On Wed, May 11, 2022 at 09:21:16AM +0200, Peter Zijlstra wrote:
> On Wed, May 11, 2022 at 05:27:46AM +0300, Kirill A. Shutemov wrote:
> > +#define __untagged_addr(addr, n) \
> > + ((__force __typeof__(addr))sign_extend64((__force u64)(addr), n))
> > +
> > +#define untagged_addr(addr) ({ \
> > + u64 __addr = (__force u64)(addr); \
> > + if (__addr >> 63 == 0) { \
> > + if (current->thread.features & X86_THREAD_LAM_U57) \
> > + __addr &= __untagged_addr(__addr, 56); \
> > + else if (current->thread.features & X86_THREAD_LAM_U48) \
> > + __addr &= __untagged_addr(__addr, 47); \
> > + } \
> > + (__force __typeof__(addr))__addr; \
> > +})
>
> Assuming you got your bits in hardware order:
>
> u64 __addr = addr;
> if ((s64)__addr >= 0) {
> int lam = (current->thread.features >> X86_THREAD_LAM_U57) & 3;
That needs a _BIT suffix or something, same in the previous reply.
> if (lam)
> __addr &= sign_extend64(__addr, 65 - 9*lam);
> }
> __addr;
>
> has less branches on and should definitely result in better code (or I
> need more morning juice).
I definitely needs more morning juice :-)
