Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 25 Mar 2020 19:13:39 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Linus Torvalds <torvalds@linux-foundation.org>
Cc:     Andy Lutomirski <luto@amacapital.net>,
        the arch/x86 maintainers <x86@kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Steven Rostedt <rostedt@goodmis.org>,
        Masami Hiramatsu <mhiramat@kernel.org>,
        Daniel Bristot de Oliveira <bristot@redhat.com>,
        Jason Baron <jbaron@akamai.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@kernel.org>, Nadav Amit <namit@vmware.com>,
        Peter Anvin <hpa@zytor.com>,
        Andrew Lutomirski <luto@kernel.org>,
        Ard Biesheuvel <ard.biesheuvel@linaro.org>,
        Josh Poimboeuf <jpoimboe@redhat.com>
Subject: Re: [RESEND][PATCH v3 14/17] static_call: Add static_cond_call()
Message-ID: <20200325181339.GK20713@hirez.programming.kicks-ass.net>
References: <CAHk-=wiumU4QxAkT+GqhBt5f-iUsoLNC0sqVfRKp0xypA6aNYg@mail.gmail.com>
 <86D80EA7-9087-4042-8119-12DD5FCEAA87@amacapital.net>
 <CAHk-=wim-2aaFi_GNs5KmX4ykkgQjnRo5D4B9ZK+1Oz=kpp_2A@mail.gmail.com>
 <20200324170306.GU20696@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200324170306.GU20696@hirez.programming.kicks-ass.net>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Mar 24, 2020 at 06:03:06PM +0100, Peter Zijlstra wrote:
> On Tue, Mar 24, 2020 at 09:33:21AM -0700, Linus Torvalds wrote:

> > Of course, one alternative is to just say "instead of using NOP, use
> > 'xorl %eax,%eax'", and then we'd have the rule that a NULL conditional
> > function returns zero (or NULL).
> > 
> > I _think_ a "xorl %eax,%eax ; retq" is just three bytes and would fit
> > in the tailcall slot too.
> 
> Correct. The only problem is that our text patching machinery can't
> replace multiple instructions :/

To clarify; the problem is a task getting preempted with its RIP at the
RET. Then when we rewrite the text to be a CALL/JMP.d32 it will read
garbage (1 byte into the displacement of the instruction) instead of a
RET when it resumes.

Now, there are ways to fix this, the easiest being calling
synchronize_rcu_tasks() just like optprobes does (see also commit
5c02ece81848 ("x86/kprobes: Fix ordering while text-patching")).

It would mean patching a call away from NULL will be 'expensive' but it
ought to work.

I'll try and do the patch, see what it looks like.