diff mbox series

[RFC,13/42] drm/i915: support object clearing via blitter engine

Message ID 20190214145740.14521-14-matthew.auld@intel.com (mailing list archive)
State New, archived
Headers show
Series Introduce memory region concept (including device local memory) | expand

Commit Message

Matthew Auld Feb. 14, 2019, 2:57 p.m. UTC
Support clearing objects via the blitter engine. This is needed for LMEM
where we need to clear the backing store before handing the object to
userspace.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h               |   2 +
 drivers/gpu/drm/i915/i915_gem.c               | 151 ++++++++++++++++++
 drivers/gpu/drm/i915/intel_gpu_commands.h     |   1 +
 .../gpu/drm/i915/selftests/i915_gem_object.c  |  70 ++++++++
 4 files changed, 224 insertions(+)

Comments

Chris Wilson Feb. 14, 2019, 3:37 p.m. UTC | #1
Quoting Matthew Auld (2019-02-14 14:57:11)
> diff --git a/drivers/gpu/drm/i915/intel_gpu_commands.h b/drivers/gpu/drm/i915/intel_gpu_commands.h
> index b96a31bc1080..f74ff1d095c2 100644
> --- a/drivers/gpu/drm/i915/intel_gpu_commands.h
> +++ b/drivers/gpu/drm/i915/intel_gpu_commands.h
> @@ -175,6 +175,7 @@
>  #define GFX_OP_DRAWRECT_INFO_I965  ((0x7900<<16)|0x2)
>  
>  #define COLOR_BLT_CMD                  (2<<29 | 0x40<<22 | (5-2))
> +#define XY_COLOR_BLT_CMD               (2<<29 | 0x50<<22 | (7-2))

Remove the dword lengths here. The XY commands are not gen8 specific,
and construction is much neater if we just add the pktlen on writing the
cmd.

>  #define SRC_COPY_BLT_CMD               ((2<<29)|(0x43<<22)|4)
>  #define XY_SRC_COPY_BLT_CMD            ((2<<29)|(0x53<<22)|6)
>  #define XY_MONO_SRC_COPY_IMM_BLT       ((2<<29)|(0x71<<22)|5)
Chris Wilson Feb. 14, 2019, 3:38 p.m. UTC | #2
Quoting Matthew Auld (2019-02-14 14:57:11)
> +static struct i915_vma *
> +__i915_gem_fill_blt(struct i915_vma *vma, u32 value)
> +{
> +       struct drm_i915_private *i915 = to_i915(vma->obj->base.dev);
> +       const int gen = INTEL_GEN(i915);
> +       struct drm_i915_gem_object *obj;
> +       struct i915_vma *batch;
> +       u32 *cmd;
> +       int err;
> +
> +       obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
> +       if (IS_ERR(obj))
> +               return ERR_CAST(obj);
> +
> +       cmd = i915_gem_object_pin_map(obj, I915_MAP_WB);
> +       if (IS_ERR(cmd)) {
> +               err = PTR_ERR(cmd);
> +               goto err;
> +       }
> +
> +       if (gen >= 8) {
> +               *cmd++ = XY_COLOR_BLT_CMD | BLT_WRITE_RGBA;
> +               *cmd++ = BLT_DEPTH_32 | BLT_ROP_COLOR_COPY | PAGE_SIZE;
> +               *cmd++ = 0;
> +               *cmd++ = vma->obj->base.size >> PAGE_SHIFT << 16 | PAGE_SIZE / 4;
> +               *cmd++ = lower_32_bits(vma->node.start);
> +               *cmd++ = upper_32_bits(vma->node.start);
> +               *cmd++ = value;
> +               *cmd++ = MI_NOOP;
> +       } else {
> +               *cmd++ = COLOR_BLT_CMD | BLT_WRITE_RGBA;
> +               *cmd++ = BLT_DEPTH_32 | BLT_ROP_COLOR_COPY | PAGE_SIZE;
> +               *cmd++ = vma->obj->base.size >> PAGE_SHIFT << 16 | PAGE_SIZE;
> +               *cmd++ = vma->node.start;
> +               *cmd++ = value;
> +               *cmd++ = MI_NOOP;
> +       }
> +
> +       *cmd = MI_BATCH_BUFFER_END;

Why did you create a batch? Just issue the command from inside the
request.
-Chris
Chris Wilson Feb. 14, 2019, 3:40 p.m. UTC | #3
Quoting Matthew Auld (2019-02-14 14:57:11)
> +static int i915_gem_object_fill_blt(struct i915_gem_context *ctx,
> +                                   struct drm_i915_gem_object *obj,
> +                                   u32 value)
> +{
> +       struct drm_i915_private *i915 = to_i915(obj->base.dev);
> +       struct i915_address_space *vm = ctx->ppgtt ? &ctx->ppgtt->vm : &i915->ggtt.vm;
> +       struct i915_vma *vma;
> +       int err;
> +
> +       lockdep_assert_held(&i915->drm.struct_mutex);
> +
> +       vma = i915_vma_instance(obj, vm, NULL);
> +       if (IS_ERR(vma))
> +               return PTR_ERR(vma);
> +
> +       err = i915_vma_pin(vma, 0, 0, PIN_USER);
> +       if (err) {
> +               i915_vma_close(vma);
> +               return err;
> +       }
> +
> +       err = i915_gem_fill_blt(ctx, vma, value);
> +       i915_vma_unpin(vma);
> +       if (err)
> +               return err;
> +
> +       return i915_gem_object_wait(obj,
> +                                   I915_WAIT_LOCKED |
> +                                   I915_WAIT_ALL,
> +                                   MAX_SCHEDULE_TIMEOUT);

Oh no, don't even start by making this synchronous.
-Chris
diff mbox series

Patch

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index a0ed000d52aa..feec3bab0f5f 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2807,6 +2807,8 @@  void *i915_gem_object_alloc(struct drm_i915_private *dev_priv);
 void i915_gem_object_free(struct drm_i915_gem_object *obj);
 void i915_gem_object_init(struct drm_i915_gem_object *obj,
 			 const struct drm_i915_gem_object_ops *ops);
+int i915_gem_object_clear_blt(struct i915_gem_context *ctx,
+			      struct drm_i915_gem_object *obj);
 struct drm_i915_gem_object *
 i915_gem_object_create(struct drm_i915_private *dev_priv, u64 size);
 struct drm_i915_gem_object *
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 7f044b643a75..032d4334c0f1 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -4288,6 +4288,157 @@  static bool discard_backing_storage(struct drm_i915_gem_object *obj)
 	return atomic_long_read(&obj->base.filp->f_count) == 1;
 }
 
+static struct i915_vma *
+__i915_gem_fill_blt(struct i915_vma *vma, u32 value)
+{
+	struct drm_i915_private *i915 = to_i915(vma->obj->base.dev);
+	const int gen = INTEL_GEN(i915);
+	struct drm_i915_gem_object *obj;
+	struct i915_vma *batch;
+	u32 *cmd;
+	int err;
+
+	obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
+	if (IS_ERR(obj))
+		return ERR_CAST(obj);
+
+	cmd = i915_gem_object_pin_map(obj, I915_MAP_WB);
+	if (IS_ERR(cmd)) {
+		err = PTR_ERR(cmd);
+		goto err;
+	}
+
+	if (gen >= 8) {
+		*cmd++ = XY_COLOR_BLT_CMD | BLT_WRITE_RGBA;
+		*cmd++ = BLT_DEPTH_32 | BLT_ROP_COLOR_COPY | PAGE_SIZE;
+		*cmd++ = 0;
+		*cmd++ = vma->obj->base.size >> PAGE_SHIFT << 16 | PAGE_SIZE / 4;
+		*cmd++ = lower_32_bits(vma->node.start);
+		*cmd++ = upper_32_bits(vma->node.start);
+		*cmd++ = value;
+		*cmd++ = MI_NOOP;
+	} else {
+		*cmd++ = COLOR_BLT_CMD | BLT_WRITE_RGBA;
+		*cmd++ = BLT_DEPTH_32 | BLT_ROP_COLOR_COPY | PAGE_SIZE;
+		*cmd++ = vma->obj->base.size >> PAGE_SHIFT << 16 | PAGE_SIZE;
+		*cmd++ = vma->node.start;
+		*cmd++ = value;
+		*cmd++ = MI_NOOP;
+	}
+
+	*cmd = MI_BATCH_BUFFER_END;
+
+	i915_gem_object_unpin_map(obj);
+
+	err = i915_gem_object_set_to_gtt_domain(obj, false);
+	if (err)
+		goto err;
+
+	batch = i915_vma_instance(obj, vma->vm, NULL);
+	if (IS_ERR(batch)) {
+		err = PTR_ERR(batch);
+		goto err;
+	}
+
+	err = i915_vma_pin(batch, 0, 0, PIN_USER);
+	if (err)
+		goto err;
+
+	return batch;
+
+err:
+	i915_gem_object_put(obj);
+	return ERR_PTR(err);
+}
+
+static int i915_gem_fill_blt(struct i915_gem_context *ctx,
+			     struct i915_vma *vma,
+			     u32 value)
+{
+	struct drm_i915_private *i915 = to_i915(vma->obj->base.dev);
+	struct intel_engine_cs *engine = i915->engine[BCS];
+	struct i915_request *rq;
+	struct i915_vma *batch;
+	int flags = 0;
+	int err;
+
+	err = i915_gem_object_set_to_gtt_domain(vma->obj, true);
+	if (err)
+		return err;
+
+	rq = i915_request_alloc(engine, ctx);
+	if (IS_ERR(rq))
+		return PTR_ERR(rq);
+
+	batch = __i915_gem_fill_blt(vma, value);
+	if (IS_ERR(batch)) {
+		err = PTR_ERR(batch);
+		goto err_request;
+	}
+
+	err = i915_vma_move_to_active(batch, rq, 0);
+	i915_vma_unpin(batch);
+	i915_vma_close(batch);
+	if (err) {
+		i915_gem_object_put(batch->obj);
+		goto err_request;
+	}
+
+	i915_gem_object_set_active_reference(batch->obj);
+
+	err = engine->emit_bb_start(rq,
+				    batch->node.start, batch->node.size,
+				    flags);
+	if (err)
+		goto err_request;
+
+	err = i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
+	if (err)
+		i915_request_skip(rq, err);
+
+err_request:
+	i915_request_add(rq);
+	return err;
+}
+
+static int i915_gem_object_fill_blt(struct i915_gem_context *ctx,
+				    struct drm_i915_gem_object *obj,
+				    u32 value)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+	struct i915_address_space *vm = ctx->ppgtt ? &ctx->ppgtt->vm : &i915->ggtt.vm;
+	struct i915_vma *vma;
+	int err;
+
+	lockdep_assert_held(&i915->drm.struct_mutex);
+
+	vma = i915_vma_instance(obj, vm, NULL);
+	if (IS_ERR(vma))
+		return PTR_ERR(vma);
+
+	err = i915_vma_pin(vma, 0, 0, PIN_USER);
+	if (err) {
+		i915_vma_close(vma);
+		return err;
+	}
+
+	err = i915_gem_fill_blt(ctx, vma, value);
+	i915_vma_unpin(vma);
+	if (err)
+		return err;
+
+	return i915_gem_object_wait(obj,
+				    I915_WAIT_LOCKED |
+				    I915_WAIT_ALL,
+				    MAX_SCHEDULE_TIMEOUT);
+}
+
+int i915_gem_object_clear_blt(struct i915_gem_context *ctx,
+			      struct drm_i915_gem_object *obj)
+{
+	return i915_gem_object_fill_blt(ctx, obj, 0);
+}
+
 static void __i915_gem_free_objects(struct drm_i915_private *i915,
 				    struct llist_node *freed)
 {
diff --git a/drivers/gpu/drm/i915/intel_gpu_commands.h b/drivers/gpu/drm/i915/intel_gpu_commands.h
index b96a31bc1080..f74ff1d095c2 100644
--- a/drivers/gpu/drm/i915/intel_gpu_commands.h
+++ b/drivers/gpu/drm/i915/intel_gpu_commands.h
@@ -175,6 +175,7 @@ 
 #define GFX_OP_DRAWRECT_INFO_I965  ((0x7900<<16)|0x2)
 
 #define COLOR_BLT_CMD			(2<<29 | 0x40<<22 | (5-2))
+#define XY_COLOR_BLT_CMD              	(2<<29 | 0x50<<22 | (7-2))
 #define SRC_COPY_BLT_CMD		((2<<29)|(0x43<<22)|4)
 #define XY_SRC_COPY_BLT_CMD		((2<<29)|(0x53<<22)|6)
 #define XY_MONO_SRC_COPY_IMM_BLT	((2<<29)|(0x71<<22)|5)
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_object.c b/drivers/gpu/drm/i915/selftests/i915_gem_object.c
index 395ae878e0f7..c83dc5e2f219 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_object.c
@@ -628,6 +628,75 @@  static int igt_mmap_offset_exhaustion(void *arg)
 	goto out;
 }
 
+static int igt_fill_blt(void *arg)
+{
+	struct drm_i915_private *i915 = arg;
+	struct drm_i915_gem_object *obj;
+	struct i915_gem_context *ctx;
+	struct drm_file *file;
+	struct rnd_state prng;
+	IGT_TIMEOUT(end);
+	u32 *vaddr;
+	int err;
+
+	file = mock_file(i915);
+	if (IS_ERR(file))
+		return PTR_ERR(file);
+
+	ctx = live_context(i915, file);
+	if (IS_ERR(ctx)) {
+		err = PTR_ERR(ctx);
+		goto err_file;
+	}
+
+	obj = i915_gem_object_create_internal(i915, SZ_2M);
+	if (IS_ERR(obj)) {
+		err = PTR_ERR(obj);
+		goto err_file;
+	}
+
+	vaddr = i915_gem_object_pin_map(obj, I915_MAP_WB);
+	if (IS_ERR(vaddr)) {
+		err = PTR_ERR(vaddr);
+		goto err_put;
+	}
+
+	prandom_seed_state(&prng, i915_selftest.random_seed);
+
+	mutex_lock(&i915->drm.struct_mutex);
+
+	do {
+		u32 val = prandom_u32_state(&prng);
+		u32 i;
+
+		err = i915_gem_object_fill_blt(ctx, obj, val);
+		if (err)
+			break;
+
+		err = i915_gem_object_set_to_cpu_domain(obj, false);
+		if (err)
+			break;
+
+		for (i = 0; i < obj->base.size / sizeof(u32); ++i) {
+			if (vaddr[i] != val) {
+				pr_err("vaddr[%d]=%u, expected=%u\n", i,
+				       val, vaddr[i]);
+				err = -EINVAL;
+				break;
+			}
+		}
+	} while (!time_after(jiffies, end));
+
+	mutex_unlock(&i915->drm.struct_mutex);
+
+	i915_gem_object_unpin_map(obj);
+err_put:
+	i915_gem_object_put(obj);
+err_file:
+	mock_file_free(i915, file);
+	return err;
+}
+
 int i915_gem_object_mock_selftests(void)
 {
 	static const struct i915_subtest tests[] = {
@@ -653,6 +722,7 @@  int i915_gem_object_live_selftests(struct drm_i915_private *i915)
 		SUBTEST(igt_gem_huge),
 		SUBTEST(igt_partial_tiling),
 		SUBTEST(igt_mmap_offset_exhaustion),
+		SUBTEST(igt_fill_blt),
 	};
 
 	return i915_subtests(tests, i915);