Re: [Qemu-devel] [PATCH v2 1/2] qxl: switch qxl.c to trace-events

[Alon Levy <alevy@redhat.com>]

On Wed, Mar 14, 2012 at 10:03:06AM +0100, Gerd Hoffmann wrote:
>   Hi,
> 
> > dprint is still used for qxl_init_common one time prints.
> > 
> > also switched parts of spice-display.c over, mainly all the callbacks to
> > spice server.
> > 
> > All qxl device trace events start with the qxl device id.
> 
> Looks pretty good overall, just some little nits left ...
> 
> 
> > @@ -948,8 +969,9 @@ static void qxl_vga_ioport_write(void *opaque, uint32_t addr, uint32_t val)
> >      VGACommonState *vga = opaque;
> >      PCIQXLDevice *qxl = container_of(vga, PCIQXLDevice, vga);
> >  
> > +    trace_qxl_io_write_vga(qxl->id, qxl_mode_to_string(qxl->mode), addr, val);
> 
> With this being added and logging qxl->mode ...
> 
> >      if (qxl->mode != QXL_MODE_VGA) {
> > -        dprint(qxl, 1, "%s\n", __FUNCTION__);
> > +        trace_qxl_vga_ioport_while_not_in_vga_mode(qxl->id);
> 
> ... this one isn't needed any more.

OK.

> 
> > @@ -1445,23 +1448,24 @@ static void qxl_send_events(PCIQXLDevice *d, uint32_t events)
> >          qxl_update_irq(d);
> >      } else {
> >          if (write(d->pipe[1], d, 1) != 1) {
> > -            dprint(d, 1, "%s: write to pipe failed\n", __FUNCTION__);
> > +            trace_qxl_send_events_write_to_pipe_failed(d->id);
> 
> I somehow feel like we can just drop this.  kernel pipe buffers are 4096
> at least, we are writing a single byte only, it is highly unlikely this
> ever happens and even it if does we just ignore the error anyway.
> 
> Have you ever seen that message in a log?

No. So I guess I'll remove the dprint as well (in a separate patch).

> 
> >  static void init_pipe_signaling(PCIQXLDevice *d)
> >  {
> > -   if (pipe(d->pipe) < 0) {
> > -       dprint(d, 1, "%s: pipe creation failed\n", __FUNCTION__);
> > -       return;
> > -   }
> > -   fcntl(d->pipe[0], F_SETFL, O_NONBLOCK);
> > -   fcntl(d->pipe[1], F_SETFL, O_NONBLOCK);
> > -   fcntl(d->pipe[0], F_SETOWN, getpid());
> > +    if (pipe(d->pipe) < 0) {
> > +        fprintf(stderr, "%s:%s: qxl pipe creation failed\n",
> > +                __FILE__, __func__);
> > +        exit(1);
> > +    }
> > +    fcntl(d->pipe[0], F_SETFL, O_NONBLOCK);
> > +    fcntl(d->pipe[1], F_SETFL, O_NONBLOCK);
> > +    fcntl(d->pipe[0], F_SETOWN, getpid());
> >  
> > -   qemu_thread_get_self(&d->main);
> > -   qemu_set_fd_handler(d->pipe[0], pipe_read, NULL, d);
> > +    qemu_thread_get_self(&d->main);
> > +    qemu_set_fd_handler(d->pipe[0], pipe_read, NULL, d);
> >  }
> 
> Intention fixup, would be nice as separate patch.

No problem.

> 
> > --- a/ui/spice-display.c
> > +++ b/ui/spice-display.c
> 
> > +    trace_qxl_spice_add_memslot(ssd->qxl.id, memslot->slot_id,
> > +                                memslot->virt_start, memslot->virt_end,
> > +                                async);
> > +
> 
> > +    trace_qxl_spice_del_memslot(ssd->qxl.id, gid, sid);
> 
> > +    trace_qemu_spice_create_primary_surface(ssd->qxl.id, id, surface, async);
> 
> > +    trace_qemu_spice_destroy_primary_surface(ssd->qxl.id, id, async);
> 
> I think these should all be qemu_spice_* for consistency.

hmm, ok - one point is that I was conflicted about renaming some of the
trace points to have more namespace-y names, like
trace_qemu_spice_surface_primary_create
instead of
trace_qemu_spice_create_primary_surface

but opted in the end to stay with the later since they match the
function names they are tracing, and there is no autocomplete yet for
trace-events anyway, so globing is the only possible issue and it isn't
affected (as long as you remember to put enough asterisks).

> 
> cheers,
>   Gerd
>