1================== 2NUMA Memory Policy 3================== 4 5What is NUMA Memory Policy? 6============================ 7 8In the Linux kernel, "memory policy" determines from which node the kernel will 9allocate memory in a NUMA system or in an emulated NUMA system. Linux has 10supported platforms with Non-Uniform Memory Access architectures since 2.4.?. 11The current memory policy support was added to Linux 2.6 around May 2004. This 12document attempts to describe the concepts and APIs of the 2.6 memory policy 13support. 14 15Memory policies should not be confused with cpusets 16(``Documentation/admin-guide/cgroup-v1/cpusets.rst``) 17which is an administrative mechanism for restricting the nodes from which 18memory may be allocated by a set of processes. Memory policies are a 19programming interface that a NUMA-aware application can take advantage of. When 20both cpusets and policies are applied to a task, the restrictions of the cpuset 21takes priority. See :ref:`Memory Policies and cpusets <mem_pol_and_cpusets>` 22below for more details. 23 24Memory Policy Concepts 25====================== 26 27Scope of Memory Policies 28------------------------ 29 30The Linux kernel supports _scopes_ of memory policy, described here from 31most general to most specific: 32 33System Default Policy 34 this policy is "hard coded" into the kernel. It is the policy 35 that governs all page allocations that aren't controlled by 36 one of the more specific policy scopes discussed below. When 37 the system is "up and running", the system default policy will 38 use "local allocation" described below. However, during boot 39 up, the system default policy will be set to interleave 40 allocations across all nodes with "sufficient" memory, so as 41 not to overload the initial boot node with boot-time 42 allocations. 43 44Task/Process Policy 45 this is an optional, per-task policy. When defined for a 46 specific task, this policy controls all page allocations made 47 by or on behalf of the task that aren't controlled by a more 48 specific scope. If a task does not define a task policy, then 49 all page allocations that would have been controlled by the 50 task policy "fall back" to the System Default Policy. 51 52 The task policy applies to the entire address space of a task. Thus, 53 it is inheritable, and indeed is inherited, across both fork() 54 [clone() w/o the CLONE_VM flag] and exec*(). This allows a parent task 55 to establish the task policy for a child task exec()'d from an 56 executable image that has no awareness of memory policy. See the 57 :ref:`Memory Policy APIs <memory_policy_apis>` section, 58 below, for an overview of the system call 59 that a task may use to set/change its task/process policy. 60 61 In a multi-threaded task, task policies apply only to the thread 62 [Linux kernel task] that installs the policy and any threads 63 subsequently created by that thread. Any sibling threads existing 64 at the time a new task policy is installed retain their current 65 policy. 66 67 A task policy applies only to pages allocated after the policy is 68 installed. Any pages already faulted in by the task when the task 69 changes its task policy remain where they were allocated based on 70 the policy at the time they were allocated. 71 72.. _vma_policy: 73 74VMA Policy 75 A "VMA" or "Virtual Memory Area" refers to a range of a task's 76 virtual address space. A task may define a specific policy for a range 77 of its virtual address space. See the 78 :ref:`Memory Policy APIs <memory_policy_apis>` section, 79 below, for an overview of the mbind() system call used to set a VMA 80 policy. 81 82 A VMA policy will govern the allocation of pages that back 83 this region of the address space. Any regions of the task's 84 address space that don't have an explicit VMA policy will fall 85 back to the task policy, which may itself fall back to the 86 System Default Policy. 87 88 VMA policies have a few complicating details: 89 90 * VMA policy applies ONLY to anonymous pages. These include 91 pages allocated for anonymous segments, such as the task 92 stack and heap, and any regions of the address space 93 mmap()ed with the MAP_ANONYMOUS flag. If a VMA policy is 94 applied to a file mapping, it will be ignored if the mapping 95 used the MAP_SHARED flag. If the file mapping used the 96 MAP_PRIVATE flag, the VMA policy will only be applied when 97 an anonymous page is allocated on an attempt to write to the 98 mapping-- i.e., at Copy-On-Write. 99 100 * VMA policies are shared between all tasks that share a 101 virtual address space--a.k.a. threads--independent of when 102 the policy is installed; and they are inherited across 103 fork(). However, because VMA policies refer to a specific 104 region of a task's address space, and because the address 105 space is discarded and recreated on exec*(), VMA policies 106 are NOT inheritable across exec(). Thus, only NUMA-aware 107 applications may use VMA policies. 108 109 * A task may install a new VMA policy on a sub-range of a 110 previously mmap()ed region. When this happens, Linux splits 111 the existing virtual memory area into 2 or 3 VMAs, each with 112 it's own policy. 113 114 * By default, VMA policy applies only to pages allocated after 115 the policy is installed. Any pages already faulted into the 116 VMA range remain where they were allocated based on the 117 policy at the time they were allocated. However, since 118 2.6.16, Linux supports page migration via the mbind() system 119 call, so that page contents can be moved to match a newly 120 installed policy. 121 122Shared Policy 123 Conceptually, shared policies apply to "memory objects" mapped 124 shared into one or more tasks' distinct address spaces. An 125 application installs shared policies the same way as VMA 126 policies--using the mbind() system call specifying a range of 127 virtual addresses that map the shared object. However, unlike 128 VMA policies, which can be considered to be an attribute of a 129 range of a task's address space, shared policies apply 130 directly to the shared object. Thus, all tasks that attach to 131 the object share the policy, and all pages allocated for the 132 shared object, by any task, will obey the shared policy. 133 134 As of 2.6.22, only shared memory segments, created by shmget() or 135 mmap(MAP_ANONYMOUS|MAP_SHARED), support shared policy. When shared 136 policy support was added to Linux, the associated data structures were 137 added to hugetlbfs shmem segments. At the time, hugetlbfs did not 138 support allocation at fault time--a.k.a lazy allocation--so hugetlbfs 139 shmem segments were never "hooked up" to the shared policy support. 140 Although hugetlbfs segments now support lazy allocation, their support 141 for shared policy has not been completed. 142 143 As mentioned above in :ref:`VMA policies <vma_policy>` section, 144 allocations of page cache pages for regular files mmap()ed 145 with MAP_SHARED ignore any VMA policy installed on the virtual 146 address range backed by the shared file mapping. Rather, 147 shared page cache pages, including pages backing private 148 mappings that have not yet been written by the task, follow 149 task policy, if any, else System Default Policy. 150 151 The shared policy infrastructure supports different policies on subset 152 ranges of the shared object. However, Linux still splits the VMA of 153 the task that installs the policy for each range of distinct policy. 154 Thus, different tasks that attach to a shared memory segment can have 155 different VMA configurations mapping that one shared object. This 156 can be seen by examining the /proc/<pid>/numa_maps of tasks sharing 157 a shared memory region, when one task has installed shared policy on 158 one or more ranges of the region. 159 160Components of Memory Policies 161----------------------------- 162 163A NUMA memory policy consists of a "mode", optional mode flags, and 164an optional set of nodes. The mode determines the behavior of the 165policy, the optional mode flags determine the behavior of the mode, 166and the optional set of nodes can be viewed as the arguments to the 167policy behavior. 168 169Internally, memory policies are implemented by a reference counted 170structure, struct mempolicy. Details of this structure will be 171discussed in context, below, as required to explain the behavior. 172 173NUMA memory policy supports the following 4 behavioral modes: 174 175Default Mode--MPOL_DEFAULT 176 This mode is only used in the memory policy APIs. Internally, 177 MPOL_DEFAULT is converted to the NULL memory policy in all 178 policy scopes. Any existing non-default policy will simply be 179 removed when MPOL_DEFAULT is specified. As a result, 180 MPOL_DEFAULT means "fall back to the next most specific policy 181 scope." 182 183 For example, a NULL or default task policy will fall back to the 184 system default policy. A NULL or default vma policy will fall 185 back to the task policy. 186 187 When specified in one of the memory policy APIs, the Default mode 188 does not use the optional set of nodes. 189 190 It is an error for the set of nodes specified for this policy to 191 be non-empty. 192 193MPOL_BIND 194 This mode specifies that memory must come from the set of 195 nodes specified by the policy. Memory will be allocated from 196 the node in the set with sufficient free memory that is 197 closest to the node where the allocation takes place. 198 199MPOL_PREFERRED 200 This mode specifies that the allocation should be attempted 201 from the single node specified in the policy. If that 202 allocation fails, the kernel will search other nodes, in order 203 of increasing distance from the preferred node based on 204 information provided by the platform firmware. 205 206 Internally, the Preferred policy uses a single node--the 207 preferred_node member of struct mempolicy. When the internal 208 mode flag MPOL_F_LOCAL is set, the preferred_node is ignored 209 and the policy is interpreted as local allocation. "Local" 210 allocation policy can be viewed as a Preferred policy that 211 starts at the node containing the cpu where the allocation 212 takes place. 213 214 It is possible for the user to specify that local allocation 215 is always preferred by passing an empty nodemask with this 216 mode. If an empty nodemask is passed, the policy cannot use 217 the MPOL_F_STATIC_NODES or MPOL_F_RELATIVE_NODES flags 218 described below. 219 220MPOL_INTERLEAVED 221 This mode specifies that page allocations be interleaved, on a 222 page granularity, across the nodes specified in the policy. 223 This mode also behaves slightly differently, based on the 224 context where it is used: 225 226 For allocation of anonymous pages and shared memory pages, 227 Interleave mode indexes the set of nodes specified by the 228 policy using the page offset of the faulting address into the 229 segment [VMA] containing the address modulo the number of 230 nodes specified by the policy. It then attempts to allocate a 231 page, starting at the selected node, as if the node had been 232 specified by a Preferred policy or had been selected by a 233 local allocation. That is, allocation will follow the per 234 node zonelist. 235 236 For allocation of page cache pages, Interleave mode indexes 237 the set of nodes specified by the policy using a node counter 238 maintained per task. This counter wraps around to the lowest 239 specified node after it reaches the highest specified node. 240 This will tend to spread the pages out over the nodes 241 specified by the policy based on the order in which they are 242 allocated, rather than based on any page offset into an 243 address range or file. During system boot up, the temporary 244 interleaved system default policy works in this mode. 245 246MPOL_PREFERRED_MANY 247 This mode specifies that the allocation should be preferably 248 satisfied from the nodemask specified in the policy. If there is 249 a memory pressure on all nodes in the nodemask, the allocation 250 can fall back to all existing numa nodes. This is effectively 251 MPOL_PREFERRED allowed for a mask rather than a single node. 252 253NUMA memory policy supports the following optional mode flags: 254 255MPOL_F_STATIC_NODES 256 This flag specifies that the nodemask passed by 257 the user should not be remapped if the task or VMA's set of allowed 258 nodes changes after the memory policy has been defined. 259 260 Without this flag, any time a mempolicy is rebound because of a 261 change in the set of allowed nodes, the preferred nodemask (Preferred 262 Many), preferred node (Preferred) or nodemask (Bind, Interleave) is 263 remapped to the new set of allowed nodes. This may result in nodes 264 being used that were previously undesired. 265 266 With this flag, if the user-specified nodes overlap with the 267 nodes allowed by the task's cpuset, then the memory policy is 268 applied to their intersection. If the two sets of nodes do not 269 overlap, the Default policy is used. 270 271 For example, consider a task that is attached to a cpuset with 272 mems 1-3 that sets an Interleave policy over the same set. If 273 the cpuset's mems change to 3-5, the Interleave will now occur 274 over nodes 3, 4, and 5. With this flag, however, since only node 275 3 is allowed from the user's nodemask, the "interleave" only 276 occurs over that node. If no nodes from the user's nodemask are 277 now allowed, the Default behavior is used. 278 279 MPOL_F_STATIC_NODES cannot be combined with the 280 MPOL_F_RELATIVE_NODES flag. It also cannot be used for 281 MPOL_PREFERRED policies that were created with an empty nodemask 282 (local allocation). 283 284MPOL_F_RELATIVE_NODES 285 This flag specifies that the nodemask passed 286 by the user will be mapped relative to the set of the task or VMA's 287 set of allowed nodes. The kernel stores the user-passed nodemask, 288 and if the allowed nodes changes, then that original nodemask will 289 be remapped relative to the new set of allowed nodes. 290 291 Without this flag (and without MPOL_F_STATIC_NODES), anytime a 292 mempolicy is rebound because of a change in the set of allowed 293 nodes, the node (Preferred) or nodemask (Bind, Interleave) is 294 remapped to the new set of allowed nodes. That remap may not 295 preserve the relative nature of the user's passed nodemask to its 296 set of allowed nodes upon successive rebinds: a nodemask of 297 1,3,5 may be remapped to 7-9 and then to 1-3 if the set of 298 allowed nodes is restored to its original state. 299 300 With this flag, the remap is done so that the node numbers from 301 the user's passed nodemask are relative to the set of allowed 302 nodes. In other words, if nodes 0, 2, and 4 are set in the user's 303 nodemask, the policy will be effected over the first (and in the 304 Bind or Interleave case, the third and fifth) nodes in the set of 305 allowed nodes. The nodemask passed by the user represents nodes 306 relative to task or VMA's set of allowed nodes. 307 308 If the user's nodemask includes nodes that are outside the range 309 of the new set of allowed nodes (for example, node 5 is set in 310 the user's nodemask when the set of allowed nodes is only 0-3), 311 then the remap wraps around to the beginning of the nodemask and, 312 if not already set, sets the node in the mempolicy nodemask. 313 314 For example, consider a task that is attached to a cpuset with 315 mems 2-5 that sets an Interleave policy over the same set with 316 MPOL_F_RELATIVE_NODES. If the cpuset's mems change to 3-7, the 317 interleave now occurs over nodes 3,5-7. If the cpuset's mems 318 then change to 0,2-3,5, then the interleave occurs over nodes 319 0,2-3,5. 320 321 Thanks to the consistent remapping, applications preparing 322 nodemasks to specify memory policies using this flag should 323 disregard their current, actual cpuset imposed memory placement 324 and prepare the nodemask as if they were always located on 325 memory nodes 0 to N-1, where N is the number of memory nodes the 326 policy is intended to manage. Let the kernel then remap to the 327 set of memory nodes allowed by the task's cpuset, as that may 328 change over time. 329 330 MPOL_F_RELATIVE_NODES cannot be combined with the 331 MPOL_F_STATIC_NODES flag. It also cannot be used for 332 MPOL_PREFERRED policies that were created with an empty nodemask 333 (local allocation). 334 335Memory Policy Reference Counting 336================================ 337 338To resolve use/free races, struct mempolicy contains an atomic reference 339count field. Internal interfaces, mpol_get()/mpol_put() increment and 340decrement this reference count, respectively. mpol_put() will only free 341the structure back to the mempolicy kmem cache when the reference count 342goes to zero. 343 344When a new memory policy is allocated, its reference count is initialized 345to '1', representing the reference held by the task that is installing the 346new policy. When a pointer to a memory policy structure is stored in another 347structure, another reference is added, as the task's reference will be dropped 348on completion of the policy installation. 349 350During run-time "usage" of the policy, we attempt to minimize atomic operations 351on the reference count, as this can lead to cache lines bouncing between cpus 352and NUMA nodes. "Usage" here means one of the following: 353 3541) querying of the policy, either by the task itself [using the get_mempolicy() 355 API discussed below] or by another task using the /proc/<pid>/numa_maps 356 interface. 357 3582) examination of the policy to determine the policy mode and associated node 359 or node lists, if any, for page allocation. This is considered a "hot 360 path". Note that for MPOL_BIND, the "usage" extends across the entire 361 allocation process, which may sleep during page reclamation, because the 362 BIND policy nodemask is used, by reference, to filter ineligible nodes. 363 364We can avoid taking an extra reference during the usages listed above as 365follows: 366 3671) we never need to get/free the system default policy as this is never 368 changed nor freed, once the system is up and running. 369 3702) for querying the policy, we do not need to take an extra reference on the 371 target task's task policy nor vma policies because we always acquire the 372 task's mm's mmap_lock for read during the query. The set_mempolicy() and 373 mbind() APIs [see below] always acquire the mmap_lock for write when 374 installing or replacing task or vma policies. Thus, there is no possibility 375 of a task or thread freeing a policy while another task or thread is 376 querying it. 377 3783) Page allocation usage of task or vma policy occurs in the fault path where 379 we hold them mmap_lock for read. Again, because replacing the task or vma 380 policy requires that the mmap_lock be held for write, the policy can't be 381 freed out from under us while we're using it for page allocation. 382 3834) Shared policies require special consideration. One task can replace a 384 shared memory policy while another task, with a distinct mmap_lock, is 385 querying or allocating a page based on the policy. To resolve this 386 potential race, the shared policy infrastructure adds an extra reference 387 to the shared policy during lookup while holding a spin lock on the shared 388 policy management structure. This requires that we drop this extra 389 reference when we're finished "using" the policy. We must drop the 390 extra reference on shared policies in the same query/allocation paths 391 used for non-shared policies. For this reason, shared policies are marked 392 as such, and the extra reference is dropped "conditionally"--i.e., only 393 for shared policies. 394 395 Because of this extra reference counting, and because we must lookup 396 shared policies in a tree structure under spinlock, shared policies are 397 more expensive to use in the page allocation path. This is especially 398 true for shared policies on shared memory regions shared by tasks running 399 on different NUMA nodes. This extra overhead can be avoided by always 400 falling back to task or system default policy for shared memory regions, 401 or by prefaulting the entire shared memory region into memory and locking 402 it down. However, this might not be appropriate for all applications. 403 404.. _memory_policy_apis: 405 406Memory Policy APIs 407================== 408 409Linux supports 4 system calls for controlling memory policy. These APIS 410always affect only the calling task, the calling task's address space, or 411some shared object mapped into the calling task's address space. 412 413.. note:: 414 the headers that define these APIs and the parameter data types for 415 user space applications reside in a package that is not part of the 416 Linux kernel. The kernel system call interfaces, with the 'sys\_' 417 prefix, are defined in <linux/syscalls.h>; the mode and flag 418 definitions are defined in <linux/mempolicy.h>. 419 420Set [Task] Memory Policy:: 421 422 long set_mempolicy(int mode, const unsigned long *nmask, 423 unsigned long maxnode); 424 425Set's the calling task's "task/process memory policy" to mode 426specified by the 'mode' argument and the set of nodes defined by 427'nmask'. 'nmask' points to a bit mask of node ids containing at least 428'maxnode' ids. Optional mode flags may be passed by combining the 429'mode' argument with the flag (for example: MPOL_INTERLEAVE | 430MPOL_F_STATIC_NODES). 431 432See the set_mempolicy(2) man page for more details 433 434 435Get [Task] Memory Policy or Related Information:: 436 437 long get_mempolicy(int *mode, 438 const unsigned long *nmask, unsigned long maxnode, 439 void *addr, int flags); 440 441Queries the "task/process memory policy" of the calling task, or the 442policy or location of a specified virtual address, depending on the 443'flags' argument. 444 445See the get_mempolicy(2) man page for more details 446 447 448Install VMA/Shared Policy for a Range of Task's Address Space:: 449 450 long mbind(void *start, unsigned long len, int mode, 451 const unsigned long *nmask, unsigned long maxnode, 452 unsigned flags); 453 454mbind() installs the policy specified by (mode, nmask, maxnodes) as a 455VMA policy for the range of the calling task's address space specified 456by the 'start' and 'len' arguments. Additional actions may be 457requested via the 'flags' argument. 458 459See the mbind(2) man page for more details. 460 461Set home node for a Range of Task's Address Spacec:: 462 463 long sys_set_mempolicy_home_node(unsigned long start, unsigned long len, 464 unsigned long home_node, 465 unsigned long flags); 466 467sys_set_mempolicy_home_node set the home node for a VMA policy present in the 468task's address range. The system call updates the home node only for the existing 469mempolicy range. Other address ranges are ignored. A home node is the NUMA node 470closest to which page allocation will come from. Specifying the home node override 471the default allocation policy to allocate memory close to the local node for an 472executing CPU. 473 474 475Memory Policy Command Line Interface 476==================================== 477 478Although not strictly part of the Linux implementation of memory policy, 479a command line tool, numactl(8), exists that allows one to: 480 481+ set the task policy for a specified program via set_mempolicy(2), fork(2) and 482 exec(2) 483 484+ set the shared policy for a shared memory segment via mbind(2) 485 486The numactl(8) tool is packaged with the run-time version of the library 487containing the memory policy system call wrappers. Some distributions 488package the headers and compile-time libraries in a separate development 489package. 490 491.. _mem_pol_and_cpusets: 492 493Memory Policies and cpusets 494=========================== 495 496Memory policies work within cpusets as described above. For memory policies 497that require a node or set of nodes, the nodes are restricted to the set of 498nodes whose memories are allowed by the cpuset constraints. If the nodemask 499specified for the policy contains nodes that are not allowed by the cpuset and 500MPOL_F_RELATIVE_NODES is not used, the intersection of the set of nodes 501specified for the policy and the set of nodes with memory is used. If the 502result is the empty set, the policy is considered invalid and cannot be 503installed. If MPOL_F_RELATIVE_NODES is used, the policy's nodes are mapped 504onto and folded into the task's set of allowed nodes as previously described. 505 506The interaction of memory policies and cpusets can be problematic when tasks 507in two cpusets share access to a memory region, such as shared memory segments 508created by shmget() of mmap() with the MAP_ANONYMOUS and MAP_SHARED flags, and 509any of the tasks install shared policy on the region, only nodes whose 510memories are allowed in both cpusets may be used in the policies. Obtaining 511this information requires "stepping outside" the memory policy APIs to use the 512cpuset information and requires that one know in what cpusets other task might 513be attaching to the shared region. Furthermore, if the cpusets' allowed 514memory sets are disjoint, "local" allocation is the only valid policy. 515