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 | |
1939c557 | 21 | Memory-hungry applications can be isolated and limited to a smaller |
1b6df3aa | 22 | amount of memory. |
1939c557 | 23 | b. Create a cgroup with a limited amount of memory; this can be used |
1b6df3aa BS |
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. | |
1939c557 | 30 | e. There are several other use cases; find one or use the controller just |
1b6df3aa BS |
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. | |
6252efcc | 37 | - pages are linked to per-memcg LRU exclusively, and there is no global LRU. |
dc10e281 KH |
38 | - optionally, memory+swap usage can be accounted and limited. |
39 | - hierarchical accounting | |
40 | - soft limit | |
1939c557 | 41 | - moving (recharging) account at moving a task is selectable. |
dc10e281 KH |
42 | - usage threshold notifier |
43 | - oom-killer disable knob and oom-notifier | |
44 | - Root cgroup has no limit controls. | |
45 | ||
1939c557 | 46 | Kernel memory support is a work in progress, and the current version provides |
65c64ce8 | 47 | basically functionality. (See Section 2.7) |
dc10e281 KH |
48 | |
49 | Brief summary of control files. | |
50 | ||
51 | tasks # attach a task(thread) and show list of threads | |
52 | cgroup.procs # show list of processes | |
53 | cgroup.event_control # an interface for event_fd() | |
a111c966 DN |
54 | memory.usage_in_bytes # show current res_counter usage for memory |
55 | (See 5.5 for details) | |
56 | memory.memsw.usage_in_bytes # show current res_counter usage for memory+Swap | |
57 | (See 5.5 for details) | |
dc10e281 KH |
58 | memory.limit_in_bytes # set/show limit of memory usage |
59 | memory.memsw.limit_in_bytes # set/show limit of memory+Swap usage | |
60 | memory.failcnt # show the number of memory usage hits limits | |
61 | memory.memsw.failcnt # show the number of memory+Swap hits limits | |
62 | memory.max_usage_in_bytes # show max memory usage recorded | |
d66c1ce7 | 63 | memory.memsw.max_usage_in_bytes # show max memory+Swap usage recorded |
dc10e281 KH |
64 | memory.soft_limit_in_bytes # set/show soft limit of memory usage |
65 | memory.stat # show various statistics | |
66 | memory.use_hierarchy # set/show hierarchical account enabled | |
67 | memory.force_empty # trigger forced move charge to parent | |
68 | memory.swappiness # set/show swappiness parameter of vmscan | |
69 | (See sysctl's vm.swappiness) | |
70 | memory.move_charge_at_immigrate # set/show controls of moving charges | |
71 | memory.oom_control # set/show oom controls. | |
50c35e5b | 72 | memory.numa_stat # show the number of memory usage per numa node |
dc10e281 | 73 | |
d5bdae7d GC |
74 | memory.kmem.limit_in_bytes # set/show hard limit for kernel memory |
75 | memory.kmem.usage_in_bytes # show current kernel memory allocation | |
76 | memory.kmem.failcnt # show the number of kernel memory usage hits limits | |
77 | memory.kmem.max_usage_in_bytes # show max kernel memory usage recorded | |
78 | ||
3aaabe23 | 79 | memory.kmem.tcp.limit_in_bytes # set/show hard limit for tcp buf memory |
5a6dd343 | 80 | memory.kmem.tcp.usage_in_bytes # show current tcp buf memory allocation |
05a73ed2 WL |
81 | memory.kmem.tcp.failcnt # show the number of tcp buf memory usage hits limits |
82 | memory.kmem.tcp.max_usage_in_bytes # show max tcp buf memory usage recorded | |
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 | ||
348b4655 JL |
152 | The accounting is done as follows: mem_cgroup_charge_common() is invoked to |
153 | set up the necessary data structures and check if the cgroup that is being | |
154 | charged is over its limit. If it is, then reclaim is invoked on the cgroup. | |
1b6df3aa BS |
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. |
6252efcc | 163 | Some pages which are never reclaimable and will not be on the LRU |
dc10e281 | 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 | ||
1939c557 | 171 | An 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 | |
1939c557 | 174 | are really freed. Such SwapCaches are also accounted. |
dc10e281 KH |
175 | A swapped-in page is not accounted until it's mapped. |
176 | ||
1939c557 | 177 | Note: The kernel does swapin-readahead and reads multiple swaps at once. |
dc10e281 KH |
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 | ||
4b91355e KH |
194 | But see section 8.2: when moving a task to another cgroup, its pages may |
195 | be recharged to the new cgroup, if move_charge_at_immigrate has been chosen. | |
196 | ||
c255a458 | 197 | Exception: If CONFIG_CGROUP_CGROUP_MEMCG_SWAP is not used. |
8c7c6e34 | 198 | When you do swapoff and make swapped-out pages of shmem(tmpfs) to |
d13d1443 KH |
199 | be backed into memory in force, charges for pages are accounted against the |
200 | caller of swapoff rather than the users of shmem. | |
201 | ||
c255a458 | 202 | 2.4 Swap Extension (CONFIG_MEMCG_SWAP) |
dc10e281 | 203 | |
8c7c6e34 KH |
204 | Swap Extension allows you to record charge for swap. A swapped-in page is |
205 | charged back to original page allocator if possible. | |
206 | ||
207 | When swap is accounted, following files are added. | |
208 | - memory.memsw.usage_in_bytes. | |
209 | - memory.memsw.limit_in_bytes. | |
210 | ||
dc10e281 KH |
211 | memsw means memory+swap. Usage of memory+swap is limited by |
212 | memsw.limit_in_bytes. | |
213 | ||
214 | Example: Assume a system with 4G of swap. A task which allocates 6G of memory | |
215 | (by mistake) under 2G memory limitation will use all swap. | |
216 | In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap. | |
1939c557 | 217 | By using the memsw limit, you can avoid system OOM which can be caused by swap |
dc10e281 | 218 | shortage. |
8c7c6e34 | 219 | |
dc10e281 | 220 | * why 'memory+swap' rather than swap. |
8c7c6e34 KH |
221 | The global LRU(kswapd) can swap out arbitrary pages. Swap-out means |
222 | to move account from memory to swap...there is no change in usage of | |
dc10e281 KH |
223 | memory+swap. In other words, when we want to limit the usage of swap without |
224 | affecting global LRU, memory+swap limit is better than just limiting swap from | |
1939c557 | 225 | an OS point of view. |
22a668d7 KH |
226 | |
227 | * What happens when a cgroup hits memory.memsw.limit_in_bytes | |
67de0162 | 228 | When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out |
22a668d7 KH |
229 | in this cgroup. Then, swap-out will not be done by cgroup routine and file |
230 | caches are dropped. But as mentioned above, global LRU can do swapout memory | |
231 | from it for sanity of the system's memory management state. You can't forbid | |
232 | it by cgroup. | |
8c7c6e34 KH |
233 | |
234 | 2.5 Reclaim | |
1b6df3aa | 235 | |
dc10e281 KH |
236 | Each cgroup maintains a per cgroup LRU which has the same structure as |
237 | global VM. When a cgroup goes over its limit, we first try | |
1b6df3aa BS |
238 | to reclaim memory from the cgroup so as to make space for the new |
239 | pages that the cgroup has touched. If the reclaim is unsuccessful, | |
240 | an OOM routine is invoked to select and kill the bulkiest task in the | |
dc10e281 | 241 | cgroup. (See 10. OOM Control below.) |
1b6df3aa BS |
242 | |
243 | The reclaim algorithm has not been modified for cgroups, except that | |
1939c557 | 244 | pages that are selected for reclaiming come from the per-cgroup LRU |
1b6df3aa BS |
245 | list. |
246 | ||
4b3bde4c BS |
247 | NOTE: Reclaim does not work for the root cgroup, since we cannot set any |
248 | limits on the root cgroup. | |
249 | ||
daaf1e68 KH |
250 | Note2: When panic_on_oom is set to "2", the whole system will panic. |
251 | ||
9490ff27 KH |
252 | When oom event notifier is registered, event will be delivered. |
253 | (See oom_control section) | |
254 | ||
dc10e281 | 255 | 2.6 Locking |
1b6df3aa | 256 | |
dc10e281 KH |
257 | lock_page_cgroup()/unlock_page_cgroup() should not be called under |
258 | mapping->tree_lock. | |
1b6df3aa | 259 | |
dc10e281 KH |
260 | Other lock order is following: |
261 | PG_locked. | |
262 | mm->page_table_lock | |
263 | zone->lru_lock | |
264 | lock_page_cgroup. | |
265 | In many cases, just lock_page_cgroup() is called. | |
266 | per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by | |
267 | zone->lru_lock, it has no lock of its own. | |
1b6df3aa | 268 | |
c255a458 | 269 | 2.7 Kernel Memory Extension (CONFIG_MEMCG_KMEM) |
e5671dfa GC |
270 | |
271 | With the Kernel memory extension, the Memory Controller is able to limit | |
272 | the amount of kernel memory used by the system. Kernel memory is fundamentally | |
273 | different than user memory, since it can't be swapped out, which makes it | |
274 | possible to DoS the system by consuming too much of this precious resource. | |
275 | ||
d5bdae7d GC |
276 | Kernel memory won't be accounted at all until limit on a group is set. This |
277 | allows for existing setups to continue working without disruption. The limit | |
278 | cannot be set if the cgroup have children, or if there are already tasks in the | |
279 | cgroup. Attempting to set the limit under those conditions will return -EBUSY. | |
280 | When use_hierarchy == 1 and a group is accounted, its children will | |
281 | automatically be accounted regardless of their limit value. | |
282 | ||
283 | After a group is first limited, it will be kept being accounted until it | |
284 | is removed. The memory limitation itself, can of course be removed by writing | |
285 | -1 to memory.kmem.limit_in_bytes. In this case, kmem will be accounted, but not | |
286 | limited. | |
287 | ||
e5671dfa | 288 | Kernel memory limits are not imposed for the root cgroup. Usage for the root |
d5bdae7d GC |
289 | cgroup may or may not be accounted. The memory used is accumulated into |
290 | memory.kmem.usage_in_bytes, or in a separate counter when it makes sense. | |
291 | (currently only for tcp). | |
292 | The main "kmem" counter is fed into the main counter, so kmem charges will | |
293 | also be visible from the user counter. | |
e5671dfa | 294 | |
e5671dfa GC |
295 | Currently no soft limit is implemented for kernel memory. It is future work |
296 | to trigger slab reclaim when those limits are reached. | |
297 | ||
298 | 2.7.1 Current Kernel Memory resources accounted | |
299 | ||
d5bdae7d GC |
300 | * stack pages: every process consumes some stack pages. By accounting into |
301 | kernel memory, we prevent new processes from being created when the kernel | |
302 | memory usage is too high. | |
303 | ||
92e79349 GC |
304 | * slab pages: pages allocated by the SLAB or SLUB allocator are tracked. A copy |
305 | of each kmem_cache is created everytime the cache is touched by the first time | |
306 | from inside the memcg. The creation is done lazily, so some objects can still be | |
307 | skipped while the cache is being created. All objects in a slab page should | |
308 | belong to the same memcg. This only fails to hold when a task is migrated to a | |
309 | different memcg during the page allocation by the cache. | |
310 | ||
e1aab161 GC |
311 | * sockets memory pressure: some sockets protocols have memory pressure |
312 | thresholds. The Memory Controller allows them to be controlled individually | |
313 | per cgroup, instead of globally. | |
e5671dfa | 314 | |
d1a4c0b3 GC |
315 | * tcp memory pressure: sockets memory pressure for the tcp protocol. |
316 | ||
d5bdae7d GC |
317 | 2.7.3 Common use cases |
318 | ||
319 | Because the "kmem" counter is fed to the main user counter, kernel memory can | |
320 | never be limited completely independently of user memory. Say "U" is the user | |
321 | limit, and "K" the kernel limit. There are three possible ways limits can be | |
322 | set: | |
323 | ||
324 | U != 0, K = unlimited: | |
325 | This is the standard memcg limitation mechanism already present before kmem | |
326 | accounting. Kernel memory is completely ignored. | |
327 | ||
328 | U != 0, K < U: | |
329 | Kernel memory is a subset of the user memory. This setup is useful in | |
330 | deployments where the total amount of memory per-cgroup is overcommited. | |
331 | Overcommiting kernel memory limits is definitely not recommended, since the | |
332 | box can still run out of non-reclaimable memory. | |
333 | In this case, the admin could set up K so that the sum of all groups is | |
334 | never greater than the total memory, and freely set U at the cost of his | |
335 | QoS. | |
336 | ||
337 | U != 0, K >= U: | |
338 | Since kmem charges will also be fed to the user counter and reclaim will be | |
339 | triggered for the cgroup for both kinds of memory. This setup gives the | |
340 | admin a unified view of memory, and it is also useful for people who just | |
341 | want to track kernel memory usage. | |
342 | ||
1b6df3aa BS |
343 | 3. User Interface |
344 | ||
345 | 0. Configuration | |
346 | ||
347 | a. Enable CONFIG_CGROUPS | |
348 | b. Enable CONFIG_RESOURCE_COUNTERS | |
c255a458 AM |
349 | c. Enable CONFIG_MEMCG |
350 | d. Enable CONFIG_MEMCG_SWAP (to use swap extension) | |
d5bdae7d | 351 | d. Enable CONFIG_MEMCG_KMEM (to use kmem extension) |
1b6df3aa | 352 | |
f6e07d38 JS |
353 | 1. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?) |
354 | # mount -t tmpfs none /sys/fs/cgroup | |
355 | # mkdir /sys/fs/cgroup/memory | |
356 | # mount -t cgroup none /sys/fs/cgroup/memory -o memory | |
1b6df3aa BS |
357 | |
358 | 2. Make the new group and move bash into it | |
f6e07d38 JS |
359 | # mkdir /sys/fs/cgroup/memory/0 |
360 | # echo $$ > /sys/fs/cgroup/memory/0/tasks | |
1b6df3aa | 361 | |
dc10e281 | 362 | Since now we're in the 0 cgroup, we can alter the memory limit: |
f6e07d38 | 363 | # echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes |
0eea1030 BS |
364 | |
365 | NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo, | |
dc10e281 KH |
366 | mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.) |
367 | ||
c5b947b2 | 368 | NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited). |
4b3bde4c | 369 | NOTE: We cannot set limits on the root cgroup any more. |
0eea1030 | 370 | |
f6e07d38 | 371 | # cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes |
2324c5dd | 372 | 4194304 |
0eea1030 | 373 | |
1b6df3aa | 374 | We can check the usage: |
f6e07d38 | 375 | # cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes |
2324c5dd | 376 | 1216512 |
0eea1030 | 377 | |
1939c557 | 378 | A successful write to this file does not guarantee a successful setting of |
dc10e281 | 379 | this limit to the value written into the file. This can be due to a |
0eea1030 | 380 | number of factors, such as rounding up to page boundaries or the total |
dc10e281 | 381 | availability of memory on the system. The user is required to re-read |
0eea1030 BS |
382 | this file after a write to guarantee the value committed by the kernel. |
383 | ||
fb78922c | 384 | # echo 1 > memory.limit_in_bytes |
0eea1030 | 385 | # cat memory.limit_in_bytes |
2324c5dd | 386 | 4096 |
1b6df3aa BS |
387 | |
388 | The memory.failcnt field gives the number of times that the cgroup limit was | |
389 | exceeded. | |
390 | ||
dfc05c25 KH |
391 | The memory.stat file gives accounting information. Now, the number of |
392 | caches, RSS and Active pages/Inactive pages are shown. | |
393 | ||
1b6df3aa BS |
394 | 4. Testing |
395 | ||
dc10e281 KH |
396 | For testing features and implementation, see memcg_test.txt. |
397 | ||
398 | Performance test is also important. To see pure memory controller's overhead, | |
399 | testing on tmpfs will give you good numbers of small overheads. | |
400 | Example: do kernel make on tmpfs. | |
401 | ||
402 | Page-fault scalability is also important. At measuring parallel | |
403 | page fault test, multi-process test may be better than multi-thread | |
404 | test because it has noise of shared objects/status. | |
405 | ||
406 | But the above two are testing extreme situations. | |
407 | Trying usual test under memory controller is always helpful. | |
1b6df3aa BS |
408 | |
409 | 4.1 Troubleshooting | |
410 | ||
411 | Sometimes a user might find that the application under a cgroup is | |
1939c557 | 412 | terminated by the OOM killer. There are several causes for this: |
1b6df3aa BS |
413 | |
414 | 1. The cgroup limit is too low (just too low to do anything useful) | |
415 | 2. The user is using anonymous memory and swap is turned off or too low | |
416 | ||
417 | A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of | |
418 | some of the pages cached in the cgroup (page cache pages). | |
419 | ||
1939c557 | 420 | To know what happens, disabling OOM_Kill as per "10. OOM Control" (below) and |
dc10e281 KH |
421 | seeing what happens will be helpful. |
422 | ||
1b6df3aa BS |
423 | 4.2 Task migration |
424 | ||
a33f3224 | 425 | When a task migrates from one cgroup to another, its charge is not |
7dc74be0 | 426 | carried forward by default. The pages allocated from the original cgroup still |
1b6df3aa BS |
427 | remain charged to it, the charge is dropped when the page is freed or |
428 | reclaimed. | |
429 | ||
dc10e281 KH |
430 | You can move charges of a task along with task migration. |
431 | See 8. "Move charges at task migration" | |
7dc74be0 | 432 | |
1b6df3aa BS |
433 | 4.3 Removing a cgroup |
434 | ||
435 | A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a | |
436 | cgroup might have some charge associated with it, even though all | |
dc10e281 KH |
437 | tasks have migrated away from it. (because we charge against pages, not |
438 | against tasks.) | |
439 | ||
cc926f78 KH |
440 | We move the stats to root (if use_hierarchy==0) or parent (if |
441 | use_hierarchy==1), and no change on the charge except uncharging | |
442 | from the child. | |
1b6df3aa | 443 | |
8c7c6e34 KH |
444 | Charges recorded in swap information is not updated at removal of cgroup. |
445 | Recorded information is discarded and a cgroup which uses swap (swapcache) | |
446 | will be charged as a new owner of it. | |
447 | ||
cc926f78 | 448 | About use_hierarchy, see Section 6. |
8c7c6e34 | 449 | |
c1e862c1 KH |
450 | 5. Misc. interfaces. |
451 | ||
452 | 5.1 force_empty | |
453 | memory.force_empty interface is provided to make cgroup's memory usage empty. | |
454 | You can use this interface only when the cgroup has no tasks. | |
455 | When writing anything to this | |
456 | ||
457 | # echo 0 > memory.force_empty | |
458 | ||
dc10e281 KH |
459 | Almost all pages tracked by this memory cgroup will be unmapped and freed. |
460 | Some pages cannot be freed because they are locked or in-use. Such pages are | |
1939c557 | 461 | moved to parent (if use_hierarchy==1) or root (if use_hierarchy==0) and this |
cc926f78 | 462 | cgroup will be empty. |
c1e862c1 | 463 | |
1939c557 | 464 | The typical use case for this interface is before calling rmdir(). |
c1e862c1 KH |
465 | Because rmdir() moves all pages to parent, some out-of-use page caches can be |
466 | moved to the parent. If you want to avoid that, force_empty will be useful. | |
467 | ||
d5bdae7d GC |
468 | Also, note that when memory.kmem.limit_in_bytes is set the charges due to |
469 | kernel pages will still be seen. This is not considered a failure and the | |
470 | write will still return success. In this case, it is expected that | |
471 | memory.kmem.usage_in_bytes == memory.usage_in_bytes. | |
472 | ||
cc926f78 KH |
473 | About use_hierarchy, see Section 6. |
474 | ||
7f016ee8 | 475 | 5.2 stat file |
c863d835 | 476 | |
185efc0f | 477 | memory.stat file includes following statistics |
c863d835 | 478 | |
dc10e281 | 479 | # per-memory cgroup local status |
c863d835 BR |
480 | cache - # of bytes of page cache memory. |
481 | rss - # of bytes of anonymous and swap cache memory. | |
dc10e281 | 482 | mapped_file - # of bytes of mapped file (includes tmpfs/shmem) |
0527b690 YH |
483 | pgpgin - # of charging events to the memory cgroup. The charging |
484 | event happens each time a page is accounted as either mapped | |
485 | anon page(RSS) or cache page(Page Cache) to the cgroup. | |
486 | pgpgout - # of uncharging events to the memory cgroup. The uncharging | |
487 | event happens each time a page is unaccounted from the cgroup. | |
dc10e281 | 488 | swap - # of bytes of swap usage |
c863d835 | 489 | inactive_anon - # of bytes of anonymous memory and swap cache memory on |
dc10e281 KH |
490 | LRU list. |
491 | active_anon - # of bytes of anonymous and swap cache memory on active | |
492 | inactive LRU list. | |
493 | inactive_file - # of bytes of file-backed memory on inactive LRU list. | |
494 | active_file - # of bytes of file-backed memory on active LRU list. | |
c863d835 BR |
495 | unevictable - # of bytes of memory that cannot be reclaimed (mlocked etc). |
496 | ||
dc10e281 KH |
497 | # status considering hierarchy (see memory.use_hierarchy settings) |
498 | ||
499 | hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy | |
500 | under which the memory cgroup is | |
501 | hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to | |
502 | hierarchy under which memory cgroup is. | |
503 | ||
eb6332a5 JW |
504 | total_<counter> - # hierarchical version of <counter>, which in |
505 | addition to the cgroup's own value includes the | |
506 | sum of all hierarchical children's values of | |
507 | <counter>, i.e. total_cache | |
dc10e281 KH |
508 | |
509 | # The following additional stats are dependent on CONFIG_DEBUG_VM. | |
c863d835 | 510 | |
c863d835 BR |
511 | recent_rotated_anon - VM internal parameter. (see mm/vmscan.c) |
512 | recent_rotated_file - VM internal parameter. (see mm/vmscan.c) | |
513 | recent_scanned_anon - VM internal parameter. (see mm/vmscan.c) | |
514 | recent_scanned_file - VM internal parameter. (see mm/vmscan.c) | |
515 | ||
516 | Memo: | |
dc10e281 KH |
517 | recent_rotated means recent frequency of LRU rotation. |
518 | recent_scanned means recent # of scans to LRU. | |
7f016ee8 KM |
519 | showing for better debug please see the code for meanings. |
520 | ||
c863d835 BR |
521 | Note: |
522 | Only anonymous and swap cache memory is listed as part of 'rss' stat. | |
523 | This should not be confused with the true 'resident set size' or the | |
dc10e281 KH |
524 | amount of physical memory used by the cgroup. |
525 | 'rss + file_mapped" will give you resident set size of cgroup. | |
526 | (Note: file and shmem may be shared among other cgroups. In that case, | |
527 | file_mapped is accounted only when the memory cgroup is owner of page | |
528 | cache.) | |
7f016ee8 | 529 | |
a7885eb8 | 530 | 5.3 swappiness |
a7885eb8 | 531 | |
dc10e281 | 532 | Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only. |
9a5a8f19 MH |
533 | Please note that unlike the global swappiness, memcg knob set to 0 |
534 | really prevents from any swapping even if there is a swap storage | |
535 | available. This might lead to memcg OOM killer if there are no file | |
536 | pages to reclaim. | |
a7885eb8 | 537 | |
dc10e281 KH |
538 | Following cgroups' swappiness can't be changed. |
539 | - root cgroup (uses /proc/sys/vm/swappiness). | |
540 | - a cgroup which uses hierarchy and it has other cgroup(s) below it. | |
541 | - a cgroup which uses hierarchy and not the root of hierarchy. | |
542 | ||
543 | 5.4 failcnt | |
544 | ||
545 | A memory cgroup provides memory.failcnt and memory.memsw.failcnt files. | |
546 | This failcnt(== failure count) shows the number of times that a usage counter | |
547 | hit its limit. When a memory cgroup hits a limit, failcnt increases and | |
548 | memory under it will be reclaimed. | |
549 | ||
550 | You can reset failcnt by writing 0 to failcnt file. | |
551 | # echo 0 > .../memory.failcnt | |
a7885eb8 | 552 | |
a111c966 DN |
553 | 5.5 usage_in_bytes |
554 | ||
555 | For efficiency, as other kernel components, memory cgroup uses some optimization | |
556 | to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the | |
1939c557 | 557 | method and doesn't show 'exact' value of memory (and swap) usage, it's a fuzz |
a111c966 DN |
558 | value for efficient access. (Of course, when necessary, it's synchronized.) |
559 | If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP) | |
560 | value in memory.stat(see 5.2). | |
561 | ||
50c35e5b YH |
562 | 5.6 numa_stat |
563 | ||
564 | This is similar to numa_maps but operates on a per-memcg basis. This is | |
565 | useful for providing visibility into the numa locality information within | |
566 | an memcg since the pages are allowed to be allocated from any physical | |
1939c557 MK |
567 | node. One of the use cases is evaluating application performance by |
568 | combining this information with the application's CPU allocation. | |
50c35e5b YH |
569 | |
570 | We export "total", "file", "anon" and "unevictable" pages per-node for | |
571 | each memcg. The ouput format of memory.numa_stat is: | |
572 | ||
573 | total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
574 | file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
575 | anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
576 | unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... | |
577 | ||
578 | And we have total = file + anon + unevictable. | |
579 | ||
52bc0d82 | 580 | 6. Hierarchy support |
c1e862c1 | 581 | |
52bc0d82 BS |
582 | The memory controller supports a deep hierarchy and hierarchical accounting. |
583 | The hierarchy is created by creating the appropriate cgroups in the | |
584 | cgroup filesystem. Consider for example, the following cgroup filesystem | |
585 | hierarchy | |
586 | ||
67de0162 | 587 | root |
52bc0d82 | 588 | / | \ |
67de0162 JS |
589 | / | \ |
590 | a b c | |
591 | | \ | |
592 | | \ | |
593 | d e | |
52bc0d82 BS |
594 | |
595 | In the diagram above, with hierarchical accounting enabled, all memory | |
596 | usage of e, is accounted to its ancestors up until the root (i.e, c and root), | |
dc10e281 | 597 | that has memory.use_hierarchy enabled. If one of the ancestors goes over its |
52bc0d82 BS |
598 | limit, the reclaim algorithm reclaims from the tasks in the ancestor and the |
599 | children of the ancestor. | |
600 | ||
601 | 6.1 Enabling hierarchical accounting and reclaim | |
602 | ||
dc10e281 | 603 | A memory cgroup by default disables the hierarchy feature. Support |
52bc0d82 BS |
604 | can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup |
605 | ||
606 | # echo 1 > memory.use_hierarchy | |
607 | ||
608 | The feature can be disabled by | |
609 | ||
610 | # echo 0 > memory.use_hierarchy | |
611 | ||
689bca3b GT |
612 | NOTE1: Enabling/disabling will fail if either the cgroup already has other |
613 | cgroups created below it, or if the parent cgroup has use_hierarchy | |
614 | enabled. | |
52bc0d82 | 615 | |
daaf1e68 | 616 | NOTE2: When panic_on_oom is set to "2", the whole system will panic in |
dc10e281 | 617 | case of an OOM event in any cgroup. |
52bc0d82 | 618 | |
a6df6361 BS |
619 | 7. Soft limits |
620 | ||
621 | Soft limits allow for greater sharing of memory. The idea behind soft limits | |
622 | is to allow control groups to use as much of the memory as needed, provided | |
623 | ||
624 | a. There is no memory contention | |
625 | b. They do not exceed their hard limit | |
626 | ||
dc10e281 | 627 | When the system detects memory contention or low memory, control groups |
a6df6361 BS |
628 | are pushed back to their soft limits. If the soft limit of each control |
629 | group is very high, they are pushed back as much as possible to make | |
630 | sure that one control group does not starve the others of memory. | |
631 | ||
1939c557 | 632 | Please note that soft limits is a best-effort feature; it comes with |
a6df6361 BS |
633 | no guarantees, but it does its best to make sure that when memory is |
634 | heavily contended for, memory is allocated based on the soft limit | |
1939c557 | 635 | hints/setup. Currently soft limit based reclaim is set up such that |
a6df6361 BS |
636 | it gets invoked from balance_pgdat (kswapd). |
637 | ||
638 | 7.1 Interface | |
639 | ||
640 | Soft limits can be setup by using the following commands (in this example we | |
dc10e281 | 641 | assume a soft limit of 256 MiB) |
a6df6361 BS |
642 | |
643 | # echo 256M > memory.soft_limit_in_bytes | |
644 | ||
645 | If we want to change this to 1G, we can at any time use | |
646 | ||
647 | # echo 1G > memory.soft_limit_in_bytes | |
648 | ||
649 | NOTE1: Soft limits take effect over a long period of time, since they involve | |
650 | reclaiming memory for balancing between memory cgroups | |
651 | NOTE2: It is recommended to set the soft limit always below the hard limit, | |
652 | otherwise the hard limit will take precedence. | |
653 | ||
7dc74be0 DN |
654 | 8. Move charges at task migration |
655 | ||
656 | Users can move charges associated with a task along with task migration, that | |
657 | is, uncharge task's pages from the old cgroup and charge them to the new cgroup. | |
02491447 DN |
658 | This feature is not supported in !CONFIG_MMU environments because of lack of |
659 | page tables. | |
7dc74be0 DN |
660 | |
661 | 8.1 Interface | |
662 | ||
1939c557 | 663 | This feature is disabled by default. It can be enabledi (and disabled again) by |
7dc74be0 DN |
664 | writing to memory.move_charge_at_immigrate of the destination cgroup. |
665 | ||
666 | If you want to enable it: | |
667 | ||
668 | # echo (some positive value) > memory.move_charge_at_immigrate | |
669 | ||
670 | Note: Each bits of move_charge_at_immigrate has its own meaning about what type | |
671 | of charges should be moved. See 8.2 for details. | |
1939c557 MK |
672 | Note: Charges are moved only when you move mm->owner, in other words, |
673 | a leader of a thread group. | |
7dc74be0 DN |
674 | Note: If we cannot find enough space for the task in the destination cgroup, we |
675 | try to make space by reclaiming memory. Task migration may fail if we | |
676 | cannot make enough space. | |
dc10e281 | 677 | Note: It can take several seconds if you move charges much. |
7dc74be0 DN |
678 | |
679 | And if you want disable it again: | |
680 | ||
681 | # echo 0 > memory.move_charge_at_immigrate | |
682 | ||
1939c557 | 683 | 8.2 Type of charges which can be moved |
7dc74be0 | 684 | |
1939c557 MK |
685 | Each bit in move_charge_at_immigrate has its own meaning about what type of |
686 | charges should be moved. But in any case, it must be noted that an account of | |
687 | a page or a swap can be moved only when it is charged to the task's current | |
688 | (old) memory cgroup. | |
7dc74be0 DN |
689 | |
690 | bit | what type of charges would be moved ? | |
691 | -----+------------------------------------------------------------------------ | |
1939c557 MK |
692 | 0 | A charge of an anonymous page (or swap of it) used by the target task. |
693 | | You must enable Swap Extension (see 2.4) to enable move of swap charges. | |
87946a72 | 694 | -----+------------------------------------------------------------------------ |
1939c557 | 695 | 1 | A charge of file pages (normal file, tmpfs file (e.g. ipc shared memory) |
dc10e281 | 696 | | and swaps of tmpfs file) mmapped by the target task. Unlike the case of |
1939c557 | 697 | | anonymous pages, file pages (and swaps) in the range mmapped by the task |
87946a72 DN |
698 | | will be moved even if the task hasn't done page fault, i.e. they might |
699 | | not be the task's "RSS", but other task's "RSS" that maps the same file. | |
1939c557 MK |
700 | | And mapcount of the page is ignored (the page can be moved even if |
701 | | page_mapcount(page) > 1). You must enable Swap Extension (see 2.4) to | |
87946a72 | 702 | | enable move of swap charges. |
7dc74be0 DN |
703 | |
704 | 8.3 TODO | |
705 | ||
7dc74be0 DN |
706 | - All of moving charge operations are done under cgroup_mutex. It's not good |
707 | behavior to hold the mutex too long, so we may need some trick. | |
708 | ||
2e72b634 KS |
709 | 9. Memory thresholds |
710 | ||
1939c557 | 711 | Memory cgroup implements memory thresholds using the cgroups notification |
2e72b634 KS |
712 | API (see cgroups.txt). It allows to register multiple memory and memsw |
713 | thresholds and gets notifications when it crosses. | |
714 | ||
1939c557 | 715 | To register a threshold, an application must: |
dc10e281 KH |
716 | - create an eventfd using eventfd(2); |
717 | - open memory.usage_in_bytes or memory.memsw.usage_in_bytes; | |
718 | - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to | |
719 | cgroup.event_control. | |
2e72b634 KS |
720 | |
721 | Application will be notified through eventfd when memory usage crosses | |
722 | threshold in any direction. | |
723 | ||
724 | It's applicable for root and non-root cgroup. | |
725 | ||
9490ff27 KH |
726 | 10. OOM Control |
727 | ||
3c11ecf4 KH |
728 | memory.oom_control file is for OOM notification and other controls. |
729 | ||
1939c557 | 730 | Memory cgroup implements OOM notifier using the cgroup notification |
dc10e281 KH |
731 | API (See cgroups.txt). It allows to register multiple OOM notification |
732 | delivery and gets notification when OOM happens. | |
9490ff27 | 733 | |
1939c557 | 734 | To register a notifier, an application must: |
9490ff27 KH |
735 | - create an eventfd using eventfd(2) |
736 | - open memory.oom_control file | |
dc10e281 KH |
737 | - write string like "<event_fd> <fd of memory.oom_control>" to |
738 | cgroup.event_control | |
9490ff27 | 739 | |
1939c557 MK |
740 | The application will be notified through eventfd when OOM happens. |
741 | OOM notification doesn't work for the root cgroup. | |
9490ff27 | 742 | |
1939c557 | 743 | You can disable the OOM-killer by writing "1" to memory.oom_control file, as: |
dc10e281 | 744 | |
3c11ecf4 KH |
745 | #echo 1 > memory.oom_control |
746 | ||
1939c557 | 747 | This operation is only allowed to the top cgroup of a sub-hierarchy. |
dc10e281 KH |
748 | If OOM-killer is disabled, tasks under cgroup will hang/sleep |
749 | in memory cgroup's OOM-waitqueue when they request accountable memory. | |
3c11ecf4 | 750 | |
dc10e281 | 751 | For running them, you have to relax the memory cgroup's OOM status by |
3c11ecf4 KH |
752 | * enlarge limit or reduce usage. |
753 | To reduce usage, | |
754 | * kill some tasks. | |
755 | * move some tasks to other group with account migration. | |
756 | * remove some files (on tmpfs?) | |
757 | ||
758 | Then, stopped tasks will work again. | |
759 | ||
760 | At reading, current status of OOM is shown. | |
761 | oom_kill_disable 0 or 1 (if 1, oom-killer is disabled) | |
dc10e281 | 762 | under_oom 0 or 1 (if 1, the memory cgroup is under OOM, tasks may |
3c11ecf4 | 763 | be stopped.) |
9490ff27 KH |
764 | |
765 | 11. TODO | |
1b6df3aa BS |
766 | |
767 | 1. Add support for accounting huge pages (as a separate controller) | |
dfc05c25 KH |
768 | 2. Make per-cgroup scanner reclaim not-shared pages first |
769 | 3. Teach controller to account for shared-pages | |
628f4235 | 770 | 4. Start reclamation in the background when the limit is |
1b6df3aa | 771 | not yet hit but the usage is getting closer |
1b6df3aa BS |
772 | |
773 | Summary | |
774 | ||
775 | Overall, the memory controller has been a stable controller and has been | |
776 | commented and discussed quite extensively in the community. | |
777 | ||
778 | References | |
779 | ||
780 | 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/ | |
781 | 2. Singh, Balbir. Memory Controller (RSS Control), | |
782 | http://lwn.net/Articles/222762/ | |
783 | 3. Emelianov, Pavel. Resource controllers based on process cgroups | |
784 | http://lkml.org/lkml/2007/3/6/198 | |
785 | 4. Emelianov, Pavel. RSS controller based on process cgroups (v2) | |
2324c5dd | 786 | http://lkml.org/lkml/2007/4/9/78 |
1b6df3aa BS |
787 | 5. Emelianov, Pavel. RSS controller based on process cgroups (v3) |
788 | http://lkml.org/lkml/2007/5/30/244 | |
789 | 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/ | |
790 | 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control | |
791 | subsystem (v3), http://lwn.net/Articles/235534/ | |
2324c5dd | 792 | 8. Singh, Balbir. RSS controller v2 test results (lmbench), |
1b6df3aa | 793 | http://lkml.org/lkml/2007/5/17/232 |
2324c5dd | 794 | 9. Singh, Balbir. RSS controller v2 AIM9 results |
1b6df3aa | 795 | http://lkml.org/lkml/2007/5/18/1 |
2324c5dd | 796 | 10. Singh, Balbir. Memory controller v6 test results, |
1b6df3aa | 797 | http://lkml.org/lkml/2007/8/19/36 |
2324c5dd LZ |
798 | 11. Singh, Balbir. Memory controller introduction (v6), |
799 | http://lkml.org/lkml/2007/8/17/69 | |
1b6df3aa BS |
800 | 12. Corbet, Jonathan, Controlling memory use in cgroups, |
801 | http://lwn.net/Articles/243795/ |