1===================================================== 2Memory Resource Controller(Memcg) Implementation Memo 3===================================================== 4 5Last Updated: 2010/2 6 7Base Kernel Version: based on 2.6.33-rc7-mm(candidate for 34). 8 9Because VM is getting complex (one of reasons is memcg...), memcg's behavior 10is complex. This is a document for memcg's internal behavior. 11Please note that implementation details can be changed. 12 13(*) Topics on API should be in Documentation/admin-guide/cgroup-v1/memory.rst) 14 150. How to record usage ? 16======================== 17 18 2 objects are used. 19 20 page_cgroup ....an object per page. 21 22 Allocated at boot or memory hotplug. Freed at memory hot removal. 23 24 swap_cgroup ... an entry per swp_entry. 25 26 Allocated at swapon(). Freed at swapoff(). 27 28 The page_cgroup has USED bit and double count against a page_cgroup never 29 occurs. swap_cgroup is used only when a charged page is swapped-out. 30 311. Charge 32========= 33 34 a page/swp_entry may be charged (usage += PAGE_SIZE) at 35 36 mem_cgroup_try_charge() 37 382. Uncharge 39=========== 40 41 a page/swp_entry may be uncharged (usage -= PAGE_SIZE) by 42 43 mem_cgroup_uncharge() 44 Called when a page's refcount goes down to 0. 45 46 mem_cgroup_uncharge_swap() 47 Called when swp_entry's refcnt goes down to 0. A charge against swap 48 disappears. 49 503. charge-commit-cancel 51======================= 52 53 Memcg pages are charged in two steps: 54 55 - mem_cgroup_try_charge() 56 - mem_cgroup_commit_charge() or mem_cgroup_cancel_charge() 57 58 At try_charge(), there are no flags to say "this page is charged". 59 at this point, usage += PAGE_SIZE. 60 61 At commit(), the page is associated with the memcg. 62 63 At cancel(), simply usage -= PAGE_SIZE. 64 65Under below explanation, we assume CONFIG_SWAP=y. 66 674. Anonymous 68============ 69 70 Anonymous page is newly allocated at 71 - page fault into MAP_ANONYMOUS mapping. 72 - Copy-On-Write. 73 74 4.1 Swap-in. 75 At swap-in, the page is taken from swap-cache. There are 2 cases. 76 77 (a) If the SwapCache is newly allocated and read, it has no charges. 78 (b) If the SwapCache has been mapped by processes, it has been 79 charged already. 80 81 4.2 Swap-out. 82 At swap-out, typical state transition is below. 83 84 (a) add to swap cache. (marked as SwapCache) 85 swp_entry's refcnt += 1. 86 (b) fully unmapped. 87 swp_entry's refcnt += # of ptes. 88 (c) write back to swap. 89 (d) delete from swap cache. (remove from SwapCache) 90 swp_entry's refcnt -= 1. 91 92 93 Finally, at task exit, 94 (e) zap_pte() is called and swp_entry's refcnt -=1 -> 0. 95 965. Page Cache 97============= 98 99 Page Cache is charged at 100 - filemap_add_folio(). 101 102 The logic is very clear. (About migration, see below) 103 104 Note: 105 __filemap_remove_folio() is called by filemap_remove_folio() 106 and __remove_mapping(). 107 1086. Shmem(tmpfs) Page Cache 109=========================== 110 111 The best way to understand shmem's page state transition is to read 112 mm/shmem.c. 113 114 But brief explanation of the behavior of memcg around shmem will be 115 helpful to understand the logic. 116 117 Shmem's page (just leaf page, not direct/indirect block) can be on 118 119 - radix-tree of shmem's inode. 120 - SwapCache. 121 - Both on radix-tree and SwapCache. This happens at swap-in 122 and swap-out, 123 124 It's charged when... 125 126 - A new page is added to shmem's radix-tree. 127 - A swp page is read. (move a charge from swap_cgroup to page_cgroup) 128 1297. Page Migration 130================= 131 132 mem_cgroup_migrate() 133 1348. LRU 135====== 136 Each memcg has its own vector of LRUs (inactive anon, active anon, 137 inactive file, active file, unevictable) of pages from each node, 138 each LRU handled under a single lru_lock for that memcg and node. 139 1409. Typical Tests. 141================= 142 143 Tests for racy cases. 144 1459.1 Small limit to memcg. 146------------------------- 147 148 When you do test to do racy case, it's good test to set memcg's limit 149 to be very small rather than GB. Many races found in the test under 150 xKB or xxMB limits. 151 152 (Memory behavior under GB and Memory behavior under MB shows very 153 different situation.) 154 1559.2 Shmem 156--------- 157 158 Historically, memcg's shmem handling was poor and we saw some amount 159 of troubles here. This is because shmem is page-cache but can be 160 SwapCache. Test with shmem/tmpfs is always good test. 161 1629.3 Migration 163------------- 164 165 For NUMA, migration is an another special case. To do easy test, cpuset 166 is useful. Following is a sample script to do migration:: 167 168 mount -t cgroup -o cpuset none /opt/cpuset 169 170 mkdir /opt/cpuset/01 171 echo 1 > /opt/cpuset/01/cpuset.cpus 172 echo 0 > /opt/cpuset/01/cpuset.mems 173 echo 1 > /opt/cpuset/01/cpuset.memory_migrate 174 mkdir /opt/cpuset/02 175 echo 1 > /opt/cpuset/02/cpuset.cpus 176 echo 1 > /opt/cpuset/02/cpuset.mems 177 echo 1 > /opt/cpuset/02/cpuset.memory_migrate 178 179 In above set, when you moves a task from 01 to 02, page migration to 180 node 0 to node 1 will occur. Following is a script to migrate all 181 under cpuset.:: 182 183 -- 184 move_task() 185 { 186 for pid in $1 187 do 188 /bin/echo $pid >$2/tasks 2>/dev/null 189 echo -n $pid 190 echo -n " " 191 done 192 echo END 193 } 194 195 G1_TASK=`cat ${G1}/tasks` 196 G2_TASK=`cat ${G2}/tasks` 197 move_task "${G1_TASK}" ${G2} & 198 -- 199 2009.4 Memory hotplug 201------------------ 202 203 memory hotplug test is one of good test. 204 205 to offline memory, do following:: 206 207 # echo offline > /sys/devices/system/memory/memoryXXX/state 208 209 (XXX is the place of memory) 210 211 This is an easy way to test page migration, too. 212 2139.5 nested cgroups 214------------------ 215 216 Use tests like the following for testing nested cgroups:: 217 218 mkdir /opt/cgroup/01/child_a 219 mkdir /opt/cgroup/01/child_b 220 221 set limit to 01. 222 add limit to 01/child_b 223 run jobs under child_a and child_b 224 225 create/delete following groups at random while jobs are running:: 226 227 /opt/cgroup/01/child_a/child_aa 228 /opt/cgroup/01/child_b/child_bb 229 /opt/cgroup/01/child_c 230 231 running new jobs in new group is also good. 232 2339.6 Mount with other subsystems 234------------------------------- 235 236 Mounting with other subsystems is a good test because there is a 237 race and lock dependency with other cgroup subsystems. 238 239 example:: 240 241 # mount -t cgroup none /cgroup -o cpuset,memory,cpu,devices 242 243 and do task move, mkdir, rmdir etc...under this. 244 2459.7 swapoff 246----------- 247 248 Besides management of swap is one of complicated parts of memcg, 249 call path of swap-in at swapoff is not same as usual swap-in path.. 250 It's worth to be tested explicitly. 251 252 For example, test like following is good: 253 254 (Shell-A):: 255 256 # mount -t cgroup none /cgroup -o memory 257 # mkdir /cgroup/test 258 # echo 40M > /cgroup/test/memory.limit_in_bytes 259 # echo 0 > /cgroup/test/tasks 260 261 Run malloc(100M) program under this. You'll see 60M of swaps. 262 263 (Shell-B):: 264 265 # move all tasks in /cgroup/test to /cgroup 266 # /sbin/swapoff -a 267 # rmdir /cgroup/test 268 # kill malloc task. 269 270 Of course, tmpfs v.s. swapoff test should be tested, too. 271 2729.8 OOM-Killer 273-------------- 274 275 Out-of-memory caused by memcg's limit will kill tasks under 276 the memcg. When hierarchy is used, a task under hierarchy 277 will be killed by the kernel. 278 279 In this case, panic_on_oom shouldn't be invoked and tasks 280 in other groups shouldn't be killed. 281 282 It's not difficult to cause OOM under memcg as following. 283 284 Case A) when you can swapoff:: 285 286 #swapoff -a 287 #echo 50M > /memory.limit_in_bytes 288 289 run 51M of malloc 290 291 Case B) when you use mem+swap limitation:: 292 293 #echo 50M > memory.limit_in_bytes 294 #echo 50M > memory.memsw.limit_in_bytes 295 296 run 51M of malloc 297 2989.9 Move charges at task migration 299---------------------------------- 300 301 Charges associated with a task can be moved along with task migration. 302 303 (Shell-A):: 304 305 #mkdir /cgroup/A 306 #echo $$ >/cgroup/A/tasks 307 308 run some programs which uses some amount of memory in /cgroup/A. 309 310 (Shell-B):: 311 312 #mkdir /cgroup/B 313 #echo 1 >/cgroup/B/memory.move_charge_at_immigrate 314 #echo "pid of the program running in group A" >/cgroup/B/tasks 315 316 You can see charges have been moved by reading ``*.usage_in_bytes`` or 317 memory.stat of both A and B. 318 319 See 8.2 of Documentation/admin-guide/cgroup-v1/memory.rst to see what value should 320 be written to move_charge_at_immigrate. 321 3229.10 Memory thresholds 323---------------------- 324 325 Memory controller implements memory thresholds using cgroups notification 326 API. You can use tools/cgroup/cgroup_event_listener.c to test it. 327 328 (Shell-A) Create cgroup and run event listener:: 329 330 # mkdir /cgroup/A 331 # ./cgroup_event_listener /cgroup/A/memory.usage_in_bytes 5M 332 333 (Shell-B) Add task to cgroup and try to allocate and free memory:: 334 335 # echo $$ >/cgroup/A/tasks 336 # a="$(dd if=/dev/zero bs=1M count=10)" 337 # a= 338 339 You will see message from cgroup_event_listener every time you cross 340 the thresholds. 341 342 Use /cgroup/A/memory.memsw.usage_in_bytes to test memsw thresholds. 343 344 It's good idea to test root cgroup as well. 345