perf: Fix build breakage for architecutes without atomic64_t

The local64.h include dependency was not dependent on PERF_EVENT=y,
which meant that arch's without atomic64_t support ended up including
it and failed to build.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 7342979..1218d05 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -462,6 +462,7 @@
 
 #ifdef CONFIG_PERF_EVENTS
 # include <asm/perf_event.h>
+# include <asm/local64.h>
 #endif
 
 struct perf_guest_info_callbacks {
@@ -487,7 +488,6 @@
 #include <linux/cpu.h>
 #include <asm/atomic.h>
 #include <asm/local.h>
-#include <asm/local64.h>
 
 #define PERF_MAX_STACK_DEPTH		255