Re: [PATCH v5] x86: use builtins to read eflags

[Andrew Cooper]

On 18/03/2022 18:19, Linus Torvalds wrote:
> Side note and kind of related: we do have this in the kernel:
>
>   register unsigned long current_stack_pointer asm(_ASM_SP);
>   #define ASM_CALL_CONSTRAINT "+r" (current_stack_pointer)
>
> which *might* also solve the redzoning issue.

Sadly not.  https://godbolt.org/z/cGx74sKE3

Given:

int pushf(void)
{
    unsigned long register sp asm("rsp");
    unsigned long x, y;

    asm ("movq $1, %0" : "=m" (x));

    asm ("pushf\n\tpop %0": "=r" (y), "+r" (sp));

    return x + y;
}

the generated code is:

pushf:
        movq $1, -8(%rsp)
        pushf
        pop %rax
        addl    -8(%rsp), %eax
        ret

so the rsp clobber doesn't prevent the push/pop pair from trashing x in
the red zone.

The builtin does cause a stack frame to be fully set up, and x to be
allocated within it, rather than in the red zone.

Experimenting with rsp clobbers leads to https://godbolt.org/z/s9scxre19
which demonstrates (for gcc at least) it does change the position of
when a stack frame gets set up (in the case that there is a path not
otherwise needing a stack frame) but doesn't unilaterally force a stack
frame to be set up.

As such, I'm not sure how current_stack_pointer can work as intended in
all cases...

~Andrew