On Thu, Mar 18, 2021 at 06:11:05PM +0100, Peter Zijlstra wrote:
> --- a/arch/x86/kernel/alternative.c
> +++ b/arch/x86/kernel/alternative.c
> @@ -345,19 +345,39 @@ recompute_jump(struct alt_instr *a, u8 *
> static void __init_or_module noinline optimize_nops(struct alt_instr *a, u8
> *instr)
> {
> unsigned long flags;
> - int i;
> + int nops = 0, i = 0;
> + struct insn insn;
> + u8 *nop = NULL;
>
> - for (i = 0; i < a->padlen; i++) {
> - if (instr[i] != 0x90)
> + do {
> + if (insn_decode_kernel(&insn, &instr[i]))
> return;
> - }
>
> - local_irq_save(flags);
> - add_nops(instr + (a->instrlen - a->padlen), a->padlen);
> - local_irq_restore(flags);
> + if (insn.length == 1 && insn.opcode.bytes[0] == 0x90) {
> + if (!nop) {
> + nop = &instr[i];
> + nops = 1;
> + } else {
> + nops++;
> + }
> + }
> + i += insn.length;
> +
> + if ((insn.length != 1 || i == a->instrlen) && nop) {
> +
> + local_irq_save(flags);
> + add_nops(nop, nops);
> + local_irq_restore(flags);
> +
> + DUMP_BYTES(instr, a->instrlen, "%px: [%d:%d) optimized
> NOPs: ",
> + instr, (int)(unsigned long)(nop-instr),
> nops);
> +
> + nop = NULL;
> + }
>
> - DUMP_BYTES(instr, a->instrlen, "%px: [%d:%d) optimized NOPs: ",
> - instr, a->instrlen - a->padlen, a->padlen);
> + } while (i < a->instrlen);
> +
> + WARN_ON_ONCE(nop);
> }
I think I've made this simpler; pasting the whole function and not the
diff because former is easier to read:
/*
* "noinline" to cause control flow change and thus invalidate I$ and
* cause refetch after modification.
*
* Jump over the non-NOP insns, the remaining bytes must be single-byte NOPs,
* optimize them.
*/
static void __init_or_module noinline optimize_nops(struct alt_instr *a, u8
*instr)
{
unsigned long flags;
struct insn insn;
int i = 0, j;
/* Skip preceding non-NOP instructions. */
do {
if (insn_decode_kernel(&insn, &instr[i]))
return;
if (insn.length == 1 && insn.opcode.bytes[0] == 0x90)
break;
i += insn.length;
} while (i < a->instrlen);
if (i >= a->instrlen - 1)
return;
/* Verify rest is NOPs - should not fire(tm) */
for (j = i; j < a->instrlen - 1; j++) {
if (WARN(instr[j] != 0x90, "Wrong insn byte 0x%hx at 0x%px\n",
instr[j], &instr[j]))
return;
}
local_irq_save(flags);
add_nops(&instr[i], a->instrlen - i);
local_irq_restore(flags);
DUMP_BYTES(instr, a->instrlen, "%px: [%d:%d) optimized NOPs: ",
instr, i, a->instrlen);
}
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
