Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <Zuri5yKtN3TNmzNW@voyager>
Date: Wed, 18 Sep 2024 16:25:43 +0200
From: Markus Wichmann <nullplan@....net>
To: musl@...ts.openwall.com
Subject: Re: SIGSEGV/stack overflow in pthread_create - race condition?

Am Tue, Sep 17, 2024 at 02:02:50PM -0400 schrieb Rich Felker:
> On Tue, Sep 17, 2024 at 03:35:20PM +0200, Markus Wichmann wrote:
> > Am Tue, Sep 17, 2024 at 08:12:41AM -0400 schrieb Rich Felker:
> > > Yes, getting minimal annotation needed for the debugger to handle this
> > > right is an open todo item. Ideally it would be contingent on whether
> > > you enabled debugging so that it wouldn't end up in no-debug-info
> > > builds. Slimming down the asm source files to the minimum that
> > > actually have to be asm source files (like clone, I think), which is
> > > also a wishlist item, would make it less of an invasive change, which
> > > is probably why I've held off on digging into it.
> >
> > Would it not be enough to clear fp in the clone child? That should
> > initialize the frame pointer chain, right? OK, it is an actual
> > instruction more for everyone, but I doubt the impact would even be
> > measurable.
>
> Oh, maybe it's not a cfi issue at all, just failure to clear frame
> pointer? Fixing that would be perfectly acceptable.
>
> Rich

I should add that I don't know if that would fix it, because I don't
know ARM very well. However, the ARM CRT code does initialize fp and the
ARM __clone code does not.

Ciao,
Markus

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.