drm/i915/bdw: Implement PPGTT clear range
authorBen Widawsky <benjamin.widawsky@intel.com>
Sun, 3 Nov 2013 04:07:23 +0000 (21:07 -0700)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Fri, 8 Nov 2013 17:09:46 +0000 (18:09 +0100)
GEN8 PPGTT range clearing is very similar to GEN6 if we assume that our
PDEs are all valid, which they should be.

v2: Rebase on top of the address space refactoring.

v3: Rebase on top of the bool use_scratch addition to the clear_range interface.

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_gem_gtt.c

index 9bee72c2c3b15aecf57282cabefcab8a97626b20..2f138ed3730ea069f00ddeb973b8e7aae97db9cc 100644 (file)
@@ -59,6 +59,7 @@ typedef gen8_gtt_pte_t gen8_ppgtt_pde_t;
 #define HSW_WB_ELLC_LLC_AGE0           HSW_CACHEABILITY_CONTROL(0xb)
 #define HSW_WT_ELLC_LLC_AGE0           HSW_CACHEABILITY_CONTROL(0x6)
 
+#define GEN8_PTES_PER_PAGE             (PAGE_SIZE / sizeof(gen8_gtt_pte_t))
 #define GEN8_PDES_PER_PAGE             (PAGE_SIZE / sizeof(gen8_ppgtt_pde_t))
 #define GEN8_LEGACY_PDPS               4
 
@@ -194,6 +195,41 @@ static gen6_gtt_pte_t iris_pte_encode(dma_addr_t addr,
        return pte;
 }
 
+static void gen8_ppgtt_clear_range(struct i915_address_space *vm,
+                                  unsigned first_entry,
+                                  unsigned num_entries,
+                                  bool use_scratch)
+{
+       struct i915_hw_ppgtt *ppgtt =
+               container_of(vm, struct i915_hw_ppgtt, base);
+       gen8_gtt_pte_t *pt_vaddr, scratch_pte;
+       unsigned act_pt = first_entry / GEN8_PTES_PER_PAGE;
+       unsigned first_pte = first_entry % GEN8_PTES_PER_PAGE;
+       unsigned last_pte, i;
+
+       scratch_pte = gen8_pte_encode(ppgtt->base.scratch.addr,
+                                     I915_CACHE_LLC, use_scratch);
+
+       while (num_entries) {
+               struct page *page_table = &ppgtt->gen8_pt_pages[act_pt];
+
+               last_pte = first_pte + num_entries;
+               if (last_pte > GEN8_PTES_PER_PAGE)
+                       last_pte = GEN8_PTES_PER_PAGE;
+
+               pt_vaddr = kmap_atomic(page_table);
+
+               for (i = first_pte; i < last_pte; i++)
+                       pt_vaddr[i] = scratch_pte;
+
+               kunmap_atomic(pt_vaddr);
+
+               num_entries -= last_pte - first_pte;
+               first_pte = 0;
+               act_pt++;
+       }
+}
+
 static void gen8_ppgtt_cleanup(struct i915_address_space *vm)
 {
        struct i915_hw_ppgtt *ppgtt =
@@ -259,7 +295,7 @@ static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
        ppgtt->num_pd_pages = 1 << get_order(max_pdp << PAGE_SHIFT);
        ppgtt->num_pt_pages = 1 << get_order(num_pt_pages << PAGE_SHIFT);
        ppgtt->num_pd_entries = max_pdp * GEN8_PDES_PER_PAGE;
-       ppgtt->base.clear_range = NULL;
+       ppgtt->base.clear_range = gen8_ppgtt_clear_range;
        ppgtt->base.insert_entries = NULL;
        ppgtt->base.cleanup = gen8_ppgtt_cleanup;
 
@@ -312,6 +348,10 @@ static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
                kunmap_atomic(pd_vaddr);
        }
 
+       ppgtt->base.clear_range(&ppgtt->base, 0,
+                               ppgtt->num_pd_entries * GEN8_PTES_PER_PAGE,
+                               true);
+
        DRM_DEBUG_DRIVER("Allocated %d pages for page directories (%d wasted)\n",
                         ppgtt->num_pd_pages, ppgtt->num_pd_pages - max_pdp);
        DRM_DEBUG_DRIVER("Allocated %d pages for page tables (%lld wasted)\n",