]>
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() | |
a111c966 DN |
55 | memory.usage_in_bytes # show current res_counter usage for memory |
56 | (See 5.5 for details) | |
57 | memory.memsw.usage_in_bytes # show current res_counter usage for memory+Swap | |
58 | (See 5.5 for details) | |
dc10e281 KH |
59 | memory.limit_in_bytes # set/show limit of memory usage |
60 | memory.memsw.limit_in_bytes # set/show limit of memory+Swap usage | |
61 | memory.failcnt # show the number of memory usage hits limits | |
62 | memory.memsw.failcnt # show the number of memory+Swap hits limits | |
63 | memory.max_usage_in_bytes # show max memory usage recorded | |
64 | memory.memsw.usage_in_bytes # show max memory+Swap usage recorded | |
65 | memory.soft_limit_in_bytes # set/show soft limit of memory usage | |
66 | memory.stat # show various statistics | |
67 | memory.use_hierarchy # set/show hierarchical account enabled | |
68 | memory.force_empty # trigger forced move charge to parent | |
69 | memory.swappiness # set/show swappiness parameter of vmscan | |
70 | (See sysctl's vm.swappiness) | |
71 | memory.move_charge_at_immigrate # set/show controls of moving charges | |
72 | memory.oom_control # set/show oom controls. | |
50c35e5b | 73 | memory.numa_stat # show the number of memory usage per numa node |
dc10e281 | 74 | |
1b6df3aa BS |
75 | 1. History |
76 | ||
77 | The memory controller has a long history. A request for comments for the memory | |
78 | controller was posted by Balbir Singh [1]. At the time the RFC was posted | |
79 | there were several implementations for memory control. The goal of the | |
80 | RFC was to build consensus and agreement for the minimal features required | |
81 | for memory control. The first RSS controller was posted by Balbir Singh[2] | |
82 | in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the | |
83 | RSS controller. At OLS, at the resource management BoF, everyone suggested | |
84 | that we handle both page cache and RSS together. Another request was raised | |
85 | to allow user space handling of OOM. The current memory controller is | |
86 | at version 6; it combines both mapped (RSS) and unmapped Page | |
87 | Cache Control [11]. | |
88 | ||
89 | 2. Memory Control | |
90 | ||
91 | Memory is a unique resource in the sense that it is present in a limited | |
92 | amount. If a task requires a lot of CPU processing, the task can spread | |
93 | its processing over a period of hours, days, months or years, but with | |
94 | memory, the same physical memory needs to be reused to accomplish the task. | |
95 | ||
96 | The memory controller implementation has been divided into phases. These | |
97 | are: | |
98 | ||
99 | 1. Memory controller | |
100 | 2. mlock(2) controller | |
101 | 3. Kernel user memory accounting and slab control | |
102 | 4. user mappings length controller | |
103 | ||
104 | The memory controller is the first controller developed. | |
105 | ||
106 | 2.1. Design | |
107 | ||
108 | The core of the design is a counter called the res_counter. The res_counter | |
109 | tracks the current memory usage and limit of the group of processes associated | |
110 | with the controller. Each cgroup has a memory controller specific data | |
111 | structure (mem_cgroup) associated with it. | |
112 | ||
113 | 2.2. Accounting | |
114 | ||
115 | +--------------------+ | |
116 | | mem_cgroup | | |
117 | | (res_counter) | | |
118 | +--------------------+ | |
119 | / ^ \ | |
120 | / | \ | |
121 | +---------------+ | +---------------+ | |
122 | | mm_struct | |.... | mm_struct | | |
123 | | | | | | | |
124 | +---------------+ | +---------------+ | |
125 | | | |
126 | + --------------+ | |
127 | | | |
128 | +---------------+ +------+--------+ | |
129 | | page +----------> page_cgroup| | |
130 | | | | | | |
131 | +---------------+ +---------------+ | |
132 | ||
133 | (Figure 1: Hierarchy of Accounting) | |
134 | ||
135 | ||
136 | Figure 1 shows the important aspects of the controller | |
137 | ||
138 | 1. Accounting happens per cgroup | |
139 | 2. Each mm_struct knows about which cgroup it belongs to | |
140 | 3. Each page has a pointer to the page_cgroup, which in turn knows the | |
141 | cgroup it belongs to | |
142 | ||
143 | The accounting is done as follows: mem_cgroup_charge() is invoked to setup | |
144 | the necessary data structures and check if the cgroup that is being charged | |
145 | is over its limit. If it is then reclaim is invoked on the cgroup. | |
146 | More details can be found in the reclaim section of this document. | |
147 | If everything goes well, a page meta-data-structure called page_cgroup is | |
dc10e281 KH |
148 | updated. page_cgroup has its own LRU on cgroup. |
149 | (*) page_cgroup structure is allocated at boot/memory-hotplug time. | |
1b6df3aa BS |
150 | |
151 | 2.2.1 Accounting details | |
152 | ||
5b4e655e | 153 | All mapped anon pages (RSS) and cache pages (Page Cache) are accounted. |
dc10e281 KH |
154 | Some pages which are never reclaimable and will not be on the global LRU |
155 | are not accounted. We just account pages under usual VM management. | |
5b4e655e KH |
156 | |
157 | RSS pages are accounted at page_fault unless they've already been accounted | |
158 | for earlier. A file page will be accounted for as Page Cache when it's | |
159 | inserted into inode (radix-tree). While it's mapped into the page tables of | |
160 | processes, duplicate accounting is carefully avoided. | |
161 | ||
162 | A RSS page is unaccounted when it's fully unmapped. A PageCache page is | |
dc10e281 KH |
163 | unaccounted when it's removed from radix-tree. Even if RSS pages are fully |
164 | unmapped (by kswapd), they may exist as SwapCache in the system until they | |
165 | are really freed. Such SwapCaches also also accounted. | |
166 | A swapped-in page is not accounted until it's mapped. | |
167 | ||
168 | Note: The kernel does swapin-readahead and read multiple swaps at once. | |
169 | This means swapped-in pages may contain pages for other tasks than a task | |
170 | causing page fault. So, we avoid accounting at swap-in I/O. | |
5b4e655e KH |
171 | |
172 | At page migration, accounting information is kept. | |
173 | ||
dc10e281 KH |
174 | Note: we just account pages-on-LRU because our purpose is to control amount |
175 | of used pages; not-on-LRU pages tend to be out-of-control from VM view. | |
1b6df3aa BS |
176 | |
177 | 2.3 Shared Page Accounting | |
178 | ||
179 | Shared pages are accounted on the basis of the first touch approach. The | |
180 | cgroup that first touches a page is accounted for the page. The principle | |
181 | behind this approach is that a cgroup that aggressively uses a shared | |
182 | page will eventually get charged for it (once it is uncharged from | |
183 | the cgroup that brought it in -- this will happen on memory pressure). | |
184 | ||
8c7c6e34 KH |
185 | Exception: If CONFIG_CGROUP_CGROUP_MEM_RES_CTLR_SWAP is not used.. |
186 | When you do swapoff and make swapped-out pages of shmem(tmpfs) to | |
d13d1443 KH |
187 | be backed into memory in force, charges for pages are accounted against the |
188 | caller of swapoff rather than the users of shmem. | |
189 | ||
190 | ||
8c7c6e34 | 191 | 2.4 Swap Extension (CONFIG_CGROUP_MEM_RES_CTLR_SWAP) |
dc10e281 | 192 | |
8c7c6e34 KH |
193 | Swap Extension allows you to record charge for swap. A swapped-in page is |
194 | charged back to original page allocator if possible. | |
195 | ||
196 | When swap is accounted, following files are added. | |
197 | - memory.memsw.usage_in_bytes. | |
198 | - memory.memsw.limit_in_bytes. | |
199 | ||
dc10e281 KH |
200 | memsw means memory+swap. Usage of memory+swap is limited by |
201 | memsw.limit_in_bytes. | |
202 | ||
203 | Example: Assume a system with 4G of swap. A task which allocates 6G of memory | |
204 | (by mistake) under 2G memory limitation will use all swap. | |
205 | In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap. | |
206 | By using memsw limit, you can avoid system OOM which can be caused by swap | |
207 | shortage. | |
8c7c6e34 | 208 | |
dc10e281 | 209 | * why 'memory+swap' rather than swap. |
8c7c6e34 KH |
210 | The global LRU(kswapd) can swap out arbitrary pages. Swap-out means |
211 | to move account from memory to swap...there is no change in usage of | |
dc10e281 KH |
212 | memory+swap. In other words, when we want to limit the usage of swap without |
213 | affecting global LRU, memory+swap limit is better than just limiting swap from | |
22a668d7 KH |
214 | OS point of view. |
215 | ||
216 | * What happens when a cgroup hits memory.memsw.limit_in_bytes | |
217 | When a cgroup his memory.memsw.limit_in_bytes, it's useless to do swap-out | |
218 | in this cgroup. Then, swap-out will not be done by cgroup routine and file | |
219 | caches are dropped. But as mentioned above, global LRU can do swapout memory | |
220 | from it for sanity of the system's memory management state. You can't forbid | |
221 | it by cgroup. | |
8c7c6e34 KH |
222 | |
223 | 2.5 Reclaim | |
1b6df3aa | 224 | |
dc10e281 KH |
225 | Each cgroup maintains a per cgroup LRU which has the same structure as |
226 | global VM. When a cgroup goes over its limit, we first try | |
1b6df3aa BS |
227 | to reclaim memory from the cgroup so as to make space for the new |
228 | pages that the cgroup has touched. If the reclaim is unsuccessful, | |
229 | an OOM routine is invoked to select and kill the bulkiest task in the | |
dc10e281 | 230 | cgroup. (See 10. OOM Control below.) |
1b6df3aa BS |
231 | |
232 | The reclaim algorithm has not been modified for cgroups, except that | |
233 | pages that are selected for reclaiming come from the per cgroup LRU | |
234 | list. | |
235 | ||
4b3bde4c BS |
236 | NOTE: Reclaim does not work for the root cgroup, since we cannot set any |
237 | limits on the root cgroup. | |
238 | ||
daaf1e68 KH |
239 | Note2: When panic_on_oom is set to "2", the whole system will panic. |
240 | ||
9490ff27 KH |
241 | When oom event notifier is registered, event will be delivered. |
242 | (See oom_control section) | |
243 | ||
dc10e281 | 244 | 2.6 Locking |
1b6df3aa | 245 | |
dc10e281 KH |
246 | lock_page_cgroup()/unlock_page_cgroup() should not be called under |
247 | mapping->tree_lock. | |
1b6df3aa | 248 | |
dc10e281 KH |
249 | Other lock order is following: |
250 | PG_locked. | |
251 | mm->page_table_lock | |
252 | zone->lru_lock | |
253 | lock_page_cgroup. | |
254 | In many cases, just lock_page_cgroup() is called. | |
255 | per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by | |
256 | zone->lru_lock, it has no lock of its own. | |
1b6df3aa BS |
257 | |
258 | 3. User Interface | |
259 | ||
260 | 0. Configuration | |
261 | ||
262 | a. Enable CONFIG_CGROUPS | |
263 | b. Enable CONFIG_RESOURCE_COUNTERS | |
00f0b825 | 264 | c. Enable CONFIG_CGROUP_MEM_RES_CTLR |
dc10e281 | 265 | d. Enable CONFIG_CGROUP_MEM_RES_CTLR_SWAP (to use swap extension) |
1b6df3aa BS |
266 | |
267 | 1. Prepare the cgroups | |
268 | # mkdir -p /cgroups | |
269 | # mount -t cgroup none /cgroups -o memory | |
270 | ||
271 | 2. Make the new group and move bash into it | |
272 | # mkdir /cgroups/0 | |
dc10e281 | 273 | # echo $$ > /cgroups/0/tasks |
1b6df3aa | 274 | |
dc10e281 | 275 | Since now we're in the 0 cgroup, we can alter the memory limit: |
fb78922c | 276 | # echo 4M > /cgroups/0/memory.limit_in_bytes |
0eea1030 BS |
277 | |
278 | NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo, | |
dc10e281 KH |
279 | mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.) |
280 | ||
c5b947b2 | 281 | NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited). |
4b3bde4c | 282 | NOTE: We cannot set limits on the root cgroup any more. |
0eea1030 BS |
283 | |
284 | # cat /cgroups/0/memory.limit_in_bytes | |
2324c5dd | 285 | 4194304 |
0eea1030 | 286 | |
1b6df3aa | 287 | We can check the usage: |
0eea1030 | 288 | # cat /cgroups/0/memory.usage_in_bytes |
2324c5dd | 289 | 1216512 |
0eea1030 BS |
290 | |
291 | A successful write to this file does not guarantee a successful set of | |
dc10e281 | 292 | this limit to the value written into the file. This can be due to a |
0eea1030 | 293 | number of factors, such as rounding up to page boundaries or the total |
dc10e281 | 294 | availability of memory on the system. The user is required to re-read |
0eea1030 BS |
295 | this file after a write to guarantee the value committed by the kernel. |
296 | ||
fb78922c | 297 | # echo 1 > memory.limit_in_bytes |
0eea1030 | 298 | # cat memory.limit_in_bytes |
2324c5dd | 299 | 4096 |
1b6df3aa BS |
300 | |
301 | The memory.failcnt field gives the number of times that the cgroup limit was | |
302 | exceeded. | |
303 | ||
dfc05c25 KH |
304 | The memory.stat file gives accounting information. Now, the number of |
305 | caches, RSS and Active pages/Inactive pages are shown. | |
306 | ||
1b6df3aa BS |
307 | 4. Testing |
308 | ||
dc10e281 KH |
309 | For testing features and implementation, see memcg_test.txt. |
310 | ||
311 | Performance test is also important. To see pure memory controller's overhead, | |
312 | testing on tmpfs will give you good numbers of small overheads. | |
313 | Example: do kernel make on tmpfs. | |
314 | ||
315 | Page-fault scalability is also important. At measuring parallel | |
316 | page fault test, multi-process test may be better than multi-thread | |
317 | test because it has noise of shared objects/status. | |
318 | ||
319 | But the above two are testing extreme situations. | |
320 | Trying usual test under memory controller is always helpful. | |
1b6df3aa BS |
321 | |
322 | 4.1 Troubleshooting | |
323 | ||
324 | Sometimes a user might find that the application under a cgroup is | |
dc10e281 | 325 | terminated by OOM killer. There are several causes for this: |
1b6df3aa BS |
326 | |
327 | 1. The cgroup limit is too low (just too low to do anything useful) | |
328 | 2. The user is using anonymous memory and swap is turned off or too low | |
329 | ||
330 | A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of | |
331 | some of the pages cached in the cgroup (page cache pages). | |
332 | ||
dc10e281 KH |
333 | To know what happens, disable OOM_Kill by 10. OOM Control(see below) and |
334 | seeing what happens will be helpful. | |
335 | ||
1b6df3aa BS |
336 | 4.2 Task migration |
337 | ||
a33f3224 | 338 | When a task migrates from one cgroup to another, its charge is not |
7dc74be0 | 339 | carried forward by default. The pages allocated from the original cgroup still |
1b6df3aa BS |
340 | remain charged to it, the charge is dropped when the page is freed or |
341 | reclaimed. | |
342 | ||
dc10e281 KH |
343 | You can move charges of a task along with task migration. |
344 | See 8. "Move charges at task migration" | |
7dc74be0 | 345 | |
1b6df3aa BS |
346 | 4.3 Removing a cgroup |
347 | ||
348 | A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a | |
349 | cgroup might have some charge associated with it, even though all | |
dc10e281 KH |
350 | tasks have migrated away from it. (because we charge against pages, not |
351 | against tasks.) | |
352 | ||
353 | Such charges are freed or moved to their parent. At moving, both of RSS | |
354 | and CACHES are moved to parent. | |
355 | rmdir() may return -EBUSY if freeing/moving fails. See 5.1 also. | |
1b6df3aa | 356 | |
8c7c6e34 KH |
357 | Charges recorded in swap information is not updated at removal of cgroup. |
358 | Recorded information is discarded and a cgroup which uses swap (swapcache) | |
359 | will be charged as a new owner of it. | |
360 | ||
361 | ||
c1e862c1 KH |
362 | 5. Misc. interfaces. |
363 | ||
364 | 5.1 force_empty | |
365 | memory.force_empty interface is provided to make cgroup's memory usage empty. | |
366 | You can use this interface only when the cgroup has no tasks. | |
367 | When writing anything to this | |
368 | ||
369 | # echo 0 > memory.force_empty | |
370 | ||
dc10e281 KH |
371 | Almost all pages tracked by this memory cgroup will be unmapped and freed. |
372 | Some pages cannot be freed because they are locked or in-use. Such pages are | |
373 | moved to parent and this cgroup will be empty. This may return -EBUSY if | |
374 | VM is too busy to free/move all pages immediately. | |
c1e862c1 KH |
375 | |
376 | Typical use case of this interface is that calling this before rmdir(). | |
377 | Because rmdir() moves all pages to parent, some out-of-use page caches can be | |
378 | moved to the parent. If you want to avoid that, force_empty will be useful. | |
379 | ||
7f016ee8 | 380 | 5.2 stat file |
c863d835 BR |
381 | |
382 | memory.stat file includes following statistics | |
383 | ||
dc10e281 | 384 | # per-memory cgroup local status |
c863d835 BR |
385 | cache - # of bytes of page cache memory. |
386 | rss - # of bytes of anonymous and swap cache memory. | |
dc10e281 | 387 | mapped_file - # of bytes of mapped file (includes tmpfs/shmem) |
c863d835 BR |
388 | pgpgin - # of pages paged in (equivalent to # of charging events). |
389 | pgpgout - # of pages paged out (equivalent to # of uncharging events). | |
dc10e281 | 390 | swap - # of bytes of swap usage |
c863d835 | 391 | inactive_anon - # of bytes of anonymous memory and swap cache memory on |
dc10e281 KH |
392 | LRU list. |
393 | active_anon - # of bytes of anonymous and swap cache memory on active | |
394 | inactive LRU list. | |
395 | inactive_file - # of bytes of file-backed memory on inactive LRU list. | |
396 | active_file - # of bytes of file-backed memory on active LRU list. | |
c863d835 BR |
397 | unevictable - # of bytes of memory that cannot be reclaimed (mlocked etc). |
398 | ||
dc10e281 KH |
399 | # status considering hierarchy (see memory.use_hierarchy settings) |
400 | ||
401 | hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy | |
402 | under which the memory cgroup is | |
403 | hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to | |
404 | hierarchy under which memory cgroup is. | |
405 | ||
406 | total_cache - sum of all children's "cache" | |
407 | total_rss - sum of all children's "rss" | |
408 | total_mapped_file - sum of all children's "cache" | |
409 | total_pgpgin - sum of all children's "pgpgin" | |
410 | total_pgpgout - sum of all children's "pgpgout" | |
411 | total_swap - sum of all children's "swap" | |
412 | total_inactive_anon - sum of all children's "inactive_anon" | |
413 | total_active_anon - sum of all children's "active_anon" | |
414 | total_inactive_file - sum of all children's "inactive_file" | |
415 | total_active_file - sum of all children's "active_file" | |
416 | total_unevictable - sum of all children's "unevictable" | |
417 | ||
418 | # The following additional stats are dependent on CONFIG_DEBUG_VM. | |
c863d835 BR |
419 | |
420 | inactive_ratio - VM internal parameter. (see mm/page_alloc.c) | |
421 | recent_rotated_anon - VM internal parameter. (see mm/vmscan.c) | |
422 | recent_rotated_file - VM internal parameter. (see mm/vmscan.c) | |
423 | recent_scanned_anon - VM internal parameter. (see mm/vmscan.c) | |
424 | recent_scanned_file - VM internal parameter. (see mm/vmscan.c) | |
425 | ||
426 | Memo: | |
dc10e281 KH |
427 | recent_rotated means recent frequency of LRU rotation. |
428 | recent_scanned means recent # of scans to LRU. | |
7f016ee8 KM |
429 | showing for better debug please see the code for meanings. |
430 | ||
c863d835 BR |
431 | Note: |
432 | Only anonymous and swap cache memory is listed as part of 'rss' stat. | |
433 | This should not be confused with the true 'resident set size' or the | |
dc10e281 KH |
434 | amount of physical memory used by the cgroup. |
435 | 'rss + file_mapped" will give you resident set size of cgroup. | |
436 | (Note: file and shmem may be shared among other cgroups. In that case, | |
437 | file_mapped is accounted only when the memory cgroup is owner of page | |
438 | cache.) | |
7f016ee8 | 439 | |
a7885eb8 | 440 | 5.3 swappiness |
a7885eb8 | 441 | |
dc10e281 | 442 | Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only. |
a7885eb8 | 443 | |
dc10e281 KH |
444 | Following cgroups' swappiness can't be changed. |
445 | - root cgroup (uses /proc/sys/vm/swappiness). | |
446 | - a cgroup which uses hierarchy and it has other cgroup(s) below it. | |
447 | - a cgroup which uses hierarchy and not the root of hierarchy. | |
448 | ||
449 | 5.4 failcnt | |
450 | ||
451 | A memory cgroup provides memory.failcnt and memory.memsw.failcnt files. | |
452 | This failcnt(== failure count) shows the number of times that a usage counter | |
453 | hit its limit. When a memory cgroup hits a limit, failcnt increases and | |
454 | memory under it will be reclaimed. | |
455 | ||
456 | You can reset failcnt by writing 0 to failcnt file. | |
457 | # echo 0 > .../memory.failcnt | |
a7885eb8 | 458 | |
a111c966 DN |
459 | 5.5 usage_in_bytes |
460 | ||
461 | For efficiency, as other kernel components, memory cgroup uses some optimization | |
462 | to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the | |
463 | method and doesn't show 'exact' value of memory(and swap) usage, it's an fuzz | |
464 | value for efficient access. (Of course, when necessary, it's synchronized.) | |
465 | If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP) | |
466 | value in memory.stat(see 5.2). | |
467 | ||
50c35e5b YH |
468 | 5.6 numa_stat |
469 | ||
470 | This is similar to numa_maps but operates on a per-memcg basis. This is | |
471 | useful for providing visibility into the numa locality information within | |
472 | an memcg since the pages are allowed to be allocated from any physical | |
473 | node. One of the usecases is evaluating application performance by | |
474 | combining this information with the application's cpu allocation. | |
475 | ||
476 | We export "total", "file", "anon" and "unevictable" pages per-node for | |
477 | each memcg. The ouput format of memory.numa_stat is: | |
478 | ||
479 | total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
480 | file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
481 | anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
482 | unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
483 | ||
484 | And we have total = file + anon + unevictable. | |
485 | ||
52bc0d82 | 486 | 6. Hierarchy support |
c1e862c1 | 487 | |
52bc0d82 BS |
488 | The memory controller supports a deep hierarchy and hierarchical accounting. |
489 | The hierarchy is created by creating the appropriate cgroups in the | |
490 | cgroup filesystem. Consider for example, the following cgroup filesystem | |
491 | hierarchy | |
492 | ||
493 | root | |
494 | / | \ | |
495 | / | \ | |
496 | a b c | |
497 | | \ | |
498 | | \ | |
499 | d e | |
500 | ||
501 | In the diagram above, with hierarchical accounting enabled, all memory | |
502 | usage of e, is accounted to its ancestors up until the root (i.e, c and root), | |
dc10e281 | 503 | that has memory.use_hierarchy enabled. If one of the ancestors goes over its |
52bc0d82 BS |
504 | limit, the reclaim algorithm reclaims from the tasks in the ancestor and the |
505 | children of the ancestor. | |
506 | ||
507 | 6.1 Enabling hierarchical accounting and reclaim | |
508 | ||
dc10e281 | 509 | A memory cgroup by default disables the hierarchy feature. Support |
52bc0d82 BS |
510 | can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup |
511 | ||
512 | # echo 1 > memory.use_hierarchy | |
513 | ||
514 | The feature can be disabled by | |
515 | ||
516 | # echo 0 > memory.use_hierarchy | |
517 | ||
689bca3b GT |
518 | NOTE1: Enabling/disabling will fail if either the cgroup already has other |
519 | cgroups created below it, or if the parent cgroup has use_hierarchy | |
520 | enabled. | |
52bc0d82 | 521 | |
daaf1e68 | 522 | NOTE2: When panic_on_oom is set to "2", the whole system will panic in |
dc10e281 | 523 | case of an OOM event in any cgroup. |
52bc0d82 | 524 | |
a6df6361 BS |
525 | 7. Soft limits |
526 | ||
527 | Soft limits allow for greater sharing of memory. The idea behind soft limits | |
528 | is to allow control groups to use as much of the memory as needed, provided | |
529 | ||
530 | a. There is no memory contention | |
531 | b. They do not exceed their hard limit | |
532 | ||
dc10e281 | 533 | When the system detects memory contention or low memory, control groups |
a6df6361 BS |
534 | are pushed back to their soft limits. If the soft limit of each control |
535 | group is very high, they are pushed back as much as possible to make | |
536 | sure that one control group does not starve the others of memory. | |
537 | ||
538 | Please note that soft limits is a best effort feature, it comes with | |
539 | no guarantees, but it does its best to make sure that when memory is | |
540 | heavily contended for, memory is allocated based on the soft limit | |
541 | hints/setup. Currently soft limit based reclaim is setup such that | |
542 | it gets invoked from balance_pgdat (kswapd). | |
543 | ||
544 | 7.1 Interface | |
545 | ||
546 | Soft limits can be setup by using the following commands (in this example we | |
dc10e281 | 547 | assume a soft limit of 256 MiB) |
a6df6361 BS |
548 | |
549 | # echo 256M > memory.soft_limit_in_bytes | |
550 | ||
551 | If we want to change this to 1G, we can at any time use | |
552 | ||
553 | # echo 1G > memory.soft_limit_in_bytes | |
554 | ||
555 | NOTE1: Soft limits take effect over a long period of time, since they involve | |
556 | reclaiming memory for balancing between memory cgroups | |
557 | NOTE2: It is recommended to set the soft limit always below the hard limit, | |
558 | otherwise the hard limit will take precedence. | |
559 | ||
7dc74be0 DN |
560 | 8. Move charges at task migration |
561 | ||
562 | Users can move charges associated with a task along with task migration, that | |
563 | is, uncharge task's pages from the old cgroup and charge them to the new cgroup. | |
02491447 DN |
564 | This feature is not supported in !CONFIG_MMU environments because of lack of |
565 | page tables. | |
7dc74be0 DN |
566 | |
567 | 8.1 Interface | |
568 | ||
569 | This feature is disabled by default. It can be enabled(and disabled again) by | |
570 | writing to memory.move_charge_at_immigrate of the destination cgroup. | |
571 | ||
572 | If you want to enable it: | |
573 | ||
574 | # echo (some positive value) > memory.move_charge_at_immigrate | |
575 | ||
576 | Note: Each bits of move_charge_at_immigrate has its own meaning about what type | |
577 | of charges should be moved. See 8.2 for details. | |
578 | Note: Charges are moved only when you move mm->owner, IOW, a leader of a thread | |
579 | group. | |
580 | Note: If we cannot find enough space for the task in the destination cgroup, we | |
581 | try to make space by reclaiming memory. Task migration may fail if we | |
582 | cannot make enough space. | |
dc10e281 | 583 | Note: It can take several seconds if you move charges much. |
7dc74be0 DN |
584 | |
585 | And if you want disable it again: | |
586 | ||
587 | # echo 0 > memory.move_charge_at_immigrate | |
588 | ||
589 | 8.2 Type of charges which can be move | |
590 | ||
591 | Each bits of move_charge_at_immigrate has its own meaning about what type of | |
87946a72 DN |
592 | charges should be moved. But in any cases, it must be noted that an account of |
593 | a page or a swap can be moved only when it is charged to the task's current(old) | |
594 | memory cgroup. | |
7dc74be0 DN |
595 | |
596 | bit | what type of charges would be moved ? | |
597 | -----+------------------------------------------------------------------------ | |
598 | 0 | A charge of an anonymous page(or swap of it) used by the target task. | |
599 | | Those pages and swaps must be used only by the target task. You must | |
600 | | enable Swap Extension(see 2.4) to enable move of swap charges. | |
87946a72 DN |
601 | -----+------------------------------------------------------------------------ |
602 | 1 | A charge of file pages(normal file, tmpfs file(e.g. ipc shared memory) | |
dc10e281 | 603 | | and swaps of tmpfs file) mmapped by the target task. Unlike the case of |
87946a72 DN |
604 | | anonymous pages, file pages(and swaps) in the range mmapped by the task |
605 | | will be moved even if the task hasn't done page fault, i.e. they might | |
606 | | not be the task's "RSS", but other task's "RSS" that maps the same file. | |
607 | | And mapcount of the page is ignored(the page can be moved even if | |
608 | | page_mapcount(page) > 1). You must enable Swap Extension(see 2.4) to | |
609 | | enable move of swap charges. | |
7dc74be0 DN |
610 | |
611 | 8.3 TODO | |
612 | ||
7dc74be0 DN |
613 | - Implement madvise(2) to let users decide the vma to be moved or not to be |
614 | moved. | |
615 | - All of moving charge operations are done under cgroup_mutex. It's not good | |
616 | behavior to hold the mutex too long, so we may need some trick. | |
617 | ||
2e72b634 KS |
618 | 9. Memory thresholds |
619 | ||
dc10e281 | 620 | Memory cgroup implements memory thresholds using cgroups notification |
2e72b634 KS |
621 | API (see cgroups.txt). It allows to register multiple memory and memsw |
622 | thresholds and gets notifications when it crosses. | |
623 | ||
624 | To register a threshold application need: | |
dc10e281 KH |
625 | - create an eventfd using eventfd(2); |
626 | - open memory.usage_in_bytes or memory.memsw.usage_in_bytes; | |
627 | - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to | |
628 | cgroup.event_control. | |
2e72b634 KS |
629 | |
630 | Application will be notified through eventfd when memory usage crosses | |
631 | threshold in any direction. | |
632 | ||
633 | It's applicable for root and non-root cgroup. | |
634 | ||
9490ff27 KH |
635 | 10. OOM Control |
636 | ||
3c11ecf4 KH |
637 | memory.oom_control file is for OOM notification and other controls. |
638 | ||
dc10e281 KH |
639 | Memory cgroup implements OOM notifier using cgroup notification |
640 | API (See cgroups.txt). It allows to register multiple OOM notification | |
641 | delivery and gets notification when OOM happens. | |
9490ff27 KH |
642 | |
643 | To register a notifier, application need: | |
644 | - create an eventfd using eventfd(2) | |
645 | - open memory.oom_control file | |
dc10e281 KH |
646 | - write string like "<event_fd> <fd of memory.oom_control>" to |
647 | cgroup.event_control | |
9490ff27 | 648 | |
dc10e281 | 649 | Application will be notified through eventfd when OOM happens. |
9490ff27 KH |
650 | OOM notification doesn't work for root cgroup. |
651 | ||
dc10e281 KH |
652 | You can disable OOM-killer by writing "1" to memory.oom_control file, as: |
653 | ||
3c11ecf4 KH |
654 | #echo 1 > memory.oom_control |
655 | ||
dc10e281 KH |
656 | This operation is only allowed to the top cgroup of sub-hierarchy. |
657 | If OOM-killer is disabled, tasks under cgroup will hang/sleep | |
658 | in memory cgroup's OOM-waitqueue when they request accountable memory. | |
3c11ecf4 | 659 | |
dc10e281 | 660 | For running them, you have to relax the memory cgroup's OOM status by |
3c11ecf4 KH |
661 | * enlarge limit or reduce usage. |
662 | To reduce usage, | |
663 | * kill some tasks. | |
664 | * move some tasks to other group with account migration. | |
665 | * remove some files (on tmpfs?) | |
666 | ||
667 | Then, stopped tasks will work again. | |
668 | ||
669 | At reading, current status of OOM is shown. | |
670 | oom_kill_disable 0 or 1 (if 1, oom-killer is disabled) | |
dc10e281 | 671 | under_oom 0 or 1 (if 1, the memory cgroup is under OOM, tasks may |
3c11ecf4 | 672 | be stopped.) |
9490ff27 KH |
673 | |
674 | 11. TODO | |
1b6df3aa BS |
675 | |
676 | 1. Add support for accounting huge pages (as a separate controller) | |
dfc05c25 KH |
677 | 2. Make per-cgroup scanner reclaim not-shared pages first |
678 | 3. Teach controller to account for shared-pages | |
628f4235 | 679 | 4. Start reclamation in the background when the limit is |
1b6df3aa | 680 | not yet hit but the usage is getting closer |
1b6df3aa BS |
681 | |
682 | Summary | |
683 | ||
684 | Overall, the memory controller has been a stable controller and has been | |
685 | commented and discussed quite extensively in the community. | |
686 | ||
687 | References | |
688 | ||
689 | 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/ | |
690 | 2. Singh, Balbir. Memory Controller (RSS Control), | |
691 | http://lwn.net/Articles/222762/ | |
692 | 3. Emelianov, Pavel. Resource controllers based on process cgroups | |
693 | http://lkml.org/lkml/2007/3/6/198 | |
694 | 4. Emelianov, Pavel. RSS controller based on process cgroups (v2) | |
2324c5dd | 695 | http://lkml.org/lkml/2007/4/9/78 |
1b6df3aa BS |
696 | 5. Emelianov, Pavel. RSS controller based on process cgroups (v3) |
697 | http://lkml.org/lkml/2007/5/30/244 | |
698 | 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/ | |
699 | 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control | |
700 | subsystem (v3), http://lwn.net/Articles/235534/ | |
2324c5dd | 701 | 8. Singh, Balbir. RSS controller v2 test results (lmbench), |
1b6df3aa | 702 | http://lkml.org/lkml/2007/5/17/232 |
2324c5dd | 703 | 9. Singh, Balbir. RSS controller v2 AIM9 results |
1b6df3aa | 704 | http://lkml.org/lkml/2007/5/18/1 |
2324c5dd | 705 | 10. Singh, Balbir. Memory controller v6 test results, |
1b6df3aa | 706 | http://lkml.org/lkml/2007/8/19/36 |
2324c5dd LZ |
707 | 11. Singh, Balbir. Memory controller introduction (v6), |
708 | http://lkml.org/lkml/2007/8/17/69 | |
1b6df3aa BS |
709 | 12. Corbet, Jonathan, Controlling memory use in cgroups, |
710 | http://lwn.net/Articles/243795/ |