I'm a little low on details, but this is a known issue with kernel 2.4.26 and is discussed on the lkml found at the link I included. I thought it may be of interest to the gentoo team. When using distcc, one of the remote machines that are being used to compile code may encounter a kernel panic. It's strange, however. I've got two 2.4.26 machines doing all the compiling for a slower machine, and one crashes while the other does not. The first line of the error is kernel BUG at page_alloc.c:98 It's easily reproducible - also discussed on the lkml. I'd suggest checking the details there, because I can reproduce the error on one machine, while I cannot on the other. Basically all I have to do is compile something with distcc, and within minutes (or less) I get a kernel panic. Sorry I havn't included too much info. It seems to be a known kernel issue and so not entirely related to Gentoo. My intent was only to notify Gentoo developers of this problem so they can decide on an appropriate action to take. Be that marking distcc as unstable, or issuing a patch to (all?) the kernel sources. I hope this is acceptable. This may be related to Bug #51689 - I see tg3 mentioned in that report and in the lkml posts. Though they're talking different kernel versions. Reproducible: Always Steps to Reproduce:
Martin, just a heads up in case you don't read lkml.
Thanks Lisa I did hear of that in passing but had not noticed it was in 2.4.26. All I can suggest is that you try different options in distcc to try to localize it: mmap on and off, compression on and off, corks on and off, and so on.
Created attachment 34357 [details, diff] distcc patch for Linux kernel 2.4.26 The patch wouldn't apply using the patch program, so I had to apply it manually.
Comment on attachment 34357 [details, diff] distcc patch for Linux kernel 2.4.26 This patch was on the lkml thread. I gave it a try this weekend on a couple machines and they've been compiling for another for the last couple of days without crashing.
Based on the kernel list thread, I think you can avoid this bug by just doing export DISTCC_MMAP=0 DISTCC_SENDFILE=0 on all the affected machines before starting distcc and distccd. (e.g. put it in /etc/profile.local or somewhere like that.)
Still an issue here? Or was this fixed in a newer version of distcc? (i don't know what version the OP was using).
I'm new to Bugzilla / bug reporting so please bare with me. I've tried the workaround from #5, by adding the vars to both /etc/env.d/02distcc and /etc/conf.d/distccd. i currently use sys-devel/distcc-2.13-r1 with 2.4.25_pre7-gss-r9 the workaround helped, at least I had no kernel panics for ~7 days. I did some "emerge -vu world"s on a 2nd machine using distcc during this time. after updating the kernel to 2.4.25_pre7-gss-r11 i get the kernel panic again, the workaround is still in place. The affected machine also uses the tg3 module, the OP mentioned for the onboard Broadcom NetXtreme BCM5702X Gigabit nic. my current CFLAGS are CFLAGS="-O3 -march=i686 -mcpu=athlon-xp -funroll-loops -pipe -fomit-frame-pointer" Would it be helpful to recompile distcc with less aggressive CFLAGS? If so, what else should I recompile...? Any more info I could possibly provide?
The problem isn't with distcc and how it's compiled. The problem is with the kernel. Even if you had some poorly designed program, it shouldn't cause a kernel panic. I'm in the process of updating to Linux kernel 2.4.27, I'll give distcc a run and see how the new kernel does.
Yes, it's a kernel bug. http://distcc.samba.org/faq.html#gbe-panic Some people have reported that the patch linked from there fixes it. I think it's in 2.4.27 but I haven't checked. Confirmation either way from people seeing the bug would be welcome. (I don't see it on my machines.)
Hmm, should have looked beyond bugzilla. Thanks for the pointers and clarifying things for me. Will go and try to patch 2.4.25_pre7-gss-r11. The only thing that astounds me, is that 2.4.25_pre7-gss-r11 panics while 2.4.25_pre7-gss-r9 does not? I guess, there shouldn't be to much difference here. But ignore me, I'm no programmer, and have no real insight to kernels besides basic theoretical stuff.
Try the ebuild/patch from bug 62739
see bug 62739 ... nothing to see here *** This bug has been marked as a duplicate of 62739 ***
