Re: [PATCH 1/2] x86/hyperv: Move hv_do_rep_hypercall to asm-generic

[Wei Liu]

On Sun, Apr 18, 2021 at 02:42:55PM +0000, Michael Kelley wrote:
> From: Wei Liu <wei.liu@xxxxxxxxxx> Sent: Sunday, April 18, 2021 6:09 AM
> > On Fri, Apr 16, 2021 at 05:43:02PM -0700, Joseph Salisbury wrote:
> > > From: Joseph Salisbury <joseph.salisbury@xxxxxxxxxxxxx>
> > >
> > > This patch makes no functional changes.  It simply moves hv_do_rep_hypercall()
> > > out of arch/x86/include/asm/mshyperv.h and into asm-generic/mshyperv.h
> > >
> > > hv_do_rep_hypercall() is architecture independent, so it makes sense that it
> > > should be in the architecture independent mshyperv.h, not in the x86-specific
> > > mshyperv.h.
> > >
> > > This is done in preperation for a follow up patch which creates a consistent
> > > pattern for checking Hyper-V hypercall status.
> > >
> > > Signed-off-by: Joseph Salisbury <joseph.salisbury@xxxxxxxxxxxxx>
> > > ---
> > [...]
> > > +static inline u64 hv_do_rep_hypercall(u16 code, u16 rep_count, u16 varhead_size,
> > > +				      void *input, void *output)
> > > +{
> > > +	u64 control = code;
> > > +	u64 status;
> > > +	u16 rep_comp;
> > > +
> > > +	control |= (u64)varhead_size << HV_HYPERCALL_VARHEAD_OFFSET;
> > > +	control |= (u64)rep_count << HV_HYPERCALL_REP_COMP_OFFSET;
> > > +
> > > +	do {
> > > +		status = hv_do_hypercall(control, input, output);
> > > +		if ((status & HV_HYPERCALL_RESULT_MASK) != HV_STATUS_SUCCESS)
> > > +			return status;
> > > +
> > > +		/* Bits 32-43 of status have 'Reps completed' data. */
> > > +		rep_comp = (status & HV_HYPERCALL_REP_COMP_MASK) >>
> > > +			HV_HYPERCALL_REP_COMP_OFFSET;
> > > +
> > > +		control &= ~HV_HYPERCALL_REP_START_MASK;
> > > +		control |= (u64)rep_comp << HV_HYPERCALL_REP_START_OFFSET;
> > > +
> > > +		touch_nmi_watchdog();
> > 
> > This seems to be missing in Arm. Does it compile?
> > 
> > Wei.
> 
> touch_nmi_watchdog() is defined as "static inline" in include/linux/nmi.h.  So
> it should be present in all architectures.  It calls arch_touch_nmi_watchdog,
> which is an empty function if CONFIG_HAVE_NMI_WATCHDOG is not defined,
> as is the case on ARM64.

I see. I couldn't find arch_touch_nmi_watchdog on Arm but missed the
stub function. Thanks for the information.

Wei.