In preparation for supporting huge gtt pages for the ppgtt, we introduce
page size members for gem objects. We fill in the page sizes by
scanning the sg table.
v2: pass the sg_mask to set_pages
v3: calculate the sg_mask inline with populating the sg_table where
possible, and pass to set_pages along with the pages.
v4: bunch of improvements from Joonas
Signed-off-by: Matthew Auld <matthew.auld@xxxxxxxxx>
Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx>
Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Daniel Vetter <daniel@xxxxxxxx>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx>
Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/i915/i915_drv.h | 5 ++-
drivers/gpu/drm/i915/i915_gem.c | 42 +++++++++++++++++++++---
drivers/gpu/drm/i915/i915_gem_dmabuf.c | 9 ++++-
drivers/gpu/drm/i915/i915_gem_internal.c | 5 ++-
drivers/gpu/drm/i915/i915_gem_object.h | 17 ++++++++++
drivers/gpu/drm/i915/i915_gem_stolen.c | 2 +-
drivers/gpu/drm/i915/i915_gem_userptr.c | 9 ++++-
drivers/gpu/drm/i915/selftests/huge_gem_object.c | 2 +-
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 5 ++-
9 files changed, 84 insertions(+), 12 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index b7ecf43080a8..bbdfb15c6183 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3085,6 +3085,8 @@ intel_info(const struct drm_i915_private *dev_priv)
#define USES_PPGTT(dev_priv) (i915_modparams.enable_ppgtt)
#define USES_FULL_PPGTT(dev_priv) (i915_modparams.enable_ppgtt >= 2)
#define USES_FULL_48BIT_PPGTT(dev_priv) (i915_modparams.enable_ppgtt == 3)
+#define HAS_PAGE_SIZES(dev_priv, sizes) \
+ ((sizes) && (((sizes) & ~(dev_priv)->info.page_sizes)) == 0)
#define HAS_OVERLAY(dev_priv) ((dev_priv)->info.has_overlay)
#define OVERLAY_NEEDS_PHYSICAL(dev_priv) \
@@ -3501,7 +3503,8 @@ i915_gem_object_get_dma_address(struct drm_i915_gem_object *obj,
unsigned long n);
void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
- struct sg_table *pages);
+ struct sg_table *pages,
+ unsigned int sg_mask);
int __i915_gem_object_get_pages(struct drm_i915_gem_object *obj);
static inline int __must_check
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index f26385732635..c9460433372a 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -228,7 +228,7 @@ static int i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj)
obj->phys_handle = phys;
- __i915_gem_object_set_pages(obj, st);
+ __i915_gem_object_set_pages(obj, st, sg->length);
return 0;
@@ -2266,6 +2266,8 @@ void __i915_gem_object_put_pages(struct drm_i915_gem_object *obj,
if (!IS_ERR(pages))
obj->ops->put_pages(obj, pages);
+ obj->mm.page_sizes.phys = obj->mm.page_sizes.sg = 0;
+
unlock:
mutex_unlock(&obj->mm.lock);
}
@@ -2308,6 +2310,7 @@ static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
struct page *page;
unsigned long last_pfn = 0; /* suppress gcc warning */
unsigned int max_segment = i915_sg_segment_size();
+ unsigned int sg_mask;
gfp_t noreclaim;
int ret;
@@ -2339,6 +2342,7 @@ static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
sg = st->sgl;
st->nents = 0;
+ sg_mask = 0;
for (i = 0; i < page_count; i++) {
const unsigned int shrink[] = {
I915_SHRINK_BOUND | I915_SHRINK_UNBOUND | I915_SHRINK_PURGEABLE,
@@ -2391,8 +2395,10 @@ static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
if (!i ||
sg->length >= max_segment ||
page_to_pfn(page) != last_pfn + 1) {
- if (i)
+ if (i) {
+ sg_mask |= sg->length;
sg = sg_next(sg);
+ }
st->nents++;
sg_set_page(sg, page, PAGE_SIZE, 0);
} else {
@@ -2403,8 +2409,10 @@ static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
/* Check that the i965g/gm workaround works. */
WARN_ON((gfp & __GFP_DMA32) && (last_pfn >= 0x00100000UL));
}
- if (sg) /* loop terminated early; short sg table */
+ if (sg) { /* loop terminated early; short sg table */
+ sg_mask |= sg->length;
sg_mark_end(sg);
+ }
/* Trim unused sg entries to avoid wasting memory. */
i915_sg_trim(st);
@@ -2433,7 +2441,7 @@ static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
if (i915_gem_object_needs_bit17_swizzle(obj))
i915_gem_object_do_bit_17_swizzle(obj, st);
- __i915_gem_object_set_pages(obj, st);
+ __i915_gem_object_set_pages(obj, st, sg_mask);
return 0;
@@ -2460,8 +2468,13 @@ static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
}
void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
- struct sg_table *pages)
+ struct sg_table *pages,
+ unsigned int sg_mask)
{
+ struct drm_i915_private *i915 = to_i915(obj->base.dev);
+ unsigned long supported_page_sizes = INTEL_INFO(i915)->page_sizes;
+ int i;
+
lockdep_assert_held(&obj->mm.lock);
obj->mm.get_page.sg_pos = pages->sgl;
@@ -2475,6 +2488,25 @@ void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
__i915_gem_object_pin_pages(obj);
obj->mm.quirked = true;
}
+
+ GEM_BUG_ON(!sg_mask);
+ obj->mm.page_sizes.phys = sg_mask;
+
+ /*
+ * Calculate the supported page-sizes which fit into the given sg_mask.
+ * This will give us the page-sizes which we may be able to use
+ * opportunistically when later inserting into the GTT. For example if
+ * phys=2G, then in theory we should be able to use 1G, 2M, 64K or 4K
+ * pages, although in practice this will depend on a number of other
+ * factors.
+ */
+ obj->mm.page_sizes.sg = 0;
+ for_each_set_bit(i, &supported_page_sizes, BITS_PER_LONG) {
+ if (obj->mm.page_sizes.phys & ~0u << i)
+ obj->mm.page_sizes.sg |= BIT(i);
+ }
+
+ GEM_BUG_ON(!HAS_PAGE_SIZES(i915, obj->mm.page_sizes.sg));
}
static int ____i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
index 4c4dc85159fb..2f80b07d7f61 100644
--- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c
+++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
@@ -259,13 +259,20 @@ struct dma_buf *i915_gem_prime_export(struct drm_device *dev,
static int i915_gem_object_get_pages_dmabuf(struct drm_i915_gem_object *obj)
{
struct sg_table *pages;
+ struct scatterlist *sg;
+ unsigned int sg_mask;
+ int n;
pages = dma_buf_map_attachment(obj->base.import_attach,
DMA_BIDIRECTIONAL);
if (IS_ERR(pages))
return PTR_ERR(pages);
- __i915_gem_object_set_pages(obj, pages);
+ sg_mask = 0;
+ for_each_sg(pages->sgl, sg, pages->nents, n)
+ sg_mask |= sg->length;
+
+ __i915_gem_object_set_pages(obj, pages, sg_mask);
return 0;
}
diff --git a/drivers/gpu/drm/i915/i915_gem_internal.c b/drivers/gpu/drm/i915/i915_gem_internal.c
index f59764da4254..bdc23c4c8783 100644
--- a/drivers/gpu/drm/i915/i915_gem_internal.c
+++ b/drivers/gpu/drm/i915/i915_gem_internal.c
@@ -49,6 +49,7 @@ static int i915_gem_object_get_pages_internal(struct drm_i915_gem_object *obj)
struct drm_i915_private *i915 = to_i915(obj->base.dev);
struct sg_table *st;
struct scatterlist *sg;
+ unsigned int sg_mask;
unsigned int npages;
int max_order;
gfp_t gfp;
@@ -87,6 +88,7 @@ static int i915_gem_object_get_pages_internal(struct drm_i915_gem_object *obj)
sg = st->sgl;
st->nents = 0;
+ sg_mask = 0;
do {
int order = min(fls(npages) - 1, max_order);
@@ -104,6 +106,7 @@ static int i915_gem_object_get_pages_internal(struct drm_i915_gem_object *obj)
} while (1);
sg_set_page(sg, page, PAGE_SIZE << order, 0);
+ sg_mask |= PAGE_SIZE << order;
st->nents++;
npages -= 1 << order;
@@ -132,7 +135,7 @@ static int i915_gem_object_get_pages_internal(struct drm_i915_gem_object *obj)
*/
obj->mm.madv = I915_MADV_DONTNEED;
- __i915_gem_object_set_pages(obj, st);
+ __i915_gem_object_set_pages(obj, st, sg_mask);
return 0;
diff --git a/drivers/gpu/drm/i915/i915_gem_object.h b/drivers/gpu/drm/i915/i915_gem_object.h
index 036e847b27f0..110672952a1c 100644
--- a/drivers/gpu/drm/i915/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/i915_gem_object.h
@@ -169,6 +169,23 @@ struct drm_i915_gem_object {
struct sg_table *pages;
void *mapping;
+ struct i915_page_sizes {
+ /**
+ * The sg mask of the pages sg_table. i.e the mask of
+ * of the lengths for each sg entry.
+ */
+ unsigned int phys;
+
+ /**
+ * The gtt page sizes we are allowed to use given the
+ * sg mask and the supported page sizes. This will
+ * express the smallest unit we can use for the whole
+ * object, as well as the larger sizes we may be able
+ * to use opportunistically.
+ */
+ unsigned int sg;
+ } page_sizes;
+
struct i915_gem_object_page_iter {
struct scatterlist *sg_pos;
unsigned int sg_idx; /* in pages, but 32bit eek! */
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index 537ecb224db0..54fd4cfa9d07 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -548,7 +548,7 @@ static int i915_gem_object_get_pages_stolen(struct drm_i915_gem_object *obj)
if (IS_ERR(pages))
return PTR_ERR(pages);
- __i915_gem_object_set_pages(obj, pages);
+ __i915_gem_object_set_pages(obj, pages, obj->stolen->size);
return 0;
}
diff --git a/drivers/gpu/drm/i915/i915_gem_userptr.c b/drivers/gpu/drm/i915/i915_gem_userptr.c
index 70ad7489827d..ad5abca1f794 100644
--- a/drivers/gpu/drm/i915/i915_gem_userptr.c
+++ b/drivers/gpu/drm/i915/i915_gem_userptr.c
@@ -405,6 +405,9 @@ __i915_gem_userptr_alloc_pages(struct drm_i915_gem_object *obj,
{
unsigned int max_segment = i915_sg_segment_size();
struct sg_table *st;
+ struct scatterlist *sg;
+ unsigned int sg_mask;
+ int n;
int ret;
st = kmalloc(sizeof(*st), GFP_KERNEL);
@@ -434,7 +437,11 @@ __i915_gem_userptr_alloc_pages(struct drm_i915_gem_object *obj,
return ERR_PTR(ret);
}
- __i915_gem_object_set_pages(obj, st);
+ sg_mask = 0;
+ for_each_sg(st->sgl, sg, num_pages, n)
+ sg_mask |= sg->length;
+
+ __i915_gem_object_set_pages(obj, st, sg_mask);
return st;
}
diff --git a/drivers/gpu/drm/i915/selftests/huge_gem_object.c b/drivers/gpu/drm/i915/selftests/huge_gem_object.c
index 41c15f3aa467..a2632df39173 100644
--- a/drivers/gpu/drm/i915/selftests/huge_gem_object.c
+++ b/drivers/gpu/drm/i915/selftests/huge_gem_object.c
@@ -80,7 +80,7 @@ static int huge_get_pages(struct drm_i915_gem_object *obj)
if (i915_gem_gtt_prepare_pages(obj, pages))
goto err;
- __i915_gem_object_set_pages(obj, pages);
+ __i915_gem_object_set_pages(obj, pages, PAGE_SIZE);
return 0;
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index aa1db375d59a..883bc19e3aaf 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -45,6 +45,7 @@ static int fake_get_pages(struct drm_i915_gem_object *obj)
#define PFN_BIAS 0x1000
struct sg_table *pages;
struct scatterlist *sg;
+ unsigned int sg_mask;
typeof(obj->base.size) rem;
pages = kmalloc(sizeof(*pages), GFP);
@@ -57,6 +58,7 @@ static int fake_get_pages(struct drm_i915_gem_object *obj)
return -ENOMEM;
}
+ sg_mask = 0;
rem = obj->base.size;
for (sg = pages->sgl; sg; sg = sg_next(sg)) {
unsigned long len = min_t(typeof(rem), rem, BIT(31));
@@ -65,6 +67,7 @@ static int fake_get_pages(struct drm_i915_gem_object *obj)
sg_set_page(sg, pfn_to_page(PFN_BIAS), len, 0);
sg_dma_address(sg) = page_to_phys(sg_page(sg));
sg_dma_len(sg) = len;
+ sg_mask |= len;
rem -= len;
}
@@ -72,7 +75,7 @@ static int fake_get_pages(struct drm_i915_gem_object *obj)
obj->mm.madv = I915_MADV_DONTNEED;
- __i915_gem_object_set_pages(obj, pages);
+ __i915_gem_object_set_pages(obj, pages, sg_mask);
return 0;
#undef GFP
--
2.13.5
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
