On 07/27/2013 11:29 PM, Pavel Emelyanov wrote: > On 07/27/2013 12:55 AM, Andy Lutomirski wrote: >> On Fri, Jul 26, 2013 at 1:18 PM, Cyrill Gorcunov <gorcunov@xxxxxxxxx> wrote: >>> Andy reported that if file page get reclaimed we loose soft-dirty bit >>> if it was there, so save _PAGE_BIT_SOFT_DIRTY bit when page address >>> get encoded into pte entry. Thus when #pf happens on such non-present >>> pte we can restore it back. >>> >> >> Unless I'm misunderstanding this, it's saving the bit in the >> non-present PTE. This sounds wrong -- what happens if the entire pmd >> (or whatever the next level is called) gets zapped? (Also, what >> happens if you unmap a file and map a different file there?) > > The whole pte gets zapped on vma unmap, and in this case forgetting > the soft-dirty bit completely is OK. I mean -- soft-dirty bits denote changes in the vm area, if you remove one, then it can be found out from the /proc/pid/maps file that the vma has disappeared. But one problem really went unnoticed here -- if we map a new vma in place of some old one with the same flags and prots. It looks like we need a vma soft-dirty mark, that is set on mmap and mremap, is cleared on soft dirty clear and is propagated into pte pagemap bits. >> --Andy Thanks, Pavel -- 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>