On 09/24/2015 09:04 PM, Rik van Riel wrote: > On 09/24/2015 05:14 PM, Joe Lawrence wrote: >> [ +cc for linux-mm mailinglist address ] >> >> On 09/24/2015 05:08 PM, Joe Lawrence wrote: >>> Hi Mel, Rik et al, >>> >>> We've encountered interesting NUMA balancing behavior on RHEL7.1, >>> reproduced with an upstream 4.2 kernel (of similar .config), that can >>> leave a user process trapped in the kernel performing task_numa_work. >>> >>> Our test group set up a server with 256GB memory running a program that >>> allocates and dirties ~50% of that memory. They reported the following >>> condition when they attempted to kill the test process -- the signal was >>> never handled, instead traces showed the task stuck here: > > Does the bug still happen with this patch applied? > > https://git.kernel.org/cgit/linux/kernel/git/tip/tip.git/commit/?id=4620f8c1fda2af4ccbd11e194e2dd785f7d7f279 > Hi Rik, Success! With 4620f8c1fda2 (-tip) cherry-picked on-top of 4.2, I could successfully kill off the memory test process, even when the numa_scan_period_max dropped to 140. I also ran kicked off the est program and let continue overnight (it restarts itself after a given time) and several iterations ran without incident. Thanks, -- Joe -- 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>