On Tue, Feb 6, 2018 at 2:52 PM, Linus Torvalds
[off-list ref] wrote:

On Tue, Feb 6, 2018 at 1:37 PM, Dan Williams [off-list ref] wrote:

quoted

At that point we're basically just back to the array_ptr() version
that returned a sanitized pointer to an array element.

.. that one does an extra unnecessary 'andq' instead of the duplicated
cmp.  But at least it avoids comparing that 32-bit integer twice, so
it's probably slightly smaller.

(And your code generation is without the "r" -> "ir" fix for the size argument)

Probably doesn't matter. But a "asm goto" would give you at least
potentially optimal code.

Should we go with array_element_nospec() in the meantime? So we're not
depending on jump labels? With the constraint fix and killing that
superfluous AND the assembly is now:

     e26:       48 81 fd 4d 01 00 00    cmp    $0x14d,%rbp
     e2d:       48 19 d2                sbb    %rdx,%rdx
                        NR_syscalls);
        if (likely(call))
     e30:       48 21 d0                and    %rdx,%rax
     e33:       74 1e                   je     e53 <do_syscall_64+0x73>
                regs->ax = (*call)(regs->di, regs->si, regs->dx,
     e35:       48 8b 4b 38             mov    0x38(%rbx),%rcx
     e39:       48 8b 53 60             mov    0x60(%rbx),%rdx
     e3d:       48 8b 73 68             mov    0x68(%rbx),%rsi
     e41:       48 8b 7b 70             mov    0x70(%rbx),%rdi
     e45:       4c 8b 4b 40             mov    0x40(%rbx),%r9
     e49:       4c 8b 43 48             mov    0x48(%rbx),%r8
     e4d:       ff 10                   callq  *(%rax)
     e4f:       48 89 43 50             mov    %rax,0x50(%rbx)
     e53:       65 48 8b 04 25 00 00    mov    %gs:0x0,%rax