Lines Matching refs:the
37 This document describes the Linux memory manager's "Unevictable LRU"
38 infrastructure and the use of this to manage several types of "unevictable"
41 The document attempts to provide the overall rationale behind this mechanism
42 and the rationale for some of the design decisions that drove the
43 implementation. The latter design rationale is discussed in the context of an
44 implementation description. Admittedly, one can obtain the implementation
45 details - the "what does it do?" - by reading the code. One hopes that the
46 descriptions below add value by provide the answer to "why does it do that?".
62 will spend a lot of time scanning the LRU lists looking for the small fraction
64 spending 100% of their time in vmscan for hours or days on end, with the system
67 The unevictable list addresses the following classes of unevictable pages:
76 unevictable, either by definition or by circumstance, in the future.
83 called the "unevictable" list and an associated page flag, PG_unevictable, to
84 indicate that the page is being managed on the unevictable list.
86 The PG_unevictable flag is analogous to, and mutually exclusive with, the
93 (1) We get to "treat unevictable pages just like we treat other pages in the
94 system - which means we get to use the same code to manipulate them, the
95 same code to isolate them (for migrate, etc.), the same code to keep track
96 of the statistics, etc..." [Rik van Riel]
100 can only migrate pages that it can successfully isolate from the LRU
103 migration, unless we reworked migration code to find the unevictable pages
108 swap-backed pages. This differentiation is only important while the pages are,
111 The unevictable list benefits from the "arrayification" of the per-zone LRU
114 The unevictable list does not use the LRU pagevec mechanism. Rather,
115 unevictable pages are placed directly on the page's zone's unevictable list
116 under the zone lru_lock. This allows us to prevent the stranding of pages on
117 the unevictable list when one task has the page isolated from the LRU and other
118 tasks are changing the "evictability" state of the page.
124 The unevictable LRU facility interacts with the memory control group [aka
125 memory controller; see Documentation/cgroups/memory.txt] by extending the
129 list as a result of the "arrayification" of the per-zone LRU lists (one per
130 lru_list enum element). The memory controller tracks the movement of pages to
131 and from the unevictable list.
133 When a memory control group comes under memory pressure, the controller will
134 not attempt to reclaim pages on the unevictable list. This has a couple of
137 (1) Because the pages are "hidden" from reclaim on the unevictable list, the
141 (2) On the other hand, if too many of the pages charged to the control group
142 are unevictable, the evictable portion of the working set of the tasks in
143 the control group may not fit into the available memory. This can cause
144 the control group to thrash or to OOM-kill tasks.
150 For facilities such as ramfs none of the pages attached to the address space
151 may be evicted. To prevent eviction of any such pages, the AS_UNEVICTABLE
157 Mark the address space as being completely unevictable.
161 Mark the address space as being evictable.
165 Query the address space, and return true if it is completely
168 These are currently used in two places in the kernel:
170 (1) By ramfs to mark the address spaces of its inodes when they are created,
171 and this mark remains for the life of the inode.
175 Note that SHM_LOCK is not required to page in the locked pages if they're
176 swapped out; the application must touch the pages manually if it wants to
184 evictable or not using the query function outlined above [see section "Marking
185 address spaces unevictable"] to check the AS_UNEVICTABLE flag.
188 might be), the lock action (eg: SHM_LOCK) can be lazy, and need not populate
189 the page tables for the region as does, for example, mlock(), nor need it make
190 any special effort to push any pages in the SHM_LOCK'd area to the unevictable
191 list. Instead, vmscan will do this if and when it encounters the pages during
194 On an unlock action (such as SHM_UNLOCK), the unlocker (eg: shmctl()) must scan
195 the pages in the region and "rescue" them from the unevictable list if no other
197 the pages are also "rescued" from the unevictable list in the process of
208 If unevictable pages are culled in the fault path, or moved to the unevictable
209 list at mlock() or mmap() time, vmscan will not encounter the pages until they
211 from the unevictable list. However, there may be situations where we decide,
212 for the sake of expediency, to leave a unevictable page on one of the regular
214 pages in all of the shrink_{active|inactive|page}_list() functions and will
215 "cull" such pages that it encounters: that is, it diverts those pages to the
216 unevictable list for the zone being scanned.
218 There may be situations where a page is mapped into a VM_LOCKED VMA, but the
219 page is not marked as PG_mlocked. Such pages will make it all the way to
220 shrink_page_list() where they will be detected when vmscan walks the reverse
222 shrink_page_list() will cull the page at that point.
224 To "cull" an unevictable page, vmscan simply puts the page back on the LRU list
225 using putback_lru_page() - the inverse operation to isolate_lru_page() - after
226 dropping the page lock. Because the condition which makes the page unevictable
227 may change once the page is unlocked, putback_lru_page() will recheck the
228 unevictable state of a page that it places on the unevictable list. If the
229 page has become unevictable, putback_lru_page() removes it from the list and
230 retries, including the page_unevictable() test. Because such a race is a rare
231 event and movement of pages onto the unevictable list should be rare, these
232 extra evictabilty checks should not occur in the majority of calls to
251 to achieve the same objective: hiding mlocked pages from vmscan.
253 In Nick's patch, he used one of the struct page LRU list link fields as a count
254 of VM_LOCKED VMAs that map the page. This use of the link field for a count
255 prevented the management of the pages on an LRU list, and thus mlocked pages
256 were not migratable as isolate_lru_page() could not find them, and the LRU list
257 link field was not available to the migration subsystem.
259 Nick resolved this by putting mlocked pages back on the lru list before
260 attempting to isolate them, thus abandoning the count of VM_LOCKED VMAs. When
261 Nick's patch was integrated with the Unevictable LRU work, the count was
262 replaced by walking the reverse map to determine whether any VM_LOCKED VMAs
263 mapped the page. More on this below.
270 pages. When such a page has been "noticed" by the memory management subsystem,
271 the page is marked with the PG_mlocked flag. This can be manipulated using the
274 A PG_mlocked page will be placed on the unevictable list when it is added to
275 the LRU. Such pages can be "noticed" by memory management in several places:
277 (1) in the mlock()/mlockall() system call handlers;
279 (2) in the mmap() system call handler when mmapping a region with the
282 (3) mmapping a region in a task that has called mlockall() with the MCL_FUTURE
285 (4) in the fault path, if mlocked pages are "culled" in the fault path,
291 all of which result in the VM_LOCKED flag being set for the VMA if it doesn't
294 mlocked pages become unlocked and rescued from the unevictable list when:
296 (1) mapped in a range unlocked via the munlock()/munlockall() system calls;
298 (2) munmap()'d out of the last VM_LOCKED VMA that maps the page, including
301 (3) when the page is truncated from the last VM_LOCKED VMA of an mmapped file;
311 for each VMA in the range specified by the call. In the case of mlockall(),
312 this is the entire active address space of the task. Note that mlock_fixup()
317 If the VMA passes some filtering as described in "Filtering Special Vmas"
318 below, mlock_fixup() will attempt to merge the VMA with its neighbors or split
319 off a subset of the VMA if the range does not cover the entire VMA. Once the
321 populate_vma_page_range() to fault in the pages via get_user_pages() and to
322 mark the pages as mlocked via mlock_vma_page().
324 Note that the VMA being mlocked might be mapped with PROT_NONE. In this case,
325 get_user_pages() will be unable to fault in the pages. That's okay. If pages
326 do end up getting faulted into this VM_LOCKED VMA, we'll handle them in the
331 populate_vma_page_range() checks page_mapping() after acquiring the page lock.
332 If the page is still associated with its mapping, we'll go ahead and call
333 mlock_vma_page(). If the mapping is gone, we just unlock the page and move on.
334 In the worst case, this will result in a page mapped in a VM_LOCKED VMA
339 get_user_pages(). We use TestSetPageMlocked() because the page might already
341 especially do not want to count an mlocked page more than once in the
342 statistics. If the page was already mlocked, mlock_vma_page() need do nothing
345 If the page was NOT already mlocked, mlock_vma_page() attempts to isolate the
346 page from the LRU, as it is likely on the appropriate active or inactive list
347 at that time. If the isolate_lru_page() succeeds, mlock_vma_page() will put
348 back the page - by calling putback_lru_page() - which will notice that the page
349 is now mlocked and divert the page to the zone's unevictable list. If
350 mlock_vma_page() is unable to isolate the page from the LRU, vmscan will handle
351 it later if and when it attempts to reclaim the page.
361 mlocked. In any case, most of the pages have no struct page in which to so
362 mark the page. Because of this, get_user_pages() will fail for these VMAs,
366 neither need nor want to mlock() these pages. However, to preserve the
367 prior behavior of mlock() - before the unevictable/mlock changes -
368 mlock_fixup() will call make_pages_present() in the hugetlbfs VMA range to
369 allocate the huge pages and populate the ptes.
372 such as the VDSO page, relay channel pages, etc. These pages
373 are inherently unevictable and are not managed on the LRU lists.
374 mlock_fixup() treats these VMAs the same as hugetlbfs VMAs. It calls
375 make_pages_present() to populate the ptes.
377 Note that for all of these special VMAs, mlock_fixup() does not set the
380 VMAs against the task's "locked_vm".
386 The munlock() and munlockall() system calls are handled by the same functions -
387 do_mlock[all]() - as the mlock() and mlockall() system calls with the unlock vs
390 mlock_fixup() simply returns. Because of the VMA filtering discussed above,
394 If the VMA is VM_LOCKED, mlock_fixup() again attempts to merge or split off the
395 specified range. The range is then munlocked via the function
396 populate_vma_page_range() - the same function used to mlock a VMA range -
399 Because the VMA access protections could have been changed to PROT_NONE after
402 get_user_pages() was enhanced to accept a flag to ignore the permissions when
403 fetching the pages - all of which should be resident as a result of previous
407 munlock_vma_page(). munlock_vma_page() unconditionally clears the PG_mlocked
409 munlock_vma_page() use the Test*PageMlocked() function to handle the case where
410 the page might have already been unlocked by another task. If the page was
411 mlocked, munlock_vma_page() updates that zone statistics for the number of
413 the page is mapped by other VM_LOCKED VMAs.
415 We can't call try_to_munlock(), the function that walks the reverse map to
416 check for other VM_LOCKED VMAs, without first isolating the page from the LRU.
417 try_to_munlock() is a variant of try_to_unmap() and thus requires that the page
418 not be on an LRU list [more on these below]. However, the call to
420 we go ahead and clear PG_mlocked up front, as this might be the only chance we
421 have. If we can successfully isolate the page, we go ahead and
422 try_to_munlock(), which will restore the PG_mlocked flag and update the zone
423 page statistics if it finds another VMA holding the page mlocked. If we fail
424 to isolate the page, we'll have left a potentially mlocked page on the LRU.
426 the page. This should be relatively rare.
432 A page that is being migrated has been isolated from the LRU lists and is held
433 locked across unmapping of the page, updating the page's address space entry
434 and copying the contents and state, until the page table entry has been
435 replaced with an entry that refers to the new page. Linux supports migration
436 of mlocked pages and other unevictable pages. This involves simply moving the
437 PG_mlocked and PG_unevictable states from the old page to the new page.
439 Note that page migration can race with mlocking or munlocking of the same page.
440 This has been discussed from the mlock/munlock perspective in the respective
441 sections above. Both processes (migration and m[un]locking) hold the page
442 locked. This provides the first level of synchronization. Page migration
443 zeros out the page_mapping of the old page before unlocking it, so m[un]lock
444 can skip these pages by testing the page mapping under page lock.
446 To complete page migration, we place the new and old pages back onto the LRU
447 after dropping the page lock. The "unneeded" page - old page on success, new
448 page on failure - will be freed when the reference count held by the migration
449 process is released. To ensure that we don't strand pages on the unevictable
450 list because of a race between munlock and migration, page migration uses the
451 putback_lru_page() function to add migrated pages back to the LRU.
457 The unevictable LRU can be scanned for compactable regions and the default
459 this behavior (see Documentation/sysctl/vm.txt). Once scanning of the
460 unevictable LRU is enabled, the work of compaction is mostly handled by
461 the page migration code and the same work flow as described in MIGRATING
468 In addition the mlock()/mlockall() system calls, an application can request
469 that a region of memory be mlocked supplying the MAP_LOCKED flag to the mmap()
470 call. Furthermore, any mmap() call or brk() call that expands the heap by a
471 task that has previously called mlockall() with the MCL_FUTURE flag will result
472 in the newly mapped memory being mlocked. Before the unevictable/mlock
473 changes, the kernel simply called make_pages_present() to allocate pages and
474 populate the page table.
476 To mlock a range of memory under the unevictable/mlock infrastructure, the
478 populate_vma_page_range() specifying the vma and the address range to mlock.
480 The callers of populate_vma_page_range() will have already added the memory range
481 to be mlocked to the task's "locked_vm". To account for filtered VMAs,
482 populate_vma_page_range() returns the number of pages NOT mlocked. All of the
483 callers then subtract a non-negative return value from the task's locked_vm. A
485 attempting to fault in a VMA with PROT_NONE access. In this case, we leave the
486 memory range accounted as locked_vm, as the protections could be changed later
495 munlock the pages if we're removing the last VM_LOCKED VMA that maps the pages.
496 Before the unevictable/mlock changes, mlocking did not mark the pages in any
499 To munlock a range of memory under the unevictable/mlock infrastructure, the
501 munlock_vma_pages_all(). The name reflects the observation that one always
502 specifies the entire VMA range when munlock()ing during unmap of a region.
503 Because of the VMA filtering when mlocking() regions, only "normal" VMAs that
506 munlock_vma_pages_all() clears the VM_LOCKED VMA flag and, like mlock_fixup()
507 for the munlock case, calls __munlock_vma_pages_range() to walk the page table
508 for the VMA's memory range and munlock_vma_page() each resident page mapped by
509 the VMA. This effectively munlocks the page, only if this is the last
510 VM_LOCKED VMA that maps the page.
518 VM_LOCKED VMAs not to have the PG_mlocked flag set and therefore reside on one
519 of the active or inactive LRU lists. This could happen if, for example, a task
520 in the process of munlocking the page could not isolate the page from the LRU.
527 migration, with the argument page locked and isolated from the LRU. Separate
533 To unmap anonymous pages, each VMA in the list anchored in the anon_vma
534 must be visited - at least until a VM_LOCKED VMA is encountered. If the
535 page is being unmapped for migration, VM_LOCKED VMAs do not stop the
537 the page is mapped into a VM_LOCKED VMA, the scan stops.
539 try_to_unmap_anon() attempts to acquire in read mode the mmap semaphore of
540 the mm_struct to which the VMA belongs. If this is successful, it will
541 mlock the page via mlock_vma_page() - we wouldn't have gotten to
542 try_to_unmap_anon() if the page were already mlocked - and will return
543 SWAP_MLOCK, indicating that the page is unevictable.
545 If the mmap semaphore cannot be acquired, we are not sure whether the page
551 Unmapping of a mapped file page works the same as for anonymous mappings,
552 except that the scan visits all VMAs that map the page's index/page offset
553 in the page's mapping's reverse map priority search tree. It also visits
554 each VMA in the page's mapping's non-linear list, if the list is
558 page, try_to_unmap_file() will attempt to acquire the associated
559 mm_struct's mmap semaphore to mlock the page, returning SWAP_MLOCK if this
566 whether the page is mapped in a VM_LOCKED VMA. Again, the scan must visit
567 all VMAs in the non-linear list to ensure that the pages is not/should not
570 If a VM_LOCKED VMA is found in the list, the scan could terminate.
571 However, there is no easy way to determine whether the page is actually
572 mapped in a given VMA - either for unmapping or testing whether the
573 VM_LOCKED VMA actually pins the page.
576 number of pages - a "cluster" - in each non-linear VMA associated with the
578 this happens to unmap the page we're trying to unmap, try_to_unmap() will
581 recirculate this page. We take advantage of the cluster scan in
584 For each non-linear VMA, try_to_unmap_cluster() attempts to acquire the
585 mmap semaphore of the associated mm_struct for read without blocking.
587 If this attempt is successful and the VMA is VM_LOCKED,
588 try_to_unmap_cluster() will retain the mmap semaphore for the scan;
591 Then, for each page in the cluster, if we're holding the mmap semaphore
593 mlock the page. This call is a no-op if the page is already locked,
594 but will mlock any pages in the non-linear mapping that happen to be
597 If one of the pages so mlocked is the page passed in to try_to_unmap(),
598 try_to_unmap_cluster() will return SWAP_MLOCK, rather than the default
599 SWAP_AGAIN. This will allow vmscan to cull the page, rather than
600 recirculating it on the inactive list.
602 Again, if try_to_unmap_cluster() cannot acquire the VMA's mmap sem, it
603 returns SWAP_AGAIN, indicating that the page is mapped by a VM_LOCKED
610 [!] TODO/FIXME: a better name might be page_mlocked() - analogous to the
615 the page is mapped by any VM_LOCKED VMA without actually attempting to unmap
616 all PTEs from the page. For this purpose, the unevictable/mlock infrastructure
619 try_to_munlock() calls the same functions as try_to_unmap() for anonymous and
621 processing. Again, these functions walk the respective reverse maps looking
623 pages mapped in linear VMAs, as in the try_to_unmap() case, the functions
624 attempt to acquire the associated mmap semaphore, mlock the page via
625 mlock_vma_page() and return SWAP_MLOCK. This effectively undoes the
626 pre-clearing of the page's PG_mlocked done by munlock_vma_page.
630 recycle the page on the inactive list and hope that it has better luck with the
633 For file pages mapped into non-linear VMAs, the try_to_munlock() logic works
635 map the page, try_to_munlock() returns SWAP_AGAIN without actually mlocking the
636 page. munlock_vma_page() will just leave the page unlocked and let vmscan deal
637 with it - the usual fallback position.
641 However, the scan can terminate when it encounters a VM_LOCKED VMA and can
642 successfully acquire the VMA's mmap semaphore for read and mlock the page.
652 !page_evictable(page) - diverting these to the unevictable list.
653 However, shrink_active_list() only sees unevictable pages that made it onto the
655 set - otherwise they would be on the unevictable list and shrink_active_list
658 Some examples of these unevictable pages on the LRU lists are:
660 (1) ramfs pages that have been placed on the LRU lists when first allocated.
663 allocate or fault in the pages in the shared memory region. This happens
664 when an application accesses the page the first time after SHM_LOCK'ing
665 the segment.
667 (3) mlocked pages that could not be isolated from the LRU and moved to the
671 acquire the VMA's mmap semaphore to test the flags and set PageMlocked.
672 munlock_vma_page() was forced to let the page back on to the normal LRU
675 shrink_inactive_list() also diverts any unevictable pages that it finds on the
676 inactive lists to the appropriate zone's unevictable list.
679 after shrink_active_list() had moved them to the inactive list, or pages mapped
680 into VM_LOCKED VMAs that munlock_vma_page() couldn't isolate from the LRU to
681 recheck via try_to_munlock(). shrink_inactive_list() won't notice the latter,
686 VM_LOCKED VMAs but without PG_mlocked set will make it all the way to
687 try_to_unmap(). shrink_page_list() will divert them to the unevictable list