[for-next][PATCH 14/20] tracing: Use stack of calling function for stack tracer

[for-next][PATCH 14/20] tracing: Use stack of calling function for stack tracer

[Steven Rostedt]

[-- Attachment #1: Type: text/plain, Size: 2040 bytes --]

From: "Steven Rostedt (Red Hat)" <rostedt@goodmis.org>

Use the stack of stack_trace_call() instead of check_stack() as
the test pointer for max stack size. It makes it a bit cleaner
and a little more accurate.

Adding stable, as a later fix depends on this patch.

Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
---
 kernel/trace/trace_stack.c |   12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c
index 42ca822..dc02e29 100644
--- a/kernel/trace/trace_stack.c
+++ b/kernel/trace/trace_stack.c
@@ -39,20 +39,21 @@ static DEFINE_MUTEX(stack_sysctl_mutex);
 int stack_tracer_enabled;
 static int last_stack_tracer_enabled;
 
-static inline void check_stack(void)
+static inline void
+check_stack(unsigned long *stack)
 {
 	unsigned long this_size, flags;
 	unsigned long *p, *top, *start;
 	int i;
 
-	this_size = ((unsigned long)&this_size) & (THREAD_SIZE-1);
+	this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
 	this_size = THREAD_SIZE - this_size;
 
 	if (this_size <= max_stack_size)
 		return;
 
 	/* we do not handle interrupt stacks yet */
-	if (!object_is_on_stack(&this_size))
+	if (!object_is_on_stack(stack))
 		return;
 
 	local_irq_save(flags);
@@ -73,7 +74,7 @@ static inline void check_stack(void)
 	 * Now find where in the stack these are.
 	 */
 	i = 0;
-	start = &this_size;
+	start = stack;
 	top = (unsigned long *)
 		(((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
 
@@ -113,6 +114,7 @@ static void
 stack_trace_call(unsigned long ip, unsigned long parent_ip,
 		 struct ftrace_ops *op, struct pt_regs *pt_regs)
 {
+	unsigned long stack;
 	int cpu;
 
 	preempt_disable_notrace();
@@ -122,7 +124,7 @@ stack_trace_call(unsigned long ip, unsigned long parent_ip,
 	if (per_cpu(trace_active, cpu)++ != 0)
 		goto out;
 
-	check_stack();
+	check_stack(&stack);
 
  out:
 	per_cpu(trace_active, cpu)--;
-- 
1.7.10.4



[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 490 bytes --]

[for-next][PATCH 15/20] tracing: Fix stack tracer with fentry use

[Steven Rostedt]

[-- Attachment #1: Type: text/plain, Size: 4237 bytes --]

From: "Steven Rostedt (Red Hat)" <rostedt@goodmis.org>

When gcc 4.6 on x86 is used, the function tracer will use the new
option -mfentry which does a call to "fentry" at every function
instead of "mcount". The significance of this is that fentry is
called as the first operation of the function instead of the mcount
usage of being called after the stack.

This causes the stack tracer to show some bogus results for the size
of the last function traced, as well as showing "ftrace_call" instead
of the function. This is due to the stack frame not being set up
by the function that is about to be traced.

 # cat stack_trace
        Depth    Size   Location    (48 entries)
        -----    ----   --------
  0)     4824     216   ftrace_call+0x5/0x2f
  1)     4608     112   ____cache_alloc+0xb7/0x22d
  2)     4496      80   kmem_cache_alloc+0x63/0x12f

The 216 size for ftrace_call includes both the ftrace_call stack
(which includes the saving of registers it does), as well as the
stack size of the parent.

To fix this, if CC_USING_FENTRY is defined, then the stack_tracer
will reserve the first item in stack_dump_trace[] array when
calling save_stack_trace(), and it will fill it in with the parent ip.
Then the code will look for the parent pointer on the stack and
give the real size of the parent's stack pointer:

 # cat stack_trace
        Depth    Size   Location    (14 entries)
        -----    ----   --------
  0)     2640      48   update_group_power+0x26/0x187
  1)     2592     224   update_sd_lb_stats+0x2a5/0x4ac
  2)     2368     160   find_busiest_group+0x31/0x1f1
  3)     2208     256   load_balance+0xd9/0x662

I'm Cc'ing stable, although it's not urgent, as it only shows bogus
size for item #0, the rest of the trace is legit. It should still be
corrected in previous stable releases.

Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
---
 kernel/trace/trace_stack.c |   33 +++++++++++++++++++++++++++++----
 1 file changed, 29 insertions(+), 4 deletions(-)

diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c
index dc02e29..ea28e4b 100644
--- a/kernel/trace/trace_stack.c
+++ b/kernel/trace/trace_stack.c
@@ -20,13 +20,27 @@
 
 #define STACK_TRACE_ENTRIES 500
 
+/*
+ * If fentry is used, then the function being traced will
+ * jump to fentry directly before it sets up its stack frame.
+ * We need to ignore that one and record the parent. Since
+ * the stack frame for the traced function wasn't set up yet,
+ * the stack_trace wont see the parent. That needs to be added
+ * manually to stack_dump_trace[] as the first element.
+ */
+#ifdef CC_USING_FENTRY
+# define add_func	1
+#else
+# define add_func	0
+#endif
+
 static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
 	 { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
 static unsigned stack_dump_index[STACK_TRACE_ENTRIES];
 
 static struct stack_trace max_stack_trace = {
-	.max_entries		= STACK_TRACE_ENTRIES,
-	.entries		= stack_dump_trace,
+	.max_entries		= STACK_TRACE_ENTRIES - add_func,
+	.entries		= &stack_dump_trace[add_func],
 };
 
 static unsigned long max_stack_size;
@@ -40,7 +54,7 @@ int stack_tracer_enabled;
 static int last_stack_tracer_enabled;
 
 static inline void
-check_stack(unsigned long *stack)
+check_stack(unsigned long ip, unsigned long *stack)
 {
 	unsigned long this_size, flags;
 	unsigned long *p, *top, *start;
@@ -71,6 +85,17 @@ check_stack(unsigned long *stack)
 	save_stack_trace(&max_stack_trace);
 
 	/*
+	 * When fentry is used, the traced function does not get
+	 * its stack frame set up, and we lose the parent.
+	 * Add that one in manally. We set up save_stack_trace()
+	 * to not touch the first element in this case.
+	 */
+	if (add_func) {
+		stack_dump_trace[0] = ip;
+		max_stack_trace.nr_entries++;
+	}
+
+	/*
 	 * Now find where in the stack these are.
 	 */
 	i = 0;
@@ -124,7 +149,7 @@ stack_trace_call(unsigned long ip, unsigned long parent_ip,
 	if (per_cpu(trace_active, cpu)++ != 0)
 		goto out;
 
-	check_stack(&stack);
+	check_stack(parent_ip, &stack);
 
  out:
 	per_cpu(trace_active, cpu)--;
-- 
1.7.10.4



[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 490 bytes --]

[for-next][PATCH 16/20] tracing: Remove most or all of stack tracer stack size from stack_max_size

[Steven Rostedt]

[-- Attachment #1: Type: text/plain, Size: 5746 bytes --]

From: "Steven Rostedt (Red Hat)" <rostedt@goodmis.org>

Currently, the depth reported in the stack tracer stack_trace file
does not match the stack_max_size file. This is because the stack_max_size
includes the overhead of stack tracer itself while the depth does not.

The first time a max is triggered, a calculation is not performed that
figures out the overhead of the stack tracer and subtracts it from
the stack_max_size variable. The overhead is stored and is subtracted
from the reported stack size for comparing for a new max.

Now the stack_max_size corresponds to the reported depth:

 # cat stack_max_size
4640

 # cat stack_trace
        Depth    Size   Location    (48 entries)
        -----    ----   --------
  0)     4640      32   _raw_spin_lock+0x18/0x24
  1)     4608     112   ____cache_alloc+0xb7/0x22d
  2)     4496      80   kmem_cache_alloc+0x63/0x12f
  3)     4416      16   mempool_alloc_slab+0x15/0x17
[...]

While testing against and older gcc on x86 that uses mcount instead
of fentry, I found that pasing in ip + MCOUNT_INSN_SIZE let the
stack trace show one more function deep which was missing before.

Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
---
 kernel/trace/trace_stack.c |   75 +++++++++++++++++++++++++++++++-------------
 1 file changed, 54 insertions(+), 21 deletions(-)

diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c
index ea28e4b..aab277b 100644
--- a/kernel/trace/trace_stack.c
+++ b/kernel/trace/trace_stack.c
@@ -20,27 +20,24 @@
 
 #define STACK_TRACE_ENTRIES 500
 
-/*
- * If fentry is used, then the function being traced will
- * jump to fentry directly before it sets up its stack frame.
- * We need to ignore that one and record the parent. Since
- * the stack frame for the traced function wasn't set up yet,
- * the stack_trace wont see the parent. That needs to be added
- * manually to stack_dump_trace[] as the first element.
- */
 #ifdef CC_USING_FENTRY
-# define add_func	1
+# define fentry		1
 #else
-# define add_func	0
+# define fentry		0
 #endif
 
 static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
 	 { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
 static unsigned stack_dump_index[STACK_TRACE_ENTRIES];
 
+/*
+ * Reserve one entry for the passed in ip. This will allow
+ * us to remove most or all of the stack size overhead
+ * added by the stack tracer itself.
+ */
 static struct stack_trace max_stack_trace = {
-	.max_entries		= STACK_TRACE_ENTRIES - add_func,
-	.entries		= &stack_dump_trace[add_func],
+	.max_entries		= STACK_TRACE_ENTRIES - 1,
+	.entries		= &stack_dump_trace[1],
 };
 
 static unsigned long max_stack_size;
@@ -58,10 +55,14 @@ check_stack(unsigned long ip, unsigned long *stack)
 {
 	unsigned long this_size, flags;
 	unsigned long *p, *top, *start;
+	static int tracer_frame;
+	int frame_size = ACCESS_ONCE(tracer_frame);
 	int i;
 
 	this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
 	this_size = THREAD_SIZE - this_size;
+	/* Remove the frame of the tracer */
+	this_size -= frame_size;
 
 	if (this_size <= max_stack_size)
 		return;
@@ -73,6 +74,10 @@ check_stack(unsigned long ip, unsigned long *stack)
 	local_irq_save(flags);
 	arch_spin_lock(&max_stack_lock);
 
+	/* In case another CPU set the tracer_frame on us */
+	if (unlikely(!frame_size))
+		this_size -= tracer_frame;
+
 	/* a race could have already updated it */
 	if (this_size <= max_stack_size)
 		goto out;
@@ -85,15 +90,12 @@ check_stack(unsigned long ip, unsigned long *stack)
 	save_stack_trace(&max_stack_trace);
 
 	/*
-	 * When fentry is used, the traced function does not get
-	 * its stack frame set up, and we lose the parent.
-	 * Add that one in manally. We set up save_stack_trace()
-	 * to not touch the first element in this case.
+	 * Add the passed in ip from the function tracer.
+	 * Searching for this on the stack will skip over
+	 * most of the overhead from the stack tracer itself.
 	 */
-	if (add_func) {
-		stack_dump_trace[0] = ip;
-		max_stack_trace.nr_entries++;
-	}
+	stack_dump_trace[0] = ip;
+	max_stack_trace.nr_entries++;
 
 	/*
 	 * Now find where in the stack these are.
@@ -123,6 +125,18 @@ check_stack(unsigned long ip, unsigned long *stack)
 				found = 1;
 				/* Start the search from here */
 				start = p + 1;
+				/*
+				 * We do not want to show the overhead
+				 * of the stack tracer stack in the
+				 * max stack. If we haven't figured
+				 * out what that is, then figure it out
+				 * now.
+				 */
+				if (unlikely(!tracer_frame) && i == 1) {
+					tracer_frame = (p - stack) *
+						sizeof(unsigned long);
+					max_stack_size -= tracer_frame;
+				}
 			}
 		}
 
@@ -149,7 +163,26 @@ stack_trace_call(unsigned long ip, unsigned long parent_ip,
 	if (per_cpu(trace_active, cpu)++ != 0)
 		goto out;
 
-	check_stack(parent_ip, &stack);
+	/*
+	 * When fentry is used, the traced function does not get
+	 * its stack frame set up, and we lose the parent.
+	 * The ip is pretty useless because the function tracer
+	 * was called before that function set up its stack frame.
+	 * In this case, we use the parent ip.
+	 *
+	 * By adding the return address of either the parent ip
+	 * or the current ip we can disregard most of the stack usage
+	 * caused by the stack tracer itself.
+	 *
+	 * The function tracer always reports the address of where the
+	 * mcount call was, but the stack will hold the return address.
+	 */
+	if (fentry)
+		ip = parent_ip;
+	else
+		ip += MCOUNT_INSN_SIZE;
+
+	check_stack(ip, &stack);
 
  out:
 	per_cpu(trace_active, cpu)--;
-- 
1.7.10.4



[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 490 bytes --]