+ /*
+ * If the mm has notifiers then we would need to invalidate them around
+ * unmap_page_range and that is risky because notifiers can sleep and
+ * what they do is basically undeterministic. So let's have a short
+ * sleep to give the oom victim some more time.
+ * TODO: we really want to get rid of this ugly hack and make sure that
+ * notifiers cannot block for unbounded amount of time and add
+ * mmu_notifier_invalidate_range_{start,end} around unmap_page_range
+ */
+ if (mm_has_notifiers(mm)) {
+ up_read(&mm->mmap_sem);
+ schedule_timeout_idle(HZ);
+ goto unlock_oom;
+ }
+