[Intel-gfx] [PATCH 14/22] drm/i915: support 64K pages for the 48b PPGTT

[Matthew Auld]

Support inserting 64K pages into the 48b PPGTT.

v2: check for 64K scratch

Signed-off-by: Matthew Auld <matthew.auld at intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
Cc: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 27 +++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_gem_gtt.h |  7 +++++++
 2 files changed, 34 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index a2178c33586c..11d0986a8c55 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -1021,6 +1021,7 @@ 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;
@@ -1052,10 +1053,18 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma,
 			max = GEN8_PTES;
 			page_size = I915_GTT_PAGE_SIZE;
 
+			if (vma->page_sizes.sg & I915_GTT_PAGE_SIZE_64K && !index)
+				maybe_64K = true;
+
 			vaddr = kmap_atomic_px(pt);
 		}
 
 		do {
+			if (maybe_64K && (index % 16 == 0) &&
+			    (!IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_64K) ||
+			     rem < I915_GTT_PAGE_SIZE_64K))
+				maybe_64K = false;
+
 			GEM_BUG_ON(iter->sg->length < page_size);
 			vaddr[index++] = encode | iter->dma;
 
@@ -1079,6 +1088,24 @@ static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma,
 
 		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) {
+			if (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 356fec26e8c9..22b8fd233f30 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -149,6 +149,7 @@ typedef u64 gen8_ppgtt_pml4e_t;
 #define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
 #define GEN8_PPAT(i, x)			((u64)(x) << ((i) * 8))
 
+#define GEN8_PDE_IPS_64K BIT(11)
 #define GEN8_PDE_PS_2M   BIT(7)
 
 #define GEN8_PDPE_PS_1G  BIT(7)
@@ -349,6 +350,12 @@ 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
-- 
2.13.4