Commit | Line | Data |
---|---|---|
00f0b825 BS |
1 | Memory Resource Controller |
2 | ||
67de0162 JS |
3 | NOTE: The Memory Resource Controller has generically been referred to as the |
4 | memory controller in this document. Do not confuse memory controller | |
5 | 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 | ||
e5671dfa GC |
47 | Hugepages is not under control yet. We just manage pages on LRU. To add more |
48 | controls, we have to take care of performance. Kernel memory support is work | |
49 | in progress, and the current version provides basically functionality. | |
dc10e281 KH |
50 | |
51 | Brief summary of control files. | |
52 | ||
53 | tasks # attach a task(thread) and show list of threads | |
54 | cgroup.procs # show list of processes | |
55 | cgroup.event_control # an interface for event_fd() | |
a111c966 DN |
56 | memory.usage_in_bytes # show current res_counter usage for memory |
57 | (See 5.5 for details) | |
58 | memory.memsw.usage_in_bytes # show current res_counter usage for memory+Swap | |
59 | (See 5.5 for details) | |
e5671dfa GC |
60 | memory.kmem.usage_in_bytes # show current res_counter usage for kmem only. |
61 | (See 2.7 for details) | |
dc10e281 KH |
62 | memory.limit_in_bytes # set/show limit of memory usage |
63 | memory.memsw.limit_in_bytes # set/show limit of memory+Swap usage | |
e5671dfa | 64 | memory.kmem.limit_in_bytes # if allowed, set/show limit of kernel memory |
dc10e281 KH |
65 | memory.failcnt # show the number of memory usage hits limits |
66 | memory.memsw.failcnt # show the number of memory+Swap hits limits | |
67 | memory.max_usage_in_bytes # show max memory usage recorded | |
68 | memory.memsw.usage_in_bytes # show max memory+Swap usage recorded | |
69 | memory.soft_limit_in_bytes # set/show soft limit of memory usage | |
70 | memory.stat # show various statistics | |
71 | memory.use_hierarchy # set/show hierarchical account enabled | |
72 | memory.force_empty # trigger forced move charge to parent | |
73 | memory.swappiness # set/show swappiness parameter of vmscan | |
74 | (See sysctl's vm.swappiness) | |
75 | memory.move_charge_at_immigrate # set/show controls of moving charges | |
76 | memory.oom_control # set/show oom controls. | |
50c35e5b | 77 | memory.numa_stat # show the number of memory usage per numa node |
dc10e281 | 78 | |
e5671dfa GC |
79 | memory.independent_kmem_limit # select whether or not kernel memory limits are |
80 | independent of user limits | |
81 | ||
1b6df3aa BS |
82 | 1. History |
83 | ||
84 | The memory controller has a long history. A request for comments for the memory | |
85 | controller was posted by Balbir Singh [1]. At the time the RFC was posted | |
86 | there were several implementations for memory control. The goal of the | |
87 | RFC was to build consensus and agreement for the minimal features required | |
88 | for memory control. The first RSS controller was posted by Balbir Singh[2] | |
89 | in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the | |
90 | RSS controller. At OLS, at the resource management BoF, everyone suggested | |
91 | that we handle both page cache and RSS together. Another request was raised | |
92 | to allow user space handling of OOM. The current memory controller is | |
93 | at version 6; it combines both mapped (RSS) and unmapped Page | |
94 | Cache Control [11]. | |
95 | ||
96 | 2. Memory Control | |
97 | ||
98 | Memory is a unique resource in the sense that it is present in a limited | |
99 | amount. If a task requires a lot of CPU processing, the task can spread | |
100 | its processing over a period of hours, days, months or years, but with | |
101 | memory, the same physical memory needs to be reused to accomplish the task. | |
102 | ||
103 | The memory controller implementation has been divided into phases. These | |
104 | are: | |
105 | ||
106 | 1. Memory controller | |
107 | 2. mlock(2) controller | |
108 | 3. Kernel user memory accounting and slab control | |
109 | 4. user mappings length controller | |
110 | ||
111 | The memory controller is the first controller developed. | |
112 | ||
113 | 2.1. Design | |
114 | ||
115 | The core of the design is a counter called the res_counter. The res_counter | |
116 | tracks the current memory usage and limit of the group of processes associated | |
117 | with the controller. Each cgroup has a memory controller specific data | |
118 | structure (mem_cgroup) associated with it. | |
119 | ||
120 | 2.2. Accounting | |
121 | ||
122 | +--------------------+ | |
123 | | mem_cgroup | | |
124 | | (res_counter) | | |
125 | +--------------------+ | |
126 | / ^ \ | |
127 | / | \ | |
128 | +---------------+ | +---------------+ | |
129 | | mm_struct | |.... | mm_struct | | |
130 | | | | | | | |
131 | +---------------+ | +---------------+ | |
132 | | | |
133 | + --------------+ | |
134 | | | |
135 | +---------------+ +------+--------+ | |
136 | | page +----------> page_cgroup| | |
137 | | | | | | |
138 | +---------------+ +---------------+ | |
139 | ||
140 | (Figure 1: Hierarchy of Accounting) | |
141 | ||
142 | ||
143 | Figure 1 shows the important aspects of the controller | |
144 | ||
145 | 1. Accounting happens per cgroup | |
146 | 2. Each mm_struct knows about which cgroup it belongs to | |
147 | 3. Each page has a pointer to the page_cgroup, which in turn knows the | |
148 | cgroup it belongs to | |
149 | ||
150 | The accounting is done as follows: mem_cgroup_charge() is invoked to setup | |
151 | the necessary data structures and check if the cgroup that is being charged | |
152 | is over its limit. If it is then reclaim is invoked on the cgroup. | |
153 | More details can be found in the reclaim section of this document. | |
154 | If everything goes well, a page meta-data-structure called page_cgroup is | |
dc10e281 KH |
155 | updated. page_cgroup has its own LRU on cgroup. |
156 | (*) page_cgroup structure is allocated at boot/memory-hotplug time. | |
1b6df3aa BS |
157 | |
158 | 2.2.1 Accounting details | |
159 | ||
5b4e655e | 160 | All mapped anon pages (RSS) and cache pages (Page Cache) are accounted. |
dc10e281 KH |
161 | Some pages which are never reclaimable and will not be on the global LRU |
162 | are not accounted. We just account pages under usual VM management. | |
5b4e655e KH |
163 | |
164 | RSS pages are accounted at page_fault unless they've already been accounted | |
165 | for earlier. A file page will be accounted for as Page Cache when it's | |
166 | inserted into inode (radix-tree). While it's mapped into the page tables of | |
167 | processes, duplicate accounting is carefully avoided. | |
168 | ||
169 | A RSS page is unaccounted when it's fully unmapped. A PageCache page is | |
dc10e281 KH |
170 | unaccounted when it's removed from radix-tree. Even if RSS pages are fully |
171 | unmapped (by kswapd), they may exist as SwapCache in the system until they | |
172 | are really freed. Such SwapCaches also also accounted. | |
173 | A swapped-in page is not accounted until it's mapped. | |
174 | ||
175 | Note: The kernel does swapin-readahead and read multiple swaps at once. | |
176 | This means swapped-in pages may contain pages for other tasks than a task | |
177 | causing page fault. So, we avoid accounting at swap-in I/O. | |
5b4e655e KH |
178 | |
179 | At page migration, accounting information is kept. | |
180 | ||
dc10e281 KH |
181 | Note: we just account pages-on-LRU because our purpose is to control amount |
182 | of used pages; not-on-LRU pages tend to be out-of-control from VM view. | |
1b6df3aa BS |
183 | |
184 | 2.3 Shared Page Accounting | |
185 | ||
186 | Shared pages are accounted on the basis of the first touch approach. The | |
187 | cgroup that first touches a page is accounted for the page. The principle | |
188 | behind this approach is that a cgroup that aggressively uses a shared | |
189 | page will eventually get charged for it (once it is uncharged from | |
190 | the cgroup that brought it in -- this will happen on memory pressure). | |
191 | ||
67de0162 | 192 | Exception: If CONFIG_CGROUP_CGROUP_MEM_RES_CTLR_SWAP is not used. |
8c7c6e34 | 193 | When you do swapoff and make swapped-out pages of shmem(tmpfs) to |
d13d1443 KH |
194 | be backed into memory in force, charges for pages are accounted against the |
195 | caller of swapoff rather than the users of shmem. | |
196 | ||
197 | ||
8c7c6e34 | 198 | 2.4 Swap Extension (CONFIG_CGROUP_MEM_RES_CTLR_SWAP) |
dc10e281 | 199 | |
8c7c6e34 KH |
200 | Swap Extension allows you to record charge for swap. A swapped-in page is |
201 | charged back to original page allocator if possible. | |
202 | ||
203 | When swap is accounted, following files are added. | |
204 | - memory.memsw.usage_in_bytes. | |
205 | - memory.memsw.limit_in_bytes. | |
206 | ||
dc10e281 KH |
207 | memsw means memory+swap. Usage of memory+swap is limited by |
208 | memsw.limit_in_bytes. | |
209 | ||
210 | Example: Assume a system with 4G of swap. A task which allocates 6G of memory | |
211 | (by mistake) under 2G memory limitation will use all swap. | |
212 | In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap. | |
213 | By using memsw limit, you can avoid system OOM which can be caused by swap | |
214 | shortage. | |
8c7c6e34 | 215 | |
dc10e281 | 216 | * why 'memory+swap' rather than swap. |
8c7c6e34 KH |
217 | The global LRU(kswapd) can swap out arbitrary pages. Swap-out means |
218 | to move account from memory to swap...there is no change in usage of | |
dc10e281 KH |
219 | memory+swap. In other words, when we want to limit the usage of swap without |
220 | affecting global LRU, memory+swap limit is better than just limiting swap from | |
22a668d7 KH |
221 | OS point of view. |
222 | ||
223 | * What happens when a cgroup hits memory.memsw.limit_in_bytes | |
67de0162 | 224 | When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out |
22a668d7 KH |
225 | in this cgroup. Then, swap-out will not be done by cgroup routine and file |
226 | caches are dropped. But as mentioned above, global LRU can do swapout memory | |
227 | from it for sanity of the system's memory management state. You can't forbid | |
228 | it by cgroup. | |
8c7c6e34 KH |
229 | |
230 | 2.5 Reclaim | |
1b6df3aa | 231 | |
dc10e281 KH |
232 | Each cgroup maintains a per cgroup LRU which has the same structure as |
233 | global VM. When a cgroup goes over its limit, we first try | |
1b6df3aa BS |
234 | to reclaim memory from the cgroup so as to make space for the new |
235 | pages that the cgroup has touched. If the reclaim is unsuccessful, | |
236 | an OOM routine is invoked to select and kill the bulkiest task in the | |
dc10e281 | 237 | cgroup. (See 10. OOM Control below.) |
1b6df3aa BS |
238 | |
239 | The reclaim algorithm has not been modified for cgroups, except that | |
240 | pages that are selected for reclaiming come from the per cgroup LRU | |
241 | list. | |
242 | ||
4b3bde4c BS |
243 | NOTE: Reclaim does not work for the root cgroup, since we cannot set any |
244 | limits on the root cgroup. | |
245 | ||
daaf1e68 KH |
246 | Note2: When panic_on_oom is set to "2", the whole system will panic. |
247 | ||
9490ff27 KH |
248 | When oom event notifier is registered, event will be delivered. |
249 | (See oom_control section) | |
250 | ||
dc10e281 | 251 | 2.6 Locking |
1b6df3aa | 252 | |
dc10e281 KH |
253 | lock_page_cgroup()/unlock_page_cgroup() should not be called under |
254 | mapping->tree_lock. | |
1b6df3aa | 255 | |
dc10e281 KH |
256 | Other lock order is following: |
257 | PG_locked. | |
258 | mm->page_table_lock | |
259 | zone->lru_lock | |
260 | lock_page_cgroup. | |
261 | In many cases, just lock_page_cgroup() is called. | |
262 | per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by | |
263 | zone->lru_lock, it has no lock of its own. | |
1b6df3aa | 264 | |
e5671dfa GC |
265 | 2.7 Kernel Memory Extension (CONFIG_CGROUP_MEM_RES_CTLR_KMEM) |
266 | ||
267 | With the Kernel memory extension, the Memory Controller is able to limit | |
268 | the amount of kernel memory used by the system. Kernel memory is fundamentally | |
269 | different than user memory, since it can't be swapped out, which makes it | |
270 | possible to DoS the system by consuming too much of this precious resource. | |
271 | ||
272 | Some kernel memory resources may be accounted and limited separately from the | |
273 | main "kmem" resource. For instance, a slab cache that is considered important | |
274 | enough to be limited separately may have its own knobs. | |
275 | ||
276 | Kernel memory limits are not imposed for the root cgroup. Usage for the root | |
277 | cgroup may or may not be accounted. | |
278 | ||
279 | Memory limits as specified by the standard Memory Controller may or may not | |
280 | take kernel memory into consideration. This is achieved through the file | |
281 | memory.independent_kmem_limit. A Value different than 0 will allow for kernel | |
282 | memory to be controlled separately. | |
283 | ||
284 | When kernel memory limits are not independent, the limit values set in | |
285 | memory.kmem files are ignored. | |
286 | ||
287 | Currently no soft limit is implemented for kernel memory. It is future work | |
288 | to trigger slab reclaim when those limits are reached. | |
289 | ||
290 | 2.7.1 Current Kernel Memory resources accounted | |
291 | ||
e1aab161 GC |
292 | * sockets memory pressure: some sockets protocols have memory pressure |
293 | thresholds. The Memory Controller allows them to be controlled individually | |
294 | per cgroup, instead of globally. | |
e5671dfa | 295 | |
1b6df3aa BS |
296 | 3. User Interface |
297 | ||
298 | 0. Configuration | |
299 | ||
300 | a. Enable CONFIG_CGROUPS | |
301 | b. Enable CONFIG_RESOURCE_COUNTERS | |
00f0b825 | 302 | c. Enable CONFIG_CGROUP_MEM_RES_CTLR |
dc10e281 | 303 | d. Enable CONFIG_CGROUP_MEM_RES_CTLR_SWAP (to use swap extension) |
1b6df3aa | 304 | |
f6e07d38 JS |
305 | 1. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?) |
306 | # mount -t tmpfs none /sys/fs/cgroup | |
307 | # mkdir /sys/fs/cgroup/memory | |
308 | # mount -t cgroup none /sys/fs/cgroup/memory -o memory | |
1b6df3aa BS |
309 | |
310 | 2. Make the new group and move bash into it | |
f6e07d38 JS |
311 | # mkdir /sys/fs/cgroup/memory/0 |
312 | # echo $$ > /sys/fs/cgroup/memory/0/tasks | |
1b6df3aa | 313 | |
dc10e281 | 314 | Since now we're in the 0 cgroup, we can alter the memory limit: |
f6e07d38 | 315 | # echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes |
0eea1030 BS |
316 | |
317 | NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo, | |
dc10e281 KH |
318 | mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.) |
319 | ||
c5b947b2 | 320 | NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited). |
4b3bde4c | 321 | NOTE: We cannot set limits on the root cgroup any more. |
0eea1030 | 322 | |
f6e07d38 | 323 | # cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes |
2324c5dd | 324 | 4194304 |
0eea1030 | 325 | |
1b6df3aa | 326 | We can check the usage: |
f6e07d38 | 327 | # cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes |
2324c5dd | 328 | 1216512 |
0eea1030 BS |
329 | |
330 | A successful write to this file does not guarantee a successful set of | |
dc10e281 | 331 | this limit to the value written into the file. This can be due to a |
0eea1030 | 332 | number of factors, such as rounding up to page boundaries or the total |
dc10e281 | 333 | availability of memory on the system. The user is required to re-read |
0eea1030 BS |
334 | this file after a write to guarantee the value committed by the kernel. |
335 | ||
fb78922c | 336 | # echo 1 > memory.limit_in_bytes |
0eea1030 | 337 | # cat memory.limit_in_bytes |
2324c5dd | 338 | 4096 |
1b6df3aa BS |
339 | |
340 | The memory.failcnt field gives the number of times that the cgroup limit was | |
341 | exceeded. | |
342 | ||
dfc05c25 KH |
343 | The memory.stat file gives accounting information. Now, the number of |
344 | caches, RSS and Active pages/Inactive pages are shown. | |
345 | ||
1b6df3aa BS |
346 | 4. Testing |
347 | ||
dc10e281 KH |
348 | For testing features and implementation, see memcg_test.txt. |
349 | ||
350 | Performance test is also important. To see pure memory controller's overhead, | |
351 | testing on tmpfs will give you good numbers of small overheads. | |
352 | Example: do kernel make on tmpfs. | |
353 | ||
354 | Page-fault scalability is also important. At measuring parallel | |
355 | page fault test, multi-process test may be better than multi-thread | |
356 | test because it has noise of shared objects/status. | |
357 | ||
358 | But the above two are testing extreme situations. | |
359 | Trying usual test under memory controller is always helpful. | |
1b6df3aa BS |
360 | |
361 | 4.1 Troubleshooting | |
362 | ||
363 | Sometimes a user might find that the application under a cgroup is | |
dc10e281 | 364 | terminated by OOM killer. There are several causes for this: |
1b6df3aa BS |
365 | |
366 | 1. The cgroup limit is too low (just too low to do anything useful) | |
367 | 2. The user is using anonymous memory and swap is turned off or too low | |
368 | ||
369 | A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of | |
370 | some of the pages cached in the cgroup (page cache pages). | |
371 | ||
dc10e281 KH |
372 | To know what happens, disable OOM_Kill by 10. OOM Control(see below) and |
373 | seeing what happens will be helpful. | |
374 | ||
1b6df3aa BS |
375 | 4.2 Task migration |
376 | ||
a33f3224 | 377 | When a task migrates from one cgroup to another, its charge is not |
7dc74be0 | 378 | carried forward by default. The pages allocated from the original cgroup still |
1b6df3aa BS |
379 | remain charged to it, the charge is dropped when the page is freed or |
380 | reclaimed. | |
381 | ||
dc10e281 KH |
382 | You can move charges of a task along with task migration. |
383 | See 8. "Move charges at task migration" | |
7dc74be0 | 384 | |
1b6df3aa BS |
385 | 4.3 Removing a cgroup |
386 | ||
387 | A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a | |
388 | cgroup might have some charge associated with it, even though all | |
dc10e281 KH |
389 | tasks have migrated away from it. (because we charge against pages, not |
390 | against tasks.) | |
391 | ||
392 | Such charges are freed or moved to their parent. At moving, both of RSS | |
393 | and CACHES are moved to parent. | |
394 | rmdir() may return -EBUSY if freeing/moving fails. See 5.1 also. | |
1b6df3aa | 395 | |
8c7c6e34 KH |
396 | Charges recorded in swap information is not updated at removal of cgroup. |
397 | Recorded information is discarded and a cgroup which uses swap (swapcache) | |
398 | will be charged as a new owner of it. | |
399 | ||
400 | ||
c1e862c1 KH |
401 | 5. Misc. interfaces. |
402 | ||
403 | 5.1 force_empty | |
404 | memory.force_empty interface is provided to make cgroup's memory usage empty. | |
405 | You can use this interface only when the cgroup has no tasks. | |
406 | When writing anything to this | |
407 | ||
408 | # echo 0 > memory.force_empty | |
409 | ||
dc10e281 KH |
410 | Almost all pages tracked by this memory cgroup will be unmapped and freed. |
411 | Some pages cannot be freed because they are locked or in-use. Such pages are | |
412 | moved to parent and this cgroup will be empty. This may return -EBUSY if | |
413 | VM is too busy to free/move all pages immediately. | |
c1e862c1 KH |
414 | |
415 | Typical use case of this interface is that calling this before rmdir(). | |
416 | Because rmdir() moves all pages to parent, some out-of-use page caches can be | |
417 | moved to the parent. If you want to avoid that, force_empty will be useful. | |
418 | ||
7f016ee8 | 419 | 5.2 stat file |
c863d835 | 420 | |
185efc0f | 421 | memory.stat file includes following statistics |
c863d835 | 422 | |
dc10e281 | 423 | # per-memory cgroup local status |
c863d835 BR |
424 | cache - # of bytes of page cache memory. |
425 | rss - # of bytes of anonymous and swap cache memory. | |
dc10e281 | 426 | mapped_file - # of bytes of mapped file (includes tmpfs/shmem) |
c863d835 BR |
427 | pgpgin - # of pages paged in (equivalent to # of charging events). |
428 | pgpgout - # of pages paged out (equivalent to # of uncharging events). | |
dc10e281 | 429 | swap - # of bytes of swap usage |
c863d835 | 430 | inactive_anon - # of bytes of anonymous memory and swap cache memory on |
dc10e281 KH |
431 | LRU list. |
432 | active_anon - # of bytes of anonymous and swap cache memory on active | |
433 | inactive LRU list. | |
434 | inactive_file - # of bytes of file-backed memory on inactive LRU list. | |
435 | active_file - # of bytes of file-backed memory on active LRU list. | |
c863d835 BR |
436 | unevictable - # of bytes of memory that cannot be reclaimed (mlocked etc). |
437 | ||
dc10e281 KH |
438 | # status considering hierarchy (see memory.use_hierarchy settings) |
439 | ||
440 | hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy | |
441 | under which the memory cgroup is | |
442 | hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to | |
443 | hierarchy under which memory cgroup is. | |
444 | ||
445 | total_cache - sum of all children's "cache" | |
446 | total_rss - sum of all children's "rss" | |
447 | total_mapped_file - sum of all children's "cache" | |
448 | total_pgpgin - sum of all children's "pgpgin" | |
449 | total_pgpgout - sum of all children's "pgpgout" | |
450 | total_swap - sum of all children's "swap" | |
451 | total_inactive_anon - sum of all children's "inactive_anon" | |
452 | total_active_anon - sum of all children's "active_anon" | |
453 | total_inactive_file - sum of all children's "inactive_file" | |
454 | total_active_file - sum of all children's "active_file" | |
455 | total_unevictable - sum of all children's "unevictable" | |
456 | ||
457 | # The following additional stats are dependent on CONFIG_DEBUG_VM. | |
c863d835 | 458 | |
c863d835 BR |
459 | recent_rotated_anon - VM internal parameter. (see mm/vmscan.c) |
460 | recent_rotated_file - VM internal parameter. (see mm/vmscan.c) | |
461 | recent_scanned_anon - VM internal parameter. (see mm/vmscan.c) | |
462 | recent_scanned_file - VM internal parameter. (see mm/vmscan.c) | |
463 | ||
464 | Memo: | |
dc10e281 KH |
465 | recent_rotated means recent frequency of LRU rotation. |
466 | recent_scanned means recent # of scans to LRU. | |
7f016ee8 KM |
467 | showing for better debug please see the code for meanings. |
468 | ||
c863d835 BR |
469 | Note: |
470 | Only anonymous and swap cache memory is listed as part of 'rss' stat. | |
471 | This should not be confused with the true 'resident set size' or the | |
dc10e281 KH |
472 | amount of physical memory used by the cgroup. |
473 | 'rss + file_mapped" will give you resident set size of cgroup. | |
474 | (Note: file and shmem may be shared among other cgroups. In that case, | |
475 | file_mapped is accounted only when the memory cgroup is owner of page | |
476 | cache.) | |
7f016ee8 | 477 | |
a7885eb8 | 478 | 5.3 swappiness |
a7885eb8 | 479 | |
dc10e281 | 480 | Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only. |
a7885eb8 | 481 | |
dc10e281 KH |
482 | Following cgroups' swappiness can't be changed. |
483 | - root cgroup (uses /proc/sys/vm/swappiness). | |
484 | - a cgroup which uses hierarchy and it has other cgroup(s) below it. | |
485 | - a cgroup which uses hierarchy and not the root of hierarchy. | |
486 | ||
487 | 5.4 failcnt | |
488 | ||
489 | A memory cgroup provides memory.failcnt and memory.memsw.failcnt files. | |
490 | This failcnt(== failure count) shows the number of times that a usage counter | |
491 | hit its limit. When a memory cgroup hits a limit, failcnt increases and | |
492 | memory under it will be reclaimed. | |
493 | ||
494 | You can reset failcnt by writing 0 to failcnt file. | |
495 | # echo 0 > .../memory.failcnt | |
a7885eb8 | 496 | |
a111c966 DN |
497 | 5.5 usage_in_bytes |
498 | ||
499 | For efficiency, as other kernel components, memory cgroup uses some optimization | |
500 | to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the | |
501 | method and doesn't show 'exact' value of memory(and swap) usage, it's an fuzz | |
502 | value for efficient access. (Of course, when necessary, it's synchronized.) | |
503 | If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP) | |
504 | value in memory.stat(see 5.2). | |
505 | ||
50c35e5b YH |
506 | 5.6 numa_stat |
507 | ||
508 | This is similar to numa_maps but operates on a per-memcg basis. This is | |
509 | useful for providing visibility into the numa locality information within | |
510 | an memcg since the pages are allowed to be allocated from any physical | |
511 | node. One of the usecases is evaluating application performance by | |
512 | combining this information with the application's cpu allocation. | |
513 | ||
514 | We export "total", "file", "anon" and "unevictable" pages per-node for | |
515 | each memcg. The ouput format of memory.numa_stat is: | |
516 | ||
517 | total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
518 | file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
519 | anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
520 | unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
521 | ||
522 | And we have total = file + anon + unevictable. | |
523 | ||
52bc0d82 | 524 | 6. Hierarchy support |
c1e862c1 | 525 | |
52bc0d82 BS |
526 | The memory controller supports a deep hierarchy and hierarchical accounting. |
527 | The hierarchy is created by creating the appropriate cgroups in the | |
528 | cgroup filesystem. Consider for example, the following cgroup filesystem | |
529 | hierarchy | |
530 | ||
67de0162 | 531 | root |
52bc0d82 | 532 | / | \ |
67de0162 JS |
533 | / | \ |
534 | a b c | |
535 | | \ | |
536 | | \ | |
537 | d e | |
52bc0d82 BS |
538 | |
539 | In the diagram above, with hierarchical accounting enabled, all memory | |
540 | usage of e, is accounted to its ancestors up until the root (i.e, c and root), | |
dc10e281 | 541 | that has memory.use_hierarchy enabled. If one of the ancestors goes over its |
52bc0d82 BS |
542 | limit, the reclaim algorithm reclaims from the tasks in the ancestor and the |
543 | children of the ancestor. | |
544 | ||
545 | 6.1 Enabling hierarchical accounting and reclaim | |
546 | ||
dc10e281 | 547 | A memory cgroup by default disables the hierarchy feature. Support |
52bc0d82 BS |
548 | can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup |
549 | ||
550 | # echo 1 > memory.use_hierarchy | |
551 | ||
552 | The feature can be disabled by | |
553 | ||
554 | # echo 0 > memory.use_hierarchy | |
555 | ||
689bca3b GT |
556 | NOTE1: Enabling/disabling will fail if either the cgroup already has other |
557 | cgroups created below it, or if the parent cgroup has use_hierarchy | |
558 | enabled. | |
52bc0d82 | 559 | |
daaf1e68 | 560 | NOTE2: When panic_on_oom is set to "2", the whole system will panic in |
dc10e281 | 561 | case of an OOM event in any cgroup. |
52bc0d82 | 562 | |
a6df6361 BS |
563 | 7. Soft limits |
564 | ||
565 | Soft limits allow for greater sharing of memory. The idea behind soft limits | |
566 | is to allow control groups to use as much of the memory as needed, provided | |
567 | ||
568 | a. There is no memory contention | |
569 | b. They do not exceed their hard limit | |
570 | ||
dc10e281 | 571 | When the system detects memory contention or low memory, control groups |
a6df6361 BS |
572 | are pushed back to their soft limits. If the soft limit of each control |
573 | group is very high, they are pushed back as much as possible to make | |
574 | sure that one control group does not starve the others of memory. | |
575 | ||
576 | Please note that soft limits is a best effort feature, it comes with | |
577 | no guarantees, but it does its best to make sure that when memory is | |
578 | heavily contended for, memory is allocated based on the soft limit | |
579 | hints/setup. Currently soft limit based reclaim is setup such that | |
580 | it gets invoked from balance_pgdat (kswapd). | |
581 | ||
582 | 7.1 Interface | |
583 | ||
584 | Soft limits can be setup by using the following commands (in this example we | |
dc10e281 | 585 | assume a soft limit of 256 MiB) |
a6df6361 BS |
586 | |
587 | # echo 256M > memory.soft_limit_in_bytes | |
588 | ||
589 | If we want to change this to 1G, we can at any time use | |
590 | ||
591 | # echo 1G > memory.soft_limit_in_bytes | |
592 | ||
593 | NOTE1: Soft limits take effect over a long period of time, since they involve | |
594 | reclaiming memory for balancing between memory cgroups | |
595 | NOTE2: It is recommended to set the soft limit always below the hard limit, | |
596 | otherwise the hard limit will take precedence. | |
597 | ||
7dc74be0 DN |
598 | 8. Move charges at task migration |
599 | ||
600 | Users can move charges associated with a task along with task migration, that | |
601 | is, uncharge task's pages from the old cgroup and charge them to the new cgroup. | |
02491447 DN |
602 | This feature is not supported in !CONFIG_MMU environments because of lack of |
603 | page tables. | |
7dc74be0 DN |
604 | |
605 | 8.1 Interface | |
606 | ||
607 | This feature is disabled by default. It can be enabled(and disabled again) by | |
608 | writing to memory.move_charge_at_immigrate of the destination cgroup. | |
609 | ||
610 | If you want to enable it: | |
611 | ||
612 | # echo (some positive value) > memory.move_charge_at_immigrate | |
613 | ||
614 | Note: Each bits of move_charge_at_immigrate has its own meaning about what type | |
615 | of charges should be moved. See 8.2 for details. | |
616 | Note: Charges are moved only when you move mm->owner, IOW, a leader of a thread | |
617 | group. | |
618 | Note: If we cannot find enough space for the task in the destination cgroup, we | |
619 | try to make space by reclaiming memory. Task migration may fail if we | |
620 | cannot make enough space. | |
dc10e281 | 621 | Note: It can take several seconds if you move charges much. |
7dc74be0 DN |
622 | |
623 | And if you want disable it again: | |
624 | ||
625 | # echo 0 > memory.move_charge_at_immigrate | |
626 | ||
627 | 8.2 Type of charges which can be move | |
628 | ||
629 | Each bits of move_charge_at_immigrate has its own meaning about what type of | |
87946a72 DN |
630 | charges should be moved. But in any cases, it must be noted that an account of |
631 | a page or a swap can be moved only when it is charged to the task's current(old) | |
632 | memory cgroup. | |
7dc74be0 DN |
633 | |
634 | bit | what type of charges would be moved ? | |
635 | -----+------------------------------------------------------------------------ | |
636 | 0 | A charge of an anonymous page(or swap of it) used by the target task. | |
637 | | Those pages and swaps must be used only by the target task. You must | |
638 | | enable Swap Extension(see 2.4) to enable move of swap charges. | |
87946a72 DN |
639 | -----+------------------------------------------------------------------------ |
640 | 1 | A charge of file pages(normal file, tmpfs file(e.g. ipc shared memory) | |
dc10e281 | 641 | | and swaps of tmpfs file) mmapped by the target task. Unlike the case of |
87946a72 DN |
642 | | anonymous pages, file pages(and swaps) in the range mmapped by the task |
643 | | will be moved even if the task hasn't done page fault, i.e. they might | |
644 | | not be the task's "RSS", but other task's "RSS" that maps the same file. | |
645 | | And mapcount of the page is ignored(the page can be moved even if | |
646 | | page_mapcount(page) > 1). You must enable Swap Extension(see 2.4) to | |
647 | | enable move of swap charges. | |
7dc74be0 DN |
648 | |
649 | 8.3 TODO | |
650 | ||
7dc74be0 DN |
651 | - Implement madvise(2) to let users decide the vma to be moved or not to be |
652 | moved. | |
653 | - All of moving charge operations are done under cgroup_mutex. It's not good | |
654 | behavior to hold the mutex too long, so we may need some trick. | |
655 | ||
2e72b634 KS |
656 | 9. Memory thresholds |
657 | ||
dc10e281 | 658 | Memory cgroup implements memory thresholds using cgroups notification |
2e72b634 KS |
659 | API (see cgroups.txt). It allows to register multiple memory and memsw |
660 | thresholds and gets notifications when it crosses. | |
661 | ||
662 | To register a threshold application need: | |
dc10e281 KH |
663 | - create an eventfd using eventfd(2); |
664 | - open memory.usage_in_bytes or memory.memsw.usage_in_bytes; | |
665 | - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to | |
666 | cgroup.event_control. | |
2e72b634 KS |
667 | |
668 | Application will be notified through eventfd when memory usage crosses | |
669 | threshold in any direction. | |
670 | ||
671 | It's applicable for root and non-root cgroup. | |
672 | ||
9490ff27 KH |
673 | 10. OOM Control |
674 | ||
3c11ecf4 KH |
675 | memory.oom_control file is for OOM notification and other controls. |
676 | ||
dc10e281 KH |
677 | Memory cgroup implements OOM notifier using cgroup notification |
678 | API (See cgroups.txt). It allows to register multiple OOM notification | |
679 | delivery and gets notification when OOM happens. | |
9490ff27 KH |
680 | |
681 | To register a notifier, application need: | |
682 | - create an eventfd using eventfd(2) | |
683 | - open memory.oom_control file | |
dc10e281 KH |
684 | - write string like "<event_fd> <fd of memory.oom_control>" to |
685 | cgroup.event_control | |
9490ff27 | 686 | |
dc10e281 | 687 | Application will be notified through eventfd when OOM happens. |
9490ff27 KH |
688 | OOM notification doesn't work for root cgroup. |
689 | ||
dc10e281 KH |
690 | You can disable OOM-killer by writing "1" to memory.oom_control file, as: |
691 | ||
3c11ecf4 KH |
692 | #echo 1 > memory.oom_control |
693 | ||
dc10e281 KH |
694 | This operation is only allowed to the top cgroup of sub-hierarchy. |
695 | If OOM-killer is disabled, tasks under cgroup will hang/sleep | |
696 | in memory cgroup's OOM-waitqueue when they request accountable memory. | |
3c11ecf4 | 697 | |
dc10e281 | 698 | For running them, you have to relax the memory cgroup's OOM status by |
3c11ecf4 KH |
699 | * enlarge limit or reduce usage. |
700 | To reduce usage, | |
701 | * kill some tasks. | |
702 | * move some tasks to other group with account migration. | |
703 | * remove some files (on tmpfs?) | |
704 | ||
705 | Then, stopped tasks will work again. | |
706 | ||
707 | At reading, current status of OOM is shown. | |
708 | oom_kill_disable 0 or 1 (if 1, oom-killer is disabled) | |
dc10e281 | 709 | under_oom 0 or 1 (if 1, the memory cgroup is under OOM, tasks may |
3c11ecf4 | 710 | be stopped.) |
9490ff27 KH |
711 | |
712 | 11. TODO | |
1b6df3aa BS |
713 | |
714 | 1. Add support for accounting huge pages (as a separate controller) | |
dfc05c25 KH |
715 | 2. Make per-cgroup scanner reclaim not-shared pages first |
716 | 3. Teach controller to account for shared-pages | |
628f4235 | 717 | 4. Start reclamation in the background when the limit is |
1b6df3aa | 718 | not yet hit but the usage is getting closer |
1b6df3aa BS |
719 | |
720 | Summary | |
721 | ||
722 | Overall, the memory controller has been a stable controller and has been | |
723 | commented and discussed quite extensively in the community. | |
724 | ||
725 | References | |
726 | ||
727 | 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/ | |
728 | 2. Singh, Balbir. Memory Controller (RSS Control), | |
729 | http://lwn.net/Articles/222762/ | |
730 | 3. Emelianov, Pavel. Resource controllers based on process cgroups | |
731 | http://lkml.org/lkml/2007/3/6/198 | |
732 | 4. Emelianov, Pavel. RSS controller based on process cgroups (v2) | |
2324c5dd | 733 | http://lkml.org/lkml/2007/4/9/78 |
1b6df3aa BS |
734 | 5. Emelianov, Pavel. RSS controller based on process cgroups (v3) |
735 | http://lkml.org/lkml/2007/5/30/244 | |
736 | 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/ | |
737 | 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control | |
738 | subsystem (v3), http://lwn.net/Articles/235534/ | |
2324c5dd | 739 | 8. Singh, Balbir. RSS controller v2 test results (lmbench), |
1b6df3aa | 740 | http://lkml.org/lkml/2007/5/17/232 |
2324c5dd | 741 | 9. Singh, Balbir. RSS controller v2 AIM9 results |
1b6df3aa | 742 | http://lkml.org/lkml/2007/5/18/1 |
2324c5dd | 743 | 10. Singh, Balbir. Memory controller v6 test results, |
1b6df3aa | 744 | http://lkml.org/lkml/2007/8/19/36 |
2324c5dd LZ |
745 | 11. Singh, Balbir. Memory controller introduction (v6), |
746 | http://lkml.org/lkml/2007/8/17/69 | |
1b6df3aa BS |
747 | 12. Corbet, Jonathan, Controlling memory use in cgroups, |
748 | http://lwn.net/Articles/243795/ |