Re: [PATCH] KVM: x86: fix undefined behavior in bit shift for __feature_bit

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Oct 31, 2022 at 05:42:56PM +0000, Sean Christopherson wrote:
> On Mon, Oct 31, 2022, Gaosheng Cui wrote:
> > Shifting signed 32-bit value by 31 bits is undefined, so changing
> > significant bit to unsigned. The UBSAN warning calltrace like below:
> > 
> > UBSAN: shift-out-of-bounds in arch/x86/kvm/reverse_cpuid.h:101:11
> > left shift of 1 by 31 places cannot be represented in type 'int'
> 
> PeterZ is contending that this isn't actually undefined behavior given how the
> kernel is compiled[*].  That said, I would be in favor of replacing the open-coded
> shift with BIT() to make the code a bit more self-documenting, and that would
> naturally fix this maybe-undefined-behavior issue. 
> 
> [*] https://lore.kernel.org/all/Y1%2FAaJOcgIc%2FINtv@hirez.programming.kicks-ass.net

I'm definitely in favour of updating this code; both your suggestion and
hpa's suggestion look like sane changes. But I do feel that whatever
UBSAN thing generated this warning needs to be fixed too.

I'm fine with the compiler warning about this code -- but it must not
claim undefined behaviour given the compiler flags we use.