drm/i915/gvt: Move common workload preparation into prepare_workload()

Move common workload preparation into prepare_workload() in scheduler.c,
as they are not specific to execlist emulation.

Signed-off-by: Zhi Wang <zhi.a.wang@intel.com>

1 files changed, 151 insertions, 1 deletions

diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c
index 3d1435f55c7b..a7a67cc65a07 100644
--- a/drivers/gpu/drm/i915/gvt/scheduler.c
+++ b/drivers/gpu/drm/i915/gvt/scheduler.c
@@ -325,13 +325,157 @@ err_scan:
 	return ret;
 }
 
+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_shadow_bb_entry *entry_obj;
+
+	/* pin the gem object to ggtt */
+	list_for_each_entry(entry_obj, &workload->shadow_bb, list) {
+		struct i915_vma *vma;
+
+		vma = i915_gem_object_ggtt_pin(entry_obj->obj, NULL, 0, 4, 0);
+		if (IS_ERR(vma))
+			return PTR_ERR(vma);
+
+		/* FIXME: we are not tracking our pinned VMA leaving it
+		 * up to the core to fix up the stray pin_count upon
+		 * free.
+		 */
+
+		/* update the relocate gma with shadow batch buffer*/
+		entry_obj->bb_start_cmd_va[1] = i915_ggtt_offset(vma);
+		if (gmadr_bytes == 8)
+			entry_obj->bb_start_cmd_va[2] = 0;
+	}
+	return 0;
+}
+
+static int update_wa_ctx_2_shadow_ctx(struct intel_shadow_wa_ctx *wa_ctx)
+{
+	struct intel_vgpu_workload *workload = container_of(wa_ctx,
+					struct intel_vgpu_workload,
+					wa_ctx);
+	int ring_id = workload->ring_id;
+	struct intel_vgpu_submission *s = &workload->vgpu->submission;
+	struct i915_gem_context *shadow_ctx = s->shadow_ctx;
+	struct drm_i915_gem_object *ctx_obj =
+		shadow_ctx->engine[ring_id].state->obj;
+	struct execlist_ring_context *shadow_ring_context;
+	struct page *page;
+
+	page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN);
+	shadow_ring_context = kmap_atomic(page);
+
+	shadow_ring_context->bb_per_ctx_ptr.val =
+		(shadow_ring_context->bb_per_ctx_ptr.val &
+		(~PER_CTX_ADDR_MASK)) | wa_ctx->per_ctx.shadow_gma;
+	shadow_ring_context->rcs_indirect_ctx.val =
+		(shadow_ring_context->rcs_indirect_ctx.val &
+		(~INDIRECT_CTX_ADDR_MASK)) | wa_ctx->indirect_ctx.shadow_gma;
+
+	kunmap_atomic(shadow_ring_context);
+	return 0;
+}
+
+static int prepare_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
+{
+	struct i915_vma *vma;
+	unsigned char *per_ctx_va =
+		(unsigned char *)wa_ctx->indirect_ctx.shadow_va +
+		wa_ctx->indirect_ctx.size;
+
+	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);
+
+	/* FIXME: we are not tracking our pinned VMA leaving it
+	 * up to the core to fix up the stray pin_count upon
+	 * free.
+	 */
+
+	wa_ctx->indirect_ctx.shadow_gma = i915_ggtt_offset(vma);
+
+	wa_ctx->per_ctx.shadow_gma = *((unsigned int *)per_ctx_va + 1);
+	memset(per_ctx_va, 0, CACHELINE_BYTES);
+
+	update_wa_ctx_2_shadow_ctx(wa_ctx);
+	return 0;
+}
+
+static void release_shadow_batch_buffer(struct intel_vgpu_workload *workload)
+{
+	/* release all the shadow batch buffer */
+	if (!list_empty(&workload->shadow_bb)) {
+		struct intel_shadow_bb_entry *entry_obj =
+			list_first_entry(&workload->shadow_bb,
+					 struct intel_shadow_bb_entry,
+					 list);
+		struct intel_shadow_bb_entry *temp;
+
+		list_for_each_entry_safe(entry_obj, temp, &workload->shadow_bb,
+					 list) {
+			i915_gem_object_unpin_map(entry_obj->obj);
+			i915_gem_object_put(entry_obj->obj);
+			list_del(&entry_obj->list);
+			kfree(entry_obj);
+		}
+	}
+}
+
 static int prepare_workload(struct intel_vgpu_workload *workload)
 {
+	struct intel_vgpu *vgpu = workload->vgpu;
 	int ret = 0;
 
-	if (workload->prepare)
+	ret = intel_vgpu_pin_mm(workload->shadow_mm);
+	if (ret) {
+		gvt_vgpu_err("fail to vgpu pin mm\n");
+		return ret;
+	}
+
+	ret = intel_vgpu_sync_oos_pages(workload->vgpu);
+	if (ret) {
+		gvt_vgpu_err("fail to vgpu sync oos pages\n");
+		goto err_unpin_mm;
+	}
+
+	ret = intel_vgpu_flush_post_shadow(workload->vgpu);
+	if (ret) {
+		gvt_vgpu_err("fail to flush post shadow\n");
+		goto err_unpin_mm;
+	}
+
+	ret = prepare_shadow_batch_buffer(workload);
+	if (ret) {
+		gvt_vgpu_err("fail to prepare_shadow_batch_buffer\n");
+		goto err_unpin_mm;
+	}
+
+	ret = prepare_shadow_wa_ctx(&workload->wa_ctx);
+	if (ret) {
+		gvt_vgpu_err("fail to prepare_shadow_wa_ctx\n");
+		goto err_shadow_batch;
+	}
+
+	if (workload->prepare) {
 		ret = workload->prepare(workload);
+		if (ret)
+			goto err_shadow_wa_ctx;
+	}
 
+	return 0;
+err_shadow_wa_ctx:
+	release_shadow_wa_ctx(&workload->wa_ctx);
+err_shadow_batch:
+	release_shadow_batch_buffer(workload);
+err_unpin_mm:
+	intel_vgpu_unpin_mm(workload->shadow_mm);
 	return ret;
 }
 
@@ -557,6 +701,12 @@ static void complete_current_workload(struct intel_gvt *gvt, int ring_id)
 	scheduler->current_workload[ring_id] = NULL;
 
 	list_del_init(&workload->list);
+
+	if (!workload->status) {
+		release_shadow_batch_buffer(workload);
+		release_shadow_wa_ctx(&workload->wa_ctx);
+	}
+
 	workload->complete(workload);
 
 	atomic_dec(&s->running_workload_num);