Re: [coreboot] r813 - coreboot-v3/lib

[Carl-Daniel Hailfinger]

On 24.08.2008 14:03, Stefan Reinauer wrote:
> Carl-Daniel Hailfinger wrote:
> > On 24.08.2008 13:29, [e-mail] wrote:
> > > Author: stepan
> > > Date: 2008-08-24 13:29:11 +0200 (Sun, 24 Aug 2008)
> > > New Revision: 813
> > >
> > > Modified:
> > >    coreboot-v3/lib/lar.c
> > > Log:
> > > needs to be const, or the segment checker will bail out. (trivial)
> > > Signed-off-by: Stefan Reinauer 
> > > Acked-by: Stefan Reinauer
> > > Modified: coreboot-v3/lib/lar.c
> > >
> > > --- coreboot-v3/lib/lar.c	2008-08-24 07:08:17 UTC (rev 812)
> > > +++ coreboot-v3/lib/lar.c	2008-08-24 11:29:11 UTC (rev 813)
> > > @@ -31,7 +31,7 @@
> > >  #define ntohl(x) (x)
> > >  #endif
> > >
> > > -static const char *algo_name[] = {
> > > +static const char * const algo_name[] = {
> > >    "none",
> > >    "lzma",
> > >    "nrv2b",
> >
> > Interesting. The checker didn't complain here. Which gcc are you using?
> > I'm using gcc 4.2.1.
>
> This happened on a box with 3.4.6

OK, that old gcc version is not smart enough to to see that nobody
writes to algo_name. It also doesn't have -funit-at-a-time on by default
which would make it smarter.
> > By the way, can we officially require at least gcc 4.1? Otherwise,
> > compiling RAMinit won't work reliably if at all.
>
> What's the problem with pre-4.1?

-funit-at-a-time is off by default before 4.1.
__attribute__((externally_visible)) doesn't work before 4.1. -combine
seems to be a new feature of 4.1 and we rely on it to get a reliably
working initram.

Regards,
Carl-Daneil