Page Frame Reclaiming Swapping

1
Page Frame ReclaimingSwapping

• Swap-in asynchronous
• Swap-out sync(kswapd)/async(Buddy failure)

2
Main Issues

• Which Kind of page to swap out
• How to distribute pages in the swap area
• How to select pages to be swapped out
• When to perform page swap-out

3
Which Kind of Pages to Swap out

• Anonymous memory regions (user mode stack or
  heap)
• Dirty (Modified) pages of a process (private
  memory mapping)
• IPC shared memory region
• Other considerations
• Swap to files map files on disk
• No swap for pages used by the kernel

4
How to distribute pages in the swap areas

• Slots for each page store pages in contiguous
  slots
• Multiple swap area faster disk for a higher
  priority, searching from the highest priority area

5
How to select the page to be swapped out

• Steal pages from the process with the largest
  number of pages in RAM
• Steal pages in the process according to some
  criterion
• LRU (least recently used), associated with a
  counter for each page table entry
• For X86, Linux uses Accessed flag.

6
The main function of the PFRA
Low on Memory Reclaiming
Periodic Reclaiming
Hibernation Reclaiming
Low memory on buffer allocation __getblk() alloc_
page_buffers()
Low memory on page allocation __alloc_pages()
Suspend to disk __pm_suspend_disk()
kswapd kernel thread
reap_work work queue
kswapd()
cache_reap()
free_more_memory()
balance_pgdat()
try_to_free_pages()
slab_destroy()
shrink_slab()
shrink_caches
shrink_zone()
out_of_memory()
shrink_cache()
refill_inactive_zone()
shrink_list()
page_referenced()
pageout()
7
When to Perform Page Swap-out

• Kernel thread kswaped, activated once every
  second when free page below a predefined
  threshold
• When Buddy system cant be satisfied due to below
  a predefined threshold.

8
Swap Area

• Disk partition or a file
• MAX_SWAPFILES
• 4k byes slot
• Swap area descriptor, swap_info_struct,

9
Main functions

• Swap-in
• do_swap_page(), swap_in(), read_swap_cache_async()
  
• Swap-out
• swap_out(),swap_out_process(),swap_out_vma(),swap_
  out_pgd(), swap_out_pmd(),try_to_swap_out()
• rw_swap_page() low-level swapping function
  ,brw_page() block device driver function

10
Swap_info_struct

• swap_map array of counter,
• 0 page slot is free
• Positive with a swap-out page
• SWAP_MAP_MAX permanent, cant be removed
• SWAP_MAP_BAD unusable
• Prio goodness of the swap area

11
Variables

• swap_infoMAX_SWAPFIELS of swap_info_struct
• nr_swapfiles, indexd of the last array element
• swap_list, sorted by swap area priority
• nr_swap_pages, total number of free, page slots
  in all active swap areas

12
Swapped-out page ID

• Index of swap_info
• Index of page slot in the swap area
• SWP_ENTRY(type, offset) -- ID
• SWP_TYPE, SWP_OFFSET
• Page ID value in page table entry
• Null entry not valid in any process
• Last 2 equal to 0 page swapped out
• Least 1 equal to 1 page is in RAM

13
Swap_duplicate()

• SWP_TYPE to extract partition type
• SWP_OFFSET to extract page slot index
• Judge If it is a valid ID
• Increment swap_infotype.swap_mapoffset. If
  counter equal to SWAP_MAP_MAX or SWAP_MAP_BAD
  unchanged

14
Swap area utility functions

• sys_swapon()
• sys_swapoff()
• Finding a free page slot
• Always from the beginning of the swap area (add
  seek time for swap out functions)
• Always from the last allocated page slot (add
  seek time for swap in functions)
• Linux always from the last allocated page slot
  unless
• End of the swap area reached
• SWPFILE_CLUSTER free page slots allocated after
  the last restart from the beginning of the swap
  area

15
Scan_swap_map()

• 1. If cluster_nr positive, scan swap_map array of
  counters at cluster_next, if null found,
  decrement cluster_nr, go to 3
• 2. Rescan from lowest_bit , set cluster_nr to
  SWAPFILE_CLUSTER
• 3. Put 1 in the entry, decrement nr_swap_pages,
  update lowest_bit, highest_bit, and cluster_next

16
Allocate and Releasing a page slot

• Get_swap_page index of a newly allocated page
  slot
• Swap_free decrement the corresponding swap_map
  counter when swap in

17
SWAP Cache

• Shared page frame
• Shared memory mapping (file mapping)
• Copy on Write
• IPC shared memory

18
Transferring swap page

• rw_swap_page
• read_swap_cache_async
• rw_swap_page_nocache

19
Page swap out

• Swap_out
• Swap_out_process
• Swap_out_vma
• Swap_out_pgd
• Swap_out_pmd
• Try_to_swap_out

20
Page Swap_in

• Page address valid
• Page not in memory
• Page table entry is not null

21
Freeing Page Frames

• Reclaim unused page frame in a cache
• Shrink_mmap
• Shrink_dcache_memry
• Shrink_cache_reap
• Swapping out a page of anonymous memory
• Swapping out a page of IPC

22
Kswaped kernel thread

• nr_free_pages gt freepages.high