drm/i915: check that rpm ref is held when accessing ringbuf in stolen mem
authorDaniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Wed, 27 Jan 2016 15:43:49 +0000 (15:43 +0000)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Thu, 11 Feb 2016 05:57:32 +0000 (06:57 +0100)
While running some tests on the scheduler patches with rpm enabled I
came across a corruption in the ringbuffer, which was root-caused to
the GPU being suspended while commands were being emitted to the
ringbuffer. The access to memory was failing because the GPU needs to
be awake when accessing stolen memory (where my ringbuffer was located).
Since we have this constraint it looks like a sensible idea to check
that we hold a refcount when we access the rungbuffer.

v2: move the check from ring_begin to ringbuffer iomap time (Chris)
v3: update comment (Chris)

Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1453909429-11024-1-git-send-email-daniele.ceraolospurio@intel.com
drivers/gpu/drm/i915/intel_ringbuffer.c

index 6f5b511bdb5d9b41eea7011f477e9bbcf663030b..133321a5b3d0799ce74bc7999370558c6bf5a353 100644 (file)
@@ -2119,6 +2119,9 @@ int intel_pin_and_map_ringbuffer_obj(struct drm_device *dev,
                        return ret;
                }
 
+               /* Access through the GTT requires the device to be awake. */
+               assert_rpm_wakelock_held(dev_priv);
+
                ringbuf->virtual_start = ioremap_wc(dev_priv->gtt.mappable_base +
                                                    i915_gem_obj_ggtt_offset(obj), ringbuf->size);
                if (ringbuf->virtual_start == NULL) {