JFS is deadlocking fairly easily in 2.5.9. Running `dbench 32' on quad CPU it lasts about 30 seconds.
Which is rather odd, because I pushed it hard in 2.5.8 with much better results (I did get one deadlock, but that's much, much harder to hit). I note that in several places we're calling lock_metapage() under spinlock, which can deadlock if lock_metapage() schedules away, but I'm not hitting that. I backed out the 2.5.8->2.5.9 JFS diff and it still deadlocks, so it looks like some external kernel change has brought this about. Lots of threads are sleeping in __lock_metapage(). The machine is not under any memory stress at the time (dbench 32 on 2.5 gigabytes). I haven't tested on uniprocessor. _______________________________________________ Jfs-discussion mailing list [EMAIL PROTECTED] http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion
