drm/i915: Move the mb() following release-mmap into release-mmap
authorChris Wilson <chris@chris-wilson.co.uk>
Wed, 13 Apr 2016 16:35:12 +0000 (17:35 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Thu, 14 Apr 2016 09:45:40 +0000 (10:45 +0100)
As paranoia, we want to ensure that the CPU's PTEs have been revoked for
the object before we return from i915_gem_release_mmap(). This allows us
to rely on there being no outstanding memory accesses from userspace
and guarantees serialisation of the code against concurrent access just
by calling i915_gem_release_mmap().

v2: Reduce the mb() into a wmb() following the revoke.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: "Goel, Akash" <akash.goel@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1460565315-7748-13-git-send-email-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/i915_gem.c

index 0c57b20532be6bfec39ed7b25db4618ffef53c8d..21fb41c93caa606cd2628afa8e56898b2a842608 100644 (file)
@@ -1936,11 +1936,27 @@ out:
 void
 i915_gem_release_mmap(struct drm_i915_gem_object *obj)
 {
+       /* Serialisation between user GTT access and our code depends upon
+        * revoking the CPU's PTE whilst the mutex is held. The next user
+        * pagefault then has to wait until we release the mutex.
+        */
+       lockdep_assert_held(&obj->base.dev->struct_mutex);
+
        if (!obj->fault_mappable)
                return;
 
        drm_vma_node_unmap(&obj->base.vma_node,
                           obj->base.dev->anon_inode->i_mapping);
+
+       /* Ensure that the CPU's PTE are revoked and there are not outstanding
+        * memory transactions from userspace before we return. The TLB
+        * flushing implied above by changing the PTE above *should* be
+        * sufficient, an extra barrier here just provides us with a bit
+        * of paranoid documentation about our requirement to serialise
+        * memory writes before touching registers / GSM.
+        */
+       wmb();
+
        obj->fault_mappable = false;
 }
 
@@ -3324,9 +3340,6 @@ static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj)
        if ((obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0)
                return;
 
-       /* Wait for any direct GTT access to complete */
-       mb();
-
        old_read_domains = obj->base.read_domains;
        old_write_domain = obj->base.write_domain;