diff mbox series

[v2,04/37] drm/i915/region: support continuous allocations

Message ID 20190627205633.1143-5-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 June 27, 2019, 8:56 p.m. UTC 
Some objects may need to be allocated as a continuous block, thinking
ahead the various kernel io_mapping interfaces seem to expect it.

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>
---
 .../gpu/drm/i915/gem/i915_gem_object_types.h  |   4 +
 drivers/gpu/drm/i915/intel_memory_region.c    |   7 +-
 .../drm/i915/selftests/intel_memory_region.c  | 152 +++++++++++++++++-
 drivers/gpu/drm/i915/selftests/mock_region.c  |   3 +
 4 files changed, 160 insertions(+), 6 deletions(-)

Comments

Chris Wilson June 27, 2019, 11:01 p.m. UTC | #1 
Quoting Matthew Auld (2019-06-27 21:56:00)
> Some objects may need to be allocated as a continuous block, thinking
> ahead the various kernel io_mapping interfaces seem to expect it.
> 
> 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>
> ---
>  .../gpu/drm/i915/gem/i915_gem_object_types.h  |   4 +
>  drivers/gpu/drm/i915/intel_memory_region.c    |   7 +-
>  .../drm/i915/selftests/intel_memory_region.c  | 152 +++++++++++++++++-
>  drivers/gpu/drm/i915/selftests/mock_region.c  |   3 +
>  4 files changed, 160 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
> index 87000fc24ab3..1c4b99e507c3 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
> @@ -133,6 +133,10 @@ struct drm_i915_gem_object {
>         struct list_head batch_pool_link;
>         I915_SELFTEST_DECLARE(struct list_head st_link);
>  
> +       unsigned long flags;
> +#define I915_BO_ALLOC_CONTIGUOUS (1<<0)
BIT(0)
> +#define I915_BO_ALLOC_FLAGS (I915_BO_ALLOC_CONTIGUOUS)
> +
>         /*
>          * Is the object to be mapped as read-only to the GPU
>          * Only honoured if hardware has relevant pte bit
> diff --git a/drivers/gpu/drm/i915/intel_memory_region.c b/drivers/gpu/drm/i915/intel_memory_region.c
> index 721b47e46492..9b6a32bfa20d 100644
> --- a/drivers/gpu/drm/i915/intel_memory_region.c
> +++ b/drivers/gpu/drm/i915/intel_memory_region.c
> @@ -90,6 +90,7 @@ i915_memory_region_get_pages_buddy(struct drm_i915_gem_object *obj)
>  {
>         struct intel_memory_region *mem = obj->memory_region;
>         resource_size_t size = obj->base.size;
> +       unsigned int flags = obj->flags;

Was unsigned long.

>         struct sg_table *st;
>         struct scatterlist *sg;
>         unsigned int sg_page_sizes;
> @@ -130,7 +131,7 @@ i915_memory_region_get_pages_buddy(struct drm_i915_gem_object *obj)
>                         if (!IS_ERR(block))
>                                 break;
>  
> -                       if (!order--) {
> +                       if (flags & I915_BO_ALLOC_CONTIGUOUS || !order--) {
>                                 resource_size_t target;
>                                 int err;
>  
> @@ -219,6 +220,9 @@ i915_gem_object_create_region(struct intel_memory_region *mem,
>         if (!mem)
>                 return ERR_PTR(-ENODEV);
>  
> +       if (flags & ~I915_BO_ALLOC_FLAGS)
> +               return ERR_PTR(-EINVAL);
> +
>         size = round_up(size, mem->min_page_size);
>  
>         GEM_BUG_ON(!size);
> @@ -236,6 +240,7 @@ i915_gem_object_create_region(struct intel_memory_region *mem,
>  
>         INIT_LIST_HEAD(&obj->blocks);
>         obj->memory_region = mem;
> +       obj->flags = flags;
>  
>         mutex_lock(&mem->obj_lock);
>         list_add(&obj->region_link, &mem->objects);
> diff --git a/drivers/gpu/drm/i915/selftests/intel_memory_region.c b/drivers/gpu/drm/i915/selftests/intel_memory_region.c
> index ece499869747..c9de8b5039e4 100644
> --- a/drivers/gpu/drm/i915/selftests/intel_memory_region.c
> +++ b/drivers/gpu/drm/i915/selftests/intel_memory_region.c
> @@ -78,17 +78,17 @@ static int igt_mock_fill(void *arg)
>  
>  static void igt_mark_evictable(struct drm_i915_gem_object *obj)
>  {
> -       i915_gem_object_unpin_pages(obj);
> +       if (i915_gem_object_has_pinned_pages(obj))
> +               i915_gem_object_unpin_pages(obj);
>         obj->mm.madv = I915_MADV_DONTNEED;
>         list_move(&obj->region_link, &obj->memory_region->purgeable);
>  }
>  
> -static int igt_mock_evict(void *arg)
> +static int igt_frag_region(struct intel_memory_region *mem,
> +                          struct list_head *objects)
>  {
> -       struct intel_memory_region *mem = arg;
>         struct drm_i915_gem_object *obj;
>         unsigned long n_objects;
> -       LIST_HEAD(objects);
>         resource_size_t target;
>         resource_size_t total;
>         int err = 0;
> @@ -104,7 +104,7 @@ static int igt_mock_evict(void *arg)
>                         goto err_close_objects;
>                 }
>  
> -               list_add(&obj->st_link, &objects);
> +               list_add(&obj->st_link, objects);
>  
>                 err = i915_gem_object_pin_pages(obj);
>                 if (err)
> @@ -118,6 +118,39 @@ static int igt_mock_evict(void *arg)
>                         igt_mark_evictable(obj);
>         }
>  
> +       return 0;
> +
> +err_close_objects:
> +       close_objects(objects);
> +       return err;
> +}
> +
> +static void igt_defrag_region(struct list_head *objects)
> +{
> +       struct drm_i915_gem_object *obj;
> +
> +       list_for_each_entry(obj, objects, st_link) {
> +               if (obj->mm.madv == I915_MADV_WILLNEED)
> +                       igt_mark_evictable(obj);
> +       }
> +}
> +
> +static int igt_mock_evict(void *arg)
> +{
> +       struct intel_memory_region *mem = arg;
> +       struct drm_i915_gem_object *obj;
> +       LIST_HEAD(objects);
> +       resource_size_t target;
> +       resource_size_t total;
> +       int err;
> +
> +       err = igt_frag_region(mem, &objects);
> +       if (err)
> +               return err;
> +
> +       total = resource_size(&mem->region);
> +       target = mem->mm.min_size;
> +
>         while (target <= total / 2) {
>                 obj = i915_gem_object_create_region(mem, target, 0);
>                 if (IS_ERR(obj)) {
> @@ -148,11 +181,120 @@ static int igt_mock_evict(void *arg)
>         return err;
>  }
>  
> +static int igt_mock_continuous(void *arg)
> +{
> +       struct intel_memory_region *mem = arg;
> +       struct drm_i915_gem_object *obj;
> +       LIST_HEAD(objects);
> +       resource_size_t target;
> +       resource_size_t total;
> +       int err;
> +
> +       err = igt_frag_region(mem, &objects);
> +       if (err)
> +               return err;
> +
> +       total = resource_size(&mem->region);
> +       target = total / 2;
> +
> +       /*
> +        * Sanity check that we can allocate all of the available fragmented
> +        * space.
> +        */
> +       obj = i915_gem_object_create_region(mem, target, 0);
> +       if (IS_ERR(obj)) {
> +               err = PTR_ERR(obj);
> +               goto err_close_objects;
> +       }
> +
> +       list_add(&obj->st_link, &objects);
> +
> +       err = i915_gem_object_pin_pages(obj);
> +       if (err) {
> +               pr_err("failed to allocate available space\n");
> +               goto err_close_objects;
> +       }
> +
> +       igt_mark_evictable(obj);
> +
> +       /* Try the smallest possible size -- should succeed */
> +       obj = i915_gem_object_create_region(mem, mem->mm.min_size,
> +                                           I915_BO_ALLOC_CONTIGUOUS);
> +       if (IS_ERR(obj)) {
> +               err = PTR_ERR(obj);
> +               goto err_close_objects;
> +       }
> +
> +       list_add(&obj->st_link, &objects);
> +
> +       err = i915_gem_object_pin_pages(obj);
> +       if (err) {
> +               pr_err("failed to allocate smallest possible size\n");
> +               goto err_close_objects;
> +       }
> +
> +       igt_mark_evictable(obj);
> +
> +       if (obj->mm.pages->nents != 1) {
> +               pr_err("[1]object spans multiple sg entries\n");
> +               err = -EINVAL;
> +               goto err_close_objects;
> +       }
> +
> +       /*
> +        * Even though there is enough free space for the allocation, we
> +        * shouldn't be able to allocate it, given that it is fragmented, and
> +        * non-continuous.
> +        */
> +       obj = i915_gem_object_create_region(mem, target, I915_BO_ALLOC_CONTIGUOUS);
> +       if (IS_ERR(obj)) {
> +               err = PTR_ERR(obj);
> +               goto err_close_objects;
> +       }
> +
> +       list_add(&obj->st_link, &objects);
> +
> +       err = i915_gem_object_pin_pages(obj);
> +       if (!err) {
> +               pr_err("expected allocation to fail\n");
> +               err = -EINVAL;
> +               goto err_close_objects;
> +       }
> +
> +       igt_defrag_region(&objects);
> +
> +       /* Should now succeed */
> +       obj = i915_gem_object_create_region(mem, target, I915_BO_ALLOC_CONTIGUOUS);
> +       if (IS_ERR(obj)) {
> +               err = PTR_ERR(obj);
> +               goto err_close_objects;
> +       }
> +
> +       list_add(&obj->st_link, &objects);
> +
> +       err = i915_gem_object_pin_pages(obj);
> +       if (err) {
> +               pr_err("failed to allocate from defraged area\n");
> +               goto err_close_objects;
> +       }
> +
> +       if (obj->mm.pages->nents != 1) {
> +               pr_err("object spans multiple sg entries\n");
> +               err = -EINVAL;
> +       }
> +
> +err_close_objects:
> +       close_objects(&objects);
> +
> +       return err;
> +}
> +
>  int intel_memory_region_mock_selftests(void)
>  {
>         static const struct i915_subtest tests[] = {
>                 SUBTEST(igt_mock_fill),
>                 SUBTEST(igt_mock_evict),
> +               SUBTEST(igt_mock_continuous),
>         };
>         struct intel_memory_region *mem;
>         struct drm_i915_private *i915;
> diff --git a/drivers/gpu/drm/i915/selftests/mock_region.c b/drivers/gpu/drm/i915/selftests/mock_region.c
> index 80eafdc54927..9eeda8f45f38 100644
> --- a/drivers/gpu/drm/i915/selftests/mock_region.c
> +++ b/drivers/gpu/drm/i915/selftests/mock_region.c
> @@ -20,6 +20,9 @@ mock_object_create(struct intel_memory_region *mem,
>         struct drm_i915_gem_object *obj;
>         unsigned int cache_level;
>  
> +       if (flags & I915_BO_ALLOC_CONTIGUOUS)
> +               size = roundup_pow_of_two(size);
> +
>         if (size > BIT(mem->mm.max_order) * mem->mm.min_size)
>                 return ERR_PTR(-E2BIG);
>  
> -- 
> 2.20.1
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
diff mbox series

Patch

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
index 87000fc24ab3..1c4b99e507c3 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
@@ -133,6 +133,10 @@  struct drm_i915_gem_object {
 	struct list_head batch_pool_link;
 	I915_SELFTEST_DECLARE(struct list_head st_link);
 
+	unsigned long flags;
+#define I915_BO_ALLOC_CONTIGUOUS (1<<0)
+#define I915_BO_ALLOC_FLAGS (I915_BO_ALLOC_CONTIGUOUS)
+
 	/*
 	 * Is the object to be mapped as read-only to the GPU
 	 * Only honoured if hardware has relevant pte bit
diff --git a/drivers/gpu/drm/i915/intel_memory_region.c b/drivers/gpu/drm/i915/intel_memory_region.c
index 721b47e46492..9b6a32bfa20d 100644
--- a/drivers/gpu/drm/i915/intel_memory_region.c
+++ b/drivers/gpu/drm/i915/intel_memory_region.c
@@ -90,6 +90,7 @@  i915_memory_region_get_pages_buddy(struct drm_i915_gem_object *obj)
 {
 	struct intel_memory_region *mem = obj->memory_region;
 	resource_size_t size = obj->base.size;
+	unsigned int flags = obj->flags;
 	struct sg_table *st;
 	struct scatterlist *sg;
 	unsigned int sg_page_sizes;
@@ -130,7 +131,7 @@  i915_memory_region_get_pages_buddy(struct drm_i915_gem_object *obj)
 			if (!IS_ERR(block))
 				break;
 
-			if (!order--) {
+			if (flags & I915_BO_ALLOC_CONTIGUOUS || !order--) {
 				resource_size_t target;
 				int err;
 
@@ -219,6 +220,9 @@  i915_gem_object_create_region(struct intel_memory_region *mem,
 	if (!mem)
 		return ERR_PTR(-ENODEV);
 
+	if (flags & ~I915_BO_ALLOC_FLAGS)
+		return ERR_PTR(-EINVAL);
+
 	size = round_up(size, mem->min_page_size);
 
 	GEM_BUG_ON(!size);
@@ -236,6 +240,7 @@  i915_gem_object_create_region(struct intel_memory_region *mem,
 
 	INIT_LIST_HEAD(&obj->blocks);
 	obj->memory_region = mem;
+	obj->flags = flags;
 
 	mutex_lock(&mem->obj_lock);
 	list_add(&obj->region_link, &mem->objects);
diff --git a/drivers/gpu/drm/i915/selftests/intel_memory_region.c b/drivers/gpu/drm/i915/selftests/intel_memory_region.c
index ece499869747..c9de8b5039e4 100644
--- a/drivers/gpu/drm/i915/selftests/intel_memory_region.c
+++ b/drivers/gpu/drm/i915/selftests/intel_memory_region.c
@@ -78,17 +78,17 @@  static int igt_mock_fill(void *arg)
 
 static void igt_mark_evictable(struct drm_i915_gem_object *obj)
 {
-	i915_gem_object_unpin_pages(obj);
+	if (i915_gem_object_has_pinned_pages(obj))
+		i915_gem_object_unpin_pages(obj);
 	obj->mm.madv = I915_MADV_DONTNEED;
 	list_move(&obj->region_link, &obj->memory_region->purgeable);
 }
 
-static int igt_mock_evict(void *arg)
+static int igt_frag_region(struct intel_memory_region *mem,
+			   struct list_head *objects)
 {
-	struct intel_memory_region *mem = arg;
 	struct drm_i915_gem_object *obj;
 	unsigned long n_objects;
-	LIST_HEAD(objects);
 	resource_size_t target;
 	resource_size_t total;
 	int err = 0;
@@ -104,7 +104,7 @@  static int igt_mock_evict(void *arg)
 			goto err_close_objects;
 		}
 
-		list_add(&obj->st_link, &objects);
+		list_add(&obj->st_link, objects);
 
 		err = i915_gem_object_pin_pages(obj);
 		if (err)
@@ -118,6 +118,39 @@  static int igt_mock_evict(void *arg)
 			igt_mark_evictable(obj);
 	}
 
+	return 0;
+
+err_close_objects:
+	close_objects(objects);
+	return err;
+}
+
+static void igt_defrag_region(struct list_head *objects)
+{
+	struct drm_i915_gem_object *obj;
+
+	list_for_each_entry(obj, objects, st_link) {
+		if (obj->mm.madv == I915_MADV_WILLNEED)
+			igt_mark_evictable(obj);
+	}
+}
+
+static int igt_mock_evict(void *arg)
+{
+	struct intel_memory_region *mem = arg;
+	struct drm_i915_gem_object *obj;
+	LIST_HEAD(objects);
+	resource_size_t target;
+	resource_size_t total;
+	int err;
+
+	err = igt_frag_region(mem, &objects);
+	if (err)
+		return err;
+
+	total = resource_size(&mem->region);
+	target = mem->mm.min_size;
+
 	while (target <= total / 2) {
 		obj = i915_gem_object_create_region(mem, target, 0);
 		if (IS_ERR(obj)) {
@@ -148,11 +181,120 @@  static int igt_mock_evict(void *arg)
 	return err;
 }
 
+static int igt_mock_continuous(void *arg)
+{
+	struct intel_memory_region *mem = arg;
+	struct drm_i915_gem_object *obj;
+	LIST_HEAD(objects);
+	resource_size_t target;
+	resource_size_t total;
+	int err;
+
+	err = igt_frag_region(mem, &objects);
+	if (err)
+		return err;
+
+	total = resource_size(&mem->region);
+	target = total / 2;
+
+	/*
+	 * Sanity check that we can allocate all of the available fragmented
+	 * space.
+	 */
+	obj = i915_gem_object_create_region(mem, target, 0);
+	if (IS_ERR(obj)) {
+		err = PTR_ERR(obj);
+		goto err_close_objects;
+	}
+
+	list_add(&obj->st_link, &objects);
+
+	err = i915_gem_object_pin_pages(obj);
+	if (err) {
+		pr_err("failed to allocate available space\n");
+		goto err_close_objects;
+	}
+
+	igt_mark_evictable(obj);
+
+	/* Try the smallest possible size -- should succeed */
+	obj = i915_gem_object_create_region(mem, mem->mm.min_size,
+					    I915_BO_ALLOC_CONTIGUOUS);
+	if (IS_ERR(obj)) {
+		err = PTR_ERR(obj);
+		goto err_close_objects;
+	}
+
+	list_add(&obj->st_link, &objects);
+
+	err = i915_gem_object_pin_pages(obj);
+	if (err) {
+		pr_err("failed to allocate smallest possible size\n");
+		goto err_close_objects;
+	}
+
+	igt_mark_evictable(obj);
+
+	if (obj->mm.pages->nents != 1) {
+		pr_err("[1]object spans multiple sg entries\n");
+		err = -EINVAL;
+		goto err_close_objects;
+	}
+
+	/*
+	 * Even though there is enough free space for the allocation, we
+	 * shouldn't be able to allocate it, given that it is fragmented, and
+	 * non-continuous.
+	 */
+	obj = i915_gem_object_create_region(mem, target, I915_BO_ALLOC_CONTIGUOUS);
+	if (IS_ERR(obj)) {
+		err = PTR_ERR(obj);
+		goto err_close_objects;
+	}
+
+	list_add(&obj->st_link, &objects);
+
+	err = i915_gem_object_pin_pages(obj);
+	if (!err) {
+		pr_err("expected allocation to fail\n");
+		err = -EINVAL;
+		goto err_close_objects;
+	}
+
+	igt_defrag_region(&objects);
+
+	/* Should now succeed */
+	obj = i915_gem_object_create_region(mem, target, I915_BO_ALLOC_CONTIGUOUS);
+	if (IS_ERR(obj)) {
+		err = PTR_ERR(obj);
+		goto err_close_objects;
+	}
+
+	list_add(&obj->st_link, &objects);
+
+	err = i915_gem_object_pin_pages(obj);
+	if (err) {
+		pr_err("failed to allocate from defraged area\n");
+		goto err_close_objects;
+	}
+
+	if (obj->mm.pages->nents != 1) {
+		pr_err("object spans multiple sg entries\n");
+		err = -EINVAL;
+	}
+
+err_close_objects:
+	close_objects(&objects);
+
+	return err;
+}
+
 int intel_memory_region_mock_selftests(void)
 {
 	static const struct i915_subtest tests[] = {
 		SUBTEST(igt_mock_fill),
 		SUBTEST(igt_mock_evict),
+		SUBTEST(igt_mock_continuous),
 	};
 	struct intel_memory_region *mem;
 	struct drm_i915_private *i915;
diff --git a/drivers/gpu/drm/i915/selftests/mock_region.c b/drivers/gpu/drm/i915/selftests/mock_region.c
index 80eafdc54927..9eeda8f45f38 100644
--- a/drivers/gpu/drm/i915/selftests/mock_region.c
+++ b/drivers/gpu/drm/i915/selftests/mock_region.c
@@ -20,6 +20,9 @@  mock_object_create(struct intel_memory_region *mem,
 	struct drm_i915_gem_object *obj;
 	unsigned int cache_level;
 
+	if (flags & I915_BO_ALLOC_CONTIGUOUS)
+		size = roundup_pow_of_two(size);
+
 	if (size > BIT(mem->mm.max_order) * mem->mm.min_size)
 		return ERR_PTR(-E2BIG);