x86/alternative: PAUSE is not a NOP
authorPeter Zijlstra <peterz@infradead.org>
Wed, 14 Jun 2023 14:35:50 +0000 (16:35 +0200)
committerBorislav Petkov (AMD) <bp@alien8.de>
Wed, 14 Jun 2023 17:02:54 +0000 (19:02 +0200)
While chasing ghosts, I did notice that optimize_nops() was replacing
'REP NOP' aka 'PAUSE' with NOP2. This is clearly not right.

Fixes: 6c480f222128 ("x86/alternative: Rewrite optimize_nops() some")
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/linux-next/20230524130104.GR83892@hirez.programming.kicks-ass.net/
arch/x86/kernel/alternative.c

index bbfbf7a..a7e1ec5 100644 (file)
@@ -169,9 +169,12 @@ void text_poke_early(void *addr, const void *opcode, size_t len);
  */
 static bool insn_is_nop(struct insn *insn)
 {
-       if (insn->opcode.bytes[0] == 0x90)
+       /* Anything NOP, but no REP NOP */
+       if (insn->opcode.bytes[0] == 0x90 &&
+           (!insn->prefixes.nbytes || insn->prefixes.bytes[0] != 0xF3))
                return true;
 
+       /* NOPL */
        if (insn->opcode.bytes[0] == 0x0F && insn->opcode.bytes[1] == 0x1F)
                return true;