Re: [PATCH 0/2] livepatch: Add compiler optimization disclaimer/docs

[Miroslav Benes <mbenes@suse.cz>]

Hi,

On Fri, 31 May 2024, Joe Lawrence wrote:

> On 5/31/24 07:23, Miroslav Benes wrote:
> > Hi,
> > 
> > On Tue, 21 Jul 2020, Joe Lawrence wrote:
> > 
> >> In light of [PATCH] Revert "kbuild: use -flive-patching when
> >> CONFIG_LIVEPATCH is enabled" [1], we should add some loud disclaimers
> >> and explanation of the impact compiler optimizations have on
> >> livepatching.
> >>
> >> The first commit provides detailed explanations and examples.  The list
> >> was taken mostly from Miroslav's LPC talk a few years back.  This is a
> >> bit rough, so corrections and additional suggestions welcome.  Expanding
> >> upon the source-based patching approach would be helpful, too.
> >>
> >> The second commit adds a small README.rst file in the livepatch samples
> >> directory pointing the reader to the doc introduced in the first commit.
> >>
> >> I didn't touch the livepatch kselftests yet as I'm still unsure about
> >> how to best account for IPA here.  We could add the same README.rst
> >> disclaimer here, too, but perhaps we have a chance to do something more.
> >> Possibilities range from checking for renamed functions as part of their
> >> build, or the selftest scripts, or even adding something to the kernel
> >> API.  I think we'll have a better idea after reviewing the compiler
> >> considerations doc.
> > 
> > thanks to Marcos for resurrecting this.
> > 
> > Joe, do you have an updated version by any chance? Some things have 
> > changed since July 2020 so it calls for a new review. If there was an 
> > improved version, it would be easier. If not, no problem at all.
> > 
> 
> Yea, it's been a little while :) I don't have any newer version than
> this one.  I can rebase,  apply all of the v1 suggestions, and see where
> it stands.  LMK if you can think of any specifics that could be added.

I will walk through the patches first to see if there is something which 
can/should be changed given the development since then.

> For example, CONFIG_KERNEL_IBT will be driving some changes soon,
> whether it be klp-convert for source-based patches or vmlinux.o binary
> comparison for kpatch-build.

True.

> I can push a v2 with a few changes, but IIRC, last time we reviewed
> this, it kinda begged the question of how someone is creating the
> livepatch in the first place.  As long as we're fine holding that
> thought for a while longer, this doc may still be useful by itself.

If I remember correctly, the conclusion was that this doc was beneficial 
on its own.

Miroslav