Re: [PATCH v2 1/2] clocksource/drivers/hyperv: add data structure for reference TSC MSR

[Anirudh Rayabharam]

On Thu, Nov 03, 2022 at 02:24:05PM +0000, Wei Liu wrote:
> On Wed, Nov 02, 2022 at 08:33:31PM +0000, Michael Kelley (LINUX) wrote:
> > From: Michael Kelley (LINUX) <mikelley@xxxxxxxxxxxxx> Sent: Thursday, October 27, 2022 6:43 AM
> > > From: Anirudh Rayabharam <anrayabh@xxxxxxxxxxxxxxxxxxx> Sent: Thursday,
> > > October 27, 2022 2:57 AM
> > > >
> > > > Add a data structure to represent the reference TSC MSR similar to
> > > > other MSRs. This simplifies the code for updating the MSR.
> > > >
> > > > Signed-off-by: Anirudh Rayabharam <anrayabh@xxxxxxxxxxxxxxxxxxx>
> > > > ---
> > > >  drivers/clocksource/hyperv_timer.c | 28 ++++++++++++++--------------
> > > >  include/asm-generic/hyperv-tlfs.h  |  9 +++++++++
> > > >  2 files changed, 23 insertions(+), 14 deletions(-)
> > > >
> > > > diff --git a/drivers/clocksource/hyperv_timer.c
> > > b/drivers/clocksource/hyperv_timer.c
> > > > index bb47610bbd1c..11332c82d1af 100644
> > > > --- a/drivers/clocksource/hyperv_timer.c
> > > > +++ b/drivers/clocksource/hyperv_timer.c
> > > > @@ -395,25 +395,25 @@ static u64 notrace read_hv_sched_clock_tsc(void)
> > > >
> > > >  static void suspend_hv_clock_tsc(struct clocksource *arg)
> > > >  {
> > > > -	u64 tsc_msr;
> > > > +	union hv_reference_tsc_msr tsc_msr;
> > > >
> > > >  	/* Disable the TSC page */
> > > > -	tsc_msr = hv_get_register(HV_REGISTER_REFERENCE_TSC);
> > > > -	tsc_msr &= ~BIT_ULL(0);
> > > > -	hv_set_register(HV_REGISTER_REFERENCE_TSC, tsc_msr);
> > > > +	tsc_msr.as_uint64 = hv_get_register(HV_REGISTER_REFERENCE_TSC);
> > > > +	tsc_msr.enable = 0;
> > > > +	hv_set_register(HV_REGISTER_REFERENCE_TSC, tsc_msr.as_uint64);
> > > >  }
> > > >
> > > >
> > > >  static void resume_hv_clock_tsc(struct clocksource *arg)
> > > >  {
> > > >  	phys_addr_t phys_addr = virt_to_phys(&tsc_pg);
> > > > -	u64 tsc_msr;
> > > > +	union hv_reference_tsc_msr tsc_msr;
> > > >
> > > >  	/* Re-enable the TSC page */
> > > > -	tsc_msr = hv_get_register(HV_REGISTER_REFERENCE_TSC);
> > > > -	tsc_msr &= GENMASK_ULL(11, 0);
> > > > -	tsc_msr |= BIT_ULL(0) | (u64)phys_addr;
> > > > -	hv_set_register(HV_REGISTER_REFERENCE_TSC, tsc_msr);
> > > > +	tsc_msr.as_uint64 = hv_get_register(HV_REGISTER_REFERENCE_TSC);
> > > > +	tsc_msr.enable = 1;
> > > > +	tsc_msr.pfn = __phys_to_pfn(phys_addr);
> > 
> > My previous review missed a problem here (and in the similar line below).
> > __phys_to_pfn() will return a PFN based on the guest page size, which might
> > be different from Hyper-V's page size that is always 4K.  This needs to be a
> > Hyper-V PFN, and we have virt_to_hvpfn() available to do just that, assuming
> > that function is safe to use here and in the case below. 
> 
> Anirudh, please take a look.

Just sent a fix for this using HVPFN_DOWN() which looks safe to use
here.

Thanks,
Anirudh.

> 
> I'm holding off sending hyperv-fixes to Linus for now.
> 
> Thanks,
> Wei.