tracing/ftrace: Introduce the big kernel lock tracer
Subsystems (2)
PATCH RESEND 1/1 tracing/ftrace: Introduce the big kernel lock tracer
Introduce the Big Kernel Lock tracer.
This new tracer lets one to observe the latencies caused by the lock_kernel() function.
It captures the time when the task request the spinlock, the time when the spinlock is hold and the time it is released. This way we can measure the latency for a task that wait for this spinlock and the latency caused by the time this lock is hold.
I will surely add some improvements like:
_ Adding a stack trace to know which function locked or unlocked the kernel. _ Adding a script to get some statistics about this: which functions hold it often, which hold it for a long time and some other things....
Signed-off-by Frederic Weisbecker <[email protected]>
Cc Steven Rostedt <[email protected]>
--- include/trace/bkl.h | 29 ++++++++ kernel/trace/Kconfig | 11 +++ kernel/trace/Makefile | 1 + kernel/trace/trace.h | 18 +++++ kernel/trace/trace_bkl.c | 170 ++++++++++++++++++++++++++++++++++++++++++++++ lib/kernel_lock.c | 22 ++++++- 6 files changed, 249 insertions(+), 2 deletions(-)
include/trace/bkl.h new file +29 -0
@@ -0,0 +1,29 @@ +#ifndef _TRACE_BKL_H +#define _TRACE_BKL_H + + +/* Entry which log the time when the task tried first to acquire the bkl +* and the time when it acquired it. +* That will let us know the latency to acquire the kernel lock. +*/ +struct bkl_trace_acquire { +unsigned long long acquire_req_time; +unsigned long long acquire_time; +}; + +/* This will log the time when the task released the bkl. +* So we will know the whole duration of its kernel lock. +*/ +struct bkl_trace_release { +unsigned long long release_time; +}; + +#ifdef CONFIG_BKL_TRACER +extern void trace_bkl_acquire (struct bkl_trace_acquire *trace); +extern void trace_bkl_release (struct bkl_trace_release *trace); +#else +static void trace_bkl_acquire (struct bkl_trace_acquire *trace) { } +static void trace_bkl_release (struct bkl_trace_release *trace) { } +#endif + +#endif
kernel/trace/Kconfig +11 -0
@@ -144,6 +144,17 @@ config STACK_TRACER This tracer records the max stack of the kernel, and displays it in debugfs/tracing/stack_trace +config BKL_TRACER + bool "Trace Big Kernel Lock latencies" + depends on HAVE_FTRACE + depends on DEBUG_KERNEL + select FTRACE + select STACKTRACE + help + This tracer records the latencies issued by the Big Kernel Lock. + It traces the time when a task request the bkl, the time when it + acquires it and the time when it releases it. + config DYNAMIC_FTRACE bool "enable/disable ftrace tracepoints dynamically" depends on FTRACE
kernel/trace/Makefile +1 -0
@@ -24,5 +24,6 @@ obj-$(CONFIG_NOP_TRACER) += trace_nop.o obj-$(CONFIG_STACK_TRACER) += trace_stack.o obj-$(CONFIG_MMIOTRACE) += trace_mmiotrace.o obj-$(CONFIG_BOOT_TRACER) += trace_boot.o +obj-$(CONFIG_BKL_TRACER) += trace_bkl.o libftrace-y := ftrace.o
kernel/trace/trace.h +18 -0
@@ -8,6 +8,7 @@#include <linux/ring_buffer.h> #include <linux/mmiotrace.h> #include <linux/ftrace.h> +#include <trace/bkl.h> enum trace_type { __TRACE_FIRST_TYPE =0 ,
@@ -22,6 +23,8 @@ enum trace_type {TRACE_MMIO_RW ,TRACE_MMIO_MAP ,TRACE_BOOT , +TRACE_BKL_ACQ , +TRACE_BKL_REL , __TRACE_LAST_TYPE };
@@ -119,6 +122,17 @@ struct trace_boot {struct boot_trace initcall; }; +/* Trace bkl entries */ +struct trace_bkl_acq { +struct trace_entry ent; +struct bkl_trace_acquire bkl_acq; +}; + +struct trace_bkl_rel { +struct trace_entry ent; +struct bkl_trace_release bkl_rel; +}; +/* * trace_flag_type is an enumeration that holds different * states when a trace occurs. These are:
@@ -219,6 +233,10 @@ extern void __ftrace_bad_type(void); IF_ASSIGN(var, ent, struct trace_mmiotrace_map, \TRACE_MMIO_MAP ); \ IF_ASSIGN(var, ent, struct trace_boot, TRACE_BOOT); \ + IF_ASSIGN(var, ent, struct trace_bkl_acq, \ +TRACE_BKL_ACQ ); \ + IF_ASSIGN(var, ent, struct trace_bkl_rel, \ +TRACE_BKL_REL ); \__ftrace_bad_type (); \ }while (0 )
kernel/trace/trace_bkl.c new file +170 -0
170 lines not on this page · raw message
lib/kernel_lock.c +20 -2
@@ -9,6 +9,7 @@#include <linux/module.h> #include <linux/kallsyms.h> #include <linux/semaphore.h> +#include <trace/bkl.h> /* * The 'big kernel lock'
@@ -116,16 +117,33 @@ static inline void __unlock_kernel(void) void __lockfunc lock_kernel(void) {int depth = current->lock_depth+1 ; -if (likely (!depth)) +if (likely (!depth)) { +#ifdef CONFIG_BKL_TRACER +struct bkl_trace_acquire trace; +preempt_disable (); + trace.acquire_req_time =cpu_clock (raw_smp_processor_id ()); +preempt_enable ();__lock_kernel (); + trace.acquire_time =cpu_clock (raw_smp_processor_id ()); +trace_bkl_acquire (&trace); +#else +__lock_kernel (); +#endif + } current->lock_depth = depth; } void __lockfunc unlock_kernel(void) {BUG_ON (current->lock_depth <0 ); -if (likely (--current->lock_depth <0 )) +if (likely (--current->lock_depth <0 )) { +#ifdef CONFIG_BKL_TRACER +struct bkl_trace_release trace; + trace.release_time =cpu_clock (raw_smp_processor_id ()); +trace_bkl_release (&trace); +#endif __unlock_kernel (); + } }EXPORT_SYMBOL (lock_kernel);
Quoted 17 lines · Frederic Weisbecker
Signed-off-by Frederic Weisbecker <[email protected]>Cc Steven Rostedt <[email protected]>--- include/trace/bkl.h | 29 ++++++++ kernel/trace/Kconfig | 11 +++ kernel/trace/Makefile | 1 + kernel/trace/trace.h | 18 +++++ kernel/trace/trace_bkl.c | 170 ++++++++++++++++++++++++++++++++++++++++++++++ lib/kernel_lock.c | 22 ++++++- 6 files changed, 249 insertions(+), 2 deletions(-)include/trace/bkl.h new file +0
-0
include/trace/bkl.h line 8
Quoted 8 lines
include/trace/bkl.h new file +8
-0@@ -0,0 +1,29 @@ +#ifndef _TRACE_BKL_H +#define _TRACE_BKL_H + + +/* Entry which log the time when the task tried first to acquire the bkl +* and the time when it acquired it. +* That will let us know the latency to acquire the kernel lock. +*/
Please use the more common/accepted long-comment style:
/*
* comment line 1
* comment line 2
*/
like you did in some other places.
include/trace/bkl.h line 16
Quoted 8 lines
include/trace/bkl.h new file +8
-0+struct bkl_trace_acquire { +unsigned long long acquire_req_time; +unsigned long long acquire_time; +}; + +/* This will log the time when the task released the bkl. +* So we will know the whole duration of its kernel lock. +*/
ditto et al
Quoted 20 lines
include/trace/bkl.h new file +13
-0+struct bkl_trace_release { +unsigned long long release_time; +}; + +#ifdef CONFIG_BKL_TRACER +extern void trace_bkl_acquire (struct bkl_trace_acquire *trace); +extern void trace_bkl_release (struct bkl_trace_release *trace); +#else +static void trace_bkl_acquire (struct bkl_trace_acquire *trace) { } +static void trace_bkl_release (struct bkl_trace_release *trace) { } +#endif + +#endif kernel/trace/Kconfig +0
-0@@ -144,6 +144,17 @@ config STACK_TRACER This tracer records the max stack of the kernel, and displays it in debugfs/tracing/stack_trace
kernel/trace/Kconfig config STACK_TRACER · line 155
Quoted 10 lines
kernel/trace/Kconfig +9
-0+config BKL_TRACER + bool "Trace Big Kernel Lock latencies" + depends on HAVE_FTRACE + depends on DEBUG_KERNEL + select FTRACE + select STACKTRACE + help + This tracer records the latencies issued by the Big Kernel Lock. + It traces the time when a task request the bkl, the time when it
requests
Quoted 23 lines
kernel/trace/Kconfig +2
-0+ acquires it and the time when it releases it. + config DYNAMIC_FTRACE bool "enable/disable ftrace tracepoints dynamically" depends on FTRACEkernel/trace/trace_bkl.c new file +11
-0@@ -0,0 +1,170 @@ +/* +* Ring buffer based Big Kernel Lock tracer +* for purpose of measuring the latencies issued +* by the BKL. +* +* Copyright (C) 2008 Frederic Weisbecker <[email protected]> +*/ + +#include <trace/bkl.h> +#include "trace.h" +#include <linux/ftrace.h>
kernel/trace/trace_bkl.c line 21
Quoted 10 lines
kernel/trace/trace_bkl.c new file +10
-0+ + +static struct trace_array *bkl_trace; +static bool bkl_tracer_enabled; + + +static void bkl_trace_init (struct trace_array *tr) +{ +int cpu; + bkl_trace = tr;
what is <bkl_trace> used for? or unused?
kernel/trace/trace_bkl.c line 28
Quoted 7 lines
kernel/trace/trace_bkl.c new file +7
-0+ +for_each_cpu_mask (cpu, cpu_possible_map) +tracing_reset (tr, cpu); + +if (tr->ctrl) + bkl_tracer_enabled =1 ; +}
~Randy
2008/10/22 Randy Dunlap <[email protected]>
Quoted 8 lines
Please use the more common/accepted long-comment style:
/*
* comment line 1
* comment line 2
*/
Ok, I will correct it.
Quoted 1 line
requests
Oops....
Quoted 1 line
what is <
bkl_trace> used for? or unused?
It is used on trace_bkl_acquire and trace_bkl_release functions. It is really necessary
to keep a reference to the global trace array.
I can send a new patch with the corrections or send a new one which applies on this one
if it is accepted.
As you want Ingo...
Steven Rostedt
maintainerQuoted 7 lines · Frederic Weisbecker
include/trace/bkl.h new file +0
-0
include/trace/bkl.h line 8
Quoted 8 lines
include/trace/bkl.h new file +8
-0@@ -0,0 +1,29 @@ +#ifndef _TRACE_BKL_H +#define _TRACE_BKL_H + + +/* Entry which log the time when the task tried first to acquire the bkl +* and the time when it acquired it. +* That will let us know the latency to acquire the kernel lock. +*/
We recommend the
/*
* Multi line
* comment style.
*/
Quoted 8 lines
include/trace/bkl.h new file +8
-0+struct bkl_trace_acquire { +unsigned long long acquire_req_time; +unsigned long long acquire_time; +}; + +/* This will log the time when the task released the bkl. +* So we will know the whole duration of its kernel lock. +*/
include/trace/bkl.h line 26
Quoted 10 lines
include/trace/bkl.h new file +10
-0+struct bkl_trace_release { +unsigned long long release_time; +}; + +#ifdef CONFIG_BKL_TRACER +extern void trace_bkl_acquire (struct bkl_trace_acquire *trace); +extern void trace_bkl_release (struct bkl_trace_release *trace); +#else +static void trace_bkl_acquire (struct bkl_trace_acquire *trace) { } +static void trace_bkl_release (struct bkl_trace_release *trace) { }
I think you want these as "static inline"
Quoted 8 lines
include/trace/bkl.h new file +3
-0+#endif + +#endif kernel/trace/Kconfig +0
-0
kernel/trace/Kconfig config STACK_TRACER · line 149
Quoted 6 lines
kernel/trace/Kconfig +3
-0@@ -144,6 +144,17 @@ config STACK_TRACER This tracer records the max stack of the kernel, and displays it in debugfs/tracing/stack_trace +config BKL_TRACER + bool "Trace Big Kernel Lock latencies" + depends on HAVE_FTRACE
You don't need FTRACE for this.
kernel/trace/Kconfig config STACK_TRACER · line 151
Quoted 2 lines
kernel/trace/Kconfig +2
-0+ depends on DEBUG_KERNEL + select FTRACE
Again, you don't need ftrace for this. You should just be able to trace the latencies and report the places that are big. If FTRACE is configured in, then you can also get the function trace. We have patches to replace FTRACE with FUNCTION_TRACER to eliminate this confusion.
Quoted 113 lines
kernel/trace/Kconfig +6
-0+ select STACKTRACE + help + This tracer records the latencies issued by the Big Kernel Lock. + It traces the time when a task request the bkl, the time when it + acquires it and the time when it releases it. + config DYNAMIC_FTRACE bool "enable/disable ftrace tracepoints dynamically" depends on FTRACEkernel/trace/Makefile +1
-0@@ -24,5 +24,6 @@ obj-$(CONFIG_NOP_TRACER) += trace_nop.o obj-$(CONFIG_STACK_TRACER) += trace_stack.o obj-$(CONFIG_MMIOTRACE) += trace_mmiotrace.o obj-$(CONFIG_BOOT_TRACER) += trace_boot.o +obj-$(CONFIG_BKL_TRACER) += trace_bkl.o libftrace-y := ftrace.okernel/trace/trace.h +18
-0@@ -8,6 +8,7 @@#include <linux/ring_buffer.h> #include <linux/mmiotrace.h> #include <linux/ftrace.h> +#include <trace/bkl.h> enum trace_type { __TRACE_FIRST_TYPE =0 ,@@ -22,6 +23,8 @@ enum trace_type {TRACE_MMIO_RW ,TRACE_MMIO_MAP ,TRACE_BOOT , +TRACE_BKL_ACQ , +TRACE_BKL_REL , __TRACE_LAST_TYPE };@@ -119,6 +122,17 @@ struct trace_boot {struct boot_trace initcall; }; +/* Trace bkl entries */ +struct trace_bkl_acq { +struct trace_entry ent; +struct bkl_trace_acquire bkl_acq; +}; + +struct trace_bkl_rel { +struct trace_entry ent; +struct bkl_trace_release bkl_rel; +}; +/* * trace_flag_type is an enumeration that holds different * states when a trace occurs. These are: @@ -219,6 +233,10 @@ extern void __ftrace_bad_type(void); IF_ASSIGN(var, ent, struct trace_mmiotrace_map, \TRACE_MMIO_MAP ); \ IF_ASSIGN(var, ent, struct trace_boot, TRACE_BOOT); \ + IF_ASSIGN(var, ent, struct trace_bkl_acq, \ +TRACE_BKL_ACQ ); \ + IF_ASSIGN(var, ent, struct trace_bkl_rel, \ +TRACE_BKL_REL ); \__ftrace_bad_type (); \ }while (0 )kernel/trace/trace_bkl.c new file +37
-0@@ -0,0 +1,170 @@ +/* +* Ring buffer based Big Kernel Lock tracer +* for purpose of measuring the latencies issued +* by the BKL. +* +* Copyright (C) 2008 Frederic Weisbecker <[email protected]> +*/ + +#include <trace/bkl.h> +#include "trace.h" +#include <linux/ftrace.h> + + +static struct trace_array *bkl_trace; +static bool bkl_tracer_enabled; + + +static void bkl_trace_init (struct trace_array *tr) +{ +int cpu; + bkl_trace = tr; + +for_each_cpu_mask (cpu, cpu_possible_map) +tracing_reset (tr, cpu); + +if (tr->ctrl) + bkl_tracer_enabled =1 ; +} + +static void reset_bkl_trace (struct trace_array *tr) +{ + bkl_tracer_enabled =0 ; +} + +static void bkl_trace_ctrl_update (struct trace_array *tr) +{ +if (tr->ctrl)
kernel/trace/trace_bkl.c line 47
Quoted 10 lines
kernel/trace/trace_bkl.c new file +10
-0+ bkl_tracer_enabled =1 ; + else + bkl_tracer_enabled =0 ; +} + +static enum print_line_t bkl_acq_print_line (struct trace_iterator *iter) +{ +int ret; +struct trace_entry *entry = iter->ent; +struct trace_bkl_acq *field = (struct trace_bkl_acq *) entry;
Could you use the trace_assign_type(field, entry) instead. See trace.c for examples.
Quoted 19 lines
kernel/trace/trace_bkl.c new file +19
-0+struct bkl_trace_acquire *bta = &field->bkl_acq; +struct trace_seq *s = &iter->seq; +unsigned long sec, nsec_rem; +unsigned long long wait_for_spin; + + wait_for_spin = bta->acquire_time - bta->acquire_req_time; + nsec_rem =do_div (bta->acquire_req_time,1000000000 ); + sec = (unsigned long ) bta->acquire_req_time; + ret =trace_seq_printf (s,"Kernel lock requested at %5lu.%09lu " , + sec, nsec_rem); +if (!ret) +return TRACE_TYPE_PARTIAL_LINE ; + + nsec_rem =do_div (bta->acquire_time,1000000000 ); + sec = (unsigned long ) bta->acquire_time; + ret =trace_seq_printf (s,"acquired at %5lu.%09lu " +"(waited %llu nsecs)\n" , + sec, nsec_rem, + wait_for_spin);
kernel/trace/trace_bkl.c line 76
Quoted 10 lines
kernel/trace/trace_bkl.c new file +10
-0+if (!ret) +return TRACE_TYPE_PARTIAL_LINE ; +return TRACE_TYPE_HANDLED ; +} + +static enum print_line_t bkl_rel_print_line (struct trace_iterator *iter) +{ +int ret; +struct trace_entry *entry = iter->ent; +struct trace_bkl_rel *field = (struct trace_bkl_rel *) entry;
Same here.
Quoted 5 lines
kernel/trace/trace_bkl.c new file +5
-0+struct bkl_trace_release *btr = &field->bkl_rel; +struct trace_seq *s = &iter->seq; +unsigned long sec, nsec_rem; + + nsec_rem =do_div (btr->release_time,1000000000 );
kernel/trace/trace_bkl.c line 91
Quoted 10 lines
kernel/trace/trace_bkl.c new file +10
-0+ sec = (unsigned long ) btr->release_time; + ret =trace_seq_printf (s,"Kernel lock released at %5lu.%09lu\n" , + sec, nsec_rem); +return TRACE_TYPE_HANDLED ; +} + +static enum print_line_t bkl_trace_print_line (struct trace_iterator *iter) +{ +struct trace_entry *entry = iter->ent; +
You might want to make this into a switch instead.
Quoted 30 lines
kernel/trace/trace_bkl.c new file +30
-0+if (entry->type ==TRACE_BKL_ACQ ) +return bkl_acq_print_line (iter); +if (entry->type ==TRACE_BKL_REL ) +return bkl_rel_print_line (iter); + +return TRACE_TYPE_UNHANDLED ; +} + +static struct tracer bkl_tracer__read_mostly = +{ + .name ="bkl" , + .init = bkl_trace_init, + .reset = reset_bkl_trace, + .ctrl_update = bkl_trace_ctrl_update, + .print_line = bkl_trace_print_line, +}; + +__init static int init_bkl_tracer (void ) +{ +return register_tracer (&bkl_tracer); +} +device_initcall (init_bkl_tracer); + +/* +* These functions don't need us to disable +* the preemption since it has been done by lock_kernel() +*/ + +void trace_bkl_acquire (struct bkl_trace_acquire *trace) +{
kernel/trace/trace_bkl.c line 131
Quoted 10 lines
kernel/trace/trace_bkl.c new file +10
-0+struct ring_buffer_event *event; +struct trace_bkl_acq *entry; +struct trace_array_cpu *data; +unsigned long irq_flags; +struct trace_array *tr = bkl_trace; + +if (!bkl_tracer_enabled) +return ; + + data = tr->data[smp_processor_id ()];
data is unused.
Quoted 16 lines
kernel/trace/trace_bkl.c new file +16
-0+ + event =ring_buffer_lock_reserve (tr->buffer,sizeof (*entry), + &irq_flags); +if (!event) +return ; + entry =ring_buffer_event_data (event); +tracing_generic_entry_update (&entry->ent,0 ,0 ); + entry->ent.type =TRACE_BKL_ACQ ; + entry->bkl_acq = *trace; +ring_buffer_unlock_commit (tr->buffer, event, irq_flags); + +trace_wake_up (); +} + +void trace_bkl_release (struct bkl_trace_release *trace) +{
kernel/trace/trace_bkl.c line 157
Quoted 10 lines
kernel/trace/trace_bkl.c new file +10
-0+struct ring_buffer_event *event; +struct trace_bkl_rel *entry; +struct trace_array_cpu *data; +unsigned long irq_flags; +struct trace_array *tr = bkl_trace; + +if (!bkl_tracer_enabled) +return ; + + data = tr->data[smp_processor_id ()];
data is unused here too.
Quoted 28 lines
kernel/trace/trace_bkl.c new file +13
-0+ + event =ring_buffer_lock_reserve (tr->buffer,sizeof (*entry), + &irq_flags); +if (!event) +return ; + entry =ring_buffer_event_data (event); +tracing_generic_entry_update (&entry->ent,0 ,0 ); + entry->ent.type =TRACE_BKL_REL ; + entry->bkl_rel = *trace; +ring_buffer_unlock_commit (tr->buffer, event, irq_flags); + +trace_wake_up (); +}lib/kernel_lock.c +1
-0@@ -9,6 +9,7 @@#include <linux/module.h> #include <linux/kallsyms.h> #include <linux/semaphore.h> +#include <trace/bkl.h> /* * The 'big kernel lock' @@ -116,16 +117,33 @@ static inline void __unlock_kernel(void) void __lockfunc lock_kernel(void) {
lib/kernel_lock.c static inline void __unlock_kernel(void) · line 127
Quoted 10 lines
lib/kernel_lock.c +7
-1int depth = current->lock_depth+1 ; -if (likely (!depth)) +if (likely (!depth)) { +#ifdef CONFIG_BKL_TRACER +struct bkl_trace_acquire trace; +preempt_disable (); + trace.acquire_req_time =cpu_clock (raw_smp_processor_id ()); +preempt_enable ();__lock_kernel (); + trace.acquire_time =cpu_clock (raw_smp_processor_id ());
Note, the cpu retuned here may not be the same as the above processor id.
lib/kernel_lock.c static inline void __unlock_kernel(void) · line 131
Quoted 4 lines
lib/kernel_lock.c +4
-0+trace_bkl_acquire (&trace); +#else +__lock_kernel (); +#endif
The above looks a bit ugly. Could you do something perhaps at the beginning of the file like:
#ifdef CONFIG_BKL_TRACER
static void lock_kernel_trace(void) {
struct bkl_trace_acquire trace;
preempt_disable();
trace.acquire_req_time = cpu_clock(raw_smp_processor_id());
preempt_enable();
__lock_kernel();
trace.acquire_time = cpu_clock(raw_smp_processor_id());
trace_bkl_acquire(&trace);
}
#else
# define lock_kernel_trace() __lock_kernel();
#endifAnd then just add the lock_kernel_trace within the code.
Quoted 4 lines
lib/kernel_lock.c +1
-0+ } current->lock_depth = depth; }
lib/kernel_lock.c static inline void __unlock_kernel(void) · line 144
Quoted 10 lines
lib/kernel_lock.c +6
-1void __lockfunc unlock_kernel(void) {BUG_ON (current->lock_depth <0 ); -if (likely (--current->lock_depth <0 )) +if (likely (--current->lock_depth <0 )) { +#ifdef CONFIG_BKL_TRACER +struct bkl_trace_release trace; + trace.release_time =cpu_clock (raw_smp_processor_id ()); +trace_bkl_release (&trace); +#endif
Same here. Have an unlock_kernel_trace();
lib/kernel_lock.c static inline void __unlock_kernel(void) · line 149
Quoted 7 lines
lib/kernel_lock.c +1
-0__unlock_kernel (); + } }EXPORT_SYMBOL (lock_kernel);
-- Steve
Ok, I will send a V2 to apply your comments.
Thanks!