[PATCH 1/2] perfcounter: Ignore the nmi call frames in the x86-64 backtraces

[Frederic Weisbecker]

About every callchains recorded with perf record are
filled up including the internal perfcounter nmi frame:

perf_callchain
perf_counter_overflow
intel_pmu_handle_irq
perf_counter_nmi_handler
notifier_call_chain
atomic_notifier_call_chain
notify_die
do_nmi
nmi

We want ignore this frame as it's not interesting for instrumentation.
To solve this, we simply ignore every frames from nmi context.

New example of "perf report -s sym -c" after this patch:

9.59%  [k] search_by_key
             4.88%
                search_by_key
                reiserfs_read_locked_inode
                reiserfs_iget
                reiserfs_lookup
                do_lookup
                __link_path_walk
                path_walk
                do_path_lookup
                user_path_at
                vfs_fstatat
                vfs_lstat
                sys_newlstat
                system_call_fastpath
                __lxstat
                0x406fb1

             3.19%
                search_by_key
                search_by_entry_key
                reiserfs_find_entry
                reiserfs_lookup
                do_lookup
                __link_path_walk
                path_walk
                do_path_lookup
                user_path_at
                vfs_fstatat
                vfs_lstat
                sys_newlstat
                system_call_fastpath
                __lxstat
                0x406fb1
[...]

For now this patch only solves the problem in x86-64.

Signed-off-by: Frederic Weisbecker <fweisbec@xxxxxxxxx>
---
 arch/x86/include/asm/stacktrace.h  |    2 ++
 arch/x86/kernel/cpu/perf_counter.c |    8 +++++++-
 arch/x86/kernel/dumpstack_32.c     |    6 ++++++
 arch/x86/kernel/dumpstack_64.c     |   22 +++++++++++++++-------
 4 files changed, 30 insertions(+), 8 deletions(-)

diff --git a/arch/x86/include/asm/stacktrace.h b/arch/x86/include/asm/stacktrace.h
index f517944..cf86a5e 100644
--- a/arch/x86/include/asm/stacktrace.h
+++ b/arch/x86/include/asm/stacktrace.h
@@ -3,6 +3,8 @@
 
 extern int kstack_depth_to_print;
 
+int x86_is_stack_id(int id, char *name);
+
 /* Generic stack tracer with callbacks */
 
 struct stacktrace_ops {
diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c
index d4cf4ce..36c3dc7 100644
--- a/arch/x86/kernel/cpu/perf_counter.c
+++ b/arch/x86/kernel/cpu/perf_counter.c
@@ -1561,6 +1561,7 @@ void callchain_store(struct perf_callchain_entry *entry, u64 ip)
 
 static DEFINE_PER_CPU(struct perf_callchain_entry, irq_entry);
 static DEFINE_PER_CPU(struct perf_callchain_entry, nmi_entry);
+static DEFINE_PER_CPU(int, in_nmi_frame);
 
 
 static void
@@ -1576,7 +1577,9 @@ static void backtrace_warning(void *data, char *msg)
 
 static int backtrace_stack(void *data, char *name)
 {
-	/* Process all stacks: */
+	per_cpu(in_nmi_frame, smp_processor_id()) =
+			x86_is_stack_id(NMI_STACK, name);
+
 	return 0;
 }
 
@@ -1584,6 +1587,9 @@ static void backtrace_address(void *data, unsigned long addr, int reliable)
 {
 	struct perf_callchain_entry *entry = data;
 
+	if (per_cpu(in_nmi_frame, smp_processor_id()))
+		return;
+
 	if (reliable)
 		callchain_store(entry, addr);
 }
diff --git a/arch/x86/kernel/dumpstack_32.c b/arch/x86/kernel/dumpstack_32.c
index d593cd1..bca5fba 100644
--- a/arch/x86/kernel/dumpstack_32.c
+++ b/arch/x86/kernel/dumpstack_32.c
@@ -19,6 +19,12 @@
 
 #include "dumpstack.h"
 
+/* Just a stub for now */
+int x86_is_stack_id(int id, char *name)
+{
+	return 0;
+}
+
 void dump_trace(struct task_struct *task, struct pt_regs *regs,
 		unsigned long *stack, unsigned long bp,
 		const struct stacktrace_ops *ops, void *data)
diff --git a/arch/x86/kernel/dumpstack_64.c b/arch/x86/kernel/dumpstack_64.c
index d35db59..54b0a32 100644
--- a/arch/x86/kernel/dumpstack_64.c
+++ b/arch/x86/kernel/dumpstack_64.c
@@ -19,10 +19,8 @@
 
 #include "dumpstack.h"
 
-static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
-					unsigned *usedp, char **idp)
-{
-	static char ids[][8] = {
+
+static char x86_stack_ids[][8] = {
 		[DEBUG_STACK - 1] = "#DB",
 		[NMI_STACK - 1] = "NMI",
 		[DOUBLEFAULT_STACK - 1] = "#DF",
@@ -33,6 +31,15 @@ static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 			N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
 #endif
 	};
+
+int x86_is_stack_id(int id, char *name)
+{
+	return x86_stack_ids[id - 1] == name;
+}
+
+static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
+					unsigned *usedp, char **idp)
+{
 	unsigned k;
 
 	/*
@@ -61,7 +68,7 @@ static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 			if (*usedp & (1U << k))
 				break;
 			*usedp |= 1U << k;
-			*idp = ids[k];
+			*idp = x86_stack_ids[k];
 			return (unsigned long *)end;
 		}
 		/*
@@ -81,12 +88,13 @@ static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
 			do {
 				++j;
 				end -= EXCEPTION_STKSZ;
-				ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
+				x86_stack_ids[j][4] = '1' +
+						(j - N_EXCEPTION_STACKS);
 			} while (stack < end - EXCEPTION_STKSZ);
 			if (*usedp & (1U << j))
 				break;
 			*usedp |= 1U << j;
-			*idp = ids[j];
+			*idp = x86_stack_ids[j];
 			return (unsigned long *)end;
 		}
 #endif
-- 
1.6.2.3

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/