Just checking the current region is not enough, if we later migrate the object somewhere else. For example if the placements are {SMEM, LMEM}, then we might get this wrong. Another idea might be to make the page_alignment part of the ttm_place, instead of the BO.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com --- drivers/gpu/drm/i915/gem/i915_gem_ttm.c | 21 ++++++++++++++++++++- 1 file changed, 20 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c index c5deb8b7227c..5d894bba6430 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c @@ -753,6 +753,25 @@ void i915_ttm_bo_destroy(struct ttm_buffer_object *bo) call_rcu(&obj->rcu, __i915_gem_free_object_rcu); }
+static u64 i915_gem_object_page_size(struct drm_i915_gem_object *obj) +{ + u64 page_size; + int i; + + if (!obj->mm.n_placements) + return obj->mm.region->min_page_size; + + page_size = 0; + for (i = 0; i < obj->mm.n_placements; i++) { + struct intel_memory_region *mr = obj->mm.placements[i]; + + page_size = max_t(u64, mr->min_page_size, page_size); + } + + GEM_BUG_ON(!page_size); + return page_size; +} + /** * __i915_gem_ttm_object_init - Initialize a ttm-backed i915 gem object * @mem: The initial memory region for the object. @@ -793,7 +812,7 @@ int __i915_gem_ttm_object_init(struct intel_memory_region *mem, obj->base.vma_node.driver_private = i915_gem_to_ttm(obj); ret = ttm_bo_init(&i915->bdev, i915_gem_to_ttm(obj), size, bo_type, &i915_sys_placement, - mem->min_page_size >> PAGE_SHIFT, + i915_gem_object_page_size(obj) >> PAGE_SHIFT, true, NULL, NULL, i915_ttm_bo_destroy); if (!ret) obj->ttm.created = true;
For some specialised objects we might need something larger than the regions min_page_size due to some hw restriction, and slightly more hairy is needing something smaller with the guarantee that such objects will never be inserted into any GTT, which is the case for the paging structures.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com --- drivers/gpu/drm/i915/gem/i915_gem_create.c | 2 +- drivers/gpu/drm/i915/gem/i915_gem_lmem.c | 33 +++++++++- drivers/gpu/drm/i915/gem/i915_gem_lmem.h | 5 ++ drivers/gpu/drm/i915/gem/i915_gem_region.c | 10 ++- drivers/gpu/drm/i915/gem/i915_gem_region.h | 1 + drivers/gpu/drm/i915/gem/i915_gem_shmem.c | 3 +- drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 3 +- drivers/gpu/drm/i915/gem/i915_gem_ttm.c | 14 +++-- drivers/gpu/drm/i915/gem/i915_gem_ttm.h | 1 + .../gpu/drm/i915/gem/selftests/huge_pages.c | 3 +- .../drm/i915/gem/selftests/i915_gem_mman.c | 8 +-- drivers/gpu/drm/i915/intel_memory_region.h | 1 + .../drm/i915/selftests/intel_memory_region.c | 63 ++++++++++++++++++- drivers/gpu/drm/i915/selftests/mock_region.c | 1 + 14 files changed, 130 insertions(+), 18 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c b/drivers/gpu/drm/i915/gem/i915_gem_create.c index 93bf63bbaff1..51f92e4b1a69 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_create.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c @@ -90,7 +90,7 @@ i915_gem_setup(struct drm_i915_gem_object *obj, u64 size) */ flags = I915_BO_ALLOC_USER;
- ret = mr->ops->init_object(mr, obj, size, flags); + ret = mr->ops->init_object(mr, obj, size, 0, flags); if (ret) return ret;
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_lmem.c b/drivers/gpu/drm/i915/gem/i915_gem_lmem.c index d539dffa1554..174a9e34ad93 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_lmem.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_lmem.c @@ -71,11 +71,42 @@ bool i915_gem_object_is_lmem(struct drm_i915_gem_object *obj) mr->type == INTEL_MEMORY_STOLEN_LOCAL); }
+/** + * __i915_gem_object_create_lmem_with_ps - Create lmem object and force the + * minimum page size for the backing pages. + * @i915: The i915 instance. + * @size: The size in bytes for the object. Note that we need to round the size + * up depending on the @page_size. The final object size can be fished out from + * the drm GEM object. + * @page_size: The requested minimum page size in bytes for this object. The is + * useful if we need something bigger than the regions min_page_size due to some + * hw restriction, or in some very specialised cases where it needs to be + * smaller, where the internal fragmentation cost is too great when rounding up + * the object size. + * @flags: the optional BO allocation flags + * + * Note that this interface assumes you know what you are doing when forcing the + * page_size. If this is smaller than the regions min_page_size then it can + * never be inserted into any GTT, otherwise it might lead to undefined + * behaviour. + * + * Return: The object pointer, which might be an ERR_PTR in the case of failure. + */ +struct drm_i915_gem_object * +__i915_gem_object_create_lmem_with_ps(struct drm_i915_private *i915, + resource_size_t size, + resource_size_t page_size, + unsigned int flags) +{ + return i915_gem_object_create_region(i915->mm.regions[INTEL_REGION_LMEM], + size, page_size, flags); +} + struct drm_i915_gem_object * i915_gem_object_create_lmem(struct drm_i915_private *i915, resource_size_t size, unsigned int flags) { return i915_gem_object_create_region(i915->mm.regions[INTEL_REGION_LMEM], - size, flags); + size, 0, flags); } diff --git a/drivers/gpu/drm/i915/gem/i915_gem_lmem.h b/drivers/gpu/drm/i915/gem/i915_gem_lmem.h index ea76fd11ccb0..e98608cebbbc 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_lmem.h +++ b/drivers/gpu/drm/i915/gem/i915_gem_lmem.h @@ -21,6 +21,11 @@ i915_gem_object_lmem_io_map(struct drm_i915_gem_object *obj,
bool i915_gem_object_is_lmem(struct drm_i915_gem_object *obj);
+struct drm_i915_gem_object * +__i915_gem_object_create_lmem_with_ps(struct drm_i915_private *i915, + resource_size_t size, + resource_size_t page_size, + unsigned int flags); struct drm_i915_gem_object * i915_gem_object_create_lmem(struct drm_i915_private *i915, resource_size_t size, diff --git a/drivers/gpu/drm/i915/gem/i915_gem_region.c b/drivers/gpu/drm/i915/gem/i915_gem_region.c index d1f1840540dd..b18182b46ee9 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_region.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_region.c @@ -36,6 +36,7 @@ void i915_gem_object_release_memory_region(struct drm_i915_gem_object *obj) struct drm_i915_gem_object * i915_gem_object_create_region(struct intel_memory_region *mem, resource_size_t size, + resource_size_t page_size, unsigned int flags) { struct drm_i915_gem_object *obj; @@ -48,11 +49,16 @@ i915_gem_object_create_region(struct intel_memory_region *mem, */
GEM_BUG_ON(flags & ~I915_BO_ALLOC_FLAGS); + GEM_BUG_ON(!is_power_of_2_u64(page_size)); + GEM_BUG_ON(page_size < PAGE_SIZE);
if (!mem) return ERR_PTR(-ENODEV);
- size = round_up(size, mem->min_page_size); + if (!page_size) + page_size = mem->min_page_size; + + size = round_up(size, page_size);
GEM_BUG_ON(!size); GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_MIN_ALIGNMENT)); @@ -64,7 +70,7 @@ i915_gem_object_create_region(struct intel_memory_region *mem, if (!obj) return ERR_PTR(-ENOMEM);
- err = mem->ops->init_object(mem, obj, size, flags); + err = mem->ops->init_object(mem, obj, size, page_size, flags); if (err) goto err_object_free;
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_region.h b/drivers/gpu/drm/i915/gem/i915_gem_region.h index 84fcb3297400..1008e580a89a 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_region.h +++ b/drivers/gpu/drm/i915/gem/i915_gem_region.h @@ -19,6 +19,7 @@ void i915_gem_object_release_memory_region(struct drm_i915_gem_object *obj); struct drm_i915_gem_object * i915_gem_object_create_region(struct intel_memory_region *mem, resource_size_t size, + resource_size_t page_size, unsigned int flags);
#endif diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c index 5d16c4462fda..f5791611b9ca 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c @@ -489,6 +489,7 @@ static int __create_shmem(struct drm_i915_private *i915, static int shmem_object_init(struct intel_memory_region *mem, struct drm_i915_gem_object *obj, resource_size_t size, + resource_size_t page_size, unsigned int flags) { static struct lock_class_key lock_class; @@ -548,7 +549,7 @@ i915_gem_object_create_shmem(struct drm_i915_private *i915, resource_size_t size) { return i915_gem_object_create_region(i915->mm.regions[INTEL_REGION_SMEM], - size, 0); + size, 0, 0); }
/* Allocate a new GEM object and fill it with the supplied data */ diff --git a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c index b0c3a7dc60d1..90708de27684 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c @@ -670,6 +670,7 @@ static int __i915_gem_object_create_stolen(struct intel_memory_region *mem, static int _i915_gem_object_stolen_init(struct intel_memory_region *mem, struct drm_i915_gem_object *obj, resource_size_t size, + resource_size_t page_size, unsigned int flags) { struct drm_i915_private *i915 = mem->i915; @@ -708,7 +709,7 @@ struct drm_i915_gem_object * i915_gem_object_create_stolen(struct drm_i915_private *i915, resource_size_t size) { - return i915_gem_object_create_region(i915->mm.stolen_region, size, 0); + return i915_gem_object_create_region(i915->mm.stolen_region, size, 0, 0); }
static int init_stolen_smem(struct intel_memory_region *mem) diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c index 5d894bba6430..2024dd0c7c22 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c @@ -758,9 +758,6 @@ static u64 i915_gem_object_page_size(struct drm_i915_gem_object *obj) u64 page_size; int i;
- if (!obj->mm.n_placements) - return obj->mm.region->min_page_size; - page_size = 0; for (i = 0; i < obj->mm.n_placements; i++) { struct intel_memory_region *mr = obj->mm.placements[i]; @@ -784,6 +781,7 @@ static u64 i915_gem_object_page_size(struct drm_i915_gem_object *obj) int __i915_gem_ttm_object_init(struct intel_memory_region *mem, struct drm_i915_gem_object *obj, resource_size_t size, + resource_size_t page_size, unsigned int flags) { static struct lock_class_key lock_class; @@ -802,6 +800,14 @@ int __i915_gem_ttm_object_init(struct intel_memory_region *mem, bo_type = (obj->flags & I915_BO_ALLOC_USER) ? ttm_bo_type_device : ttm_bo_type_kernel;
+ GEM_BUG_ON(!page_size); + + if (obj->mm.n_placements) { + /* Forcing the page size is kernel internal only */ + GEM_BUG_ON(page_size != mem->min_page_size); + page_size = i915_gem_object_page_size(obj); + } + /* * If this function fails, it will call the destructor, but * our caller still owns the object. So no freeing in the @@ -812,7 +818,7 @@ int __i915_gem_ttm_object_init(struct intel_memory_region *mem, obj->base.vma_node.driver_private = i915_gem_to_ttm(obj); ret = ttm_bo_init(&i915->bdev, i915_gem_to_ttm(obj), size, bo_type, &i915_sys_placement, - i915_gem_object_page_size(obj) >> PAGE_SHIFT, + page_size >> PAGE_SHIFT, true, NULL, NULL, i915_ttm_bo_destroy); if (!ret) obj->ttm.created = true; diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm.h b/drivers/gpu/drm/i915/gem/i915_gem_ttm.h index b8d3dcbb50df..40927f67b6d9 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm.h +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm.h @@ -44,5 +44,6 @@ i915_ttm_to_gem(struct ttm_buffer_object *bo) int __i915_gem_ttm_object_init(struct intel_memory_region *mem, struct drm_i915_gem_object *obj, resource_size_t size, + resource_size_t page_size, unsigned int flags); #endif diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c index dadd485bc52f..9d17bff8b36d 100644 --- a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c +++ b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c @@ -497,7 +497,8 @@ static int igt_mock_memory_region_huge_pages(void *arg) int i;
for (i = 0; i < ARRAY_SIZE(flags); ++i) { - obj = i915_gem_object_create_region(mem, page_size, + obj = i915_gem_object_create_region(mem, + page_size, page_size, flags[i]); if (IS_ERR(obj)) { err = PTR_ERR(obj); diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c index 44b5de06ce64..7398b502738a 100644 --- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c +++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c @@ -955,7 +955,7 @@ static int igt_mmap(void *arg) struct drm_i915_gem_object *obj; int err;
- obj = i915_gem_object_create_region(mr, sizes[i], I915_BO_ALLOC_USER); + obj = i915_gem_object_create_region(mr, sizes[i], 0, I915_BO_ALLOC_USER); if (obj == ERR_PTR(-ENODEV)) continue;
@@ -1075,7 +1075,7 @@ static int igt_mmap_access(void *arg) struct drm_i915_gem_object *obj; int err;
- obj = i915_gem_object_create_region(mr, PAGE_SIZE, I915_BO_ALLOC_USER); + obj = i915_gem_object_create_region(mr, PAGE_SIZE, 0, I915_BO_ALLOC_USER); if (obj == ERR_PTR(-ENODEV)) continue;
@@ -1220,7 +1220,7 @@ static int igt_mmap_gpu(void *arg) struct drm_i915_gem_object *obj; int err;
- obj = i915_gem_object_create_region(mr, PAGE_SIZE, I915_BO_ALLOC_USER); + obj = i915_gem_object_create_region(mr, PAGE_SIZE, 0, I915_BO_ALLOC_USER); if (obj == ERR_PTR(-ENODEV)) continue;
@@ -1375,7 +1375,7 @@ static int igt_mmap_revoke(void *arg) struct drm_i915_gem_object *obj; int err;
- obj = i915_gem_object_create_region(mr, PAGE_SIZE, I915_BO_ALLOC_USER); + obj = i915_gem_object_create_region(mr, PAGE_SIZE, 0, I915_BO_ALLOC_USER); if (obj == ERR_PTR(-ENODEV)) continue;
diff --git a/drivers/gpu/drm/i915/intel_memory_region.h b/drivers/gpu/drm/i915/intel_memory_region.h index 2be8433d373a..359cd8424b9a 100644 --- a/drivers/gpu/drm/i915/intel_memory_region.h +++ b/drivers/gpu/drm/i915/intel_memory_region.h @@ -55,6 +55,7 @@ struct intel_memory_region_ops { int (*init_object)(struct intel_memory_region *mem, struct drm_i915_gem_object *obj, resource_size_t size, + resource_size_t page_size, unsigned int flags); };
diff --git a/drivers/gpu/drm/i915/selftests/intel_memory_region.c b/drivers/gpu/drm/i915/selftests/intel_memory_region.c index ecc3b9e6c22b..1aaccb9841a0 100644 --- a/drivers/gpu/drm/i915/selftests/intel_memory_region.c +++ b/drivers/gpu/drm/i915/selftests/intel_memory_region.c @@ -68,7 +68,7 @@ static int igt_mock_fill(void *arg) resource_size_t size = page_num * page_size; struct drm_i915_gem_object *obj;
- obj = i915_gem_object_create_region(mem, size, 0); + obj = i915_gem_object_create_region(mem, size, 0, 0); if (IS_ERR(obj)) { err = PTR_ERR(obj); break; @@ -110,7 +110,7 @@ igt_object_create(struct intel_memory_region *mem, struct drm_i915_gem_object *obj; int err;
- obj = i915_gem_object_create_region(mem, size, flags); + obj = i915_gem_object_create_region(mem, size, 0, flags); if (IS_ERR(obj)) return obj;
@@ -647,6 +647,62 @@ static int igt_lmem_create(void *arg) return err; }
+static int igt_lmem_create_with_ps(void *arg) +{ + struct drm_i915_private *i915 = arg; + int err = 0; + u32 ps; + + for (ps = PAGE_SIZE; ps <= SZ_1G; ps <<= 1) { + struct drm_i915_gem_object *obj; + dma_addr_t daddr; + + obj = __i915_gem_object_create_lmem_with_ps(i915, ps, ps, 0); + if (IS_ERR(obj)) { + err = PTR_ERR(obj); + if (err == -ENXIO || err == -E2BIG) { + pr_info("%s not enough lmem for ps(%u) err=%d\n", + __func__, ps, err); + err = 0; + } + + break; + } + + if (obj->base.size != ps) { + pr_err("%s size(%zu) != ps(%u)\n", + __func__, obj->base.size, ps); + err = -EINVAL; + goto out_put; + } + + i915_gem_object_lock(obj, NULL); + err = i915_gem_object_pin_pages(obj); + if (err) + goto out_put; + + daddr = i915_gem_object_get_dma_address(obj, 0); + if (!IS_ALIGNED(daddr, ps)) { + pr_err("%s daddr(%pa) not aligned with ps(%u)\n", + __func__, &daddr, ps); + err = -EINVAL; + goto out_unpin; + } + +out_unpin: + i915_gem_object_unpin_pages(obj); + __i915_gem_object_put_pages(obj); +out_put: + i915_gem_object_unlock(obj); + i915_gem_object_put(obj); + + if (err) + break; + } + + return err; +} + static int igt_lmem_create_cleared_cpu(void *arg) { struct drm_i915_private *i915 = arg; @@ -932,7 +988,7 @@ create_region_for_mapping(struct intel_memory_region *mr, u64 size, u32 type, struct drm_i915_gem_object *obj; void *addr;
- obj = i915_gem_object_create_region(mr, size, 0); + obj = i915_gem_object_create_region(mr, size, 0, 0); if (IS_ERR(obj)) { if (PTR_ERR(obj) == -ENOSPC) /* Stolen memory */ return ERR_PTR(-ENODEV); @@ -1149,6 +1205,7 @@ int intel_memory_region_live_selftests(struct drm_i915_private *i915) { static const struct i915_subtest tests[] = { SUBTEST(igt_lmem_create), + SUBTEST(igt_lmem_create_with_ps), SUBTEST(igt_lmem_create_cleared_cpu), SUBTEST(igt_lmem_write_cpu), SUBTEST(igt_lmem_write_gpu), diff --git a/drivers/gpu/drm/i915/selftests/mock_region.c b/drivers/gpu/drm/i915/selftests/mock_region.c index fa786dede608..efa86dffe3c6 100644 --- a/drivers/gpu/drm/i915/selftests/mock_region.c +++ b/drivers/gpu/drm/i915/selftests/mock_region.c @@ -63,6 +63,7 @@ static const struct drm_i915_gem_object_ops mock_region_obj_ops = { static int mock_object_init(struct intel_memory_region *mem, struct drm_i915_gem_object *obj, resource_size_t size, + resource_size_t page_size, unsigned int flags) { static struct lock_class_key lock_class;
On Wed, 2021-06-23 at 12:26 +0100, Matthew Auld wrote:
For some specialised objects we might need something larger than the regions min_page_size due to some hw restriction, and slightly more hairy is needing something smaller with the guarantee that such objects will never be inserted into any GTT, which is the case for the paging structures.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com
Reviewed-by: Thomas Hellström thomas.hellstrom@linux.intel.com
The min_page_size is only needed for pages inserted into the GTT, and for our paging structures we only need at most 4K bytes, so simply ignore the min_page_size restrictions here, otherwise we might see some severe overallocation on some devices.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com --- drivers/gpu/drm/i915/gt/intel_gtt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c index 084ea65d59c0..61e8a8c25374 100644 --- a/drivers/gpu/drm/i915/gt/intel_gtt.c +++ b/drivers/gpu/drm/i915/gt/intel_gtt.c @@ -16,7 +16,7 @@ struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz) { struct drm_i915_gem_object *obj;
- obj = i915_gem_object_create_lmem(vm->i915, sz, 0); + obj = __i915_gem_object_create_lmem_with_ps(vm->i915, sz, sz, 0); /* * Ensure all paging structures for this vm share the same dma-resv * object underneath, with the idea that one object_lock() will lock
On 6/23/21 1:26 PM, Matthew Auld wrote:
The min_page_size is only needed for pages inserted into the GTT, and for our paging structures we only need at most 4K bytes, so simply ignore the min_page_size restrictions here, otherwise we might see some severe overallocation on some devices.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com
drivers/gpu/drm/i915/gt/intel_gtt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c index 084ea65d59c0..61e8a8c25374 100644 --- a/drivers/gpu/drm/i915/gt/intel_gtt.c +++ b/drivers/gpu/drm/i915/gt/intel_gtt.c @@ -16,7 +16,7 @@ struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz) { struct drm_i915_gem_object *obj;
- obj = i915_gem_object_create_lmem(vm->i915, sz, 0);
- obj = __i915_gem_object_create_lmem_with_ps(vm->i915, sz, sz, 0); /*
- Ensure all paging structures for this vm share the same dma-resv
- object underneath, with the idea that one object_lock() will lock
I think for this one the new gt migration code might break, because there we insert even PT pages into the GTT, so it might need a special interface? Ram is looking at supporter larger GPU PTE sizes with that code..
/Thomas
On 23/06/2021 12:51, Thomas Hellström wrote:
On 6/23/21 1:26 PM, Matthew Auld wrote:
The min_page_size is only needed for pages inserted into the GTT, and for our paging structures we only need at most 4K bytes, so simply ignore the min_page_size restrictions here, otherwise we might see some severe overallocation on some devices.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com
drivers/gpu/drm/i915/gt/intel_gtt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c index 084ea65d59c0..61e8a8c25374 100644 --- a/drivers/gpu/drm/i915/gt/intel_gtt.c +++ b/drivers/gpu/drm/i915/gt/intel_gtt.c @@ -16,7 +16,7 @@ struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz) { struct drm_i915_gem_object *obj; - obj = i915_gem_object_create_lmem(vm->i915, sz, 0); + obj = __i915_gem_object_create_lmem_with_ps(vm->i915, sz, sz, 0); /* * Ensure all paging structures for this vm share the same dma-resv * object underneath, with the idea that one object_lock() will lock
I think for this one the new gt migration code might break, because there we insert even PT pages into the GTT, so it might need a special interface? Ram is looking at supporter larger GPU PTE sizes with that code..
For DG1 at least we don't need this. But yeah we can always just pass along the page size when allocating the stash I guess, if we need something special for migration?
But when we need to support huge PTEs for stuff other than DG1, then it's still a pile of work I assume, since we still need all the special PTE insertion routines specifically for insert_pte() which will differ wildly between generations, also each has quite different restrictions wrt min physical alignment of lmem, whether you can mix 64K/4K PTEs in the same 2M va range, whether 4K PTEs are even supported for lmem etc.
Not sure if it's simpler to go with mapping all of lmem upfront with the flat-ppGTT? Maybe that sidesteps some of these issues? At least for the physical alignment of paging structures that would no longer be a concern.
/Thomas
On Wed, 2021-06-23 at 13:25 +0100, Matthew Auld wrote:
On 23/06/2021 12:51, Thomas Hellström wrote:
On 6/23/21 1:26 PM, Matthew Auld wrote:
The min_page_size is only needed for pages inserted into the GTT, and for our paging structures we only need at most 4K bytes, so simply ignore the min_page_size restrictions here, otherwise we might see some severe overallocation on some devices.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com
drivers/gpu/drm/i915/gt/intel_gtt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c index 084ea65d59c0..61e8a8c25374 100644 --- a/drivers/gpu/drm/i915/gt/intel_gtt.c +++ b/drivers/gpu/drm/i915/gt/intel_gtt.c @@ -16,7 +16,7 @@ struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz) { struct drm_i915_gem_object *obj; - obj = i915_gem_object_create_lmem(vm->i915, sz, 0); + obj = __i915_gem_object_create_lmem_with_ps(vm->i915, sz, sz, 0); /* * Ensure all paging structures for this vm share the same dma-resv * object underneath, with the idea that one object_lock() will lock
I think for this one the new gt migration code might break, because there we insert even PT pages into the GTT, so it might need a special interface? Ram is looking at supporter larger GPU PTE sizes with that code..
For DG1 at least we don't need this. But yeah we can always just pass along the page size when allocating the stash I guess, if we need something special for migration?
But when we need to support huge PTEs for stuff other than DG1, then it's still a pile of work I assume, since we still need all the special PTE insertion routines specifically for insert_pte() which will differ wildly between generations, also each has quite different restrictions wrt min physical alignment of lmem, whether you can mix 64K/4K PTEs in the same 2M va range, whether 4K PTEs are even supported for lmem etc.
Not sure if it's simpler to go with mapping all of lmem upfront with the flat-ppGTT? Maybe that sidesteps some of these issues? At least for the physical alignment of paging structures that would no longer be a concern.
Yes, that might be the simplest way forward.
/Thomas
/Thomas
On 6/23/21 1:26 PM, Matthew Auld wrote:
The min_page_size is only needed for pages inserted into the GTT, and for our paging structures we only need at most 4K bytes, so simply ignore the min_page_size restrictions here, otherwise we might see some severe overallocation on some devices.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com
drivers/gpu/drm/i915/gt/intel_gtt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c index 084ea65d59c0..61e8a8c25374 100644 --- a/drivers/gpu/drm/i915/gt/intel_gtt.c +++ b/drivers/gpu/drm/i915/gt/intel_gtt.c @@ -16,7 +16,7 @@ struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz) { struct drm_i915_gem_object *obj;
- obj = i915_gem_object_create_lmem(vm->i915, sz, 0);
- obj = __i915_gem_object_create_lmem_with_ps(vm->i915, sz, sz, 0);
So is this memory always required to be size aligned? or should it say sz, PAGE_SIZE?
/Thomas
On 23/06/2021 14:32, Thomas Hellström wrote:
On 6/23/21 1:26 PM, Matthew Auld wrote:
The min_page_size is only needed for pages inserted into the GTT, and for our paging structures we only need at most 4K bytes, so simply ignore the min_page_size restrictions here, otherwise we might see some severe overallocation on some devices.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com
drivers/gpu/drm/i915/gt/intel_gtt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c index 084ea65d59c0..61e8a8c25374 100644 --- a/drivers/gpu/drm/i915/gt/intel_gtt.c +++ b/drivers/gpu/drm/i915/gt/intel_gtt.c @@ -16,7 +16,7 @@ struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz) { struct drm_i915_gem_object *obj; - obj = i915_gem_object_create_lmem(vm->i915, sz, 0); + obj = __i915_gem_object_create_lmem_with_ps(vm->i915, sz, sz, 0);
So is this memory always required to be size aligned? or should it say sz, PAGE_SIZE?
The scratch page also hits this path, which is another can of worms. In terms of size it might need to be 64K(with proper physical alignment), which is why we can't force 4K here, and instead need to use the passed in size, where the returned page will have the same alignment.
/Thomas
On 6/23/21 3:38 PM, Matthew Auld wrote:
On 23/06/2021 14:32, Thomas Hellström wrote:
On 6/23/21 1:26 PM, Matthew Auld wrote:
The min_page_size is only needed for pages inserted into the GTT, and for our paging structures we only need at most 4K bytes, so simply ignore the min_page_size restrictions here, otherwise we might see some severe overallocation on some devices.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com
drivers/gpu/drm/i915/gt/intel_gtt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c index 084ea65d59c0..61e8a8c25374 100644 --- a/drivers/gpu/drm/i915/gt/intel_gtt.c +++ b/drivers/gpu/drm/i915/gt/intel_gtt.c @@ -16,7 +16,7 @@ struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz) { struct drm_i915_gem_object *obj; - obj = i915_gem_object_create_lmem(vm->i915, sz, 0); + obj = __i915_gem_object_create_lmem_with_ps(vm->i915, sz, sz, 0);
So is this memory always required to be size aligned? or should it say sz, PAGE_SIZE?
The scratch page also hits this path, which is another can of worms. In terms of size it might need to be 64K(with proper physical alignment), which is why we can't force 4K here, and instead need to use the passed in size, where the returned page will have the same alignment.
OK. Perhaps a comment to explain that?
Reviewed-by: Thomas Hellström thomas.hellstrom@linux.intel.com
/Thomas
On Wed, 2021-06-23 at 12:26 +0100, Matthew Auld wrote:
Just checking the current region is not enough, if we later migrate the object somewhere else. For example if the placements are {SMEM, LMEM}, then we might get this wrong. Another idea might be to make the page_alignment part of the ttm_place, instead of the BO.
Signed-off-by: Matthew Auld matthew.auld@intel.com Cc: Thomas Hellström thomas.hellstrom@linux.intel.com
drivers/gpu/drm/i915/gem/i915_gem_ttm.c | 21 ++++++++++++++++++++- 1 file changed, 20 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c index c5deb8b7227c..5d894bba6430 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_ttm.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_ttm.c @@ -753,6 +753,25 @@ void i915_ttm_bo_destroy(struct ttm_buffer_object *bo) call_rcu(&obj->rcu, __i915_gem_free_object_rcu); } +static u64 i915_gem_object_page_size(struct drm_i915_gem_object *obj) +{ + u64 page_size; + int i;
+ if (!obj->mm.n_placements) + return obj->mm.region->min_page_size;
+ page_size = 0; + for (i = 0; i < obj->mm.n_placements; i++) { + struct intel_memory_region *mr = obj-
mm.placements[i];
+ page_size = max_t(u64, mr->min_page_size, page_size); + }
+ GEM_BUG_ON(!page_size); + return page_size; +}
I think if at all possible, we really should try to avoid the above. Could we, just like in your next patch, perhaps set alignment to 0, indicating that we don't care at the per-object level and something else, indicating that we care.
Then the manager could use its default if we don't care and the indicated alignment, even if it's less, if we care at the per object level?
/Thomas
/** * __i915_gem_ttm_object_init - Initialize a ttm-backed i915 gem object * @mem: The initial memory region for the object. @@ -793,7 +812,7 @@ int __i915_gem_ttm_object_init(struct intel_memory_region *mem, obj->base.vma_node.driver_private = i915_gem_to_ttm(obj); ret = ttm_bo_init(&i915->bdev, i915_gem_to_ttm(obj), size, bo_type, &i915_sys_placement, - mem->min_page_size >> PAGE_SHIFT, + i915_gem_object_page_size(obj) >> PAGE_SHIFT,
true, NULL, NULL, i915_ttm_bo_destroy); if (!ret) obj->ttm.created = true;
dri-devel@lists.freedesktop.org
-
Matthew Auld -
Thomas Hellström