Re: [PATCH] x86: Implement Linear Address Masking support

[Thomas Gleixner]

On Thu, May 12 2022 at 15:07, Matthew Wilcox wrote:
> On Thu, May 12, 2022 at 01:01:07PM +0000, David Laight wrote:
>> > +static inline int64_t sign_extend64(uint64_t value, int index)
>> > +{
>> > +    int shift = 63 - index;
>> > +    return (int64_t)(value << shift) >> shift;
>> > +}
>> 
>> Shift of signed integers are UB.
>
> Citation needed.

I'll bite :)

C11/19: 6.5.7 Bitwise shift operators

  4 The result of E1 << E2 is E1 left-shifted E2 bit positions; vacated
    bits are filled with zeros. If E1 has an unsigned type, the value of
    the result is E1 × 2E2, reduced modulo one more than the maximum
    value representable in the result type. If E1 has a signed type and
    nonnegative value, and E1 × 2E2 is representable in the result type,
    then that is the resulting value; otherwise, the behavior is
    undefined.

This is irrelevant for the case above because the left shift is on an
unsigned integer. The interesting part is this:

  5 The result of E1 >> E2 is E1 right-shifted E2 bit positions. If E1
    has an unsigned type or if E1 has a signed type and a nonnegative
    value, the value of the result is the integral part of the quotient
    of E1/2E2.  If E1 has a signed type and a negative value, the
    resulting value is implementation-defined.

So it's not UB, it's implementation defined. The obvious choice is to
keep LSB set, i.e. arithmetic shift, what both GCC and clang do.

Thanks,

        tglx