On Sun, Jan 27, 2013 at 01:40:40PM -0800, Hugh Dickins wrote: > On Sun, 27 Jan 2013, Shaohua Li wrote: > > On Sat, Jan 26, 2013 at 06:16:05PM -0800, Hugh Dickins wrote: > > > On Fri, 25 Jan 2013, Shaohua Li wrote: > > > > On Thu, Jan 24, 2013 at 10:45:57PM -0500, Sasha Levin wrote: > > > > > > > > Subject: give-each-swapper-space-separate-backing_dev_info > > > > > > > > The backing_dev_info can't be shared by all swapper address space. > > > > > > Whyever not? It's perfectly normal for different inodes/address_spaces > > > to share a single backing_dev! Sasha's trace says that it's wrong to > > > initialize it MAX_SWAPFILES times: fair enough. But why should I now > > > want to spend 32kB (not even counting their __percpu counters) on all > > > these pseudo-backing_devs? > > > > That's correct, silly me. Updated it. > > Looks much more to my taste, thank you! > > > > > > > p.s. a grand little change would be to move page_cluster and swap_setup() > > > from mm/swap.c to mm/swap_state.c: they have nothing to do with the other > > > contents of swap.c, and everything to do with the contents of swap_state.c. > > > Why swap.c is called swap.c is rather a mystery. > > > > Tried, but looks page_cluster is used in sysctl, moving to swap_state.c will > > make it optional. don't want to add another #ifdef, so give up. > > Good point, thanks for trying, maybe I'll attack it next time it > irritates me. > > I don't yet know whether I approve of your changes or not, but running > with them to see (and I'll send another bugfix separately in a moment). > > I was the one who removed the swap_device_lock() which 2.4 used, > because it almost always ended up having to take both swap_list_lock() > and swap_device_lock(si). You seem to have done a much better job of > separating them usefully, but I need to convince myself that it does > end up safely. > > My reservations so far would be: how many installations actually have > more than one swap area, so is it a good tradeoff to add more overhead > to help those at the (slight) expense of everyone else? The increasingly > ugly page_mapping() worries me, and the static array of swapper_spaces > annoys me a little. > > I'm glad Minchan has now pointed you to Rik's posting of two years ago: > I think there are more important changes to be made in that direction. Not sure how others use multiple swaps, but current lock contention forces us to use multiple swaps. I haven't carefully think about Rik's posting, but looks it doesn't solve the lock contention problem. Thanks, Shaohua -- 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>