System : CPU: Dual Opteron 240 Board: TYAN S2895 with BIOS 1.05 RAM: 6GB ( 4x 1GiB + 4x 512 MiB) cards: GF7600 GSin PCIe slot 2x NIC with Broadcom BCM5704 in PCI-X slots 1x NIC with Realtek 8169 in PCI slot 1x Areca HW RAID ARC-1160 in PCI-X slot with 16 disks connected to the card 1x SATA disk, connected directly to the board, from which I boot and where gentoo system is 1x DVD-RW Gentoo: /etc/make.profile« -> »/usr/portage/profiles/selinux/2007.0/amd64/hardened kernel: 2.6.23-hardened-r3 ( error happens with any other rev also ) System is updated regularly. Error description: Booting with 2.6.20* kernel works fine. Booting with 2.6.23* works up to a point when init should start /sbin/rc and then /sbin/rc fails. System is stuck at a point of a prompt and I can't login, since nothing is done in the background ( fs in /etc/fstab are not mounted, no service is started etc). I tried to start the system with kernel parameter "init=/bin/bash" and to check things manually and it worked. Bash started normally. I could mount filesestems manually etc, but when i tried to start /sbin/rc, it crashed. Since rc is simply a script that is interpreted by runscript, I have tried recompiling everything related ( baselayout, glibc, policycoreutils etc ) but result was exactly the same. Then I straced the /sbin/rc and it seems that everything works fine up to a point when runscript repeatedly mmaps and unmaps a small block, always of the same size ( so it always gets te same address). On one of the attempts it gets SIGSEGV when trying to unmap the block it just successfully mmapped. It seems as some kind of race condition in multisocket NUMA system. Practically same SW configuration works fine on unisocket dual-core X2 6000+ system ( hardened-sources-2.6.23-r1) I will append in subequent post the output of "strace -o stracelog.txt /sbin/rc default" and "emerge --info"... Reproducible: Always
Created attachment 137775 [details] emerge --info output
Created attachment 137777 [details] ouptut of "strace -o logfile.txt /sbin/rc default"
I have just tried gentoo-sources-2.6.23-r3. It works.
I need your /proc/config.gz (or .config) of hardened-sources.
Created attachment 139434 [details] config of hardened-sources-2.6.23-r5 here is the config of the latest hardened-sources-2.6.23 i have tried ( -r5), with same results as with any previous attempt...
Maybe pipacs has an idea. I certainly don't.
I am having the same issues as the previous posters. In order to resolve the issue, I switched to gcc-vanilla (using gcc-config), ran source /etc/profile and then recompiled the latest kernel (2.6.23-hardened-r4). Lo and behold, upon booting, no issues whatsoever. -Noah Jaehnert
(In reply to comment #0) > Then I straced the /sbin/rc and it seems that everything works fine up to a > point when runscript repeatedly mmaps and unmaps a small block, always of the > same size ( so it always gets te same address). On one of the attempts it gets > SIGSEGV when trying to unmap the block it just successfully mmapped. you could gather more information by running rc inside gdb: 1. in the shell: gdb /sbin/rc 2. in gdb: run default this will run rc under the control of gdb and will cause gdb to break in at the time the segfault happens. then you can issue further commands (for best results you should have debug info stored, not stripped): 1. x/8i $pc 2. x/8x $sp 3. info reg 4. bt you can also suspend gdb and then get the /proc/<pid of rc>/maps file which will be helpful to see the memory layout of rc.
(In reply to comment #8) > (In reply to comment #0) Sorry, I didn't have the time to check that. In any case, after several rounds worth of system updates it works now. I have no idea what was the cause ( kernel, baselayout, sme lib or something else )...
