perf, trace: Fix !x86 build bug

Patch b7e2ece (perf, trace: Optimize tracepoints by removing
IRQ-disable from perf/tracepoint interaction) made the
unfortunate mistake of assuming the world is x86 only, correct
this.

The problem was that perf_fetch_caller_regs() did
local_save_flags() into regs->flags, and I re-used that to
remove another local_save_flags(), forgetting !x86 doesn't have
regs->flags.

Do the reverse, remove the local_save_flags() from
perf_fetch_caller_regs() and let the ftrace site do the
local_save_flags() instead.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Cc: acme@redhat.com
Cc: efault@gmx.de
Cc: fweisbec@gmail.com
Cc: rostedt@goodmis.org
LKML-Reference: <1274778175.5882.623.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

arch/x86/kernel/cpu/perf_event.c

@@ -1717,7 +1717,11 @@ void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int ski
 	 */
 	regs->bp = rewind_frame_pointer(skip + 1);
 	regs->cs = __KERNEL_CS;
-	local_save_flags(regs->flags);
+	/*
+	 * We abuse bit 3 to pass exact information, see perf_misc_flags
+	 * and the comment with PERF_EFLAGS_EXACT.
+	 */
+	regs->flags = 0;
 }
 
 unsigned long perf_instruction_pointer(struct pt_regs *regs)

kernel/trace/trace_event_perf.c

@@ -157,6 +157,7 @@ __kprobes void *perf_trace_buf_prepare(int size, unsigned short type,
 				       struct pt_regs *regs, int *rctxp)
 {
 	struct trace_entry *entry;
+	unsigned long flags;
 	char *raw_data;
 	int pc;
 
@@ -174,7 +175,8 @@ __kprobes void *perf_trace_buf_prepare(int size, unsigned short type,
 	memset(&raw_data[size - sizeof(u64)], 0, sizeof(u64));
 
 	entry = (struct trace_entry *)raw_data;
-	tracing_generic_entry_update(entry, regs->flags, pc);
+	local_save_flags(flags);
+	tracing_generic_entry_update(entry, flags, pc);
 	entry->type = type;
 
 	return raw_data;