Michal Hocko wrote: > The comment above the check is misleading but now you are allowing to > fail all ALLOC_NO_WATERMARKS (without __GFP_NOFAIL) allocations before > entering the direct reclaim and compaction. This seems incorrect. What > about __GFP_MEMALLOC requests? So, you want __GPP_MEMALLOC to retry forever unless TIF_MEMDIE is set, don't you? > I think the check for TIF_MEMDIE makes more sense here. Since we already failed to allocate from memory reserves, I don't know if direct reclaim and compaction can work as expected under such situation. Maybe the OOM killer is invoked, but I worry that the OOM victim gets stuck because we already failed to allocate from memory reserves. Unless next OOM victims are chosen via timeout, I think that this can be one of triggers that lead to silent hangup... (Just my suspect. I can't prove it because I can't go to in front of customers' servers and check SysRq.) -- 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>