The function graph tracer is currently the most invasive tracer
in the ftrace family. It can easily overflow the buffer even with
10megs per CPU. This means that events can often be lost.
On start up, or after events are lost, if the function return is
recorded but the function enter was lost, all we get to see is the
exiting '}'.
Here is how a typical trace output starts:
[tracing] cat trace
# tracer: function_graph
#
# CPU DURATION FUNCTION CALLS
# | | | | | | |
0) + 91.897 us | }
0) ! 567.961 us | }
0) <========== |
0) ! 579.083 us | _raw_spin_lock_irqsave();
0) 4.694 us | _raw_spin_unlock_irqrestore();
0) ! 594.862 us | }
0) ! 603.361 us | }
0) ! 613.574 us | }
0) ! 623.554 us | }
0) 3.653 us | fget_light();
0) | sock_poll() {
There are a series of '}' with no matching "func() {". There's no information
to what functions these ending brackets belong to.
This patch adds a stack on the per cpu structure used in outputting
the function graph tracer to keep track of what function was outputted.
Then on a function exit event, it checks the depth to see if the
function exit has a matching entry event. If it does, then it only
prints the '}', otherwise it adds the function name after the '}'.
This allows function exit events to show what function they belong to
at trace output startup, when the entry was lost due to ring buffer
overflow, or even after a new task is scheduled in.
Here is what the above trace will look like after this patch:
[tracing] cat trace
# tracer: function_graph
#
# CPU DURATION FUNCTION CALLS
# | | | | | | |
0) + 91.897 us | } (irq_exit)
0) ! 567.961 us | } (smp_apic_timer_interrupt)
0) <========== |
0) ! 579.083 us | _raw_spin_lock_irqsave();
0) 4.694 us | _raw_spin_unlock_irqrestore();
0) ! 594.862 us | } (add_wait_queue)
0) ! 603.361 us | } (__pollwait)
0) ! 613.574 us | } (tcp_poll)
0) ! 623.554 us | } (sock_poll)
0) 3.653 us | fget_light();
0) | sock_poll() {
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
In the function graph tracer, a calling function is to be traced
only when it is enabled through the set_graph_function file,
or when it is nested in an enabled function.
Current code uses TSK_TRACE_FL_GRAPH to test whether it is nested
or not. Looking at the code, we can get this:
(trace->depth > 0) <==> (TSK_TRACE_FL_GRAPH is set)
trace->depth is more explicit to tell that it is nested.
So we use trace->depth directly and simplify the code.
No functionality is changed.
TSK_TRACE_FL_GRAPH is not removed yet, it is left for future usage.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <4B4DB0B6.7040607@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Each time we save a function entry from the function graph
tracer, we check if the trace array is set, which is wasteful
because it is set anyway before we start the tracer. All we need
is to ensure we have good read and write orderings. When we set
the trace array, we just need to guarantee it to be visible
before starting tracing.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
LKML-Reference: <1263453795-7496-1-git-send-regression-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu: (34 commits)
m68k: rename global variable vmalloc_end to m68k_vmalloc_end
percpu: add missing per_cpu_ptr_to_phys() definition for UP
percpu: Fix kdump failure if booted with percpu_alloc=page
percpu: make misc percpu symbols unique
percpu: make percpu symbols in ia64 unique
percpu: make percpu symbols in powerpc unique
percpu: make percpu symbols in x86 unique
percpu: make percpu symbols in xen unique
percpu: make percpu symbols in cpufreq unique
percpu: make percpu symbols in oprofile unique
percpu: make percpu symbols in tracer unique
percpu: make percpu symbols under kernel/ and mm/ unique
percpu: remove some sparse warnings
percpu: make alloc_percpu() handle array types
vmalloc: fix use of non-existent percpu variable in put_cpu_var()
this_cpu: Use this_cpu_xx in trace_functions_graph.c
this_cpu: Use this_cpu_xx for ftrace
this_cpu: Use this_cpu_xx in nmi handling
this_cpu: Use this_cpu operations in RCU
this_cpu: Use this_cpu ops for VM statistics
...
Fix up trivial (famous last words) global per-cpu naming conflicts in
arch/x86/kvm/svm.c
mm/slab.c
There is a case where the graph tracer might get confused and omits
displaying of a single record. This applies mostly with the trace_pipe
since it is unlikely that the trace_seq buffer will overflow with the
trace file.
As the function_graph tracer goes through the trace entries keeping a
pointer to the current record:
current -> func1 ENTRY
func2 ENTRY
func2 RETURN
func1 RETURN
When an function ENTRY is encountered, it moves the pointer to the
next entry to check if the function is a nested or leaf function.
func1 ENTRY
current -> func2 ENTRY
func2 RETURN
func1 RETURN
If the rest of the writing of the function fills the trace_seq buffer,
then the trace_pipe read will ignore this entry. The next read will
Now start at the current location, but the first entry (func1) will
be discarded.
This patch keeps a copy of the current entry in the iterator private
storage and will keep track of when the trace_seq buffer fills. When
the trace_seq buffer fills, it will reuse the copy of the entry in the
next iteration.
[
This patch has been largely modified by Steven Rostedt in order to
clean it up and simplify it. The original idea and concept was from
Jirka and for that, this patch will go under his name to give him
the credit he deserves. But because this was modify by Steven Rostedt
anything wrong with the patch should be blamed on Steven.
]
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1259067458-27143-1-git-send-email-jolsa@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
ftrace_cpu_disabled usage in trace_functions_graph.c were left out
during this_cpu_xx conversion in commit 9288f99a causing compile
failure. Convert them.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Christoph Lameter <cl@linux-foundation.org>
For direct function pointers (like what mcount provides) PowerPC64
requires the use of %ps, otherwise nothing is printed.
This patch converts all prints of functions retrieved through mcount
to use the %ps format from the %pf.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Both trace_output.c and trace_function_graph.c do basically the same
thing to handle the printing of the latency-format. This patch moves
the code into one function that both can use.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
While debugging something with the function_graph tracer, I found the
need to see the preempt count of the traces. Unfortunately, since
the function graph tracer has its own output formatting, it does not
honor the latency-format option.
This patch makes the function_graph tracer honor the latency-format
option, but still keeps control of the output. But now we have the
same details that the latency-format supplies.
# tracer: function_graph
#
# _-----=> irqs-off
# / _----=> need-resched
# | / _---=> hardirq/softirq
# || / _--=> preempt-depth
# ||| /
# ||||
# CPU|||| DURATION FUNCTION CALLS
# | |||| | | | | | |
3) d..1 1.333 us | idle_cpu();
3) d.h1 | tick_check_idle() {
3) d.h1 0.550 us | tick_check_oneshot_broadcast();
3) d.h1 | tick_nohz_stop_idle() {
3) d.h1 | ktime_get() {
3) d.h1 | ktime_get_ts() {
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The latency tracers (irqsoff and wakeup) can swap trace buffers
on the fly. If an event is happening and has reserved data on one of
the buffers, and the latency tracer swaps the global buffer with the
max buffer, the result is that the event may commit the data to the
wrong buffer.
This patch changes the API to the trace recording to be recieve the
buffer that was used to reserve a commit. Then this buffer can be passed
in to the commit.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The function graph events helpers which insert the function entry and
return events into the ring buffer currently reside in trace.c
But this file is quite overloaded and the right place for these helpers
is in the function graph tracer file.
Then move them to trace_functions_graph.c
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
print_graph_cpu() is little over-designed.
And "log10_all" may be wrong when there are holes in cpu_online_mask:
the max online cpu id > cpumask_weight(cpu_online_mask)
So change it by using a static column length for the cpu matching
nr_cpu_ids number of decimal characters.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <4A6EEE5E.2000001@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
When print_graph_entry() computes a function call entry event, it needs
to also check the next entry to guess if it matches the return event of
the current function entry.
In order to look at this next event, it needs to consume the current
entry before going ahead in the ring buffer.
However, if the current event that gets consumed is the last one in the
ring buffer head page, the ring_buffer may reuse the page for writers.
The consumed entry will then become invalid because of possible
racy overwriting.
Me must then handle this entry by making a copy of it.
The fix also applies on 2.6.30
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: stable@kernel.org
LKML-Reference: <4A6EEAEC.3050508@cn.fujitsu.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
In case gcc does something funny with the stack frames, or the return
from function code, we would like to detect that.
An arch may implement passing of a variable that is unique to the
function and can be saved on entering a function and can be tested
when exiting the function. Usually the frame pointer can be used for
this purpose.
This patch also implements this for x86. Where it passes in the stack
frame of the parent function, and will test that frame on exit.
There was a case in x86_32 with optimize for size (-Os) where, for a
few functions, gcc would align the stack frame and place a copy of the
return address into it. The function graph tracer modified the copy and
not the actual return address. On return from the funtion, it did not go
to the tracer hook, but returned to the parent. This broke the function
graph tracer, because the return of the parent (where gcc did not do
this funky manipulation) returned to the location that the child function
was suppose to. This caused strange kernel crashes.
This test detected the problem and pointed out where the issue was.
This modifies the parameters of one of the functions that the arch
specific code calls, so it includes changes to arch code to accommodate
the new prototype.
Note, I notice that the parsic arch implements its own push_return_trace.
This is now a generic function and the ftrace_push_return_trace should be
used instead. This patch does not touch that code.
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
* 'tracing-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
function-graph: always initialize task ret_stack
function-graph: move initialization of new tasks up in fork
function-graph: add memory barriers for accessing task's ret_stack
function-graph: enable the stack after initialization of other variables
function-graph: only allocate init tasks if it was not already done
Manually fix trivial conflict in kernel/trace/ftrace.c
The function graph tracer checks if the task_struct has ret_stack defined
to know if it is OK or not to use it. The initialization is done for
all tasks by one process, but the idle tasks use the same initialization
used by new tasks.
If an interrupt happens on an idle task that just had the ret_stack
created, but before the rest of the initialization took place, then
we can corrupt the return address of the functions.
This patch moves the setting of the task_struct's ret_stack to after
the other variables have been initialized.
[ Impact: prevent kernel panic on idle task when starting function graph ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
graph time is the time that a function is executing another function.
Thus if function A calls B, if graph-time is set, then the time for
A includes B. This is the default behavior. But if graph-time is off,
then the time spent executing B is subtracted from A.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
If the function graph trace is enabled, the function profiler will
use it to take the timing of the functions.
cat /debug/tracing/trace_stat/functions
Function Hit Time
-------- --- ----
mwait_idle 127 183028.4 us
schedule 26 151997.7 us
__schedule 31 151975.1 us
sys_wait4 2 74080.53 us
do_wait 2 74077.80 us
sys_newlstat 138 39929.16 us
do_path_lookup 179 39845.79 us
vfs_lstat_fd 138 39761.97 us
user_path_at 153 39469.58 us
path_walk 179 39435.76 us
__link_path_walk 189 39143.73 us
[...]
Note the times are skewed due to the function graph tracer not taking
into account schedules.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
This patch move the timestamp from happening in the arch specific
code into the general code. This allows for better control by the tracer
to time manipulation.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
With the added TRACE_EVENT macro, the events no longer appear in
the function graph tracer. This was because the function graph
did not know how to display the entries. The graph tracer was
only aware of its own entries and the printk entries.
By using the event call back feature, the graph tracer can now display
the events.
# echo irq > /debug/tracing/set_event
Which can show:
0) | handle_IRQ_event() {
0) | /* irq_handler_entry: irq=48 handler=eth0 */
0) | e1000_intr() {
0) 0.926 us | __napi_schedule();
0) 3.888 us | }
0) | /* irq_handler_exit: irq=48 return=handled */
0) 0.655 us | runqueue_is_locked();
0) | __wake_up() {
0) 0.831 us | _spin_lock_irqsave();
The irq entry and exit events show up as comments.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Currently, the function graph tracer depends on the trace_printk
to record the depth. All the information is already there in the trace
to calculate function depth, with the exception of having the printk
be the first item. But as soon as a entry or exit is reached, then
we know the depth.
This patch changes the iter->private data from recording a per cpu
last_pid, to a structure that holds both the last_pid and the current
depth. This data is used to determine the function depth for the
printks.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
