Re: [PATCH v10 7/7] x86/tsc: use tsc early

From: Feng Tang
Date: Wed Jun 20 2018 - 06:27:42 EST


On Wed, Jun 20, 2018 at 12:16:36PM +0200, Thomas Gleixner wrote:
> On Wed, 20 Jun 2018, Feng Tang wrote:
> > On Wed, Jun 20, 2018 at 01:52:10AM +0200, Thomas Gleixner wrote:
> > >
> > > static DEFINE_STATIC_KEY_FALSE(__use_tsc);
> > > +static DEFINE_STATIC_KEY_TRUE(tsc_early_enabled);
> >
> > One potential problem may be the several static_keys used here,
> > the "__use_tsc", the "__sched_clock_stable", it may not be used
> > very early in boot phase. As the the static_branch_enable() will
> > use pageing related code while the paging is not setup ready yet.
>
> I know how static keys work and thats the reason for having the extra
> conditional. The key is disabled at a point where paging is available.

Maybe I'm wrong, for printk timestamp, it uses the local_clock() thus
sched_clock_cpu()
sched_clock_stable
check the static_key "__sched_clock_stable" which is
defined FALSE, and need a static_branch_enable() to
take effect

Thanks,
Feng

>
> > > /*
> > > * Scheduler clock - returns current time in nanosec units.
> > > */
> > > u64 native_sched_clock(void)
> > > {
> > > - if (static_branch_likely(&__use_tsc)) {
> > > - u64 tsc_now = rdtsc();
> > > + if (static_branch_likely(&__use_tsc))
> > > + return cycles_2_ns(rdtsc());
> > >
> > > - /* return the value in ns */
> > > - return cycles_2_ns(tsc_now);
> > > + if (static_branch_unlikely(&tsc_early_enabled)) {
> > > + if (tsc_early_sched_clock)
> > > + return cycles_2_ns(rdtsc());
> > > }
>
> Thanks,
>
> tglx