Re: [PATCH v4 3/3] tracing: format non-nanosec times from tsc clockwithout a decimal point.

From: David Sharp
Date: Tue Sep 25 2012 - 18:29:45 EST

Next message: David Sharp: "[PATCH v4 3/3] tracing: format non-nanosec times from tsc clock without a decimal point."
Previous message: Dave Jiang: "[PATCH] MAINTAINERS: update Intel C600 SAS driver maintainers"
In reply to: Steven Rostedt: "Re: [PATCH v4 3/3] tracing: format non-nanosec times from tsc clockwithout a decimal point."
Next in thread: David Sharp: "[PATCH v4 3/3] tracing: format non-nanosec times from tsc clock without a decimal point."
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Tue, Sep 25, 2012 at 2:42 PM, Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> Sorry, I should have been more picky before. I haven't totally tested
> this yet.
>
> On Tue, 2012-09-25 at 13:49 -0700, David Sharp wrote:
>> With the addition of the "tsc" clock, formatting timestamps to look like
>> fractional seconds is misleading. Mark clocks as either in nanoseconds or
>> not, and format non-nanosecond timestamps as decimal integers.
>>
>> Tested:
>> $ cd /sys/kernel/debug/tracing/
>> $ cat trace_clock
>> [local] global tsc
>> $ echo sched_switch > set_event
>> $ echo 1 > tracing_enabled ; sleep 0.0005 ; echo 0 > tracing_enabled
>> $ cat trace
>> <idle>-0 [000] 6330.555552: sched_switch: prev_comm=swapper prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=bash next_pid=29964 next_prio=120
>> sleep-29964 [000] 6330.555628: sched_switch: prev_comm=bash prev_pid=29964 prev_prio=120 prev_state=S ==> next_comm=swapper next_pid=0 next_prio=120
>> ...
>> $ echo 1 > options/latency-format
>> $ cat trace
>> <idle>-0 0 4104553247us+: sched_switch: prev_comm=swapper prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=bash next_pid=29964 next_prio=120
>> sleep-29964 0 4104553322us+: sched_switch: prev_comm=bash prev_pid=29964 prev_prio=120 prev_state=S ==> next_comm=swapper next_pid=0 next_prio=120
>> ...
>> $ echo tsc > trace_clock
>> $ cat trace
>> $ echo 1 > tracing_enabled ; sleep 0.0005 ; echo 0 > tracing_enabled
>> $ echo 0 > options/latency-format
>> $ cat trace
>> <idle>-0 [000] 16490053398357: sched_switch: prev_comm=swapper prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=bash next_pid=31128 next_prio=120
>> sleep-31128 [000] 16490053588518: sched_switch: prev_comm=bash prev_pid=31128 prev_prio=120 prev_state=S ==> next_comm=swapper next_pid=0 next_prio=120
>> ...
>> echo 1 > options/latency-format
>> $ cat trace
>> <idle>-0 0 91557653238+: sched_switch: prev_comm=swapper prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=bash next_pid=31128 next_prio=120
>> sleep-31128 0 91557843399+: sched_switch: prev_comm=bash prev_pid=31128 prev_prio=120 prev_state=S ==> next_comm=swapper next_pid=0 next_prio=120
>> ...
>>
>> v2:
>> Move arch-specific bits out of generic code.
>> v4:
>> Fix x86_32 build due to 64-bit division.
>>
>> Google-Bug-Id: 6980623
>> Signed-off-by: David Sharp <dhsharp@xxxxxxxxxx>
>> Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
>> Cc: Masami Hiramatsu <masami.hiramatsu.pt@xxxxxxxxxxx>
>> ---
>> arch/x86/include/asm/trace_clock.h | 2 +-
>> include/linux/ftrace_event.h | 6 +++
>> kernel/trace/trace.c | 15 +++++-
>> kernel/trace/trace.h | 4 --
>> kernel/trace/trace_output.c | 84 +++++++++++++++++++++++++-----------
>> 5 files changed, 78 insertions(+), 33 deletions(-)
>>
>> diff --git a/arch/x86/include/asm/trace_clock.h b/arch/x86/include/asm/trace_clock.h
>> index 7ee0d8c..45e17f5 100644
>> --- a/arch/x86/include/asm/trace_clock.h
>> +++ b/arch/x86/include/asm/trace_clock.h
>> @@ -9,7 +9,7 @@
>> extern u64 notrace trace_clock_x86_tsc(void);
>>
>> # define ARCH_TRACE_CLOCKS \
>> - { trace_clock_x86_tsc, "x86-tsc" },
>> + { trace_clock_x86_tsc, "x86-tsc", .in_ns = 0 },
>>
>> #endif
>>
>> diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h
>> index 642928c..c760670 100644
>> --- a/include/linux/ftrace_event.h
>> +++ b/include/linux/ftrace_event.h
>> @@ -86,6 +86,12 @@ struct trace_iterator {
>> cpumask_var_t started;
>> };
>>
>> +enum trace_iter_flags {
>> + TRACE_FILE_LAT_FMT = 1,
>> + TRACE_FILE_ANNOTATE = 2,
>> + TRACE_FILE_TIME_IN_NS = 4,
>> +};
>> +
>>
>> struct trace_event;
>>
>> diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
>> index 4e26df3..3fe4c5b 100644
>> --- a/kernel/trace/trace.c
>> +++ b/kernel/trace/trace.c
>> @@ -476,10 +476,11 @@ static const char *trace_options[] = {
>> static struct {
>> u64 (*func)(void);
>> const char *name;
>> + int in_ns; /* is this clock in nanoseconds? */
>
> Add a few tabs between the ns; and /*

Done.

>
>
>> } trace_clocks[] = {
>> - { trace_clock_local, "local" },
>> - { trace_clock_global, "global" },
>> - { trace_clock_counter, "counter" },
>> + { trace_clock_local, "local", 1 },
>> + { trace_clock_global, "global", 1 },
>> + { trace_clock_counter, "counter", 0 },
>> ARCH_TRACE_CLOCKS
>> };
>>
>> @@ -2425,6 +2426,10 @@ __tracing_open(struct inode *inode, struct file *file)
>> if (ring_buffer_overruns(iter->tr->buffer))
>> iter->iter_flags |= TRACE_FILE_ANNOTATE;
>>
>> + /* Output in nanoseconds only if we are using a clock in nanoseconds. */
>> + if (trace_clocks[trace_clock_id].in_ns)
>> + iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
>> +
>> /* stop the trace while dumping */
>> tracing_stop();
>>
>> @@ -3324,6 +3329,10 @@ static int tracing_open_pipe(struct inode *inode, struct file *filp)
>> if (trace_flags & TRACE_ITER_LATENCY_FMT)
>> iter->iter_flags |= TRACE_FILE_LAT_FMT;
>>
>> + /* Output in nanoseconds only if we are using a clock in nanoseconds. */
>> + if (trace_clocks[trace_clock_id].in_ns)
>> + iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
>> +
>> iter->cpu_file = cpu_file;
>> iter->tr = &global_trace;
>> mutex_init(&iter->mutex);
>> diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
>> index 55e1f7f..84fefed 100644
>> --- a/kernel/trace/trace.h
>> +++ b/kernel/trace/trace.h
>> @@ -409,10 +409,6 @@ void tracing_start_sched_switch_record(void);
>> int register_tracer(struct tracer *type);
>> void unregister_tracer(struct tracer *type);
>> int is_tracing_stopped(void);
>> -enum trace_file_type {
>> - TRACE_FILE_LAT_FMT = 1,
>> - TRACE_FILE_ANNOTATE = 2,
>> -};
>>
>> extern cpumask_var_t __read_mostly tracing_buffer_mask;
>>
>> diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c
>> index 123b189..ca640ff 100644
>> --- a/kernel/trace/trace_output.c
>> +++ b/kernel/trace/trace_output.c
>> @@ -610,24 +610,59 @@ lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
>> return trace_print_lat_fmt(s, entry);
>> }
>>
>> -static unsigned long preempt_mark_thresh = 100;
>> +static unsigned long preempt_mark_thresh_us = 100;
>> +/* roughly the same at 2.0GHz: */
>> +static unsigned long preempt_mark_thresh_cycles = 200000;
>>
>> static int
>> -lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
>> - unsigned long rel_usecs)
>> +lat_print_timestamp(struct trace_iterator *iter, u64 next_ts)
>> {
>> - return trace_seq_printf(s, " %4lldus%c: ", abs_usecs,
>> - rel_usecs > preempt_mark_thresh ? '!' :
>> - rel_usecs > 1 ? '+' : ' ');
>> + int ret;
>> + struct trace_seq *s = &iter->seq;
>
> Move the above down below, to give a nicer aesthetic look.

Well, that's a matter of opinion... you're the maintainer. :) Done.

>
>> + unsigned long verbose = trace_flags & TRACE_ITER_VERBOSE;
>> + unsigned long in_ns = iter->iter_flags & TRACE_FILE_TIME_IN_NS;
>> + unsigned long long abs_ts = iter->ts - iter->tr->time_start;
>> + unsigned long long rel_ts = next_ts - iter->ts;
> struct trace_seq *s = &iter->seq;
>> + unsigned long mark_thresh;
> int ret;
>
>> +
>> + if (in_ns) {
>> + abs_ts = ns2usecs(abs_ts);
>> + rel_ts = ns2usecs(rel_ts);
>> + mark_thresh = preempt_mark_thresh_us;
>> + } else
>> + mark_thresh = preempt_mark_thresh_cycles;
>> +
>> + if (verbose && in_ns) {
>> + unsigned long abs_msec = abs_ts;
>> + unsigned long abs_usec = do_div(abs_msec, USEC_PER_MSEC);
>> + unsigned long rel_msec = rel_ts;
>> + unsigned long rel_usec = do_div(rel_msec, USEC_PER_MSEC);
>
> Either add a space here, or move the declarations to the top of the
> function and keep the code part here.

Done.

>
>> + ret = trace_seq_printf(
>> + s, "[%08llx] %ld.%03ldms (+%ld.%03ldms): ",
>> + ns2usecs(iter->ts),
>> + abs_msec, abs_usec,
>> + rel_msec, rel_usec);
>> + } else if (verbose && !in_ns) {
>> + ret = trace_seq_printf(
>> + s, "[%016llx] %lld (+%lld): ",
>> + iter->ts, abs_ts, rel_ts);
>> + } else { /* !verbose */
>> + ret = trace_seq_printf(
>> + s, " %4lld%s%c: ",
>> + abs_ts,
>> + in_ns ? "us" : "",
>> + rel_ts > mark_thresh ? '!' :
>> + rel_ts > 1 ? '+' : ' ');

I just noticed something about this: with x86-tsc clock, this will
always print a '+'. Does it matter? Also, is the 200k cycle threshold
for '!' okay? I guess the counter clock will always end up with rel_ts
== 1, so marks should never appear.

>> + }
>> + return ret;
>> }
>>
>> int trace_print_context(struct trace_iterator *iter)
>> {
>> struct trace_seq *s = &iter->seq;
>> struct trace_entry *entry = iter->ent;
>> - unsigned long long t = ns2usecs(iter->ts);
>> - unsigned long usec_rem = do_div(t, USEC_PER_SEC);
>> - unsigned long secs = (unsigned long)t;
>> + unsigned long long t;
>> + unsigned long secs, usec_rem;
>> char comm[TASK_COMM_LEN];
>> int ret;
>>
>> @@ -644,8 +679,13 @@ int trace_print_context(struct trace_iterator *iter)
>> return 0;
>> }
>>
>> - return trace_seq_printf(s, " %5lu.%06lu: ",
>> - secs, usec_rem);
>> + if (iter->iter_flags & TRACE_FILE_TIME_IN_NS) {
>> + t = ns2usecs(iter->ts);
>> + usec_rem = do_div(t, USEC_PER_SEC);
>> + secs = (unsigned long)t;
>> + return trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
>> + } else
>> + return trace_seq_printf(s, "%12llu: ", iter->ts);
>> }
>>
>> int trace_print_lat_context(struct trace_iterator *iter)
>> @@ -659,36 +699,30 @@ int trace_print_lat_context(struct trace_iterator *iter)
>> *next_entry = trace_find_next_entry(iter, NULL,
>> &next_ts);
>> unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
>> - unsigned long abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
>> - unsigned long rel_usecs;
>> +
>
> Is this adding an extra newline?

Yup. Fixed.

>
> -- Steve
>
>>
>> /* Restore the original ent_size */
>> iter->ent_size = ent_size;
>>
>> if (!next_entry)
>> next_ts = iter->ts;
>> - rel_usecs = ns2usecs(next_ts - iter->ts);
>>
>> if (verbose) {
>> char comm[TASK_COMM_LEN];
>>
>> trace_find_cmdline(entry->pid, comm);
>>
>> - ret = trace_seq_printf(s, "%16s %5d %3d %d %08x %08lx [%08llx]"
>> - " %ld.%03ldms (+%ld.%03ldms): ", comm,
>> - entry->pid, iter->cpu, entry->flags,
>> - entry->preempt_count, iter->idx,
>> - ns2usecs(iter->ts),
>> - abs_usecs / USEC_PER_MSEC,
>> - abs_usecs % USEC_PER_MSEC,
>> - rel_usecs / USEC_PER_MSEC,
>> - rel_usecs % USEC_PER_MSEC);
>> + ret = trace_seq_printf(
>> + s, "%16s %5d %3d %d %08x %08lx ",
>> + comm, entry->pid, iter->cpu, entry->flags,
>> + entry->preempt_count, iter->idx);
>> } else {
>> ret = lat_print_generic(s, entry, iter->cpu);
>> - if (ret)
>> - ret = lat_print_timestamp(s, abs_usecs, rel_usecs);
>> }
>>
>> + if (ret)
>> + ret = lat_print_timestamp(iter, next_ts);
>> +
>> return ret;
>> }
>>
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Next message: David Sharp: "[PATCH v4 3/3] tracing: format non-nanosec times from tsc clock without a decimal point."
Previous message: Dave Jiang: "[PATCH] MAINTAINERS: update Intel C600 SAS driver maintainers"
In reply to: Steven Rostedt: "Re: [PATCH v4 3/3] tracing: format non-nanosec times from tsc clockwithout a decimal point."
Next in thread: David Sharp: "[PATCH v4 3/3] tracing: format non-nanosec times from tsc clock without a decimal point."
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]