Re: [PATCH v1 1/1] binder: transaction latency tracking for user build

[Frankie Chang]

On Wed, 2020-02-05 at 10:49 -0500, Joel Fernandes wrote:
> On Wed, Feb 05, 2020 at 09:36:12AM +0000, Greg Kroah-Hartman wrote:
> > On Wed, Feb 05, 2020 at 02:52:52PM +0800, Frankie Chang wrote:
> > > Record start/end timestamp to binder transaction.
> > > When transaction is completed or transaction is free,
> > > it would be checked if transaction latency over threshold (2 sec),
> > > if yes, printing related information for tracing.
> > > 
> > > Signed-off-by: Frankie Chang <Frankie.Chang@xxxxxxxxxxxx>
> > > ---
> > >  drivers/android/Kconfig           |    8 +++
> > >  drivers/android/binder.c          |  107 +++++++++++++++++++++++++++++++++++++
> > >  drivers/android/binder_internal.h |    4 ++
> > >  3 files changed, 119 insertions(+)
> > > 
> > > diff --git a/drivers/android/Kconfig b/drivers/android/Kconfig
> > > index 6fdf2ab..7ba80eb 100644
> > > --- a/drivers/android/Kconfig
> > > +++ b/drivers/android/Kconfig
> > > @@ -54,6 +54,14 @@ config ANDROID_BINDER_IPC_SELFTEST
> > >  	  exhaustively with combinations of various buffer sizes and
> > >  	  alignments.
> > >  
> > > +config BINDER_USER_TRACKING
> > > +	bool "Android Binder transaction tracking"
> > > +	help
> > > +	  Used for track abnormal binder transaction which is over 2 seconds,
> > > +	  when the transaction is done or be free, this transaction would be
> > > +	  checked whether it executed overtime.
> > > +	  If yes, printing out the detail info about it.
> > > +
> > >  endif # if ANDROID
> > >  
> > >  endmenu
> > > diff --git a/drivers/android/binder.c b/drivers/android/binder.c
> > > index e9bc9fc..5a352ee 100644
> > > --- a/drivers/android/binder.c
> > > +++ b/drivers/android/binder.c
> > > @@ -76,6 +76,11 @@
> > >  #include "binder_internal.h"
> > >  #include "binder_trace.h"
> > >  
> > > +#ifdef CONFIG_BINDER_USER_TRACKING
> > > +#include <linux/rtc.h>
> > > +#include <linux/time.h>
> > > +#endif
> > > +
> > >  static HLIST_HEAD(binder_deferred_list);
> > >  static DEFINE_MUTEX(binder_deferred_lock);
> > >  
> > > @@ -591,8 +596,104 @@ struct binder_transaction {
> > >  	 * during thread teardown
> > >  	 */
> > >  	spinlock_t lock;
> > > +#ifdef CONFIG_BINDER_USER_TRACKING
> > > +	struct timespec timestamp;
> > > +	struct timeval tv;
> > > +#endif
> > >  };
> > >  
> > > +#ifdef CONFIG_BINDER_USER_TRACKING
> > > +
> > > +/*
> > > + * binder_print_delay - Output info of a delay transaction
> > > + * @t:          pointer to the over-time transaction
> > > + */
> > > +static void binder_print_delay(struct binder_transaction *t)
> > > +{
> > > +	struct rtc_time tm;
> > > +	struct timespec *startime;
> > > +	struct timespec cur, sub_t;
> > > +
> > > +	ktime_get_ts(&cur);
> > > +	startime = &t->timestamp;
> > > +	sub_t = timespec_sub(cur, *startime);
> > > +
> > > +	/* if transaction time is over than 2 sec,
> > > +	 * show timeout warning log.
> > > +	 */
> > > +	if (sub_t.tv_sec < 2)
> > > +		return;
> > > +
> > > +	rtc_time_to_tm(t->tv.tv_sec, &tm);
> > > +
> > > +	spin_lock(&t->lock);
> > > +	pr_info_ratelimited("%d: from %d:%d to %d:%d",
> > > +			    t->debug_id,
> > > +			    t->from ? t->from->proc->pid : 0,
> > > +			    t->from ? t->from->pid : 0,
> > > +			    t->to_proc ? t->to_proc->pid : 0,
> > > +			    t->to_thread ? t->to_thread->pid : 0);
> > > +	spin_unlock(&t->lock);
> > > +
> > > +	pr_info_ratelimited(" total %u.%03ld s code %u start %lu.%03ld android %d-%02d-%02d %02d:%02d:%02d.%03lu\n",
> > > +			    (unsigned int)sub_t.tv_sec,
> > > +			    (sub_t.tv_nsec / NSEC_PER_MSEC),
> > > +			    t->code,
> > > +			    (unsigned long)startime->tv_sec,
> > > +			    (startime->tv_nsec / NSEC_PER_MSEC),
> > > +			    (tm.tm_year + 1900), (tm.tm_mon + 1), tm.tm_mday,
> > > +			    tm.tm_hour, tm.tm_min, tm.tm_sec,
> > > +			    (unsigned long)(t->tv.tv_usec / USEC_PER_MSEC));
> > > +}
> > 
> > Ick, why not use a tracepoint for this instead?
> > 
> > And what is userspace supposed to do with this if they see it?
> 
> Or another option is to implement this separately outside of binder.c using
> register_trace_* on the existing binder tracepoints, similar to what say the
> block tracer or preempt-off tracers do. Call it, say, "binder-latency tracer".
> 
> That way all of this tracing code is in-kernel but outside of binder.c.
> 
> thanks,
> 
>  - Joel
> 
Time limitation of recording is the reason why we don't use tracepoint.
In some situations, the exception is caused by a series of transactions
interaction.
Some abnormal transactions may be pending for a long time ago, they
could not be recorded due to buffer limited.
Therefore, it is difficult to dig out the root causes which caused by
the earlier transactions occurred.

Another point is that we'd just like to record the abnormal
transactions.
But most of transactions are less than 2 seconds, these are not the key
point we need to focus on.

thanks,

Frankie