This version looks really clean. Thank for keeping working on this through 9 versions! A couple of small issues. 1) In arch/ia64/mm/init.c: __ia64_sync_icache_dcache() - if (!pte_exec(pte)) - return; /* not an executable page... */ + BUG_ON(!pte_exec(pte)); In this latest version the only route to this routine is from set_pte() inside the test : if (pte_exec(pteval) && ....) { } So this BUG_ON is now redundant. 2) In include/asm-ia64/pgtable.h + if (pte_exec(pteval) && // flush only new executable page. + pte_present(pteval) && // swap out ? + pte_user(pteval) && // ignore kernel page + (!pte_present(*ptep) ||// do_no_page or swap in, migration, + pte_pfn(*ptep) != pte_pfn(pteval))) // do_wp_page(), page copy + /* load_module() calles flush_icache_range() explicitly*/ + __ia64_sync_icache_dcache(pteval); Just above this there is a comment saying that pte_exec() only works when pte_present() is true. So we must re-order the conditions so that we check that the pteval satisfies pte_present() before using either of pte_exec() or pte_user() on it like this: if (pte_present(pteval) && pte_exec(pteval) && pte_user(pteval) && I put in some crude counters to see whether we should check pte_exec() or pte_user() next ... and it was very clear that the pte_exec() check gets us out of the if() faster (at least during a kernel build). I also compared how often the old code called lazy_mmu_prot_update() with how often the new code calls __ia64_sync_icache_dcache() (again using kernel build as my workload) ... and the answer is about the same (less than 0.2% change ... probably less than run-to-run variation). So now the only remaining task is to convince myself that this new version covers all the cases. -Tony - To unsubscribe from this list: send the line "unsubscribe linux-ia64" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
![]() |