Re: [PATCH] linux-user,bsd-user: re-exec with G_SLICE=always-malloc

[Emilio Cota <cota@braap.org>]

On Thu, Dec 01, 2022 at 10:49:27 +0000, Alex Bennée wrote:
> Emilio Cota <cota@braap.org> writes:
> > On Tue, Oct 04, 2022 at 13:00:47 +0100, Daniel P. Berrangé wrote:
> > (snip)
> >> Can't say I especially like this but I'm out of other ideas for how
> >> to guarantee a solution. Users can't set env vars prior to launching
> >> QEMU user emulators when using binfmt.
> >
> > An alternative is to not use GSlice between fork/exec. I'm
> > not sure if within that region there are other users besides
> > GTree (GArray perhaps?), but if there aren't, then just using
> > a different binary tree implementation should do.
> 
> Hmm my distros version of GArray certainly does and that is used quite
> heavily across gdbstub and plugins.

Then we might have to also import a GSlice-free GArray ("QArray").
Currently we just deadlock on POSIX-compliant code, which is
unacceptable.

> > Untested patches using ccan's AVL tree: 
> >   https://github.com/cota/qemu/commits/avl
> >
> > Would that be more palatable?
> 
> I think generally we wouldn't want to have multiple implementations
> unless there was a definite benefit (c.f. QHT). That said I think
> Richard's latest optimisation work:
> 
>   Subject: [PATCH v2 0/7] accel/tcg: Rewrite user-only vma tracking
>   Date: Thu, 27 Oct 2022 22:12:51 +1100
>   Message-Id: <20221027111258.348196-1-richard.henderson@linaro.org>
> 
> brings in the kernel's interval tree (with unit tests). I wonder if the
> page_collection use of GTree could be converted to that?

Thanks. I looked into reusing this but I don't think it's a drop-in
replacement for GTree.

> I don't know how you would defend against re-introducing it into
> linux-user though aside from commentary.

To close the loop:
I've sent a patch series that imports GTree-sans-GSlice as QTree,
and uses that for TCG:
  https://lists.gnu.org/archive/html/qemu-devel/2023-01/msg02080.html

Thanks,
		Emilio