aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation
diff options
context:
space:
mode:
authorChristoph Lameter <clameter@sgi.com>2006-03-14 22:50:20 -0500
committerLinus Torvalds <torvalds@g5.osdl.org>2006-03-15 00:43:02 -0500
commitb4fb376628e63bfc8071fc915b921da3db4a3385 (patch)
tree01863773f7d48792072b2a49aba9fa78df39eecc /Documentation
parent4983da07f1e2e8dc81cb9d640fbf35b899cdbdf2 (diff)
[PATCH] Page migration documentation update
Update the documentation for page migration. - Fix up bits and pieces in cpusets.txt - Rework text in vm/page-migration to be clearer and reflect the final version of page migration in 2.6.16. Mention Andi Kleen's numactl package that contains user space tools for page migration via libnuma. Add reference to numa_maps and to the manpage in numactl. - Add todo list for outstanding issues Signed-off-by: Christoph Lameter <clameter@sgi.com> Acked-by: Paul Jackson <pj@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'Documentation')
-rw-r--r--Documentation/cpusets.txt41
-rw-r--r--Documentation/vm/page_migration118
2 files changed, 96 insertions, 63 deletions
diff --git a/Documentation/cpusets.txt b/Documentation/cpusets.txt
index 990998ee10b6..30c41459953c 100644
--- a/Documentation/cpusets.txt
+++ b/Documentation/cpusets.txt
@@ -4,8 +4,9 @@
4Copyright (C) 2004 BULL SA. 4Copyright (C) 2004 BULL SA.
5Written by Simon.Derr@bull.net 5Written by Simon.Derr@bull.net
6 6
7Portions Copyright (c) 2004 Silicon Graphics, Inc. 7Portions Copyright (c) 2004-2006 Silicon Graphics, Inc.
8Modified by Paul Jackson <pj@sgi.com> 8Modified by Paul Jackson <pj@sgi.com>
9Modified by Christoph Lameter <clameter@sgi.com>
9 10
10CONTENTS: 11CONTENTS:
11========= 12=========
@@ -90,7 +91,8 @@ This can be especially valuable on:
90 91
91These subsets, or "soft partitions" must be able to be dynamically 92These subsets, or "soft partitions" must be able to be dynamically
92adjusted, as the job mix changes, without impacting other concurrently 93adjusted, as the job mix changes, without impacting other concurrently
93executing jobs. 94executing jobs. The location of the running jobs pages may also be moved
95when the memory locations are changed.
94 96
95The kernel cpuset patch provides the minimum essential kernel 97The kernel cpuset patch provides the minimum essential kernel
96mechanisms required to efficiently implement such subsets. It 98mechanisms required to efficiently implement such subsets. It
@@ -102,8 +104,8 @@ memory allocator code.
1021.3 How are cpusets implemented ? 1041.3 How are cpusets implemented ?
103--------------------------------- 105---------------------------------
104 106
105Cpusets provide a Linux kernel (2.6.7 and above) mechanism to constrain 107Cpusets provide a Linux kernel mechanism to constrain which CPUs and
106which CPUs and Memory Nodes are used by a process or set of processes. 108Memory Nodes are used by a process or set of processes.
107 109
108The Linux kernel already has a pair of mechanisms to specify on which 110The Linux kernel already has a pair of mechanisms to specify on which
109CPUs a task may be scheduled (sched_setaffinity) and on which Memory 111CPUs a task may be scheduled (sched_setaffinity) and on which Memory
@@ -371,22 +373,17 @@ cpusets memory placement policy 'mems' subsequently changes.
371If the cpuset flag file 'memory_migrate' is set true, then when 373If the cpuset flag file 'memory_migrate' is set true, then when
372tasks are attached to that cpuset, any pages that task had 374tasks are attached to that cpuset, any pages that task had
373allocated to it on nodes in its previous cpuset are migrated 375allocated to it on nodes in its previous cpuset are migrated
374to the tasks new cpuset. Depending on the implementation, 376to the tasks new cpuset. The relative placement of the page within
375this migration may either be done by swapping the page out, 377the cpuset is preserved during these migration operations if possible.
376so that the next time the page is referenced, it will be paged 378For example if the page was on the second valid node of the prior cpuset
377into the tasks new cpuset, usually on the node where it was 379then the page will be placed on the second valid node of the new cpuset.
378referenced, or this migration may be done by directly copying 380
379the pages from the tasks previous cpuset to the new cpuset,
380where possible to the same node, relative to the new cpuset,
381as the node that held the page, relative to the old cpuset.
382Also if 'memory_migrate' is set true, then if that cpusets 381Also if 'memory_migrate' is set true, then if that cpusets
383'mems' file is modified, pages allocated to tasks in that 382'mems' file is modified, pages allocated to tasks in that
384cpuset, that were on nodes in the previous setting of 'mems', 383cpuset, that were on nodes in the previous setting of 'mems',
385will be moved to nodes in the new setting of 'mems.' Again, 384will be moved to nodes in the new setting of 'mems.'
386depending on the implementation, this might be done by swapping, 385Pages that were not in the tasks prior cpuset, or in the cpusets
387or by direct copying. In either case, pages that were not in 386prior 'mems' setting, will not be moved.
388the tasks prior cpuset, or in the cpusets prior 'mems' setting,
389will not be moved.
390 387
391There is an exception to the above. If hotplug functionality is used 388There is an exception to the above. If hotplug functionality is used
392to remove all the CPUs that are currently assigned to a cpuset, 389to remove all the CPUs that are currently assigned to a cpuset,
@@ -434,16 +431,6 @@ and then start a subshell 'sh' in that cpuset:
434 # The next line should display '/Charlie' 431 # The next line should display '/Charlie'
435 cat /proc/self/cpuset 432 cat /proc/self/cpuset
436 433
437In the case that a change of cpuset includes wanting to move already
438allocated memory pages, consider further the work of IWAMOTO
439Toshihiro <iwamoto@valinux.co.jp> for page remapping and memory
440hotremoval, which can be found at:
441
442 http://people.valinux.co.jp/~iwamoto/mh.html
443
444The integration of cpusets with such memory migration is not yet
445available.
446
447In the future, a C library interface to cpusets will likely be 434In the future, a C library interface to cpusets will likely be
448available. For now, the only way to query or modify cpusets is 435available. For now, the only way to query or modify cpusets is
449via the cpuset file system, using the various cd, mkdir, echo, cat, 436via the cpuset file system, using the various cd, mkdir, echo, cat,
diff --git a/Documentation/vm/page_migration b/Documentation/vm/page_migration
index c52820fcf500..0dd4ef30c361 100644
--- a/Documentation/vm/page_migration
+++ b/Documentation/vm/page_migration
@@ -12,12 +12,18 @@ is running.
12 12
13Page migration allows a process to manually relocate the node on which its 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 14pages are located through the MF_MOVE and MF_MOVE_ALL options while setting
15a new memory policy. The pages of process can also be relocated 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 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 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. 18process that are located on the from nodes to the destination nodes.
19 19Page migration functions are provided by the numactl package by Andi Kleen
20Manual migration is very useful if for example the scheduler has relocated 20(a version later than 0.9.3 is required. Get it from
21ftp://ftp.suse.com/pub/people/ak). numactl provided libnuma which
22provides an interface similar to other numa functionality for page migration.
23cat /proc/<pid>/numa_maps allows an easy review of where the pages of
24a process are located. See also the numa_maps manpage in the numactl package.
25
26Manual migration is useful if for example the scheduler has relocated
21a process to a processor on a distant node. A batch scheduler or an 27a process to a processor on a distant node. A batch scheduler or an
22administrator may detect the situation and move the pages of the process 28administrator may detect the situation and move the pages of the process
23nearer to the new processor. At some point in the future we may have 29nearer to the new processor. At some point in the future we may have
@@ -25,10 +31,12 @@ some mechanism in the scheduler that will automatically move the pages.
25 31
26Larger installations usually partition the system using cpusets into 32Larger installations usually partition the system using cpusets into
27sections of nodes. Paul Jackson has equipped cpusets with the ability to 33sections of nodes. Paul Jackson has equipped cpusets with the ability to
28move pages when a task is moved to another cpuset. This allows automatic 34move pages when a task is moved to another cpuset (See ../cpusets.txt).
29control over locality of a process. If a task is moved to a new cpuset 35Cpusets allows the automation of process locality. If a task is moved to
30then also all its pages are moved with it so that the performance of the 36a new cpuset then also all its pages are moved with it so that the
31process does not sink dramatically (as is the case today). 37performance of the process does not sink dramatically. Also the pages
38of processes in a cpuset are moved if the allowed memory nodes of a
39cpuset are changed.
32 40
33Page migration allows the preservation of the relative location of pages 41Page migration allows the preservation of the relative location of pages
34within a group of nodes for all migration techniques which will preserve a 42within a group of nodes for all migration techniques which will preserve a
@@ -37,22 +45,26 @@ process. This is necessary in order to preserve the memory latencies.
37Processes will run with similar performance after migration. 45Processes will run with similar performance after migration.
38 46
39Page migration occurs in several steps. First a high level 47Page migration occurs in several steps. First a high level
40description for those trying to use migrate_pages() and then 48description for those trying to use migrate_pages() from the kernel
41a low level description of how the low level details work. 49(for userspace usage see the Andi Kleen's numactl package mentioned above)
50and then a low level description of how the low level details work.
42 51
43A. Use of migrate_pages() 52A. In kernel use of migrate_pages()
44------------------------- 53-----------------------------------
45 54
461. Remove pages from the LRU. 551. Remove pages from the LRU.
47 56
48 Lists of pages to be migrated are generated by scanning over 57 Lists of pages to be migrated are generated by scanning over
49 pages and moving them into lists. This is done by 58 pages and moving them into lists. This is done by
50 calling isolate_lru_page() or __isolate_lru_page(). 59 calling isolate_lru_page().
51 Calling isolate_lru_page increases the references to the page 60 Calling isolate_lru_page increases the references to the page
52 so that it cannot vanish under us. 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.
53 64
542. Generate a list of newly allocates page to move the contents 652. Generate a list of newly allocates page. These pages will contain the
55 of the first list to. 66 contents of the pages from the first list after page migration is
67 complete.
56 68
573. The migrate_pages() function is called which attempts 693. The migrate_pages() function is called which attempts
58 to do the migration. It returns the moved pages in the 70 to do the migration. It returns the moved pages in the
@@ -63,13 +75,17 @@ A. Use of migrate_pages()
634. The leftover pages of various types are returned 754. The leftover pages of various types are returned
64 to the LRU using putback_to_lru_pages() or otherwise 76 to the LRU using putback_to_lru_pages() or otherwise
65 disposed of. The pages will still have the refcount as 77 disposed of. The pages will still have the refcount as
66 increased by isolate_lru_pages()! 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.
67 81
68B. Operation of migrate_pages() 82B. How migrate_pages() works
69-------------------------------- 83----------------------------
70 84
71migrate_pages does several passes over its list of pages. A page is moved 85migrate_pages() does several passes over its list of pages. A page is moved
72if all references to a page are removable at the time. 86if all references to a page are removable at the time. The page has
87already been removed from the LRU via isolate_lru_page() and the refcount
88is increased so that the page cannot be freed while page migration occurs.
73 89
74Steps: 90Steps:
75 91
@@ -79,36 +95,40 @@ Steps:
79 95
803. Make sure that the page has assigned swap cache entry if 963. Make sure that the page has assigned swap cache entry if
81 it is an anonyous page. The swap cache reference is necessary 97 it is an anonyous page. The swap cache reference is necessary
82 to preserve the information contain in the page table maps. 98 to preserve the information contain in the page table maps while
99 page migration occurs.
83 100
844. Prep the new page that we want to move to. It is locked 1014. Prep the new page that we want to move to. It is locked
85 and set to not being uptodate so that all accesses to the new 102 and set to not being uptodate so that all accesses to the new
86 page immediately lock while we are moving references. 103 page immediately lock while the move is in progress.
87 104
885. All the page table references to the page are either dropped (file backed) 1055. All the page table references to the page are either dropped (file
89 or converted to swap references (anonymous pages). This should decrease the 106 backed pages) or converted to swap references (anonymous pages).
90 reference count. 107 This should decrease the reference count.
91 108
926. The radix tree lock is taken 1096. The radix tree lock is taken. This will cause all processes trying
110 to reestablish a pte to block on the radix tree spinlock.
93 111
947. The refcount of the page is examined and we back out if references remain 1127. The refcount of the page is examined and we back out if references remain
95 otherwise we know that we are the only one referencing this page. 113 otherwise we know that we are the only one referencing this page.
96 114
978. The radix tree is checked and if it does not contain the pointer to this 1158. The radix tree is checked and if it does not contain the pointer to this
98 page then we back out. 116 page then we back out because someone else modified the mapping first.
99 117
1009. The mapping is checked. If the mapping is gone then a truncate action may 1189. The mapping is checked. If the mapping is gone then a truncate action may
101 be in progress and we back out. 119 be in progress and we back out.
102 120
10310. The new page is prepped with some settings from the old page so that accesses 12110. The new page is prepped with some settings from the old page so that
104 to the new page will be discovered to have the correct settings. 122 accesses to the new page will be discovered to have the correct settings.
105 123
10611. The radix tree is changed to point to the new page. 12411. The radix tree is changed to point to the new page.
107 125
10812. The reference count of the old page is dropped because the reference has now 12612. The reference count of the old page is dropped because the radix tree
109 been removed. 127 reference is gone.
110 128
11113. The radix tree lock is dropped. 12913. 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.
112 132
11314. The page contents are copied to the new page. 13314. The page contents are copied to the new page.
114 134
@@ -119,11 +139,37 @@ Steps:
119 139
12017. Queued up writeback on the new page is triggered. 14017. Queued up writeback on the new page is triggered.
121 141
12218. If swap pte's were generated for the page then remove them again. 14218. 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
14519. The page locks are dropped from the old and new page.
146 Processes waiting on the page lock can continue.
147
14820. The new page is moved to the LRU and can be scanned by the swapper
149 etc again.
150
151TODO 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).
123 162
12419. The locks are dropped from the old and new page. 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.
125 168
12620. The new page is moved to the LRU. 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.
127 173
128Christoph Lameter, December 19, 2005. 174Christoph Lameter, March 8, 2006.
129 175