Merge tag 'gvt-next-fixes-2021-02-22' of https://github.com/intel/gvt-linux into drm-intel-next-fixes

gvt-next-fixes-2021-02-22

- Fix to use i915 default state for cmd parser on all engines (Chris)
- Purge dev_priv->gt (Chris)
- Fix gvt object ww locking (Zhi)

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
From: Zhenyu Wang <zhenyuw@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210222083402.GD1551@zhen-hp.sh.intel.com

Author: rodrigovivi

drivers/gpu/drm/i915/gvt/cmd_parser.c

@@ -41,6 +41,7 @@
 #include "gt/intel_lrc.h"
 #include "gt/intel_ring.h"
 #include "gt/intel_gt_requests.h"
+#include "gt/shmem_utils.h"
 #include "gvt.h"
 #include "i915_pvinfo.h"
 #include "trace.h"
@@ -3094,111 +3095,57 @@ int intel_gvt_scan_and_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
  */
 void intel_gvt_update_reg_whitelist(struct intel_vgpu *vgpu)
 {
+	const unsigned long start = LRC_STATE_PN * PAGE_SIZE;
 	struct intel_gvt *gvt = vgpu->gvt;
-	struct drm_i915_private *dev_priv = gvt->gt->i915;
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
-	const unsigned long start = LRC_STATE_PN * PAGE_SIZE;
-	struct i915_request *rq;
-	struct intel_vgpu_submission *s = &vgpu->submission;
-	struct i915_request *requests[I915_NUM_ENGINES] = {};
-	bool is_ctx_pinned[I915_NUM_ENGINES] = {};
-	int ret = 0;
 
 	if (gvt->is_reg_whitelist_updated)
 		return;
 
-	for_each_engine(engine, &dev_priv->gt, id) {
-		ret = intel_context_pin(s->shadow[id]);
-		if (ret) {
-			gvt_vgpu_err("fail to pin shadow ctx\n");
-			goto out;
-		}
-		is_ctx_pinned[id] = true;
-
-		rq = i915_request_create(s->shadow[id]);
-		if (IS_ERR(rq)) {
-			gvt_vgpu_err("fail to alloc default request\n");
-			ret = -EIO;
-			goto out;
-		}
-		requests[id] = i915_request_get(rq);
-		i915_request_add(rq);
-	}
-
-	if (intel_gt_wait_for_idle(&dev_priv->gt,
-				I915_GEM_IDLE_TIMEOUT) == -ETIME) {
-		ret = -EIO;
-		goto out;
-	}
-
 	/* scan init ctx to update cmd accessible list */
-	for_each_engine(engine, &dev_priv->gt, id) {
-		int size = engine->context_size - PAGE_SIZE;
-		void *vaddr;
+	for_each_engine(engine, gvt->gt, id) {
 		struct parser_exec_state s;
-		struct drm_i915_gem_object *obj;
-		struct i915_request *rq;
-
-		rq = requests[id];
-		GEM_BUG_ON(!i915_request_completed(rq));
-		GEM_BUG_ON(!intel_context_is_pinned(rq->context));
-		obj = rq->context->state->obj;
-
-		if (!obj) {
-			ret = -EIO;
-			goto out;
-		}
+		void *vaddr;
+		int ret;
 
-		i915_gem_object_set_cache_coherency(obj,
-						    I915_CACHE_LLC);
+		if (!engine->default_state)
+			continue;
 
-		vaddr = i915_gem_object_pin_map(obj, I915_MAP_WB);
+		vaddr = shmem_pin_map(engine->default_state);
 		if (IS_ERR(vaddr)) {
-			gvt_err("failed to pin init ctx obj, ring=%d, err=%lx\n",
-				id, PTR_ERR(vaddr));
-			ret = PTR_ERR(vaddr);
-			goto out;
+			gvt_err("failed to map %s->default state, err:%zd\n",
+				engine->name, PTR_ERR(vaddr));
+			return;
 		}
 
 		s.buf_type = RING_BUFFER_CTX;
 		s.buf_addr_type = GTT_BUFFER;
 		s.vgpu = vgpu;
 		s.engine = engine;
 		s.ring_start = 0;
-		s.ring_size = size;
+		s.ring_size = engine->context_size - start;
 		s.ring_head = 0;
-		s.ring_tail = size;
+		s.ring_tail = s.ring_size;
 		s.rb_va = vaddr + start;
 		s.workload = NULL;
 		s.is_ctx_wa = false;
 		s.is_init_ctx = true;
 
 		/* skipping the first RING_CTX_SIZE(0x50) dwords */
 		ret = ip_gma_set(&s, RING_CTX_SIZE);
-		if (ret) {
-			i915_gem_object_unpin_map(obj);
-			goto out;
+		if (ret == 0) {
+			ret = command_scan(&s, 0, s.ring_size, 0, s.ring_size);
+			if (ret)
+				gvt_err("Scan init ctx error\n");
 		}
 
-		ret = command_scan(&s, 0, size, 0, size);
+		shmem_unpin_map(engine->default_state, vaddr);
 		if (ret)
-			gvt_err("Scan init ctx error\n");
-
-		i915_gem_object_unpin_map(obj);
+			return;
 	}
 
-out:
-	if (!ret)
-		gvt->is_reg_whitelist_updated = true;
-
-	for (id = 0; id < I915_NUM_ENGINES ; id++) {
-		if (requests[id])
-			i915_request_put(requests[id]);
-
-		if (is_ctx_pinned[id])
-			intel_context_unpin(s->shadow[id]);
-	}
+	gvt->is_reg_whitelist_updated = true;
 }
 
 int intel_gvt_scan_engine_context(struct intel_vgpu_workload *workload)

drivers/gpu/drm/i915/gvt/execlist.c

@@ -522,12 +522,11 @@ static void init_vgpu_execlist(struct intel_vgpu *vgpu,
 static void clean_execlist(struct intel_vgpu *vgpu,
 			   intel_engine_mask_t engine_mask)
 {
-	struct drm_i915_private *dev_priv = vgpu->gvt->gt->i915;
-	struct intel_engine_cs *engine;
 	struct intel_vgpu_submission *s = &vgpu->submission;
+	struct intel_engine_cs *engine;
 	intel_engine_mask_t tmp;
 
-	for_each_engine_masked(engine, &dev_priv->gt, engine_mask, tmp) {
+	for_each_engine_masked(engine, vgpu->gvt->gt, engine_mask, tmp) {
 		kfree(s->ring_scan_buffer[engine->id]);
 		s->ring_scan_buffer[engine->id] = NULL;
 		s->ring_scan_buffer_size[engine->id] = 0;
@@ -537,11 +536,10 @@ static void clean_execlist(struct intel_vgpu *vgpu,
 static void reset_execlist(struct intel_vgpu *vgpu,
 			   intel_engine_mask_t engine_mask)
 {
-	struct drm_i915_private *dev_priv = vgpu->gvt->gt->i915;
 	struct intel_engine_cs *engine;
 	intel_engine_mask_t tmp;
 
-	for_each_engine_masked(engine, &dev_priv->gt, engine_mask, tmp)
+	for_each_engine_masked(engine, vgpu->gvt->gt, engine_mask, tmp)
 		init_vgpu_execlist(vgpu, engine);
 }
 

drivers/gpu/drm/i915/gvt/scheduler.c

@@ -412,7 +412,9 @@ static void release_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
 	if (!wa_ctx->indirect_ctx.obj)
 		return;
 
+	i915_gem_object_lock(wa_ctx->indirect_ctx.obj, NULL);
 	i915_gem_object_unpin_map(wa_ctx->indirect_ctx.obj);
+	i915_gem_object_unlock(wa_ctx->indirect_ctx.obj);
 	i915_gem_object_put(wa_ctx->indirect_ctx.obj);
 
 	wa_ctx->indirect_ctx.obj = NULL;
@@ -520,6 +522,7 @@ static int prepare_shadow_batch_buffer(struct intel_vgpu_workload *workload)
 	struct intel_gvt *gvt = workload->vgpu->gvt;
 	const int gmadr_bytes = gvt->device_info.gmadr_bytes_in_cmd;
 	struct intel_vgpu_shadow_bb *bb;
+	struct i915_gem_ww_ctx ww;
 	int ret;
 
 	list_for_each_entry(bb, &workload->shadow_bb, list) {
@@ -544,10 +547,19 @@ static int prepare_shadow_batch_buffer(struct intel_vgpu_workload *workload)
 		 * directly
 		 */
 		if (!bb->ppgtt) {
-			bb->vma = i915_gem_object_ggtt_pin(bb->obj,
-							   NULL, 0, 0, 0);
+			i915_gem_ww_ctx_init(&ww, false);
+retry:
+			i915_gem_object_lock(bb->obj, &ww);
+
+			bb->vma = i915_gem_object_ggtt_pin_ww(bb->obj, &ww,
+							      NULL, 0, 0, 0);
 			if (IS_ERR(bb->vma)) {
 				ret = PTR_ERR(bb->vma);
+				if (ret == -EDEADLK) {
+					ret = i915_gem_ww_ctx_backoff(&ww);
+					if (!ret)
+						goto retry;
+				}
 				goto err;
 			}
 
@@ -561,13 +573,15 @@ static int prepare_shadow_batch_buffer(struct intel_vgpu_workload *workload)
 						      0);
 			if (ret)
 				goto err;
-		}
 
-		/* No one is going to touch shadow bb from now on. */
-		i915_gem_object_flush_map(bb->obj);
+			/* No one is going to touch shadow bb from now on. */
+			i915_gem_object_flush_map(bb->obj);
+			i915_gem_object_unlock(bb->obj);
+		}
 	}
 	return 0;
 err:
+	i915_gem_ww_ctx_fini(&ww);
 	release_shadow_batch_buffer(workload);
 	return ret;
 }
@@ -594,14 +608,29 @@ static int prepare_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
 	unsigned char *per_ctx_va =
 		(unsigned char *)wa_ctx->indirect_ctx.shadow_va +
 		wa_ctx->indirect_ctx.size;
+	struct i915_gem_ww_ctx ww;
+	int ret;
 
 	if (wa_ctx->indirect_ctx.size == 0)
 		return 0;
 
-	vma = i915_gem_object_ggtt_pin(wa_ctx->indirect_ctx.obj, NULL,
-				       0, CACHELINE_BYTES, 0);
-	if (IS_ERR(vma))
-		return PTR_ERR(vma);
+	i915_gem_ww_ctx_init(&ww, false);
+retry:
+	i915_gem_object_lock(wa_ctx->indirect_ctx.obj, &ww);
+
+	vma = i915_gem_object_ggtt_pin_ww(wa_ctx->indirect_ctx.obj, &ww, NULL,
+					  0, CACHELINE_BYTES, 0);
+	if (IS_ERR(vma)) {
+		ret = PTR_ERR(vma);
+		if (ret == -EDEADLK) {
+			ret = i915_gem_ww_ctx_backoff(&ww);
+			if (!ret)
+				goto retry;
+		}
+		return ret;
+	}
+
+	i915_gem_object_unlock(wa_ctx->indirect_ctx.obj);
 
 	/* FIXME: we are not tracking our pinned VMA leaving it
 	 * up to the core to fix up the stray pin_count upon
@@ -635,12 +664,14 @@ static void release_shadow_batch_buffer(struct intel_vgpu_workload *workload)
 
 	list_for_each_entry_safe(bb, pos, &workload->shadow_bb, list) {
 		if (bb->obj) {
+			i915_gem_object_lock(bb->obj, NULL);
 			if (bb->va && !IS_ERR(bb->va))
 				i915_gem_object_unpin_map(bb->obj);
 
 			if (bb->vma && !IS_ERR(bb->vma))
 				i915_vma_unpin(bb->vma);
 
+			i915_gem_object_unlock(bb->obj);
 			i915_gem_object_put(bb->obj);
 		}
 		list_del(&bb->list);
@@ -1015,13 +1046,12 @@ void intel_vgpu_clean_workloads(struct intel_vgpu *vgpu,
 				intel_engine_mask_t engine_mask)
 {
 	struct intel_vgpu_submission *s = &vgpu->submission;
-	struct drm_i915_private *dev_priv = vgpu->gvt->gt->i915;
 	struct intel_engine_cs *engine;
 	struct intel_vgpu_workload *pos, *n;
 	intel_engine_mask_t tmp;
 
 	/* free the unsubmited workloads in the queues. */
-	for_each_engine_masked(engine, &dev_priv->gt, engine_mask, tmp) {
+	for_each_engine_masked(engine, vgpu->gvt->gt, engine_mask, tmp) {
 		list_for_each_entry_safe(pos, n,
 			&s->workload_q_head[engine->id], list) {
 			list_del_init(&pos->list);