On Tue, Mar 13, 2012 at 07:42:17AM +0100, Gerd Hoffmann wrote: > On 03/12/12 16:50, Alon Levy wrote: > > On Mon, Mar 12, 2012 at 01:43:11PM +0200, Alon Levy wrote: > >> On Mon, Mar 12, 2012 at 11:20:55AM +0100, Gerd Hoffmann wrote: > >>> On 03/11/12 20:26, Alon Levy wrote: > >>>> dprint is still used for qxl_init_common one time prints. > >>> > >>> I think we shouldn't simply convert the dprintf's into trace-points. > >>> > >>> We should look at each dprintf and check whenever it makes sense at all, > >>> whenever it makes sense at that place before converting it over to a > >>> tracepoint. > > > > I'll also add qxl_spice_* trace points for the next patch. Does that > > sound excessive? you could just trace the qxl_io_write to get the io > > itself, or trace just qxl_spice_* to get the qxl<->spice interface, or > > both (qxl_*). > > Makes sense to place trace points systematically like that. >
What about having the frequent (read: too frequent to use stderr to dump them since they clutter the screen, unless you 'stop' before each monitor command) have a postfix "_freq"? This is a stopgap, but helpful one, you can then do: trace-event qxl* on trace-event qxl*freq off Instead of remembering / having conveniently ready a longer list: trace-event qxl* on trace-event qxl_interface_get_command_enter off trace-event qxl_interface_release_resource off trace-event qxl_interface_get_command_ret off trace-event qxl_push_free_res off > cheers, > Gerd > >
