On Sun, 21 Mar 2010 07:00, alexbestms@ wrote:
Garrett Cooper schrieb am 2010-03-21:
On Sat, Mar 20, 2010 at 5:55 PM, Alexander Best <alexbes...@wwu.de>
wrote:
ok. i think i finally solved this riddle. the cause for the problem
seems to
have been my CPUTYPE in /etc/make.conf. it is set to 'native'.
actually i've
been using the 'native' keyword for years now and never had any
problems with
it, but it seems a recent commit broke 'native' as CPUTYPE. for me
this is
100% reproducable:
1. put 'CPUTYPE = native' in /etc/make.conf
2. build and install gnu/usr.bin/cc
3. do 'buildkernel' or 'buildworld' and observe the segfault. for
some reason
this always occurs in lib/libc/string/strlen.c (r205108). i've
tested this
with older version of libc.so (built 22. Feb) and got the same
result. so i
assume this is not a libc problem, but a problem with gcc tripping
over some
code in libc. i have no clue however why this happend just now and
not
earlier. i don't think there has been a recent commit to gcc.
to solve this there are two ways:
1. set CPUTYPE to 'nocona' (i'm running amd64). this will let you
compile
everything just fine even with a gcc that has itself been built
with 'CPUTYPE
= native'.
2. build and install gnu/usr.bin/cc with 'CPUTYPE = nocona'. the
gcc version
that has been built this way will compile everything just fine even
with
'CPUTYPE = native'. the only way to break this is to go and compile
gcc again
with the CPUTYPE set to 'native'.
so to summarize: compiling gnu/usr.bin/cc with CPUTYPE set to
'native' will
give you a broken gcc!
What does -march=native yield in your case? There haven't been
any
recent commits to gcc, so I'm not sure whether or not that's the
issue. The libraries that I provided could have just been built from
a
sane system -- maybe it's something else that needs to be explored a
bit more closely to root cause the issue.
i've experimented with setting CPUTYPE to native yesterday, but still couldn't
figure out what the cause of it is. only a few points i'd like to point out:
1. this is very easily reproducible for me. i just need to set CPUTYPE=native
in my /etc/make.conf and everything that gets linked against libc and uses the
strlen() function won't compile due to gcc segfaulting. this is the case with
/usr/src/bin/cat e.g. as well as kernel, world and probably lots of other
stuff.
also the following gcc command segfaults too (no need for setting
CPUTYPE=native in this case, because -mtune gets set manually):
gcc -v -x c -E -mtune=native /dev/null -o /dev/null 2>&1
2. there seems to be a connection with strlen.c because gcc alaways segfaults
here. also i've been using CPUTYPE=native for years now and never had any
problems with it. for me the segfault always happens in:
#0 strlen (str=Variable "str" is not available.
) at /usr/src/lib/libc/string/strlen.c:100
100 va = (*lp - mask01);
it would be nice if people with arch i386 and amd64 could try to reproduce
this (i believe the other archs don't support CPUTYPE=native). again the
easiest way to trigger this (you don't need to edit your /etc/make.conf for
this) should be running:
gcc -v -x c -E -mtune=native /dev/null -o /dev/null 2>&1
for now i'm using the attached patch to prevent myself from shooting me in the
foot again. ;)
cheers.
alex
Cheers,
-Garrett
Native is equal to CPUTYPE not being defined right ?
So if the above is true why would you set CPUTYPE to native in the first
place ? when you can just leave it unset and its equal to native.
Nowhere in /usr/share/examples/etc/make.conf do I see a referance to this
type of CPUTYPE so unless I am totally missing the point of this thread, I
do not see a problem that has to be fixed.
With all due respect,
--
jhell
_______________________________________________
freebsd-current@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
