[PATCH v4] clocksource: arch_timer: Fix code to use physical timers when requested

[Catalin Marinas]

Doug,

On Thu, Nov 20, 2014 at 04:24:09PM +0000, Doug Anderson wrote:
> On Thu, Nov 20, 2014 at 8:10 AM, Catalin Marinas
> <catalin.marinas at arm.com> wrote:
> > On Wed, Oct 08, 2014 at 08:38:57AM +0100, Sonny Rao wrote:
> >> This is a bug fix for using physical arch timers when
> >> the arch_timer_use_virtual boolean is false.  It restores the
> >> arch_counter_get_cntpct() function after removal in
> >>
> >> 0d651e4e "clocksource: arch_timer: use virtual counters"
> >>
> >> We need this on certain ARMv7 systems which are architected like this:
> >>
> >> * The firmware doesn't know and doesn't care about hypervisor mode and
> >>   we don't want to add the complexity of hypervisor there.
> >>
> >> * The firmware isn't involved in SMP bringup or resume.
> >>
> >> * The ARCH timer come up with an uninitialized offset between the
> >>   virtual and physical counters.  Each core gets a different random
> >>   offset.
> >>
> >> * The device boots in "Secure SVC" mode.
> >>
> >> * Nothing has touched the reset value of CNTHCTL.PL1PCEN or
> >>   CNTHCTL.PL1PCTEN (both default to 1 at reset)
> >>
> >> One example of such as system is RK3288 where it is much simpler to
> >> use the physical counter since there's nobody managing the offset and
> >> each time a core goes down and comes back up it will get reinitialized
> >> to some other random value.
> >>
> >> Fixes: 0d651e4e65e9 ("clocksource: arch_timer: use virtual counters")
> >> Cc: stable at vger.kernel.org
> >> Signed-off-by: Sonny Rao <sonnyrao at chromium.org>
> >> Acked-by: Olof Johansson <olof at lixom.net>
> > [...]
> >> --- a/arch/arm64/include/asm/arch_timer.h
> >> +++ b/arch/arm64/include/asm/arch_timer.h
> >> @@ -135,6 +135,16 @@ static inline void arch_timer_evtstrm_enable(int divider)
> >>  #endif
> >>  }
> >>
> >> +static inline u64 arch_counter_get_cntpct(void)
> >> +{
> >> +     u64 cval;
> >> +
> >> +     isb();
> >> +     asm volatile("mrs %0, cntpct_el0" : "=r" (cval));
> >> +
> >> +     return cval;
> >> +}
> >
> > Sorry but I have to NAK the arm64 changes here. If the firmware is
> > broken and does not initialise CNTVOFF properly, please fix it (at least
> > on ARMv8 hardware). Also, on arm64 the vdso gettimeofday()
> > implementation relies on using the virtual counter, so correct
> > initialisation of CNTVOFF is essential.
> 
> Sonny's patch here just makes it so that we honor the global variable.
> My patch at <https://patchwork.kernel.org/patch/5051881/> is the one
> that allows the global variable to be set.  You can see in that patch
> that it's impossible for the variable to be set on ARM64.

It just gives people ideas ;), thinking they only need to remove
IS_ENABLED(CONFIG_ARM) in your patch and get this working on arm64.

> In previous discussions it was agreed that on ARM64 psci (or something
> similar) was a requirement anyway and that gave us a way to get the
> firmware involved again if we ever need to bring down a processor and
> bring it back up in the kernel.  PSCI is not a requirement for ARM32.
> There are systems that don't get the firmware involved when a
> processor loses state (like if it is powered off and powered on again,
> maybe for suspend/resume) and there was pushback against the kernel
> itself transitioning into monitor mode to init CNTVOFF in these cases.
> People agreed a month ago that these two patches were a reasonable
> approach for ARM32.

I'm not complaining about about arm32 here, just the arm64
implementation. If you want to avoid #ifdefs in the arch timer driver,
what about, for arm64, defining something like:

static inline u64 arch_counter_get_cntpct(void)
{
	/*
	 * AArch64 kernel and user space mandate the use of CNTVCT.
	 */
	BUG();
	return 0;
}

-- 
Catalin