[gentoo-embedded] Re: (was) working hxgcc, related to uclibc

[Ned Ludd]

The code is not ready for any sort of public consumption yet. Please
hold off on sending it to any mailing lists yet. 
It does not use the DYNAMIC_LINKER correctly yet, it only handles i386,
and the macro's fully override the built in gcc which we don't want.
I'll give you another update over the weekened which can be tested with
uClibc.


On Fri, 2004-01-23 at 04:43, Peter S. Mazinger wrote:
> Some comments regarding uclibc:
>
> On 22 Jan 2004, Ned Ludd wrote:
>
> > pappy,
> >
> > I'm CC: peter so he can get a heads up on what we are thinking.
> >
> > Attachments included.
> >
> > This is also building on your box.
> > In the screen chroot but it's not in the not in an overlay.
> >
> > Also pappy to make Martin happy could you confirm the guard scan takes
> > place every single time gcc is built. You suggested the idea to him so a
> > patch to have the ebuild behave the way you said in a previous mail
> > would be great. The base-system herd wants this so -r5 can be pushed to
> > stable. I also want see 3.3.2 go stable because it will allow us get
> > everybody on the same page with basic toolchain pie support.
> >
> > ...
> > * Done with patching
> >  * Applying
> > gcc331-pp-fixup.patch...                                                                       [ ok ]
> >  * Applying
> > protector.dif...                                                                                                  [ ok ]
> >  * Applying
> > gcc-3.3.2-gentoo-branch-update-20031218-pie-ssp.patch...                                                          [ ok ]
>
> This patch has fixed crt1.o, uclibc works with pie also for crt0.o. One
> solution would be to patch uclibc so, that it copies crt0.o as crt1.o (as
> I did with Scrt1.o for the case CTOR_DTOR is not defined), then no change
> is needed to work with uclibc. I'll check with Erik if he will do the
> change to uclibc, if not this patch has to be somehow adapted to check for
> uClibc-config.h file.
>
> >  * Applying
> > gcc332-gentoo-branding.patch...                                                                                   [ ok ]
> >
> >  * This sys-libs/glibc has __guard object and __stack_smash_handler
> > functions
> >  * scanning the system for binaries with __guard - this may take 5-10
> > minutes
> >  * Please do not press ctrl-C or ctrl-Z during this period - it will
> > continue
> >
> >  * Scanning system for __guard@GCC symbols...
> >  *  Scanning 01 of 10 /lib...
> >  *  Scanning 02 of 10 /usr/lib...
>
> The ssp_scan is easy to implement for uClibc (check for ld-uClibc.so.0
> presence, and check libc.so.0), but how do we know, if uClibc is main
> libc, should we check first for ld-uClibc.so, then if it is present,
> check if also libc.so.6 is present, if yes, then glibc is main libc, else
> uClibc. The check is needed because propolice is an option (yet) in
> uClibc, so there could be versions not having __guard/__smash...
>
> Peter,
-- 
Ned Ludd <solar@g.o>
Gentoo Linux Developer