• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1Page migration
2--------------
3
4Page migration allows the moving of the physical location of pages between
5nodes in a numa system while the process is running. This means that the
6virtual addresses that the process sees do not change. However, the
7system rearranges the physical location of those pages.
8
9The main intend of page migration is to reduce the latency of memory access
10by moving pages near to the processor where the process accessing that memory
11is running.
12
13Page migration allows a process to manually relocate the node on which its
14pages are located through the MF_MOVE and MF_MOVE_ALL options while setting
15a new memory policy via mbind(). The pages of process can also be relocated
16from another process using the sys_migrate_pages() function call. The
17migrate_pages function call takes two sets of nodes and moves pages of a
18process that are located on the from nodes to the destination nodes.
19Page migration functions are provided by the numactl package by Andi Kleen
20(a version later than 0.9.3 is required. Get it from
21ftp://oss.sgi.com/www/projects/libnuma/download/). numactl provides libnuma
22which provides an interface similar to other numa functionality for page
23migration.  cat /proc/<pid>/numa_maps allows an easy review of where the
24pages of a process are located. See also the numa_maps documentation in the
25proc(5) man page.
26
27Manual migration is useful if for example the scheduler has relocated
28a process to a processor on a distant node. A batch scheduler or an
29administrator may detect the situation and move the pages of the process
30nearer to the new processor. The kernel itself does only provide
31manual page migration support. Automatic page migration may be implemented
32through user space processes that move pages. A special function call
33"move_pages" allows the moving of individual pages within a process.
34A NUMA profiler may f.e. obtain a log showing frequent off node
35accesses and may use the result to move pages to more advantageous
36locations.
37
38Larger installations usually partition the system using cpusets into
39sections of nodes. Paul Jackson has equipped cpusets with the ability to
40move pages when a task is moved to another cpuset (See ../cpusets.txt).
41Cpusets allows the automation of process locality. If a task is moved to
42a new cpuset then also all its pages are moved with it so that the
43performance of the process does not sink dramatically. Also the pages
44of processes in a cpuset are moved if the allowed memory nodes of a
45cpuset are changed.
46
47Page migration allows the preservation of the relative location of pages
48within a group of nodes for all migration techniques which will preserve a
49particular memory allocation pattern generated even after migrating a
50process. This is necessary in order to preserve the memory latencies.
51Processes will run with similar performance after migration.
52
53Page migration occurs in several steps. First a high level
54description for those trying to use migrate_pages() from the kernel
55(for userspace usage see the Andi Kleen's numactl package mentioned above)
56and then a low level description of how the low level details work.
57
58A. In kernel use of migrate_pages()
59-----------------------------------
60
611. Remove pages from the LRU.
62
63   Lists of pages to be migrated are generated by scanning over
64   pages and moving them into lists. This is done by
65   calling isolate_lru_page().
66   Calling isolate_lru_page increases the references to the page
67   so that it cannot vanish while the page migration occurs.
68   It also prevents the swapper or other scans to encounter
69   the page.
70
712. We need to have a function of type new_page_t that can be
72   passed to migrate_pages(). This function should figure out
73   how to allocate the correct new page given the old page.
74
753. The migrate_pages() function is called which attempts
76   to do the migration. It will call the function to allocate
77   the new page for each page that is considered for
78   moving.
79
80B. How migrate_pages() works
81----------------------------
82
83migrate_pages() does several passes over its list of pages. A page is moved
84if all references to a page are removable at the time. The page has
85already been removed from the LRU via isolate_lru_page() and the refcount
86is increased so that the page cannot be freed while page migration occurs.
87
88Steps:
89
901. Lock the page to be migrated
91
922. Insure that writeback is complete.
93
943. Prep the new page that we want to move to. It is locked
95   and set to not being uptodate so that all accesses to the new
96   page immediately lock while the move is in progress.
97
984. The new page is prepped with some settings from the old page so that
99   accesses to the new page will discover a page with the correct settings.
100
1015. All the page table references to the page are converted
102   to migration entries or dropped (nonlinear vmas).
103   This decrease the mapcount of a page. If the resulting
104   mapcount is not zero then we do not migrate the page.
105   All user space processes that attempt to access the page
106   will now wait on the page lock.
107
1086. The radix tree lock is taken. This will cause all processes trying
109   to access the page via the mapping to block on the radix tree spinlock.
110
1117. The refcount of the page is examined and we back out if references remain
112   otherwise we know that we are the only one referencing this page.
113
1148. The radix tree is checked and if it does not contain the pointer to this
115   page then we back out because someone else modified the radix tree.
116
1179. The radix tree is changed to point to the new page.
118
11910. The reference count of the old page is dropped because the radix tree
120    reference is gone. A reference to the new page is established because
121    the new page is referenced to by the radix tree.
122
12311. The radix tree lock is dropped. With that lookups in the mapping
124    become possible again. Processes will move from spinning on the tree_lock
125    to sleeping on the locked new page.
126
12712. The page contents are copied to the new page.
128
12913. The remaining page flags are copied to the new page.
130
13114. The old page flags are cleared to indicate that the page does
132    not provide any information anymore.
133
13415. Queued up writeback on the new page is triggered.
135
13616. If migration entries were page then replace them with real ptes. Doing
137    so will enable access for user space processes not already waiting for
138    the page lock.
139
14019. The page locks are dropped from the old and new page.
141    Processes waiting on the page lock will redo their page faults
142    and will reach the new page.
143
14420. The new page is moved to the LRU and can be scanned by the swapper
145    etc again.
146
147Christoph Lameter, May 8, 2006.
148
149