alloc_migrate_target() is called from migrate_pages(), so the page is always from user space, so why not add __GFP_HIGHMEM directly, instead of the following code. if (PageHighMem(page)) // it always return false in x86_64 gfp_mask |= __GFP_HIGHMEM; Another question, when we do migration, why should split THP first? e.g. 2M(512*4kb) should flush 512 times TLB, and 2M(2M*1) only need one. I find flush TLB takes a lot of time, especially multithreaded app. Thanks, Xishi Qiu -- 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>