Merge branch 'for-airlied' of git://people.freedesktop.org/~danvet/drm-intel into drm-next

Last pile of stuff for 3.7, essentially just a bunch of bigger fixes and a
few less intrusive features:

- cpu freq interface in sysfs from Ben
- cpu edp fixes and some related cleanups
- write-combining ptes for pre-gen6 (Chris)
- basic CADL support (Peter Wu), this fixes quite a few issues with
  backlights ...
- rework of the gem backing pages handling (preps for stolen mem handling)
  from Chris
- some more cleanup-fallout from the modeset-rework

On top of that I've done a backmerge of -rc7(since the conflicts got too
messy and I've pushed out broken merged trees too often). I've also
included 3 fixes on top of what QA beat on:

- Fix for a infoframe handling regression in 3.5 - infoframe blows up too
  often and 3.6 is pretty much done, so I'd like to merge that through
  -next and the stable process and give it more exposure before it lands
  in a stable tree.
- ioctl cosmetics^Wspelling fix in the structs (userspace won't be
  affected, since all existing userspace uses private copies of the ioctl
  struct definitions, and the struct layout itself is abi compatible).
- Bugfix for a regression introduced in this pull's testing cycle.

* 'for-airlied' of git://people.freedesktop.org/~danvet/drm-intel: (695 commits)
  drm/i915: Wrap external callers to IPS state with appropriate locks
  drm/i915: s/cacheing/caching/
  drm/i915: make sure we write all the DIP data bytes
  drm/i915: BUG() on unexpected HDMI register
  ...

1 files changed, 9 insertions, 2 deletions

diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
index 0321007086f..8d4214afc21 100644
--- a/arch/powerpc/kernel/smp.c
+++ b/arch/powerpc/kernel/smp.c
@@ -198,8 +198,15 @@ void smp_muxed_ipi_message_pass(int cpu, int msg)
 	struct cpu_messages *info = &per_cpu(ipi_message, cpu);
 	char *message = (char *)&info->messages;
 
+	/*
+	 * Order previous accesses before accesses in the IPI handler.
+	 */
+	smp_mb();
 	message[msg] = 1;
-	mb();
+	/*
+	 * cause_ipi functions are required to include a full barrier
+	 * before doing whatever causes the IPI.
+	 */
 	smp_ops->cause_ipi(cpu, info->data);
 }
 
@@ -211,7 +218,7 @@ irqreturn_t smp_ipi_demux(void)
 	mb();	/* order any irq clear */
 
 	do {
-		all = xchg_local(&info->messages, 0);
+		all = xchg(&info->messages, 0);
 
 #ifdef __BIG_ENDIAN
 		if (all & (1 << (24 - 8 * PPC_MSG_CALL_FUNCTION)))