Skip to contentSkip to conversation

drm/i915/gvt: Make elsp_dwords in the right order

PATCH In review Zhi Wang 2 participants 12 messages Sep 12, 2017

Subsystems (3)
lore
Share & tools
lore

Zhi Wang

author
#0 lore raw

PATCH v2 1/9 drm/i915/gvt: Make elsp_dwords in the right order

The context descriptors in elsp_dwords are stored in a reversed order and the definition of context descriptor is also reversed. The revesred stuff is hard to be used and might cause misunderstanding. Make them in the right oder for following code re-factoring.

Tested on my SKL NUC.

---
 drivers/gpu/drm/i915/gvt/execlist.c | 22 +++++++++++-----------
 drivers/gpu/drm/i915/gvt/execlist.h |  8 ++++----
 drivers/gpu/drm/i915/gvt/handlers.c |  2 +-
 3 files changed, 16 insertions(+), 16 deletions(-)
drivers/gpu/drm/i915/gvt/execlist.c +11 -11
index 5ec07ec..9402aa5 100644
@@ -511,8 +511,8 @@ static int prepare_execlist_workload(struct intel_vgpu_workload *workload)
 	if (!workload->emulate_schedule_in)
 		return 0;
 
-	ctx[0] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 1);
-	ctx[1] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 0);
+	ctx[0] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 0);
+	ctx[1] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 1);
 
 	ret = emulate_execlist_schedule_in(&vgpu->execlist[ring_id], ctx);
 	if (!ret)
@@ -771,21 +771,21 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
 int intel_vgpu_submit_execlist(struct intel_vgpu *vgpu, int ring_id)
 {
 	struct intel_vgpu_execlist *execlist = &vgpu->execlist[ring_id];
-	struct execlist_ctx_descriptor_format desc[2];
+	struct execlist_ctx_descriptor_format *desc[2];
 	int i, ret;
 
-	desc[0] = *get_desc_from_elsp_dwords(&execlist->elsp_dwords, 1);
-	desc[1] = *get_desc_from_elsp_dwords(&execlist->elsp_dwords, 0);
+	desc[0] = get_desc_from_elsp_dwords(&execlist->elsp_dwords, 0);
+	desc[1] = get_desc_from_elsp_dwords(&execlist->elsp_dwords, 1);
 
-	if (!desc[0].valid) {
+	if (!desc[0]->valid) {
 		gvt_vgpu_err("invalid elsp submission, desc0 is invalid\n");
 		goto inv_desc;
 	}
 
 	for (i = 0; i < ARRAY_SIZE(desc); i++) {
-		if (!desc[i].valid)
+		if (!desc[i]->valid)
 			continue;
-		if (!desc[i].privilege_access) {
+		if (!desc[i]->privilege_access) {
 			gvt_vgpu_err("unexpected GGTT elsp submission\n");
 			goto inv_desc;
 		}
@@ -793,9 +793,9 @@ int intel_vgpu_submit_execlist(struct intel_vgpu *vgpu, int ring_id)
 
 	/* submit workload */
 	for (i = 0; i < ARRAY_SIZE(desc); i++) {
-		if (!desc[i].valid)
+		if (!desc[i]->valid)
 			continue;
-		ret = submit_context(vgpu, ring_id, &desc[i], i == 0);
+		ret = submit_context(vgpu, ring_id, desc[i], i == 0);
 		if (ret) {
 			gvt_vgpu_err("failed to submit desc %d\n", i);
 			return ret;
@@ -806,7 +806,7 @@ int intel_vgpu_submit_execlist(struct intel_vgpu *vgpu, int ring_id)
 
 inv_desc:
 	gvt_vgpu_err("descriptors content: desc0 %08x %08x desc1 %08x %08x\n",
-		     desc[0].udw, desc[0].ldw, desc[1].udw, desc[1].ldw);
+		     desc[0]->udw, desc[0]->ldw, desc[1]->udw, desc[1]->ldw);
 	return -EINVAL;
 }
 
drivers/gpu/drm/i915/gvt/execlist.h +4 -4
index 7eced40..427e40e 100644
@@ -37,10 +37,6 @@
 
 struct execlist_ctx_descriptor_format {
 	union {
-		u32 udw;
-		u32 context_id;
-	};
-	union {
 		u32 ldw;
 		struct {
 			u32 valid                  : 1;
@@ -54,6 +50,10 @@ struct execlist_ctx_descriptor_format {
 			u32 lrca                   : 20;
 		};
 	};
+	union {
+		u32 udw;
+		u32 context_id;
+	};
 };
 
 struct execlist_status_format {
drivers/gpu/drm/i915/gvt/handlers.c +1 -1
index 2294466..dcee645 100644
@@ -1464,7 +1464,7 @@ static int elsp_mmio_write(struct intel_vgpu *vgpu, unsigned int offset,
 
 	execlist = &vgpu->execlist[ring_id];
 
-	execlist->elsp_dwords.data[execlist->elsp_dwords.index] = data;
+	execlist->elsp_dwords.data[3 - execlist->elsp_dwords.index] = data;
 	if (execlist->elsp_dwords.index == 3) {
 		ret = intel_vgpu_submit_execlist(vgpu, ring_id);
 		if(ret)
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#1 lore raw

PATCH v2 2/9 drm/i915/gvt: Rename intel_vgpu_{init, clean}_gvt_context()

To move workload related functions into scheduler.c, an expected way is to collect all the init/clean functions related to vGPU workload submission into fewer functions.

Rename intel_vgpu_{init, clean}_gvt_context() for above usage in future.

---
 drivers/gpu/drm/i915/gvt/scheduler.c | 21 +++++++++++++++++++--
 drivers/gpu/drm/i915/gvt/scheduler.h |  4 ++--
 drivers/gpu/drm/i915/gvt/vgpu.c      |  6 +++---
 3 files changed, 24 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/gvt/scheduler.c +19 -2
index 6fb9b58..d14d910 100644
@@ -708,12 +708,29 @@ int intel_gvt_init_workload_scheduler(struct intel_gvt *gvt)
 	return ret;
 }
 
-void intel_vgpu_clean_gvt_context(struct intel_vgpu *vgpu)
+/**
+ * intel_vgpu_clean_submission - free submission-related resource for vGPU
+ * @vgpu: a vGPU
+ *
+ * This function is called when a vGPU is being destroyed.
+ *
+ */
+void intel_vgpu_clean_submission(struct intel_vgpu *vgpu)
 {
 	i915_gem_context_put(vgpu->shadow_ctx);
 }
 
-int intel_vgpu_init_gvt_context(struct intel_vgpu *vgpu)
+/**
+ * intel_vgpu_setup_submission - setup submission-related resource for vGPU
+ * @vgpu: a vGPU
+ *
+ * This function is called when a vGPU is being created.
+ *
+ * Returns:
+ * Zero on success, negative error code if failed.
+ *
+ */
+int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
 {
 	atomic_set(&vgpu->running_workload_num, 0);
 
drivers/gpu/drm/i915/gvt/scheduler.h +2 -2
index f36b85f..b6ca44e 100644
@@ -136,9 +136,9 @@ void intel_gvt_clean_workload_scheduler(struct intel_gvt *gvt);
 
 void intel_gvt_wait_vgpu_idle(struct intel_vgpu *vgpu);
 
-int intel_vgpu_init_gvt_context(struct intel_vgpu *vgpu);
+int intel_vgpu_setup_submission(struct intel_vgpu *vgpu);
 
-void intel_vgpu_clean_gvt_context(struct intel_vgpu *vgpu);
+void intel_vgpu_clean_submission(struct intel_vgpu *vgpu);
 
 void release_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx);
 #endif
drivers/gpu/drm/i915/gvt/vgpu.c +3 -3
index 02c61a1..3d69871 100644
@@ -254,7 +254,7 @@ void intel_gvt_destroy_vgpu(struct intel_vgpu *vgpu)
 
 	idr_remove(&gvt->vgpu_idr, vgpu->id);
 	intel_vgpu_clean_sched_policy(vgpu);
-	intel_vgpu_clean_gvt_context(vgpu);
+	intel_vgpu_clean_submission(vgpu);
 	intel_vgpu_clean_execlist(vgpu);
 	intel_vgpu_clean_display(vgpu);
 	intel_vgpu_clean_opregion(vgpu);
@@ -376,7 +376,7 @@ static struct intel_vgpu *__intel_gvt_create_vgpu(struct intel_gvt *gvt,
 	if (ret)
 		goto out_clean_display;
 
-	ret = intel_vgpu_init_gvt_context(vgpu);
+	ret = intel_vgpu_setup_submission(vgpu);
 	if (ret)
 		goto out_clean_execlist;
 
@@ -389,7 +389,7 @@ static struct intel_vgpu *__intel_gvt_create_vgpu(struct intel_gvt *gvt,
 	return vgpu;
 
 out_clean_shadow_ctx:
-	intel_vgpu_clean_gvt_context(vgpu);
+	intel_vgpu_clean_submission(vgpu);
 out_clean_execlist:
 	intel_vgpu_clean_execlist(vgpu);
 out_clean_display:
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#2 lore raw

PATCH v2 3/9 drm/i915/gvt: Move workload cache init/clean into intel_vgpu_{setup, clean}_submission()

In reply to Zhi Wang

Move vGPU workload cache initialization/de-initialization into intel_vgpu_{setup, clean}_submission() since they are not specific to execlist stuffs.

---
 drivers/gpu/drm/i915/gvt/execlist.c  | 15 +--------------
 drivers/gpu/drm/i915/gvt/scheduler.c | 24 +++++++++++++++++++++++-
 2 files changed, 24 insertions(+), 15 deletions(-)
drivers/gpu/drm/i915/gvt/execlist.c +1 -14
index 9402aa5..bdf78ab 100644
@@ -857,14 +857,12 @@ void intel_vgpu_clean_execlist(struct intel_vgpu *vgpu)
 	struct intel_engine_cs *engine;
 
 	clean_workloads(vgpu, ALL_ENGINES);
-	kmem_cache_destroy(vgpu->workloads);
 
 	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
 		kfree(vgpu->reserve_ring_buffer_va[i]);
 		vgpu->reserve_ring_buffer_va[i] = NULL;
 		vgpu->reserve_ring_buffer_size[i] = 0;
 	}
-
 }
 
 #define RESERVE_RING_BUFFER_SIZE		((1 * PAGE_SIZE)/8)
@@ -873,19 +871,8 @@ int intel_vgpu_init_execlist(struct intel_vgpu *vgpu)
 	enum intel_engine_id i;
 	struct intel_engine_cs *engine;
 
-	/* each ring has a virtual execlist engine */
-	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
+	for_each_engine(engine, vgpu->gvt->dev_priv, i)
 		init_vgpu_execlist(vgpu, i);
-		INIT_LIST_HEAD(&vgpu->workload_q_head[i]);
-	}
-
-	vgpu->workloads = kmem_cache_create("gvt-g_vgpu_workload",
-			sizeof(struct intel_vgpu_workload), 0,
-			SLAB_HWCACHE_ALIGN,
-			NULL);
-
-	if (!vgpu->workloads)
-		return -ENOMEM;
 
 	/* each ring has a shadow ring buffer until vgpu destroyed */
 	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
drivers/gpu/drm/i915/gvt/scheduler.c +23 -1
index d14d910..c5d7baf 100644
@@ -718,6 +718,7 @@ int intel_gvt_init_workload_scheduler(struct intel_gvt *gvt)
 void intel_vgpu_clean_submission(struct intel_vgpu *vgpu)
 {
 	i915_gem_context_put(vgpu->shadow_ctx);
+	kmem_cache_destroy(vgpu->workloads);
 }
 
 /**
@@ -732,7 +733,9 @@ void intel_vgpu_clean_submission(struct intel_vgpu *vgpu)
  */
 int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
 {
-	atomic_set(&vgpu->running_workload_num, 0);
+	enum intel_engine_id i;
+	struct intel_engine_cs *engine;
+	int ret;
 
 	vgpu->shadow_ctx = i915_gem_context_create_gvt(
 			&vgpu->gvt->dev_priv->drm);
@@ -743,5 +746,24 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
 
 	bitmap_zero(vgpu->shadow_ctx_desc_updated, I915_NUM_ENGINES);
 
+	vgpu->workloads = kmem_cache_create("gvt-g_vgpu_workload",
+			sizeof(struct intel_vgpu_workload), 0,
+			SLAB_HWCACHE_ALIGN,
+			NULL);
+
+	if (!vgpu->workloads) {
+		ret = -ENOMEM;
+		goto out_shadow_ctx;
+	}
+
+	for_each_engine(engine, vgpu->gvt->dev_priv, i)
+		INIT_LIST_HEAD(&vgpu->workload_q_head[i]);
+
+	atomic_set(&vgpu->running_workload_num, 0);
+
 	return 0;
+
+out_shadow_ctx:
+	i915_gem_context_put(vgpu->shadow_ctx);
+	return ret;
 }
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#3 lore raw

PATCH v2 4/9 drm/i915/gvt: Introduce intel_vgpu_submission

In reply to Zhi Wang

Introduce intel_vgpu_submission to hold all members related to submission in struct intel_vgpu before.

---
 drivers/gpu/drm/i915/gvt/execlist.c  | 30 +++++++++-------
 drivers/gpu/drm/i915/gvt/gvt.h       | 17 +++++----
 drivers/gpu/drm/i915/gvt/handlers.c  |  2 +-
 drivers/gpu/drm/i915/gvt/kvmgt.c     |  2 +-
 drivers/gpu/drm/i915/gvt/render.c    |  9 ++---
 drivers/gpu/drm/i915/gvt/scheduler.c | 67 ++++++++++++++++++++----------------
 drivers/gpu/drm/i915/gvt/scheduler.h |  2 +-
 drivers/gpu/drm/i915/gvt/vgpu.c      |  4 +--
 8 files changed, 76 insertions(+), 57 deletions(-)
drivers/gpu/drm/i915/gvt/execlist.c +18 -12
index bdf78ab..d28bc89 100644
@@ -362,7 +362,7 @@ static void free_workload(struct intel_vgpu_workload *workload)
 {
 	intel_vgpu_unpin_mm(workload->shadow_mm);
 	intel_gvt_mm_unreference(workload->shadow_mm);
-	kmem_cache_free(workload->vgpu->workloads, workload);
+	kmem_cache_free(workload->vgpu->submission.workloads, workload);
 }
 
 #define get_desc_from_elsp_dwords(ed, i) \
@@ -401,7 +401,8 @@ static int update_wa_ctx_2_shadow_ctx(struct intel_shadow_wa_ctx *wa_ctx)
 					struct intel_vgpu_workload,
 					wa_ctx);
 	int ring_id = workload->ring_id;
-	struct i915_gem_context *shadow_ctx = workload->vgpu->shadow_ctx;
+	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;
@@ -474,6 +475,7 @@ static void release_shadow_batch_buffer(struct intel_vgpu_workload *workload)
 static int prepare_execlist_workload(struct intel_vgpu_workload *workload)
 {
 	struct intel_vgpu *vgpu = workload->vgpu;
+	struct intel_vgpu_submission *s = &vgpu->submission;
 	struct execlist_ctx_descriptor_format ctx[2];
 	int ring_id = workload->ring_id;
 	int ret;
@@ -514,7 +516,7 @@ static int prepare_execlist_workload(struct intel_vgpu_workload *workload)
 	ctx[0] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 0);
 	ctx[1] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 1);
 
-	ret = emulate_execlist_schedule_in(&vgpu->execlist[ring_id], ctx);
+	ret = emulate_execlist_schedule_in(&s->execlist[ring_id], ctx);
 	if (!ret)
 		goto out;
 	else
@@ -533,7 +535,8 @@ static int complete_execlist_workload(struct intel_vgpu_workload *workload)
 {
 	struct intel_vgpu *vgpu = workload->vgpu;
 	int ring_id = workload->ring_id;
-	struct intel_vgpu_execlist *execlist = &vgpu->execlist[ring_id];
+	struct intel_vgpu_submission *s = &vgpu->submission;
+	struct intel_vgpu_execlist *execlist = &s->execlist[ring_id];
 	struct intel_vgpu_workload *next_workload;
 	struct list_head *next = workload_q_head(vgpu, ring_id)->next;
 	bool lite_restore = false;
@@ -652,6 +655,7 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
 		struct execlist_ctx_descriptor_format *desc,
 		bool emulate_schedule_in)
 {
+	struct intel_vgpu_submission *s = &vgpu->submission;
 	struct list_head *q = workload_q_head(vgpu, ring_id);
 	struct intel_vgpu_workload *last_workload = get_last_workload(q);
 	struct intel_vgpu_workload *workload = NULL;
@@ -689,7 +693,7 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
 
 	gvt_dbg_el("ring id %d begin a new workload\n", ring_id);
 
-	workload = kmem_cache_zalloc(vgpu->workloads, GFP_KERNEL);
+	workload = kmem_cache_zalloc(s->workloads, GFP_KERNEL);
 	if (!workload)
 		return -ENOMEM;
 
@@ -739,7 +743,7 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
 	}
 
 	if (emulate_schedule_in)
-		workload->elsp_dwords = vgpu->execlist[ring_id].elsp_dwords;
+		workload->elsp_dwords = s->execlist[ring_id].elsp_dwords;
 
 	gvt_dbg_el("workload %p ring id %d head %x tail %x start %x ctl %x\n",
 			workload, ring_id, head, tail, start, ctl);
@@ -749,7 +753,7 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
 
 	ret = prepare_mm(workload);
 	if (ret) {
-		kmem_cache_free(vgpu->workloads, workload);
+		kmem_cache_free(s->workloads, workload);
 		return ret;
 	}
 
@@ -770,7 +774,8 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
 
 int intel_vgpu_submit_execlist(struct intel_vgpu *vgpu, int ring_id)
 {
-	struct intel_vgpu_execlist *execlist = &vgpu->execlist[ring_id];
+	struct intel_vgpu_submission *s = &vgpu->submission;
+	struct intel_vgpu_execlist *execlist = &s->execlist[ring_id];
 	struct execlist_ctx_descriptor_format *desc[2];
 	int i, ret;
 
@@ -812,7 +817,8 @@ int intel_vgpu_submit_execlist(struct intel_vgpu *vgpu, int ring_id)
 
 static void init_vgpu_execlist(struct intel_vgpu *vgpu, int ring_id)
 {
-	struct intel_vgpu_execlist *execlist = &vgpu->execlist[ring_id];
+	struct intel_vgpu_submission *s = &vgpu->submission;
+	struct intel_vgpu_execlist *execlist = &s->execlist[ring_id];
 	struct execlist_context_status_pointer_format ctx_status_ptr;
 	u32 ctx_status_ptr_reg;
 
@@ -834,6 +840,7 @@ static void init_vgpu_execlist(struct intel_vgpu *vgpu, int ring_id)
 
 static void clean_workloads(struct intel_vgpu *vgpu, unsigned long engine_mask)
 {
+	struct intel_vgpu_submission *s = &vgpu->submission;
 	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
 	struct intel_engine_cs *engine;
 	struct intel_vgpu_workload *pos, *n;
@@ -842,12 +849,11 @@ static void clean_workloads(struct intel_vgpu *vgpu, unsigned long engine_mask)
 	/* free the unsubmited workloads in the queues. */
 	for_each_engine_masked(engine, dev_priv, engine_mask, tmp) {
 		list_for_each_entry_safe(pos, n,
-			&vgpu->workload_q_head[engine->id], list) {
+			&s->workload_q_head[engine->id], list) {
 			list_del_init(&pos->list);
 			free_workload(pos);
 		}
-
-		clear_bit(engine->id, vgpu->shadow_ctx_desc_updated);
+		clear_bit(engine->id, s->shadow_ctx_desc_updated);
 	}
 }
 
drivers/gpu/drm/i915/gvt/gvt.h +11 -6
index 9c2e7c0..c3f84f2 100644
@@ -142,6 +142,15 @@ struct vgpu_sched_ctl {
 	int weight;
 };
 
+struct intel_vgpu_submission {
+	struct intel_vgpu_execlist execlist[I915_NUM_ENGINES];
+	struct list_head workload_q_head[I915_NUM_ENGINES];
+	struct kmem_cache *workloads;
+	atomic_t running_workload_num;
+	struct i915_gem_context *shadow_ctx;
+	DECLARE_BITMAP(shadow_ctx_desc_updated, I915_NUM_ENGINES);
+};
+
 struct intel_vgpu {
 	struct intel_gvt *gvt;
 	int id;
@@ -161,16 +170,12 @@ struct intel_vgpu {
 	struct intel_vgpu_gtt gtt;
 	struct intel_vgpu_opregion opregion;
 	struct intel_vgpu_display display;
-	struct intel_vgpu_execlist execlist[I915_NUM_ENGINES];
-	struct list_head workload_q_head[I915_NUM_ENGINES];
-	struct kmem_cache *workloads;
-	atomic_t running_workload_num;
+	struct intel_vgpu_submission submission;
 	/* 1/2K for each reserve ring buffer */
 	void *reserve_ring_buffer_va[I915_NUM_ENGINES];
 	int reserve_ring_buffer_size[I915_NUM_ENGINES];
 	DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES);
-	struct i915_gem_context *shadow_ctx;
-	DECLARE_BITMAP(shadow_ctx_desc_updated, I915_NUM_ENGINES);
+
 
 #if IS_ENABLED(CONFIG_DRM_I915_GVT_KVMGT)
 	struct {
drivers/gpu/drm/i915/gvt/handlers.c +1 -1
index dcee645..a1963fc 100644
@@ -1462,7 +1462,7 @@ static int elsp_mmio_write(struct intel_vgpu *vgpu, unsigned int offset,
 	if (WARN_ON(ring_id < 0 || ring_id > I915_NUM_ENGINES - 1))
 		return -EINVAL;
 
-	execlist = &vgpu->execlist[ring_id];
+	execlist = &vgpu->submission.execlist[ring_id];
 
 	execlist->elsp_dwords.data[3 - execlist->elsp_dwords.index] = data;
 	if (execlist->elsp_dwords.index == 3) {
drivers/gpu/drm/i915/gvt/kvmgt.c +1 -1
index 9606092..fc6878b 100644
@@ -1188,7 +1188,7 @@ hw_id_show(struct device *dev, struct device_attribute *attr,
 		struct intel_vgpu *vgpu = (struct intel_vgpu *)
 			mdev_get_drvdata(mdev);
 		return sprintf(buf, "%u\n",
-			       vgpu->shadow_ctx->hw_id);
+			       vgpu->submission.shadow_ctx->hw_id);
 	}
 	return sprintf(buf, "\n");
 }
drivers/gpu/drm/i915/gvt/render.c +5 -4
index 2ea5422..7ed76ae 100644
@@ -261,14 +261,15 @@ static void restore_mocs(struct intel_vgpu *vgpu, int ring_id)
 static void switch_mmio_to_vgpu(struct intel_vgpu *vgpu, int ring_id)
 {
 	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
-	struct render_mmio *mmio;
-	u32 v;
-	int i, array_size;
-	u32 *reg_state = vgpu->shadow_ctx->engine[ring_id].lrc_reg_state;
+	struct intel_vgpu_submission *s = &vgpu->submission;
+	u32 *reg_state = s->shadow_ctx->engine[ring_id].lrc_reg_state;
 	u32 ctx_ctrl = reg_state[CTX_CONTEXT_CONTROL_VAL];
 	u32 inhibit_mask =
 		_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
 	i915_reg_t last_reg = _MMIO(0);
+	struct render_mmio *mmio;
+	u32 v;
+	int i, array_size;
 
 	if (IS_SKYLAKE(vgpu->gvt->dev_priv)
 		|| IS_KABYLAKE(vgpu->gvt->dev_priv)) {
drivers/gpu/drm/i915/gvt/scheduler.c +37 -30

175 lines not on this page · raw message

drivers/gpu/drm/i915/gvt/scheduler.h +1 -1
index b6ca44e..c6e47ef 100644
@@ -121,7 +121,7 @@ struct intel_shadow_bb_entry {
 };
 
 #define workload_q_head(vgpu, ring_id) \
-	(&(vgpu->workload_q_head[ring_id]))
+	(&(vgpu->submission.workload_q_head[ring_id]))
 
 #define queue_workload(workload) do { \
 	list_add_tail(&workload->list, \
drivers/gpu/drm/i915/gvt/vgpu.c +2 -2
index 3d69871..35a5ec2 100644
@@ -226,7 +226,7 @@ void intel_gvt_deactivate_vgpu(struct intel_vgpu *vgpu)
 
 	vgpu->active = false;
 
-	if (atomic_read(&vgpu->running_workload_num)) {
+	if (atomic_read(&vgpu->submission.running_workload_num)) {
 		mutex_unlock(&gvt->lock);
 		intel_gvt_wait_vgpu_idle(vgpu);
 		mutex_lock(&gvt->lock);
@@ -293,7 +293,7 @@ struct intel_vgpu *intel_gvt_create_idle_vgpu(struct intel_gvt *gvt)
 	vgpu->gvt = gvt;
 
 	for (i = 0; i < I915_NUM_ENGINES; i++)
-		INIT_LIST_HEAD(&vgpu->workload_q_head[i]);
+		INIT_LIST_HEAD(&vgpu->submission.workload_q_head[i]);
 
 	ret = intel_vgpu_init_sched_policy(vgpu);
 	if (ret)
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#4 lore raw

PATCH v2 5/9 drm/i915/gvt: Move tlb_handle_pending into intel_vgpu_submission

In reply to Zhi Wang

Move tlb_handle_pending into intel_vgpu_submssion since it belongs to a part of vGPU submission stuffs

---
 drivers/gpu/drm/i915/gvt/gvt.h       | 2 +-
 drivers/gpu/drm/i915/gvt/handlers.c  | 2 +-
 drivers/gpu/drm/i915/gvt/render.c    | 3 ++-
 drivers/gpu/drm/i915/gvt/scheduler.c | 1 +
 drivers/gpu/drm/i915/gvt/vgpu.c      | 1 -
 5 files changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/gvt/gvt.h +1 -1
index c3f84f2..93ff530 100644
@@ -149,6 +149,7 @@ struct intel_vgpu_submission {
 	atomic_t running_workload_num;
 	struct i915_gem_context *shadow_ctx;
 	DECLARE_BITMAP(shadow_ctx_desc_updated, I915_NUM_ENGINES);
+	DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES);
 };
 
 struct intel_vgpu {
@@ -174,7 +175,6 @@ struct intel_vgpu {
 	/* 1/2K for each reserve ring buffer */
 	void *reserve_ring_buffer_va[I915_NUM_ENGINES];
 	int reserve_ring_buffer_size[I915_NUM_ENGINES];
-	DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES);
 
 
 #if IS_ENABLED(CONFIG_DRM_I915_GVT_KVMGT)
drivers/gpu/drm/i915/gvt/handlers.c +1 -1
index a1963fc..2ee494b 100644
@@ -1537,7 +1537,7 @@ static int gvt_reg_tlb_control_handler(struct intel_vgpu *vgpu,
 	default:
 		return -EINVAL;
 	}
-	set_bit(id, (void *)vgpu->tlb_handle_pending);
+	set_bit(id, (void *)vgpu->submission.tlb_handle_pending);
 
 	return 0;
 }
drivers/gpu/drm/i915/gvt/render.c +2 -1
index 7ed76ae..3910832 100644
@@ -147,6 +147,7 @@ static u32 gen9_render_mocs_L3[32];
 static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id)
 {
 	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
+	struct intel_vgpu_submission *s = &vgpu->submission;
 	enum forcewake_domains fw;
 	i915_reg_t reg;
 	u32 regs[] = {
@@ -160,7 +161,7 @@ static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id)
 	if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
 		return;
 
-	if (!test_and_clear_bit(ring_id, (void *)vgpu->tlb_handle_pending))
+	if (!test_and_clear_bit(ring_id, (void *)s->tlb_handle_pending))
 		return;
 
 	reg = _MMIO(regs[ring_id]);
drivers/gpu/drm/i915/gvt/scheduler.c +1 -0
index 3527daf..c880866 100644
@@ -767,6 +767,7 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
 		INIT_LIST_HEAD(&s->workload_q_head[i]);
 
 	atomic_set(&s->running_workload_num, 0);
+	bitmap_zero(s->tlb_handle_pending, I915_NUM_ENGINES);
 
 	return 0;
 
drivers/gpu/drm/i915/gvt/vgpu.c +0 -1
index 35a5ec2..1c9818d 100644
@@ -346,7 +346,6 @@ static struct intel_vgpu *__intel_gvt_create_vgpu(struct intel_gvt *gvt,
 	vgpu->handle = param->handle;
 	vgpu->gvt = gvt;
 	vgpu->sched_ctl.weight = param->weight;
-	bitmap_zero(vgpu->tlb_handle_pending, I915_NUM_ENGINES);
 
 	intel_vgpu_init_cfg_space(vgpu, param->primary);
 
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#5 lore raw

PATCH v2 6/9 drm/i915/gvt: Fix a memory leak in cmd_parser.c

In reply to Zhi Wang

The pointer points to the original memory can never take the return value of krealloc().

---
 drivers/gpu/drm/i915/gvt/cmd_parser.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/gvt/cmd_parser.c +6 -4
index 2c0ccbb..22d33be 100644
@@ -2620,14 +2620,16 @@ static int shadow_workload_ring_buffer(struct intel_vgpu_workload *workload)
 	gma_top = workload->rb_start + guest_rb_size;
 
 	if (workload->rb_len > vgpu->reserve_ring_buffer_size[ring_id]) {
-		void *va = vgpu->reserve_ring_buffer_va[ring_id];
+		void *va, *p;
+
 		/* realloc the new ring buffer if needed */
-		vgpu->reserve_ring_buffer_va[ring_id] =
-			krealloc(va, workload->rb_len, GFP_KERNEL);
-		if (!vgpu->reserve_ring_buffer_va[ring_id]) {
+		va = vgpu->reserve_ring_buffer_va[ring_id];
+		p = krealloc(va, workload->rb_len, GFP_KERNEL);
+		if (!p) {
 			gvt_vgpu_err("fail to alloc reserve ring buffer\n");
 			return -ENOMEM;
 		}
+		vgpu->reserve_ring_buffer_va[ring_id] = p;
 		vgpu->reserve_ring_buffer_size[ring_id] = workload->rb_len;
 	}
 
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#6 lore raw

PATCH v2 7/9 drm/i915/gvt: Rename reserved ring buffer

In reply to Zhi Wang

"reserved" means reserve something from somewhere. Actually they are buffers used by command scanner. Rename it to ring_scan_buffer.

v2:

- Remove the usage of an extra variable. (Zhenyu)

---
 drivers/gpu/drm/i915/gvt/cmd_parser.c | 16 ++++++++--------
 drivers/gpu/drm/i915/gvt/execlist.c   | 22 +++++++++++-----------
 drivers/gpu/drm/i915/gvt/gvt.h        |  6 +++---
 3 files changed, 22 insertions(+), 22 deletions(-)
drivers/gpu/drm/i915/gvt/cmd_parser.c +8 -8
index 22d33be..23922ac 100644
@@ -2619,21 +2619,21 @@ static int shadow_workload_ring_buffer(struct intel_vgpu_workload *workload)
 	gma_tail = workload->rb_start + workload->rb_tail;
 	gma_top = workload->rb_start + guest_rb_size;
 
-	if (workload->rb_len > vgpu->reserve_ring_buffer_size[ring_id]) {
-		void *va, *p;
+	if (workload->rb_len > vgpu->ring_scan_buffer_size[ring_id]) {
+		void *p;
 
 		/* realloc the new ring buffer if needed */
-		va = vgpu->reserve_ring_buffer_va[ring_id];
-		p = krealloc(va, workload->rb_len, GFP_KERNEL);
+		p = krealloc(vgpu->ring_scan_buffer[ring_id], workload->rb_len,
+				GFP_KERNEL);
 		if (!p) {
-			gvt_vgpu_err("fail to alloc reserve ring buffer\n");
+			gvt_vgpu_err("fail to re-alloc ring scan buffer\n");
 			return -ENOMEM;
 		}
-		vgpu->reserve_ring_buffer_va[ring_id] = p;
-		vgpu->reserve_ring_buffer_size[ring_id] = workload->rb_len;
+		vgpu->ring_scan_buffer[ring_id] = p;
+		vgpu->ring_scan_buffer_size[ring_id] = workload->rb_len;
 	}
 
-	shadow_ring_buffer_va = vgpu->reserve_ring_buffer_va[ring_id];
+	shadow_ring_buffer_va = vgpu->ring_scan_buffer[ring_id];
 
 	/* get shadow ring buffer va */
 	workload->shadow_ring_buffer_va = shadow_ring_buffer_va;
drivers/gpu/drm/i915/gvt/execlist.c +11 -11
index d28bc89..733be7b 100644
@@ -865,9 +865,9 @@ void intel_vgpu_clean_execlist(struct intel_vgpu *vgpu)
 	clean_workloads(vgpu, ALL_ENGINES);
 
 	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		kfree(vgpu->reserve_ring_buffer_va[i]);
-		vgpu->reserve_ring_buffer_va[i] = NULL;
-		vgpu->reserve_ring_buffer_size[i] = 0;
+		kfree(vgpu->ring_scan_buffer[i]);
+		vgpu->ring_scan_buffer[i] = NULL;
+		vgpu->ring_scan_buffer_size[i] = 0;
 	}
 }
 
@@ -882,21 +882,21 @@ int intel_vgpu_init_execlist(struct intel_vgpu *vgpu)
 
 	/* each ring has a shadow ring buffer until vgpu destroyed */
 	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		vgpu->reserve_ring_buffer_va[i] =
+		vgpu->ring_scan_buffer[i] =
 			kmalloc(RESERVE_RING_BUFFER_SIZE, GFP_KERNEL);
-		if (!vgpu->reserve_ring_buffer_va[i]) {
-			gvt_vgpu_err("fail to alloc reserve ring buffer\n");
+		if (!vgpu->ring_scan_buffer[i]) {
+			gvt_vgpu_err("fail to alloc ring scan buffer\n");
 			goto out;
 		}
-		vgpu->reserve_ring_buffer_size[i] = RESERVE_RING_BUFFER_SIZE;
+		vgpu->ring_scan_buffer_size[i] = RESERVE_RING_BUFFER_SIZE;
 	}
 	return 0;
 out:
 	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		if (vgpu->reserve_ring_buffer_size[i]) {
-			kfree(vgpu->reserve_ring_buffer_va[i]);
-			vgpu->reserve_ring_buffer_va[i] = NULL;
-			vgpu->reserve_ring_buffer_size[i] = 0;
+		if (vgpu->ring_scan_buffer_size[i]) {
+			kfree(vgpu->ring_scan_buffer[i]);
+			vgpu->ring_scan_buffer[i] = NULL;
+			vgpu->ring_scan_buffer_size[i] = 0;
 		}
 	}
 	return -ENOMEM;
drivers/gpu/drm/i915/gvt/gvt.h +3 -3
index 93ff530..5b723fa 100644
@@ -172,9 +172,9 @@ struct intel_vgpu {
 	struct intel_vgpu_opregion opregion;
 	struct intel_vgpu_display display;
 	struct intel_vgpu_submission submission;
-	/* 1/2K for each reserve ring buffer */
-	void *reserve_ring_buffer_va[I915_NUM_ENGINES];
-	int reserve_ring_buffer_size[I915_NUM_ENGINES];
+	/* 1/2K for each engine */
+	void *ring_scan_buffer[I915_NUM_ENGINES];
+	int ring_scan_buffer_size[I915_NUM_ENGINES];
 
 
 #if IS_ENABLED(CONFIG_DRM_I915_GVT_KVMGT)
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#7 lore raw

PATCH v2 8/9 drm/i915/gvt: Move ring scan buffers into intel_vgpu_submission

In reply to Zhi Wang

Move ring scan buffers into intel_vgpu_submission since they belongs to a part of vGPU submission stuffs.

---
 drivers/gpu/drm/i915/gvt/cmd_parser.c | 11 ++++++-----
 drivers/gpu/drm/i915/gvt/execlist.c   | 23 +++++++++++++----------
 drivers/gpu/drm/i915/gvt/gvt.h        |  7 +++----
 3 files changed, 22 insertions(+), 19 deletions(-)
drivers/gpu/drm/i915/gvt/cmd_parser.c +6 -5
index 23922ac..136367f 100644
@@ -2604,6 +2604,7 @@ static int scan_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
 static int shadow_workload_ring_buffer(struct intel_vgpu_workload *workload)
 {
 	struct intel_vgpu *vgpu = workload->vgpu;
+	struct intel_vgpu_submission *s = &vgpu->submission;
 	unsigned long gma_head, gma_tail, gma_top, guest_rb_size;
 	void *shadow_ring_buffer_va;
 	int ring_id = workload->ring_id;
@@ -2619,21 +2620,21 @@ static int shadow_workload_ring_buffer(struct intel_vgpu_workload *workload)
 	gma_tail = workload->rb_start + workload->rb_tail;
 	gma_top = workload->rb_start + guest_rb_size;
 
-	if (workload->rb_len > vgpu->ring_scan_buffer_size[ring_id]) {
+	if (workload->rb_len > s->ring_scan_buffer_size[ring_id]) {
 		void *p;
 
 		/* realloc the new ring buffer if needed */
-		p = krealloc(vgpu->ring_scan_buffer[ring_id], workload->rb_len,
+		p = krealloc(s->ring_scan_buffer[ring_id], workload->rb_len,
 				GFP_KERNEL);
 		if (!p) {
 			gvt_vgpu_err("fail to re-alloc ring scan buffer\n");
 			return -ENOMEM;
 		}
-		vgpu->ring_scan_buffer[ring_id] = p;
-		vgpu->ring_scan_buffer_size[ring_id] = workload->rb_len;
+		s->ring_scan_buffer[ring_id] = p;
+		s->ring_scan_buffer_size[ring_id] = workload->rb_len;
 	}
 
-	shadow_ring_buffer_va = vgpu->ring_scan_buffer[ring_id];
+	shadow_ring_buffer_va = s->ring_scan_buffer[ring_id];
 
 	/* get shadow ring buffer va */
 	workload->shadow_ring_buffer_va = shadow_ring_buffer_va;
drivers/gpu/drm/i915/gvt/execlist.c +13 -10
index 733be7b..fae7459 100644
@@ -865,15 +865,18 @@ void intel_vgpu_clean_execlist(struct intel_vgpu *vgpu)
 	clean_workloads(vgpu, ALL_ENGINES);
 
 	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		kfree(vgpu->ring_scan_buffer[i]);
-		vgpu->ring_scan_buffer[i] = NULL;
-		vgpu->ring_scan_buffer_size[i] = 0;
+		struct intel_vgpu_submission *s = &vgpu->submission;
+
+		kfree(s->ring_scan_buffer[i]);
+		s->ring_scan_buffer[i] = NULL;
+		s->ring_scan_buffer_size[i] = 0;
 	}
 }
 
 #define RESERVE_RING_BUFFER_SIZE		((1 * PAGE_SIZE)/8)
 int intel_vgpu_init_execlist(struct intel_vgpu *vgpu)
 {
+	struct intel_vgpu_submission *s = &vgpu->submission;
 	enum intel_engine_id i;
 	struct intel_engine_cs *engine;
 
@@ -882,21 +885,21 @@ int intel_vgpu_init_execlist(struct intel_vgpu *vgpu)
 
 	/* each ring has a shadow ring buffer until vgpu destroyed */
 	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		vgpu->ring_scan_buffer[i] =
+		s->ring_scan_buffer[i] =
 			kmalloc(RESERVE_RING_BUFFER_SIZE, GFP_KERNEL);
-		if (!vgpu->ring_scan_buffer[i]) {
+		if (!s->ring_scan_buffer[i]) {
 			gvt_vgpu_err("fail to alloc ring scan buffer\n");
 			goto out;
 		}
-		vgpu->ring_scan_buffer_size[i] = RESERVE_RING_BUFFER_SIZE;
+		s->ring_scan_buffer_size[i] = RESERVE_RING_BUFFER_SIZE;
 	}
 	return 0;
 out:
 	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		if (vgpu->ring_scan_buffer_size[i]) {
-			kfree(vgpu->ring_scan_buffer[i]);
-			vgpu->ring_scan_buffer[i] = NULL;
-			vgpu->ring_scan_buffer_size[i] = 0;
+		if (s->ring_scan_buffer_size[i]) {
+			kfree(s->ring_scan_buffer[i]);
+			s->ring_scan_buffer[i] = NULL;
+			s->ring_scan_buffer_size[i] = 0;
 		}
 	}
 	return -ENOMEM;
drivers/gpu/drm/i915/gvt/gvt.h +3 -4
index 5b723fa..49fe548 100644
@@ -150,6 +150,9 @@ struct intel_vgpu_submission {
 	struct i915_gem_context *shadow_ctx;
 	DECLARE_BITMAP(shadow_ctx_desc_updated, I915_NUM_ENGINES);
 	DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES);
+	/* 1/2K for each engine */
+	void *ring_scan_buffer[I915_NUM_ENGINES];
+	int ring_scan_buffer_size[I915_NUM_ENGINES];
 };
 
 struct intel_vgpu {
@@ -172,10 +175,6 @@ struct intel_vgpu {
 	struct intel_vgpu_opregion opregion;
 	struct intel_vgpu_display display;
 	struct intel_vgpu_submission submission;
-	/* 1/2K for each engine */
-	void *ring_scan_buffer[I915_NUM_ENGINES];
-	int ring_scan_buffer_size[I915_NUM_ENGINES];
-
 
 #if IS_ENABLED(CONFIG_DRM_I915_GVT_KVMGT)
 	struct {
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#8 lore raw

PATCH v2 9/9 drm/i915/gvt: Do not allocate initial ring scan buffer

In reply to Zhi Wang

Theoretically, the largest bulk of commands in the ring buffer of an engine might be the first submission, which usually contains a lot of commands to initialize the HW. After removing the initial allocation of the ring scan buffer and let krealloc() do everything we need, we still have a big chance to get the buffer of suitable size in the first submission.

Tested on my SKL NUC.

---
 drivers/gpu/drm/i915/gvt/execlist.c | 21 ---------------------
 drivers/gpu/drm/i915/gvt/gvt.h      |  1 -
 2 files changed, 22 deletions(-)
drivers/gpu/drm/i915/gvt/execlist.c +0 -21
index fae7459..86d3cc8 100644
@@ -873,36 +873,15 @@ void intel_vgpu_clean_execlist(struct intel_vgpu *vgpu)
 	}
 }
 
-#define RESERVE_RING_BUFFER_SIZE		((1 * PAGE_SIZE)/8)
 int intel_vgpu_init_execlist(struct intel_vgpu *vgpu)
 {
-	struct intel_vgpu_submission *s = &vgpu->submission;
 	enum intel_engine_id i;
 	struct intel_engine_cs *engine;
 
 	for_each_engine(engine, vgpu->gvt->dev_priv, i)
 		init_vgpu_execlist(vgpu, i);
 
-	/* each ring has a shadow ring buffer until vgpu destroyed */
-	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		s->ring_scan_buffer[i] =
-			kmalloc(RESERVE_RING_BUFFER_SIZE, GFP_KERNEL);
-		if (!s->ring_scan_buffer[i]) {
-			gvt_vgpu_err("fail to alloc ring scan buffer\n");
-			goto out;
-		}
-		s->ring_scan_buffer_size[i] = RESERVE_RING_BUFFER_SIZE;
-	}
 	return 0;
-out:
-	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		if (s->ring_scan_buffer_size[i]) {
-			kfree(s->ring_scan_buffer[i]);
-			s->ring_scan_buffer[i] = NULL;
-			s->ring_scan_buffer_size[i] = 0;
-		}
-	}
-	return -ENOMEM;
 }
 
 void intel_vgpu_reset_execlist(struct intel_vgpu *vgpu,
drivers/gpu/drm/i915/gvt/gvt.h +0 -1
index 49fe548..7a770b1 100644
@@ -150,7 +150,6 @@ struct intel_vgpu_submission {
 	struct i915_gem_context *shadow_ctx;
 	DECLARE_BITMAP(shadow_ctx_desc_updated, I915_NUM_ENGINES);
 	DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES);
-	/* 1/2K for each engine */
 	void *ring_scan_buffer[I915_NUM_ENGINES];
 	int ring_scan_buffer_size[I915_NUM_ENGINES];
 };
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#9 lore raw

Re: [PATCH v2 9/9] drm/i915/gvt: Do not allocate initial ring scan buffer

Ah... sorry. This patchset shouldn't be here.

-----Original Message-----
From: Wang, Zhi A
Sent: Tuesday, September 12, 2017 8:40 AM
To: [email protected]
Cc: Wang, Zhi A <[email protected]>
Subject: [PATCH v2 9/9] drm/i915/gvt: Do not allocate initial ring scan buffer

Theoretically, the largest bulk of commands in the ring buffer of an engine might be the first submission, which usually contains a lot of commands to initialize the HW. After removing the initial allocation of the ring scan buffer and let krealloc() do everything we need, we still have a big chance to get the buffer of suitable size in the first submission.

Tested on my SKL NUC.

---
 drivers/gpu/drm/i915/gvt/execlist.c | 21 ---------------------
 drivers/gpu/drm/i915/gvt/gvt.h      |  1 -
 2 files changed, 22 deletions(-)
drivers/gpu/drm/i915/gvt/execlist.c +1 -23
index fae7459..86d3cc8 100644
@@ -873,36 +873,15 @@ void intel_vgpu_clean_execlist(struct intel_vgpu *vgpu)
 	}
 }
 
-#define RESERVE_RING_BUFFER_SIZE		((1 * PAGE_SIZE)/8)
 int intel_vgpu_init_execlist(struct intel_vgpu *vgpu)  {
-	struct intel_vgpu_submission *s = &vgpu->submission;
 	enum intel_engine_id i;
 	struct intel_engine_cs *engine;
 
 	for_each_engine(engine, vgpu->gvt->dev_priv, i)
 		init_vgpu_execlist(vgpu, i);
 
-	/* each ring has a shadow ring buffer until vgpu destroyed */
-	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		s->ring_scan_buffer[i] =
-			kmalloc(RESERVE_RING_BUFFER_SIZE, GFP_KERNEL);
-		if (!s->ring_scan_buffer[i]) {
-			gvt_vgpu_err("fail to alloc ring scan buffer\n");
-			goto out;
-		}
-		s->ring_scan_buffer_size[i] = RESERVE_RING_BUFFER_SIZE;
-	}
 	return 0;
-out:
-	for_each_engine(engine, vgpu->gvt->dev_priv, i) {
-		if (s->ring_scan_buffer_size[i]) {
-			kfree(s->ring_scan_buffer[i]);
-			s->ring_scan_buffer[i] = NULL;
-			s->ring_scan_buffer_size[i] = 0;
-		}
-	}
-	return -ENOMEM;
 }
 
 void intel_vgpu_reset_execlist(struct intel_vgpu *vgpu, diff --git a/drivers/gpu/drm/i915/gvt/gvt.h b/drivers/gpu/drm/i915/gvt/gvt.h index 49fe548..7a770b1 100644
--- a/drivers/gpu/drm/i915/gvt/gvt.h
+++ b/drivers/gpu/drm/i915/gvt/gvt.h
@@ -150,7 +150,6 @@ struct intel_vgpu_submission {
 	struct i915_gem_context *shadow_ctx;
 	DECLARE_BITMAP(shadow_ctx_desc_updated, I915_NUM_ENGINES);
 	DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES);
-	/* 1/2K for each engine */
 	void *ring_scan_buffer[I915_NUM_ENGINES];
 	int ring_scan_buffer_size[I915_NUM_ENGINES];
 };

--
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#10 lore raw

✓ Fi.CI.BAT: success for series starting with [v2,1/9] drm/i915/gvt: Make elsp_dwords in the right order

In reply to Zhi Wang
== Series Details ==

== Series Details ==

Series: series starting with [v2,1/9] drm/i915/gvt: Make elsp_dwords in the right order
URL : https://patchwork.freedesktop.org/series/30164/
State : success

== Summary ==

Series 30164v1 series starting with [v2,1/9] drm/i915/gvt: Make elsp_dwords in the right order
https://patchwork.freedesktop.org/api/1.0/series/30164/revisions/1/mbox/

Test chamelium:

        Subgroup dp-edid-read:
                fail       -> PASS       (fi-kbl-7500u)
Test gem_ringfill:
        Subgroup basic-default-hang:
                incomplete -> DMESG-WARN (fi-pnv-d510) fdo#101600
Test kms_cursor_legacy:
        Subgroup basic-busy-flip-before-cursor-atomic:
                pass       -> FAIL       (fi-snb-2600) fdo#100215 +1
Test kms_flip:
        Subgroup basic-flip-vs-modeset:
                skip       -> PASS       (fi-skl-x1585l) fdo#101781
Test kms_frontbuffer_tracking:
        Subgroup basic:
                dmesg-warn -> PASS       (fi-bdw-5557u) fdo#102473

fdo#101600 https://bugs.freedesktop.org/show_bug.cgi?id=101600
fdo#100215 https://bugs.freedesktop.org/show_bug.cgi?id=100215
fdo#101781 https://bugs.freedesktop.org/show_bug.cgi?id=101781
fdo#102473 https://bugs.freedesktop.org/show_bug.cgi?id=102473

fi-bdw-5557u     total:289  pass:268  dwarn:0   dfail:0   fail:0   skip:21  time:446s
fi-bdw-gvtdvm    total:289  pass:265  dwarn:0   dfail:0   fail:0   skip:24  time:453s
fi-blb-e6850     total:289  pass:224  dwarn:1   dfail:0   fail:0   skip:64  time:375s
fi-bsw-n3050     total:289  pass:243  dwarn:0   dfail:0   fail:0   skip:46  time:525s
fi-bwr-2160      total:289  pass:184  dwarn:0   dfail:0   fail:0   skip:105 time:269s
fi-bxt-j4205     total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:517s
fi-byt-j1900     total:289  pass:254  dwarn:1   dfail:0   fail:0   skip:34  time:499s
fi-byt-n2820     total:289  pass:250  dwarn:1   dfail:0   fail:0   skip:38  time:494s
fi-cfl-s         total:289  pass:250  dwarn:4   dfail:0   fail:0   skip:35  time:450s
fi-elk-e7500     total:289  pass:230  dwarn:0   dfail:0   fail:0   skip:59  time:451s
fi-glk-2a        total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:598s
fi-hsw-4770      total:289  pass:263  dwarn:0   dfail:0   fail:0   skip:26  time:430s
fi-hsw-4770r     total:289  pass:263  dwarn:0   dfail:0   fail:0   skip:26  time:407s
fi-ilk-650       total:289  pass:229  dwarn:0   dfail:0   fail:0   skip:60  time:440s
fi-ivb-3520m     total:289  pass:261  dwarn:0   dfail:0   fail:0   skip:28  time:488s
fi-ivb-3770      total:289  pass:261  dwarn:0   dfail:0   fail:0   skip:28  time:464s
fi-kbl-7500u     total:289  pass:264  dwarn:1   dfail:0   fail:0   skip:24  time:494s
fi-kbl-7560u     total:289  pass:270  dwarn:0   dfail:0   fail:0   skip:19  time:586s
fi-kbl-r         total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:586s
fi-pnv-d510      total:289  pass:223  dwarn:1   dfail:0   fail:0   skip:65  time:549s
fi-skl-6260u     total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:451s
fi-skl-6700k     total:289  pass:265  dwarn:0   dfail:0   fail:0   skip:24  time:520s
fi-skl-6770hq    total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:499s
fi-skl-gvtdvm    total:289  pass:266  dwarn:0   dfail:0   fail:0   skip:23  time:458s
fi-skl-x1585l    total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:489s
fi-snb-2520m     total:289  pass:251  dwarn:0   dfail:0   fail:0   skip:38  time:568s
fi-snb-2600      total:289  pass:249  dwarn:0   dfail:0   fail:1   skip:39  time:421s

3cbfafbde5bb92b6ecdbfd47e189d2742aa6fbec drm-tip: 2017y-09m-11d-23h-04m-21s UTC integration manifest e519b3fb4ddb drm/i915/gvt: Do not allocate initial ring scan buffer
132e169bcb33 drm/i915/gvt: Move ring scan buffers into intel_vgpu_submission 4f2f14ab38b2 drm/i915/gvt: Rename reserved ring buffer
3dab59fa6ba7 drm/i915/gvt: Fix a memory leak in cmd_parser.c
d4be4ef3a152 drm/i915/gvt: Move tlb_handle_pending into intel_vgpu_submission b5f5c6684678 drm/i915/gvt: Introduce intel_vgpu_submission
4cf1e7d0b878 drm/i915/gvt: Move workload cache init/clean into intel_vgpu_{setup, clean}_submission() bac89f98cad3 drm/i915/gvt: Rename intel_vgpu_{init, clean}_gvt_context()
8912340c75ad drm/i915/gvt: Make elsp_dwords in the right order

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_5650/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#11 lore raw

✓ Fi.CI.IGT: success for series starting with [v2,1/9] drm/i915/gvt: Make elsp_dwords in the right order

In reply to Zhi Wang
== Series Details ==

== Series Details ==

Series: series starting with [v2,1/9] drm/i915/gvt: Make elsp_dwords in the right order
URL : https://patchwork.freedesktop.org/series/30164/
State : success

== Summary ==

Test kms_setmode:

        Subgroup basic:
                pass       -> FAIL       (shard-hsw) fdo#99912

fdo#99912 https://bugs.freedesktop.org/show_bug.cgi?id=99912

shard-hsw total:2301 pass:1235 dwarn:0 dfail:0 fail:14 skip:1052 time:9348s

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_5650/shards.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx