Re: [PATCH] KVM: selftests: Make guest_snprintf() a macro and use it as appropriate

[Sean Christopherson <seanjc@google.com>]

On Wed, Sep 30, 2026, sashiko-bot@kernel.org wrote:
> Closes: https://lore.kernel.org/all/20260918113631.634233-4-gokul02k@gmail.com
> 
> > diff --git a/tools/testing/selftests/kvm/include/test_util.h b/tools/testing/selftests/kvm/include/test_util.h
> > index a6a3e16578959..817d4d5ca7cc7 100644
> > --- a/tools/testing/selftests/kvm/include/test_util.h
> > +++ b/tools/testing/selftests/kvm/include/test_util.h
> > @@ -238,7 +238,17 @@ static inline u32 atoi_non_negative(const char *name, const char *num_str)
> >  }
> >  
> >  int guest_vsnprintf(char *buf, int n, const char *fmt, va_list args);
> > -__printf(3, 4) int guest_snprintf(char *buf, int n, const char *fmt, ...);
> > +
> > +#define guest_snprintf(__buf, __n, __fmt)		\
> > +({							\
> > +	va_list va;					\
> > +	int len;					\
> 
> [Severity: Medium]
> Does this macro shadow caller arguments? If a caller passes an argument
> named len to guest_snprintf(), it will be shadowed by the uninitialized
> local len due to C scoping rules for statement expressions. The inner len
> is then passed to guest_vsnprintf(), which could result in a random or
> uninitialized buffer size.
> 
> > +							\
> > +	va_start(va, __fmt);				\
> 
> [Severity: Medium]
> Does this implicitly capture the caller's variadic arguments? 
> 
> The macro internally invokes va_start(va, __fmt), which couples it to the
> caller's stack frame. If this macro is used in a non-variadic function,
> or if __fmt is not the rightmost named argument of the calling function,
> could va_start cause undefined behavior? 

Ugh, and now I remember why we ended up with duplicate code.  Lame.  I'll drop
this and grab Gokul's original patch.  As much as I want to keep guest_snprintf()
around "just in case", in practice I highly doubt we'll ever use it.