2 * Copyright (c) 1982, 1986, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)vmmeter.h 8.2 (Berkeley) 7/10/94
30 * $FreeBSD: src/sys/sys/vmmeter.h,v 1.21.2.2 2002/10/10 19:28:21 dillon Exp $
33 #ifndef _VM_VM_PAGE2_H_
34 #define _VM_VM_PAGE2_H_
38 #ifndef _SYS_VMMETER_H_
39 #include <sys/vmmeter.h>
42 #include <sys/queue.h>
44 #ifndef _VM_VM_PAGE_H_
45 #include <vm/vm_page.h>
47 #ifndef _SYS_SPINLOCK_H_
48 #include <sys/spinlock.h>
50 #ifndef _SYS_SPINLOCK2_H_
51 #include <sys/spinlock2.h>
57 * VM fault rates are highly dependent on SMP locking conflicts and, on
58 * multi-socket systems, cache mastership changes for globals due to atomic
59 * ops (even simple atomic_add_*() calls). Cache mastership changes can
60 * limit the aggregate fault rate.
62 * For this reason we go through some hoops to access VM statistics for
63 * low-memory handling, pageout, and other triggers. Each cpu collects
64 * adjustments in gd->gd_vmstats_adj. These get rolled up into the global
65 * vmstats structure. The global vmstats structure is then pulled into
66 * gd->gd_vmstats by each cpu when it needs it. Critical path checks always
67 * use the pcpu gd->gd_vmstats structure.
70 * Return TRUE if we are under our severe low-free-pages threshold
72 * This causes user processes to stall to avoid exhausting memory that
73 * the kernel might need.
75 * reserved < severe < minimum < wait < start < target1 < target2
79 vm_paging_severe(void)
81 globaldata_t gd
= mycpu
;
83 if (__predict_false(gd
->gd_vmstats
.v_free_severe
>
84 gd
->gd_vmstats
.v_free_count
+
85 gd
->gd_vmstats
.v_cache_count
))
89 if (__predict_false(gd
->gd_vmstats
.v_free_reserved
>
90 gd
->gd_vmstats
.v_free_count
))
98 * Return TRUE if we are under our minimum low-free-pages threshold. We
99 * will not count (donotcount) free pages as being free (used mainly for
102 * This will cause most normal page faults to block and activate the
105 * The pageout daemon should already be active due to vm_paging_start(n)
106 * and will typically continue running until it hits target2
108 * reserved < severe < minimum < wait < start < target1 < target2
112 vm_paging_min_dnc(long donotcount
)
114 globaldata_t gd
= mycpu
;
116 if (__predict_false(gd
->gd_vmstats
.v_free_min
+ donotcount
>
117 (gd
->gd_vmstats
.v_free_count
+
118 gd
->gd_vmstats
.v_cache_count
)))
122 if (__predict_false(gd
->gd_vmstats
.v_free_reserved
>
123 gd
->gd_vmstats
.v_free_count
))
131 * Returns TRUE if the number of FREE+CACHE pages falls below vm_paging_wait,
132 * based on the nice value the trip point can be between vm_paging_min and
135 * Used by vm_fault (see vm_wait_pfault()) to block a process on low-memory
136 * based on the process 'nice' value (-20 to +20).
140 vm_paging_min_nice(int nice
)
147 delta
= vmstats
.v_paging_wait
- vmstats
.v_free_min
- 1;
150 /* range 0-40, 0 is high priority, 40 is low */
151 count
= (nice
+ 20) * delta
/ 40;
154 return vm_paging_min_dnc(count
);
161 return vm_paging_min_dnc(0);
165 * Return TRUE if nominal userland / VM-system allocations should slow
166 * down (but not stop) due to low free pages in the system. This is
167 * typically 1/2 way between min and start.
169 * reserved < severe < minimum < wait < start < target1 < target2
175 globaldata_t gd
= mycpu
;
177 if (__predict_false(gd
->gd_vmstats
.v_paging_wait
>
178 (gd
->gd_vmstats
.v_free_count
+
179 gd
->gd_vmstats
.v_cache_count
)))
183 if (__predict_false(gd
->gd_vmstats
.v_free_reserved
>
184 gd
->gd_vmstats
.v_free_count
))
192 * Return TRUE if the pageout daemon should be started up or continue
193 * running. Available pages have dropped to a level where we need to
194 * think about freeing some up.
196 * Also handles edge cases for required 'actually-free' pages.
198 * reserved < severe < minimum < wait < start < target1 < target2
202 vm_paging_start(int adj
)
204 globaldata_t gd
= mycpu
;
206 if (__predict_false(gd
->gd_vmstats
.v_paging_start
>
207 (gd
->gd_vmstats
.v_free_count
+
208 gd
->gd_vmstats
.v_cache_count
+ adj
)))
212 if (__predict_false(gd
->gd_vmstats
.v_free_min
>
213 gd
->gd_vmstats
.v_free_count
+ adj
))
217 if (__predict_false(gd
->gd_vmstats
.v_free_reserved
>
218 gd
->gd_vmstats
.v_free_count
))
226 * Return TRUE if the pageout daemon has not yet reached its initial target.
227 * The pageout daemon works hard to reach target1.
229 * reserved < severe < minimum < wait < start < target1 < target2
233 vm_paging_target1(void)
235 globaldata_t gd
= mycpu
;
237 if (__predict_false(gd
->gd_vmstats
.v_paging_target1
>
238 (gd
->gd_vmstats
.v_free_count
+
239 gd
->gd_vmstats
.v_cache_count
)))
243 if (__predict_false(gd
->gd_vmstats
.v_free_reserved
>
244 gd
->gd_vmstats
.v_free_count
))
253 vm_paging_target1_count(void)
255 globaldata_t gd
= mycpu
;
258 delta
= gd
->gd_vmstats
.v_paging_target1
-
259 (gd
->gd_vmstats
.v_free_count
+ gd
->gd_vmstats
.v_cache_count
);
264 * Return TRUE if the pageout daemon has not yet reached its final target.
265 * The pageout daemon takes it easy on its way between target1 and target2.
267 * reserved < severe < minimum < wait < start < target1 < target2
271 vm_paging_target2(void)
273 globaldata_t gd
= mycpu
;
275 if (__predict_false(gd
->gd_vmstats
.v_paging_target2
>
276 (gd
->gd_vmstats
.v_free_count
+
277 gd
->gd_vmstats
.v_cache_count
)))
281 if (__predict_false(gd
->gd_vmstats
.v_free_reserved
>
282 gd
->gd_vmstats
.v_free_count
))
291 vm_paging_target2_count(void)
293 globaldata_t gd
= mycpu
;
296 delta
= gd
->gd_vmstats
.v_paging_target2
-
297 (gd
->gd_vmstats
.v_free_count
+ gd
->gd_vmstats
.v_cache_count
);
302 * Returns TRUE if additional pages must be deactivated, either during a
303 * pageout operation or during the page stats scan.
305 * Inactive tests are used in two places. During heavy paging the
306 * inactive_target is used to refill the inactive queue in staged.
307 * Those pages are then ultimately flushed and moved to the cache or free
310 * The inactive queue is also used to manage scans to update page stats
311 * (m->act_count). The page stats scan occurs lazily in small batches to
312 * update m->act_count for pages in the active queue and to move pages
313 * (limited by inactive_target) to the inactive queue. Page stats scanning
314 * and active deactivations only run while the inactive queue is below target.
315 * After this, additional page stats scanning just to update m->act_count
316 * (but not do further deactivations) continues to run for a limited period
317 * of time after any pageout daemon activity.
321 vm_paging_inactive(void)
323 globaldata_t gd
= mycpu
;
325 if (__predict_false((gd
->gd_vmstats
.v_free_count
+
326 gd
->gd_vmstats
.v_cache_count
+
327 gd
->gd_vmstats
.v_inactive_count
) <
328 (gd
->gd_vmstats
.v_free_min
+
329 gd
->gd_vmstats
.v_inactive_target
)))
337 * Return number of pages that need to be deactivated to achieve the inactive
338 * target as a positive number. A negative number indicates that there are
339 * already a sufficient number of inactive pages.
343 vm_paging_inactive_count(void)
345 globaldata_t gd
= mycpu
;
348 delta
= (gd
->gd_vmstats
.v_free_min
+ gd
->gd_vmstats
.v_inactive_target
) -
349 (gd
->gd_vmstats
.v_free_count
+ gd
->gd_vmstats
.v_cache_count
+
350 gd
->gd_vmstats
.v_inactive_count
);
356 * Clear dirty bits in the VM page but truncate the
357 * end to a DEV_BSIZE'd boundary.
359 * Used when reading data in, typically via getpages.
360 * The partial device block at the end of the truncation
361 * range should not lose its dirty bit.
363 * NOTE: This function does not clear the pmap modified bit.
367 vm_page_clear_dirty_end_nonincl(vm_page_t m
, int base
, int size
)
369 size
= (base
+ size
) & ~DEV_BMASK
;
371 vm_page_clear_dirty(m
, base
, size
- base
);
375 * Clear dirty bits in the VM page but truncate the
376 * beginning to a DEV_BSIZE'd boundary.
378 * Used when truncating a buffer. The partial device
379 * block at the beginning of the truncation range
380 * should not lose its dirty bit.
382 * NOTE: This function does not clear the pmap modified bit.
386 vm_page_clear_dirty_beg_nonincl(vm_page_t m
, int base
, int size
)
389 base
= (base
+ DEV_BMASK
) & ~DEV_BMASK
;
391 vm_page_clear_dirty(m
, base
, size
- base
);
396 vm_page_spin_lock(vm_page_t m
)
403 vm_page_spin_unlock(vm_page_t m
)
405 spin_unlock(&m
->spin
);
409 * Wire a vm_page that is already wired. Does not require a busied
414 vm_page_wire_quick(vm_page_t m
)
416 if (atomic_fetchadd_int(&m
->wire_count
, 1) == 0)
417 panic("vm_page_wire_quick: wire_count was 0");
421 * Unwire a vm_page quickly, does not require a busied page.
423 * This routine refuses to drop the wire_count to 0 and will return
424 * TRUE if it would have had to (instead of decrementing it to 0).
425 * The caller can then busy the page and deal with it.
429 vm_page_unwire_quick(vm_page_t m
)
431 KKASSERT(m
->wire_count
> 0);
433 u_int wire_count
= m
->wire_count
;
438 if (atomic_cmpset_int(&m
->wire_count
, wire_count
, wire_count
- 1))
444 * Functions implemented as macros
448 vm_page_flag_set(vm_page_t m
, unsigned int bits
)
450 atomic_set_int(&(m
)->flags
, bits
);
454 vm_page_flag_clear(vm_page_t m
, unsigned int bits
)
456 atomic_clear_int(&(m
)->flags
, bits
);
460 * Wakeup anyone waiting for the page after potentially unbusying
461 * (hard or soft) or doing other work on a page that might make a
462 * waiter ready. The setting of PBUSY_WANTED is integrated into the
463 * related flags and it can't be set once the flags are already
464 * clear, so there should be no races here.
467 vm_page_flash(vm_page_t m
)
469 if (m
->busy_count
& PBUSY_WANTED
) {
470 atomic_clear_int(&m
->busy_count
, PBUSY_WANTED
);
476 * Adjust the soft-busy count on a page. The drop code will issue an
477 * integrated wakeup if busy_count becomes 0.
480 vm_page_sbusy_hold(vm_page_t m
)
482 atomic_add_int(&m
->busy_count
, 1);
486 vm_page_sbusy_drop(vm_page_t m
)
490 ocount
= atomic_fetchadd_int(&m
->busy_count
, -1);
491 if (ocount
- 1 == PBUSY_WANTED
) {
492 /* WANTED and no longer BUSY or SBUSY */
493 atomic_clear_int(&m
->busy_count
, PBUSY_WANTED
);
499 * Reduce the protection of a page. This routine never raises the
500 * protection and therefore can be safely called if the page is already
501 * at VM_PROT_NONE (it will be a NOP effectively ).
503 * VM_PROT_NONE will remove all user mappings of a page. This is often
504 * necessary when a page changes state (for example, turns into a copy-on-write
505 * page or needs to be frozen for write I/O) in order to force a fault, or
506 * to force a page's dirty bits to be synchronized and avoid hardware
507 * (modified/accessed) bit update races with pmap changes.
509 * Since 'prot' is usually a constant, this inline usually winds up optimizing
510 * out the primary conditional.
512 * Must be called with (m) hard-busied.
514 * WARNING: VM_PROT_NONE can block, but will loop until all mappings have
515 * been cleared. Callers should be aware that other page related
516 * elements might have changed, however.
519 vm_page_protect(vm_page_t m
, int prot
)
521 KKASSERT(m
->busy_count
& PBUSY_LOCKED
);
522 if (prot
== VM_PROT_NONE
) {
523 if (pmap_mapped_sync(m
) & (PG_MAPPED
| PG_WRITEABLE
)) {
524 pmap_page_protect(m
, VM_PROT_NONE
);
525 /* PG_WRITEABLE & PG_MAPPED cleared by call */
527 } else if ((prot
== VM_PROT_READ
) &&
528 (m
->flags
& PG_WRITEABLE
) &&
529 (pmap_mapped_sync(m
) & PG_WRITEABLE
)) {
530 pmap_page_protect(m
, VM_PROT_READ
);
531 /* PG_WRITEABLE cleared by call */
536 * Zero-fill the specified page. The entire contents of the page will be
539 static __inline boolean_t
540 vm_page_zero_fill(vm_page_t m
)
542 pmap_zero_page(VM_PAGE_TO_PHYS(m
));
547 * Copy the contents of src_m to dest_m. The pages must be stable but spl
548 * and other protections depend on context.
551 vm_page_copy(vm_page_t src_m
, vm_page_t dest_m
)
553 pmap_copy_page(VM_PAGE_TO_PHYS(src_m
), VM_PAGE_TO_PHYS(dest_m
));
554 dest_m
->valid
= VM_PAGE_BITS_ALL
;
555 dest_m
->dirty
= VM_PAGE_BITS_ALL
;
559 * Free a page. The page must be marked BUSY.
562 vm_page_free(vm_page_t m
)
568 * Free a page to the zerod-pages queue. The caller must ensure that the
569 * page has been zerod.
572 vm_page_free_zero(vm_page_t m
)
576 char *p
= (char *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m
));
579 for (i
= 0; i
< PAGE_SIZE
; i
++) {
581 panic("non-zero page in vm_page_free_zero()");
590 * Set page to not be dirty. Note: does not clear pmap modify bits .
593 vm_page_undirty(vm_page_t m
)
599 #endif /* _VM_VM_PAGE2_H_ */