On Fri, 25 Jul 2014 23:38:06 +0200 Joerg Roedel <joro@xxxxxxxxxx> wrote: > On Fri, Jul 25, 2014 at 01:16:39PM -0700, Jesse Barnes wrote: > > > To allow managing external TLBs the MMU-notifiers need to > > > catch the moment when pages are unmapped but not yet freed. > > > This new notifier catches that moment and notifies the > > > interested subsytem when pages that were unmapped are about > > > to be freed. The new notifier will only be called between > > > invalidate_range_start()/end(). > > > > So if we were actually sharing page tables, we should be able to make > > start/end no-ops and just use this new callback, assuming we didn't > > need to do any other serialization or debug stuff, right? > > Well, not completly. What you need with this patch-set is a > invalidate_range and an invalidate_end call-back. There are call sites > of the start/end functions where the TLB flush happens after the _end > notifier (or at least can wait until _end is called). I did not add > invalidate_range calls to these places (yet). But you can easily discard > invalidate_range_start, any flush done in there is useless with shared > page-tables. > > I though about removing the need for invalidate_range_end too when > writing the patches, and possible solutions are > > 1) Add mmu_notifier_invalidate_range() to all places where > start/end is called too. This might add some unnecessary > overhead. > > 2) Call the invalidate_range() call-back from the > mmu_notifier_invalidate_range_end too. > > 3) Just let the user register the same function for > invalidate_range and invalidate_range_end > > I though that option 1) adds overhead that is not needed (but it might > not be too bad, the overhead is an additional iteration over the > mmu_notifer list when there are no call-backs registered). > > Option 2) might also be overhead if a user registers different functions > for invalidate_range() and invalidate_range_end(). In the end I came to > the conclusion that option 3) is the best one from an overhead POV. > > But probably targeting better usability with one of the other options is > a better choice? I am open for thoughts and suggestions on that. Making the _end callback just do another TLB flush is fine too, but it would be nice to have the consistency of (1). I can live with either though, as long as the callbacks are well documented. Thanks, -- Jesse Barnes, Intel Open Source Technology Center -- 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>