Le 05.06.2006 03:10, Andrew Morton a écrit : > On Sun, 4 Jun 2006 20:59:50 -0400 > Chuck Ebbert <76306.1226@compuserve.com> wrote: > >> In-Reply-To: <44833955.9000104@free.fr> >> >> On Sun, 04 Jun 2006 21:49:41 +0200, Laurent Riffard wrote: >> >>>> Something strange is happening with the stack. Can you try with 8K stacks? >>>> >>>> kernel hacking ---> >>>> [ ] Use 4Kb for kernel stacks instead of 8Kb >>>> >>> Good catch! >> Jan Beulich was the one who noticed the stack overflow. >> >>> I just tried with 2.6.17-rc5-mm3. The BUG still happens with 4K stacks, >>> but the system runs fine with 8K stacks. >> Can you try -mm3 with "check for stack overflows" and 4k stacks? > > Maybe we could modify the check-for-stack-overflows code with 8k stacks to > be "check for overflows which would kill a 4k stack kernel". By decreasing > the thresholds to 3k (say). > >>> Another info: the system is able to run fine with the following scenario, >>> even with 4K stack: >>> - boot to runlevel 1 >>> - load pktcdvd (modprobe + pktsetup) >>> - then go to runlevel 5 >>> It fails if pktcdvd is loaded at runlevel 2 or higher. >> I have no idea why that would happen. > > I'd be surprised if it has to do with the packet code - more likely someone > screwed something up in the lockdep/genirq/unwind area and the kernel went > recursive. > > I tried it here, couldn't reproduce it. Laurent, can you (re?)send the > offending .config? .config is attached I run a few additional tests: - 2.6.17-rc5-mm3 + CONFIG_DEBUG_STACKOVERFLOW=y, CONFIG_DEBUG_STACK_USAGE=y, CONFIG_4KSTACKS=y: system freezes right after typing "pktsetup test /dev/dvd". No stack trace printed. - Same config as above + lockdep-combo-2.6.17-rc5-mm3.patch + lockdep-tracer-2.6.17-rc5-mm3.patch: Random Oops (recursive dies) on early boot. I sometimes succeded on booting but the system dies on "pktsetup test /dev/dvd". I can't get a full trace since I don't have my second box here for some days. I tried to boot with vga=791, but system hangs with a blank screen. -- laurent