On 12/28/2015 02:55 AM, Dan Williams wrote: > On Sun, Dec 27, 2015 at 12:33 AM, Bob Liu <lliubbo@xxxxxxxxx> wrote: >> Hey Dan, >> > [..] >> What about space for page tables? >> Page tables(mapping all memory in PMEM to virtual address space) may >> also consume significantly DRAM space if huge page is not enabled or >> split. >> Should we also consider to allocate pte page tables from PMEM in future? > > On x86_64 these ranges are covered by gigabyte pages by default (see > init_memory_mapping()). I don't see much incremental benefit from > allocating pte's from pmem. > Oh, that's the direct mapping. I mean ptes consumed in: __dax_fault > dax_insert_mapping > insert_pfn > __get_locked_pte In some bad situations e.g mmap a lot of large ext4 files exist in pmem but hugepage can't be used because of fragmentation, then the consumption of pte page tables may can't be ignored? Anyway, this is not a blocker of these patches. Thanks, Bob -- 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>