> I think that perhaps this commit could be combined with a previous one.
> Another comment below
>
I think so, there is no reason to have patch and refactoring on the same patch series.
> On Thu, 2015-08-13 at 16:25 +0100, Frediano Ziglio wrote:
> > From: Marc-André Lureau <marcandre.lureau@xxxxxxxxx>
> >
> > ---
> > server/red_record_qxl.c | 12 ++++++++++++
> > server/red_record_qxl.h | 2 ++
> > server/red_worker.c | 24 ++++--------------------
> > 3 files changed, 18 insertions(+), 20 deletions(-)
> >
> > diff --git a/server/red_record_qxl.c b/server/red_record_qxl.c
> > index 481eb16..a72a200 100644
> > --- a/server/red_record_qxl.c
> > +++ b/server/red_record_qxl.c
> > @@ -792,3 +792,15 @@ void red_record_dev_input_primary_surface_create(FILE
> > *fd,
> > write_binary(fd, "data", line_0 ? abs(surface->stride)*surface->height
> > : 0,
> > line_0);
> > }
> > +
> > +void red_record_event(FILE *fd, int what, uint32_t type, unsigned long ts)
> > +{
> > + static int counter = 0;
> > +
> > + // TODO: record the size of the packet in the header. This would make
> > + // navigating it much faster (well, I can add an index while I'm at
> > it..)
> > + // and make it trivial to get a histogram from a file.
> > + // But to implement that I would need some temporary buffer for each
> > event.
> > + // (that can be up to VGA_FRAMEBUFFER large)
> > + fprintf(fd, "event %d %d %u %lu\n", counter++, what, type, ts);
> > +}
> > diff --git a/server/red_record_qxl.h b/server/red_record_qxl.h
> > index 1be9e41..f7dcdc0 100644
> > --- a/server/red_record_qxl.h
> > +++ b/server/red_record_qxl.h
> > @@ -41,4 +41,6 @@ void red_record_cursor_cmd(FILE *fd, RedMemSlotInfo
> > *slots, int group_id,
> > void red_record_dev_input_primary_surface_create(
> > FILE *fd, QXLDevSurfaceCreate *surface, uint8_t
> > *line_0);
> >
> > +void red_record_event(FILE *fd, int what, uint32_t type, unsigned long
> > ts);
> > +
> > #endif
> > diff --git a/server/red_worker.c b/server/red_worker.c
> > index 0540249..98a39dd 100644
> > --- a/server/red_worker.c
> > +++ b/server/red_worker.c
> > @@ -167,7 +167,6 @@ static inline red_time_t timespec_to_red_time(struct
> > timespec *time)
> > return time->tv_sec * (1000 * 1000 * 1000) + time->tv_nsec;
> > }
> >
> > -#if defined(RED_WORKER_STAT) || defined(COMPRESS_STAT)
> > static clockid_t clock_id;
> >
> > typedef unsigned long stat_time_t;
> > @@ -180,6 +179,7 @@ static stat_time_t stat_now(void)
> > return ts.tv_nsec + ts.tv_sec * 1000 * 1000 * 1000;
> > }
> >
> > +#if defined(RED_WORKER_STAT) || defined(COMPRESS_STAT)
> > double stat_cpu_time_to_sec(stat_time_t time)
> > {
> > return (double)time / (1000 * 1000 * 1000);
> > @@ -5019,24 +5019,10 @@ static RedDrawable *red_drawable_new(void)
> > return red;
> > }
> >
> > -static void red_record_event(RedWorker *worker, int what, uint32_t type)
> > -{
> > - struct timespec ts;
> > - static int counter = 0;
> > -
> > - // TODO: record the size of the packet in the header. This would make
> > - // navigating it much faster (well, I can add an index while I'm at
> > it..)
> > - // and make it trivial to get a histogram from a file.
> > - // But to implement that I would need some temporary buffer for each
> > event.
> > - // (that can be up to VGA_FRAMEBUFFER large)
> > - clock_gettime(worker->record_clock_id, &ts);
> > - fprintf(worker->record_fd, "event %d %d %d %ld\n", counter++, what,
> > type,
> > - ts.tv_nsec + ts.tv_sec * 1000 * 1000 * 1000);
> > -}
> > -
> > static void red_record_command(RedWorker *worker, QXLCommandExt ext_cmd)
> > {
> > - red_record_event(worker, 0, ext_cmd.cmd.type);
> > + red_record_event(worker->record_fd, 0, ext_cmd.cmd.type, stat_now());
>
> Here we're now using stat_now() instead of getting the timestamp with :
>
> clock_gettime(worker->record_clock_id, &ts);
>
> stat_now() gets the timestamp by calling:
>
> clock_gettime(clock_id, &ts);
>
> where 'clock_id' is a global variable in red_worker.c
>
> This means that the 'worker->record_clock_id' variable is now completely
> unused.
>
>
Yes, and also are the same timer.
Could you have multiple red_workers in the same process ??
However I think that these timers are user in later patches but I agree should
be removed in this series.
> > +
> > switch (ext_cmd.cmd.type) {
> > case QXL_CMD_DRAW:
> > red_record_drawable(worker->record_fd, &worker->mem_slots,
> > ext_cmd.group_id,
> > @@ -11939,7 +11925,7 @@ static void worker_dispatcher_record(void *opaque,
> > uint32_t message_type, void *
> > RedWorker *worker = opaque;
> >
> > if (worker->record_fd) {
> > - red_record_event(worker, 1, message_type);
> > + red_record_event(worker->record_fd, 1, message_type, stat_now());
> > }
> > }
> >
> > @@ -12247,11 +12233,9 @@ SPICE_GNUC_NORETURN void *red_worker_main(void
> > *arg)
> > spice_assert(MAX_PIPE_SIZE > WIDE_CLIENT_ACK_WINDOW &&
> > MAX_PIPE_SIZE > NARROW_CLIENT_ACK_WINDOW); //ensure wakeup by
> > ack message
> >
> > -#if defined(RED_WORKER_STAT) || defined(COMPRESS_STAT)
> > if (pthread_getcpuclockid(pthread_self(), &clock_id)) {
> > spice_error("pthread_getcpuclockid failed");
> > }
> > -#endif
> >
> > red_init(worker, (WorkerInitData *)arg);
> >
>
>
>
Frediano
_______________________________________________
Spice-devel mailing list
Spice-devel@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/spice-devel