1.. _unevictable_lru: 2 3============================== 4Unevictable LRU Infrastructure 5============================== 6 7.. contents:: :local: 8 9 10Introduction 11============ 12 13This document describes the Linux memory manager's "Unevictable LRU" 14infrastructure and the use of this to manage several types of "unevictable" 15pages. 16 17The document attempts to provide the overall rationale behind this mechanism 18and the rationale for some of the design decisions that drove the 19implementation. The latter design rationale is discussed in the context of an 20implementation description. Admittedly, one can obtain the implementation 21details - the "what does it do?" - by reading the code. One hopes that the 22descriptions below add value by provide the answer to "why does it do that?". 23 24 25 26The Unevictable LRU 27=================== 28 29The Unevictable LRU facility adds an additional LRU list to track unevictable 30pages and to hide these pages from vmscan. This mechanism is based on a patch 31by Larry Woodman of Red Hat to address several scalability problems with page 32reclaim in Linux. The problems have been observed at customer sites on large 33memory x86_64 systems. 34 35To illustrate this with an example, a non-NUMA x86_64 platform with 128GB of 36main memory will have over 32 million 4k pages in a single zone. When a large 37fraction of these pages are not evictable for any reason [see below], vmscan 38will spend a lot of time scanning the LRU lists looking for the small fraction 39of pages that are evictable. This can result in a situation where all CPUs are 40spending 100% of their time in vmscan for hours or days on end, with the system 41completely unresponsive. 42 43The unevictable list addresses the following classes of unevictable pages: 44 45 * Those owned by ramfs. 46 47 * Those mapped into SHM_LOCK'd shared memory regions. 48 49 * Those mapped into VM_LOCKED [mlock()ed] VMAs. 50 51The infrastructure may also be able to handle other conditions that make pages 52unevictable, either by definition or by circumstance, in the future. 53 54 55The Unevictable Page List 56------------------------- 57 58The Unevictable LRU infrastructure consists of an additional, per-zone, LRU list 59called the "unevictable" list and an associated page flag, PG_unevictable, to 60indicate that the page is being managed on the unevictable list. 61 62The PG_unevictable flag is analogous to, and mutually exclusive with, the 63PG_active flag in that it indicates on which LRU list a page resides when 64PG_lru is set. 65 66The Unevictable LRU infrastructure maintains unevictable pages on an additional 67LRU list for a few reasons: 68 69 (1) We get to "treat unevictable pages just like we treat other pages in the 70 system - which means we get to use the same code to manipulate them, the 71 same code to isolate them (for migrate, etc.), the same code to keep track 72 of the statistics, etc..." [Rik van Riel] 73 74 (2) We want to be able to migrate unevictable pages between nodes for memory 75 defragmentation, workload management and memory hotplug. The linux kernel 76 can only migrate pages that it can successfully isolate from the LRU 77 lists. If we were to maintain pages elsewhere than on an LRU-like list, 78 where they can be found by isolate_lru_page(), we would prevent their 79 migration, unless we reworked migration code to find the unevictable pages 80 itself. 81 82 83The unevictable list does not differentiate between file-backed and anonymous, 84swap-backed pages. This differentiation is only important while the pages are, 85in fact, evictable. 86 87The unevictable list benefits from the "arrayification" of the per-zone LRU 88lists and statistics originally proposed and posted by Christoph Lameter. 89 90The unevictable list does not use the LRU pagevec mechanism. Rather, 91unevictable pages are placed directly on the page's zone's unevictable list 92under the zone lru_lock. This allows us to prevent the stranding of pages on 93the unevictable list when one task has the page isolated from the LRU and other 94tasks are changing the "evictability" state of the page. 95 96 97Memory Control Group Interaction 98-------------------------------- 99 100The unevictable LRU facility interacts with the memory control group [aka 101memory controller; see Documentation/admin-guide/cgroup-v1/memory.rst] by extending the 102lru_list enum. 103 104The memory controller data structure automatically gets a per-zone unevictable 105list as a result of the "arrayification" of the per-zone LRU lists (one per 106lru_list enum element). The memory controller tracks the movement of pages to 107and from the unevictable list. 108 109When a memory control group comes under memory pressure, the controller will 110not attempt to reclaim pages on the unevictable list. This has a couple of 111effects: 112 113 (1) Because the pages are "hidden" from reclaim on the unevictable list, the 114 reclaim process can be more efficient, dealing only with pages that have a 115 chance of being reclaimed. 116 117 (2) On the other hand, if too many of the pages charged to the control group 118 are unevictable, the evictable portion of the working set of the tasks in 119 the control group may not fit into the available memory. This can cause 120 the control group to thrash or to OOM-kill tasks. 121 122 123.. _mark_addr_space_unevict: 124 125Marking Address Spaces Unevictable 126---------------------------------- 127 128For facilities such as ramfs none of the pages attached to the address space 129may be evicted. To prevent eviction of any such pages, the AS_UNEVICTABLE 130address space flag is provided, and this can be manipulated by a filesystem 131using a number of wrapper functions: 132 133 * ``void mapping_set_unevictable(struct address_space *mapping);`` 134 135 Mark the address space as being completely unevictable. 136 137 * ``void mapping_clear_unevictable(struct address_space *mapping);`` 138 139 Mark the address space as being evictable. 140 141 * ``int mapping_unevictable(struct address_space *mapping);`` 142 143 Query the address space, and return true if it is completely 144 unevictable. 145 146These are currently used in three places in the kernel: 147 148 (1) By ramfs to mark the address spaces of its inodes when they are created, 149 and this mark remains for the life of the inode. 150 151 (2) By SYSV SHM to mark SHM_LOCK'd address spaces until SHM_UNLOCK is called. 152 153 Note that SHM_LOCK is not required to page in the locked pages if they're 154 swapped out; the application must touch the pages manually if it wants to 155 ensure they're in memory. 156 157 (3) By the i915 driver to mark pinned address space until it's unpinned. The 158 amount of unevictable memory marked by i915 driver is roughly the bounded 159 object size in debugfs/dri/0/i915_gem_objects. 160 161 162Detecting Unevictable Pages 163--------------------------- 164 165The function page_evictable() in vmscan.c determines whether a page is 166evictable or not using the query function outlined above [see section 167:ref:`Marking address spaces unevictable <mark_addr_space_unevict>`] 168to check the AS_UNEVICTABLE flag. 169 170For address spaces that are so marked after being populated (as SHM regions 171might be), the lock action (eg: SHM_LOCK) can be lazy, and need not populate 172the page tables for the region as does, for example, mlock(), nor need it make 173any special effort to push any pages in the SHM_LOCK'd area to the unevictable 174list. Instead, vmscan will do this if and when it encounters the pages during 175a reclamation scan. 176 177On an unlock action (such as SHM_UNLOCK), the unlocker (eg: shmctl()) must scan 178the pages in the region and "rescue" them from the unevictable list if no other 179condition is keeping them unevictable. If an unevictable region is destroyed, 180the pages are also "rescued" from the unevictable list in the process of 181freeing them. 182 183page_evictable() also checks for mlocked pages by testing an additional page 184flag, PG_mlocked (as wrapped by PageMlocked()), which is set when a page is 185faulted into a VM_LOCKED vma, or found in a vma being VM_LOCKED. 186 187 188Vmscan's Handling of Unevictable Pages 189-------------------------------------- 190 191If unevictable pages are culled in the fault path, or moved to the unevictable 192list at mlock() or mmap() time, vmscan will not encounter the pages until they 193have become evictable again (via munlock() for example) and have been "rescued" 194from the unevictable list. However, there may be situations where we decide, 195for the sake of expediency, to leave a unevictable page on one of the regular 196active/inactive LRU lists for vmscan to deal with. vmscan checks for such 197pages in all of the shrink_{active|inactive|page}_list() functions and will 198"cull" such pages that it encounters: that is, it diverts those pages to the 199unevictable list for the zone being scanned. 200 201There may be situations where a page is mapped into a VM_LOCKED VMA, but the 202page is not marked as PG_mlocked. Such pages will make it all the way to 203shrink_page_list() where they will be detected when vmscan walks the reverse 204map in try_to_unmap(). If try_to_unmap() returns SWAP_MLOCK, 205shrink_page_list() will cull the page at that point. 206 207To "cull" an unevictable page, vmscan simply puts the page back on the LRU list 208using putback_lru_page() - the inverse operation to isolate_lru_page() - after 209dropping the page lock. Because the condition which makes the page unevictable 210may change once the page is unlocked, putback_lru_page() will recheck the 211unevictable state of a page that it places on the unevictable list. If the 212page has become unevictable, putback_lru_page() removes it from the list and 213retries, including the page_unevictable() test. Because such a race is a rare 214event and movement of pages onto the unevictable list should be rare, these 215extra evictabilty checks should not occur in the majority of calls to 216putback_lru_page(). 217 218 219MLOCKED Pages 220============= 221 222The unevictable page list is also useful for mlock(), in addition to ramfs and 223SYSV SHM. Note that mlock() is only available in CONFIG_MMU=y situations; in 224NOMMU situations, all mappings are effectively mlocked. 225 226 227History 228------- 229 230The "Unevictable mlocked Pages" infrastructure is based on work originally 231posted by Nick Piggin in an RFC patch entitled "mm: mlocked pages off LRU". 232Nick posted his patch as an alternative to a patch posted by Christoph Lameter 233to achieve the same objective: hiding mlocked pages from vmscan. 234 235In Nick's patch, he used one of the struct page LRU list link fields as a count 236of VM_LOCKED VMAs that map the page. This use of the link field for a count 237prevented the management of the pages on an LRU list, and thus mlocked pages 238were not migratable as isolate_lru_page() could not find them, and the LRU list 239link field was not available to the migration subsystem. 240 241Nick resolved this by putting mlocked pages back on the lru list before 242attempting to isolate them, thus abandoning the count of VM_LOCKED VMAs. When 243Nick's patch was integrated with the Unevictable LRU work, the count was 244replaced by walking the reverse map to determine whether any VM_LOCKED VMAs 245mapped the page. More on this below. 246 247 248Basic Management 249---------------- 250 251mlocked pages - pages mapped into a VM_LOCKED VMA - are a class of unevictable 252pages. When such a page has been "noticed" by the memory management subsystem, 253the page is marked with the PG_mlocked flag. This can be manipulated using the 254PageMlocked() functions. 255 256A PG_mlocked page will be placed on the unevictable list when it is added to 257the LRU. Such pages can be "noticed" by memory management in several places: 258 259 (1) in the mlock()/mlockall() system call handlers; 260 261 (2) in the mmap() system call handler when mmapping a region with the 262 MAP_LOCKED flag; 263 264 (3) mmapping a region in a task that has called mlockall() with the MCL_FUTURE 265 flag 266 267 (4) in the fault path, if mlocked pages are "culled" in the fault path, 268 and when a VM_LOCKED stack segment is expanded; or 269 270 (5) as mentioned above, in vmscan:shrink_page_list() when attempting to 271 reclaim a page in a VM_LOCKED VMA via try_to_unmap() 272 273all of which result in the VM_LOCKED flag being set for the VMA if it doesn't 274already have it set. 275 276mlocked pages become unlocked and rescued from the unevictable list when: 277 278 (1) mapped in a range unlocked via the munlock()/munlockall() system calls; 279 280 (2) munmap()'d out of the last VM_LOCKED VMA that maps the page, including 281 unmapping at task exit; 282 283 (3) when the page is truncated from the last VM_LOCKED VMA of an mmapped file; 284 or 285 286 (4) before a page is COW'd in a VM_LOCKED VMA. 287 288 289mlock()/mlockall() System Call Handling 290--------------------------------------- 291 292Both [do\_]mlock() and [do\_]mlockall() system call handlers call mlock_fixup() 293for each VMA in the range specified by the call. In the case of mlockall(), 294this is the entire active address space of the task. Note that mlock_fixup() 295is used for both mlocking and munlocking a range of memory. A call to mlock() 296an already VM_LOCKED VMA, or to munlock() a VMA that is not VM_LOCKED is 297treated as a no-op, and mlock_fixup() simply returns. 298 299If the VMA passes some filtering as described in "Filtering Special Vmas" 300below, mlock_fixup() will attempt to merge the VMA with its neighbors or split 301off a subset of the VMA if the range does not cover the entire VMA. Once the 302VMA has been merged or split or neither, mlock_fixup() will call 303populate_vma_page_range() to fault in the pages via get_user_pages() and to 304mark the pages as mlocked via mlock_vma_page(). 305 306Note that the VMA being mlocked might be mapped with PROT_NONE. In this case, 307get_user_pages() will be unable to fault in the pages. That's okay. If pages 308do end up getting faulted into this VM_LOCKED VMA, we'll handle them in the 309fault path or in vmscan. 310 311Also note that a page returned by get_user_pages() could be truncated or 312migrated out from under us, while we're trying to mlock it. To detect this, 313populate_vma_page_range() checks page_mapping() after acquiring the page lock. 314If the page is still associated with its mapping, we'll go ahead and call 315mlock_vma_page(). If the mapping is gone, we just unlock the page and move on. 316In the worst case, this will result in a page mapped in a VM_LOCKED VMA 317remaining on a normal LRU list without being PageMlocked(). Again, vmscan will 318detect and cull such pages. 319 320mlock_vma_page() will call TestSetPageMlocked() for each page returned by 321get_user_pages(). We use TestSetPageMlocked() because the page might already 322be mlocked by another task/VMA and we don't want to do extra work. We 323especially do not want to count an mlocked page more than once in the 324statistics. If the page was already mlocked, mlock_vma_page() need do nothing 325more. 326 327If the page was NOT already mlocked, mlock_vma_page() attempts to isolate the 328page from the LRU, as it is likely on the appropriate active or inactive list 329at that time. If the isolate_lru_page() succeeds, mlock_vma_page() will put 330back the page - by calling putback_lru_page() - which will notice that the page 331is now mlocked and divert the page to the zone's unevictable list. If 332mlock_vma_page() is unable to isolate the page from the LRU, vmscan will handle 333it later if and when it attempts to reclaim the page. 334 335 336Filtering Special VMAs 337---------------------- 338 339mlock_fixup() filters several classes of "special" VMAs: 340 3411) VMAs with VM_IO or VM_PFNMAP set are skipped entirely. The pages behind 342 these mappings are inherently pinned, so we don't need to mark them as 343 mlocked. In any case, most of the pages have no struct page in which to so 344 mark the page. Because of this, get_user_pages() will fail for these VMAs, 345 so there is no sense in attempting to visit them. 346 3472) VMAs mapping hugetlbfs page are already effectively pinned into memory. We 348 neither need nor want to mlock() these pages. However, to preserve the 349 prior behavior of mlock() - before the unevictable/mlock changes - 350 mlock_fixup() will call make_pages_present() in the hugetlbfs VMA range to 351 allocate the huge pages and populate the ptes. 352 3533) VMAs with VM_DONTEXPAND are generally userspace mappings of kernel pages, 354 such as the VDSO page, relay channel pages, etc. These pages 355 are inherently unevictable and are not managed on the LRU lists. 356 mlock_fixup() treats these VMAs the same as hugetlbfs VMAs. It calls 357 make_pages_present() to populate the ptes. 358 359Note that for all of these special VMAs, mlock_fixup() does not set the 360VM_LOCKED flag. Therefore, we won't have to deal with them later during 361munlock(), munmap() or task exit. Neither does mlock_fixup() account these 362VMAs against the task's "locked_vm". 363 364.. _munlock_munlockall_handling: 365 366munlock()/munlockall() System Call Handling 367------------------------------------------- 368 369The munlock() and munlockall() system calls are handled by the same functions - 370do_mlock[all]() - as the mlock() and mlockall() system calls with the unlock vs 371lock operation indicated by an argument. So, these system calls are also 372handled by mlock_fixup(). Again, if called for an already munlocked VMA, 373mlock_fixup() simply returns. Because of the VMA filtering discussed above, 374VM_LOCKED will not be set in any "special" VMAs. So, these VMAs will be 375ignored for munlock. 376 377If the VMA is VM_LOCKED, mlock_fixup() again attempts to merge or split off the 378specified range. The range is then munlocked via the function 379populate_vma_page_range() - the same function used to mlock a VMA range - 380passing a flag to indicate that munlock() is being performed. 381 382Because the VMA access protections could have been changed to PROT_NONE after 383faulting in and mlocking pages, get_user_pages() was unreliable for visiting 384these pages for munlocking. Because we don't want to leave pages mlocked, 385get_user_pages() was enhanced to accept a flag to ignore the permissions when 386fetching the pages - all of which should be resident as a result of previous 387mlocking. 388 389For munlock(), populate_vma_page_range() unlocks individual pages by calling 390munlock_vma_page(). munlock_vma_page() unconditionally clears the PG_mlocked 391flag using TestClearPageMlocked(). As with mlock_vma_page(), 392munlock_vma_page() use the Test*PageMlocked() function to handle the case where 393the page might have already been unlocked by another task. If the page was 394mlocked, munlock_vma_page() updates that zone statistics for the number of 395mlocked pages. Note, however, that at this point we haven't checked whether 396the page is mapped by other VM_LOCKED VMAs. 397 398We can't call try_to_munlock(), the function that walks the reverse map to 399check for other VM_LOCKED VMAs, without first isolating the page from the LRU. 400try_to_munlock() is a variant of try_to_unmap() and thus requires that the page 401not be on an LRU list [more on these below]. However, the call to 402isolate_lru_page() could fail, in which case we couldn't try_to_munlock(). So, 403we go ahead and clear PG_mlocked up front, as this might be the only chance we 404have. If we can successfully isolate the page, we go ahead and 405try_to_munlock(), which will restore the PG_mlocked flag and update the zone 406page statistics if it finds another VMA holding the page mlocked. If we fail 407to isolate the page, we'll have left a potentially mlocked page on the LRU. 408This is fine, because we'll catch it later if and if vmscan tries to reclaim 409the page. This should be relatively rare. 410 411 412Migrating MLOCKED Pages 413----------------------- 414 415A page that is being migrated has been isolated from the LRU lists and is held 416locked across unmapping of the page, updating the page's address space entry 417and copying the contents and state, until the page table entry has been 418replaced with an entry that refers to the new page. Linux supports migration 419of mlocked pages and other unevictable pages. This involves simply moving the 420PG_mlocked and PG_unevictable states from the old page to the new page. 421 422Note that page migration can race with mlocking or munlocking of the same page. 423This has been discussed from the mlock/munlock perspective in the respective 424sections above. Both processes (migration and m[un]locking) hold the page 425locked. This provides the first level of synchronization. Page migration 426zeros out the page_mapping of the old page before unlocking it, so m[un]lock 427can skip these pages by testing the page mapping under page lock. 428 429To complete page migration, we place the new and old pages back onto the LRU 430after dropping the page lock. The "unneeded" page - old page on success, new 431page on failure - will be freed when the reference count held by the migration 432process is released. To ensure that we don't strand pages on the unevictable 433list because of a race between munlock and migration, page migration uses the 434putback_lru_page() function to add migrated pages back to the LRU. 435 436 437Compacting MLOCKED Pages 438------------------------ 439 440The unevictable LRU can be scanned for compactable regions and the default 441behavior is to do so. /proc/sys/vm/compact_unevictable_allowed controls 442this behavior (see Documentation/admin-guide/sysctl/vm.rst). Once scanning of the 443unevictable LRU is enabled, the work of compaction is mostly handled by 444the page migration code and the same work flow as described in MIGRATING 445MLOCKED PAGES will apply. 446 447MLOCKING Transparent Huge Pages 448------------------------------- 449 450A transparent huge page is represented by a single entry on an LRU list. 451Therefore, we can only make unevictable an entire compound page, not 452individual subpages. 453 454If a user tries to mlock() part of a huge page, we want the rest of the 455page to be reclaimable. 456 457We cannot just split the page on partial mlock() as split_huge_page() can 458fail and new intermittent failure mode for the syscall is undesirable. 459 460We handle this by keeping PTE-mapped huge pages on normal LRU lists: the 461PMD on border of VM_LOCKED VMA will be split into PTE table. 462 463This way the huge page is accessible for vmscan. Under memory pressure the 464page will be split, subpages which belong to VM_LOCKED VMAs will be moved 465to unevictable LRU and the rest can be reclaimed. 466 467See also comment in follow_trans_huge_pmd(). 468 469mmap(MAP_LOCKED) System Call Handling 470------------------------------------- 471 472In addition the mlock()/mlockall() system calls, an application can request 473that a region of memory be mlocked supplying the MAP_LOCKED flag to the mmap() 474call. There is one important and subtle difference here, though. mmap() + mlock() 475will fail if the range cannot be faulted in (e.g. because mm_populate fails) 476and returns with ENOMEM while mmap(MAP_LOCKED) will not fail. The mmaped 477area will still have properties of the locked area - aka. pages will not get 478swapped out - but major page faults to fault memory in might still happen. 479 480Furthermore, any mmap() call or brk() call that expands the heap by a 481task that has previously called mlockall() with the MCL_FUTURE flag will result 482in the newly mapped memory being mlocked. Before the unevictable/mlock 483changes, the kernel simply called make_pages_present() to allocate pages and 484populate the page table. 485 486To mlock a range of memory under the unevictable/mlock infrastructure, the 487mmap() handler and task address space expansion functions call 488populate_vma_page_range() specifying the vma and the address range to mlock. 489 490The callers of populate_vma_page_range() will have already added the memory range 491to be mlocked to the task's "locked_vm". To account for filtered VMAs, 492populate_vma_page_range() returns the number of pages NOT mlocked. All of the 493callers then subtract a non-negative return value from the task's locked_vm. A 494negative return value represent an error - for example, from get_user_pages() 495attempting to fault in a VMA with PROT_NONE access. In this case, we leave the 496memory range accounted as locked_vm, as the protections could be changed later 497and pages allocated into that region. 498 499 500munmap()/exit()/exec() System Call Handling 501------------------------------------------- 502 503When unmapping an mlocked region of memory, whether by an explicit call to 504munmap() or via an internal unmap from exit() or exec() processing, we must 505munlock the pages if we're removing the last VM_LOCKED VMA that maps the pages. 506Before the unevictable/mlock changes, mlocking did not mark the pages in any 507way, so unmapping them required no processing. 508 509To munlock a range of memory under the unevictable/mlock infrastructure, the 510munmap() handler and task address space call tear down function 511munlock_vma_pages_all(). The name reflects the observation that one always 512specifies the entire VMA range when munlock()ing during unmap of a region. 513Because of the VMA filtering when mlocking() regions, only "normal" VMAs that 514actually contain mlocked pages will be passed to munlock_vma_pages_all(). 515 516munlock_vma_pages_all() clears the VM_LOCKED VMA flag and, like mlock_fixup() 517for the munlock case, calls __munlock_vma_pages_range() to walk the page table 518for the VMA's memory range and munlock_vma_page() each resident page mapped by 519the VMA. This effectively munlocks the page, only if this is the last 520VM_LOCKED VMA that maps the page. 521 522 523try_to_unmap() 524-------------- 525 526Pages can, of course, be mapped into multiple VMAs. Some of these VMAs may 527have VM_LOCKED flag set. It is possible for a page mapped into one or more 528VM_LOCKED VMAs not to have the PG_mlocked flag set and therefore reside on one 529of the active or inactive LRU lists. This could happen if, for example, a task 530in the process of munlocking the page could not isolate the page from the LRU. 531As a result, vmscan/shrink_page_list() might encounter such a page as described 532in section "vmscan's handling of unevictable pages". To handle this situation, 533try_to_unmap() checks for VM_LOCKED VMAs while it is walking a page's reverse 534map. 535 536try_to_unmap() is always called, by either vmscan for reclaim or for page 537migration, with the argument page locked and isolated from the LRU. Separate 538functions handle anonymous and mapped file and KSM pages, as these types of 539pages have different reverse map lookup mechanisms, with different locking. 540In each case, whether rmap_walk_anon() or rmap_walk_file() or rmap_walk_ksm(), 541it will call try_to_unmap_one() for every VMA which might contain the page. 542 543When trying to reclaim, if try_to_unmap_one() finds the page in a VM_LOCKED 544VMA, it will then mlock the page via mlock_vma_page() instead of unmapping it, 545and return SWAP_MLOCK to indicate that the page is unevictable: and the scan 546stops there. 547 548mlock_vma_page() is called while holding the page table's lock (in addition 549to the page lock, and the rmap lock): to serialize against concurrent mlock or 550munlock or munmap system calls, mm teardown (munlock_vma_pages_all), reclaim, 551holepunching, and truncation of file pages and their anonymous COWed pages. 552 553 554try_to_munlock() Reverse Map Scan 555--------------------------------- 556 557.. warning:: 558 [!] TODO/FIXME: a better name might be page_mlocked() - analogous to the 559 page_referenced() reverse map walker. 560 561When munlock_vma_page() [see section :ref:`munlock()/munlockall() System Call 562Handling <munlock_munlockall_handling>` above] tries to munlock a 563page, it needs to determine whether or not the page is mapped by any 564VM_LOCKED VMA without actually attempting to unmap all PTEs from the 565page. For this purpose, the unevictable/mlock infrastructure 566introduced a variant of try_to_unmap() called try_to_munlock(). 567 568try_to_munlock() calls the same functions as try_to_unmap() for anonymous and 569mapped file and KSM pages with a flag argument specifying unlock versus unmap 570processing. Again, these functions walk the respective reverse maps looking 571for VM_LOCKED VMAs. When such a VMA is found, as in the try_to_unmap() case, 572the functions mlock the page via mlock_vma_page() and return SWAP_MLOCK. This 573undoes the pre-clearing of the page's PG_mlocked done by munlock_vma_page. 574 575Note that try_to_munlock()'s reverse map walk must visit every VMA in a page's 576reverse map to determine that a page is NOT mapped into any VM_LOCKED VMA. 577However, the scan can terminate when it encounters a VM_LOCKED VMA. 578Although try_to_munlock() might be called a great many times when munlocking a 579large region or tearing down a large address space that has been mlocked via 580mlockall(), overall this is a fairly rare event. 581 582 583Page Reclaim in shrink_*_list() 584------------------------------- 585 586shrink_active_list() culls any obviously unevictable pages - i.e. 587!page_evictable(page) - diverting these to the unevictable list. 588However, shrink_active_list() only sees unevictable pages that made it onto the 589active/inactive lru lists. Note that these pages do not have PageUnevictable 590set - otherwise they would be on the unevictable list and shrink_active_list 591would never see them. 592 593Some examples of these unevictable pages on the LRU lists are: 594 595 (1) ramfs pages that have been placed on the LRU lists when first allocated. 596 597 (2) SHM_LOCK'd shared memory pages. shmctl(SHM_LOCK) does not attempt to 598 allocate or fault in the pages in the shared memory region. This happens 599 when an application accesses the page the first time after SHM_LOCK'ing 600 the segment. 601 602 (3) mlocked pages that could not be isolated from the LRU and moved to the 603 unevictable list in mlock_vma_page(). 604 605shrink_inactive_list() also diverts any unevictable pages that it finds on the 606inactive lists to the appropriate zone's unevictable list. 607 608shrink_inactive_list() should only see SHM_LOCK'd pages that became SHM_LOCK'd 609after shrink_active_list() had moved them to the inactive list, or pages mapped 610into VM_LOCKED VMAs that munlock_vma_page() couldn't isolate from the LRU to 611recheck via try_to_munlock(). shrink_inactive_list() won't notice the latter, 612but will pass on to shrink_page_list(). 613 614shrink_page_list() again culls obviously unevictable pages that it could 615encounter for similar reason to shrink_inactive_list(). Pages mapped into 616VM_LOCKED VMAs but without PG_mlocked set will make it all the way to 617try_to_unmap(). shrink_page_list() will divert them to the unevictable list 618when try_to_unmap() returns SWAP_MLOCK, as discussed above. 619