| Message ID | 20190809222643.23142-5-matthew.auld@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Introduce memory region concept (including device local memory) | expand |
Quoting Matthew Auld (2019-08-09 23:26:10) > Some objects may need to be allocated as a continuous block, thinking > ahead the various kernel io_mapping interfaces seem to expect it. But we could always use scattergather over top... > @@ -98,10 +101,12 @@ i915_gem_object_get_pages_buddy(struct drm_i915_gem_object *obj) > } > > void i915_gem_object_init_memory_region(struct drm_i915_gem_object *obj, > - struct intel_memory_region *mem) > + struct intel_memory_region *mem, > + unsigned long flags) > { > INIT_LIST_HEAD(&obj->mm.blocks); > obj->mm.region= mem; > + obj->flags = flags; > > mutex_lock(&mem->obj_lock); > list_add(&obj->mm.region_link, &mem->objects); > @@ -125,6 +130,9 @@ 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); This is a programmer error not a user. > + > size = round_up(size, mem->min_page_size);
On Fri, Aug 09, 2019 at 11:26:10PM +0100, Matthew Auld wrote: > Some objects may need to be allocated as a continuous block, thinking > ahead the various kernel io_mapping interfaces seem to expect it. Not really, we can vmalloc for iomappings too. -Daniel > > 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/gem/i915_gem_region.c | 10 +- > drivers/gpu/drm/i915/gem/i915_gem_region.h | 3 +- > .../drm/i915/selftests/intel_memory_region.c | 152 +++++++++++++++++- > drivers/gpu/drm/i915/selftests/mock_region.c | 5 +- > 5 files changed, 166 insertions(+), 8 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 5e2fa37e9bc0..eb92243d473b 100644 > --- a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h > +++ b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h > @@ -116,6 +116,10 @@ struct drm_i915_gem_object { > > I915_SELFTEST_DECLARE(struct list_head st_link); > > + unsigned long flags; > +#define I915_BO_ALLOC_CONTIGUOUS 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/gem/i915_gem_region.c b/drivers/gpu/drm/i915/gem/i915_gem_region.c > index be126e70c90f..d9cd722b5dbf 100644 > --- a/drivers/gpu/drm/i915/gem/i915_gem_region.c > +++ b/drivers/gpu/drm/i915/gem/i915_gem_region.c > @@ -42,6 +42,9 @@ i915_gem_object_get_pages_buddy(struct drm_i915_gem_object *obj) > return -ENOMEM; > } > > + if (obj->flags & I915_BO_ALLOC_CONTIGUOUS) > + flags = I915_ALLOC_CONTIGUOUS; > + > ret = __intel_memory_region_get_pages_buddy(mem, size, flags, blocks); > if (ret) > goto err_free_sg; > @@ -98,10 +101,12 @@ i915_gem_object_get_pages_buddy(struct drm_i915_gem_object *obj) > } > > void i915_gem_object_init_memory_region(struct drm_i915_gem_object *obj, > - struct intel_memory_region *mem) > + struct intel_memory_region *mem, > + unsigned long flags) > { > INIT_LIST_HEAD(&obj->mm.blocks); > obj->mm.region= mem; > + obj->flags = flags; > > mutex_lock(&mem->obj_lock); > list_add(&obj->mm.region_link, &mem->objects); > @@ -125,6 +130,9 @@ 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); > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_region.h b/drivers/gpu/drm/i915/gem/i915_gem_region.h > index ebddc86d78f7..f2ff6f8bff74 100644 > --- a/drivers/gpu/drm/i915/gem/i915_gem_region.h > +++ b/drivers/gpu/drm/i915/gem/i915_gem_region.h > @@ -17,7 +17,8 @@ void i915_gem_object_put_pages_buddy(struct drm_i915_gem_object *obj, > struct sg_table *pages); > > void i915_gem_object_init_memory_region(struct drm_i915_gem_object *obj, > - struct intel_memory_region *mem); > + struct intel_memory_region *mem, > + unsigned long flags); > void i915_gem_object_release_memory_region(struct drm_i915_gem_object *obj); > > struct drm_i915_gem_object * > diff --git a/drivers/gpu/drm/i915/selftests/intel_memory_region.c b/drivers/gpu/drm/i915/selftests/intel_memory_region.c > index 2f13e4c1d999..70b467d4e811 100644 > --- a/drivers/gpu/drm/i915/selftests/intel_memory_region.c > +++ b/drivers/gpu/drm/i915/selftests/intel_memory_region.c > @@ -81,17 +81,17 @@ 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->mm.region_link, &obj->mm.region->purgeable); > } > > -static int igt_mock_shrink(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; > @@ -109,7 +109,7 @@ static int igt_mock_shrink(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) > @@ -123,6 +123,39 @@ static int igt_mock_shrink(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_shrink(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.chunk_size; > + > while (target <= total / 2) { > obj = i915_gem_object_create_region(mem, target, 0); > if (IS_ERR(obj)) { > @@ -154,11 +187,120 @@ static int igt_mock_shrink(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.chunk_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_shrink), > + 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 cc97250dca62..d73f37712c44 100644 > --- a/drivers/gpu/drm/i915/selftests/mock_region.c > +++ b/drivers/gpu/drm/i915/selftests/mock_region.c > @@ -23,6 +23,9 @@ 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.chunk_size) > return ERR_PTR(-E2BIG); > > @@ -38,7 +41,7 @@ mock_object_create(struct intel_memory_region *mem, > cache_level = HAS_LLC(i915) ? I915_CACHE_LLC : I915_CACHE_NONE; > i915_gem_object_set_cache_coherency(obj, cache_level); > > - i915_gem_object_init_memory_region(obj, mem); > + i915_gem_object_init_memory_region(obj, mem, flags); > > return obj; > } > -- > 2.20.1 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/dri-devel
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 5e2fa37e9bc0..eb92243d473b 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h +++ b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h @@ -116,6 +116,10 @@ struct drm_i915_gem_object { I915_SELFTEST_DECLARE(struct list_head st_link); + unsigned long flags; +#define I915_BO_ALLOC_CONTIGUOUS 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/gem/i915_gem_region.c b/drivers/gpu/drm/i915/gem/i915_gem_region.c index be126e70c90f..d9cd722b5dbf 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_region.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_region.c @@ -42,6 +42,9 @@ i915_gem_object_get_pages_buddy(struct drm_i915_gem_object *obj) return -ENOMEM; } + if (obj->flags & I915_BO_ALLOC_CONTIGUOUS) + flags = I915_ALLOC_CONTIGUOUS; + ret = __intel_memory_region_get_pages_buddy(mem, size, flags, blocks); if (ret) goto err_free_sg; @@ -98,10 +101,12 @@ i915_gem_object_get_pages_buddy(struct drm_i915_gem_object *obj) } void i915_gem_object_init_memory_region(struct drm_i915_gem_object *obj, - struct intel_memory_region *mem) + struct intel_memory_region *mem, + unsigned long flags) { INIT_LIST_HEAD(&obj->mm.blocks); obj->mm.region= mem; + obj->flags = flags; mutex_lock(&mem->obj_lock); list_add(&obj->mm.region_link, &mem->objects); @@ -125,6 +130,9 @@ 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); diff --git a/drivers/gpu/drm/i915/gem/i915_gem_region.h b/drivers/gpu/drm/i915/gem/i915_gem_region.h index ebddc86d78f7..f2ff6f8bff74 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_region.h +++ b/drivers/gpu/drm/i915/gem/i915_gem_region.h @@ -17,7 +17,8 @@ void i915_gem_object_put_pages_buddy(struct drm_i915_gem_object *obj, struct sg_table *pages); void i915_gem_object_init_memory_region(struct drm_i915_gem_object *obj, - struct intel_memory_region *mem); + struct intel_memory_region *mem, + unsigned long flags); void i915_gem_object_release_memory_region(struct drm_i915_gem_object *obj); struct drm_i915_gem_object * diff --git a/drivers/gpu/drm/i915/selftests/intel_memory_region.c b/drivers/gpu/drm/i915/selftests/intel_memory_region.c index 2f13e4c1d999..70b467d4e811 100644 --- a/drivers/gpu/drm/i915/selftests/intel_memory_region.c +++ b/drivers/gpu/drm/i915/selftests/intel_memory_region.c @@ -81,17 +81,17 @@ 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->mm.region_link, &obj->mm.region->purgeable); } -static int igt_mock_shrink(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; @@ -109,7 +109,7 @@ static int igt_mock_shrink(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) @@ -123,6 +123,39 @@ static int igt_mock_shrink(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_shrink(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.chunk_size; + while (target <= total / 2) { obj = i915_gem_object_create_region(mem, target, 0); if (IS_ERR(obj)) { @@ -154,11 +187,120 @@ static int igt_mock_shrink(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.chunk_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_shrink), + 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 cc97250dca62..d73f37712c44 100644 --- a/drivers/gpu/drm/i915/selftests/mock_region.c +++ b/drivers/gpu/drm/i915/selftests/mock_region.c @@ -23,6 +23,9 @@ 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.chunk_size) return ERR_PTR(-E2BIG); @@ -38,7 +41,7 @@ mock_object_create(struct intel_memory_region *mem, cache_level = HAS_LLC(i915) ? I915_CACHE_LLC : I915_CACHE_NONE; i915_gem_object_set_cache_coherency(obj, cache_level); - i915_gem_object_init_memory_region(obj, mem); + i915_gem_object_init_memory_region(obj, mem, flags); return obj; }
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/gem/i915_gem_region.c | 10 +- drivers/gpu/drm/i915/gem/i915_gem_region.h | 3 +- .../drm/i915/selftests/intel_memory_region.c | 152 +++++++++++++++++- drivers/gpu/drm/i915/selftests/mock_region.c | 5 +- 5 files changed, 166 insertions(+), 8 deletions(-)