Re: [PATCH v25 10/16] trace-cmd: Add time sync protocol flags

[Tzvetomir Stoyanov <tz.stoyanov@gmail.com>]

On Thu, Dec 3, 2020 at 4:09 AM Steven Rostedt <rostedt@goodmis.org> wrote:
>
> On Thu, 29 Oct 2020 13:18:10 +0200
> "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote:
>
> > Added 32bit flags for the time synchronization protocols. The first added
> > flag is TRACECMD_TSYNC_FLAG_INTERPOLATE, used to specify how the
> > timestamps must be corrected.
> >  - If the flag is set, an interpolation is performed:
> >    Find the (min, max) interval from the offsets array and calculate
> > offset specific to the given timestamp using interpolation in that
> > interval.
> >  - If the flag is not set, do not interpolate:
> >    Find the (min, max) interval from the offsets array and use the
> >    min offset for all timespamps within the interval.
>
>     "timestamps"
>
> >
> > These flags are set by the timestamp synchronization protocols at the
> > protocol initialization time.
> >
> > Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
> > ---
> > --- a/tracecmd/trace-tsync.c
> > +++ b/tracecmd/trace-tsync.c
> > @@ -132,7 +132,8 @@ out:
> >  static void write_guest_time_shift(struct buffer_instance *instance)
> >  {
> >       struct tracecmd_output *handle;
> > -     struct iovec vector[5];
> > +     struct iovec vector[6];
> > +     unsigned int flags;
> >       long long *scalings = NULL;
> >       long long *offsets = NULL;
> >       long long *ts = NULL;
> > @@ -145,6 +146,9 @@ static void write_guest_time_shift(struct buffer_instance *instance)
> >                                        &ts, &offsets, &scalings);
> >       if (ret < 0 || !count || !ts || !offsets || !scalings)
> >               return;
> > +     ret = tracecmd_tsync_get_proto_flags(&instance->tsync, &flags);
> > +     if (ret < 0)
> > +             return;
> >
> >       file = instance->output_file;
> >       fd = open(file, O_RDWR);
> > @@ -154,14 +158,16 @@ static void write_guest_time_shift(struct buffer_instance *instance)
> >       vector[0].iov_len = 8;
> >       vector[0].iov_base = &top_instance.trace_id;
> >       vector[1].iov_len = 4;
> > -     vector[1].iov_base = &count;
> > -     vector[2].iov_len = 8 * count;
> > -     vector[2].iov_base = ts;
> > +     vector[1].iov_base = &flags;
> > +     vector[2].iov_len = 4;
> > +     vector[2].iov_base = &count;
> >       vector[3].iov_len = 8 * count;
> > -     vector[3].iov_base = offsets;
> > +     vector[3].iov_base = ts;
> >       vector[4].iov_len = 8 * count;
> > -     vector[4].iov_base = scalings;
> > -     tracecmd_add_option_v(handle, TRACECMD_OPTION_TIME_SHIFT, vector, 5);
> > +     vector[4].iov_base = offsets;
> > +     vector[5].iov_len = 8 * count;
> > +     vector[5].iov_base = scalings;
> > +     tracecmd_add_option_v(handle, TRACECMD_OPTION_TIME_SHIFT, vector, 6);
> >       tracecmd_append_options(handle);
> >       tracecmd_output_close(handle);
> >  #ifdef TSYNC_DEBUG
>
> To make the above cleaner, I would use an enum to define the vector
> indexes:
>
> enum {
>         VECTOR_TRACE_ID,
>         VECTOR_FLAGS,
>         VECTOR_COUNT,
>         VECTOR_TIMES,
>         VECTOR_OFFSETS,
>         VECTOR_SCALINGS,
> }
>
> And then you can make it:
>
>         vector[VECTOR_TRACE_ID].iov_len = 8;
>         vector[VECTOR_TRACE_ID].iov_base = &top_instance.trace_id;
>         vector[VECTOR_FLAGS].iov_len = 4;
>         vector[VECTOR_FLAGS].iov_base = &flags;
>         vector[VECTOR_COUNT].iov_len = 4;
>         vector[VECTOR_COUNT].iov_base = &count;
>         vector[VECTOR_TIMES].iov_len = 8 * count;
>         vector[VECTOR_TIMES].iov_base = ts;
>         vector[VECTOR_OFFSETS].iov_len = 8 * count;
>         vector[VECTOR_OFFSETS].iov_base = offsets;
>         vector[VECTOR_SCALINGS].iov_len = 8 * count;
>         vector[VECTOR_SCALINGS].iov_base = scalings;
>
> It makes it obvious what each vector is used for.
>
> This is just an opinion. You don't need to implement it. I just hate
> hard coded numbers ;-)

The code is changed in "trace-cmd: Add timestamp synchronization per
vCPU" patch,
mo more hard coded numbers. The vector size is dynamic, depending on
the VCPU count.

>
> -- Steve



-- 
Tzvetomir (Ceco) Stoyanov
VMware Open Source Technology Center