On Sun, 13 Jun 1999, Konstantin Boldyshev wrote:
> If sys_exit call in a user program is missing such program hangs and
> cannot be killed. At all (kill -9 and SysRq will not help). Even if you
> kill that terminal - program itself remains anyway. Only way to kill -
> reboot.
Thanks - finally a better place to start debugging this problem.
I'm currently building kernel 26mdk hoping it will fix the problem. I
won't release it before I've verified the problem is fixed.
Looking at the patches we've applied, I don't think one of them can be
causing the problem:
- linux-2.2.9-ISDN.diff.bz2
New ISDN drivers. They're more reliable than the ones in the stock
kernel, and 100% inactive unless you're using ISDN.
- kernel-2.2.7-ac3-pentium3.diff.bz2
Pentium III identification.
It's just a one-liner patch to a string array, can't hurt.
- 2.2.9-multisession.patch.bz2
Support for reading old sessions from a multisession CD (using
mount -o session=<nr>).
Does something only while mounting the CD.
- aic7xxx-5.1.13-5.1.17.patch.bz2
Update for the Adaptec AIC7xxx SCSI drivers, known to be more stable
than the drivers currently in the kernel.
Only used with Adaptec SCSI cards, can't be the cause.
- 2.2.9.uniform-ide-6.19.hotel.patch.bz2
Updated IDE drivers, including support for the PIIX4 chipset.
- linux-2.2.9-security.patch.bz2
Fix for a security bug with directory handling
- fat-patch-20.bz2
FAT bugfixes (only in 25mdk)
- ibcs-2.1-rh.patch.bz2
iBCS patches from RH 6.0
- pcmcia-cs-%{pcmciaver}-script.patch.bz2
chkconfig support for PCMCIA
- linux-2.2.9-ac5-DAC960.patch.bz2
DAC960 drivers
- patch-pcsp-soundcore-2.2.5.bz2
PC-Speaker driver
- e2compr-0.4.31-patch-2.2.9-ac5.bz2
ext2 compression support - disabled by default, so it can't be the cause
- linux-2.2.6ac3-compile.patch.bz2
Patch to make the kernel compile in all configurations (SMP+3c507
doesn't work without it)
> I've tested this with 2.0.36,2.2.1,2.2.8 and non mandrake 2.2.9
Which compiler have you used to compile these kernels? I'm starting to
think we've stumbled across a pgcc bug...
If the problem still occurs in 26mdk, I'll try recompiling it with egcs
1.1.2 to see if that fixes it.
LLaP
bero
