oom: oom_kill_process() doesn't select kthread child
in [Kernel]
|
Prev: [PATCH 6/9] oom: unify CAP_SYS_RAWIO check into other superuser check
Next: + oom-fold-__out_of_memory-into-out_of_memory.patch added to -mm tree
From: KOSAKI Motohiro on 16 Jun 2010 22:00 > On Wed, Jun 16, 2010 at 08:32:08PM +0900, KOSAKI Motohiro wrote: > > Now, select_bad_process() have PF_KTHREAD check, but oom_kill_process > > doesn't. It mean oom_kill_process() may choose wrong task, especially, > > when the child are using use_mm(). > Now oom_kill_process is called by three place. > > 1. mem_cgroup_out_of_memory > 2. out_of_memory with sysctl_oom_kill_allocating_task > 3. out_of_memory with non-sysctl_oom_kill_allocating_task > > I think it's no problem in 1 and 3 since select_bad_process already checks > PF_KTHREAD. The problem in in 2. > So How about put the check before calling oom_kill_process in case of > sysctl_oom_kill_allocating task? > > if (sysctl_oom_kill_allocating_task) { > if (!current->flags & PF_KTHREAD) > oom_kill_process(); > > It can remove duplicated PF_KTHREAD check in select_bad_process and > oom_kill_process. This patch changed child selection logic. select_bad_process() doesn't check victim's child. IOW, this is necessary when all 1-3. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: KOSAKI Motohiro on 30 Jun 2010 05:30 > On Thu, 17 Jun 2010, KOSAKI Motohiro wrote: > > > > > Now, select_bad_process() have PF_KTHREAD check, but oom_kill_process > > doesn't. It mean oom_kill_process() may choose wrong task, especially, > > when the child are using use_mm(). > > > > This type of check should be moved to badness(), it will prevent these > types of tasks from being selected both in select_bad_process() and > oom_kill_process() if the score it returns is 0. No, process check order of select_bad_process() is certain meaningful. Only PF_KTHREAD check can move into badness(). Okey, that's fix to incorrect /proc/<pid>/oom_score issue. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: KOSAKI Motohiro on 30 Jun 2010 20:10
> On Wed, Jun 30, 2010 at 06:27:52PM +0900, KOSAKI Motohiro wrote: > > Now, select_bad_process() have PF_KTHREAD check, but oom_kill_process > > doesn't. It mean oom_kill_process() may choose wrong task, especially, > > when the child are using use_mm(). > > Is it possible child is kthread even though parent isn't kthread? Usually unhappen. but crappy driver can do any strange thing freely. As I said, oom code should have conservative assumption as far as possible. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |