Re: [RFC][PATCH] x86,nospec: Simplify {JMP,CALL}_NOSPEC

From: Peter Zijlstra <peterz@infradead.org>
To: "Maciej S. Szmigiero" <mail@maciej.szmigiero.name>
Cc: linux-kernel@vger.kernel.org, keescook@chromium.org,
	hjl.tools@gmail.com, andrew.cooper3@citrix.com,
	mark.rutland@arm.com, will@kernel.org, ndesaulniers@google.com,
	x86@kernel.org, Ankur Arora <ankur.a.arora@oracle.com>,
	joao@overdrivepizza.com
Subject: Re: [RFC][PATCH] x86,nospec: Simplify {JMP,CALL}_NOSPEC
Date: Wed, 20 Jul 2022 11:12:04 +0200	[thread overview]
Message-ID: <YtfG5NYAgfTzboid@worktop.programming.kicks-ass.net> (raw)
In-Reply-To: <32be2ea4-b60b-fbca-10c2-8c06f5bc135b@maciej.szmigiero.name>

On Wed, Jul 20, 2022 at 02:01:39AM +0200, Maciej S. Szmigiero wrote:
> > Obviously I meant: apply_retpolines() ...
> 
> Will apply_retpolines() actually patch in that trailing int3 in
> the X86_FEATURE_RETPOLINE_LFENCE case?
> 
> Looking at its code it uses just ordinary NOPs as fill:
> > 	/*
> > 	 * For RETPOLINE_LFENCE: prepend the indirect CALL/JMP with an LFENCE.
> > 	 */
> > 	if (cpu_feature_enabled(X86_FEATURE_RETPOLINE_LFENCE)) {
> > 		bytes[i++] = 0x0f;
> > 		bytes[i++] = 0xae;
> > 		bytes[i++] = 0xe8; /* LFENCE */
> > 	}
> > 
> > 	ret = emit_indirect(op, reg, bytes + i);
> > 	if (ret < 0)
> > 		return ret;
> > 	i += ret;
> > 
> > 	for (; i < insn->length;)
> > 		bytes[i++] = BYTES_NOP1;

There is no space for int3 in that case. You get 3 bytes for LFENCE and
{2,3} bytes for 'jmp *%reg', which fully consumes the {5,6} bytes
available.

There will be no nops added.

But this is what all regular retpolines get to look like.

The plan was; and that's still pending, to get the INT3 from the AMD BTC
mitigation that adds INT3 after regular JMPs but those compiler patches
still need to happen I think.