netdev - Re: [PATCH 0/2] vhost: improve livepatch switching for heavily loaded vhost worker kthreads

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <Y9qdWzOfiUZP2dET@FVFF77S0Q05N>
Date:   Wed, 1 Feb 2023 17:11:55 +0000
From:   Mark Rutland <mark.rutland@....com>
To:     Josh Poimboeuf <jpoimboe@...nel.org>
Cc:     Petr Mladek <pmladek@...e.com>,
        Joe Lawrence <joe.lawrence@...hat.com>, kvm@...r.kernel.org,
        "Michael S. Tsirkin" <mst@...hat.com>,
        Peter Zijlstra <peterz@...radead.org>, netdev@...r.kernel.org,
        Jiri Kosina <jikos@...nel.org>, linux-kernel@...r.kernel.org,
        virtualization@...ts.linux-foundation.org,
        "Seth Forshee (DigitalOcean)" <sforshee@...italocean.com>,
        live-patching@...r.kernel.org, Miroslav Benes <mbenes@...e.cz>
Subject: Re: [PATCH 0/2] vhost: improve livepatch switching for heavily
 loaded vhost worker kthreads

On Wed, Feb 01, 2023 at 08:57:27AM -0800, Josh Poimboeuf wrote:
> On Wed, Feb 01, 2023 at 11:10:20AM +0000, Mark Rutland wrote:
> > On Tue, Jan 31, 2023 at 08:38:32AM -0800, Josh Poimboeuf wrote:
> > > On Tue, Jan 31, 2023 at 10:22:09AM +0000, Mark Rutland wrote:
> > > > > Hm, it might be nice if our out-of-line static call implementation would
> > > > > automatically do a static key check as part of static_call_cond() for
> > > > > NULL-type static calls.
> > > > > 
> > > > > But the best answer is probably to just add inline static calls to
> > > > > arm64.  Is the lack of objtool the only thing blocking that?
> > > > 
> > > > The major issues were branch range limitations (and needing the linker to add
> > > > PLTs),
> > > 
> > > Does the compiler do the right thing (e.g., force PLT) if the branch
> > > target is outside the translation unit?  I'm wondering if we could for
> > > example use objtool to help enforce such rules at the call site.
> > 
> > It's the linker (rather than the compiler) that'll generate the PLT if the
> > caller and callee are out of range at link time. There are a few other issues
> > too (e.g. no guarnatee that the PLT isn't used by multiple distinct callers,
> > CMODX patching requirements), so we'd have to generate a pseudo-PLT ourselves
> > at build time with a patching-friendly code sequence. Ard had a prototype for
> > that:
> > 
> >   https://lore.kernel.org/linux-arm-kernel/20211105145917.2828911-1-ardb@kernel.org/
> > 
> > ... but that was sufficiently painful that we went with the current static key
> > approach:
> > 
> >   https://lore.kernel.org/all/20211109172408.49641-1-mark.rutland@arm.com/
> 
> Thanks for the background.
> 
> Was there a reason for putting it out-of-line rather than directly in
> _cond_resched()?

I think that's mostly a historical accident; I'm not aware of a reaason we
can't put that directly in _cond_resched(). 

Since we started from out-of-line static call trampolines, even the out-of-line
static key check looked nicer, and I think we just never considered moving the
static key check inline.

> If it were inline then it wouldn't be that much different from the
> static called version and I wonder if we could simplify by just using
> the static key for all PREEMPT_DYNAMIC configs.

That would be nice!

> > > > If we knew each call-site would only call a particular function or skip the
> > > > call, then we could do better (and would probably need something like objtool
> > > > to NOP that out at compile time), but since we don't know the callee at build
> > > > time we can't ensure we have a PLT in range when necessary.
> > > 
> > > Unfortunately most static calls have multiple destinations.
> > 
> > Sure, but here we're just enabling/disabling a call, which we could treat
> > differently, or wrap at a different level within the scheduler code. I'm happy
> > to take a look at that.
> 
> I can try to emulate what you did for PREEMPT_DYNAMIC.  I'll Cc you on
> my actual patch to come soon-ish.

I look forward to it! :)

> > > And most don't have the option of being NULL.
> > 
> > Oh, I was under the impression that all could be disabled/skipped, which is
> > what a NULL target implied.
> 
> I guess what I was trying to say is that if the target can be NULL, the
> call site has to use static_call_cond() to not break the
> !HAVE_STATIC_CALL case.  But most call sites use static_call().

Ah, sorry -- I had missed that we had distinct static_call_cond() and
static_call() helpers.

Thanks,
Mark.