2
0
mirror of https://github.com/edk2-porting/linux-next.git synced 2024-12-23 12:43:55 +08:00
linux-next/kernel/trace/trace_functions_graph.c

326 lines
7.5 KiB
C
Raw Normal View History

/*
*
* Function graph tracer.
* Copyright (c) 2008 Frederic Weisbecker <fweisbec@gmail.com>
* Mostly borrowed from function tracer which
* is Copyright (c) Steven Rostedt <srostedt@redhat.com>
*
*/
#include <linux/debugfs.h>
#include <linux/uaccess.h>
#include <linux/ftrace.h>
#include <linux/fs.h>
#include "trace.h"
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
#define TRACE_GRAPH_INDENT 2
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
/* Spaces between function call and time duration */
#define TRACE_GRAPH_TIMESPACE_ENTRY " "
/* Spaces between function call and closing braces */
#define TRACE_GRAPH_TIMESPACE_RET " "
#define TRACE_GRAPH_PRINT_OVERRUN 0x1
static struct tracer_opt trace_opts[] = {
/* Display overruns or not */
{ TRACER_OPT(overrun, TRACE_GRAPH_PRINT_OVERRUN) },
{ } /* Empty entry */
};
static struct tracer_flags tracer_flags = {
.val = 0, /* Don't display overruns by default */
.opts = trace_opts
};
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
/* pid on the last trace processed */
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
static pid_t last_pid[NR_CPUS] = { [0 ... NR_CPUS-1] = -1 };
static int graph_trace_init(struct trace_array *tr)
{
int cpu, ret;
for_each_online_cpu(cpu)
tracing_reset(tr, cpu);
ret = register_ftrace_graph(&trace_graph_return,
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
&trace_graph_entry);
if (ret)
return ret;
tracing_start_cmdline_record();
return 0;
}
static void graph_trace_reset(struct trace_array *tr)
{
tracing_stop_cmdline_record();
unregister_ftrace_graph();
}
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
/* If the pid changed since the last trace, output this event */
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
static int verif_pid(struct trace_seq *s, pid_t pid, int cpu)
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
{
char *comm;
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
if (last_pid[cpu] != -1 && last_pid[cpu] == pid)
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
return 1;
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
last_pid[cpu] = pid;
comm = trace_find_cmdline(pid);
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
return trace_seq_printf(s, "\nCPU[%03d] "
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
" ------------8<---------- thread %s-%d"
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
" ------------8<----------\n\n",
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
cpu, comm, pid);
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
}
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
static bool
trace_branch_is_leaf(struct trace_iterator *iter,
struct ftrace_graph_ent_entry *curr)
{
struct ring_buffer_iter *ring_iter;
struct ring_buffer_event *event;
struct ftrace_graph_ret_entry *next;
ring_iter = iter->buffer_iter[iter->cpu];
if (!ring_iter)
return false;
event = ring_buffer_iter_peek(iter->buffer_iter[iter->cpu], NULL);
if (!event)
return false;
next = ring_buffer_event_data(event);
if (next->ent.type != TRACE_GRAPH_RET)
return false;
if (curr->ent.pid != next->ent.pid ||
curr->graph_ent.func != next->ret.func)
return false;
return true;
}
static inline int
print_graph_duration(unsigned long long duration, struct trace_seq *s)
{
unsigned long nsecs_rem = do_div(duration, 1000);
return trace_seq_printf(s, "+ %llu.%lu us\n", duration, nsecs_rem);
}
/* Signal a overhead of time execution to the output */
static int
print_graph_overhead(unsigned long long duration, struct trace_seq *s)
{
/* Duration exceeded 100 msecs */
if (duration > 100000ULL)
return trace_seq_printf(s, "! ");
/* Duration exceeded 10 msecs */
if (duration > 10000ULL)
return trace_seq_printf(s, "+ ");
return trace_seq_printf(s, " ");
}
/* Case of a leaf function on its call entry */
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
static enum print_line_t
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
print_graph_entry_leaf(struct trace_iterator *iter,
struct ftrace_graph_ent_entry *entry, struct trace_seq *s)
{
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
struct ftrace_graph_ret_entry *ret_entry;
struct ftrace_graph_ret *graph_ret;
struct ring_buffer_event *event;
struct ftrace_graph_ent *call;
unsigned long long duration;
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
int i;
int ret;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
event = ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
ret_entry = ring_buffer_event_data(event);
graph_ret = &ret_entry->ret;
call = &entry->graph_ent;
duration = graph_ret->rettime - graph_ret->calltime;
/* Overhead */
ret = print_graph_overhead(duration, s);
if (!ret)
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
/* Function */
for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
ret = trace_seq_printf(s, " ");
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
}
ret = seq_print_ip_sym(s, call->func, 0);
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
ret = trace_seq_printf(s, "();");
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
if (!ret)
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
/* Duration */
ret = trace_seq_printf(s, TRACE_GRAPH_TIMESPACE_ENTRY);
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
ret = print_graph_duration(duration, s);
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
return TRACE_TYPE_HANDLED;
}
static enum print_line_t
print_graph_entry_nested(struct ftrace_graph_ent_entry *entry,
struct trace_seq *s)
{
int i;
int ret;
struct ftrace_graph_ent *call = &entry->graph_ent;
/* No overhead */
ret = trace_seq_printf(s, " ");
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
/* Function */
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
ret = trace_seq_printf(s, " ");
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
}
ret = seq_print_ip_sym(s, call->func, 0);
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
ret = trace_seq_printf(s, "() {");
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
/* No duration to print at this state */
ret = trace_seq_printf(s, TRACE_GRAPH_TIMESPACE_ENTRY "-\n");
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
return TRACE_TYPE_HANDLED;
}
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
static enum print_line_t
print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
struct trace_iterator *iter, int cpu)
{
int ret;
struct trace_entry *ent = iter->ent;
if (!verif_pid(s, ent->pid, cpu))
return TRACE_TYPE_PARTIAL_LINE;
ret = trace_seq_printf(s, "CPU[%03d] ", cpu);
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
if (trace_branch_is_leaf(iter, field))
return print_graph_entry_leaf(iter, field, s);
else
return print_graph_entry_nested(field, s);
}
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
static enum print_line_t
print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
struct trace_entry *ent, int cpu)
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
{
int i;
int ret;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
unsigned long long duration = trace->rettime - trace->calltime;
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
/* Pid */
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
if (!verif_pid(s, ent->pid, cpu))
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
/* Cpu */
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
ret = trace_seq_printf(s, "CPU[%03d] ", cpu);
if (!ret)
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
/* Overhead */
ret = print_graph_overhead(duration, s);
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
/* Closing brace */
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
ret = trace_seq_printf(s, " ");
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
}
ret = trace_seq_printf(s, "} ");
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
/* Duration */
ret = trace_seq_printf(s, TRACE_GRAPH_TIMESPACE_RET);
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
ret = print_graph_duration(duration, s);
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
/* Overrun */
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERRUN) {
ret = trace_seq_printf(s, " (Overruns: %lu)\n",
trace->overrun);
if (!ret)
return TRACE_TYPE_PARTIAL_LINE;
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
}
return TRACE_TYPE_HANDLED;
}
enum print_line_t
print_graph_function(struct trace_iterator *iter)
{
struct trace_seq *s = &iter->seq;
struct trace_entry *entry = iter->ent;
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
switch (entry->type) {
case TRACE_GRAPH_ENT: {
struct ftrace_graph_ent_entry *field;
trace_assign_type(field, entry);
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
return print_graph_entry(field, s, iter,
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
iter->cpu);
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
}
case TRACE_GRAPH_RET: {
struct ftrace_graph_ret_entry *field;
trace_assign_type(field, entry);
ftrace: add cpu annotation for function graph tracer 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>
2008-11-26 13:16:27 +08:00
return print_graph_return(&field->ret, s, entry, iter->cpu);
tracing/function-return-tracer: set a more human readable output 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>
2008-11-26 07:57:25 +08:00
}
default:
return TRACE_TYPE_UNHANDLED;
}
}
static struct tracer graph_trace __read_mostly = {
tracing/function-graph-tracer: enhancements for the trace output 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>
2008-11-27 08:46:33 +08:00
.name = "function_graph",
.init = graph_trace_init,
.reset = graph_trace_reset,
.print_line = print_graph_function,
.flags = &tracer_flags,
};
static __init int init_graph_trace(void)
{
return register_tracer(&graph_trace);
}
device_initcall(init_graph_trace);