Am 03.04.2013 16:03, schrieb Mel Gorman: >> I've now tested 3.9-rc5 this gaves me a slightly different kernel log: >> [ 197.236518] pigz[2908]: segfault at 0 ip (null) sp >> 00007f347bffed00 error 14 >> [ 197.237632] traps: pigz[2915] general protection ip:7f3482dbce2d >> sp:7f3473ffec10 error:0 in libz.so.1.2.3.4[7f3482db7000+17000] >> [ 197.330615] in pigz[400000+10000] >> >> With 3.8 it is the same as with 3.8.4 or 3.8.5. >> > > Ok. Are there NUMA machines were you do *not* see this problem? Sadly no. I can really fast reproduce it with this one: 1.) Machine with only 16GB Mem 2.) compressing two 60GB Files in parallel with pigz consuming all cores > If so, can you spot what the common configuration, software or hardware, that > affects the broken machines versus the working machines? I'm wondering > if there is a bug in a migration handler. > > Do you know if a NUMA nodes are low on memory when the segfaults occur? One of them is but the others aren't. (196GB Mem just 20GB in use) Stefan -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>