Support inserting 64K pages into the 48b PPGTT.
v2: check for 64K scratch
v3: we should only have to re-adjust maybe_64K at every sg interval
Signed-off-by: Matthew Auld <matthew.auld@xxxxxxxxx>
Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx>
Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/i915/i915_gem_gtt.c | 31 +++++++++++++++++++++++++++++++
drivers/gpu/drm/i915/i915_gem_gtt.h | 7 +++++++
2 files changed, 38 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 86e9b6c3a987..5b6829990657 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -1072,6 +1072,7 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma,
struct i915_page_directory_pointer *pdp = pdps[idx.pml4e];
struct i915_page_directory *pd = pdp->page_directory[idx.pdpe];
unsigned int page_size;
+ bool maybe_64K = false;
gen8_pte_t encode = pte_encode;
gen8_pte_t *vaddr;
u16 index, max;
@@ -1093,6 +1094,13 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma,
max = GEN8_PTES;
page_size = I915_GTT_PAGE_SIZE;
+ if (!index &&
+ vma->page_sizes.sg & I915_GTT_PAGE_SIZE_64K &&
+ IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_64K) &&
+ (IS_ALIGNED(rem, I915_GTT_PAGE_SIZE_64K) ||
+ rem >= (max - index) << PAGE_SHIFT))
+ maybe_64K = true;
+
vaddr = kmap_atomic_px(pt);
}
@@ -1112,12 +1120,35 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma,
iter->dma = sg_dma_address(iter->sg);
iter->max = iter->dma + rem;
+ if (maybe_64K && index < max &&
+ !(IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_64K) &&
+ (IS_ALIGNED(rem, I915_GTT_PAGE_SIZE_64K) ||
+ rem >= (max - index) << PAGE_SHIFT)))
+ maybe_64K = false;
+
if (unlikely(!IS_ALIGNED(iter->dma, page_size)))
break;
}
} while (rem >= page_size && index < max);
kunmap_atomic(vaddr);
+
+ /*
+ * Is it safe to mark the 2M block as 64K? -- Either we have
+ * filled whole page-table with 64K entries, or filled part of
+ * it and have reached the end of the sg table and we have
+ * enough padding.
+ */
+ if (maybe_64K &&
+ (index == max ||
+ (i915_vm_has_scratch_64K(vma->vm) &&
+ !iter->sg && IS_ALIGNED(vma->node.start +
+ vma->node.size,
+ I915_GTT_PAGE_SIZE_2M)))) {
+ vaddr = kmap_atomic_px(pd);
+ vaddr[idx.pde] |= GEN8_PDE_IPS_64K;
+ kunmap_atomic(vaddr);
+ }
} while (iter->sg);
}
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index 0a31dc369c28..475e4cf042be 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -153,6 +153,7 @@ typedef u64 gen8_ppgtt_pml4e_t;
#define GEN8_PPAT_GET_AGE(x) ((x) & (3 << 4))
#define CHV_PPAT_GET_SNOOP(x) ((x) & (1 << 6))
+#define GEN8_PDE_IPS_64K BIT(11)
#define GEN8_PDE_PS_2M BIT(7)
struct sg_table;
@@ -351,6 +352,12 @@ i915_vm_is_48bit(const struct i915_address_space *vm)
return (vm->total - 1) >> 32;
}
+static inline bool
+i915_vm_has_scratch_64K(struct i915_address_space *vm)
+{
+ return vm->scratch_page.order == get_order(I915_GTT_PAGE_SIZE_64K);
+}
+
/* The Graphics Translation Table is the way in which GEN hardware translates a
* Graphics Virtual Address into a Physical Address. In addition to the normal
* collateral associated with any va->pa translations GEN hardware also has a
--
2.13.5
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
