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 | ||
65c64ce8 GC |
47 | Kernel memory support is work in progress, and the current version provides |
48 | basically functionality. (See Section 2.7) | |
dc10e281 KH |
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 | |
d66c1ce7 | 64 | memory.memsw.max_usage_in_bytes # show max memory+Swap usage recorded |
dc10e281 KH |
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 | |
3aaabe23 | 75 | memory.kmem.tcp.limit_in_bytes # set/show hard limit for tcp buf memory |
5a6dd343 | 76 | memory.kmem.tcp.usage_in_bytes # show current tcp buf memory allocation |
e5671dfa | 77 | |
1b6df3aa BS |
78 | 1. History |
79 | ||
80 | The memory controller has a long history. A request for comments for the memory | |
81 | controller was posted by Balbir Singh [1]. At the time the RFC was posted | |
82 | there were several implementations for memory control. The goal of the | |
83 | RFC was to build consensus and agreement for the minimal features required | |
84 | for memory control. The first RSS controller was posted by Balbir Singh[2] | |
85 | in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the | |
86 | RSS controller. At OLS, at the resource management BoF, everyone suggested | |
87 | that we handle both page cache and RSS together. Another request was raised | |
88 | to allow user space handling of OOM. The current memory controller is | |
89 | at version 6; it combines both mapped (RSS) and unmapped Page | |
90 | Cache Control [11]. | |
91 | ||
92 | 2. Memory Control | |
93 | ||
94 | Memory is a unique resource in the sense that it is present in a limited | |
95 | amount. If a task requires a lot of CPU processing, the task can spread | |
96 | its processing over a period of hours, days, months or years, but with | |
97 | memory, the same physical memory needs to be reused to accomplish the task. | |
98 | ||
99 | The memory controller implementation has been divided into phases. These | |
100 | are: | |
101 | ||
102 | 1. Memory controller | |
103 | 2. mlock(2) controller | |
104 | 3. Kernel user memory accounting and slab control | |
105 | 4. user mappings length controller | |
106 | ||
107 | The memory controller is the first controller developed. | |
108 | ||
109 | 2.1. Design | |
110 | ||
111 | The core of the design is a counter called the res_counter. The res_counter | |
112 | tracks the current memory usage and limit of the group of processes associated | |
113 | with the controller. Each cgroup has a memory controller specific data | |
114 | structure (mem_cgroup) associated with it. | |
115 | ||
116 | 2.2. Accounting | |
117 | ||
118 | +--------------------+ | |
119 | | mem_cgroup | | |
120 | | (res_counter) | | |
121 | +--------------------+ | |
122 | / ^ \ | |
123 | / | \ | |
124 | +---------------+ | +---------------+ | |
125 | | mm_struct | |.... | mm_struct | | |
126 | | | | | | | |
127 | +---------------+ | +---------------+ | |
128 | | | |
129 | + --------------+ | |
130 | | | |
131 | +---------------+ +------+--------+ | |
132 | | page +----------> page_cgroup| | |
133 | | | | | | |
134 | +---------------+ +---------------+ | |
135 | ||
136 | (Figure 1: Hierarchy of Accounting) | |
137 | ||
138 | ||
139 | Figure 1 shows the important aspects of the controller | |
140 | ||
141 | 1. Accounting happens per cgroup | |
142 | 2. Each mm_struct knows about which cgroup it belongs to | |
143 | 3. Each page has a pointer to the page_cgroup, which in turn knows the | |
144 | cgroup it belongs to | |
145 | ||
146 | The accounting is done as follows: mem_cgroup_charge() is invoked to setup | |
147 | the necessary data structures and check if the cgroup that is being charged | |
148 | is over its limit. If it is then reclaim is invoked on the cgroup. | |
149 | More details can be found in the reclaim section of this document. | |
150 | If everything goes well, a page meta-data-structure called page_cgroup is | |
dc10e281 KH |
151 | updated. page_cgroup has its own LRU on cgroup. |
152 | (*) page_cgroup structure is allocated at boot/memory-hotplug time. | |
1b6df3aa BS |
153 | |
154 | 2.2.1 Accounting details | |
155 | ||
5b4e655e | 156 | All mapped anon pages (RSS) and cache pages (Page Cache) are accounted. |
dc10e281 KH |
157 | Some pages which are never reclaimable and will not be on the global LRU |
158 | are not accounted. We just account pages under usual VM management. | |
5b4e655e KH |
159 | |
160 | RSS pages are accounted at page_fault unless they've already been accounted | |
161 | for earlier. A file page will be accounted for as Page Cache when it's | |
162 | inserted into inode (radix-tree). While it's mapped into the page tables of | |
163 | processes, duplicate accounting is carefully avoided. | |
164 | ||
165 | A RSS page is unaccounted when it's fully unmapped. A PageCache page is | |
dc10e281 KH |
166 | unaccounted when it's removed from radix-tree. Even if RSS pages are fully |
167 | unmapped (by kswapd), they may exist as SwapCache in the system until they | |
168 | are really freed. Such SwapCaches also also accounted. | |
169 | A swapped-in page is not accounted until it's mapped. | |
170 | ||
171 | Note: The kernel does swapin-readahead and read multiple swaps at once. | |
172 | This means swapped-in pages may contain pages for other tasks than a task | |
173 | causing page fault. So, we avoid accounting at swap-in I/O. | |
5b4e655e KH |
174 | |
175 | At page migration, accounting information is kept. | |
176 | ||
dc10e281 KH |
177 | Note: we just account pages-on-LRU because our purpose is to control amount |
178 | of used pages; not-on-LRU pages tend to be out-of-control from VM view. | |
1b6df3aa BS |
179 | |
180 | 2.3 Shared Page Accounting | |
181 | ||
182 | Shared pages are accounted on the basis of the first touch approach. The | |
183 | cgroup that first touches a page is accounted for the page. The principle | |
184 | behind this approach is that a cgroup that aggressively uses a shared | |
185 | page will eventually get charged for it (once it is uncharged from | |
186 | the cgroup that brought it in -- this will happen on memory pressure). | |
187 | ||
67de0162 | 188 | Exception: If CONFIG_CGROUP_CGROUP_MEM_RES_CTLR_SWAP is not used. |
8c7c6e34 | 189 | When you do swapoff and make swapped-out pages of shmem(tmpfs) to |
d13d1443 KH |
190 | be backed into memory in force, charges for pages are accounted against the |
191 | caller of swapoff rather than the users of shmem. | |
192 | ||
193 | ||
8c7c6e34 | 194 | 2.4 Swap Extension (CONFIG_CGROUP_MEM_RES_CTLR_SWAP) |
dc10e281 | 195 | |
8c7c6e34 KH |
196 | Swap Extension allows you to record charge for swap. A swapped-in page is |
197 | charged back to original page allocator if possible. | |
198 | ||
199 | When swap is accounted, following files are added. | |
200 | - memory.memsw.usage_in_bytes. | |
201 | - memory.memsw.limit_in_bytes. | |
202 | ||
dc10e281 KH |
203 | memsw means memory+swap. Usage of memory+swap is limited by |
204 | memsw.limit_in_bytes. | |
205 | ||
206 | Example: Assume a system with 4G of swap. A task which allocates 6G of memory | |
207 | (by mistake) under 2G memory limitation will use all swap. | |
208 | In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap. | |
209 | By using memsw limit, you can avoid system OOM which can be caused by swap | |
210 | shortage. | |
8c7c6e34 | 211 | |
dc10e281 | 212 | * why 'memory+swap' rather than swap. |
8c7c6e34 KH |
213 | The global LRU(kswapd) can swap out arbitrary pages. Swap-out means |
214 | to move account from memory to swap...there is no change in usage of | |
dc10e281 KH |
215 | memory+swap. In other words, when we want to limit the usage of swap without |
216 | affecting global LRU, memory+swap limit is better than just limiting swap from | |
22a668d7 KH |
217 | OS point of view. |
218 | ||
219 | * What happens when a cgroup hits memory.memsw.limit_in_bytes | |
67de0162 | 220 | When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out |
22a668d7 KH |
221 | in this cgroup. Then, swap-out will not be done by cgroup routine and file |
222 | caches are dropped. But as mentioned above, global LRU can do swapout memory | |
223 | from it for sanity of the system's memory management state. You can't forbid | |
224 | it by cgroup. | |
8c7c6e34 KH |
225 | |
226 | 2.5 Reclaim | |
1b6df3aa | 227 | |
dc10e281 KH |
228 | Each cgroup maintains a per cgroup LRU which has the same structure as |
229 | global VM. When a cgroup goes over its limit, we first try | |
1b6df3aa BS |
230 | to reclaim memory from the cgroup so as to make space for the new |
231 | pages that the cgroup has touched. If the reclaim is unsuccessful, | |
232 | an OOM routine is invoked to select and kill the bulkiest task in the | |
dc10e281 | 233 | cgroup. (See 10. OOM Control below.) |
1b6df3aa BS |
234 | |
235 | The reclaim algorithm has not been modified for cgroups, except that | |
236 | pages that are selected for reclaiming come from the per cgroup LRU | |
237 | list. | |
238 | ||
4b3bde4c BS |
239 | NOTE: Reclaim does not work for the root cgroup, since we cannot set any |
240 | limits on the root cgroup. | |
241 | ||
daaf1e68 KH |
242 | Note2: When panic_on_oom is set to "2", the whole system will panic. |
243 | ||
9490ff27 KH |
244 | When oom event notifier is registered, event will be delivered. |
245 | (See oom_control section) | |
246 | ||
dc10e281 | 247 | 2.6 Locking |
1b6df3aa | 248 | |
dc10e281 KH |
249 | lock_page_cgroup()/unlock_page_cgroup() should not be called under |
250 | mapping->tree_lock. | |
1b6df3aa | 251 | |
dc10e281 KH |
252 | Other lock order is following: |
253 | PG_locked. | |
254 | mm->page_table_lock | |
255 | zone->lru_lock | |
256 | lock_page_cgroup. | |
257 | In many cases, just lock_page_cgroup() is called. | |
258 | per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by | |
259 | zone->lru_lock, it has no lock of its own. | |
1b6df3aa | 260 | |
e5671dfa GC |
261 | 2.7 Kernel Memory Extension (CONFIG_CGROUP_MEM_RES_CTLR_KMEM) |
262 | ||
263 | With the Kernel memory extension, the Memory Controller is able to limit | |
264 | the amount of kernel memory used by the system. Kernel memory is fundamentally | |
265 | different than user memory, since it can't be swapped out, which makes it | |
266 | possible to DoS the system by consuming too much of this precious resource. | |
267 | ||
e5671dfa GC |
268 | Kernel memory limits are not imposed for the root cgroup. Usage for the root |
269 | cgroup may or may not be accounted. | |
270 | ||
e5671dfa GC |
271 | Currently no soft limit is implemented for kernel memory. It is future work |
272 | to trigger slab reclaim when those limits are reached. | |
273 | ||
274 | 2.7.1 Current Kernel Memory resources accounted | |
275 | ||
e1aab161 GC |
276 | * sockets memory pressure: some sockets protocols have memory pressure |
277 | thresholds. The Memory Controller allows them to be controlled individually | |
278 | per cgroup, instead of globally. | |
e5671dfa | 279 | |
d1a4c0b3 GC |
280 | * tcp memory pressure: sockets memory pressure for the tcp protocol. |
281 | ||
1b6df3aa BS |
282 | 3. User Interface |
283 | ||
284 | 0. Configuration | |
285 | ||
286 | a. Enable CONFIG_CGROUPS | |
287 | b. Enable CONFIG_RESOURCE_COUNTERS | |
00f0b825 | 288 | c. Enable CONFIG_CGROUP_MEM_RES_CTLR |
dc10e281 | 289 | d. Enable CONFIG_CGROUP_MEM_RES_CTLR_SWAP (to use swap extension) |
1b6df3aa | 290 | |
f6e07d38 JS |
291 | 1. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?) |
292 | # mount -t tmpfs none /sys/fs/cgroup | |
293 | # mkdir /sys/fs/cgroup/memory | |
294 | # mount -t cgroup none /sys/fs/cgroup/memory -o memory | |
1b6df3aa BS |
295 | |
296 | 2. Make the new group and move bash into it | |
f6e07d38 JS |
297 | # mkdir /sys/fs/cgroup/memory/0 |
298 | # echo $$ > /sys/fs/cgroup/memory/0/tasks | |
1b6df3aa | 299 | |
dc10e281 | 300 | Since now we're in the 0 cgroup, we can alter the memory limit: |
f6e07d38 | 301 | # echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes |
0eea1030 BS |
302 | |
303 | NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo, | |
dc10e281 KH |
304 | mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.) |
305 | ||
c5b947b2 | 306 | NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited). |
4b3bde4c | 307 | NOTE: We cannot set limits on the root cgroup any more. |
0eea1030 | 308 | |
f6e07d38 | 309 | # cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes |
2324c5dd | 310 | 4194304 |
0eea1030 | 311 | |
1b6df3aa | 312 | We can check the usage: |
f6e07d38 | 313 | # cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes |
2324c5dd | 314 | 1216512 |
0eea1030 BS |
315 | |
316 | A successful write to this file does not guarantee a successful set of | |
dc10e281 | 317 | this limit to the value written into the file. This can be due to a |
0eea1030 | 318 | number of factors, such as rounding up to page boundaries or the total |
dc10e281 | 319 | availability of memory on the system. The user is required to re-read |
0eea1030 BS |
320 | this file after a write to guarantee the value committed by the kernel. |
321 | ||
fb78922c | 322 | # echo 1 > memory.limit_in_bytes |
0eea1030 | 323 | # cat memory.limit_in_bytes |
2324c5dd | 324 | 4096 |
1b6df3aa BS |
325 | |
326 | The memory.failcnt field gives the number of times that the cgroup limit was | |
327 | exceeded. | |
328 | ||
dfc05c25 KH |
329 | The memory.stat file gives accounting information. Now, the number of |
330 | caches, RSS and Active pages/Inactive pages are shown. | |
331 | ||
1b6df3aa BS |
332 | 4. Testing |
333 | ||
dc10e281 KH |
334 | For testing features and implementation, see memcg_test.txt. |
335 | ||
336 | Performance test is also important. To see pure memory controller's overhead, | |
337 | testing on tmpfs will give you good numbers of small overheads. | |
338 | Example: do kernel make on tmpfs. | |
339 | ||
340 | Page-fault scalability is also important. At measuring parallel | |
341 | page fault test, multi-process test may be better than multi-thread | |
342 | test because it has noise of shared objects/status. | |
343 | ||
344 | But the above two are testing extreme situations. | |
345 | Trying usual test under memory controller is always helpful. | |
1b6df3aa BS |
346 | |
347 | 4.1 Troubleshooting | |
348 | ||
349 | Sometimes a user might find that the application under a cgroup is | |
dc10e281 | 350 | terminated by OOM killer. There are several causes for this: |
1b6df3aa BS |
351 | |
352 | 1. The cgroup limit is too low (just too low to do anything useful) | |
353 | 2. The user is using anonymous memory and swap is turned off or too low | |
354 | ||
355 | A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of | |
356 | some of the pages cached in the cgroup (page cache pages). | |
357 | ||
dc10e281 KH |
358 | To know what happens, disable OOM_Kill by 10. OOM Control(see below) and |
359 | seeing what happens will be helpful. | |
360 | ||
1b6df3aa BS |
361 | 4.2 Task migration |
362 | ||
a33f3224 | 363 | When a task migrates from one cgroup to another, its charge is not |
7dc74be0 | 364 | carried forward by default. The pages allocated from the original cgroup still |
1b6df3aa BS |
365 | remain charged to it, the charge is dropped when the page is freed or |
366 | reclaimed. | |
367 | ||
dc10e281 KH |
368 | You can move charges of a task along with task migration. |
369 | See 8. "Move charges at task migration" | |
7dc74be0 | 370 | |
1b6df3aa BS |
371 | 4.3 Removing a cgroup |
372 | ||
373 | A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a | |
374 | cgroup might have some charge associated with it, even though all | |
dc10e281 KH |
375 | tasks have migrated away from it. (because we charge against pages, not |
376 | against tasks.) | |
377 | ||
378 | Such charges are freed or moved to their parent. At moving, both of RSS | |
379 | and CACHES are moved to parent. | |
380 | rmdir() may return -EBUSY if freeing/moving fails. See 5.1 also. | |
1b6df3aa | 381 | |
8c7c6e34 KH |
382 | Charges recorded in swap information is not updated at removal of cgroup. |
383 | Recorded information is discarded and a cgroup which uses swap (swapcache) | |
384 | will be charged as a new owner of it. | |
385 | ||
386 | ||
c1e862c1 KH |
387 | 5. Misc. interfaces. |
388 | ||
389 | 5.1 force_empty | |
390 | memory.force_empty interface is provided to make cgroup's memory usage empty. | |
391 | You can use this interface only when the cgroup has no tasks. | |
392 | When writing anything to this | |
393 | ||
394 | # echo 0 > memory.force_empty | |
395 | ||
dc10e281 KH |
396 | Almost all pages tracked by this memory cgroup will be unmapped and freed. |
397 | Some pages cannot be freed because they are locked or in-use. Such pages are | |
398 | moved to parent and this cgroup will be empty. This may return -EBUSY if | |
399 | VM is too busy to free/move all pages immediately. | |
c1e862c1 KH |
400 | |
401 | Typical use case of this interface is that calling this before rmdir(). | |
402 | Because rmdir() moves all pages to parent, some out-of-use page caches can be | |
403 | moved to the parent. If you want to avoid that, force_empty will be useful. | |
404 | ||
7f016ee8 | 405 | 5.2 stat file |
c863d835 | 406 | |
185efc0f | 407 | memory.stat file includes following statistics |
c863d835 | 408 | |
dc10e281 | 409 | # per-memory cgroup local status |
c863d835 BR |
410 | cache - # of bytes of page cache memory. |
411 | rss - # of bytes of anonymous and swap cache memory. | |
dc10e281 | 412 | mapped_file - # of bytes of mapped file (includes tmpfs/shmem) |
0527b690 YH |
413 | pgpgin - # of charging events to the memory cgroup. The charging |
414 | event happens each time a page is accounted as either mapped | |
415 | anon page(RSS) or cache page(Page Cache) to the cgroup. | |
416 | pgpgout - # of uncharging events to the memory cgroup. The uncharging | |
417 | event happens each time a page is unaccounted from the cgroup. | |
dc10e281 | 418 | swap - # of bytes of swap usage |
c863d835 | 419 | inactive_anon - # of bytes of anonymous memory and swap cache memory on |
dc10e281 KH |
420 | LRU list. |
421 | active_anon - # of bytes of anonymous and swap cache memory on active | |
422 | inactive LRU list. | |
423 | inactive_file - # of bytes of file-backed memory on inactive LRU list. | |
424 | active_file - # of bytes of file-backed memory on active LRU list. | |
c863d835 BR |
425 | unevictable - # of bytes of memory that cannot be reclaimed (mlocked etc). |
426 | ||
dc10e281 KH |
427 | # status considering hierarchy (see memory.use_hierarchy settings) |
428 | ||
429 | hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy | |
430 | under which the memory cgroup is | |
431 | hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to | |
432 | hierarchy under which memory cgroup is. | |
433 | ||
434 | total_cache - sum of all children's "cache" | |
435 | total_rss - sum of all children's "rss" | |
436 | total_mapped_file - sum of all children's "cache" | |
437 | total_pgpgin - sum of all children's "pgpgin" | |
438 | total_pgpgout - sum of all children's "pgpgout" | |
439 | total_swap - sum of all children's "swap" | |
440 | total_inactive_anon - sum of all children's "inactive_anon" | |
441 | total_active_anon - sum of all children's "active_anon" | |
442 | total_inactive_file - sum of all children's "inactive_file" | |
443 | total_active_file - sum of all children's "active_file" | |
444 | total_unevictable - sum of all children's "unevictable" | |
445 | ||
446 | # The following additional stats are dependent on CONFIG_DEBUG_VM. | |
c863d835 | 447 | |
c863d835 BR |
448 | recent_rotated_anon - VM internal parameter. (see mm/vmscan.c) |
449 | recent_rotated_file - VM internal parameter. (see mm/vmscan.c) | |
450 | recent_scanned_anon - VM internal parameter. (see mm/vmscan.c) | |
451 | recent_scanned_file - VM internal parameter. (see mm/vmscan.c) | |
452 | ||
453 | Memo: | |
dc10e281 KH |
454 | recent_rotated means recent frequency of LRU rotation. |
455 | recent_scanned means recent # of scans to LRU. | |
7f016ee8 KM |
456 | showing for better debug please see the code for meanings. |
457 | ||
c863d835 BR |
458 | Note: |
459 | Only anonymous and swap cache memory is listed as part of 'rss' stat. | |
460 | This should not be confused with the true 'resident set size' or the | |
dc10e281 KH |
461 | amount of physical memory used by the cgroup. |
462 | 'rss + file_mapped" will give you resident set size of cgroup. | |
463 | (Note: file and shmem may be shared among other cgroups. In that case, | |
464 | file_mapped is accounted only when the memory cgroup is owner of page | |
465 | cache.) | |
7f016ee8 | 466 | |
a7885eb8 | 467 | 5.3 swappiness |
a7885eb8 | 468 | |
dc10e281 | 469 | Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only. |
a7885eb8 | 470 | |
dc10e281 KH |
471 | Following cgroups' swappiness can't be changed. |
472 | - root cgroup (uses /proc/sys/vm/swappiness). | |
473 | - a cgroup which uses hierarchy and it has other cgroup(s) below it. | |
474 | - a cgroup which uses hierarchy and not the root of hierarchy. | |
475 | ||
476 | 5.4 failcnt | |
477 | ||
478 | A memory cgroup provides memory.failcnt and memory.memsw.failcnt files. | |
479 | This failcnt(== failure count) shows the number of times that a usage counter | |
480 | hit its limit. When a memory cgroup hits a limit, failcnt increases and | |
481 | memory under it will be reclaimed. | |
482 | ||
483 | You can reset failcnt by writing 0 to failcnt file. | |
484 | # echo 0 > .../memory.failcnt | |
a7885eb8 | 485 | |
a111c966 DN |
486 | 5.5 usage_in_bytes |
487 | ||
488 | For efficiency, as other kernel components, memory cgroup uses some optimization | |
489 | to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the | |
490 | method and doesn't show 'exact' value of memory(and swap) usage, it's an fuzz | |
491 | value for efficient access. (Of course, when necessary, it's synchronized.) | |
492 | If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP) | |
493 | value in memory.stat(see 5.2). | |
494 | ||
50c35e5b YH |
495 | 5.6 numa_stat |
496 | ||
497 | This is similar to numa_maps but operates on a per-memcg basis. This is | |
498 | useful for providing visibility into the numa locality information within | |
499 | an memcg since the pages are allowed to be allocated from any physical | |
500 | node. One of the usecases is evaluating application performance by | |
501 | combining this information with the application's cpu allocation. | |
502 | ||
503 | We export "total", "file", "anon" and "unevictable" pages per-node for | |
504 | each memcg. The ouput format of memory.numa_stat is: | |
505 | ||
506 | total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
507 | file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
508 | anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
509 | unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
510 | ||
511 | And we have total = file + anon + unevictable. | |
512 | ||
52bc0d82 | 513 | 6. Hierarchy support |
c1e862c1 | 514 | |
52bc0d82 BS |
515 | The memory controller supports a deep hierarchy and hierarchical accounting. |
516 | The hierarchy is created by creating the appropriate cgroups in the | |
517 | cgroup filesystem. Consider for example, the following cgroup filesystem | |
518 | hierarchy | |
519 | ||
67de0162 | 520 | root |
52bc0d82 | 521 | / | \ |
67de0162 JS |
522 | / | \ |
523 | a b c | |
524 | | \ | |
525 | | \ | |
526 | d e | |
52bc0d82 BS |
527 | |
528 | In the diagram above, with hierarchical accounting enabled, all memory | |
529 | usage of e, is accounted to its ancestors up until the root (i.e, c and root), | |
dc10e281 | 530 | that has memory.use_hierarchy enabled. If one of the ancestors goes over its |
52bc0d82 BS |
531 | limit, the reclaim algorithm reclaims from the tasks in the ancestor and the |
532 | children of the ancestor. | |
533 | ||
534 | 6.1 Enabling hierarchical accounting and reclaim | |
535 | ||
dc10e281 | 536 | A memory cgroup by default disables the hierarchy feature. Support |
52bc0d82 BS |
537 | can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup |
538 | ||
539 | # echo 1 > memory.use_hierarchy | |
540 | ||
541 | The feature can be disabled by | |
542 | ||
543 | # echo 0 > memory.use_hierarchy | |
544 | ||
689bca3b GT |
545 | NOTE1: Enabling/disabling will fail if either the cgroup already has other |
546 | cgroups created below it, or if the parent cgroup has use_hierarchy | |
547 | enabled. | |
52bc0d82 | 548 | |
daaf1e68 | 549 | NOTE2: When panic_on_oom is set to "2", the whole system will panic in |
dc10e281 | 550 | case of an OOM event in any cgroup. |
52bc0d82 | 551 | |
a6df6361 BS |
552 | 7. Soft limits |
553 | ||
554 | Soft limits allow for greater sharing of memory. The idea behind soft limits | |
555 | is to allow control groups to use as much of the memory as needed, provided | |
556 | ||
557 | a. There is no memory contention | |
558 | b. They do not exceed their hard limit | |
559 | ||
dc10e281 | 560 | When the system detects memory contention or low memory, control groups |
a6df6361 BS |
561 | are pushed back to their soft limits. If the soft limit of each control |
562 | group is very high, they are pushed back as much as possible to make | |
563 | sure that one control group does not starve the others of memory. | |
564 | ||
565 | Please note that soft limits is a best effort feature, it comes with | |
566 | no guarantees, but it does its best to make sure that when memory is | |
567 | heavily contended for, memory is allocated based on the soft limit | |
568 | hints/setup. Currently soft limit based reclaim is setup such that | |
569 | it gets invoked from balance_pgdat (kswapd). | |
570 | ||
571 | 7.1 Interface | |
572 | ||
573 | Soft limits can be setup by using the following commands (in this example we | |
dc10e281 | 574 | assume a soft limit of 256 MiB) |
a6df6361 BS |
575 | |
576 | # echo 256M > memory.soft_limit_in_bytes | |
577 | ||
578 | If we want to change this to 1G, we can at any time use | |
579 | ||
580 | # echo 1G > memory.soft_limit_in_bytes | |
581 | ||
582 | NOTE1: Soft limits take effect over a long period of time, since they involve | |
583 | reclaiming memory for balancing between memory cgroups | |
584 | NOTE2: It is recommended to set the soft limit always below the hard limit, | |
585 | otherwise the hard limit will take precedence. | |
586 | ||
7dc74be0 DN |
587 | 8. Move charges at task migration |
588 | ||
589 | Users can move charges associated with a task along with task migration, that | |
590 | is, uncharge task's pages from the old cgroup and charge them to the new cgroup. | |
02491447 DN |
591 | This feature is not supported in !CONFIG_MMU environments because of lack of |
592 | page tables. | |
7dc74be0 DN |
593 | |
594 | 8.1 Interface | |
595 | ||
596 | This feature is disabled by default. It can be enabled(and disabled again) by | |
597 | writing to memory.move_charge_at_immigrate of the destination cgroup. | |
598 | ||
599 | If you want to enable it: | |
600 | ||
601 | # echo (some positive value) > memory.move_charge_at_immigrate | |
602 | ||
603 | Note: Each bits of move_charge_at_immigrate has its own meaning about what type | |
604 | of charges should be moved. See 8.2 for details. | |
605 | Note: Charges are moved only when you move mm->owner, IOW, a leader of a thread | |
606 | group. | |
607 | Note: If we cannot find enough space for the task in the destination cgroup, we | |
608 | try to make space by reclaiming memory. Task migration may fail if we | |
609 | cannot make enough space. | |
dc10e281 | 610 | Note: It can take several seconds if you move charges much. |
7dc74be0 DN |
611 | |
612 | And if you want disable it again: | |
613 | ||
614 | # echo 0 > memory.move_charge_at_immigrate | |
615 | ||
616 | 8.2 Type of charges which can be move | |
617 | ||
618 | Each bits of move_charge_at_immigrate has its own meaning about what type of | |
87946a72 DN |
619 | charges should be moved. But in any cases, it must be noted that an account of |
620 | a page or a swap can be moved only when it is charged to the task's current(old) | |
621 | memory cgroup. | |
7dc74be0 DN |
622 | |
623 | bit | what type of charges would be moved ? | |
624 | -----+------------------------------------------------------------------------ | |
625 | 0 | A charge of an anonymous page(or swap of it) used by the target task. | |
626 | | Those pages and swaps must be used only by the target task. You must | |
627 | | enable Swap Extension(see 2.4) to enable move of swap charges. | |
87946a72 DN |
628 | -----+------------------------------------------------------------------------ |
629 | 1 | A charge of file pages(normal file, tmpfs file(e.g. ipc shared memory) | |
dc10e281 | 630 | | and swaps of tmpfs file) mmapped by the target task. Unlike the case of |
87946a72 DN |
631 | | anonymous pages, file pages(and swaps) in the range mmapped by the task |
632 | | will be moved even if the task hasn't done page fault, i.e. they might | |
633 | | not be the task's "RSS", but other task's "RSS" that maps the same file. | |
634 | | And mapcount of the page is ignored(the page can be moved even if | |
635 | | page_mapcount(page) > 1). You must enable Swap Extension(see 2.4) to | |
636 | | enable move of swap charges. | |
7dc74be0 DN |
637 | |
638 | 8.3 TODO | |
639 | ||
7dc74be0 DN |
640 | - Implement madvise(2) to let users decide the vma to be moved or not to be |
641 | moved. | |
642 | - All of moving charge operations are done under cgroup_mutex. It's not good | |
643 | behavior to hold the mutex too long, so we may need some trick. | |
644 | ||
2e72b634 KS |
645 | 9. Memory thresholds |
646 | ||
dc10e281 | 647 | Memory cgroup implements memory thresholds using cgroups notification |
2e72b634 KS |
648 | API (see cgroups.txt). It allows to register multiple memory and memsw |
649 | thresholds and gets notifications when it crosses. | |
650 | ||
651 | To register a threshold application need: | |
dc10e281 KH |
652 | - create an eventfd using eventfd(2); |
653 | - open memory.usage_in_bytes or memory.memsw.usage_in_bytes; | |
654 | - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to | |
655 | cgroup.event_control. | |
2e72b634 KS |
656 | |
657 | Application will be notified through eventfd when memory usage crosses | |
658 | threshold in any direction. | |
659 | ||
660 | It's applicable for root and non-root cgroup. | |
661 | ||
9490ff27 KH |
662 | 10. OOM Control |
663 | ||
3c11ecf4 KH |
664 | memory.oom_control file is for OOM notification and other controls. |
665 | ||
dc10e281 KH |
666 | Memory cgroup implements OOM notifier using cgroup notification |
667 | API (See cgroups.txt). It allows to register multiple OOM notification | |
668 | delivery and gets notification when OOM happens. | |
9490ff27 KH |
669 | |
670 | To register a notifier, application need: | |
671 | - create an eventfd using eventfd(2) | |
672 | - open memory.oom_control file | |
dc10e281 KH |
673 | - write string like "<event_fd> <fd of memory.oom_control>" to |
674 | cgroup.event_control | |
9490ff27 | 675 | |
dc10e281 | 676 | Application will be notified through eventfd when OOM happens. |
9490ff27 KH |
677 | OOM notification doesn't work for root cgroup. |
678 | ||
dc10e281 KH |
679 | You can disable OOM-killer by writing "1" to memory.oom_control file, as: |
680 | ||
3c11ecf4 KH |
681 | #echo 1 > memory.oom_control |
682 | ||
dc10e281 KH |
683 | This operation is only allowed to the top cgroup of sub-hierarchy. |
684 | If OOM-killer is disabled, tasks under cgroup will hang/sleep | |
685 | in memory cgroup's OOM-waitqueue when they request accountable memory. | |
3c11ecf4 | 686 | |
dc10e281 | 687 | For running them, you have to relax the memory cgroup's OOM status by |
3c11ecf4 KH |
688 | * enlarge limit or reduce usage. |
689 | To reduce usage, | |
690 | * kill some tasks. | |
691 | * move some tasks to other group with account migration. | |
692 | * remove some files (on tmpfs?) | |
693 | ||
694 | Then, stopped tasks will work again. | |
695 | ||
696 | At reading, current status of OOM is shown. | |
697 | oom_kill_disable 0 or 1 (if 1, oom-killer is disabled) | |
dc10e281 | 698 | under_oom 0 or 1 (if 1, the memory cgroup is under OOM, tasks may |
3c11ecf4 | 699 | be stopped.) |
9490ff27 KH |
700 | |
701 | 11. TODO | |
1b6df3aa BS |
702 | |
703 | 1. Add support for accounting huge pages (as a separate controller) | |
dfc05c25 KH |
704 | 2. Make per-cgroup scanner reclaim not-shared pages first |
705 | 3. Teach controller to account for shared-pages | |
628f4235 | 706 | 4. Start reclamation in the background when the limit is |
1b6df3aa | 707 | not yet hit but the usage is getting closer |
1b6df3aa BS |
708 | |
709 | Summary | |
710 | ||
711 | Overall, the memory controller has been a stable controller and has been | |
712 | commented and discussed quite extensively in the community. | |
713 | ||
714 | References | |
715 | ||
716 | 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/ | |
717 | 2. Singh, Balbir. Memory Controller (RSS Control), | |
718 | http://lwn.net/Articles/222762/ | |
719 | 3. Emelianov, Pavel. Resource controllers based on process cgroups | |
720 | http://lkml.org/lkml/2007/3/6/198 | |
721 | 4. Emelianov, Pavel. RSS controller based on process cgroups (v2) | |
2324c5dd | 722 | http://lkml.org/lkml/2007/4/9/78 |
1b6df3aa BS |
723 | 5. Emelianov, Pavel. RSS controller based on process cgroups (v3) |
724 | http://lkml.org/lkml/2007/5/30/244 | |
725 | 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/ | |
726 | 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control | |
727 | subsystem (v3), http://lwn.net/Articles/235534/ | |
2324c5dd | 728 | 8. Singh, Balbir. RSS controller v2 test results (lmbench), |
1b6df3aa | 729 | http://lkml.org/lkml/2007/5/17/232 |
2324c5dd | 730 | 9. Singh, Balbir. RSS controller v2 AIM9 results |
1b6df3aa | 731 | http://lkml.org/lkml/2007/5/18/1 |
2324c5dd | 732 | 10. Singh, Balbir. Memory controller v6 test results, |
1b6df3aa | 733 | http://lkml.org/lkml/2007/8/19/36 |
2324c5dd LZ |
734 | 11. Singh, Balbir. Memory controller introduction (v6), |
735 | http://lkml.org/lkml/2007/8/17/69 | |
1b6df3aa BS |
736 | 12. Corbet, Jonathan, Controlling memory use in cgroups, |
737 | http://lwn.net/Articles/243795/ |