[RFC] Clean up sparc64 timecounters

[Super Bisquit]

On Wed, Jun 29, 2011 at 4:56 PM, Marius Strobl <marius at alchemy.franken.de>wrote:

> On Tue, Jun 28, 2011 at 01:37:52PM -0400, Jung-uk Kim wrote:
> > On Tuesday 28 June 2011 01:26 pm, Jung-uk Kim wrote:
> > > Can you please review the attached patch?
> > >
> > > sys/sparc64/pci/fire.c:
> > > - Remove redundant timecounter masking from tc_get_timecount
> > > method. - Remove an unnecessary macro for timecounter mask.
> > > - Remove a redundant NULL assignment.
> > >
> > > sys/sparc64/pci/schizo.c:
> > > - Remove redundant timecounter masking from tc_get_timecount
> > > method. - Correct timecounter mask.  Note this is a no-op because
> > > the STX_CTRL_PERF_CNT_CNT0_SHIFT is actually zero.
> > > - Remove a redundant NULL assignment.
>
> I'm not sure whether you correctly understand how that timer works.
> The hardware actually provides a pair of 32-bit timers which are
> read via a single 64-bit register so the existing tc_counter_mask
> is correct and your change is wrong. For the same reason the masking
> and shifting in schizo_get_timecount() only happens to be unnecessary
> in so far as we currently use the lower 32-bit counter and the
> tc_get_timecount methods return u_int. If we'd either switch to
> the upper 32-bit counter or the timecounter code would be enhanced
> to support up to 64-bit counters it wouldn't be redundant. There's
> actually a right-shift missing in schizo_get_timecount() though,
> i.e. it should actually do:
>        return ((SCHIZO_CTRL_READ_8(sc, STX_CTRL_PERF_CNT) &
>            (STX_CTRL_PERF_CNT_MASK << STX_CTRL_PERF_CNT_CNT0_SHIFT) >>
>            STX_CTRL_PERF_CNT_CNT0_SHIFT);
> The compiler should be smart enough to boil all that down to a
> single 64-bit to 32-bit conversion when returning though.
> For similar reasons I'd prefer to also keep the masking in
> fire_get_timecount(), besides using the macro IMO is cleaner
> than using ~0u(l).
>
> >
> > @@ -686,8 +684,7 @@ fire_attach(device_t dev)
> >                 if (tc == NULL)
> >                         panic("%s: could not malloc timecounter",
> __func__);
> >                 tc->tc_get_timecount = fire_get_timecount;
> > -               tc->tc_poll_pps = NULL;
> > -               tc->tc_counter_mask = TC_COUNTER_MAX_MASK;
> > +               tc->tc_counter_mask = ~0ul;
> >                                       ^^^^
> >                                       ~0u
> >                 if (OF_getprop(OF_peer(0), "clock-frequency", &prop,
> >                     sizeof(prop)) == -1)
> >                         panic("%s: could not determine clock frequency",
> >
>
> Well, if you really remove the masking from fire_get_timecount()
> then you should actually also use ~0ul here for consistency as it's
> an 64-bit counter.
>
> Marius
>
> _______________________________________________
> freebsd-sparc64 at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-sparc64
> To unsubscribe, send any mail to "freebsd-sparc64-unsubscribe at freebsd.org"
>


I'm curious as to how this would improve performance on the overall SPARC64
port or is it only for one model?