On Tue, 2012-07-10 at 19:47 -0500, Scott Wood wrote:
On 07/10/2012 07:44 PM, Alexander Graf wrote:
quoted
On 11.07.2012, at 02:34, Scott Wood wrote:
quoted
+#ifdef CONFIG_BOOKE
+ /*
+ * We're not changing address space on Book E, and the extra rfi
+ * can hurt when virtualized without hardware support -- whereas
+ * mtmsr can be paravirtualized.
We can always paravirtualize RFI as well if it makes sense.
I'm not sure that's possible. We thought about it a while back, but
IIRC the difficulty was not leaving a register clobbered.
Besides mtmsr is slow on real HW as well. Also paravirt as done today
for complex instructions like mtmsr is racy :-) (I already had a chat
about that with Alex a while back, we might want to re-consider what
kind of fix can be done at some point).
Cheers,
Ben.