Page 1 of 1

kernel freeze at boot time if CONFIG_PAX_NOEXEC enabled

PostPosted: Sat Apr 21, 2007 9:48 am
by Netzpython
hi there,
have a weird problem here while booting a linux 2.6 kernel on a quite old SMP box.
(Dual-P3@500mhz + HP Netserver E60 board) Maybe its PaX related cause
if CONFIG_PAX_NOEXEC and related options (CONFIG_PAX_SEGMEXEC, CONFIG_PAX_PAGEEXEC, CONFIG_PAX_MPROTECT, CONFIG_PAX_KERNEXEC) are set, the kernel freezes at boottime. Last printed msg was:

checking if image is initramfs...it isn't (bad gzip magic numbers); looks like an initrd

Then the box locks up completely, no kbd, no blinkin' LEDs, simply hangs here and sit forever so ive to cycle power. Anyway, iam sure the initramfs is valid, it works when CONFIG_PAX_NOEXEC is disabled! useing a cramfs-based initrd, the screen shows vertical columns of blue and green lines and then reboots...
Its debian etch 4.0, error occurs with kernel 2.6.17.9, 2.6.19.7, 2.6.20.7, gcc 3.4 and 4.x.

Btw:
CONFIG_COMPAT_VDSO is not set
CONFIG_PCI_GODIRECT=y

though thats a more or less productive box iam able to do some more debuggin or experimentin if that could be useful!

best regards,netzpython

(...and »THANK YOU« spender && paxteam, GRSEC is great work!)

8)

Re: kernel freeze at boot time if CONFIG_PAX_NOEXEC enabled

PostPosted: Sat Apr 21, 2007 5:44 pm
by PaX Team
Netzpython wrote:Maybe its PaX related cause if CONFIG_PAX_NOEXEC and related options (CONFIG_PAX_SEGMEXEC, CONFIG_PAX_PAGEEXEC, CONFIG_PAX_MPROTECT, CONFIG_PAX_KERNEXEC) are set, the kernel freezes at boottime. Last printed msg was:

checking if image is initramfs...it isn't (bad gzip magic numbers); looks like an initrd
that's an interesting one, could you send me a failing kernel image + .config please (preferably with the 2.6.20.7 kernel)?

Re: kernel freeze at boot time if CONFIG_PAX_NOEXEC enabled

PostPosted: Fri May 04, 2007 7:58 am
by PaX Team
PaX Team wrote:that's an interesting one, could you send me a failing kernel image + .config please (preferably with the 2.6.20.7 kernel)?
after some debugging it turned out that apparently LILO loaded the initrd at the wrong address that got overwritten when the kernel was decompressed, hence the magic number mismatch, etc. so if you're experiencing similar problems with initrd, try GRUB instead.