Re: [PATCH v3 14/17] KVM: arm64: Macros for setting/clearing FGT bits

[Marc Zyngier <maz@kernel.org>]

On Mon, 18 Dec 2023 09:56:32 +0000,
Fuad Tabba <tabba@google.com> wrote:
> 
> Hi Marc,
> 
> On Mon, Dec 18, 2023 at 9:40 AM Marc Zyngier <maz@kernel.org> wrote:
> >
> > On Thu, 14 Dec 2023 10:01:54 +0000,
> > Fuad Tabba <tabba@google.com> wrote:
> > >
> > > There's a lot of boilerplate code for setting and clearing FGT
> > > bits when activating guest traps. Refactor it into macros. These
> > > macros will also be used in future patch series.
> > >
> > > No functional change intended.
> > >
> > > Signed-off-by: Fuad Tabba <tabba@google.com>
> > > ---
> > >  arch/arm64/kvm/hyp/include/hyp/switch.h | 60 +++++++++----------------
> > >  1 file changed, 21 insertions(+), 39 deletions(-)
> > >
> > > diff --git a/arch/arm64/kvm/hyp/include/hyp/switch.h b/arch/arm64/kvm/hyp/include/hyp/switch.h
> > > index 17ce40f5b006..e223fc0d5193 100644
> > > --- a/arch/arm64/kvm/hyp/include/hyp/switch.h
> > > +++ b/arch/arm64/kvm/hyp/include/hyp/switch.h
> > > @@ -79,6 +79,23 @@ static inline void __activate_traps_fpsimd32(struct kvm_vcpu *vcpu)
> > >               clr |= ~hfg & __ ## reg ## _nMASK;                      \
> > >       } while(0)
> > >
> > > +#define update_fgt_traps_cs(reg, clr, set)                           \
> > > +     do {                                                            \
> > > +             struct kvm_cpu_context *hctxt =                         \
> > > +                     &this_cpu_ptr(&kvm_host_data)->host_ctxt;       \
> > > +             u64 val, c = 0, s = 0;                                  \
> > > +                                                                     \
> > > +             ctxt_sys_reg(hctxt, reg) = read_sysreg_s(SYS_ ## reg);  \
> > > +             compute_clr_set(vcpu, reg, c, s);                       \
> >
> > You are referring to a variable name that is in the scope of the macro
> > user, and not the macro itself. It is so fragile it isn't funny.
> >
> > Why don't you simply pass the vcpu as a parameter to the function?
> >
> > Another thing is that this read/write can be expensive. How about not
> > doing anything when there is no change to the value of the sysreg?
> 
> What do you think of this (spacing will be fixed):
> 
> #define update_fgt_traps_cs(vcpu, reg, clr, set)                         \
>     do {
>                            \
>         struct kvm_cpu_context *hctxt =
>          \
>             &this_cpu_ptr(&kvm_host_data)->host_ctxt;                  \
>         u64 val, c = 0, s = 0;
>                   \
> 
>                              \
>         ctxt_sys_reg(hctxt, reg) = read_sysreg_s(SYS_ ## reg);   \
>         compute_clr_set(vcpu, reg, c, s);
>           \
>         val = __ ## reg ## _nMASK;
>           \
>         val |= (s | set);
>                         \
>         val &= ~(c | clr);
>                        \
>         if (ctxt_sys_reg(hctxt, reg) != val)
>               \
>             write_sysreg_s(val, SYS_ ## reg);
>            \
>     } while(0)
> 
> If it looks good to you, I'll fix it on the respin.

This is what I have on top at the moment:

diff --git a/arch/arm64/kvm/hyp/include/hyp/switch.h b/arch/arm64/kvm/hyp/include/hyp/switch.h
index 27077e1f7de2..d56fef44dc31 100644
--- a/arch/arm64/kvm/hyp/include/hyp/switch.h
+++ b/arch/arm64/kvm/hyp/include/hyp/switch.h
@@ -79,22 +79,26 @@ static inline void __activate_traps_fpsimd32(struct kvm_vcpu *vcpu)
 		clr |= ~hfg & __ ## reg ## _nMASK; 			\
 	} while(0)
 
-#define update_fgt_traps_cs(reg, clr, set)				\
+#define update_fgt_traps_cs(vcpu, reg, clr, set)			\
 	do {								\
 		struct kvm_cpu_context *hctxt =				\
 			&this_cpu_ptr(&kvm_host_data)->host_ctxt;	\
-		u64 val, c = 0, s = 0;					\
+		u64 c = 0, s = 0;					\
 									\
 		ctxt_sys_reg(hctxt, reg) = read_sysreg_s(SYS_ ## reg);	\
 		compute_clr_set(vcpu, reg, c, s);			\
-		val = __ ## reg ## _nMASK;				\
-		val |= (s | set);					\
-		val &= ~(c | clr);					\
-		write_sysreg_s(val, SYS_ ## reg);			\
+		s |= set;						\
+		c |= clr;						\
+		if (c || s) {						\
+			u64 val = __ ## reg ## _nMASK;			\
+			val |= s;					\
+			val &= ~c;					\
+			write_sysreg_s(val, SYS_ ## reg);		\
+		}							\
 	} while(0)
 
-#define update_fgt_traps(reg)						\
-	update_fgt_traps_cs(reg, 0, 0)
+#define update_fgt_traps(vcpu, reg)		\
+	update_fgt_traps_cs(vcpu, reg, 0, 0)
 
 /*
  * Validate the fine grain trap masks.
@@ -171,9 +175,9 @@ static inline void __activate_traps_hfgxtr(struct kvm_vcpu *vcpu)
 	if (!vcpu_has_nv(vcpu) || is_hyp_ctxt(vcpu))
 		return;
 
-	update_fgt_traps(HFGITR_EL2);
-	update_fgt_traps(HDFGRTR_EL2);
-	update_fgt_traps(HDFGWTR_EL2);
+	update_fgt_traps(vcpu, HFGITR_EL2);
+	update_fgt_traps(vcpu, HDFGRTR_EL2);
+	update_fgt_traps(vcpu, HDFGWTR_EL2);
 
 	if (cpu_has_amu())
 		update_fgt_traps(vcpu, HAFGRTR_EL2);

If that seems sensible to you, I'll fold it in, as I'd like to do
without a respin if we can avoid it.

`	M.

-- 
Without deviation from the norm, progress is not possible.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel