Re: [PATCH v4] livepatch: Cleanup module page permission changes

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Fri, Dec 04, 2015 at 10:57:45PM +0100, Jiri Kosina wrote:
> On Fri, 4 Dec 2015, Josh Poimboeuf wrote:
> 
> > > (1) I pull your 'modules-next' branch, apply this patch on top, and wait 
> > >     for your merge with Linus and send merge request afterwards
> > > (2) If you are okay with rebasing your tree (seems like this is 
> > >     ocassionally happening), how about you prepare a branch that'd have 
> > >     just b3212ec77 ("module: keep percpu symbols in module's symtab") on 
> > >     top of some common base, I merge it, and the cross-dependency is gone
> > > (3) I cherry-pick b3212ec77 ("module: keep percpu symbols in 
> > >     module's symtab") from your tree, and apply this on top. git will 
> > >     handle duplicate commits when Linus merges both just fine
> > > (4) I sign this patch off and you merge it
> > > 
> > > (4) seems really outside the regular process. (1) is really tricky wrt. 
> > > coordination of timing during the merge window. I'd prefer (2) (more 
> > > git-ish way of doing things, but would require you rebasing your tree) or 
> > > eventually (3) (git will handle this with grace).
> > 
> > [ off-list ]
> 
> :-)
> 
> > Quick question.  Just curious, because I'm new at this...
> > 
> > My impression was that #1 was standard operating procedure.  Merge a
> > (non-rebasable) modules branch into livepatch, and then make sure to
> > submit the livepatch pull request after Rusty sends his, with a note in
> > the mail to Linus stating the dependency.  That seems pretty
> > straightforward to me.  Or am I missing something?
> 
> It's one of the options, yes. The only drawback is that it introduces, in 
> addition to the actual code cross-dependency, also maintainer timing 
> cross-dependency, and it might easily go wrong during merge window. But 
> I've done this quite a few times already, and it was rather smooth.
> 
> What I actually prefer doing in this case is have a common merge base as a 
> separate branch that gets merged to both trees, and then it's not really 
> important who merges first. But that'd require in-advance planning and 
> structuring Rusty's tree for that, and that's probably not worth the 
> hassle for these few patches.

Ah, got it.  That does sound better, assuming there's some advance
planning.  Thanks for educating me :-)

-- 
Josh