Re: [PATCH 08/15] x86/alternatives: Teach text_poke_bp() to emulate instructions

[Linus Torvalds]

On Fri, Jun 7, 2019 at 10:34 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> I was/am lazy and didn't want to deal with:
>
> arch/x86/include/asm/nops.h:#define GENERIC_NOP5_ATOMIC NOP_DS_PREFIX,GENERIC_NOP4
> arch/x86/include/asm/nops.h:#define K8_NOP5_ATOMIC 0x66,K8_NOP4
> arch/x86/include/asm/nops.h:#define K7_NOP5_ATOMIC NOP_DS_PREFIX,K7_NOP4
> arch/x86/include/asm/nops.h:#define P6_NOP5_ATOMIC P6_NOP5

Ugh. Maybe we could just pick one atomic sequence, and not have the
magic atomic nops be dynamic.

It's essentially what STATIC_KEY_INIT_NOP #define seems to do anyway.

NOP5_ATOMIC is already special, and not used for the normal nop
rewriting, only for kprobe/jump_label/ftrace.

So I suspect we could just replace all cases of

   ideal_nops[NOP_ATOMIC5]

with

   STATIC_KEY_INIT_NOP

and get rid of the whole "let's optimize the atomic 5-byte nop" entirely.

Hmm?

By definition, NOP_ATOMIC5 is just a single nop anyway, it's not used
for the potentially more complex alternative rewriting cases.

                Linus