Re: [GIT PULL] x86/urgent for v5.11-rc7

From: Andy Lutomirski <luto@amacapital.net>
To: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>,
	Miroslav Benes <mbenes@suse.cz>,
	Peter Zijlstra <peterz@infradead.org>,
	Josh Poimboeuf <jpoimboe@redhat.com>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Borislav Petkov <bp@suse.de>, Dave Hansen <dave.hansen@intel.com>,
	x86-ml <x86@kernel.org>, lkml <linux-kernel@vger.kernel.org>,
	Alexei Starovoitov <ast@kernel.org>,
	live-patching@vger.kernel.org
Subject: Re: [GIT PULL] x86/urgent for v5.11-rc7
Date: Tue, 9 Feb 2021 08:55:51 -0800	[thread overview]
Message-ID: <73175691-4AE1-496D-80D1-DC85AE1E9C27@amacapital.net> (raw)
In-Reply-To: <CAADnVQK-qdE6mHUogeaQo9Av_58cLptosmfUVmdMzW7gJn5UVw@mail.gmail.com>

> On Feb 9, 2021, at 8:45 AM, Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
> 
> On Tue, Feb 9, 2021 at 6:49 AM Steven Rostedt <rostedt@goodmis.org> wrote:
>> 
>>> On Tue, 9 Feb 2021 09:32:34 +0100 (CET)
>>> Miroslav Benes <mbenes@suse.cz> wrote:
>>> 
>>> powerpc has this
>>> 
>>> static inline unsigned long klp_get_ftrace_location(unsigned long faddr)
>>> {
>>>        /*
>>>         * Live patch works only with -mprofile-kernel on PPC. In this case,
>>>         * the ftrace location is always within the first 16 bytes.
>>>         */
>>>        return ftrace_location_range(faddr, faddr + 16);
>>> }
>>> 
>>>>> I suppose the trivial fix is to see if it points to endbr64 and if so,
>>>>> increment the addr by the length of that.
>>>> 
>>>> I thought of that too. But one thing that may be possible, is to use
>>>> kallsym. I believe you can get the range of a function (start and end of
>>>> the function) from kallsyms. Then ask ftrace for the addr in that range
>>>> (there should only be one).
>>> 
>>> And we can do this if a hard-coded value live above is not welcome. If I
>>> remember correctly, we used to have exactly this in the old versions of
>>> kGraft. We walked through all ftrace records, called
>>> kallsyms_lookup_size_offset() on every record's ip and if the offset+ip
>>> matched faddr (in this case), we returned the ip.
>> 
>> Either way is fine. Question is, should we just wait till CET is
>> implemented for the kernel before making any of these changes? Just knowing
>> that we have a solution to handle it may be good enough for now.
> 
> I think the issue is more fundamental than what appears on the surface.
> According to endbr64 documentation it's not just any instruction.
> The cpu will wait for it and if it's replaced with int3 or not seen at
> the branch target the cpu will throw an exception.
> If I understood the doc correctly it means that endbr64 can never be
> replaced with a breakpoint. If that's the case text_poke_bp and kprobe
> need to do extra safety checks.

Ugh.

Or we hack up #CP to handle this case. I don’t quite know how I feel about this.