| Christoph Lameter | a48d07a | 2006-02-01 03:05:38 -0800 | [diff] [blame^] | 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. 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 | |
| 20 | Manual migration is very useful if for example the scheduler has relocated |
| 21 | a process to a processor on a distant node. A batch scheduler or an |
| 22 | administrator may detect the situation and move the pages of the process |
| 23 | nearer to the new processor. At some point in the future we may have |
| 24 | some mechanism in the scheduler that will automatically move the pages. |
| 25 | |
| 26 | Larger installations usually partition the system using cpusets into |
| 27 | 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). |
| 32 | |
| 33 | Page migration allows the preservation of the relative location of pages |
| 34 | within a group of nodes for all migration techniques which will preserve a |
| 35 | particular memory allocation pattern generated even after migrating a |
| 36 | process. This is necessary in order to preserve the memory latencies. |
| 37 | Processes will run with similar performance after migration. |
| 38 | |
| 39 | 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. |
| 42 | |
| 43 | A. Use of migrate_pages() |
| 44 | ------------------------- |
| 45 | |
| 46 | 1. Remove pages from the LRU. |
| 47 | |
| 48 | Lists of pages to be migrated are generated by scanning over |
| 49 | pages and moving them into lists. This is done by |
| 50 | calling isolate_lru_page() or __isolate_lru_page(). |
| 51 | Calling isolate_lru_page increases the references to the page |
| 52 | so that it cannot vanish under us. |
| 53 | |
| 54 | 2. Generate a list of newly allocates page to move the contents |
| 55 | of the first list to. |
| 56 | |
| 57 | 3. The migrate_pages() function is called which attempts |
| 58 | to do the migration. It returns the moved pages in the |
| 59 | list specified as the third parameter and the failed |
| 60 | migrations in the fourth parameter. The first parameter |
| 61 | will contain the pages that could still be retried. |
| 62 | |
| 63 | 4. The leftover pages of various types are returned |
| 64 | to the LRU using putback_to_lru_pages() or otherwise |
| 65 | disposed of. The pages will still have the refcount as |
| 66 | increased by isolate_lru_pages()! |
| 67 | |
| 68 | B. Operation of migrate_pages() |
| 69 | -------------------------------- |
| 70 | |
| 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. |
| 73 | |
| 74 | Steps: |
| 75 | |
| 76 | 1. Lock the page to be migrated |
| 77 | |
| 78 | 2. Insure that writeback is complete. |
| 79 | |
| 80 | 3. Make sure that the page has assigned swap cache entry if |
| 81 | it is an anonyous page. The swap cache reference is necessary |
| 82 | to preserve the information contain in the page table maps. |
| 83 | |
| 84 | 4. 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 |
| 86 | page immediately lock while we are moving references. |
| 87 | |
| 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. |
| 91 | |
| 92 | 6. The radix tree lock is taken |
| 93 | |
| 94 | 7. 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. |
| 96 | |
| 97 | 8. The radix tree is checked and if it does not contain the pointer to this |
| 98 | page then we back out. |
| 99 | |
| 100 | 9. The mapping is checked. If the mapping is gone then a truncate action may |
| 101 | be in progress and we back out. |
| 102 | |
| 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. |
| 105 | |
| 106 | 11. The radix tree is changed to point to the new page. |
| 107 | |
| 108 | 12. The reference count of the old page is dropped because the reference has now |
| 109 | been removed. |
| 110 | |
| 111 | 13. The radix tree lock is dropped. |
| 112 | |
| 113 | 14. The page contents are copied to the new page. |
| 114 | |
| 115 | 15. The remaining page flags are copied to the new page. |
| 116 | |
| 117 | 16. The old page flags are cleared to indicate that the page does |
| 118 | not use any information anymore. |
| 119 | |
| 120 | 17. Queued up writeback on the new page is triggered. |
| 121 | |
| 122 | 18. If swap pte's were generated for the page then remove them again. |
| 123 | |
| 124 | 19. The locks are dropped from the old and new page. |
| 125 | |
| 126 | 20. The new page is moved to the LRU. |
| 127 | |
| 128 | Christoph Lameter, December 19, 2005. |
| 129 | |