Messages in this thread | | | Date | Wed, 11 Mar 2015 23:03:37 +0300 | Subject | Re: [PATCH 2/4] mm, procfs: account for shmem swap in /proc/pid/smaps | From | Konstantin Khlebnikov <> |
| |
On Wed, Mar 11, 2015 at 10:10 PM, Vlastimil Babka <vbabka@suse.cz> wrote: > On 03/11/2015 01:30 PM, Konstantin Khlebnikov wrote: >> >> On Thu, Feb 26, 2015 at 4:51 PM, Vlastimil Babka <vbabka@suse.cz> wrote: >>> >>> Currently, /proc/pid/smaps will always show "Swap: 0 kB" for shmem-backed >>> mappings, even if the mapped portion does contain pages that were swapped >>> out. >>> This is because unlike private anonymous mappings, shmem does not change >>> pte >>> to swap entry, but pte_none when swapping the page out. In the smaps page >>> walk, such page thus looks like it was never faulted in. >> >> >> Maybe just add count of swap entries allocated by mapped shmem into >> swap usage of this vma? That's isn't exactly correct for partially >> mapped shmem but this is something weird anyway. > > > Yeah for next version I want to add a patch optimizing for the (hopefully) > common cases: > > 1. SHMEM_I(inode)->swapped is 0 - no need to consult radix tree > 2. shmem inode is mapped fully (I hope it's ok to just compare its size and > mapping size) - just use the value of SHMEM_I(inode)->swapped like you > suggest >
BTW using radix tree iterator you can count swap entries without touching page->count.
Also long time ago I've suggested to mark swap entries in shmem with one of radix tree tag -- tagged iterator is much faster for sparse trees. (just for this case it's overkill but these tags can speedup swapoff)
| |