blob: fa3b527086fabdb44f7ebb9ef97ef00923ec63c6 [file] [log] [blame]
David Howellsc24b7202009-04-13 14:40:01 -07001 ==============================
2 UNEVICTABLE LRU INFRASTRUCTURE
3 ==============================
Lee Schermerhornfa07e782008-10-18 20:26:47 -07004
David Howellsc24b7202009-04-13 14:40:01 -07005========
6CONTENTS
7========
Lee Schermerhornfa07e782008-10-18 20:26:47 -07008
David Howellsc24b7202009-04-13 14:40:01 -07009 (*) The Unevictable LRU
Lee Schermerhornfa07e782008-10-18 20:26:47 -070010
David Howellsc24b7202009-04-13 14:40:01 -070011 - The unevictable page list.
12 - Memory control group interaction.
13 - Marking address spaces unevictable.
14 - Detecting Unevictable Pages.
15 - vmscan's handling of unevictable pages.
16
17 (*) mlock()'d pages.
18
19 - History.
20 - Basic management.
21 - mlock()/mlockall() system call handling.
22 - Filtering special vmas.
23 - munlock()/munlockall() system call handling.
24 - Migrating mlocked pages.
Eric B Munson922c0552015-04-15 16:13:23 -070025 - Compacting mlocked pages.
David Howellsc24b7202009-04-13 14:40:01 -070026 - mmap(MAP_LOCKED) system call handling.
27 - munmap()/exit()/exec() system call handling.
28 - try_to_unmap().
29 - try_to_munlock() reverse map scan.
30 - Page reclaim in shrink_*_list().
31
32
33============
34INTRODUCTION
35============
36
37This document describes the Linux memory manager's "Unevictable LRU"
38infrastructure and the use of this to manage several types of "unevictable"
39pages.
40
41The document attempts to provide the overall rationale behind this mechanism
42and the rationale for some of the design decisions that drove the
43implementation. The latter design rationale is discussed in the context of an
44implementation description. Admittedly, one can obtain the implementation
45details - the "what does it do?" - by reading the code. One hopes that the
46descriptions below add value by provide the answer to "why does it do that?".
47
48
49===================
50THE UNEVICTABLE LRU
51===================
52
53The Unevictable LRU facility adds an additional LRU list to track unevictable
54pages and to hide these pages from vmscan. This mechanism is based on a patch
55by Larry Woodman of Red Hat to address several scalability problems with page
Lee Schermerhornfa07e782008-10-18 20:26:47 -070056reclaim in Linux. The problems have been observed at customer sites on large
David Howellsc24b7202009-04-13 14:40:01 -070057memory x86_64 systems.
Lee Schermerhornfa07e782008-10-18 20:26:47 -070058
David Howellsc24b7202009-04-13 14:40:01 -070059To illustrate this with an example, a non-NUMA x86_64 platform with 128GB of
60main memory will have over 32 million 4k pages in a single zone. When a large
61fraction of these pages are not evictable for any reason [see below], vmscan
62will spend a lot of time scanning the LRU lists looking for the small fraction
63of pages that are evictable. This can result in a situation where all CPUs are
64spending 100% of their time in vmscan for hours or days on end, with the system
65completely unresponsive.
Lee Schermerhornfa07e782008-10-18 20:26:47 -070066
David Howellsc24b7202009-04-13 14:40:01 -070067The unevictable list addresses the following classes of unevictable pages:
Lee Schermerhornfa07e782008-10-18 20:26:47 -070068
David Howellsc24b7202009-04-13 14:40:01 -070069 (*) Those owned by ramfs.
70
71 (*) Those mapped into SHM_LOCK'd shared memory regions.
72
73 (*) Those mapped into VM_LOCKED [mlock()ed] VMAs.
74
75The infrastructure may also be able to handle other conditions that make pages
Lee Schermerhornfa07e782008-10-18 20:26:47 -070076unevictable, either by definition or by circumstance, in the future.
77
78
David Howellsc24b7202009-04-13 14:40:01 -070079THE UNEVICTABLE PAGE LIST
80-------------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -070081
82The Unevictable LRU infrastructure consists of an additional, per-zone, LRU list
83called the "unevictable" list and an associated page flag, PG_unevictable, to
David Howellsc24b7202009-04-13 14:40:01 -070084indicate that the page is being managed on the unevictable list.
85
86The PG_unevictable flag is analogous to, and mutually exclusive with, the
87PG_active flag in that it indicates on which LRU list a page resides when
Michal Hockoe6e8dd52011-03-16 15:01:37 +010088PG_lru is set.
Lee Schermerhornfa07e782008-10-18 20:26:47 -070089
90The Unevictable LRU infrastructure maintains unevictable pages on an additional
91LRU list for a few reasons:
92
David Howellsc24b7202009-04-13 14:40:01 -070093 (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]
Lee Schermerhornfa07e782008-10-18 20:26:47 -070097
David Howellsc24b7202009-04-13 14:40:01 -070098 (2) We want to be able to migrate unevictable pages between nodes for memory
99 defragmentation, workload management and memory hotplug. The linux kernel
100 can only migrate pages that it can successfully isolate from the LRU
101 lists. If we were to maintain pages elsewhere than on an LRU-like list,
102 where they can be found by isolate_lru_page(), we would prevent their
103 migration, unless we reworked migration code to find the unevictable pages
104 itself.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700105
106
David Howellsc24b7202009-04-13 14:40:01 -0700107The unevictable list does not differentiate between file-backed and anonymous,
108swap-backed pages. This differentiation is only important while the pages are,
109in fact, evictable.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700110
David Howellsc24b7202009-04-13 14:40:01 -0700111The unevictable list benefits from the "arrayification" of the per-zone LRU
112lists and statistics originally proposed and posted by Christoph Lameter.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700113
David Howellsc24b7202009-04-13 14:40:01 -0700114The unevictable list does not use the LRU pagevec mechanism. Rather,
115unevictable pages are placed directly on the page's zone's unevictable list
116under the zone lru_lock. This allows us to prevent the stranding of pages on
117the unevictable list when one task has the page isolated from the LRU and other
118tasks are changing the "evictability" state of the page.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700119
120
David Howellsc24b7202009-04-13 14:40:01 -0700121MEMORY CONTROL GROUP INTERACTION
122--------------------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700123
David Howellsc24b7202009-04-13 14:40:01 -0700124The unevictable LRU facility interacts with the memory control group [aka
125memory controller; see Documentation/cgroups/memory.txt] by extending the
126lru_list enum.
127
128The memory controller data structure automatically gets a per-zone unevictable
129list as a result of the "arrayification" of the per-zone LRU lists (one per
130lru_list enum element). The memory controller tracks the movement of pages to
131and from the unevictable list.
132
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700133When a memory control group comes under memory pressure, the controller will
134not attempt to reclaim pages on the unevictable list. This has a couple of
David Howellsc24b7202009-04-13 14:40:01 -0700135effects:
136
137 (1) Because the pages are "hidden" from reclaim on the unevictable list, the
138 reclaim process can be more efficient, dealing only with pages that have a
139 chance of being reclaimed.
140
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.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700145
146
David Howellsc24b7202009-04-13 14:40:01 -0700147MARKING ADDRESS SPACES UNEVICTABLE
148----------------------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700149
David Howellsc24b7202009-04-13 14:40:01 -0700150For facilities such as ramfs none of the pages attached to the address space
151may be evicted. To prevent eviction of any such pages, the AS_UNEVICTABLE
152address space flag is provided, and this can be manipulated by a filesystem
153using a number of wrapper functions:
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700154
David Howellsc24b7202009-04-13 14:40:01 -0700155 (*) void mapping_set_unevictable(struct address_space *mapping);
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700156
David Howellsc24b7202009-04-13 14:40:01 -0700157 Mark the address space as being completely unevictable.
158
159 (*) void mapping_clear_unevictable(struct address_space *mapping);
160
161 Mark the address space as being evictable.
162
163 (*) int mapping_unevictable(struct address_space *mapping);
164
165 Query the address space, and return true if it is completely
166 unevictable.
167
168These are currently used in two places in the kernel:
169
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.
172
173 (2) By SYSV SHM to mark SHM_LOCK'd address spaces until SHM_UNLOCK is called.
174
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
177 ensure they're in memory.
178
179
180DETECTING UNEVICTABLE PAGES
181---------------------------
182
183The function page_evictable() in vmscan.c determines whether a page is
184evictable or not using the query function outlined above [see section "Marking
185address spaces unevictable"] to check the AS_UNEVICTABLE flag.
186
187For address spaces that are so marked after being populated (as SHM regions
188might be), the lock action (eg: SHM_LOCK) can be lazy, and need not populate
189the page tables for the region as does, for example, mlock(), nor need it make
190any special effort to push any pages in the SHM_LOCK'd area to the unevictable
191list. Instead, vmscan will do this if and when it encounters the pages during
192a reclamation scan.
193
194On an unlock action (such as SHM_UNLOCK), the unlocker (eg: shmctl()) must scan
195the pages in the region and "rescue" them from the unevictable list if no other
196condition is keeping them unevictable. If an unevictable region is destroyed,
197the pages are also "rescued" from the unevictable list in the process of
198freeing them.
199
200page_evictable() also checks for mlocked pages by testing an additional page
Hugh Dickins39b5f292012-10-08 16:33:18 -0700201flag, PG_mlocked (as wrapped by PageMlocked()), which is set when a page is
202faulted into a VM_LOCKED vma, or found in a vma being VM_LOCKED.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700203
204
David Howellsc24b7202009-04-13 14:40:01 -0700205VMSCAN'S HANDLING OF UNEVICTABLE PAGES
206--------------------------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700207
208If unevictable pages are culled in the fault path, or moved to the unevictable
David Howellsc24b7202009-04-13 14:40:01 -0700209list at mlock() or mmap() time, vmscan will not encounter the pages until they
210have become evictable again (via munlock() for example) and have been "rescued"
211from the unevictable list. However, there may be situations where we decide,
212for the sake of expediency, to leave a unevictable page on one of the regular
213active/inactive LRU lists for vmscan to deal with. vmscan checks for such
214pages 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
216unevictable list for the zone being scanned.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700217
David Howellsc24b7202009-04-13 14:40:01 -0700218There may be situations where a page is mapped into a VM_LOCKED VMA, but the
219page is not marked as PG_mlocked. Such pages will make it all the way to
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700220shrink_page_list() where they will be detected when vmscan walks the reverse
David Howellsc24b7202009-04-13 14:40:01 -0700221map in try_to_unmap(). If try_to_unmap() returns SWAP_MLOCK,
222shrink_page_list() will cull the page at that point.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700223
David Howellsc24b7202009-04-13 14:40:01 -0700224To "cull" an unevictable page, vmscan simply puts the page back on the LRU list
225using putback_lru_page() - the inverse operation to isolate_lru_page() - after
226dropping the page lock. Because the condition which makes the page unevictable
227may change once the page is unlocked, putback_lru_page() will recheck the
228unevictable state of a page that it places on the unevictable list. If the
229page has become unevictable, putback_lru_page() removes it from the list and
230retries, including the page_unevictable() test. Because such a race is a rare
231event and movement of pages onto the unevictable list should be rare, these
232extra evictabilty checks should not occur in the majority of calls to
233putback_lru_page().
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700234
235
David Howellsc24b7202009-04-13 14:40:01 -0700236=============
237MLOCKED PAGES
238=============
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700239
David Howellsc24b7202009-04-13 14:40:01 -0700240The unevictable page list is also useful for mlock(), in addition to ramfs and
241SYSV SHM. Note that mlock() is only available in CONFIG_MMU=y situations; in
242NOMMU situations, all mappings are effectively mlocked.
243
244
245HISTORY
246-------
247
248The "Unevictable mlocked Pages" infrastructure is based on work originally
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700249posted by Nick Piggin in an RFC patch entitled "mm: mlocked pages off LRU".
David Howellsc24b7202009-04-13 14:40:01 -0700250Nick posted his patch as an alternative to a patch posted by Christoph Lameter
251to achieve the same objective: hiding mlocked pages from vmscan.
252
253In Nick's patch, he used one of the struct page LRU list link fields as a count
254of VM_LOCKED VMAs that map the page. This use of the link field for a count
255prevented the management of the pages on an LRU list, and thus mlocked pages
256were not migratable as isolate_lru_page() could not find them, and the LRU list
257link field was not available to the migration subsystem.
258
259Nick resolved this by putting mlocked pages back on the lru list before
260attempting to isolate them, thus abandoning the count of VM_LOCKED VMAs. When
261Nick's patch was integrated with the Unevictable LRU work, the count was
262replaced by walking the reverse map to determine whether any VM_LOCKED VMAs
263mapped the page. More on this below.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700264
265
David Howellsc24b7202009-04-13 14:40:01 -0700266BASIC MANAGEMENT
267----------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700268
David Howellsc24b7202009-04-13 14:40:01 -0700269mlocked pages - pages mapped into a VM_LOCKED VMA - are a class of unevictable
270pages. When such a page has been "noticed" by the memory management subsystem,
271the page is marked with the PG_mlocked flag. This can be manipulated using the
272PageMlocked() functions.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700273
David Howellsc24b7202009-04-13 14:40:01 -0700274A PG_mlocked page will be placed on the unevictable list when it is added to
275the LRU. Such pages can be "noticed" by memory management in several places:
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700276
David Howellsc24b7202009-04-13 14:40:01 -0700277 (1) in the mlock()/mlockall() system call handlers;
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700278
David Howellsc24b7202009-04-13 14:40:01 -0700279 (2) in the mmap() system call handler when mmapping a region with the
280 MAP_LOCKED flag;
281
282 (3) mmapping a region in a task that has called mlockall() with the MCL_FUTURE
283 flag
284
285 (4) in the fault path, if mlocked pages are "culled" in the fault path,
286 and when a VM_LOCKED stack segment is expanded; or
287
288 (5) as mentioned above, in vmscan:shrink_page_list() when attempting to
289 reclaim a page in a VM_LOCKED VMA via try_to_unmap()
290
291all of which result in the VM_LOCKED flag being set for the VMA if it doesn't
292already have it set.
293
294mlocked pages become unlocked and rescued from the unevictable list when:
295
296 (1) mapped in a range unlocked via the munlock()/munlockall() system calls;
297
298 (2) munmap()'d out of the last VM_LOCKED VMA that maps the page, including
299 unmapping at task exit;
300
301 (3) when the page is truncated from the last VM_LOCKED VMA of an mmapped file;
302 or
303
304 (4) before a page is COW'd in a VM_LOCKED VMA.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700305
306
David Howellsc24b7202009-04-13 14:40:01 -0700307mlock()/mlockall() SYSTEM CALL HANDLING
308---------------------------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700309
310Both [do_]mlock() and [do_]mlockall() system call handlers call mlock_fixup()
David Howellsc24b7202009-04-13 14:40:01 -0700311for each VMA in the range specified by the call. In the case of mlockall(),
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700312this is the entire active address space of the task. Note that mlock_fixup()
David Howellsc24b7202009-04-13 14:40:01 -0700313is used for both mlocking and munlocking a range of memory. A call to mlock()
314an already VM_LOCKED VMA, or to munlock() a VMA that is not VM_LOCKED is
315treated as a no-op, and mlock_fixup() simply returns.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700316
David Howellsc24b7202009-04-13 14:40:01 -0700317If the VMA passes some filtering as described in "Filtering Special Vmas"
318below, mlock_fixup() will attempt to merge the VMA with its neighbors or split
319off a subset of the VMA if the range does not cover the entire VMA. Once the
320VMA has been merged or split or neither, mlock_fixup() will call
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700321populate_vma_page_range() to fault in the pages via get_user_pages() and to
David Howellsc24b7202009-04-13 14:40:01 -0700322mark the pages as mlocked via mlock_vma_page().
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700323
David Howellsc24b7202009-04-13 14:40:01 -0700324Note that the VMA being mlocked might be mapped with PROT_NONE. In this case,
325get_user_pages() will be unable to fault in the pages. That's okay. If pages
326do end up getting faulted into this VM_LOCKED VMA, we'll handle them in the
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700327fault path or in vmscan.
328
329Also note that a page returned by get_user_pages() could be truncated or
David Howellsc24b7202009-04-13 14:40:01 -0700330migrated out from under us, while we're trying to mlock it. To detect this,
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700331populate_vma_page_range() checks page_mapping() after acquiring the page lock.
David Howellsc24b7202009-04-13 14:40:01 -0700332If the page is still associated with its mapping, we'll go ahead and call
333mlock_vma_page(). If the mapping is gone, we just unlock the page and move on.
334In the worst case, this will result in a page mapped in a VM_LOCKED VMA
335remaining on a normal LRU list without being PageMlocked(). Again, vmscan will
336detect and cull such pages.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700337
David Howellsc24b7202009-04-13 14:40:01 -0700338mlock_vma_page() will call TestSetPageMlocked() for each page returned by
339get_user_pages(). We use TestSetPageMlocked() because the page might already
340be mlocked by another task/VMA and we don't want to do extra work. We
341especially do not want to count an mlocked page more than once in the
342statistics. If the page was already mlocked, mlock_vma_page() need do nothing
343more.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700344
345If the page was NOT already mlocked, mlock_vma_page() attempts to isolate the
346page from the LRU, as it is likely on the appropriate active or inactive list
David Howellsc24b7202009-04-13 14:40:01 -0700347at that time. If the isolate_lru_page() succeeds, mlock_vma_page() will put
348back the page - by calling putback_lru_page() - which will notice that the page
349is now mlocked and divert the page to the zone's unevictable list. If
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700350mlock_vma_page() is unable to isolate the page from the LRU, vmscan will handle
David Howellsc24b7202009-04-13 14:40:01 -0700351it later if and when it attempts to reclaim the page.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700352
353
David Howellsc24b7202009-04-13 14:40:01 -0700354FILTERING SPECIAL VMAS
355----------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700356
David Howellsc24b7202009-04-13 14:40:01 -0700357mlock_fixup() filters several classes of "special" VMAs:
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700358
David Howellsc24b7202009-04-13 14:40:01 -07003591) VMAs with VM_IO or VM_PFNMAP set are skipped entirely. The pages behind
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700360 these mappings are inherently pinned, so we don't need to mark them as
David Howellsc24b7202009-04-13 14:40:01 -0700361 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,
363 so there is no sense in attempting to visit them.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700364
David Howellsc24b7202009-04-13 14:40:01 -07003652) VMAs mapping hugetlbfs page are already effectively pinned into memory. We
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.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700370
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07003713) VMAs with VM_DONTEXPAND are generally userspace mappings of kernel pages,
372 such as the VDSO page, relay channel pages, etc. These pages
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700373 are inherently unevictable and are not managed on the LRU lists.
David Howellsc24b7202009-04-13 14:40:01 -0700374 mlock_fixup() treats these VMAs the same as hugetlbfs VMAs. It calls
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700375 make_pages_present() to populate the ptes.
376
David Howellsc24b7202009-04-13 14:40:01 -0700377Note that for all of these special VMAs, mlock_fixup() does not set the
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700378VM_LOCKED flag. Therefore, we won't have to deal with them later during
David Howellsc24b7202009-04-13 14:40:01 -0700379munlock(), munmap() or task exit. Neither does mlock_fixup() account these
380VMAs against the task's "locked_vm".
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700381
382
David Howellsc24b7202009-04-13 14:40:01 -0700383munlock()/munlockall() SYSTEM CALL HANDLING
384-------------------------------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700385
David Howellsc24b7202009-04-13 14:40:01 -0700386The munlock() and munlockall() system calls are handled by the same functions -
387do_mlock[all]() - as the mlock() and mlockall() system calls with the unlock vs
388lock operation indicated by an argument. So, these system calls are also
389handled by mlock_fixup(). Again, if called for an already munlocked VMA,
390mlock_fixup() simply returns. Because of the VMA filtering discussed above,
391VM_LOCKED will not be set in any "special" VMAs. So, these VMAs will be
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700392ignored for munlock.
393
David Howellsc24b7202009-04-13 14:40:01 -0700394If the VMA is VM_LOCKED, mlock_fixup() again attempts to merge or split off the
395specified range. The range is then munlocked via the function
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700396populate_vma_page_range() - the same function used to mlock a VMA range -
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700397passing a flag to indicate that munlock() is being performed.
398
David Howellsc24b7202009-04-13 14:40:01 -0700399Because the VMA access protections could have been changed to PROT_NONE after
Hugh Dickins63d6c5a2009-01-06 14:39:38 -0800400faulting in and mlocking pages, get_user_pages() was unreliable for visiting
David Howellsc24b7202009-04-13 14:40:01 -0700401these pages for munlocking. Because we don't want to leave pages mlocked,
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700402get_user_pages() was enhanced to accept a flag to ignore the permissions when
David Howellsc24b7202009-04-13 14:40:01 -0700403fetching the pages - all of which should be resident as a result of previous
404mlocking.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700405
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700406For munlock(), populate_vma_page_range() unlocks individual pages by calling
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700407munlock_vma_page(). munlock_vma_page() unconditionally clears the PG_mlocked
David Howellsc24b7202009-04-13 14:40:01 -0700408flag using TestClearPageMlocked(). As with mlock_vma_page(),
409munlock_vma_page() use the Test*PageMlocked() function to handle the case where
410the page might have already been unlocked by another task. If the page was
411mlocked, munlock_vma_page() updates that zone statistics for the number of
412mlocked pages. Note, however, that at this point we haven't checked whether
413the page is mapped by other VM_LOCKED VMAs.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700414
David Howellsc24b7202009-04-13 14:40:01 -0700415We can't call try_to_munlock(), the function that walks the reverse map to
416check for other VM_LOCKED VMAs, without first isolating the page from the LRU.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700417try_to_munlock() is a variant of try_to_unmap() and thus requires that the page
David Howellsc24b7202009-04-13 14:40:01 -0700418not be on an LRU list [more on these below]. However, the call to
419isolate_lru_page() could fail, in which case we couldn't try_to_munlock(). So,
420we go ahead and clear PG_mlocked up front, as this might be the only chance we
421have. If we can successfully isolate the page, we go ahead and
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700422try_to_munlock(), which will restore the PG_mlocked flag and update the zone
David Howellsc24b7202009-04-13 14:40:01 -0700423page statistics if it finds another VMA holding the page mlocked. If we fail
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700424to isolate the page, we'll have left a potentially mlocked page on the LRU.
David Howellsc24b7202009-04-13 14:40:01 -0700425This is fine, because we'll catch it later if and if vmscan tries to reclaim
426the page. This should be relatively rare.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700427
428
David Howellsc24b7202009-04-13 14:40:01 -0700429MIGRATING MLOCKED PAGES
430-----------------------
431
432A page that is being migrated has been isolated from the LRU lists and is held
433locked across unmapping of the page, updating the page's address space entry
434and copying the contents and state, until the page table entry has been
435replaced with an entry that refers to the new page. Linux supports migration
436of mlocked pages and other unevictable pages. This involves simply moving the
437PG_mlocked and PG_unevictable states from the old page to the new page.
438
439Note that page migration can race with mlocking or munlocking of the same page.
440This has been discussed from the mlock/munlock perspective in the respective
441sections above. Both processes (migration and m[un]locking) hold the page
442locked. This provides the first level of synchronization. Page migration
443zeros out the page_mapping of the old page before unlocking it, so m[un]lock
444can skip these pages by testing the page mapping under page lock.
445
446To complete page migration, we place the new and old pages back onto the LRU
447after dropping the page lock. The "unneeded" page - old page on success, new
448page on failure - will be freed when the reference count held by the migration
449process is released. To ensure that we don't strand pages on the unevictable
450list because of a race between munlock and migration, page migration uses the
451putback_lru_page() function to add migrated pages back to the LRU.
452
453
Eric B Munson922c0552015-04-15 16:13:23 -0700454COMPACTING MLOCKED PAGES
455------------------------
456
457The unevictable LRU can be scanned for compactable regions and the default
458behavior is to do so. /proc/sys/vm/compact_unevictable_allowed controls
459this behavior (see Documentation/sysctl/vm.txt). Once scanning of the
460unevictable LRU is enabled, the work of compaction is mostly handled by
461the page migration code and the same work flow as described in MIGRATING
462MLOCKED PAGES will apply.
463
464
David Howellsc24b7202009-04-13 14:40:01 -0700465mmap(MAP_LOCKED) SYSTEM CALL HANDLING
466-------------------------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700467
Masanari Iidadf5cbb22014-03-21 10:04:30 +0900468In addition the mlock()/mlockall() system calls, an application can request
David Howellsc24b7202009-04-13 14:40:01 -0700469that a region of memory be mlocked supplying the MAP_LOCKED flag to the mmap()
Michal Hocko9b012a22015-06-24 16:57:50 -0700470call. There is one important and subtle difference here, though. mmap() + mlock()
471will fail if the range cannot be faulted in (e.g. because mm_populate fails)
472and returns with ENOMEM while mmap(MAP_LOCKED) will not fail. The mmaped
473area will still have properties of the locked area - aka. pages will not get
474swapped out - but major page faults to fault memory in might still happen.
475
476Furthermore, any mmap() call or brk() call that expands the heap by a
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700477task that has previously called mlockall() with the MCL_FUTURE flag will result
David Howellsc24b7202009-04-13 14:40:01 -0700478in the newly mapped memory being mlocked. Before the unevictable/mlock
479changes, the kernel simply called make_pages_present() to allocate pages and
480populate the page table.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700481
482To mlock a range of memory under the unevictable/mlock infrastructure, the
483mmap() handler and task address space expansion functions call
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700484populate_vma_page_range() specifying the vma and the address range to mlock.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700485
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700486The callers of populate_vma_page_range() will have already added the memory range
David Howellsc24b7202009-04-13 14:40:01 -0700487to be mlocked to the task's "locked_vm". To account for filtered VMAs,
Kirill A. Shutemovfc05f562015-04-14 15:44:39 -0700488populate_vma_page_range() returns the number of pages NOT mlocked. All of the
David Howellsc24b7202009-04-13 14:40:01 -0700489callers then subtract a non-negative return value from the task's locked_vm. A
490negative return value represent an error - for example, from get_user_pages()
491attempting to fault in a VMA with PROT_NONE access. In this case, we leave the
492memory range accounted as locked_vm, as the protections could be changed later
493and pages allocated into that region.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700494
495
David Howellsc24b7202009-04-13 14:40:01 -0700496munmap()/exit()/exec() SYSTEM CALL HANDLING
497-------------------------------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700498
499When unmapping an mlocked region of memory, whether by an explicit call to
500munmap() or via an internal unmap from exit() or exec() processing, we must
David Howellsc24b7202009-04-13 14:40:01 -0700501munlock the pages if we're removing the last VM_LOCKED VMA that maps the pages.
Hugh Dickins63d6c5a2009-01-06 14:39:38 -0800502Before the unevictable/mlock changes, mlocking did not mark the pages in any
503way, so unmapping them required no processing.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700504
505To munlock a range of memory under the unevictable/mlock infrastructure, the
David Howellsc24b7202009-04-13 14:40:01 -0700506munmap() handler and task address space call tear down function
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700507munlock_vma_pages_all(). The name reflects the observation that one always
David Howellsc24b7202009-04-13 14:40:01 -0700508specifies the entire VMA range when munlock()ing during unmap of a region.
509Because of the VMA filtering when mlocking() regions, only "normal" VMAs that
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700510actually contain mlocked pages will be passed to munlock_vma_pages_all().
511
David Howellsc24b7202009-04-13 14:40:01 -0700512munlock_vma_pages_all() clears the VM_LOCKED VMA flag and, like mlock_fixup()
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700513for the munlock case, calls __munlock_vma_pages_range() to walk the page table
David Howellsc24b7202009-04-13 14:40:01 -0700514for the VMA's memory range and munlock_vma_page() each resident page mapped by
515the VMA. This effectively munlocks the page, only if this is the last
516VM_LOCKED VMA that maps the page.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700517
518
David Howellsc24b7202009-04-13 14:40:01 -0700519try_to_unmap()
520--------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700521
David Howellsc24b7202009-04-13 14:40:01 -0700522Pages can, of course, be mapped into multiple VMAs. Some of these VMAs may
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700523have VM_LOCKED flag set. It is possible for a page mapped into one or more
David Howellsc24b7202009-04-13 14:40:01 -0700524VM_LOCKED VMAs not to have the PG_mlocked flag set and therefore reside on one
525of the active or inactive LRU lists. This could happen if, for example, a task
526in the process of munlocking the page could not isolate the page from the LRU.
527As a result, vmscan/shrink_page_list() might encounter such a page as described
528in section "vmscan's handling of unevictable pages". To handle this situation,
529try_to_unmap() checks for VM_LOCKED VMAs while it is walking a page's reverse
530map.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700531
532try_to_unmap() is always called, by either vmscan for reclaim or for page
David Howellsc24b7202009-04-13 14:40:01 -0700533migration, with the argument page locked and isolated from the LRU. Separate
Hugh Dickinsb87537d9e2015-11-05 18:49:33 -0800534functions handle anonymous and mapped file and KSM pages, as these types of
535pages have different reverse map lookup mechanisms, with different locking.
536In each case, whether rmap_walk_anon() or rmap_walk_file() or rmap_walk_ksm(),
537it will call try_to_unmap_one() for every VMA which might contain the page.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700538
Hugh Dickinsb87537d9e2015-11-05 18:49:33 -0800539When trying to reclaim, if try_to_unmap_one() finds the page in a VM_LOCKED
540VMA, it will then mlock the page via mlock_vma_page() instead of unmapping it,
541and return SWAP_MLOCK to indicate that the page is unevictable: and the scan
542stops there.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700543
Hugh Dickinsb87537d9e2015-11-05 18:49:33 -0800544mlock_vma_page() is called while holding the page table's lock (in addition
545to the page lock, and the rmap lock): to serialize against concurrent mlock or
546munlock or munmap system calls, mm teardown (munlock_vma_pages_all), reclaim,
547holepunching, and truncation of file pages and their anonymous COWed pages.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700548
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700549
David Howellsc24b7202009-04-13 14:40:01 -0700550try_to_munlock() REVERSE MAP SCAN
551---------------------------------
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700552
David Howellsc24b7202009-04-13 14:40:01 -0700553 [!] TODO/FIXME: a better name might be page_mlocked() - analogous to the
554 page_referenced() reverse map walker.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700555
David Howellsc24b7202009-04-13 14:40:01 -0700556When munlock_vma_page() [see section "munlock()/munlockall() System Call
557Handling" above] tries to munlock a page, it needs to determine whether or not
558the page is mapped by any VM_LOCKED VMA without actually attempting to unmap
559all PTEs from the page. For this purpose, the unevictable/mlock infrastructure
560introduced a variant of try_to_unmap() called try_to_munlock().
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700561
562try_to_munlock() calls the same functions as try_to_unmap() for anonymous and
Hugh Dickinsb87537d9e2015-11-05 18:49:33 -0800563mapped file and KSM pages with a flag argument specifying unlock versus unmap
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700564processing. Again, these functions walk the respective reverse maps looking
Hugh Dickins7a142392015-11-05 18:49:30 -0800565for VM_LOCKED VMAs. When such a VMA is found, as in the try_to_unmap() case,
Hugh Dickinsb87537d9e2015-11-05 18:49:33 -0800566the functions mlock the page via mlock_vma_page() and return SWAP_MLOCK. This
567undoes the pre-clearing of the page's PG_mlocked done by munlock_vma_page.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700568
David Howellsc24b7202009-04-13 14:40:01 -0700569Note that try_to_munlock()'s reverse map walk must visit every VMA in a page's
570reverse map to determine that a page is NOT mapped into any VM_LOCKED VMA.
Hugh Dickinsb87537d9e2015-11-05 18:49:33 -0800571However, the scan can terminate when it encounters a VM_LOCKED VMA.
David Howellsc24b7202009-04-13 14:40:01 -0700572Although try_to_munlock() might be called a great many times when munlocking a
573large region or tearing down a large address space that has been mlocked via
574mlockall(), overall this is a fairly rare event.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700575
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700576
David Howellsc24b7202009-04-13 14:40:01 -0700577PAGE RECLAIM IN shrink_*_list()
578-------------------------------
579
580shrink_active_list() culls any obviously unevictable pages - i.e.
Hugh Dickins39b5f292012-10-08 16:33:18 -0700581!page_evictable(page) - diverting these to the unevictable list.
David Howellsc24b7202009-04-13 14:40:01 -0700582However, shrink_active_list() only sees unevictable pages that made it onto the
583active/inactive lru lists. Note that these pages do not have PageUnevictable
584set - otherwise they would be on the unevictable list and shrink_active_list
585would never see them.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700586
587Some examples of these unevictable pages on the LRU lists are:
588
David Howellsc24b7202009-04-13 14:40:01 -0700589 (1) ramfs pages that have been placed on the LRU lists when first allocated.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700590
David Howellsc24b7202009-04-13 14:40:01 -0700591 (2) SHM_LOCK'd shared memory pages. shmctl(SHM_LOCK) does not attempt to
592 allocate or fault in the pages in the shared memory region. This happens
593 when an application accesses the page the first time after SHM_LOCK'ing
594 the segment.
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700595
David Howellsc24b7202009-04-13 14:40:01 -0700596 (3) mlocked pages that could not be isolated from the LRU and moved to the
597 unevictable list in mlock_vma_page().
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700598
David Howellsc24b7202009-04-13 14:40:01 -0700599shrink_inactive_list() also diverts any unevictable pages that it finds on the
600inactive lists to the appropriate zone's unevictable list.
601
602shrink_inactive_list() should only see SHM_LOCK'd pages that became SHM_LOCK'd
603after shrink_active_list() had moved them to the inactive list, or pages mapped
604into VM_LOCKED VMAs that munlock_vma_page() couldn't isolate from the LRU to
605recheck via try_to_munlock(). shrink_inactive_list() won't notice the latter,
606but will pass on to shrink_page_list().
Lee Schermerhornfa07e782008-10-18 20:26:47 -0700607
608shrink_page_list() again culls obviously unevictable pages that it could
Hugh Dickins63d6c5a2009-01-06 14:39:38 -0800609encounter for similar reason to shrink_inactive_list(). Pages mapped into
David Howellsc24b7202009-04-13 14:40:01 -0700610VM_LOCKED VMAs but without PG_mlocked set will make it all the way to
Hugh Dickins63d6c5a2009-01-06 14:39:38 -0800611try_to_unmap(). shrink_page_list() will divert them to the unevictable list
612when try_to_unmap() returns SWAP_MLOCK, as discussed above.