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 | |
3aaabe23 | 81 | memory.kmem.tcp.limit_in_bytes # set/show hard limit for tcp buf memory |
5a6dd343 | 82 | memory.kmem.tcp.usage_in_bytes # show current tcp buf memory allocation |
e5671dfa | 83 | |
1b6df3aa BS |
84 | 1. History |
85 | ||
86 | The memory controller has a long history. A request for comments for the memory | |
87 | controller was posted by Balbir Singh [1]. At the time the RFC was posted | |
88 | there were several implementations for memory control. The goal of the | |
89 | RFC was to build consensus and agreement for the minimal features required | |
90 | for memory control. The first RSS controller was posted by Balbir Singh[2] | |
91 | in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the | |
92 | RSS controller. At OLS, at the resource management BoF, everyone suggested | |
93 | that we handle both page cache and RSS together. Another request was raised | |
94 | to allow user space handling of OOM. The current memory controller is | |
95 | at version 6; it combines both mapped (RSS) and unmapped Page | |
96 | Cache Control [11]. | |
97 | ||
98 | 2. Memory Control | |
99 | ||
100 | Memory is a unique resource in the sense that it is present in a limited | |
101 | amount. If a task requires a lot of CPU processing, the task can spread | |
102 | its processing over a period of hours, days, months or years, but with | |
103 | memory, the same physical memory needs to be reused to accomplish the task. | |
104 | ||
105 | The memory controller implementation has been divided into phases. These | |
106 | are: | |
107 | ||
108 | 1. Memory controller | |
109 | 2. mlock(2) controller | |
110 | 3. Kernel user memory accounting and slab control | |
111 | 4. user mappings length controller | |
112 | ||
113 | The memory controller is the first controller developed. | |
114 | ||
115 | 2.1. Design | |
116 | ||
117 | The core of the design is a counter called the res_counter. The res_counter | |
118 | tracks the current memory usage and limit of the group of processes associated | |
119 | with the controller. Each cgroup has a memory controller specific data | |
120 | structure (mem_cgroup) associated with it. | |
121 | ||
122 | 2.2. Accounting | |
123 | ||
124 | +--------------------+ | |
125 | | mem_cgroup | | |
126 | | (res_counter) | | |
127 | +--------------------+ | |
128 | / ^ \ | |
129 | / | \ | |
130 | +---------------+ | +---------------+ | |
131 | | mm_struct | |.... | mm_struct | | |
132 | | | | | | | |
133 | +---------------+ | +---------------+ | |
134 | | | |
135 | + --------------+ | |
136 | | | |
137 | +---------------+ +------+--------+ | |
138 | | page +----------> page_cgroup| | |
139 | | | | | | |
140 | +---------------+ +---------------+ | |
141 | ||
142 | (Figure 1: Hierarchy of Accounting) | |
143 | ||
144 | ||
145 | Figure 1 shows the important aspects of the controller | |
146 | ||
147 | 1. Accounting happens per cgroup | |
148 | 2. Each mm_struct knows about which cgroup it belongs to | |
149 | 3. Each page has a pointer to the page_cgroup, which in turn knows the | |
150 | cgroup it belongs to | |
151 | ||
152 | The accounting is done as follows: mem_cgroup_charge() is invoked to setup | |
153 | the necessary data structures and check if the cgroup that is being charged | |
154 | is over its limit. If it is then reclaim is invoked on the cgroup. | |
155 | More details can be found in the reclaim section of this document. | |
156 | If everything goes well, a page meta-data-structure called page_cgroup is | |
dc10e281 KH |
157 | updated. page_cgroup has its own LRU on cgroup. |
158 | (*) page_cgroup structure is allocated at boot/memory-hotplug time. | |
1b6df3aa BS |
159 | |
160 | 2.2.1 Accounting details | |
161 | ||
5b4e655e | 162 | All mapped anon pages (RSS) and cache pages (Page Cache) are accounted. |
dc10e281 KH |
163 | Some pages which are never reclaimable and will not be on the global LRU |
164 | are not accounted. We just account pages under usual VM management. | |
5b4e655e KH |
165 | |
166 | RSS pages are accounted at page_fault unless they've already been accounted | |
167 | for earlier. A file page will be accounted for as Page Cache when it's | |
168 | inserted into inode (radix-tree). While it's mapped into the page tables of | |
169 | processes, duplicate accounting is carefully avoided. | |
170 | ||
171 | A RSS page is unaccounted when it's fully unmapped. A PageCache page is | |
dc10e281 KH |
172 | unaccounted when it's removed from radix-tree. Even if RSS pages are fully |
173 | unmapped (by kswapd), they may exist as SwapCache in the system until they | |
174 | are really freed. Such SwapCaches also also accounted. | |
175 | A swapped-in page is not accounted until it's mapped. | |
176 | ||
177 | Note: The kernel does swapin-readahead and read multiple swaps at once. | |
178 | This means swapped-in pages may contain pages for other tasks than a task | |
179 | causing page fault. So, we avoid accounting at swap-in I/O. | |
5b4e655e KH |
180 | |
181 | At page migration, accounting information is kept. | |
182 | ||
dc10e281 KH |
183 | Note: we just account pages-on-LRU because our purpose is to control amount |
184 | of used pages; not-on-LRU pages tend to be out-of-control from VM view. | |
1b6df3aa BS |
185 | |
186 | 2.3 Shared Page Accounting | |
187 | ||
188 | Shared pages are accounted on the basis of the first touch approach. The | |
189 | cgroup that first touches a page is accounted for the page. The principle | |
190 | behind this approach is that a cgroup that aggressively uses a shared | |
191 | page will eventually get charged for it (once it is uncharged from | |
192 | the cgroup that brought it in -- this will happen on memory pressure). | |
193 | ||
67de0162 | 194 | Exception: If CONFIG_CGROUP_CGROUP_MEM_RES_CTLR_SWAP is not used. |
8c7c6e34 | 195 | When you do swapoff and make swapped-out pages of shmem(tmpfs) to |
d13d1443 KH |
196 | be backed into memory in force, charges for pages are accounted against the |
197 | caller of swapoff rather than the users of shmem. | |
198 | ||
199 | ||
8c7c6e34 | 200 | 2.4 Swap Extension (CONFIG_CGROUP_MEM_RES_CTLR_SWAP) |
dc10e281 | 201 | |
8c7c6e34 KH |
202 | Swap Extension allows you to record charge for swap. A swapped-in page is |
203 | charged back to original page allocator if possible. | |
204 | ||
205 | When swap is accounted, following files are added. | |
206 | - memory.memsw.usage_in_bytes. | |
207 | - memory.memsw.limit_in_bytes. | |
208 | ||
dc10e281 KH |
209 | memsw means memory+swap. Usage of memory+swap is limited by |
210 | memsw.limit_in_bytes. | |
211 | ||
212 | Example: Assume a system with 4G of swap. A task which allocates 6G of memory | |
213 | (by mistake) under 2G memory limitation will use all swap. | |
214 | In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap. | |
215 | By using memsw limit, you can avoid system OOM which can be caused by swap | |
216 | shortage. | |
8c7c6e34 | 217 | |
dc10e281 | 218 | * why 'memory+swap' rather than swap. |
8c7c6e34 KH |
219 | The global LRU(kswapd) can swap out arbitrary pages. Swap-out means |
220 | to move account from memory to swap...there is no change in usage of | |
dc10e281 KH |
221 | memory+swap. In other words, when we want to limit the usage of swap without |
222 | affecting global LRU, memory+swap limit is better than just limiting swap from | |
22a668d7 KH |
223 | OS point of view. |
224 | ||
225 | * What happens when a cgroup hits memory.memsw.limit_in_bytes | |
67de0162 | 226 | When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out |
22a668d7 KH |
227 | in this cgroup. Then, swap-out will not be done by cgroup routine and file |
228 | caches are dropped. But as mentioned above, global LRU can do swapout memory | |
229 | from it for sanity of the system's memory management state. You can't forbid | |
230 | it by cgroup. | |
8c7c6e34 KH |
231 | |
232 | 2.5 Reclaim | |
1b6df3aa | 233 | |
dc10e281 KH |
234 | Each cgroup maintains a per cgroup LRU which has the same structure as |
235 | global VM. When a cgroup goes over its limit, we first try | |
1b6df3aa BS |
236 | to reclaim memory from the cgroup so as to make space for the new |
237 | pages that the cgroup has touched. If the reclaim is unsuccessful, | |
238 | an OOM routine is invoked to select and kill the bulkiest task in the | |
dc10e281 | 239 | cgroup. (See 10. OOM Control below.) |
1b6df3aa BS |
240 | |
241 | The reclaim algorithm has not been modified for cgroups, except that | |
242 | pages that are selected for reclaiming come from the per cgroup LRU | |
243 | list. | |
244 | ||
4b3bde4c BS |
245 | NOTE: Reclaim does not work for the root cgroup, since we cannot set any |
246 | limits on the root cgroup. | |
247 | ||
daaf1e68 KH |
248 | Note2: When panic_on_oom is set to "2", the whole system will panic. |
249 | ||
9490ff27 KH |
250 | When oom event notifier is registered, event will be delivered. |
251 | (See oom_control section) | |
252 | ||
dc10e281 | 253 | 2.6 Locking |
1b6df3aa | 254 | |
dc10e281 KH |
255 | lock_page_cgroup()/unlock_page_cgroup() should not be called under |
256 | mapping->tree_lock. | |
1b6df3aa | 257 | |
dc10e281 KH |
258 | Other lock order is following: |
259 | PG_locked. | |
260 | mm->page_table_lock | |
261 | zone->lru_lock | |
262 | lock_page_cgroup. | |
263 | In many cases, just lock_page_cgroup() is called. | |
264 | per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by | |
265 | zone->lru_lock, it has no lock of its own. | |
1b6df3aa | 266 | |
e5671dfa GC |
267 | 2.7 Kernel Memory Extension (CONFIG_CGROUP_MEM_RES_CTLR_KMEM) |
268 | ||
269 | With the Kernel memory extension, the Memory Controller is able to limit | |
270 | the amount of kernel memory used by the system. Kernel memory is fundamentally | |
271 | different than user memory, since it can't be swapped out, which makes it | |
272 | possible to DoS the system by consuming too much of this precious resource. | |
273 | ||
274 | Some kernel memory resources may be accounted and limited separately from the | |
275 | main "kmem" resource. For instance, a slab cache that is considered important | |
276 | enough to be limited separately may have its own knobs. | |
277 | ||
278 | Kernel memory limits are not imposed for the root cgroup. Usage for the root | |
279 | cgroup may or may not be accounted. | |
280 | ||
281 | Memory limits as specified by the standard Memory Controller may or may not | |
282 | take kernel memory into consideration. This is achieved through the file | |
283 | memory.independent_kmem_limit. A Value different than 0 will allow for kernel | |
284 | memory to be controlled separately. | |
285 | ||
286 | When kernel memory limits are not independent, the limit values set in | |
287 | memory.kmem files are ignored. | |
288 | ||
289 | Currently no soft limit is implemented for kernel memory. It is future work | |
290 | to trigger slab reclaim when those limits are reached. | |
291 | ||
292 | 2.7.1 Current Kernel Memory resources accounted | |
293 | ||
e1aab161 GC |
294 | * sockets memory pressure: some sockets protocols have memory pressure |
295 | thresholds. The Memory Controller allows them to be controlled individually | |
296 | per cgroup, instead of globally. | |
e5671dfa | 297 | |
d1a4c0b3 GC |
298 | * tcp memory pressure: sockets memory pressure for the tcp protocol. |
299 | ||
1b6df3aa BS |
300 | 3. User Interface |
301 | ||
302 | 0. Configuration | |
303 | ||
304 | a. Enable CONFIG_CGROUPS | |
305 | b. Enable CONFIG_RESOURCE_COUNTERS | |
00f0b825 | 306 | c. Enable CONFIG_CGROUP_MEM_RES_CTLR |
dc10e281 | 307 | d. Enable CONFIG_CGROUP_MEM_RES_CTLR_SWAP (to use swap extension) |
1b6df3aa | 308 | |
f6e07d38 JS |
309 | 1. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?) |
310 | # mount -t tmpfs none /sys/fs/cgroup | |
311 | # mkdir /sys/fs/cgroup/memory | |
312 | # mount -t cgroup none /sys/fs/cgroup/memory -o memory | |
1b6df3aa BS |
313 | |
314 | 2. Make the new group and move bash into it | |
f6e07d38 JS |
315 | # mkdir /sys/fs/cgroup/memory/0 |
316 | # echo $$ > /sys/fs/cgroup/memory/0/tasks | |
1b6df3aa | 317 | |
dc10e281 | 318 | Since now we're in the 0 cgroup, we can alter the memory limit: |
f6e07d38 | 319 | # echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes |
0eea1030 BS |
320 | |
321 | NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo, | |
dc10e281 KH |
322 | mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.) |
323 | ||
c5b947b2 | 324 | NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited). |
4b3bde4c | 325 | NOTE: We cannot set limits on the root cgroup any more. |
0eea1030 | 326 | |
f6e07d38 | 327 | # cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes |
2324c5dd | 328 | 4194304 |
0eea1030 | 329 | |
1b6df3aa | 330 | We can check the usage: |
f6e07d38 | 331 | # cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes |
2324c5dd | 332 | 1216512 |
0eea1030 BS |
333 | |
334 | A successful write to this file does not guarantee a successful set of | |
dc10e281 | 335 | this limit to the value written into the file. This can be due to a |
0eea1030 | 336 | number of factors, such as rounding up to page boundaries or the total |
dc10e281 | 337 | availability of memory on the system. The user is required to re-read |
0eea1030 BS |
338 | this file after a write to guarantee the value committed by the kernel. |
339 | ||
fb78922c | 340 | # echo 1 > memory.limit_in_bytes |
0eea1030 | 341 | # cat memory.limit_in_bytes |
2324c5dd | 342 | 4096 |
1b6df3aa BS |
343 | |
344 | The memory.failcnt field gives the number of times that the cgroup limit was | |
345 | exceeded. | |
346 | ||
dfc05c25 KH |
347 | The memory.stat file gives accounting information. Now, the number of |
348 | caches, RSS and Active pages/Inactive pages are shown. | |
349 | ||
1b6df3aa BS |
350 | 4. Testing |
351 | ||
dc10e281 KH |
352 | For testing features and implementation, see memcg_test.txt. |
353 | ||
354 | Performance test is also important. To see pure memory controller's overhead, | |
355 | testing on tmpfs will give you good numbers of small overheads. | |
356 | Example: do kernel make on tmpfs. | |
357 | ||
358 | Page-fault scalability is also important. At measuring parallel | |
359 | page fault test, multi-process test may be better than multi-thread | |
360 | test because it has noise of shared objects/status. | |
361 | ||
362 | But the above two are testing extreme situations. | |
363 | Trying usual test under memory controller is always helpful. | |
1b6df3aa BS |
364 | |
365 | 4.1 Troubleshooting | |
366 | ||
367 | Sometimes a user might find that the application under a cgroup is | |
dc10e281 | 368 | terminated by OOM killer. There are several causes for this: |
1b6df3aa BS |
369 | |
370 | 1. The cgroup limit is too low (just too low to do anything useful) | |
371 | 2. The user is using anonymous memory and swap is turned off or too low | |
372 | ||
373 | A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of | |
374 | some of the pages cached in the cgroup (page cache pages). | |
375 | ||
dc10e281 KH |
376 | To know what happens, disable OOM_Kill by 10. OOM Control(see below) and |
377 | seeing what happens will be helpful. | |
378 | ||
1b6df3aa BS |
379 | 4.2 Task migration |
380 | ||
a33f3224 | 381 | When a task migrates from one cgroup to another, its charge is not |
7dc74be0 | 382 | carried forward by default. The pages allocated from the original cgroup still |
1b6df3aa BS |
383 | remain charged to it, the charge is dropped when the page is freed or |
384 | reclaimed. | |
385 | ||
dc10e281 KH |
386 | You can move charges of a task along with task migration. |
387 | See 8. "Move charges at task migration" | |
7dc74be0 | 388 | |
1b6df3aa BS |
389 | 4.3 Removing a cgroup |
390 | ||
391 | A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a | |
392 | cgroup might have some charge associated with it, even though all | |
dc10e281 KH |
393 | tasks have migrated away from it. (because we charge against pages, not |
394 | against tasks.) | |
395 | ||
396 | Such charges are freed or moved to their parent. At moving, both of RSS | |
397 | and CACHES are moved to parent. | |
398 | rmdir() may return -EBUSY if freeing/moving fails. See 5.1 also. | |
1b6df3aa | 399 | |
8c7c6e34 KH |
400 | Charges recorded in swap information is not updated at removal of cgroup. |
401 | Recorded information is discarded and a cgroup which uses swap (swapcache) | |
402 | will be charged as a new owner of it. | |
403 | ||
404 | ||
c1e862c1 KH |
405 | 5. Misc. interfaces. |
406 | ||
407 | 5.1 force_empty | |
408 | memory.force_empty interface is provided to make cgroup's memory usage empty. | |
409 | You can use this interface only when the cgroup has no tasks. | |
410 | When writing anything to this | |
411 | ||
412 | # echo 0 > memory.force_empty | |
413 | ||
dc10e281 KH |
414 | Almost all pages tracked by this memory cgroup will be unmapped and freed. |
415 | Some pages cannot be freed because they are locked or in-use. Such pages are | |
416 | moved to parent and this cgroup will be empty. This may return -EBUSY if | |
417 | VM is too busy to free/move all pages immediately. | |
c1e862c1 KH |
418 | |
419 | Typical use case of this interface is that calling this before rmdir(). | |
420 | Because rmdir() moves all pages to parent, some out-of-use page caches can be | |
421 | moved to the parent. If you want to avoid that, force_empty will be useful. | |
422 | ||
7f016ee8 | 423 | 5.2 stat file |
c863d835 | 424 | |
185efc0f | 425 | memory.stat file includes following statistics |
c863d835 | 426 | |
dc10e281 | 427 | # per-memory cgroup local status |
c863d835 BR |
428 | cache - # of bytes of page cache memory. |
429 | rss - # of bytes of anonymous and swap cache memory. | |
dc10e281 | 430 | mapped_file - # of bytes of mapped file (includes tmpfs/shmem) |
c863d835 BR |
431 | pgpgin - # of pages paged in (equivalent to # of charging events). |
432 | pgpgout - # of pages paged out (equivalent to # of uncharging events). | |
dc10e281 | 433 | swap - # of bytes of swap usage |
c863d835 | 434 | inactive_anon - # of bytes of anonymous memory and swap cache memory on |
dc10e281 KH |
435 | LRU list. |
436 | active_anon - # of bytes of anonymous and swap cache memory on active | |
437 | inactive LRU list. | |
438 | inactive_file - # of bytes of file-backed memory on inactive LRU list. | |
439 | active_file - # of bytes of file-backed memory on active LRU list. | |
c863d835 BR |
440 | unevictable - # of bytes of memory that cannot be reclaimed (mlocked etc). |
441 | ||
dc10e281 KH |
442 | # status considering hierarchy (see memory.use_hierarchy settings) |
443 | ||
444 | hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy | |
445 | under which the memory cgroup is | |
446 | hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to | |
447 | hierarchy under which memory cgroup is. | |
448 | ||
449 | total_cache - sum of all children's "cache" | |
450 | total_rss - sum of all children's "rss" | |
451 | total_mapped_file - sum of all children's "cache" | |
452 | total_pgpgin - sum of all children's "pgpgin" | |
453 | total_pgpgout - sum of all children's "pgpgout" | |
454 | total_swap - sum of all children's "swap" | |
455 | total_inactive_anon - sum of all children's "inactive_anon" | |
456 | total_active_anon - sum of all children's "active_anon" | |
457 | total_inactive_file - sum of all children's "inactive_file" | |
458 | total_active_file - sum of all children's "active_file" | |
459 | total_unevictable - sum of all children's "unevictable" | |
460 | ||
461 | # The following additional stats are dependent on CONFIG_DEBUG_VM. | |
c863d835 | 462 | |
c863d835 BR |
463 | recent_rotated_anon - VM internal parameter. (see mm/vmscan.c) |
464 | recent_rotated_file - VM internal parameter. (see mm/vmscan.c) | |
465 | recent_scanned_anon - VM internal parameter. (see mm/vmscan.c) | |
466 | recent_scanned_file - VM internal parameter. (see mm/vmscan.c) | |
467 | ||
468 | Memo: | |
dc10e281 KH |
469 | recent_rotated means recent frequency of LRU rotation. |
470 | recent_scanned means recent # of scans to LRU. | |
7f016ee8 KM |
471 | showing for better debug please see the code for meanings. |
472 | ||
c863d835 BR |
473 | Note: |
474 | Only anonymous and swap cache memory is listed as part of 'rss' stat. | |
475 | This should not be confused with the true 'resident set size' or the | |
dc10e281 KH |
476 | amount of physical memory used by the cgroup. |
477 | 'rss + file_mapped" will give you resident set size of cgroup. | |
478 | (Note: file and shmem may be shared among other cgroups. In that case, | |
479 | file_mapped is accounted only when the memory cgroup is owner of page | |
480 | cache.) | |
7f016ee8 | 481 | |
a7885eb8 | 482 | 5.3 swappiness |
a7885eb8 | 483 | |
dc10e281 | 484 | Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only. |
a7885eb8 | 485 | |
dc10e281 KH |
486 | Following cgroups' swappiness can't be changed. |
487 | - root cgroup (uses /proc/sys/vm/swappiness). | |
488 | - a cgroup which uses hierarchy and it has other cgroup(s) below it. | |
489 | - a cgroup which uses hierarchy and not the root of hierarchy. | |
490 | ||
491 | 5.4 failcnt | |
492 | ||
493 | A memory cgroup provides memory.failcnt and memory.memsw.failcnt files. | |
494 | This failcnt(== failure count) shows the number of times that a usage counter | |
495 | hit its limit. When a memory cgroup hits a limit, failcnt increases and | |
496 | memory under it will be reclaimed. | |
497 | ||
498 | You can reset failcnt by writing 0 to failcnt file. | |
499 | # echo 0 > .../memory.failcnt | |
a7885eb8 | 500 | |
a111c966 DN |
501 | 5.5 usage_in_bytes |
502 | ||
503 | For efficiency, as other kernel components, memory cgroup uses some optimization | |
504 | to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the | |
505 | method and doesn't show 'exact' value of memory(and swap) usage, it's an fuzz | |
506 | value for efficient access. (Of course, when necessary, it's synchronized.) | |
507 | If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP) | |
508 | value in memory.stat(see 5.2). | |
509 | ||
50c35e5b YH |
510 | 5.6 numa_stat |
511 | ||
512 | This is similar to numa_maps but operates on a per-memcg basis. This is | |
513 | useful for providing visibility into the numa locality information within | |
514 | an memcg since the pages are allowed to be allocated from any physical | |
515 | node. One of the usecases is evaluating application performance by | |
516 | combining this information with the application's cpu allocation. | |
517 | ||
518 | We export "total", "file", "anon" and "unevictable" pages per-node for | |
519 | each memcg. The ouput format of memory.numa_stat is: | |
520 | ||
521 | total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
522 | file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
523 | anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
524 | unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
525 | ||
526 | And we have total = file + anon + unevictable. | |
527 | ||
52bc0d82 | 528 | 6. Hierarchy support |
c1e862c1 | 529 | |
52bc0d82 BS |
530 | The memory controller supports a deep hierarchy and hierarchical accounting. |
531 | The hierarchy is created by creating the appropriate cgroups in the | |
532 | cgroup filesystem. Consider for example, the following cgroup filesystem | |
533 | hierarchy | |
534 | ||
67de0162 | 535 | root |
52bc0d82 | 536 | / | \ |
67de0162 JS |
537 | / | \ |
538 | a b c | |
539 | | \ | |
540 | | \ | |
541 | d e | |
52bc0d82 BS |
542 | |
543 | In the diagram above, with hierarchical accounting enabled, all memory | |
544 | usage of e, is accounted to its ancestors up until the root (i.e, c and root), | |
dc10e281 | 545 | that has memory.use_hierarchy enabled. If one of the ancestors goes over its |
52bc0d82 BS |
546 | limit, the reclaim algorithm reclaims from the tasks in the ancestor and the |
547 | children of the ancestor. | |
548 | ||
549 | 6.1 Enabling hierarchical accounting and reclaim | |
550 | ||
dc10e281 | 551 | A memory cgroup by default disables the hierarchy feature. Support |
52bc0d82 BS |
552 | can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup |
553 | ||
554 | # echo 1 > memory.use_hierarchy | |
555 | ||
556 | The feature can be disabled by | |
557 | ||
558 | # echo 0 > memory.use_hierarchy | |
559 | ||
689bca3b GT |
560 | NOTE1: Enabling/disabling will fail if either the cgroup already has other |
561 | cgroups created below it, or if the parent cgroup has use_hierarchy | |
562 | enabled. | |
52bc0d82 | 563 | |
daaf1e68 | 564 | NOTE2: When panic_on_oom is set to "2", the whole system will panic in |
dc10e281 | 565 | case of an OOM event in any cgroup. |
52bc0d82 | 566 | |
a6df6361 BS |
567 | 7. Soft limits |
568 | ||
569 | Soft limits allow for greater sharing of memory. The idea behind soft limits | |
570 | is to allow control groups to use as much of the memory as needed, provided | |
571 | ||
572 | a. There is no memory contention | |
573 | b. They do not exceed their hard limit | |
574 | ||
dc10e281 | 575 | When the system detects memory contention or low memory, control groups |
a6df6361 BS |
576 | are pushed back to their soft limits. If the soft limit of each control |
577 | group is very high, they are pushed back as much as possible to make | |
578 | sure that one control group does not starve the others of memory. | |
579 | ||
580 | Please note that soft limits is a best effort feature, it comes with | |
581 | no guarantees, but it does its best to make sure that when memory is | |
582 | heavily contended for, memory is allocated based on the soft limit | |
583 | hints/setup. Currently soft limit based reclaim is setup such that | |
584 | it gets invoked from balance_pgdat (kswapd). | |
585 | ||
586 | 7.1 Interface | |
587 | ||
588 | Soft limits can be setup by using the following commands (in this example we | |
dc10e281 | 589 | assume a soft limit of 256 MiB) |
a6df6361 BS |
590 | |
591 | # echo 256M > memory.soft_limit_in_bytes | |
592 | ||
593 | If we want to change this to 1G, we can at any time use | |
594 | ||
595 | # echo 1G > memory.soft_limit_in_bytes | |
596 | ||
597 | NOTE1: Soft limits take effect over a long period of time, since they involve | |
598 | reclaiming memory for balancing between memory cgroups | |
599 | NOTE2: It is recommended to set the soft limit always below the hard limit, | |
600 | otherwise the hard limit will take precedence. | |
601 | ||
7dc74be0 DN |
602 | 8. Move charges at task migration |
603 | ||
604 | Users can move charges associated with a task along with task migration, that | |
605 | is, uncharge task's pages from the old cgroup and charge them to the new cgroup. | |
02491447 DN |
606 | This feature is not supported in !CONFIG_MMU environments because of lack of |
607 | page tables. | |
7dc74be0 DN |
608 | |
609 | 8.1 Interface | |
610 | ||
611 | This feature is disabled by default. It can be enabled(and disabled again) by | |
612 | writing to memory.move_charge_at_immigrate of the destination cgroup. | |
613 | ||
614 | If you want to enable it: | |
615 | ||
616 | # echo (some positive value) > memory.move_charge_at_immigrate | |
617 | ||
618 | Note: Each bits of move_charge_at_immigrate has its own meaning about what type | |
619 | of charges should be moved. See 8.2 for details. | |
620 | Note: Charges are moved only when you move mm->owner, IOW, a leader of a thread | |
621 | group. | |
622 | Note: If we cannot find enough space for the task in the destination cgroup, we | |
623 | try to make space by reclaiming memory. Task migration may fail if we | |
624 | cannot make enough space. | |
dc10e281 | 625 | Note: It can take several seconds if you move charges much. |
7dc74be0 DN |
626 | |
627 | And if you want disable it again: | |
628 | ||
629 | # echo 0 > memory.move_charge_at_immigrate | |
630 | ||
631 | 8.2 Type of charges which can be move | |
632 | ||
633 | Each bits of move_charge_at_immigrate has its own meaning about what type of | |
87946a72 DN |
634 | charges should be moved. But in any cases, it must be noted that an account of |
635 | a page or a swap can be moved only when it is charged to the task's current(old) | |
636 | memory cgroup. | |
7dc74be0 DN |
637 | |
638 | bit | what type of charges would be moved ? | |
639 | -----+------------------------------------------------------------------------ | |
640 | 0 | A charge of an anonymous page(or swap of it) used by the target task. | |
641 | | Those pages and swaps must be used only by the target task. You must | |
642 | | enable Swap Extension(see 2.4) to enable move of swap charges. | |
87946a72 DN |
643 | -----+------------------------------------------------------------------------ |
644 | 1 | A charge of file pages(normal file, tmpfs file(e.g. ipc shared memory) | |
dc10e281 | 645 | | and swaps of tmpfs file) mmapped by the target task. Unlike the case of |
87946a72 DN |
646 | | anonymous pages, file pages(and swaps) in the range mmapped by the task |
647 | | will be moved even if the task hasn't done page fault, i.e. they might | |
648 | | not be the task's "RSS", but other task's "RSS" that maps the same file. | |
649 | | And mapcount of the page is ignored(the page can be moved even if | |
650 | | page_mapcount(page) > 1). You must enable Swap Extension(see 2.4) to | |
651 | | enable move of swap charges. | |
7dc74be0 DN |
652 | |
653 | 8.3 TODO | |
654 | ||
7dc74be0 DN |
655 | - Implement madvise(2) to let users decide the vma to be moved or not to be |
656 | moved. | |
657 | - All of moving charge operations are done under cgroup_mutex. It's not good | |
658 | behavior to hold the mutex too long, so we may need some trick. | |
659 | ||
2e72b634 KS |
660 | 9. Memory thresholds |
661 | ||
dc10e281 | 662 | Memory cgroup implements memory thresholds using cgroups notification |
2e72b634 KS |
663 | API (see cgroups.txt). It allows to register multiple memory and memsw |
664 | thresholds and gets notifications when it crosses. | |
665 | ||
666 | To register a threshold application need: | |
dc10e281 KH |
667 | - create an eventfd using eventfd(2); |
668 | - open memory.usage_in_bytes or memory.memsw.usage_in_bytes; | |
669 | - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to | |
670 | cgroup.event_control. | |
2e72b634 KS |
671 | |
672 | Application will be notified through eventfd when memory usage crosses | |
673 | threshold in any direction. | |
674 | ||
675 | It's applicable for root and non-root cgroup. | |
676 | ||
9490ff27 KH |
677 | 10. OOM Control |
678 | ||
3c11ecf4 KH |
679 | memory.oom_control file is for OOM notification and other controls. |
680 | ||
dc10e281 KH |
681 | Memory cgroup implements OOM notifier using cgroup notification |
682 | API (See cgroups.txt). It allows to register multiple OOM notification | |
683 | delivery and gets notification when OOM happens. | |
9490ff27 KH |
684 | |
685 | To register a notifier, application need: | |
686 | - create an eventfd using eventfd(2) | |
687 | - open memory.oom_control file | |
dc10e281 KH |
688 | - write string like "<event_fd> <fd of memory.oom_control>" to |
689 | cgroup.event_control | |
9490ff27 | 690 | |
dc10e281 | 691 | Application will be notified through eventfd when OOM happens. |
9490ff27 KH |
692 | OOM notification doesn't work for root cgroup. |
693 | ||
dc10e281 KH |
694 | You can disable OOM-killer by writing "1" to memory.oom_control file, as: |
695 | ||
3c11ecf4 KH |
696 | #echo 1 > memory.oom_control |
697 | ||
dc10e281 KH |
698 | This operation is only allowed to the top cgroup of sub-hierarchy. |
699 | If OOM-killer is disabled, tasks under cgroup will hang/sleep | |
700 | in memory cgroup's OOM-waitqueue when they request accountable memory. | |
3c11ecf4 | 701 | |
dc10e281 | 702 | For running them, you have to relax the memory cgroup's OOM status by |
3c11ecf4 KH |
703 | * enlarge limit or reduce usage. |
704 | To reduce usage, | |
705 | * kill some tasks. | |
706 | * move some tasks to other group with account migration. | |
707 | * remove some files (on tmpfs?) | |
708 | ||
709 | Then, stopped tasks will work again. | |
710 | ||
711 | At reading, current status of OOM is shown. | |
712 | oom_kill_disable 0 or 1 (if 1, oom-killer is disabled) | |
dc10e281 | 713 | under_oom 0 or 1 (if 1, the memory cgroup is under OOM, tasks may |
3c11ecf4 | 714 | be stopped.) |
9490ff27 KH |
715 | |
716 | 11. TODO | |
1b6df3aa BS |
717 | |
718 | 1. Add support for accounting huge pages (as a separate controller) | |
dfc05c25 KH |
719 | 2. Make per-cgroup scanner reclaim not-shared pages first |
720 | 3. Teach controller to account for shared-pages | |
628f4235 | 721 | 4. Start reclamation in the background when the limit is |
1b6df3aa | 722 | not yet hit but the usage is getting closer |
1b6df3aa BS |
723 | |
724 | Summary | |
725 | ||
726 | Overall, the memory controller has been a stable controller and has been | |
727 | commented and discussed quite extensively in the community. | |
728 | ||
729 | References | |
730 | ||
731 | 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/ | |
732 | 2. Singh, Balbir. Memory Controller (RSS Control), | |
733 | http://lwn.net/Articles/222762/ | |
734 | 3. Emelianov, Pavel. Resource controllers based on process cgroups | |
735 | http://lkml.org/lkml/2007/3/6/198 | |
736 | 4. Emelianov, Pavel. RSS controller based on process cgroups (v2) | |
2324c5dd | 737 | http://lkml.org/lkml/2007/4/9/78 |
1b6df3aa BS |
738 | 5. Emelianov, Pavel. RSS controller based on process cgroups (v3) |
739 | http://lkml.org/lkml/2007/5/30/244 | |
740 | 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/ | |
741 | 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control | |
742 | subsystem (v3), http://lwn.net/Articles/235534/ | |
2324c5dd | 743 | 8. Singh, Balbir. RSS controller v2 test results (lmbench), |
1b6df3aa | 744 | http://lkml.org/lkml/2007/5/17/232 |
2324c5dd | 745 | 9. Singh, Balbir. RSS controller v2 AIM9 results |
1b6df3aa | 746 | http://lkml.org/lkml/2007/5/18/1 |
2324c5dd | 747 | 10. Singh, Balbir. Memory controller v6 test results, |
1b6df3aa | 748 | http://lkml.org/lkml/2007/8/19/36 |
2324c5dd LZ |
749 | 11. Singh, Balbir. Memory controller introduction (v6), |
750 | http://lkml.org/lkml/2007/8/17/69 | |
1b6df3aa BS |
751 | 12. Corbet, Jonathan, Controlling memory use in cgroups, |
752 | http://lwn.net/Articles/243795/ |