On Thu, May 17, 2012 at 06:22:15PM +0100, Russell King wrote: > On Thu, May 17, 2012 at 05:00:12PM +0100, Catalin Marinas wrote: > > On Thu, May 17, 2012 at 12:28:06PM +0100, Peter Zijlstra wrote: > > > On Thu, 2012-05-17 at 10:51 +0100, Russell King wrote: > > > > On Thu, May 17, 2012 at 10:30:23AM +0100, Catalin Marinas wrote: > > > > > Another minor thing is that on newer ARM processors (Cortex-A15) we > > > > > need the TLB shootdown even on UP systems, so tlb_fast_mode should > > > > > always return 0. Something like below (untested): > > > > > > > > No Catalin, we need this for virtually all ARMv7 CPUs whether they're UP > > > > or SMP, not just for A15, because of the speculative prefetch which can > > > > re-load TLB entries from the page tables at _any_ time. > > > > > > Hmm,. so this is mostly because of the confusion/coupling between > > > tlb_remove_page() and tlb_remove_table() I guess. Since I don't see the > > > freeing of the actual pages being a problem with speculative TLB > > > reloads, just the page-tables. > > > > > > Should we introduce a tlb_remove_table() regardless of > > > HAVE_RCU_TABLE_FREE which always queues the tables regardless of > > > tlb_fast_mode()? > > > > BTW, looking at your tlb-unify branch, does tlb_remove_table() call > > tlb_flush/tlb_flush_mmu before freeing the tables? I can only see > > tlb_remove_page() doing this. On ARM, even UP, we need the TLB flushing > > after clearing the pmd and before freeing the pte page table (and > > ideally doing it less often than at every pte_free_tlb() call). > > Catalin, > > The way TLB shootdown stuff works is that _every_ single bit of memory > which gets freed, whether its a page or a page table, gets added to a > list of pages to be freed. > > So, the sequence is: > - remove pte/pmd/pud/pgd pointers > - add pages, whether they be pages pointed to by pte entries or page tables > to be freed to a list > - when list is sufficiently full, invalidate TLBs > - free list of pages > > That means the pages will not be freed, whether it be a page mapped > into userspace or a page table until such time that the TLB has been > invalidated. > > For page tables, this is done via pXX_free_tlb(), which then calls out > to the arch specific __pXX_free_tlb(), which ultimately then hands the > page table over to tlb_remove_page() to add to the list of to-be-freed > pages. I know that already, not sure why you explained it again (but it's good for future reference). My point was that if we move to HAVE_RCU_FREE_TLB, the other architectures doing this are calling tlb_remove_table() instead of tlb_remove_page() in __p??_free_tlb(). And tlb_remove_table() does not do any TLB maintenance when it can no longer queue pages (batch table overflow). -- Catalin -- 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/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>