cpusets.txt 26 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621
  1. CPUSETS
  2. -------
  3. Copyright (C) 2004 BULL SA.
  4. Written by Simon.Derr@bull.net
  5. Portions Copyright (c) 2004-2006 Silicon Graphics, Inc.
  6. Modified by Paul Jackson <pj@sgi.com>
  7. Modified by Christoph Lameter <clameter@sgi.com>
  8. CONTENTS:
  9. =========
  10. 1. Cpusets
  11. 1.1 What are cpusets ?
  12. 1.2 Why are cpusets needed ?
  13. 1.3 How are cpusets implemented ?
  14. 1.4 What are exclusive cpusets ?
  15. 1.5 What does notify_on_release do ?
  16. 1.6 What is memory_pressure ?
  17. 1.7 What is memory spread ?
  18. 1.8 How do I use cpusets ?
  19. 2. Usage Examples and Syntax
  20. 2.1 Basic Usage
  21. 2.2 Adding/removing cpus
  22. 2.3 Setting flags
  23. 2.4 Attaching processes
  24. 3. Questions
  25. 4. Contact
  26. 1. Cpusets
  27. ==========
  28. 1.1 What are cpusets ?
  29. ----------------------
  30. Cpusets provide a mechanism for assigning a set of CPUs and Memory
  31. Nodes to a set of tasks.
  32. Cpusets constrain the CPU and Memory placement of tasks to only
  33. the resources within a tasks current cpuset. They form a nested
  34. hierarchy visible in a virtual file system. These are the essential
  35. hooks, beyond what is already present, required to manage dynamic
  36. job placement on large systems.
  37. Each task has a pointer to a cpuset. Multiple tasks may reference
  38. the same cpuset. Requests by a task, using the sched_setaffinity(2)
  39. system call to include CPUs in its CPU affinity mask, and using the
  40. mbind(2) and set_mempolicy(2) system calls to include Memory Nodes
  41. in its memory policy, are both filtered through that tasks cpuset,
  42. filtering out any CPUs or Memory Nodes not in that cpuset. The
  43. scheduler will not schedule a task on a CPU that is not allowed in
  44. its cpus_allowed vector, and the kernel page allocator will not
  45. allocate a page on a node that is not allowed in the requesting tasks
  46. mems_allowed vector.
  47. User level code may create and destroy cpusets by name in the cpuset
  48. virtual file system, manage the attributes and permissions of these
  49. cpusets and which CPUs and Memory Nodes are assigned to each cpuset,
  50. specify and query to which cpuset a task is assigned, and list the
  51. task pids assigned to a cpuset.
  52. 1.2 Why are cpusets needed ?
  53. ----------------------------
  54. The management of large computer systems, with many processors (CPUs),
  55. complex memory cache hierarchies and multiple Memory Nodes having
  56. non-uniform access times (NUMA) presents additional challenges for
  57. the efficient scheduling and memory placement of processes.
  58. Frequently more modest sized systems can be operated with adequate
  59. efficiency just by letting the operating system automatically share
  60. the available CPU and Memory resources amongst the requesting tasks.
  61. But larger systems, which benefit more from careful processor and
  62. memory placement to reduce memory access times and contention,
  63. and which typically represent a larger investment for the customer,
  64. can benefit from explicitly placing jobs on properly sized subsets of
  65. the system.
  66. This can be especially valuable on:
  67. * Web Servers running multiple instances of the same web application,
  68. * Servers running different applications (for instance, a web server
  69. and a database), or
  70. * NUMA systems running large HPC applications with demanding
  71. performance characteristics.
  72. * Also cpu_exclusive cpusets are useful for servers running orthogonal
  73. workloads such as RT applications requiring low latency and HPC
  74. applications that are throughput sensitive
  75. These subsets, or "soft partitions" must be able to be dynamically
  76. adjusted, as the job mix changes, without impacting other concurrently
  77. executing jobs. The location of the running jobs pages may also be moved
  78. when the memory locations are changed.
  79. The kernel cpuset patch provides the minimum essential kernel
  80. mechanisms required to efficiently implement such subsets. It
  81. leverages existing CPU and Memory Placement facilities in the Linux
  82. kernel to avoid any additional impact on the critical scheduler or
  83. memory allocator code.
  84. 1.3 How are cpusets implemented ?
  85. ---------------------------------
  86. Cpusets provide a Linux kernel mechanism to constrain which CPUs and
  87. Memory Nodes are used by a process or set of processes.
  88. The Linux kernel already has a pair of mechanisms to specify on which
  89. CPUs a task may be scheduled (sched_setaffinity) and on which Memory
  90. Nodes it may obtain memory (mbind, set_mempolicy).
  91. Cpusets extends these two mechanisms as follows:
  92. - Cpusets are sets of allowed CPUs and Memory Nodes, known to the
  93. kernel.
  94. - Each task in the system is attached to a cpuset, via a pointer
  95. in the task structure to a reference counted cpuset structure.
  96. - Calls to sched_setaffinity are filtered to just those CPUs
  97. allowed in that tasks cpuset.
  98. - Calls to mbind and set_mempolicy are filtered to just
  99. those Memory Nodes allowed in that tasks cpuset.
  100. - The root cpuset contains all the systems CPUs and Memory
  101. Nodes.
  102. - For any cpuset, one can define child cpusets containing a subset
  103. of the parents CPU and Memory Node resources.
  104. - The hierarchy of cpusets can be mounted at /dev/cpuset, for
  105. browsing and manipulation from user space.
  106. - A cpuset may be marked exclusive, which ensures that no other
  107. cpuset (except direct ancestors and descendents) may contain
  108. any overlapping CPUs or Memory Nodes.
  109. Also a cpu_exclusive cpuset would be associated with a sched
  110. domain.
  111. - You can list all the tasks (by pid) attached to any cpuset.
  112. The implementation of cpusets requires a few, simple hooks
  113. into the rest of the kernel, none in performance critical paths:
  114. - in init/main.c, to initialize the root cpuset at system boot.
  115. - in fork and exit, to attach and detach a task from its cpuset.
  116. - in sched_setaffinity, to mask the requested CPUs by what's
  117. allowed in that tasks cpuset.
  118. - in sched.c migrate_all_tasks(), to keep migrating tasks within
  119. the CPUs allowed by their cpuset, if possible.
  120. - in sched.c, a new API partition_sched_domains for handling
  121. sched domain changes associated with cpu_exclusive cpusets
  122. and related changes in both sched.c and arch/ia64/kernel/domain.c
  123. - in the mbind and set_mempolicy system calls, to mask the requested
  124. Memory Nodes by what's allowed in that tasks cpuset.
  125. - in page_alloc.c, to restrict memory to allowed nodes.
  126. - in vmscan.c, to restrict page recovery to the current cpuset.
  127. In addition a new file system, of type "cpuset" may be mounted,
  128. typically at /dev/cpuset, to enable browsing and modifying the cpusets
  129. presently known to the kernel. No new system calls are added for
  130. cpusets - all support for querying and modifying cpusets is via
  131. this cpuset file system.
  132. Each task under /proc has an added file named 'cpuset', displaying
  133. the cpuset name, as the path relative to the root of the cpuset file
  134. system.
  135. The /proc/<pid>/status file for each task has two added lines,
  136. displaying the tasks cpus_allowed (on which CPUs it may be scheduled)
  137. and mems_allowed (on which Memory Nodes it may obtain memory),
  138. in the format seen in the following example:
  139. Cpus_allowed: ffffffff,ffffffff,ffffffff,ffffffff
  140. Mems_allowed: ffffffff,ffffffff
  141. Each cpuset is represented by a directory in the cpuset file system
  142. containing the following files describing that cpuset:
  143. - cpus: list of CPUs in that cpuset
  144. - mems: list of Memory Nodes in that cpuset
  145. - memory_migrate flag: if set, move pages to cpusets nodes
  146. - cpu_exclusive flag: is cpu placement exclusive?
  147. - mem_exclusive flag: is memory placement exclusive?
  148. - tasks: list of tasks (by pid) attached to that cpuset
  149. - notify_on_release flag: run /sbin/cpuset_release_agent on exit?
  150. - memory_pressure: measure of how much paging pressure in cpuset
  151. In addition, the root cpuset only has the following file:
  152. - memory_pressure_enabled flag: compute memory_pressure?
  153. New cpusets are created using the mkdir system call or shell
  154. command. The properties of a cpuset, such as its flags, allowed
  155. CPUs and Memory Nodes, and attached tasks, are modified by writing
  156. to the appropriate file in that cpusets directory, as listed above.
  157. The named hierarchical structure of nested cpusets allows partitioning
  158. a large system into nested, dynamically changeable, "soft-partitions".
  159. The attachment of each task, automatically inherited at fork by any
  160. children of that task, to a cpuset allows organizing the work load
  161. on a system into related sets of tasks such that each set is constrained
  162. to using the CPUs and Memory Nodes of a particular cpuset. A task
  163. may be re-attached to any other cpuset, if allowed by the permissions
  164. on the necessary cpuset file system directories.
  165. Such management of a system "in the large" integrates smoothly with
  166. the detailed placement done on individual tasks and memory regions
  167. using the sched_setaffinity, mbind and set_mempolicy system calls.
  168. The following rules apply to each cpuset:
  169. - Its CPUs and Memory Nodes must be a subset of its parents.
  170. - It can only be marked exclusive if its parent is.
  171. - If its cpu or memory is exclusive, they may not overlap any sibling.
  172. These rules, and the natural hierarchy of cpusets, enable efficient
  173. enforcement of the exclusive guarantee, without having to scan all
  174. cpusets every time any of them change to ensure nothing overlaps a
  175. exclusive cpuset. Also, the use of a Linux virtual file system (vfs)
  176. to represent the cpuset hierarchy provides for a familiar permission
  177. and name space for cpusets, with a minimum of additional kernel code.
  178. The cpus and mems files in the root (top_cpuset) cpuset are
  179. read-only. The cpus file automatically tracks the value of
  180. cpu_online_map using a CPU hotplug notifier, and the mems file
  181. automatically tracks the value of node_online_map using the
  182. cpuset_track_online_nodes() hook.
  183. 1.4 What are exclusive cpusets ?
  184. --------------------------------
  185. If a cpuset is cpu or mem exclusive, no other cpuset, other than
  186. a direct ancestor or descendent, may share any of the same CPUs or
  187. Memory Nodes.
  188. A cpuset that is cpu_exclusive has a scheduler (sched) domain
  189. associated with it. The sched domain consists of all CPUs in the
  190. current cpuset that are not part of any exclusive child cpusets.
  191. This ensures that the scheduler load balancing code only balances
  192. against the CPUs that are in the sched domain as defined above and
  193. not all of the CPUs in the system. This removes any overhead due to
  194. load balancing code trying to pull tasks outside of the cpu_exclusive
  195. cpuset only to be prevented by the tasks' cpus_allowed mask.
  196. A cpuset that is mem_exclusive restricts kernel allocations for
  197. page, buffer and other data commonly shared by the kernel across
  198. multiple users. All cpusets, whether mem_exclusive or not, restrict
  199. allocations of memory for user space. This enables configuring a
  200. system so that several independent jobs can share common kernel data,
  201. such as file system pages, while isolating each jobs user allocation in
  202. its own cpuset. To do this, construct a large mem_exclusive cpuset to
  203. hold all the jobs, and construct child, non-mem_exclusive cpusets for
  204. each individual job. Only a small amount of typical kernel memory,
  205. such as requests from interrupt handlers, is allowed to be taken
  206. outside even a mem_exclusive cpuset.
  207. 1.5 What does notify_on_release do ?
  208. ------------------------------------
  209. If the notify_on_release flag is enabled (1) in a cpuset, then whenever
  210. the last task in the cpuset leaves (exits or attaches to some other
  211. cpuset) and the last child cpuset of that cpuset is removed, then
  212. the kernel runs the command /sbin/cpuset_release_agent, supplying the
  213. pathname (relative to the mount point of the cpuset file system) of the
  214. abandoned cpuset. This enables automatic removal of abandoned cpusets.
  215. The default value of notify_on_release in the root cpuset at system
  216. boot is disabled (0). The default value of other cpusets at creation
  217. is the current value of their parents notify_on_release setting.
  218. 1.6 What is memory_pressure ?
  219. -----------------------------
  220. The memory_pressure of a cpuset provides a simple per-cpuset metric
  221. of the rate that the tasks in a cpuset are attempting to free up in
  222. use memory on the nodes of the cpuset to satisfy additional memory
  223. requests.
  224. This enables batch managers monitoring jobs running in dedicated
  225. cpusets to efficiently detect what level of memory pressure that job
  226. is causing.
  227. This is useful both on tightly managed systems running a wide mix of
  228. submitted jobs, which may choose to terminate or re-prioritize jobs that
  229. are trying to use more memory than allowed on the nodes assigned them,
  230. and with tightly coupled, long running, massively parallel scientific
  231. computing jobs that will dramatically fail to meet required performance
  232. goals if they start to use more memory than allowed to them.
  233. This mechanism provides a very economical way for the batch manager
  234. to monitor a cpuset for signs of memory pressure. It's up to the
  235. batch manager or other user code to decide what to do about it and
  236. take action.
  237. ==> Unless this feature is enabled by writing "1" to the special file
  238. /dev/cpuset/memory_pressure_enabled, the hook in the rebalance
  239. code of __alloc_pages() for this metric reduces to simply noticing
  240. that the cpuset_memory_pressure_enabled flag is zero. So only
  241. systems that enable this feature will compute the metric.
  242. Why a per-cpuset, running average:
  243. Because this meter is per-cpuset, rather than per-task or mm,
  244. the system load imposed by a batch scheduler monitoring this
  245. metric is sharply reduced on large systems, because a scan of
  246. the tasklist can be avoided on each set of queries.
  247. Because this meter is a running average, instead of an accumulating
  248. counter, a batch scheduler can detect memory pressure with a
  249. single read, instead of having to read and accumulate results
  250. for a period of time.
  251. Because this meter is per-cpuset rather than per-task or mm,
  252. the batch scheduler can obtain the key information, memory
  253. pressure in a cpuset, with a single read, rather than having to
  254. query and accumulate results over all the (dynamically changing)
  255. set of tasks in the cpuset.
  256. A per-cpuset simple digital filter (requires a spinlock and 3 words
  257. of data per-cpuset) is kept, and updated by any task attached to that
  258. cpuset, if it enters the synchronous (direct) page reclaim code.
  259. A per-cpuset file provides an integer number representing the recent
  260. (half-life of 10 seconds) rate of direct page reclaims caused by
  261. the tasks in the cpuset, in units of reclaims attempted per second,
  262. times 1000.
  263. 1.7 What is memory spread ?
  264. ---------------------------
  265. There are two boolean flag files per cpuset that control where the
  266. kernel allocates pages for the file system buffers and related in
  267. kernel data structures. They are called 'memory_spread_page' and
  268. 'memory_spread_slab'.
  269. If the per-cpuset boolean flag file 'memory_spread_page' is set, then
  270. the kernel will spread the file system buffers (page cache) evenly
  271. over all the nodes that the faulting task is allowed to use, instead
  272. of preferring to put those pages on the node where the task is running.
  273. If the per-cpuset boolean flag file 'memory_spread_slab' is set,
  274. then the kernel will spread some file system related slab caches,
  275. such as for inodes and dentries evenly over all the nodes that the
  276. faulting task is allowed to use, instead of preferring to put those
  277. pages on the node where the task is running.
  278. The setting of these flags does not affect anonymous data segment or
  279. stack segment pages of a task.
  280. By default, both kinds of memory spreading are off, and memory
  281. pages are allocated on the node local to where the task is running,
  282. except perhaps as modified by the tasks NUMA mempolicy or cpuset
  283. configuration, so long as sufficient free memory pages are available.
  284. When new cpusets are created, they inherit the memory spread settings
  285. of their parent.
  286. Setting memory spreading causes allocations for the affected page
  287. or slab caches to ignore the tasks NUMA mempolicy and be spread
  288. instead. Tasks using mbind() or set_mempolicy() calls to set NUMA
  289. mempolicies will not notice any change in these calls as a result of
  290. their containing tasks memory spread settings. If memory spreading
  291. is turned off, then the currently specified NUMA mempolicy once again
  292. applies to memory page allocations.
  293. Both 'memory_spread_page' and 'memory_spread_slab' are boolean flag
  294. files. By default they contain "0", meaning that the feature is off
  295. for that cpuset. If a "1" is written to that file, then that turns
  296. the named feature on.
  297. The implementation is simple.
  298. Setting the flag 'memory_spread_page' turns on a per-process flag
  299. PF_SPREAD_PAGE for each task that is in that cpuset or subsequently
  300. joins that cpuset. The page allocation calls for the page cache
  301. is modified to perform an inline check for this PF_SPREAD_PAGE task
  302. flag, and if set, a call to a new routine cpuset_mem_spread_node()
  303. returns the node to prefer for the allocation.
  304. Similarly, setting 'memory_spread_cache' turns on the flag
  305. PF_SPREAD_SLAB, and appropriately marked slab caches will allocate
  306. pages from the node returned by cpuset_mem_spread_node().
  307. The cpuset_mem_spread_node() routine is also simple. It uses the
  308. value of a per-task rotor cpuset_mem_spread_rotor to select the next
  309. node in the current tasks mems_allowed to prefer for the allocation.
  310. This memory placement policy is also known (in other contexts) as
  311. round-robin or interleave.
  312. This policy can provide substantial improvements for jobs that need
  313. to place thread local data on the corresponding node, but that need
  314. to access large file system data sets that need to be spread across
  315. the several nodes in the jobs cpuset in order to fit. Without this
  316. policy, especially for jobs that might have one thread reading in the
  317. data set, the memory allocation across the nodes in the jobs cpuset
  318. can become very uneven.
  319. 1.8 How do I use cpusets ?
  320. --------------------------
  321. In order to minimize the impact of cpusets on critical kernel
  322. code, such as the scheduler, and due to the fact that the kernel
  323. does not support one task updating the memory placement of another
  324. task directly, the impact on a task of changing its cpuset CPU
  325. or Memory Node placement, or of changing to which cpuset a task
  326. is attached, is subtle.
  327. If a cpuset has its Memory Nodes modified, then for each task attached
  328. to that cpuset, the next time that the kernel attempts to allocate
  329. a page of memory for that task, the kernel will notice the change
  330. in the tasks cpuset, and update its per-task memory placement to
  331. remain within the new cpusets memory placement. If the task was using
  332. mempolicy MPOL_BIND, and the nodes to which it was bound overlap with
  333. its new cpuset, then the task will continue to use whatever subset
  334. of MPOL_BIND nodes are still allowed in the new cpuset. If the task
  335. was using MPOL_BIND and now none of its MPOL_BIND nodes are allowed
  336. in the new cpuset, then the task will be essentially treated as if it
  337. was MPOL_BIND bound to the new cpuset (even though its numa placement,
  338. as queried by get_mempolicy(), doesn't change). If a task is moved
  339. from one cpuset to another, then the kernel will adjust the tasks
  340. memory placement, as above, the next time that the kernel attempts
  341. to allocate a page of memory for that task.
  342. If a cpuset has its CPUs modified, then each task using that
  343. cpuset does _not_ change its behavior automatically. In order to
  344. minimize the impact on the critical scheduling code in the kernel,
  345. tasks will continue to use their prior CPU placement until they
  346. are rebound to their cpuset, by rewriting their pid to the 'tasks'
  347. file of their cpuset. If a task had been bound to some subset of its
  348. cpuset using the sched_setaffinity() call, and if any of that subset
  349. is still allowed in its new cpuset settings, then the task will be
  350. restricted to the intersection of the CPUs it was allowed on before,
  351. and its new cpuset CPU placement. If, on the other hand, there is
  352. no overlap between a tasks prior placement and its new cpuset CPU
  353. placement, then the task will be allowed to run on any CPU allowed
  354. in its new cpuset. If a task is moved from one cpuset to another,
  355. its CPU placement is updated in the same way as if the tasks pid is
  356. rewritten to the 'tasks' file of its current cpuset.
  357. In summary, the memory placement of a task whose cpuset is changed is
  358. updated by the kernel, on the next allocation of a page for that task,
  359. but the processor placement is not updated, until that tasks pid is
  360. rewritten to the 'tasks' file of its cpuset. This is done to avoid
  361. impacting the scheduler code in the kernel with a check for changes
  362. in a tasks processor placement.
  363. Normally, once a page is allocated (given a physical page
  364. of main memory) then that page stays on whatever node it
  365. was allocated, so long as it remains allocated, even if the
  366. cpusets memory placement policy 'mems' subsequently changes.
  367. If the cpuset flag file 'memory_migrate' is set true, then when
  368. tasks are attached to that cpuset, any pages that task had
  369. allocated to it on nodes in its previous cpuset are migrated
  370. to the tasks new cpuset. The relative placement of the page within
  371. the cpuset is preserved during these migration operations if possible.
  372. For example if the page was on the second valid node of the prior cpuset
  373. then the page will be placed on the second valid node of the new cpuset.
  374. Also if 'memory_migrate' is set true, then if that cpusets
  375. 'mems' file is modified, pages allocated to tasks in that
  376. cpuset, that were on nodes in the previous setting of 'mems',
  377. will be moved to nodes in the new setting of 'mems.'
  378. Pages that were not in the tasks prior cpuset, or in the cpusets
  379. prior 'mems' setting, will not be moved.
  380. There is an exception to the above. If hotplug functionality is used
  381. to remove all the CPUs that are currently assigned to a cpuset,
  382. then the kernel will automatically update the cpus_allowed of all
  383. tasks attached to CPUs in that cpuset to allow all CPUs. When memory
  384. hotplug functionality for removing Memory Nodes is available, a
  385. similar exception is expected to apply there as well. In general,
  386. the kernel prefers to violate cpuset placement, over starving a task
  387. that has had all its allowed CPUs or Memory Nodes taken offline. User
  388. code should reconfigure cpusets to only refer to online CPUs and Memory
  389. Nodes when using hotplug to add or remove such resources.
  390. There is a second exception to the above. GFP_ATOMIC requests are
  391. kernel internal allocations that must be satisfied, immediately.
  392. The kernel may drop some request, in rare cases even panic, if a
  393. GFP_ATOMIC alloc fails. If the request cannot be satisfied within
  394. the current tasks cpuset, then we relax the cpuset, and look for
  395. memory anywhere we can find it. It's better to violate the cpuset
  396. than stress the kernel.
  397. To start a new job that is to be contained within a cpuset, the steps are:
  398. 1) mkdir /dev/cpuset
  399. 2) mount -t cpuset none /dev/cpuset
  400. 3) Create the new cpuset by doing mkdir's and write's (or echo's) in
  401. the /dev/cpuset virtual file system.
  402. 4) Start a task that will be the "founding father" of the new job.
  403. 5) Attach that task to the new cpuset by writing its pid to the
  404. /dev/cpuset tasks file for that cpuset.
  405. 6) fork, exec or clone the job tasks from this founding father task.
  406. For example, the following sequence of commands will setup a cpuset
  407. named "Charlie", containing just CPUs 2 and 3, and Memory Node 1,
  408. and then start a subshell 'sh' in that cpuset:
  409. mount -t cpuset none /dev/cpuset
  410. cd /dev/cpuset
  411. mkdir Charlie
  412. cd Charlie
  413. /bin/echo 2-3 > cpus
  414. /bin/echo 1 > mems
  415. /bin/echo $$ > tasks
  416. sh
  417. # The subshell 'sh' is now running in cpuset Charlie
  418. # The next line should display '/Charlie'
  419. cat /proc/self/cpuset
  420. In the future, a C library interface to cpusets will likely be
  421. available. For now, the only way to query or modify cpusets is
  422. via the cpuset file system, using the various cd, mkdir, echo, cat,
  423. rmdir commands from the shell, or their equivalent from C.
  424. The sched_setaffinity calls can also be done at the shell prompt using
  425. SGI's runon or Robert Love's taskset. The mbind and set_mempolicy
  426. calls can be done at the shell prompt using the numactl command
  427. (part of Andi Kleen's numa package).
  428. 2. Usage Examples and Syntax
  429. ============================
  430. 2.1 Basic Usage
  431. ---------------
  432. Creating, modifying, using the cpusets can be done through the cpuset
  433. virtual filesystem.
  434. To mount it, type:
  435. # mount -t cpuset none /dev/cpuset
  436. Then under /dev/cpuset you can find a tree that corresponds to the
  437. tree of the cpusets in the system. For instance, /dev/cpuset
  438. is the cpuset that holds the whole system.
  439. If you want to create a new cpuset under /dev/cpuset:
  440. # cd /dev/cpuset
  441. # mkdir my_cpuset
  442. Now you want to do something with this cpuset.
  443. # cd my_cpuset
  444. In this directory you can find several files:
  445. # ls
  446. cpus cpu_exclusive mems mem_exclusive tasks
  447. Reading them will give you information about the state of this cpuset:
  448. the CPUs and Memory Nodes it can use, the processes that are using
  449. it, its properties. By writing to these files you can manipulate
  450. the cpuset.
  451. Set some flags:
  452. # /bin/echo 1 > cpu_exclusive
  453. Add some cpus:
  454. # /bin/echo 0-7 > cpus
  455. Add some mems:
  456. # /bin/echo 0-7 > mems
  457. Now attach your shell to this cpuset:
  458. # /bin/echo $$ > tasks
  459. You can also create cpusets inside your cpuset by using mkdir in this
  460. directory.
  461. # mkdir my_sub_cs
  462. To remove a cpuset, just use rmdir:
  463. # rmdir my_sub_cs
  464. This will fail if the cpuset is in use (has cpusets inside, or has
  465. processes attached).
  466. 2.2 Adding/removing cpus
  467. ------------------------
  468. This is the syntax to use when writing in the cpus or mems files
  469. in cpuset directories:
  470. # /bin/echo 1-4 > cpus -> set cpus list to cpus 1,2,3,4
  471. # /bin/echo 1,2,3,4 > cpus -> set cpus list to cpus 1,2,3,4
  472. 2.3 Setting flags
  473. -----------------
  474. The syntax is very simple:
  475. # /bin/echo 1 > cpu_exclusive -> set flag 'cpu_exclusive'
  476. # /bin/echo 0 > cpu_exclusive -> unset flag 'cpu_exclusive'
  477. 2.4 Attaching processes
  478. -----------------------
  479. # /bin/echo PID > tasks
  480. Note that it is PID, not PIDs. You can only attach ONE task at a time.
  481. If you have several tasks to attach, you have to do it one after another:
  482. # /bin/echo PID1 > tasks
  483. # /bin/echo PID2 > tasks
  484. ...
  485. # /bin/echo PIDn > tasks
  486. 3. Questions
  487. ============
  488. Q: what's up with this '/bin/echo' ?
  489. A: bash's builtin 'echo' command does not check calls to write() against
  490. errors. If you use it in the cpuset file system, you won't be
  491. able to tell whether a command succeeded or failed.
  492. Q: When I attach processes, only the first of the line gets really attached !
  493. A: We can only return one error code per call to write(). So you should also
  494. put only ONE pid.
  495. 4. Contact
  496. ==========
  497. Web: http://www.bullopensource.org/cpuset