blob: 842f0d1ab2165781420fdd53b3875a16ba91adb6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001 CPUSETS
2 -------
3
4Copyright (C) 2004 BULL SA.
5Written by Simon.Derr@bull.net
6
Christoph Lameterb4fb3762006-03-14 19:50:20 -08007Portions Copyright (c) 2004-2006 Silicon Graphics, Inc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07008Modified by Paul Jackson <pj@sgi.com>
Christoph Lameterb4fb3762006-03-14 19:50:20 -08009Modified by Christoph Lameter <clameter@sgi.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010
11CONTENTS:
12=========
13
141. Cpusets
15 1.1 What are cpusets ?
16 1.2 Why are cpusets needed ?
17 1.3 How are cpusets implemented ?
Paul Jacksonbd5e09c2006-01-08 01:01:50 -080018 1.4 What are exclusive cpusets ?
19 1.5 What does notify_on_release do ?
Paul Jackson90c9cc42006-01-08 01:01:51 -080020 1.6 What is memory_pressure ?
Paul Jackson825a46a2006-03-24 03:16:03 -080021 1.7 What is memory spread ?
22 1.8 How do I use cpusets ?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232. Usage Examples and Syntax
24 2.1 Basic Usage
25 2.2 Adding/removing cpus
26 2.3 Setting flags
27 2.4 Attaching processes
283. Questions
294. Contact
30
311. Cpusets
32==========
33
341.1 What are cpusets ?
35----------------------
36
37Cpusets provide a mechanism for assigning a set of CPUs and Memory
38Nodes to a set of tasks.
39
40Cpusets constrain the CPU and Memory placement of tasks to only
41the resources within a tasks current cpuset. They form a nested
42hierarchy visible in a virtual file system. These are the essential
43hooks, beyond what is already present, required to manage dynamic
44job placement on large systems.
45
46Each task has a pointer to a cpuset. Multiple tasks may reference
47the same cpuset. Requests by a task, using the sched_setaffinity(2)
48system call to include CPUs in its CPU affinity mask, and using the
49mbind(2) and set_mempolicy(2) system calls to include Memory Nodes
50in its memory policy, are both filtered through that tasks cpuset,
51filtering out any CPUs or Memory Nodes not in that cpuset. The
52scheduler will not schedule a task on a CPU that is not allowed in
53its cpus_allowed vector, and the kernel page allocator will not
54allocate a page on a node that is not allowed in the requesting tasks
55mems_allowed vector.
56
Linus Torvalds1da177e2005-04-16 15:20:36 -070057User level code may create and destroy cpusets by name in the cpuset
58virtual file system, manage the attributes and permissions of these
59cpusets and which CPUs and Memory Nodes are assigned to each cpuset,
60specify and query to which cpuset a task is assigned, and list the
61task pids assigned to a cpuset.
62
63
641.2 Why are cpusets needed ?
65----------------------------
66
67The management of large computer systems, with many processors (CPUs),
68complex memory cache hierarchies and multiple Memory Nodes having
69non-uniform access times (NUMA) presents additional challenges for
70the efficient scheduling and memory placement of processes.
71
72Frequently more modest sized systems can be operated with adequate
73efficiency just by letting the operating system automatically share
74the available CPU and Memory resources amongst the requesting tasks.
75
76But larger systems, which benefit more from careful processor and
77memory placement to reduce memory access times and contention,
78and which typically represent a larger investment for the customer,
Jean Delvare33430dc2005-10-30 15:02:20 -080079can benefit from explicitly placing jobs on properly sized subsets of
Linus Torvalds1da177e2005-04-16 15:20:36 -070080the system.
81
82This can be especially valuable on:
83
84 * Web Servers running multiple instances of the same web application,
85 * Servers running different applications (for instance, a web server
86 and a database), or
87 * NUMA systems running large HPC applications with demanding
88 performance characteristics.
Dinakar Guniguntala85d7b942005-06-25 14:57:34 -070089 * Also cpu_exclusive cpusets are useful for servers running orthogonal
90 workloads such as RT applications requiring low latency and HPC
91 applications that are throughput sensitive
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
93These subsets, or "soft partitions" must be able to be dynamically
94adjusted, as the job mix changes, without impacting other concurrently
Christoph Lameterb4fb3762006-03-14 19:50:20 -080095executing jobs. The location of the running jobs pages may also be moved
96when the memory locations are changed.
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
98The kernel cpuset patch provides the minimum essential kernel
99mechanisms required to efficiently implement such subsets. It
100leverages existing CPU and Memory Placement facilities in the Linux
101kernel to avoid any additional impact on the critical scheduler or
102memory allocator code.
103
104
1051.3 How are cpusets implemented ?
106---------------------------------
107
Christoph Lameterb4fb3762006-03-14 19:50:20 -0800108Cpusets provide a Linux kernel mechanism to constrain which CPUs and
109Memory Nodes are used by a process or set of processes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
111The Linux kernel already has a pair of mechanisms to specify on which
112CPUs a task may be scheduled (sched_setaffinity) and on which Memory
113Nodes it may obtain memory (mbind, set_mempolicy).
114
115Cpusets extends these two mechanisms as follows:
116
117 - Cpusets are sets of allowed CPUs and Memory Nodes, known to the
118 kernel.
119 - Each task in the system is attached to a cpuset, via a pointer
120 in the task structure to a reference counted cpuset structure.
121 - Calls to sched_setaffinity are filtered to just those CPUs
122 allowed in that tasks cpuset.
123 - Calls to mbind and set_mempolicy are filtered to just
124 those Memory Nodes allowed in that tasks cpuset.
125 - The root cpuset contains all the systems CPUs and Memory
126 Nodes.
127 - For any cpuset, one can define child cpusets containing a subset
128 of the parents CPU and Memory Node resources.
129 - The hierarchy of cpusets can be mounted at /dev/cpuset, for
130 browsing and manipulation from user space.
131 - A cpuset may be marked exclusive, which ensures that no other
132 cpuset (except direct ancestors and descendents) may contain
133 any overlapping CPUs or Memory Nodes.
Dinakar Guniguntala85d7b942005-06-25 14:57:34 -0700134 Also a cpu_exclusive cpuset would be associated with a sched
135 domain.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 - You can list all the tasks (by pid) attached to any cpuset.
137
138The implementation of cpusets requires a few, simple hooks
139into the rest of the kernel, none in performance critical paths:
140
Paul Jackson864913f2006-01-11 02:01:38 +0100141 - in init/main.c, to initialize the root cpuset at system boot.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 - in fork and exit, to attach and detach a task from its cpuset.
143 - in sched_setaffinity, to mask the requested CPUs by what's
144 allowed in that tasks cpuset.
145 - in sched.c migrate_all_tasks(), to keep migrating tasks within
146 the CPUs allowed by their cpuset, if possible.
Dinakar Guniguntala85d7b942005-06-25 14:57:34 -0700147 - in sched.c, a new API partition_sched_domains for handling
148 sched domain changes associated with cpu_exclusive cpusets
149 and related changes in both sched.c and arch/ia64/kernel/domain.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 - in the mbind and set_mempolicy system calls, to mask the requested
151 Memory Nodes by what's allowed in that tasks cpuset.
Paul Jackson864913f2006-01-11 02:01:38 +0100152 - in page_alloc.c, to restrict memory to allowed nodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 - in vmscan.c, to restrict page recovery to the current cpuset.
154
155In addition a new file system, of type "cpuset" may be mounted,
156typically at /dev/cpuset, to enable browsing and modifying the cpusets
157presently known to the kernel. No new system calls are added for
158cpusets - all support for querying and modifying cpusets is via
159this cpuset file system.
160
161Each task under /proc has an added file named 'cpuset', displaying
162the cpuset name, as the path relative to the root of the cpuset file
163system.
164
165The /proc/<pid>/status file for each task has two added lines,
166displaying the tasks cpus_allowed (on which CPUs it may be scheduled)
167and mems_allowed (on which Memory Nodes it may obtain memory),
168in the format seen in the following example:
169
170 Cpus_allowed: ffffffff,ffffffff,ffffffff,ffffffff
171 Mems_allowed: ffffffff,ffffffff
172
173Each cpuset is represented by a directory in the cpuset file system
174containing the following files describing that cpuset:
175
176 - cpus: list of CPUs in that cpuset
177 - mems: list of Memory Nodes in that cpuset
Paul Jackson45b07ef2006-01-08 01:00:56 -0800178 - memory_migrate flag: if set, move pages to cpusets nodes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 - cpu_exclusive flag: is cpu placement exclusive?
180 - mem_exclusive flag: is memory placement exclusive?
181 - tasks: list of tasks (by pid) attached to that cpuset
Paul Jacksonbd5e09c2006-01-08 01:01:50 -0800182 - notify_on_release flag: run /sbin/cpuset_release_agent on exit?
Paul Jacksonbd5e09c2006-01-08 01:01:50 -0800183 - memory_pressure: measure of how much paging pressure in cpuset
184
185In addition, the root cpuset only has the following file:
186 - memory_pressure_enabled flag: compute memory_pressure?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188New cpusets are created using the mkdir system call or shell
189command. The properties of a cpuset, such as its flags, allowed
190CPUs and Memory Nodes, and attached tasks, are modified by writing
191to the appropriate file in that cpusets directory, as listed above.
192
193The named hierarchical structure of nested cpusets allows partitioning
194a large system into nested, dynamically changeable, "soft-partitions".
195
196The attachment of each task, automatically inherited at fork by any
197children of that task, to a cpuset allows organizing the work load
198on a system into related sets of tasks such that each set is constrained
199to using the CPUs and Memory Nodes of a particular cpuset. A task
200may be re-attached to any other cpuset, if allowed by the permissions
201on the necessary cpuset file system directories.
202
203Such management of a system "in the large" integrates smoothly with
204the detailed placement done on individual tasks and memory regions
205using the sched_setaffinity, mbind and set_mempolicy system calls.
206
207The following rules apply to each cpuset:
208
209 - Its CPUs and Memory Nodes must be a subset of its parents.
210 - It can only be marked exclusive if its parent is.
211 - If its cpu or memory is exclusive, they may not overlap any sibling.
212
213These rules, and the natural hierarchy of cpusets, enable efficient
214enforcement of the exclusive guarantee, without having to scan all
215cpusets every time any of them change to ensure nothing overlaps a
216exclusive cpuset. Also, the use of a Linux virtual file system (vfs)
217to represent the cpuset hierarchy provides for a familiar permission
218and name space for cpusets, with a minimum of additional kernel code.
219
Paul Jackson38837fc2006-09-29 02:01:16 -0700220The cpus and mems files in the root (top_cpuset) cpuset are
221read-only. The cpus file automatically tracks the value of
222cpu_online_map using a CPU hotplug notifier, and the mems file
223automatically tracks the value of node_online_map using the
224cpuset_track_online_nodes() hook.
Paul Jackson4c4d50f2006-08-27 01:23:51 -0700225
Paul Jacksonbd5e09c2006-01-08 01:01:50 -0800226
2271.4 What are exclusive cpusets ?
228--------------------------------
229
230If a cpuset is cpu or mem exclusive, no other cpuset, other than
231a direct ancestor or descendent, may share any of the same CPUs or
232Memory Nodes.
233
234A cpuset that is cpu_exclusive has a scheduler (sched) domain
235associated with it. The sched domain consists of all CPUs in the
236current cpuset that are not part of any exclusive child cpusets.
237This ensures that the scheduler load balancing code only balances
238against the CPUs that are in the sched domain as defined above and
239not all of the CPUs in the system. This removes any overhead due to
240load balancing code trying to pull tasks outside of the cpu_exclusive
241cpuset only to be prevented by the tasks' cpus_allowed mask.
242
243A cpuset that is mem_exclusive restricts kernel allocations for
244page, buffer and other data commonly shared by the kernel across
245multiple users. All cpusets, whether mem_exclusive or not, restrict
246allocations of memory for user space. This enables configuring a
247system so that several independent jobs can share common kernel data,
248such as file system pages, while isolating each jobs user allocation in
249its own cpuset. To do this, construct a large mem_exclusive cpuset to
250hold all the jobs, and construct child, non-mem_exclusive cpusets for
251each individual job. Only a small amount of typical kernel memory,
252such as requests from interrupt handlers, is allowed to be taken
253outside even a mem_exclusive cpuset.
254
255
2561.5 What does notify_on_release do ?
257------------------------------------
258
259If the notify_on_release flag is enabled (1) in a cpuset, then whenever
260the last task in the cpuset leaves (exits or attaches to some other
261cpuset) and the last child cpuset of that cpuset is removed, then
262the kernel runs the command /sbin/cpuset_release_agent, supplying the
263pathname (relative to the mount point of the cpuset file system) of the
264abandoned cpuset. This enables automatic removal of abandoned cpusets.
265The default value of notify_on_release in the root cpuset at system
266boot is disabled (0). The default value of other cpusets at creation
267is the current value of their parents notify_on_release setting.
268
269
Paul Jackson90c9cc42006-01-08 01:01:51 -08002701.6 What is memory_pressure ?
Paul Jacksonbd5e09c2006-01-08 01:01:50 -0800271-----------------------------
272The memory_pressure of a cpuset provides a simple per-cpuset metric
273of the rate that the tasks in a cpuset are attempting to free up in
274use memory on the nodes of the cpuset to satisfy additional memory
275requests.
276
277This enables batch managers monitoring jobs running in dedicated
278cpusets to efficiently detect what level of memory pressure that job
279is causing.
280
281This is useful both on tightly managed systems running a wide mix of
282submitted jobs, which may choose to terminate or re-prioritize jobs that
283are trying to use more memory than allowed on the nodes assigned them,
284and with tightly coupled, long running, massively parallel scientific
285computing jobs that will dramatically fail to meet required performance
286goals if they start to use more memory than allowed to them.
287
288This mechanism provides a very economical way for the batch manager
289to monitor a cpuset for signs of memory pressure. It's up to the
290batch manager or other user code to decide what to do about it and
291take action.
292
293==> Unless this feature is enabled by writing "1" to the special file
294 /dev/cpuset/memory_pressure_enabled, the hook in the rebalance
295 code of __alloc_pages() for this metric reduces to simply noticing
296 that the cpuset_memory_pressure_enabled flag is zero. So only
297 systems that enable this feature will compute the metric.
298
299Why a per-cpuset, running average:
300
301 Because this meter is per-cpuset, rather than per-task or mm,
302 the system load imposed by a batch scheduler monitoring this
303 metric is sharply reduced on large systems, because a scan of
304 the tasklist can be avoided on each set of queries.
305
306 Because this meter is a running average, instead of an accumulating
307 counter, a batch scheduler can detect memory pressure with a
308 single read, instead of having to read and accumulate results
309 for a period of time.
310
311 Because this meter is per-cpuset rather than per-task or mm,
312 the batch scheduler can obtain the key information, memory
313 pressure in a cpuset, with a single read, rather than having to
314 query and accumulate results over all the (dynamically changing)
315 set of tasks in the cpuset.
316
317A per-cpuset simple digital filter (requires a spinlock and 3 words
318of data per-cpuset) is kept, and updated by any task attached to that
319cpuset, if it enters the synchronous (direct) page reclaim code.
320
321A per-cpuset file provides an integer number representing the recent
322(half-life of 10 seconds) rate of direct page reclaims caused by
323the tasks in the cpuset, in units of reclaims attempted per second,
324times 1000.
325
326
Paul Jackson825a46a2006-03-24 03:16:03 -08003271.7 What is memory spread ?
328---------------------------
329There are two boolean flag files per cpuset that control where the
330kernel allocates pages for the file system buffers and related in
331kernel data structures. They are called 'memory_spread_page' and
332'memory_spread_slab'.
333
334If the per-cpuset boolean flag file 'memory_spread_page' is set, then
335the kernel will spread the file system buffers (page cache) evenly
336over all the nodes that the faulting task is allowed to use, instead
337of preferring to put those pages on the node where the task is running.
338
339If the per-cpuset boolean flag file 'memory_spread_slab' is set,
340then the kernel will spread some file system related slab caches,
341such as for inodes and dentries evenly over all the nodes that the
342faulting task is allowed to use, instead of preferring to put those
343pages on the node where the task is running.
344
345The setting of these flags does not affect anonymous data segment or
346stack segment pages of a task.
347
348By default, both kinds of memory spreading are off, and memory
349pages are allocated on the node local to where the task is running,
350except perhaps as modified by the tasks NUMA mempolicy or cpuset
351configuration, so long as sufficient free memory pages are available.
352
353When new cpusets are created, they inherit the memory spread settings
354of their parent.
355
356Setting memory spreading causes allocations for the affected page
357or slab caches to ignore the tasks NUMA mempolicy and be spread
358instead. Tasks using mbind() or set_mempolicy() calls to set NUMA
359mempolicies will not notice any change in these calls as a result of
360their containing tasks memory spread settings. If memory spreading
361is turned off, then the currently specified NUMA mempolicy once again
362applies to memory page allocations.
363
364Both 'memory_spread_page' and 'memory_spread_slab' are boolean flag
365files. By default they contain "0", meaning that the feature is off
366for that cpuset. If a "1" is written to that file, then that turns
367the named feature on.
368
369The implementation is simple.
370
371Setting the flag 'memory_spread_page' turns on a per-process flag
372PF_SPREAD_PAGE for each task that is in that cpuset or subsequently
373joins that cpuset. The page allocation calls for the page cache
374is modified to perform an inline check for this PF_SPREAD_PAGE task
375flag, and if set, a call to a new routine cpuset_mem_spread_node()
376returns the node to prefer for the allocation.
377
378Similarly, setting 'memory_spread_cache' turns on the flag
379PF_SPREAD_SLAB, and appropriately marked slab caches will allocate
380pages from the node returned by cpuset_mem_spread_node().
381
382The cpuset_mem_spread_node() routine is also simple. It uses the
383value of a per-task rotor cpuset_mem_spread_rotor to select the next
384node in the current tasks mems_allowed to prefer for the allocation.
385
386This memory placement policy is also known (in other contexts) as
387round-robin or interleave.
388
389This policy can provide substantial improvements for jobs that need
390to place thread local data on the corresponding node, but that need
391to access large file system data sets that need to be spread across
392the several nodes in the jobs cpuset in order to fit. Without this
393policy, especially for jobs that might have one thread reading in the
394data set, the memory allocation across the nodes in the jobs cpuset
395can become very uneven.
396
397
3981.8 How do I use cpusets ?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399--------------------------
400
401In order to minimize the impact of cpusets on critical kernel
402code, such as the scheduler, and due to the fact that the kernel
403does not support one task updating the memory placement of another
404task directly, the impact on a task of changing its cpuset CPU
405or Memory Node placement, or of changing to which cpuset a task
406is attached, is subtle.
407
408If a cpuset has its Memory Nodes modified, then for each task attached
409to that cpuset, the next time that the kernel attempts to allocate
410a page of memory for that task, the kernel will notice the change
411in the tasks cpuset, and update its per-task memory placement to
412remain within the new cpusets memory placement. If the task was using
413mempolicy MPOL_BIND, and the nodes to which it was bound overlap with
414its new cpuset, then the task will continue to use whatever subset
415of MPOL_BIND nodes are still allowed in the new cpuset. If the task
416was using MPOL_BIND and now none of its MPOL_BIND nodes are allowed
417in the new cpuset, then the task will be essentially treated as if it
418was MPOL_BIND bound to the new cpuset (even though its numa placement,
419as queried by get_mempolicy(), doesn't change). If a task is moved
420from one cpuset to another, then the kernel will adjust the tasks
421memory placement, as above, the next time that the kernel attempts
422to allocate a page of memory for that task.
423
424If a cpuset has its CPUs modified, then each task using that
425cpuset does _not_ change its behavior automatically. In order to
426minimize the impact on the critical scheduling code in the kernel,
427tasks will continue to use their prior CPU placement until they
428are rebound to their cpuset, by rewriting their pid to the 'tasks'
429file of their cpuset. If a task had been bound to some subset of its
430cpuset using the sched_setaffinity() call, and if any of that subset
431is still allowed in its new cpuset settings, then the task will be
432restricted to the intersection of the CPUs it was allowed on before,
433and its new cpuset CPU placement. If, on the other hand, there is
434no overlap between a tasks prior placement and its new cpuset CPU
435placement, then the task will be allowed to run on any CPU allowed
436in its new cpuset. If a task is moved from one cpuset to another,
437its CPU placement is updated in the same way as if the tasks pid is
438rewritten to the 'tasks' file of its current cpuset.
439
440In summary, the memory placement of a task whose cpuset is changed is
441updated by the kernel, on the next allocation of a page for that task,
442but the processor placement is not updated, until that tasks pid is
443rewritten to the 'tasks' file of its cpuset. This is done to avoid
444impacting the scheduler code in the kernel with a check for changes
445in a tasks processor placement.
446
Paul Jackson45b07ef2006-01-08 01:00:56 -0800447Normally, once a page is allocated (given a physical page
448of main memory) then that page stays on whatever node it
449was allocated, so long as it remains allocated, even if the
450cpusets memory placement policy 'mems' subsequently changes.
451If the cpuset flag file 'memory_migrate' is set true, then when
452tasks are attached to that cpuset, any pages that task had
453allocated to it on nodes in its previous cpuset are migrated
Christoph Lameterb4fb3762006-03-14 19:50:20 -0800454to the tasks new cpuset. The relative placement of the page within
455the cpuset is preserved during these migration operations if possible.
456For example if the page was on the second valid node of the prior cpuset
457then the page will be placed on the second valid node of the new cpuset.
458
Paul Jackson45b07ef2006-01-08 01:00:56 -0800459Also if 'memory_migrate' is set true, then if that cpusets
460'mems' file is modified, pages allocated to tasks in that
461cpuset, that were on nodes in the previous setting of 'mems',
Christoph Lameterb4fb3762006-03-14 19:50:20 -0800462will be moved to nodes in the new setting of 'mems.'
463Pages that were not in the tasks prior cpuset, or in the cpusets
464prior 'mems' setting, will not be moved.
Paul Jackson45b07ef2006-01-08 01:00:56 -0800465
Tobias Klauserd533f672005-09-10 00:26:46 -0700466There is an exception to the above. If hotplug functionality is used
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467to remove all the CPUs that are currently assigned to a cpuset,
468then the kernel will automatically update the cpus_allowed of all
Paul Jacksonb39c4fa2005-05-20 13:59:15 -0700469tasks attached to CPUs in that cpuset to allow all CPUs. When memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470hotplug functionality for removing Memory Nodes is available, a
471similar exception is expected to apply there as well. In general,
472the kernel prefers to violate cpuset placement, over starving a task
473that has had all its allowed CPUs or Memory Nodes taken offline. User
474code should reconfigure cpusets to only refer to online CPUs and Memory
475Nodes when using hotplug to add or remove such resources.
476
477There is a second exception to the above. GFP_ATOMIC requests are
478kernel internal allocations that must be satisfied, immediately.
479The kernel may drop some request, in rare cases even panic, if a
480GFP_ATOMIC alloc fails. If the request cannot be satisfied within
481the current tasks cpuset, then we relax the cpuset, and look for
482memory anywhere we can find it. It's better to violate the cpuset
483than stress the kernel.
484
485To start a new job that is to be contained within a cpuset, the steps are:
486
487 1) mkdir /dev/cpuset
488 2) mount -t cpuset none /dev/cpuset
489 3) Create the new cpuset by doing mkdir's and write's (or echo's) in
490 the /dev/cpuset virtual file system.
491 4) Start a task that will be the "founding father" of the new job.
492 5) Attach that task to the new cpuset by writing its pid to the
493 /dev/cpuset tasks file for that cpuset.
494 6) fork, exec or clone the job tasks from this founding father task.
495
496For example, the following sequence of commands will setup a cpuset
497named "Charlie", containing just CPUs 2 and 3, and Memory Node 1,
498and then start a subshell 'sh' in that cpuset:
499
500 mount -t cpuset none /dev/cpuset
501 cd /dev/cpuset
502 mkdir Charlie
503 cd Charlie
504 /bin/echo 2-3 > cpus
505 /bin/echo 1 > mems
506 /bin/echo $$ > tasks
507 sh
508 # The subshell 'sh' is now running in cpuset Charlie
509 # The next line should display '/Charlie'
510 cat /proc/self/cpuset
511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512In the future, a C library interface to cpusets will likely be
513available. For now, the only way to query or modify cpusets is
514via the cpuset file system, using the various cd, mkdir, echo, cat,
515rmdir commands from the shell, or their equivalent from C.
516
517The sched_setaffinity calls can also be done at the shell prompt using
518SGI's runon or Robert Love's taskset. The mbind and set_mempolicy
519calls can be done at the shell prompt using the numactl command
520(part of Andi Kleen's numa package).
521
5222. Usage Examples and Syntax
523============================
524
5252.1 Basic Usage
526---------------
527
528Creating, modifying, using the cpusets can be done through the cpuset
529virtual filesystem.
530
531To mount it, type:
532# mount -t cpuset none /dev/cpuset
533
534Then under /dev/cpuset you can find a tree that corresponds to the
535tree of the cpusets in the system. For instance, /dev/cpuset
536is the cpuset that holds the whole system.
537
538If you want to create a new cpuset under /dev/cpuset:
539# cd /dev/cpuset
540# mkdir my_cpuset
541
542Now you want to do something with this cpuset.
543# cd my_cpuset
544
545In this directory you can find several files:
546# ls
547cpus cpu_exclusive mems mem_exclusive tasks
548
549Reading them will give you information about the state of this cpuset:
550the CPUs and Memory Nodes it can use, the processes that are using
551it, its properties. By writing to these files you can manipulate
552the cpuset.
553
554Set some flags:
555# /bin/echo 1 > cpu_exclusive
556
557Add some cpus:
558# /bin/echo 0-7 > cpus
559
560Now attach your shell to this cpuset:
561# /bin/echo $$ > tasks
562
563You can also create cpusets inside your cpuset by using mkdir in this
564directory.
565# mkdir my_sub_cs
566
567To remove a cpuset, just use rmdir:
568# rmdir my_sub_cs
569This will fail if the cpuset is in use (has cpusets inside, or has
570processes attached).
571
5722.2 Adding/removing cpus
573------------------------
574
575This is the syntax to use when writing in the cpus or mems files
576in cpuset directories:
577
578# /bin/echo 1-4 > cpus -> set cpus list to cpus 1,2,3,4
579# /bin/echo 1,2,3,4 > cpus -> set cpus list to cpus 1,2,3,4
580
5812.3 Setting flags
582-----------------
583
584The syntax is very simple:
585
586# /bin/echo 1 > cpu_exclusive -> set flag 'cpu_exclusive'
587# /bin/echo 0 > cpu_exclusive -> unset flag 'cpu_exclusive'
588
5892.4 Attaching processes
590-----------------------
591
592# /bin/echo PID > tasks
593
594Note that it is PID, not PIDs. You can only attach ONE task at a time.
595If you have several tasks to attach, you have to do it one after another:
596
597# /bin/echo PID1 > tasks
598# /bin/echo PID2 > tasks
599 ...
600# /bin/echo PIDn > tasks
601
602
6033. Questions
604============
605
606Q: what's up with this '/bin/echo' ?
607A: bash's builtin 'echo' command does not check calls to write() against
608 errors. If you use it in the cpuset file system, you won't be
609 able to tell whether a command succeeded or failed.
610
611Q: When I attach processes, only the first of the line gets really attached !
612A: We can only return one error code per call to write(). So you should also
613 put only ONE pid.
614
6154. Contact
616==========
617
618Web: http://www.bullopensource.org/cpuset