Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit b4fb376628e63bfc8071fc915b921da3db4a3385

Authored by Christoph Lameter
Committed by Linus Torvalds
1 parent 4983da07f1
Exists in master and in 7 other branches imx_3.0.35_4.1.0, imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga, smarc-l5.0.0_1.0.0-ga

[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>

Showing 2 changed files with 96 additions and 63 deletions Side-by-side Diff

Documentation/cpusets.txt
Diff comments   View file @ b4fb376
... ... @@ -4,8 +4,9 @@
4 4 Copyright (C) 2004 BULL SA.
5 5 Written by Simon.Derr@bull.net
6 6  
7   -Portions Copyright (c) 2004 Silicon Graphics, Inc.
  7 +Portions Copyright (c) 2004-2006 Silicon Graphics, Inc.
8 8 Modified by Paul Jackson <pj@sgi.com>
  9 +Modified by Christoph Lameter <clameter@sgi.com>
9 10  
10 11 CONTENTS:
11 12 =========
... ... @@ -90,7 +91,8 @@
90 91  
91 92 These subsets, or "soft partitions" must be able to be dynamically
92 93 adjusted, as the job mix changes, without impacting other concurrently
93   -executing jobs.
  94 +executing jobs. The location of the running jobs pages may also be moved
  95 +when the memory locations are changed.
94 96  
95 97 The kernel cpuset patch provides the minimum essential kernel
96 98 mechanisms required to efficiently implement such subsets. It
... ... @@ -102,8 +104,8 @@
102 104 1.3 How are cpusets implemented ?
103 105 ---------------------------------
104 106  
105   -Cpusets provide a Linux kernel (2.6.7 and above) mechanism to constrain
106   -which CPUs and Memory Nodes are used by a process or set of processes.
  107 +Cpusets provide a Linux kernel mechanism to constrain which CPUs and
  108 +Memory Nodes are used by a process or set of processes.
107 109  
108 110 The Linux kernel already has a pair of mechanisms to specify on which
109 111 CPUs a task may be scheduled (sched_setaffinity) and on which Memory
110 112  
... ... @@ -371,22 +373,17 @@
371 373 If the cpuset flag file 'memory_migrate' is set true, then when
372 374 tasks are attached to that cpuset, any pages that task had
373 375 allocated to it on nodes in its previous cpuset are migrated
374   -to the tasks new cpuset. Depending on the implementation,
375   -this migration may either be done by swapping the page out,
376   -so that the next time the page is referenced, it will be paged
377   -into the tasks new cpuset, usually on the node where it was
378   -referenced, or this migration may be done by directly copying
379   -the pages from the tasks previous cpuset to the new cpuset,
380   -where possible to the same node, relative to the new cpuset,
381   -as the node that held the page, relative to the old cpuset.
  376 +to the tasks new cpuset. The relative placement of the page within
  377 +the cpuset is preserved during these migration operations if possible.
  378 +For example if the page was on the second valid node of the prior cpuset
  379 +then the page will be placed on the second valid node of the new cpuset.
  380 +
382 381 Also if 'memory_migrate' is set true, then if that cpusets
383 382 'mems' file is modified, pages allocated to tasks in that
384 383 cpuset, that were on nodes in the previous setting of 'mems',
385   -will be moved to nodes in the new setting of 'mems.' Again,
386   -depending on the implementation, this might be done by swapping,
387   -or by direct copying. In either case, pages that were not in
388   -the tasks prior cpuset, or in the cpusets prior 'mems' setting,
389   -will not be moved.
  384 +will be moved to nodes in the new setting of 'mems.'
  385 +Pages that were not in the tasks prior cpuset, or in the cpusets
  386 +prior 'mems' setting, will not be moved.
390 387  
391 388 There is an exception to the above. If hotplug functionality is used
392 389 to remove all the CPUs that are currently assigned to a cpuset,
... ... @@ -433,16 +430,6 @@
433 430 # The subshell 'sh' is now running in cpuset Charlie
434 431 # The next line should display '/Charlie'
435 432 cat /proc/self/cpuset
436   -
437   -In the case that a change of cpuset includes wanting to move already
438   -allocated memory pages, consider further the work of IWAMOTO
439   -Toshihiro <iwamoto@valinux.co.jp> for page remapping and memory
440   -hotremoval, which can be found at:
441   -
442   - http://people.valinux.co.jp/~iwamoto/mh.html
443   -
444   -The integration of cpusets with such memory migration is not yet
445   -available.
446 433  
447 434 In the future, a C library interface to cpusets will likely be
448 435 available. For now, the only way to query or modify cpusets is
Documentation/vm/page_migration
Diff comments   View file @ b4fb376
... ... @@ -12,12 +12,18 @@
12 12  
13 13 Page migration allows a process to manually relocate the node on which its
14 14 pages are located through the MF_MOVE and MF_MOVE_ALL options while setting
15   -a new memory policy. The pages of process can also be relocated
  15 +a new memory policy via mbind(). The pages of process can also be relocated
16 16 from another process using the sys_migrate_pages() function call. The
17 17 migrate_pages function call takes two sets of nodes and moves pages of a
18 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.
19 25  
20   -Manual migration is very useful if for example the scheduler has relocated
  26 +Manual migration is useful if for example the scheduler has relocated
21 27 a process to a processor on a distant node. A batch scheduler or an
22 28 administrator may detect the situation and move the pages of the process
23 29 nearer to the new processor. At some point in the future we may have
... ... @@ -25,10 +31,12 @@
25 31  
26 32 Larger installations usually partition the system using cpusets into
27 33 sections of nodes. Paul Jackson has equipped cpusets with the ability to
28   -move pages when a task is moved to another cpuset. This allows automatic
29   -control over locality of a process. If a task is moved to a new cpuset
30   -then also all its pages are moved with it so that the performance of the
31   -process does not sink dramatically (as is the case today).
  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.
32 40  
33 41 Page migration allows the preservation of the relative location of pages
34 42 within a group of nodes for all migration techniques which will preserve a
35 43  
36 44  
37 45  
38 46  
... ... @@ -37,22 +45,26 @@
37 45 Processes will run with similar performance after migration.
38 46  
39 47 Page migration occurs in several steps. First a high level
40   -description for those trying to use migrate_pages() and then
41   -a low level description of how the low level details work.
  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.
42 51  
43   -A. Use of migrate_pages()
44   --------------------------
  52 +A. In kernel use of migrate_pages()
  53 +-----------------------------------
45 54  
46 55 1. Remove pages from the LRU.
47 56  
48 57 Lists of pages to be migrated are generated by scanning over
49 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 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  
54   -2. Generate a list of newly allocates page to move the contents
55   - of the first list to.
  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.
56 68  
57 69 3. The migrate_pages() function is called which attempts
58 70 to do the migration. It returns the moved pages in the
59 71  
60 72  
... ... @@ -63,13 +75,17 @@
63 75 4. The leftover pages of various types are returned
64 76 to the LRU using putback_to_lru_pages() or otherwise
65 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  
68   -B. Operation of migrate_pages()
69   ---------------------------------
  82 +B. How migrate_pages() works
  83 +----------------------------
70 84  
71   -migrate_pages does several passes over its list of pages. A page is moved
72   -if all references to a page are removable at the time.
  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.
73 89  
74 90 Steps:
75 91  
76 92  
77 93  
78 94  
79 95  
80 96  
81 97  
82 98  
... ... @@ -79,36 +95,40 @@
79 95  
80 96 3. Make sure that the page has assigned swap cache entry if
81 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  
84 101 4. Prep the new page that we want to move to. It is locked
85 102 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  
88   -5. All the page table references to the page are either dropped (file backed)
89   - or converted to swap references (anonymous pages). This should decrease the
90   - reference count.
  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.
91 108  
92   -6. The radix tree lock is taken
  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.
93 111  
94 112 7. The refcount of the page is examined and we back out if references remain
95 113 otherwise we know that we are the only one referencing this page.
96 114  
97 115 8. 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  
100 118 9. The mapping is checked. If the mapping is gone then a truncate action may
101 119 be in progress and we back out.
102 120  
103   -10. The new page is prepped with some settings from the old page so that accesses
104   - to the new page will be discovered to have the correct settings.
  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.
105 123  
106 124 11. The radix tree is changed to point to the new page.
107 125  
108   -12. The reference count of the old page is dropped because the reference has now
109   - been removed.
  126 +12. The reference count of the old page is dropped because the radix tree
  127 + reference is gone.
110 128  
111   -13. The radix tree lock is dropped.
  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.
112 132  
113 133 14. The page contents are copied to the new page.
114 134  
115 135  
116 136  
117 137  
... ... @@ -119,11 +139,37 @@
119 139  
120 140 17. Queued up writeback on the new page is triggered.
121 141  
122   -18. If swap pte's were generated for the page then remove them again.
  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.
123 144  
124   -19. The locks are dropped from the old and new page.
  145 +19. The page locks are dropped from the old and new page.
  146 + Processes waiting on the page lock can continue.
125 147  
126   -20. The new page is moved to the LRU.
  148 +20. The new page is moved to the LRU and can be scanned by the swapper
  149 + etc again.
127 150  
128   -Christoph Lameter, December 19, 2005.
  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.