drm/i915: Fallback to single page GTT mmappings for relocations

If we cannot pin the entire object into the mappable region of the GTT,
try to pin a single page instead. This is much more likely to succeed,
and prevents us falling back to the clflush slow path.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-14-chris@chris-wilson.co.uk

diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 8d0df7d81d8b1122b422a4715a46a965513d12e2..c970aabfffa3fe01246263ebd37cfeb69a0394d9 100644 (file)
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -331,6 +331,7 @@ static void reloc_cache_init(struct reloc_cache *cache,
        cache->vaddr = 0;
        cache->i915 = i915;
        cache->use_64bit_reloc = INTEL_GEN(cache->i915) >= 8;
+       cache->node.allocated = false;
 }
 
 static inline void *unmask_page(unsigned long p)
@@ -360,8 +361,19 @@ static void reloc_cache_fini(struct reloc_cache *cache)
                kunmap_atomic(vaddr);
                i915_gem_obj_finish_shmem_access((struct drm_i915_gem_object *)cache->node.mm);
        } else {
+               wmb();
                io_mapping_unmap_atomic((void __iomem *)vaddr);
-               i915_vma_unpin((struct i915_vma *)cache->node.mm);
+               if (cache->node.allocated) {
+                       struct i915_ggtt *ggtt = &cache->i915->ggtt;
+
+                       ggtt->base.clear_range(&ggtt->base,
+                                              cache->node.start,
+                                              cache->node.size,
+                                              true);
+                       drm_mm_remove_node(&cache->node);
+               } else {
+                       i915_vma_unpin((struct i915_vma *)cache->node.mm);
+               }
        }
 }
 
@@ -401,8 +413,19 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
                         struct reloc_cache *cache,
                         int page)
 {
+       struct i915_ggtt *ggtt = &cache->i915->ggtt;
+       unsigned long offset;
        void *vaddr;
 
+       if (cache->node.allocated) {
+               wmb();
+               ggtt->base.insert_page(&ggtt->base,
+                                      i915_gem_object_get_dma_address(obj, page),
+                                      cache->node.start, I915_CACHE_NONE, 0);
+               cache->page = page;
+               return unmask_page(cache->vaddr);
+       }
+
        if (cache->vaddr) {
                io_mapping_unmap_atomic(unmask_page(cache->vaddr));
        } else {
@@ -418,21 +441,38 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
 
                vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0,
                                               PIN_MAPPABLE | PIN_NONBLOCK);
-               if (IS_ERR(vma))
-                       return NULL;
+               if (IS_ERR(vma)) {
+                       memset(&cache->node, 0, sizeof(cache->node));
+                       ret = drm_mm_insert_node_in_range_generic
+                               (&ggtt->base.mm, &cache->node,
+                                4096, 0, 0,
+                                0, ggtt->mappable_end,
+                                DRM_MM_SEARCH_DEFAULT,
+                                DRM_MM_CREATE_DEFAULT);
+                       if (ret)
+                               return ERR_PTR(ret);
+               } else {
+                       ret = i915_gem_object_put_fence(obj);
+                       if (ret) {
+                               i915_vma_unpin(vma);
+                               return ERR_PTR(ret);
+                       }
 
-               ret = i915_gem_object_put_fence(obj);
-               if (ret) {
-                       i915_vma_unpin(vma);
-                       return ERR_PTR(ret);
+                       cache->node.start = vma->node.start;
+                       cache->node.mm = (void *)vma;
                }
+       }
 
-               cache->node.start = vma->node.start;
-               cache->node.mm = (void *)vma;
+       offset = cache->node.start;
+       if (cache->node.allocated) {
+               ggtt->base.insert_page(&ggtt->base,
+                                      i915_gem_object_get_dma_address(obj, page),
+                                      offset, I915_CACHE_NONE, 0);
+       } else {
+               offset += page << PAGE_SHIFT;
        }
 
-       vaddr = io_mapping_map_atomic_wc(cache->i915->ggtt.mappable,
-                                        cache->node.start + (page << PAGE_SHIFT));
+       vaddr = io_mapping_map_atomic_wc(cache->i915->ggtt.mappable, offset);
        cache->page = page;
        cache->vaddr = (unsigned long)vaddr;