On Thu, 2004-05-20 at 09:44, Philippe Gerum wrote: > On Wed, 2004-05-19 at 13:26, Rus V. Brushkoff wrote: > > On Tue, 18 May 2004, Philippe Gerum wrote: > > > > :> lines after stop (probably - made partialy from of some other succesful > > boot log, line feeding is too fast)> > > :> VFS: Mounted root (ext2 filesystem) readonly. > > :> Trying to move old root to /initrd ... okay > > :> Freeing unused kernel memory: 156k freed > > :> Usage: init 0123456SsQqAaBbCcUu > > :> Usage: init 0123456SsQqAaBbCcUu > > :> Usage: init 0123456SsQqAaBbCcUu > > :> Usage: init 0123456SsQqAaBbCcUu > > :> Usage: init 0123456SsQqAaBbCcUu > > :> ... endlessly repeating > > :> > > : > > :Ok, this one is framed now. It should also explain the lockup at boot, > > :which looks like a variation of the same problem. I need a bit more time > > :banging my head on the wall in order to solve it, but there's hope. In > > :any case, it's a 2.6-specific Adeos bug in r5c2. Adeos patches for 2.4 > > :kernels are totally immune from this one. > > > > Strange, I tried 2.6.5, and now using 2.6.6 with slightly modified r5c2 - > > no problem at boot. Though there are some problems with fifo - now I'm > > investigating it ... > > Looks like corrupted init or corrupted kernel patch - in kernel source > > (init/main.c:run_init_process()) init started with no options at all. > > > > In fact, init is indeed called with no arg by the booting kernel. The > problem is that some additions from r5c2 to the sysenter/sysexit support > utterly trash the registers on the syscall return path; oh, well... > So when init asks for its pid by calling getpid(), it gets back zero, > instead of one. In reaction, it simply emits a plain usage message and > exits, before being respawned a number of times, failing the same way. > You and I did not experiment this problem because our glibc install > still uses the traditional 'int 0x80' method for syscalling which is > immune to this bug, not the newer sysenter/sysexit protocol. > > I intend to come up with r5c3 fixing this bug next week, after a few > days of stress testing. >
This nasty bug has been fixed: http://download.gna.org/adeos/patches/v2.6/i386/candidates/adeos-linux-2.6.5-i386-r5c3.patch -- Philippe.
