Documentation/vm/page_migration

   1 Page migration
   2 --------------
   3
   4 Page migration allows the moving of the physical location of pages between
   5 nodes in a numa system while the process is running. This means that the
   6 virtual addresses that the process sees do not change. However, the
   7 system rearranges the physical location of those pages.
   8
   9 The main intend of page migration is to reduce the latency of memory access
  10 by moving pages near to the processor where the process accessing that memory
  11 is running.
  12
  13 Page migration allows a process to manually relocate the node on which its
  14 pages are located through the MF_MOVE and MF_MOVE_ALL options while setting
  15 a new memory policy via mbind(). The pages of process can also be relocated
  16 from another process using the sys_migrate_pages() function call. The
  17 migrate_pages function call takes two sets of nodes and moves pages of a
  18 process that are located on the from nodes to the destination nodes.
  19 Page migration functions are provided by the numactl package by Andi Kleen
  20 (a version later than 0.9.3 is required. Get it from
  21 ftp://ftp.suse.com/pub/people/ak). numactl provided libnuma which
  22 provides an interface similar to other numa functionality for page migration.
  23 cat /proc/<pid>/numa_maps allows an easy review of where the pages of
  24 a process are located. See also the numa_maps manpage in the numactl package.
  25
  26 Manual migration is useful if for example the scheduler has relocated
  27 a process to a processor on a distant node. A batch scheduler or an
  28 administrator may detect the situation and move the pages of the process
  29 nearer to the new processor. At some point in the future we may have
  30 some mechanism in the scheduler that will automatically move the pages.
  31
  32 Larger installations usually partition the system using cpusets into
  33 sections of nodes. Paul Jackson has equipped cpusets with the ability to
  34 move pages when a task is moved to another cpuset (See ../cpusets.txt).
  35 Cpusets allows the automation of process locality. If a task is moved to
  36 a new cpuset then also all its pages are moved with it so that the
  37 performance of the process does not sink dramatically. Also the pages
  38 of processes in a cpuset are moved if the allowed memory nodes of a
  39 cpuset are changed.
  40
  41 Page migration allows the preservation of the relative location of pages
  42 within a group of nodes for all migration techniques which will preserve a
  43 particular memory allocation pattern generated even after migrating a
  44 process. This is necessary in order to preserve the memory latencies.
  45 Processes will run with similar performance after migration.
  46
  47 Page migration occurs in several steps. First a high level
  48 description for those trying to use migrate_pages() from the kernel
  49 (for userspace usage see the Andi Kleen's numactl package mentioned above)
  50 and then a low level description of how the low level details work.
  51
  52 A. In kernel use of migrate_pages()
  53 -----------------------------------
  54
  55 1. Remove pages from the LRU.
  56
  57    Lists of pages to be migrated are generated by scanning over
  58    pages and moving them into lists. This is done by
  59    calling isolate_lru_page().
  60    Calling isolate_lru_page increases the references to the page
  61    so that it cannot vanish while the page migration occurs.
  62    It also prevents the swapper or other scans to encounter
  63    the page.
  64
  65 2. Generate a list of newly allocates page. These pages will contain the
  66    contents of the pages from the first list after page migration is
  67    complete.
  68
  69 3. The migrate_pages() function is called which attempts
  70    to do the migration. It returns the moved pages in the
  71    list specified as the third parameter and the failed
  72    migrations in the fourth parameter. The first parameter
  73    will contain the pages that could still be retried.
  74
  75 4. The leftover pages of various types are returned
  76    to the LRU using putback_to_lru_pages() or otherwise
  77    disposed of. The pages will still have the refcount as
  78    increased by isolate_lru_pages() if putback_to_lru_pages() is not
  79    used! The kernel may want to handle the various cases of failures in
  80    different ways.
  81
  82 B. How migrate_pages() works
  83 ----------------------------
  84
  85 migrate_pages() does several passes over its list of pages. A page is moved
  86 if all references to a page are removable at the time. The page has
  87 already been removed from the LRU via isolate_lru_page() and the refcount
  88 is increased so that the page cannot be freed while page migration occurs.
  89
  90 Steps:
  91
  92 1. Lock the page to be migrated
  93
  94 2. Insure that writeback is complete.
  95
  96 3. Make sure that the page has assigned swap cache entry if
  97    it is an anonyous page. The swap cache reference is necessary
  98    to preserve the information contain in the page table maps while
  99    page migration occurs.
 100
 101 4. Prep the new page that we want to move to. It is locked
 102    and set to not being uptodate so that all accesses to the new
 103    page immediately lock while the move is in progress.
 104
 105 5. All the page table references to the page are either dropped (file
 106    backed pages) or converted to swap references (anonymous pages).
 107    This should decrease the reference count.
 108
 109 6. The radix tree lock is taken. This will cause all processes trying
 110    to reestablish a pte to block on the radix tree spinlock.
 111
 112 7. The refcount of the page is examined and we back out if references remain
 113    otherwise we know that we are the only one referencing this page.
 114
 115 8. The radix tree is checked and if it does not contain the pointer to this
 116    page then we back out because someone else modified the mapping first.
 117
 118 9. The mapping is checked. If the mapping is gone then a truncate action may
 119    be in progress and we back out.
 120
 121 10. The new page is prepped with some settings from the old page so that
 122    accesses to the new page will be discovered to have the correct settings.
 123
 124 11. The radix tree is changed to point to the new page.
 125
 126 12. The reference count of the old page is dropped because the radix tree
 127     reference is gone.
 128
 129 13. The radix tree lock is dropped. With that lookups become possible again
 130     and other processes will move from spinning on the tree lock to sleeping on
 131     the locked new page.
 132
 133 14. The page contents are copied to the new page.
 134
 135 15. The remaining page flags are copied to the new page.
 136
 137 16. The old page flags are cleared to indicate that the page does
 138     not use any information anymore.
 139
 140 17. Queued up writeback on the new page is triggered.
 141
 142 18. If swap pte's were generated for the page then replace them with real
 143     ptes. This will reenable access for processes not blocked by the page lock.
 144
 145 19. The page locks are dropped from the old and new page.
 146     Processes waiting on the page lock can continue.
 147
 148 20. The new page is moved to the LRU and can be scanned by the swapper
 149     etc again.
 150
 151 TODO list
 152 ---------
 153
 154 - Page migration requires the use of swap handles to preserve the
 155   information of the anonymous page table entries. This means that swap
 156   space is reserved but never used. The maximum number of swap handles used
 157   is determined by CHUNK_SIZE (see mm/mempolicy.c) per ongoing migration.
 158   Reservation of pages could be avoided by having a special type of swap
 159   handle that does not require swap space and that would only track the page
 160   references. Something like that was proposed by Marcelo Tosatti in the
 161   past (search for migration cache on lkml or linux-mm@kvack.org).
 162
 163 - Page migration unmaps ptes for file backed pages and requires page
 164   faults to reestablish these ptes. This could be optimized by somehow
 165   recording the references before migration and then reestablish them later.
 166   However, there are several locking challenges that have to be overcome
 167   before this is possible.
 168
 169 - Page migration generates read ptes for anonymous pages. Dirty page
 170   faults are required to make the pages writable again. It may be possible
 171   to generate a pte marked dirty if it is known that the page is dirty and
 172   that this process has the only reference to that page.
 173
 174 Christoph Lameter, March 8, 2006.
 175