OSDN Git Service

drm/i915/ringbuffer: Use the HEAD auto-reporting mechanism
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 29 Oct 2010 20:44:37 +0000 (21:44 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Fri, 29 Oct 2010 20:49:41 +0000 (21:49 +0100)
My Sandybridge only reports 0 for the ring buffer registers, causing it
to hang as soon as we exhaust the available ring. As a workaround, take
advantage of our huge ring buffers and use the auto-reporting mechanism
to update the status page with the HEAD location every 64 KiB.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
drivers/gpu/drm/i915/intel_ringbuffer.c

index 2e72d3a..390aa21 100644 (file)
@@ -174,7 +174,7 @@ static int init_ring_common(struct intel_ring_buffer *ring)
 
        I915_WRITE_CTL(ring,
                        ((ring->gem_object->size - PAGE_SIZE) & RING_NR_PAGES)
-                       | RING_NO_REPORT | RING_VALID);
+                       | RING_REPORT_64K | RING_VALID);
 
        /* If the head is still not zero, the ring is dead */
        if ((I915_READ_CTL(ring) & RING_VALID) == 0 ||
@@ -691,6 +691,17 @@ int intel_wait_ring_buffer(struct intel_ring_buffer *ring, int n)
        struct drm_device *dev = ring->dev;
        drm_i915_private_t *dev_priv = dev->dev_private;
        unsigned long end;
+       u32 head;
+
+       head = intel_read_status_page(ring, 4);
+       if (head) {
+               ring->head = head & HEAD_ADDR;
+               ring->space = ring->head - (ring->tail + 8);
+               if (ring->space < 0)
+                       ring->space += ring->size;
+               if (ring->space >= n)
+                       return 0;
+       }
 
        trace_i915_ring_wait_begin (dev);
        end = jiffies + 3 * HZ;