drm/i915: fixup forcewake spinlock fallout in drpc debugfs function
authorDaniel Vetter <daniel.vetter@ffwll.ch>
Wed, 25 Jan 2012 12:52:43 +0000 (13:52 +0100)
committerKeith Packard <keithp@keithp.com>
Wed, 25 Jan 2012 17:15:33 +0000 (09:15 -0800)
My forcewake spinlock patches have a functional conflict with Ben
Widawsky's gen6 drpc support for debugfs. Result was a benign warning
about trying to read an non-atomic variabla with atomic_read.

Note that the entire check is racy anyway and purely informational.
Also update it to reflect the forcewake voodoo changes, the kernel can
now also hold onto a forcewake reference for longer times.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Keith Packard <keithp@keithp.com>
drivers/gpu/drm/i915/i915_debugfs.c

index 99f1407..deaa657 100644 (file)
@@ -1075,6 +1075,7 @@ static int gen6_drpc_info(struct seq_file *m)
        struct drm_device *dev = node->minor->dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
        u32 rpmodectl1, gt_core_status, rcctl1;
+       unsigned forcewake_count;
        int count=0, ret;
 
 
@@ -1082,9 +1083,13 @@ static int gen6_drpc_info(struct seq_file *m)
        if (ret)
                return ret;
 
-       if (atomic_read(&dev_priv->forcewake_count)) {
-               seq_printf(m, "RC information inaccurate because userspace "
-                             "holds a reference \n");
+       spin_lock_irq(&dev_priv->gt_lock);
+       forcewake_count = dev_priv->forcewake_count;
+       spin_unlock_irq(&dev_priv->gt_lock);
+
+       if (forcewake_count) {
+               seq_printf(m, "RC information inaccurate because somebody "
+                             "holds a forcewake reference \n");
        } else {
                /* NB: we cannot use forcewake, else we read the wrong values */
                while (count++ < 50 && (I915_READ_NOTRACE(FORCEWAKE_ACK) & 1))