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>
The "cpu" variable is declared at the start of the function and
also within a branch, with the exact same initialization.
Remove the local variable of the same name in the branch.
Signed-off-by: Wenji Huang <wenji.huang@oracle.com>
LKML-Reference: <1266997226-6833-3-git-send-email-wenji.huang@oracle.com>
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>
Now that the return from alloc_percpu is compatible with the address
of per-cpu vars, it makes sense to hand around the address of per-cpu
variables. To make this sane, we remove the per_cpu__ prefix we used
created to stop people accidentally using these vars directly.
Now we have sparse, we can use that (next patch).
tj: * Updated to convert stuff which were missed by or added after the
original patch.
* Kill per_cpu_var() macro.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Christoph Lameter <cl@linux-foundation.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>
Remove the obsolete seq_print_ip_sym() usage and replace it
by the %pf format in order to print function symbols.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <1247107590-6428-2-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
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>
Impact: clean up
The prologue of the function graph entry, return and comments all
start out pretty much the same. Each of these duplicate code and
do so slightly differently.
This patch consolidates the printing of the pid, absolute time,
cpu and proc (and for entry, the interrupt).
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Impact: fix to one cause of incorrect comm outputs in trace
The spinlock only protected the creation of a comm <=> pid pair.
But it was possible that a reader could look up a pid, and get the
wrong comm because it had no locking.
This also required changing trace_find_cmdline to copy the comm cache
and not just send back a pointer to it.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Impact: fix callsites with dynamic format strings
Since its new binary implementation, trace_printk() internally uses static
containers for the format strings on each callsites. But the value is
assigned once at build time, which means that it can't take dynamic
formats.
So this patch unearthes the raw trace_printk implementation for the callers
that will need trace_printk to be able to carry these dynamic format
strings. The trace_printk() macro will use the appropriate implementation
for each callsite. Most of the time however, the binary implementation will
still be used.
The other impact of this patch is that mmiotrace_printk() will use the old
implementation because it calls the low level trace_vprintk and we can't
guess here whether the format passed in it is dynamic or not.
Some parts of this patch have been written by Steven Rostedt (most notably
the part that chooses the appropriate implementation for each callsites).
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Impact: clean up
There existed a lot of <space><tab>'s in the tracing code. This
patch removes them.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Impact: faster and lighter tracing
Now that we have trace_bprintk() which is faster and consume lesser
memory than trace_printk() and has the same purpose, we can now drop
the old implementation in favour of the binary one from trace_bprintk(),
which means we move all the implementation of trace_bprintk() to
trace_printk(), so the Api doesn't change except that we must now use
trace_seq_bprintk() to print the TRACE_PRINT entries.
Some changes result of this:
- Previously, trace_bprintk depended of a single tracer and couldn't
work without. This tracer has been dropped and the whole implementation
of trace_printk() (like the module formats management) is now integrated
in the tracing core (comes with CONFIG_TRACING), though we keep the file
trace_printk (previously trace_bprintk.c) where we can find the module
management. Thus we don't overflow trace.c
- changes some parts to use trace_seq_bprintk() to print TRACE_PRINT entries.
- change a bit trace_printk/trace_vprintk macros to support non-builtin formats
constants, and fix 'const' qualifiers warnings. But this is all transparent for
developers.
- etc...
V2:
- Rebase against last changes
- Fix mispell on the changelog
V3:
- Rebase against last changes (moving trace_printk() to kernel.h)
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1236356510-8381-5-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: decrease hangs risks with the graph tracer on slow systems
Since the function graph tracer can spend too much time on timer
interrupts, it's better now to use the more lightweight local
clock. Anyway, the function graph traces are more reliable on a
per cpu trace.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <49af243d.06e9300a.53ad.ffff840c@mx.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
There is nothing really arch specific of the push and pop functions
used by the function graph tracer. This patch moves them to generic
code.
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Impact: api and pipe waiting change
Currently, the waiting used in tracing_read_pipe() is done through a
100 msecs schedule_timeout() loop which periodically check if there
are traces on the buffer.
This can cause small latencies for programs which are reading the incoming
events.
This patch makes the reader waiting for the trace_wait waitqueue except
for few tracers such as the sched and functions tracers which might be
already hold the runqueue lock while waking up the reader.
This is performed through a new callback wait_pipe() on struct tracer.
If none is implemented on a specific tracer, the default waiting for
trace_wait queue is attached.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch is to make the function return early on failure, and give
correct return value on success.
Signed-off-by: Wenji Huang <wenji.huang@oracle.com>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
When one cats the trace file, the leaf functions are printed without brackets:
function();
whereas in the trace_pipe file we'll see the following:
function() {
}
This is because the ring_buffer handling is not the same between those two files.
On the trace file, when an entry is printed, the iterator advanced and then we can
check the next entry.
There is no iterator with trace_pipe, the current entry to print has been peeked
and not consumed. So checking the next entry will still return the current one while
we don't consume it.
This patch introduces a new value for the output callbacks to ask the tracing
core to not consume the current entry after printing it.
We need it because we will have to consume the current entry ourself to check
the next one.
Now the trace_pipe is able to handle well the leaf functions.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Making it more easy to do a basic regression test for this tracer.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
To make it easy for ftrace plugin writers, as this was open coded in
the existing plugins
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Frédéric Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch brings various bugfixes:
- Drop the first irrelevant task switch on the very beginning of a trace.
- Drop the OVERHEAD word from the headers, the DURATION word is sufficient
and will not overlap other columns.
- Make the headers fit well their respective columns whatever the
selected options.
Ie, default options:
# tracer: function_graph
#
# CPU DURATION FUNCTION CALLS
# | | | | | | |
1) 0.646 us | }
1) | mem_cgroup_del_lru_list() {
1) 0.624 us | lookup_page_cgroup();
1) 1.970 us | }
echo funcgraph-proc > trace_options
# tracer: function_graph
#
# CPU TASK/PID DURATION FUNCTION CALLS
# | | | | | | | | |
0) bash-2937 | 0.895 us | }
0) bash-2937 | 0.888 us | __rcu_read_unlock();
0) bash-2937 | 0.864 us | conv_uni_to_pc();
0) bash-2937 | 1.015 us | __rcu_read_lock();
echo nofuncgraph-cpu > trace_options
echo nofuncgraph-proc > trace_options
# tracer: function_graph
#
# DURATION FUNCTION CALLS
# | | | | | |
3.752 us | native_pud_val();
0.616 us | native_pud_val();
0.624 us | native_pmd_val();
About features, one can now disable the duration (this will hide the
overhead too for convenient reasons and because on doesn't need
overhead if it hasn't the duration):
echo nofuncgraph-duration > trace_options
# tracer: function_graph
#
# FUNCTION CALLS
# | | | |
cap_vm_enough_memory() {
__vm_enough_memory() {
vm_acct_memory();
}
}
}
And at last, an option to print the absolute time:
//Restart from default options
echo funcgraph-abstime > trace_options
# tracer: function_graph
#
# TIME CPU DURATION FUNCTION CALLS
# | | | | | | | |
261.339774 | 1) + 42.823 us | }
261.339775 | 1) 1.045 us | _spin_lock_irq();
261.339777 | 1) 0.940 us | _spin_lock_irqsave();
261.339778 | 1) 0.752 us | _spin_unlock_irqrestore();
261.339780 | 1) 0.857 us | _spin_unlock_irq();
261.339782 | 1) | flush_to_ldisc() {
261.339783 | 1) | tty_ldisc_ref() {
261.339783 | 1) | tty_ldisc_try() {
261.339784 | 1) 1.075 us | _spin_lock_irqsave();
261.339786 | 1) 0.842 us | _spin_unlock_irqrestore();
261.339788 | 1) 4.211 us | }
261.339788 | 1) 5.662 us | }
The format is seconds.usecs.
I guess no one needs the nanosec precision here, the main goal is to have
an overview about the general timings of events, and to see the place when
the trace switches from one cpu to another.
ie:
274.874760 | 1) 0.676 us | _spin_unlock();
274.874762 | 1) 0.609 us | native_load_sp0();
274.874763 | 1) 0.602 us | native_load_tls();
274.878739 | 0) 0.722 us | }
274.878740 | 0) 0.714 us | native_pmd_val();
274.878741 | 0) 0.730 us | native_pmd_val();
Here there is a 4000 usecs difference when we switch the cpu.
Changes in V2:
- Completely fix the first pointless task switch.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: Reduce future memory usage, use new cpumask API.
Since the last patch was created and acked, more old cpumask users
slipped into kernel/trace.
Mostly trivial conversions, except struct trace_iterator's "started"
member becomes a cpumask_var_t.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Impact: simplify/generalize/refactor trace.c
The trace.c file is becoming more difficult to maintain due to the
growing number of events. There is several formats that an event may
be printed. This patch sets up the infrastructure of an event hash to
allow for events to register how they should be printed.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup, remove obsolete code
Now that the ring buffer used by ftrace allows for variable length
entries, we do not need the 'cont' feature of the buffer. This code
makes other parts of ftrace more complex and by removing this it
simplifies the ftrace code.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: tracer output improvement
Ending newlines are appended automatically on comments by the function
graph tracer because the newline needs to be placed after the "*/"
comment characters.
So if the user puts an ending newline, we want to strip it.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: provide trace headers to explain a bit the output
This patch implements the print_headers callback for the function graph
tracer. These headers are output according to the current trace options.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Handle the TRACE_PRINT entries from the function grapg tracer
and output them as a C comment just below the function that called
it, as if it was a comment inside this function.
Example with an ftrace_printk inside might_sleep() function:
void __might_sleep(char *file, int line)
{
static unsigned long prev_jiffy; /* ratelimiting */
ftrace_printk("Hi I'm a comment in might_sleep() :-)");
A chunk of a resulting trace:
0) | _reiserfs_free_block() {
0) | reiserfs_read_bitmap_block() {
0) | __bread() {
0) | __getblk() {
0) | __find_get_block() {
0) 0.698 us | mark_page_accessed();
0) 2.267 us | }
0) | __might_sleep() {
0) | /* Hi I'm a comment in might_sleep() :-) */
0) 1.321 us | }
0) 5.872 us | }
0) 7.313 us | }
0) 8.718 us | }
And this patch brings two minor fixes:
- The newline after a switch-out task has disappeared
- The "|" sign just before the cpu number on task-switch has been deleted.
0) 0.616 us | pick_next_task_rt();
0) 1.457 us | _spin_trylock();
0) 0.653 us | _spin_unlock();
0) 0.728 us | _spin_trylock();
0) 0.631 us | _spin_unlock();
0) 0.729 us | native_load_sp0();
0) 0.593 us | native_load_tls();
------------------------------------------
0) cat-2834 => migrati-3
------------------------------------------
0) | finish_task_switch() {
0) 0.841 us | _spin_unlock_irq();
0) 0.616 us | post_schedule_rt();
0) 3.882 us | }
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: better trace output of duration for long calls
The old duration output didn't exceeded 9999.999 us to fit the column
and the nanosecs were always 3 numbers. As Ingo suggested, it's better
to have the whole microseconds elapsed time and shift the nanosecs precision
if needed to fit the maximum 7 numbers. And usec need more number, the case
should be rare and important enough to break a bit the column alignment to
show it.
So, depending of the duration value, we now have these patterns:
u.nnn us
uu.nnn us
uuu.nnn us
uuuu.nnn us
uuuuu.nn us
uuuuuu.n us
uuuuuuuu..... us
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: extend function-graph output: let one know which thread called a function
This patch implements a helper function to print the couple cmdline/pid.
Its output is provided during task switching and on each row if the new
"funcgraph-proc" defualt-off option is set through trace_options file.
The output is center aligned and never exceeds 14 characters. The cmdline
is truncated over 7 chars.
But note that if the pid exceeds 6 characters, the column will overflow (but
the situation is abnormal).
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: increase the visual qualities of the call-graph-tracer output
This patch applies various trace output formatting changes:
- CPU is now a decimal number, followed by a parenthesis.
- Overhead is now on the second column (gives a good visibility)
- Cost is now on the third column, can't exceed 9999.99 us. It is
followed by a virtual line based on a "|" character.
- Functions calls are now the last column on the right. This way, we
haven't dynamic column (which flow is harder to follow) on its right.
- CPU and Overhead have their own option flag. They are default-on but you
can disable them easily:
echo nofuncgraph-cpu > trace_options
echo nofuncgraph-overhead > trace_options
TODO:
_ Refactoring of the thread switch output.
_ Give a default-off option to output the thread and its pid on each row.
_ Provide headers
_ ....
Here is an example of the new trace style:
0) | mutex_unlock() {
0) 0.639 us | __mutex_unlock_slowpath();
0) 1.607 us | }
0) | remove_wait_queue() {
0) 0.616 us | _spin_lock_irqsave();
0) 0.616 us | _spin_unlock_irqrestore();
0) 2.779 us | }
0) 0.495 us | n_tty_set_room();
0) ! 9999.999 us | }
0) | tty_ldisc_deref() {
0) 0.615 us | _spin_lock_irqsave();
0) 0.616 us | _spin_unlock_irqrestore();
0) 2.793 us | }
0) | current_fs_time() {
0) 0.488 us | current_kernel_time();
0) 0.495 us | timespec_trunc();
0) 2.486 us | }
0) ! 9999.999 us | }
0) ! 9999.999 us | }
0) ! 9999.999 us | }
0) | sys_read() {
0) 0.796 us | fget_light();
0) | vfs_read() {
0) | rw_verify_area() {
0) | security_file_permission() {
0) 0.488 us | cap_file_permission();
0) 1.720 us | }
0) 3. 4 us | }
0) | tty_read() {
0) 0.488 us | tty_paranoia_check();
0) | tty_ldisc_ref_wait() {
0) | tty_ldisc_try() {
0) 0.615 us | _spin_lock_irqsave();
0) 0.615 us | _spin_unlock_irqrestore();
0) 5.436 us | }
0) 6.427 us | }
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: enhance the output of the graph-tracer
This patch applies some ideas of Ingo Molnar and Steven Rostedt.
* Output leaf functions in one line with parenthesis, semicolon and duration
output.
* Add a second column (after cpu) for an overhead sign.
if duration > 100 us, "!"
if duration > 10 us, "+"
else " "
* Print output in us with remaining nanosec: u.n
* Print duration on the right end, following the indentation of the functions.
Use also visual clues: "-" on entry call (no duration to output) and "+" on
return (duration output).
The name of the tracer has been fixed as well: function-branch becomes
function_branch.
Here is an example of the new output:
CPU[000] dequeue_entity() { -
CPU[000] update_curr() { -
CPU[000] update_min_vruntime(); + 0.512 us
CPU[000] } + 1.504 us
CPU[000] clear_buddies(); + 0.481 us
CPU[000] update_min_vruntime(); + 0.504 us
CPU[000] } + 4.557 us
CPU[000] hrtick_update() { -
CPU[000] hrtick_start_fair(); + 0.489 us
CPU[000] } + 1.443 us
CPU[000] + } + 14.655 us
CPU[000] + } + 15.678 us
CPU[000] + } + 16.686 us
CPU[000] msecs_to_jiffies(); + 0.481 us
CPU[000] put_prev_task_fair(); + 0.504 us
CPU[000] pick_next_task_fair(); + 0.482 us
CPU[000] pick_next_task_rt(); + 0.504 us
CPU[000] pick_next_task_fair(); + 0.481 us
CPU[000] pick_next_task_idle(); + 0.489 us
CPU[000] _spin_trylock(); + 0.655 us
CPU[000] _spin_unlock(); + 0.609 us
CPU[000] ------------8<---------- thread bash-2794 ------------8<----------
CPU[000] finish_task_switch() { -
CPU[000] _spin_unlock_irq(); + 0.722 us
CPU[000] } + 2.369 us
CPU[000] ! } + 501972.605 us
CPU[000] ! } + 501973.763 us
CPU[000] copy_from_read_buf() { -
CPU[000] _spin_lock_irqsave(); + 0.670 us
CPU[000] _spin_unlock_irqrestore(); + 0.699 us
CPU[000] copy_to_user() { -
CPU[000] might_fault() { -
CPU[000] __might_sleep(); + 0.503 us
CPU[000] } + 1.632 us
CPU[000] __copy_to_user_ll(); + 0.542 us
CPU[000] } + 3.858 us
CPU[000] tty_audit_add_data() { -
CPU[000] _spin_lock_irq(); + 0.609 us
CPU[000] _spin_unlock_irq(); + 0.624 us
CPU[000] } + 3.196 us
CPU[000] _spin_lock_irqsave(); + 0.624 us
CPU[000] _spin_unlock_irqrestore(); + 0.625 us
CPU[000] + } + 13.611 us
CPU[000] copy_from_read_buf() { -
CPU[000] _spin_lock_irqsave(); + 0.624 us
CPU[000] _spin_unlock_irqrestore(); + 0.616 us
CPU[000] } + 2.820 us
CPU[000]
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: enhancement for function graph tracer
When run on a SMP box, the function graph tracer is confusing because
it shows the different CPUS as changes in the trace.
This patch adds the annotation of 'CPU[###]' where ### is a three digit
number. The output will look similar to this:
CPU[001] dput() {
CPU[000] } 726
CPU[001] } 487
CPU[000] do_softirq() {
CPU[001] } 2221
CPU[000] __do_softirq() {
CPU[000] __local_bh_disable() {
CPU[001] unroll_tree_refs() {
CPU[000] } 569
CPU[001] } 501
CPU[000] rcu_process_callbacks() {
CPU[001] kfree() {
What makes this nice is that now you can grep the file and produce
readable format for a particular CPU.
# cat /debug/tracing/trace > /tmp/trace
# grep '^CPU\[000\]' /tmp/trace > /tmp/trace0
# grep '^CPU\[001\]' /tmp/trace > /tmp/trace1
Will give you:
# head /tmp/trace0
CPU[000] ------------8<---------- thread sshd-3899 ------------8<----------
CPU[000] inotify_dentry_parent_queue_event() {
CPU[000] } 2531
CPU[000] inotify_inode_queue_event() {
CPU[000] } 505
CPU[000] } 69626
CPU[000] } 73089
CPU[000] audit_syscall_exit() {
CPU[000] path_put() {
CPU[000] dput() {
# head /tmp/trace1
CPU[001] ------------8<---------- thread pcscd-3446 ------------8<----------
CPU[001] } 4186
CPU[001] dput() {
CPU[001] } 543
CPU[001] vfs_permission() {
CPU[001] inode_permission() {
CPU[001] shmem_permission() {
CPU[001] generic_permission() {
CPU[001] } 501
CPU[001] } 2205
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: enhancement to function graph tracer
Export the trace_find_cmdline so the function graph tracer can
use it to print the comms of the threads.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: feature
This patch sets a C-like output for the function graph tracing.
For this aim, we now call two handler for each function: one on the entry
and one other on return. This way we can draw a well-ordered call stack.
The pid of the previous trace is loosely stored to be compared against
the one of the current trace to see if there were a context switch.
Without this little feature, the call tree would seem broken at
some locations.
We could use the sched_tracer to capture these sched_events but this
way of processing is much more simpler.
2 spaces have been chosen for indentation to fit the screen while deep
calls. The time of execution in nanosecs is printed just after closed
braces, it seems more easy this way to find the corresponding function.
If the time was printed as a first column, it would be not so easy to
find the corresponding function if it is called on a deep depth.
I plan to output the return value but on 32 bits CPU, the return value
can be 32 or 64, and its difficult to guess on which case we are.
I don't know what would be the better solution on X86-32: only print
eax (low-part) or even edx (high-part).
Actually it's thee same problem when a function return a 8 bits value, the
high part of eax could contain junk values...
Here is an example of trace:
sys_read() {
fget_light() {
} 526
vfs_read() {
rw_verify_area() {
security_file_permission() {
cap_file_permission() {
} 519
} 1564
} 2640
do_sync_read() {
pipe_read() {
__might_sleep() {
} 511
pipe_wait() {
prepare_to_wait() {
} 760
deactivate_task() {
dequeue_task() {
dequeue_task_fair() {
dequeue_entity() {
update_curr() {
update_min_vruntime() {
} 504
} 1587
clear_buddies() {
} 512
add_cfs_task_weight() {
} 519
update_min_vruntime() {
} 511
} 5602
dequeue_entity() {
update_curr() {
update_min_vruntime() {
} 496
} 1631
clear_buddies() {
} 496
update_min_vruntime() {
} 527
} 4580
hrtick_update() {
hrtick_start_fair() {
} 488
} 1489
} 13700
} 14949
} 16016
msecs_to_jiffies() {
} 496
put_prev_task_fair() {
} 504
pick_next_task_fair() {
} 489
pick_next_task_rt() {
} 496
pick_next_task_fair() {
} 489
pick_next_task_idle() {
} 489
------------8<---------- thread 4 ------------8<----------
finish_task_switch() {
} 1203
do_softirq() {
__do_softirq() {
__local_bh_disable() {
} 669
rcu_process_callbacks() {
__rcu_process_callbacks() {
cpu_quiet() {
rcu_start_batch() {
} 503
} 1647
} 3128
__rcu_process_callbacks() {
} 542
} 5362
_local_bh_enable() {
} 587
} 8880
} 9986
kthread_should_stop() {
} 669
deactivate_task() {
dequeue_task() {
dequeue_task_fair() {
dequeue_entity() {
update_curr() {
calc_delta_mine() {
} 511
update_min_vruntime() {
} 511
} 2813
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
This patch changes the name of the "return function tracer" into
function-graph-tracer which is a more suitable name for a tracing
which makes one able to retrieve the ordered call stack during
the code flow.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Linus Torvalds