drm/i915: merge shmem_pwrite slow&fast-path
authorDaniel Vetter <daniel.vetter@ffwll.ch>
Sun, 25 Mar 2012 17:47:28 +0000 (19:47 +0200)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Tue, 27 Mar 2012 11:18:58 +0000 (13:18 +0200)
With the previous rewrite, they've become essential identical.

v2: Simplify the page_do_bit17_swizzling logic as suggested by Chris
Wilson.

Tested-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_gem.c

index 09c033e5e02bb26c81dbd61808e709ceb06fc92f..a2547528a9d44b3296884feedd45b684fa2f9527 100644 (file)
@@ -691,84 +691,11 @@ out_unpin_pages:
        return ret;
 }
 
-/**
- * This is the fast shmem pwrite path, which attempts to directly
- * copy_from_user into the kmapped pages backing the object.
- */
-static int
-i915_gem_shmem_pwrite_fast(struct drm_device *dev,
-                          struct drm_i915_gem_object *obj,
-                          struct drm_i915_gem_pwrite *args,
-                          struct drm_file *file)
-{
-       struct address_space *mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping;
-       ssize_t remain;
-       loff_t offset;
-       char __user *user_data;
-       int page_offset, page_length;
-
-       user_data = (char __user *) (uintptr_t) args->data_ptr;
-       remain = args->size;
-
-       offset = args->offset;
-       obj->dirty = 1;
-
-       while (remain > 0) {
-               struct page *page;
-               char *vaddr;
-               int ret;
-
-               /* Operation in this page
-                *
-                * page_offset = offset within page
-                * page_length = bytes to copy for this page
-                */
-               page_offset = offset_in_page(offset);
-               page_length = remain;
-               if ((page_offset + remain) > PAGE_SIZE)
-                       page_length = PAGE_SIZE - page_offset;
-
-               page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
-               if (IS_ERR(page))
-                       return PTR_ERR(page);
-
-               vaddr = kmap_atomic(page);
-               ret = __copy_from_user_inatomic(vaddr + page_offset,
-                                               user_data,
-                                               page_length);
-               kunmap_atomic(vaddr);
-
-               set_page_dirty(page);
-               mark_page_accessed(page);
-               page_cache_release(page);
-
-               /* If we get a fault while copying data, then (presumably) our
-                * source page isn't available.  Return the error and we'll
-                * retry in the slow path.
-                */
-               if (ret)
-                       return -EFAULT;
-
-               remain -= page_length;
-               user_data += page_length;
-               offset += page_length;
-       }
-
-       return 0;
-}
-
-/**
- * This is the fallback shmem pwrite path, which uses get_user_pages to pin
- * the memory and maps it using kmap_atomic for copying.
- *
- * This avoids taking mmap_sem for faulting on the user's address while the
- * struct_mutex is held.
- */
 static int
-i915_gem_shmem_pwrite_slow(struct drm_device *dev,
-                          struct drm_i915_gem_object *obj,
-                          struct drm_i915_gem_pwrite *args,
-                          struct drm_file *file)
+i915_gem_shmem_pwrite(struct drm_device *dev,
+                     struct drm_i915_gem_object *obj,
+                     struct drm_i915_gem_pwrite *args,
+                     struct drm_file *file)
 {
        struct address_space *mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping;
        ssize_t remain;
@@ -776,6 +703,7 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev,
        char __user *user_data;
        int shmem_page_offset, page_length, ret = 0;
        int obj_do_bit17_swizzling, page_do_bit17_swizzling;
+       int hit_slowpath = 0;
 
        user_data = (char __user *) (uintptr_t) args->data_ptr;
        remain = args->size;
@@ -785,8 +713,6 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev,
        offset = args->offset;
        obj->dirty = 1;
 
-       mutex_unlock(&dev->struct_mutex);
-
        while (remain > 0) {
                struct page *page;
                char *vaddr;
@@ -811,6 +737,21 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev,
                page_do_bit17_swizzling = obj_do_bit17_swizzling &&
                        (page_to_phys(page) & (1 << 17)) != 0;
 
+               if (!page_do_bit17_swizzling) {
+                       vaddr = kmap_atomic(page);
+                       ret = __copy_from_user_inatomic(vaddr + shmem_page_offset,
+                                                       user_data,
+                                                       page_length);
+                       kunmap_atomic(vaddr);
+
+                       if (ret == 0)
+                               goto next_page;
+               }
+
+               hit_slowpath = 1;
+
+               mutex_unlock(&dev->struct_mutex);
+
                vaddr = kmap(page);
                if (page_do_bit17_swizzling)
                        ret = __copy_from_user_swizzled(vaddr, shmem_page_offset,
@@ -822,6 +763,8 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev,
                                               page_length);
                kunmap(page);
 
+               mutex_lock(&dev->struct_mutex);
+next_page:
                set_page_dirty(page);
                mark_page_accessed(page);
                page_cache_release(page);
@@ -837,15 +780,16 @@ i915_gem_shmem_pwrite_slow(struct drm_device *dev,
        }
 
 out:
-       mutex_lock(&dev->struct_mutex);
-       /* Fixup: Kill any reinstated backing storage pages */
-       if (obj->madv == __I915_MADV_PURGED)
-               i915_gem_object_truncate(obj);
-       /* and flush dirty cachelines in case the object isn't in the cpu write
-        * domain anymore. */
-       if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) {
-               i915_gem_clflush_object(obj);
-               intel_gtt_chipset_flush();
+       if (hit_slowpath) {
+               /* Fixup: Kill any reinstated backing storage pages */
+               if (obj->madv == __I915_MADV_PURGED)
+                       i915_gem_object_truncate(obj);
+               /* and flush dirty cachelines in case the object isn't in the cpu write
+                * domain anymore. */
+               if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) {
+                       i915_gem_clflush_object(obj);
+                       intel_gtt_chipset_flush();
+               }
        }
 
        return ret;
@@ -939,11 +883,7 @@ out_unpin:
        if (ret)
                goto out;
 
-       ret = -EFAULT;
-       if (!i915_gem_object_needs_bit17_swizzle(obj))
-               ret = i915_gem_shmem_pwrite_fast(dev, obj, args, file);
-       if (ret == -EFAULT)
-               ret = i915_gem_shmem_pwrite_slow(dev, obj, args, file);
+       ret = i915_gem_shmem_pwrite(dev, obj, args, file);
 
 out:
        drm_gem_object_unreference(&obj->base);