On Wed, Jan 21, 2004 at 04:37:44PM +0100, Nicolas Joly wrote:
 [...]
> cd src/x11
> make build
> make install DESTDIR=/

....
rm -f SetSOwner.o.tmp
#   compile  static/SetSOwner.po
/usr/netbsd/current/src/obj.amd64/tooldir.NetBSD-1.6ZH-x86_64/bin/x86_64--netbsd-gcc -O2  -Werror    -DXTHREADS -D_REENTRANT -DXUSE_MTSAFE_API  -DXNO_MTSAFE_PWDAPI -DUSE_NBSD_THREADLIB -I. -DHAS_SNPRINTF -DLIBX11 -DPOSTLOCALELIBDIR=\"lib\" -DCSRG_BASED -DFUNCPROTO=15 -DNARROWPROTO -I/usr/X11R6/include   -nostdinc -isystem /usr/include -c    -pg /usr/netbsd/current/xsrc/xfree/xc/lib/X11/SetSOwner.c -o SetSOwner.po.tmp
[1]   Segmentation fault (core dumped) make realall
*** Error code 139


...Some points to note:

I have a /usr/netbsd/xsrc (not src/x11) where I built everything successfully
last night, and am running X from that build.

Second, the segfault happens at different times in the build, but is always
accompanied by a stream of NFS errors about "server not responding"
and "server alive again".  (Sources are on an NFS server since NetBSD does
not yet know how to do DMA with the nVidia pciide controller.  NFS is
almost exactly as fast as PIO for this machine, and has much lower CPU over-
head, so I use NFS.  (^&)

Third: The NFS freezes seem to be systemic since upgrading to -current
yesterday.  With the previous snapshot, hitting CapsLock would cause a
~1 second freeze of the entire system (as near as I can tell).  That bug
has gone away, now, but has been replaced by more random-seeming freezes.
(The freezes seem to generally be caused by filesystem, and/or NFS,
activity.  But it could be broader than that.)

Fourth...when I did the command as you had it above, I initially had a
problem with an "install" command that was unable to find the tooldir-built
"install" program.  (So maybe I had it 99% okay until I did a
"sh build.sh tools" and retried the "make build"?)  That failure was not
a signal, but was simply a "couldn't find .../nbinstall", so I did the
obvious thing of updating my tools.)


Another interesting side note: It used to be that when I (re)started
X, the server would have a pretty sharp picture of whatever I had on
the display before I last quit X.  Now, I see the old display is pepper
with hundreds, maybe thousands, of tiny dots, as if something was hitting
random places in memory while X was down.


Here's the gdb "bt" output from the resulting make.core in
...src/x11/lib/X11/static/obj.amd64:

Core was generated by `make'.
Program terminated with signal 11, Segmentation fault.
Reading symbols from /usr/libexec/ld.elf_so...(no debugging symbols found)...
done.
Loaded symbols for /usr/libexec/ld.elf_so
Reading symbols from /usr/lib/libc.so.12...(no debugging symbols found)...done.
Loaded symbols for /usr/lib/libc.so.12
#0  0x000000000041495b in free ()
(gdb) bt
#0  0x000000000041495b in free ()
#1  0x0000000000418533 in free ()
#2  0x0000000000418abe in free ()
#3  0x0000000000403c09 in free ()
#4  0x00000000004193cf in free ()
#5  0x0000000000404232 in free ()
#6  0x00000000004193cf in free ()
#7  0x000000000040410e in free ()
#8  0x0000000000404537 in free ()
#9  0x000000000040c8ed in free ()
#10 0x0000000000402528 in free ()

...I suppose that bad memory is a possible explanation, though I didn't
see this before the upgrade---and, again, I was able to do "make build"
yesterday in "xsrc".


-- 
  "I probably don't know what I'm talking about."  http://www.olib.org/~rkr/