]>
Commit | Line | Data |
---|---|---|
00f0b825 BS |
1 | Memory Resource Controller |
2 | ||
3 | NOTE: The Memory Resource Controller has been generically been referred | |
dc10e281 KH |
4 | to as the memory controller in this document. Do not confuse memory |
5 | controller used here with the memory controller that is used in hardware. | |
1b6df3aa | 6 | |
dc10e281 KH |
7 | (For editors) |
8 | In this document: | |
9 | When we mention a cgroup (cgroupfs's directory) with memory controller, | |
10 | we call it "memory cgroup". When you see git-log and source code, you'll | |
11 | see patch's title and function names tend to use "memcg". | |
12 | In this document, we avoid using it. | |
1b6df3aa | 13 | |
1b6df3aa BS |
14 | Benefits and Purpose of the memory controller |
15 | ||
16 | The memory controller isolates the memory behaviour of a group of tasks | |
17 | from the rest of the system. The article on LWN [12] mentions some probable | |
18 | uses of the memory controller. The memory controller can be used to | |
19 | ||
20 | a. Isolate an application or a group of applications | |
21 | Memory hungry applications can be isolated and limited to a smaller | |
22 | amount of memory. | |
23 | b. Create a cgroup with limited amount of memory, this can be used | |
24 | as a good alternative to booting with mem=XXXX. | |
25 | c. Virtualization solutions can control the amount of memory they want | |
26 | to assign to a virtual machine instance. | |
27 | d. A CD/DVD burner could control the amount of memory used by the | |
28 | rest of the system to ensure that burning does not fail due to lack | |
29 | of available memory. | |
30 | e. There are several other use cases, find one or use the controller just | |
31 | for fun (to learn and hack on the VM subsystem). | |
32 | ||
dc10e281 KH |
33 | Current Status: linux-2.6.34-mmotm(development version of 2010/April) |
34 | ||
35 | Features: | |
36 | - accounting anonymous pages, file caches, swap caches usage and limiting them. | |
37 | - private LRU and reclaim routine. (system's global LRU and private LRU | |
38 | work independently from each other) | |
39 | - optionally, memory+swap usage can be accounted and limited. | |
40 | - hierarchical accounting | |
41 | - soft limit | |
42 | - moving(recharging) account at moving a task is selectable. | |
43 | - usage threshold notifier | |
44 | - oom-killer disable knob and oom-notifier | |
45 | - Root cgroup has no limit controls. | |
46 | ||
47 | Kernel memory and Hugepages are not under control yet. We just manage | |
48 | pages on LRU. To add more controls, we have to take care of performance. | |
49 | ||
50 | Brief summary of control files. | |
51 | ||
52 | tasks # attach a task(thread) and show list of threads | |
53 | cgroup.procs # show list of processes | |
54 | cgroup.event_control # an interface for event_fd() | |
55 | memory.usage_in_bytes # show current memory(RSS+Cache) usage. | |
56 | memory.memsw.usage_in_bytes # show current memory+Swap usage | |
57 | memory.limit_in_bytes # set/show limit of memory usage | |
58 | memory.memsw.limit_in_bytes # set/show limit of memory+Swap usage | |
59 | memory.failcnt # show the number of memory usage hits limits | |
60 | memory.memsw.failcnt # show the number of memory+Swap hits limits | |
61 | memory.max_usage_in_bytes # show max memory usage recorded | |
62 | memory.memsw.usage_in_bytes # show max memory+Swap usage recorded | |
63 | memory.soft_limit_in_bytes # set/show soft limit of memory usage | |
64 | memory.stat # show various statistics | |
65 | memory.use_hierarchy # set/show hierarchical account enabled | |
66 | memory.force_empty # trigger forced move charge to parent | |
67 | memory.swappiness # set/show swappiness parameter of vmscan | |
68 | (See sysctl's vm.swappiness) | |
69 | memory.move_charge_at_immigrate # set/show controls of moving charges | |
70 | memory.oom_control # set/show oom controls. | |
71 | ||
1b6df3aa BS |
72 | 1. History |
73 | ||
74 | The memory controller has a long history. A request for comments for the memory | |
75 | controller was posted by Balbir Singh [1]. At the time the RFC was posted | |
76 | there were several implementations for memory control. The goal of the | |
77 | RFC was to build consensus and agreement for the minimal features required | |
78 | for memory control. The first RSS controller was posted by Balbir Singh[2] | |
79 | in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the | |
80 | RSS controller. At OLS, at the resource management BoF, everyone suggested | |
81 | that we handle both page cache and RSS together. Another request was raised | |
82 | to allow user space handling of OOM. The current memory controller is | |
83 | at version 6; it combines both mapped (RSS) and unmapped Page | |
84 | Cache Control [11]. | |
85 | ||
86 | 2. Memory Control | |
87 | ||
88 | Memory is a unique resource in the sense that it is present in a limited | |
89 | amount. If a task requires a lot of CPU processing, the task can spread | |
90 | its processing over a period of hours, days, months or years, but with | |
91 | memory, the same physical memory needs to be reused to accomplish the task. | |
92 | ||
93 | The memory controller implementation has been divided into phases. These | |
94 | are: | |
95 | ||
96 | 1. Memory controller | |
97 | 2. mlock(2) controller | |
98 | 3. Kernel user memory accounting and slab control | |
99 | 4. user mappings length controller | |
100 | ||
101 | The memory controller is the first controller developed. | |
102 | ||
103 | 2.1. Design | |
104 | ||
105 | The core of the design is a counter called the res_counter. The res_counter | |
106 | tracks the current memory usage and limit of the group of processes associated | |
107 | with the controller. Each cgroup has a memory controller specific data | |
108 | structure (mem_cgroup) associated with it. | |
109 | ||
110 | 2.2. Accounting | |
111 | ||
112 | +--------------------+ | |
113 | | mem_cgroup | | |
114 | | (res_counter) | | |
115 | +--------------------+ | |
116 | / ^ \ | |
117 | / | \ | |
118 | +---------------+ | +---------------+ | |
119 | | mm_struct | |.... | mm_struct | | |
120 | | | | | | | |
121 | +---------------+ | +---------------+ | |
122 | | | |
123 | + --------------+ | |
124 | | | |
125 | +---------------+ +------+--------+ | |
126 | | page +----------> page_cgroup| | |
127 | | | | | | |
128 | +---------------+ +---------------+ | |
129 | ||
130 | (Figure 1: Hierarchy of Accounting) | |
131 | ||
132 | ||
133 | Figure 1 shows the important aspects of the controller | |
134 | ||
135 | 1. Accounting happens per cgroup | |
136 | 2. Each mm_struct knows about which cgroup it belongs to | |
137 | 3. Each page has a pointer to the page_cgroup, which in turn knows the | |
138 | cgroup it belongs to | |
139 | ||
140 | The accounting is done as follows: mem_cgroup_charge() is invoked to setup | |
141 | the necessary data structures and check if the cgroup that is being charged | |
142 | is over its limit. If it is then reclaim is invoked on the cgroup. | |
143 | More details can be found in the reclaim section of this document. | |
144 | If everything goes well, a page meta-data-structure called page_cgroup is | |
dc10e281 KH |
145 | updated. page_cgroup has its own LRU on cgroup. |
146 | (*) page_cgroup structure is allocated at boot/memory-hotplug time. | |
1b6df3aa BS |
147 | |
148 | 2.2.1 Accounting details | |
149 | ||
5b4e655e | 150 | All mapped anon pages (RSS) and cache pages (Page Cache) are accounted. |
dc10e281 KH |
151 | Some pages which are never reclaimable and will not be on the global LRU |
152 | are not accounted. We just account pages under usual VM management. | |
5b4e655e KH |
153 | |
154 | RSS pages are accounted at page_fault unless they've already been accounted | |
155 | for earlier. A file page will be accounted for as Page Cache when it's | |
156 | inserted into inode (radix-tree). While it's mapped into the page tables of | |
157 | processes, duplicate accounting is carefully avoided. | |
158 | ||
159 | A RSS page is unaccounted when it's fully unmapped. A PageCache page is | |
dc10e281 KH |
160 | unaccounted when it's removed from radix-tree. Even if RSS pages are fully |
161 | unmapped (by kswapd), they may exist as SwapCache in the system until they | |
162 | are really freed. Such SwapCaches also also accounted. | |
163 | A swapped-in page is not accounted until it's mapped. | |
164 | ||
165 | Note: The kernel does swapin-readahead and read multiple swaps at once. | |
166 | This means swapped-in pages may contain pages for other tasks than a task | |
167 | causing page fault. So, we avoid accounting at swap-in I/O. | |
5b4e655e KH |
168 | |
169 | At page migration, accounting information is kept. | |
170 | ||
dc10e281 KH |
171 | Note: we just account pages-on-LRU because our purpose is to control amount |
172 | of used pages; not-on-LRU pages tend to be out-of-control from VM view. | |
1b6df3aa BS |
173 | |
174 | 2.3 Shared Page Accounting | |
175 | ||
176 | Shared pages are accounted on the basis of the first touch approach. The | |
177 | cgroup that first touches a page is accounted for the page. The principle | |
178 | behind this approach is that a cgroup that aggressively uses a shared | |
179 | page will eventually get charged for it (once it is uncharged from | |
180 | the cgroup that brought it in -- this will happen on memory pressure). | |
181 | ||
8c7c6e34 KH |
182 | Exception: If CONFIG_CGROUP_CGROUP_MEM_RES_CTLR_SWAP is not used.. |
183 | When you do swapoff and make swapped-out pages of shmem(tmpfs) to | |
d13d1443 KH |
184 | be backed into memory in force, charges for pages are accounted against the |
185 | caller of swapoff rather than the users of shmem. | |
186 | ||
187 | ||
8c7c6e34 | 188 | 2.4 Swap Extension (CONFIG_CGROUP_MEM_RES_CTLR_SWAP) |
dc10e281 | 189 | |
8c7c6e34 KH |
190 | Swap Extension allows you to record charge for swap. A swapped-in page is |
191 | charged back to original page allocator if possible. | |
192 | ||
193 | When swap is accounted, following files are added. | |
194 | - memory.memsw.usage_in_bytes. | |
195 | - memory.memsw.limit_in_bytes. | |
196 | ||
dc10e281 KH |
197 | memsw means memory+swap. Usage of memory+swap is limited by |
198 | memsw.limit_in_bytes. | |
199 | ||
200 | Example: Assume a system with 4G of swap. A task which allocates 6G of memory | |
201 | (by mistake) under 2G memory limitation will use all swap. | |
202 | In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap. | |
203 | By using memsw limit, you can avoid system OOM which can be caused by swap | |
204 | shortage. | |
8c7c6e34 | 205 | |
dc10e281 | 206 | * why 'memory+swap' rather than swap. |
8c7c6e34 KH |
207 | The global LRU(kswapd) can swap out arbitrary pages. Swap-out means |
208 | to move account from memory to swap...there is no change in usage of | |
dc10e281 KH |
209 | memory+swap. In other words, when we want to limit the usage of swap without |
210 | affecting global LRU, memory+swap limit is better than just limiting swap from | |
22a668d7 KH |
211 | OS point of view. |
212 | ||
213 | * What happens when a cgroup hits memory.memsw.limit_in_bytes | |
214 | When a cgroup his memory.memsw.limit_in_bytes, it's useless to do swap-out | |
215 | in this cgroup. Then, swap-out will not be done by cgroup routine and file | |
216 | caches are dropped. But as mentioned above, global LRU can do swapout memory | |
217 | from it for sanity of the system's memory management state. You can't forbid | |
218 | it by cgroup. | |
8c7c6e34 KH |
219 | |
220 | 2.5 Reclaim | |
1b6df3aa | 221 | |
dc10e281 KH |
222 | Each cgroup maintains a per cgroup LRU which has the same structure as |
223 | global VM. When a cgroup goes over its limit, we first try | |
1b6df3aa BS |
224 | to reclaim memory from the cgroup so as to make space for the new |
225 | pages that the cgroup has touched. If the reclaim is unsuccessful, | |
226 | an OOM routine is invoked to select and kill the bulkiest task in the | |
dc10e281 | 227 | cgroup. (See 10. OOM Control below.) |
1b6df3aa BS |
228 | |
229 | The reclaim algorithm has not been modified for cgroups, except that | |
230 | pages that are selected for reclaiming come from the per cgroup LRU | |
231 | list. | |
232 | ||
4b3bde4c BS |
233 | NOTE: Reclaim does not work for the root cgroup, since we cannot set any |
234 | limits on the root cgroup. | |
235 | ||
daaf1e68 KH |
236 | Note2: When panic_on_oom is set to "2", the whole system will panic. |
237 | ||
9490ff27 KH |
238 | When oom event notifier is registered, event will be delivered. |
239 | (See oom_control section) | |
240 | ||
dc10e281 | 241 | 2.6 Locking |
1b6df3aa | 242 | |
dc10e281 KH |
243 | lock_page_cgroup()/unlock_page_cgroup() should not be called under |
244 | mapping->tree_lock. | |
1b6df3aa | 245 | |
dc10e281 KH |
246 | Other lock order is following: |
247 | PG_locked. | |
248 | mm->page_table_lock | |
249 | zone->lru_lock | |
250 | lock_page_cgroup. | |
251 | In many cases, just lock_page_cgroup() is called. | |
252 | per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by | |
253 | zone->lru_lock, it has no lock of its own. | |
1b6df3aa BS |
254 | |
255 | 3. User Interface | |
256 | ||
257 | 0. Configuration | |
258 | ||
259 | a. Enable CONFIG_CGROUPS | |
260 | b. Enable CONFIG_RESOURCE_COUNTERS | |
00f0b825 | 261 | c. Enable CONFIG_CGROUP_MEM_RES_CTLR |
dc10e281 | 262 | d. Enable CONFIG_CGROUP_MEM_RES_CTLR_SWAP (to use swap extension) |
1b6df3aa BS |
263 | |
264 | 1. Prepare the cgroups | |
265 | # mkdir -p /cgroups | |
266 | # mount -t cgroup none /cgroups -o memory | |
267 | ||
268 | 2. Make the new group and move bash into it | |
269 | # mkdir /cgroups/0 | |
dc10e281 | 270 | # echo $$ > /cgroups/0/tasks |
1b6df3aa | 271 | |
dc10e281 | 272 | Since now we're in the 0 cgroup, we can alter the memory limit: |
fb78922c | 273 | # echo 4M > /cgroups/0/memory.limit_in_bytes |
0eea1030 BS |
274 | |
275 | NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo, | |
dc10e281 KH |
276 | mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.) |
277 | ||
c5b947b2 | 278 | NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited). |
4b3bde4c | 279 | NOTE: We cannot set limits on the root cgroup any more. |
0eea1030 BS |
280 | |
281 | # cat /cgroups/0/memory.limit_in_bytes | |
2324c5dd | 282 | 4194304 |
0eea1030 | 283 | |
1b6df3aa | 284 | We can check the usage: |
0eea1030 | 285 | # cat /cgroups/0/memory.usage_in_bytes |
2324c5dd | 286 | 1216512 |
0eea1030 BS |
287 | |
288 | A successful write to this file does not guarantee a successful set of | |
dc10e281 | 289 | this limit to the value written into the file. This can be due to a |
0eea1030 | 290 | number of factors, such as rounding up to page boundaries or the total |
dc10e281 | 291 | availability of memory on the system. The user is required to re-read |
0eea1030 BS |
292 | this file after a write to guarantee the value committed by the kernel. |
293 | ||
fb78922c | 294 | # echo 1 > memory.limit_in_bytes |
0eea1030 | 295 | # cat memory.limit_in_bytes |
2324c5dd | 296 | 4096 |
1b6df3aa BS |
297 | |
298 | The memory.failcnt field gives the number of times that the cgroup limit was | |
299 | exceeded. | |
300 | ||
dfc05c25 KH |
301 | The memory.stat file gives accounting information. Now, the number of |
302 | caches, RSS and Active pages/Inactive pages are shown. | |
303 | ||
1b6df3aa BS |
304 | 4. Testing |
305 | ||
dc10e281 KH |
306 | For testing features and implementation, see memcg_test.txt. |
307 | ||
308 | Performance test is also important. To see pure memory controller's overhead, | |
309 | testing on tmpfs will give you good numbers of small overheads. | |
310 | Example: do kernel make on tmpfs. | |
311 | ||
312 | Page-fault scalability is also important. At measuring parallel | |
313 | page fault test, multi-process test may be better than multi-thread | |
314 | test because it has noise of shared objects/status. | |
315 | ||
316 | But the above two are testing extreme situations. | |
317 | Trying usual test under memory controller is always helpful. | |
1b6df3aa BS |
318 | |
319 | 4.1 Troubleshooting | |
320 | ||
321 | Sometimes a user might find that the application under a cgroup is | |
dc10e281 | 322 | terminated by OOM killer. There are several causes for this: |
1b6df3aa BS |
323 | |
324 | 1. The cgroup limit is too low (just too low to do anything useful) | |
325 | 2. The user is using anonymous memory and swap is turned off or too low | |
326 | ||
327 | A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of | |
328 | some of the pages cached in the cgroup (page cache pages). | |
329 | ||
dc10e281 KH |
330 | To know what happens, disable OOM_Kill by 10. OOM Control(see below) and |
331 | seeing what happens will be helpful. | |
332 | ||
1b6df3aa BS |
333 | 4.2 Task migration |
334 | ||
a33f3224 | 335 | When a task migrates from one cgroup to another, its charge is not |
7dc74be0 | 336 | carried forward by default. The pages allocated from the original cgroup still |
1b6df3aa BS |
337 | remain charged to it, the charge is dropped when the page is freed or |
338 | reclaimed. | |
339 | ||
dc10e281 KH |
340 | You can move charges of a task along with task migration. |
341 | See 8. "Move charges at task migration" | |
7dc74be0 | 342 | |
1b6df3aa BS |
343 | 4.3 Removing a cgroup |
344 | ||
345 | A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a | |
346 | cgroup might have some charge associated with it, even though all | |
dc10e281 KH |
347 | tasks have migrated away from it. (because we charge against pages, not |
348 | against tasks.) | |
349 | ||
350 | Such charges are freed or moved to their parent. At moving, both of RSS | |
351 | and CACHES are moved to parent. | |
352 | rmdir() may return -EBUSY if freeing/moving fails. See 5.1 also. | |
1b6df3aa | 353 | |
8c7c6e34 KH |
354 | Charges recorded in swap information is not updated at removal of cgroup. |
355 | Recorded information is discarded and a cgroup which uses swap (swapcache) | |
356 | will be charged as a new owner of it. | |
357 | ||
358 | ||
c1e862c1 KH |
359 | 5. Misc. interfaces. |
360 | ||
361 | 5.1 force_empty | |
362 | memory.force_empty interface is provided to make cgroup's memory usage empty. | |
363 | You can use this interface only when the cgroup has no tasks. | |
364 | When writing anything to this | |
365 | ||
366 | # echo 0 > memory.force_empty | |
367 | ||
dc10e281 KH |
368 | Almost all pages tracked by this memory cgroup will be unmapped and freed. |
369 | Some pages cannot be freed because they are locked or in-use. Such pages are | |
370 | moved to parent and this cgroup will be empty. This may return -EBUSY if | |
371 | VM is too busy to free/move all pages immediately. | |
c1e862c1 KH |
372 | |
373 | Typical use case of this interface is that calling this before rmdir(). | |
374 | Because rmdir() moves all pages to parent, some out-of-use page caches can be | |
375 | moved to the parent. If you want to avoid that, force_empty will be useful. | |
376 | ||
7f016ee8 | 377 | 5.2 stat file |
c863d835 BR |
378 | |
379 | memory.stat file includes following statistics | |
380 | ||
dc10e281 | 381 | # per-memory cgroup local status |
c863d835 BR |
382 | cache - # of bytes of page cache memory. |
383 | rss - # of bytes of anonymous and swap cache memory. | |
dc10e281 | 384 | mapped_file - # of bytes of mapped file (includes tmpfs/shmem) |
c863d835 BR |
385 | pgpgin - # of pages paged in (equivalent to # of charging events). |
386 | pgpgout - # of pages paged out (equivalent to # of uncharging events). | |
dc10e281 | 387 | swap - # of bytes of swap usage |
c863d835 | 388 | inactive_anon - # of bytes of anonymous memory and swap cache memory on |
dc10e281 KH |
389 | LRU list. |
390 | active_anon - # of bytes of anonymous and swap cache memory on active | |
391 | inactive LRU list. | |
392 | inactive_file - # of bytes of file-backed memory on inactive LRU list. | |
393 | active_file - # of bytes of file-backed memory on active LRU list. | |
c863d835 BR |
394 | unevictable - # of bytes of memory that cannot be reclaimed (mlocked etc). |
395 | ||
dc10e281 KH |
396 | # status considering hierarchy (see memory.use_hierarchy settings) |
397 | ||
398 | hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy | |
399 | under which the memory cgroup is | |
400 | hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to | |
401 | hierarchy under which memory cgroup is. | |
402 | ||
403 | total_cache - sum of all children's "cache" | |
404 | total_rss - sum of all children's "rss" | |
405 | total_mapped_file - sum of all children's "cache" | |
406 | total_pgpgin - sum of all children's "pgpgin" | |
407 | total_pgpgout - sum of all children's "pgpgout" | |
408 | total_swap - sum of all children's "swap" | |
409 | total_inactive_anon - sum of all children's "inactive_anon" | |
410 | total_active_anon - sum of all children's "active_anon" | |
411 | total_inactive_file - sum of all children's "inactive_file" | |
412 | total_active_file - sum of all children's "active_file" | |
413 | total_unevictable - sum of all children's "unevictable" | |
414 | ||
415 | # The following additional stats are dependent on CONFIG_DEBUG_VM. | |
c863d835 BR |
416 | |
417 | inactive_ratio - VM internal parameter. (see mm/page_alloc.c) | |
418 | recent_rotated_anon - VM internal parameter. (see mm/vmscan.c) | |
419 | recent_rotated_file - VM internal parameter. (see mm/vmscan.c) | |
420 | recent_scanned_anon - VM internal parameter. (see mm/vmscan.c) | |
421 | recent_scanned_file - VM internal parameter. (see mm/vmscan.c) | |
422 | ||
423 | Memo: | |
dc10e281 KH |
424 | recent_rotated means recent frequency of LRU rotation. |
425 | recent_scanned means recent # of scans to LRU. | |
7f016ee8 KM |
426 | showing for better debug please see the code for meanings. |
427 | ||
c863d835 BR |
428 | Note: |
429 | Only anonymous and swap cache memory is listed as part of 'rss' stat. | |
430 | This should not be confused with the true 'resident set size' or the | |
dc10e281 KH |
431 | amount of physical memory used by the cgroup. |
432 | 'rss + file_mapped" will give you resident set size of cgroup. | |
433 | (Note: file and shmem may be shared among other cgroups. In that case, | |
434 | file_mapped is accounted only when the memory cgroup is owner of page | |
435 | cache.) | |
7f016ee8 | 436 | |
a7885eb8 | 437 | 5.3 swappiness |
a7885eb8 | 438 | |
dc10e281 | 439 | Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only. |
a7885eb8 | 440 | |
dc10e281 KH |
441 | Following cgroups' swappiness can't be changed. |
442 | - root cgroup (uses /proc/sys/vm/swappiness). | |
443 | - a cgroup which uses hierarchy and it has other cgroup(s) below it. | |
444 | - a cgroup which uses hierarchy and not the root of hierarchy. | |
445 | ||
446 | 5.4 failcnt | |
447 | ||
448 | A memory cgroup provides memory.failcnt and memory.memsw.failcnt files. | |
449 | This failcnt(== failure count) shows the number of times that a usage counter | |
450 | hit its limit. When a memory cgroup hits a limit, failcnt increases and | |
451 | memory under it will be reclaimed. | |
452 | ||
453 | You can reset failcnt by writing 0 to failcnt file. | |
454 | # echo 0 > .../memory.failcnt | |
a7885eb8 | 455 | |
52bc0d82 | 456 | 6. Hierarchy support |
c1e862c1 | 457 | |
52bc0d82 BS |
458 | The memory controller supports a deep hierarchy and hierarchical accounting. |
459 | The hierarchy is created by creating the appropriate cgroups in the | |
460 | cgroup filesystem. Consider for example, the following cgroup filesystem | |
461 | hierarchy | |
462 | ||
463 | root | |
464 | / | \ | |
465 | / | \ | |
466 | a b c | |
467 | | \ | |
468 | | \ | |
469 | d e | |
470 | ||
471 | In the diagram above, with hierarchical accounting enabled, all memory | |
472 | usage of e, is accounted to its ancestors up until the root (i.e, c and root), | |
dc10e281 | 473 | that has memory.use_hierarchy enabled. If one of the ancestors goes over its |
52bc0d82 BS |
474 | limit, the reclaim algorithm reclaims from the tasks in the ancestor and the |
475 | children of the ancestor. | |
476 | ||
477 | 6.1 Enabling hierarchical accounting and reclaim | |
478 | ||
dc10e281 | 479 | A memory cgroup by default disables the hierarchy feature. Support |
52bc0d82 BS |
480 | can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup |
481 | ||
482 | # echo 1 > memory.use_hierarchy | |
483 | ||
484 | The feature can be disabled by | |
485 | ||
486 | # echo 0 > memory.use_hierarchy | |
487 | ||
488 | NOTE1: Enabling/disabling will fail if the cgroup already has other | |
dc10e281 | 489 | cgroups created below it. |
52bc0d82 | 490 | |
daaf1e68 | 491 | NOTE2: When panic_on_oom is set to "2", the whole system will panic in |
dc10e281 | 492 | case of an OOM event in any cgroup. |
52bc0d82 | 493 | |
a6df6361 BS |
494 | 7. Soft limits |
495 | ||
496 | Soft limits allow for greater sharing of memory. The idea behind soft limits | |
497 | is to allow control groups to use as much of the memory as needed, provided | |
498 | ||
499 | a. There is no memory contention | |
500 | b. They do not exceed their hard limit | |
501 | ||
dc10e281 | 502 | When the system detects memory contention or low memory, control groups |
a6df6361 BS |
503 | are pushed back to their soft limits. If the soft limit of each control |
504 | group is very high, they are pushed back as much as possible to make | |
505 | sure that one control group does not starve the others of memory. | |
506 | ||
507 | Please note that soft limits is a best effort feature, it comes with | |
508 | no guarantees, but it does its best to make sure that when memory is | |
509 | heavily contended for, memory is allocated based on the soft limit | |
510 | hints/setup. Currently soft limit based reclaim is setup such that | |
511 | it gets invoked from balance_pgdat (kswapd). | |
512 | ||
513 | 7.1 Interface | |
514 | ||
515 | Soft limits can be setup by using the following commands (in this example we | |
dc10e281 | 516 | assume a soft limit of 256 MiB) |
a6df6361 BS |
517 | |
518 | # echo 256M > memory.soft_limit_in_bytes | |
519 | ||
520 | If we want to change this to 1G, we can at any time use | |
521 | ||
522 | # echo 1G > memory.soft_limit_in_bytes | |
523 | ||
524 | NOTE1: Soft limits take effect over a long period of time, since they involve | |
525 | reclaiming memory for balancing between memory cgroups | |
526 | NOTE2: It is recommended to set the soft limit always below the hard limit, | |
527 | otherwise the hard limit will take precedence. | |
528 | ||
7dc74be0 DN |
529 | 8. Move charges at task migration |
530 | ||
531 | Users can move charges associated with a task along with task migration, that | |
532 | is, uncharge task's pages from the old cgroup and charge them to the new cgroup. | |
02491447 DN |
533 | This feature is not supported in !CONFIG_MMU environments because of lack of |
534 | page tables. | |
7dc74be0 DN |
535 | |
536 | 8.1 Interface | |
537 | ||
538 | This feature is disabled by default. It can be enabled(and disabled again) by | |
539 | writing to memory.move_charge_at_immigrate of the destination cgroup. | |
540 | ||
541 | If you want to enable it: | |
542 | ||
543 | # echo (some positive value) > memory.move_charge_at_immigrate | |
544 | ||
545 | Note: Each bits of move_charge_at_immigrate has its own meaning about what type | |
546 | of charges should be moved. See 8.2 for details. | |
547 | Note: Charges are moved only when you move mm->owner, IOW, a leader of a thread | |
548 | group. | |
549 | Note: If we cannot find enough space for the task in the destination cgroup, we | |
550 | try to make space by reclaiming memory. Task migration may fail if we | |
551 | cannot make enough space. | |
dc10e281 | 552 | Note: It can take several seconds if you move charges much. |
7dc74be0 DN |
553 | |
554 | And if you want disable it again: | |
555 | ||
556 | # echo 0 > memory.move_charge_at_immigrate | |
557 | ||
558 | 8.2 Type of charges which can be move | |
559 | ||
560 | Each bits of move_charge_at_immigrate has its own meaning about what type of | |
87946a72 DN |
561 | charges should be moved. But in any cases, it must be noted that an account of |
562 | a page or a swap can be moved only when it is charged to the task's current(old) | |
563 | memory cgroup. | |
7dc74be0 DN |
564 | |
565 | bit | what type of charges would be moved ? | |
566 | -----+------------------------------------------------------------------------ | |
567 | 0 | A charge of an anonymous page(or swap of it) used by the target task. | |
568 | | Those pages and swaps must be used only by the target task. You must | |
569 | | enable Swap Extension(see 2.4) to enable move of swap charges. | |
87946a72 DN |
570 | -----+------------------------------------------------------------------------ |
571 | 1 | A charge of file pages(normal file, tmpfs file(e.g. ipc shared memory) | |
dc10e281 | 572 | | and swaps of tmpfs file) mmapped by the target task. Unlike the case of |
87946a72 DN |
573 | | anonymous pages, file pages(and swaps) in the range mmapped by the task |
574 | | will be moved even if the task hasn't done page fault, i.e. they might | |
575 | | not be the task's "RSS", but other task's "RSS" that maps the same file. | |
576 | | And mapcount of the page is ignored(the page can be moved even if | |
577 | | page_mapcount(page) > 1). You must enable Swap Extension(see 2.4) to | |
578 | | enable move of swap charges. | |
7dc74be0 DN |
579 | |
580 | 8.3 TODO | |
581 | ||
7dc74be0 DN |
582 | - Implement madvise(2) to let users decide the vma to be moved or not to be |
583 | moved. | |
584 | - All of moving charge operations are done under cgroup_mutex. It's not good | |
585 | behavior to hold the mutex too long, so we may need some trick. | |
586 | ||
2e72b634 KS |
587 | 9. Memory thresholds |
588 | ||
dc10e281 | 589 | Memory cgroup implements memory thresholds using cgroups notification |
2e72b634 KS |
590 | API (see cgroups.txt). It allows to register multiple memory and memsw |
591 | thresholds and gets notifications when it crosses. | |
592 | ||
593 | To register a threshold application need: | |
dc10e281 KH |
594 | - create an eventfd using eventfd(2); |
595 | - open memory.usage_in_bytes or memory.memsw.usage_in_bytes; | |
596 | - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to | |
597 | cgroup.event_control. | |
2e72b634 KS |
598 | |
599 | Application will be notified through eventfd when memory usage crosses | |
600 | threshold in any direction. | |
601 | ||
602 | It's applicable for root and non-root cgroup. | |
603 | ||
9490ff27 KH |
604 | 10. OOM Control |
605 | ||
3c11ecf4 KH |
606 | memory.oom_control file is for OOM notification and other controls. |
607 | ||
dc10e281 KH |
608 | Memory cgroup implements OOM notifier using cgroup notification |
609 | API (See cgroups.txt). It allows to register multiple OOM notification | |
610 | delivery and gets notification when OOM happens. | |
9490ff27 KH |
611 | |
612 | To register a notifier, application need: | |
613 | - create an eventfd using eventfd(2) | |
614 | - open memory.oom_control file | |
dc10e281 KH |
615 | - write string like "<event_fd> <fd of memory.oom_control>" to |
616 | cgroup.event_control | |
9490ff27 | 617 | |
dc10e281 | 618 | Application will be notified through eventfd when OOM happens. |
9490ff27 KH |
619 | OOM notification doesn't work for root cgroup. |
620 | ||
dc10e281 KH |
621 | You can disable OOM-killer by writing "1" to memory.oom_control file, as: |
622 | ||
3c11ecf4 KH |
623 | #echo 1 > memory.oom_control |
624 | ||
dc10e281 KH |
625 | This operation is only allowed to the top cgroup of sub-hierarchy. |
626 | If OOM-killer is disabled, tasks under cgroup will hang/sleep | |
627 | in memory cgroup's OOM-waitqueue when they request accountable memory. | |
3c11ecf4 | 628 | |
dc10e281 | 629 | For running them, you have to relax the memory cgroup's OOM status by |
3c11ecf4 KH |
630 | * enlarge limit or reduce usage. |
631 | To reduce usage, | |
632 | * kill some tasks. | |
633 | * move some tasks to other group with account migration. | |
634 | * remove some files (on tmpfs?) | |
635 | ||
636 | Then, stopped tasks will work again. | |
637 | ||
638 | At reading, current status of OOM is shown. | |
639 | oom_kill_disable 0 or 1 (if 1, oom-killer is disabled) | |
dc10e281 | 640 | under_oom 0 or 1 (if 1, the memory cgroup is under OOM, tasks may |
3c11ecf4 | 641 | be stopped.) |
9490ff27 KH |
642 | |
643 | 11. TODO | |
1b6df3aa BS |
644 | |
645 | 1. Add support for accounting huge pages (as a separate controller) | |
dfc05c25 KH |
646 | 2. Make per-cgroup scanner reclaim not-shared pages first |
647 | 3. Teach controller to account for shared-pages | |
628f4235 | 648 | 4. Start reclamation in the background when the limit is |
1b6df3aa | 649 | not yet hit but the usage is getting closer |
1b6df3aa BS |
650 | |
651 | Summary | |
652 | ||
653 | Overall, the memory controller has been a stable controller and has been | |
654 | commented and discussed quite extensively in the community. | |
655 | ||
656 | References | |
657 | ||
658 | 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/ | |
659 | 2. Singh, Balbir. Memory Controller (RSS Control), | |
660 | http://lwn.net/Articles/222762/ | |
661 | 3. Emelianov, Pavel. Resource controllers based on process cgroups | |
662 | http://lkml.org/lkml/2007/3/6/198 | |
663 | 4. Emelianov, Pavel. RSS controller based on process cgroups (v2) | |
2324c5dd | 664 | http://lkml.org/lkml/2007/4/9/78 |
1b6df3aa BS |
665 | 5. Emelianov, Pavel. RSS controller based on process cgroups (v3) |
666 | http://lkml.org/lkml/2007/5/30/244 | |
667 | 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/ | |
668 | 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control | |
669 | subsystem (v3), http://lwn.net/Articles/235534/ | |
2324c5dd | 670 | 8. Singh, Balbir. RSS controller v2 test results (lmbench), |
1b6df3aa | 671 | http://lkml.org/lkml/2007/5/17/232 |
2324c5dd | 672 | 9. Singh, Balbir. RSS controller v2 AIM9 results |
1b6df3aa | 673 | http://lkml.org/lkml/2007/5/18/1 |
2324c5dd | 674 | 10. Singh, Balbir. Memory controller v6 test results, |
1b6df3aa | 675 | http://lkml.org/lkml/2007/8/19/36 |
2324c5dd LZ |
676 | 11. Singh, Balbir. Memory controller introduction (v6), |
677 | http://lkml.org/lkml/2007/8/17/69 | |
1b6df3aa BS |
678 | 12. Corbet, Jonathan, Controlling memory use in cgroups, |
679 | http://lwn.net/Articles/243795/ |