser *p)
@@ -1254,9 +1302,12 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 			continue;
 
 		fence = &p->jobs[i]->base.s_fence->scheduled;
+		dma_fence_get(fence);
 		r = drm_sched_job_add_dependency(&leader->base, fence);
-		if (r)
-			goto error_cleanup;
+		if (r) {
+			dma_fence_put(fence);
+			return r;
+		}
 	}
 
 	if (p->gang_size > 1) {
@@ -1282,7 +1333,8 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	}
 	if (r) {
 		r = -EAGAIN;
-		goto error_unlock;
+		mutex_unlock(&p->adev->notifier_lock);
+		return r;
 	}
 
 	p->fence = dma_fence_get(&leader->base.s_fence->finished);
@@ -1329,21 +1381,14 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	mutex_unlock(&p->adev->notifier_lock);
 	mutex_unlock(&p->bo_list->bo_list_mutex);
 	return 0;
-
-error_unlock:
-	mutex_unlock(&p->adev->notifier_lock);
-
-error_cleanup:
-	for (i = 0; i < p->gang_size; ++i)
-		drm_sched_job_cleanup(&p->jobs[i]->base);
-	return r;
 }
 
 /* Cleanup the parser structure */
 static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser)
 {
-	unsigned i;
+	unsigned int i;
 
+	amdgpu_sync_free(&parser->sync);
 	for (i = 0; i < parser->num_post_deps; i++) {
 		drm_syncobj_put(parser->post_deps[i].syncobj);
 		kfree(parser->post_deps[i].chain);
@@ -1609,15 +1654,15 @@ static int amdgpu_cs_wait_all_fences(struct amdgpu_device *adev,
 			continue;
 
 		r = dma_fence_wait_timeout(fence, true, timeout);
+		if (r > 0 && fence->error)
+			r = fence->error;
+
 		dma_fence_put(fence);
 		if (r < 0)
 			return r;
 
 		if (r == 0)
 			break;
-
-		if (fence->error)
-			return fence->error;
 	}
 
 	memset(wait, 0, sizeof(*wait));