xref: /freebsd/sys/vm/vm_pagequeue.h (revision d6eb98610fa65663bf0df4574b7cb2c5c4ffda71)
1 /*-
2  * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
3  *
4  * Copyright (c) 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * The Mach Operating System project at Carnegie-Mellon University.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from: @(#)vm_page.h	8.2 (Berkeley) 12/13/93
35  *
36  *
37  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38  * All rights reserved.
39  *
40  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41  *
42  * Permission to use, copy, modify and distribute this software and
43  * its documentation is hereby granted, provided that both the copyright
44  * notice and this permission notice appear in all copies of the
45  * software, derivative works or modified versions, and any portions
46  * thereof, and that both notices appear in supporting documentation.
47  *
48  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51  *
52  * Carnegie Mellon requests users of this software to return to
53  *
54  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55  *  School of Computer Science
56  *  Carnegie Mellon University
57  *  Pittsburgh PA 15213-3890
58  *
59  * any improvements or extensions that they make and grant Carnegie the
60  * rights to redistribute these changes.
61  *
62  * $FreeBSD$
63  */
64 
65 #ifndef	_VM_PAGEQUEUE_
66 #define	_VM_PAGEQUEUE_
67 
68 #ifdef _KERNEL
69 struct vm_pagequeue {
70 	struct mtx	pq_mutex;
71 	struct pglist	pq_pl;
72 	int		pq_cnt;
73 	const char	* const pq_name;
74 	uint64_t	pq_pdpages;
75 } __aligned(CACHE_LINE_SIZE);
76 
77 #ifndef VM_BATCHQUEUE_SIZE
78 #define	VM_BATCHQUEUE_SIZE	7
79 #endif
80 
81 struct vm_batchqueue {
82 	vm_page_t	bq_pa[VM_BATCHQUEUE_SIZE];
83 	int		bq_cnt;
84 } __aligned(CACHE_LINE_SIZE);
85 
86 #include <vm/uma.h>
87 #include <sys/pidctrl.h>
88 struct sysctl_oid;
89 
90 /*
91  * One vm_domain per-numa domain.  Contains pagequeues, free page structures,
92  * and accounting.
93  *
94  * Lock Key:
95  * f	vmd_free_mtx
96  * p	vmd_pageout_mtx
97  * d	vm_domainset_lock
98  * a	atomic
99  * c	const after boot
100  * q	page queue lock
101 */
102 struct vm_domain {
103 	struct vm_pagequeue vmd_pagequeues[PQ_COUNT];
104 	struct mtx_padalign vmd_free_mtx;
105 	struct mtx_padalign vmd_pageout_mtx;
106 	struct vm_pgcache {
107 		int domain;
108 		int pool;
109 		uma_zone_t zone;
110 	} vmd_pgcache[VM_NFREEPOOL];
111 	struct vmem *vmd_kernel_arena;	/* (c) per-domain kva R/W arena. */
112 	struct vmem *vmd_kernel_rwx_arena; /* (c) per-domain kva R/W/X arena. */
113 	u_int vmd_domain;		/* (c) Domain number. */
114 	u_int vmd_page_count;		/* (c) Total page count. */
115 	long vmd_segs;			/* (c) bitmask of the segments */
116 	u_int __aligned(CACHE_LINE_SIZE) vmd_free_count; /* (a,f) free page count */
117 	u_int vmd_pageout_deficit;	/* (a) Estimated number of pages deficit */
118 	uint8_t vmd_pad[CACHE_LINE_SIZE - (sizeof(u_int) * 2)];
119 
120 	/* Paging control variables, used within single threaded page daemon. */
121 	struct pidctrl vmd_pid;		/* Pageout controller. */
122 	boolean_t vmd_oom;
123 	int vmd_oom_seq;
124 	int vmd_last_active_scan;
125 	struct vm_page vmd_markers[PQ_COUNT]; /* (q) markers for queue scans */
126 	struct vm_page vmd_inacthead; /* marker for LRU-defeating insertions */
127 	struct vm_page vmd_clock[2]; /* markers for active queue scan */
128 
129 	int vmd_pageout_wanted;		/* (a, p) pageout daemon wait channel */
130 	int vmd_pageout_pages_needed;	/* (d) page daemon waiting for pages? */
131 	bool vmd_minset;		/* (d) Are we in vm_min_domains? */
132 	bool vmd_severeset;		/* (d) Are we in vm_severe_domains? */
133 	enum {
134 		VM_LAUNDRY_IDLE = 0,
135 		VM_LAUNDRY_BACKGROUND,
136 		VM_LAUNDRY_SHORTFALL
137 	} vmd_laundry_request;
138 
139 	/* Paging thresholds and targets. */
140 	u_int vmd_clean_pages_freed;	/* (q) accumulator for laundry thread */
141 	u_int vmd_background_launder_target; /* (c) */
142 	u_int vmd_free_reserved;	/* (c) pages reserved for deadlock */
143 	u_int vmd_free_target;		/* (c) pages desired free */
144 	u_int vmd_free_min;		/* (c) pages desired free */
145 	u_int vmd_inactive_target;	/* (c) pages desired inactive */
146 	u_int vmd_pageout_free_min;	/* (c) min pages reserved for kernel */
147 	u_int vmd_pageout_wakeup_thresh;/* (c) min pages to wake pagedaemon */
148 	u_int vmd_interrupt_free_min;	/* (c) reserved pages for int code */
149 	u_int vmd_free_severe;		/* (c) severe page depletion point */
150 
151 	/* Name for sysctl etc. */
152 	struct sysctl_oid *vmd_oid;
153 	char vmd_name[sizeof(__XSTRING(MAXMEMDOM))];
154 } __aligned(CACHE_LINE_SIZE);
155 
156 extern struct vm_domain vm_dom[MAXMEMDOM];
157 
158 #define	VM_DOMAIN(n)		(&vm_dom[(n)])
159 #define	VM_DOMAIN_EMPTY(n)	(vm_dom[(n)].vmd_page_count == 0)
160 
161 #define	vm_pagequeue_assert_locked(pq)	mtx_assert(&(pq)->pq_mutex, MA_OWNED)
162 #define	vm_pagequeue_lock(pq)		mtx_lock(&(pq)->pq_mutex)
163 #define	vm_pagequeue_lockptr(pq)	(&(pq)->pq_mutex)
164 #define	vm_pagequeue_trylock(pq)	mtx_trylock(&(pq)->pq_mutex)
165 #define	vm_pagequeue_unlock(pq)		mtx_unlock(&(pq)->pq_mutex)
166 
167 #define	vm_domain_free_assert_locked(n)					\
168 	    mtx_assert(vm_domain_free_lockptr((n)), MA_OWNED)
169 #define	vm_domain_free_assert_unlocked(n)				\
170 	    mtx_assert(vm_domain_free_lockptr((n)), MA_NOTOWNED)
171 #define	vm_domain_free_lock(d)						\
172 	    mtx_lock(vm_domain_free_lockptr((d)))
173 #define	vm_domain_free_lockptr(d)					\
174 	    (&(d)->vmd_free_mtx)
175 #define	vm_domain_free_trylock(d)					\
176 	    mtx_trylock(vm_domain_free_lockptr((d)))
177 #define	vm_domain_free_unlock(d)					\
178 	    mtx_unlock(vm_domain_free_lockptr((d)))
179 
180 #define	vm_domain_pageout_lockptr(d)					\
181 	    (&(d)->vmd_pageout_mtx)
182 #define	vm_domain_pageout_assert_locked(n)				\
183 	    mtx_assert(vm_domain_pageout_lockptr((n)), MA_OWNED)
184 #define	vm_domain_pageout_assert_unlocked(n)				\
185 	    mtx_assert(vm_domain_pageout_lockptr((n)), MA_NOTOWNED)
186 #define	vm_domain_pageout_lock(d)					\
187 	    mtx_lock(vm_domain_pageout_lockptr((d)))
188 #define	vm_domain_pageout_unlock(d)					\
189 	    mtx_unlock(vm_domain_pageout_lockptr((d)))
190 
191 static __inline void
192 vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend)
193 {
194 
195 	vm_pagequeue_assert_locked(pq);
196 	pq->pq_cnt += addend;
197 }
198 #define	vm_pagequeue_cnt_inc(pq)	vm_pagequeue_cnt_add((pq), 1)
199 #define	vm_pagequeue_cnt_dec(pq)	vm_pagequeue_cnt_add((pq), -1)
200 
201 static inline void
202 vm_batchqueue_init(struct vm_batchqueue *bq)
203 {
204 
205 	bq->bq_cnt = 0;
206 }
207 
208 static inline bool
209 vm_batchqueue_insert(struct vm_batchqueue *bq, vm_page_t m)
210 {
211 
212 	if (bq->bq_cnt < nitems(bq->bq_pa)) {
213 		bq->bq_pa[bq->bq_cnt++] = m;
214 		return (true);
215 	}
216 	return (false);
217 }
218 
219 static inline vm_page_t
220 vm_batchqueue_pop(struct vm_batchqueue *bq)
221 {
222 
223 	if (bq->bq_cnt == 0)
224 		return (NULL);
225 	return (bq->bq_pa[--bq->bq_cnt]);
226 }
227 
228 void vm_domain_set(struct vm_domain *vmd);
229 void vm_domain_clear(struct vm_domain *vmd);
230 int vm_domain_allocate(struct vm_domain *vmd, int req, int npages);
231 
232 /*
233  *      vm_pagequeue_domain:
234  *
235  *      Return the memory domain the page belongs to.
236  */
237 static inline struct vm_domain *
238 vm_pagequeue_domain(vm_page_t m)
239 {
240 
241 	return (VM_DOMAIN(vm_phys_domain(m)));
242 }
243 
244 /*
245  * Return the number of pages we need to free-up or cache
246  * A positive number indicates that we do not have enough free pages.
247  */
248 static inline int
249 vm_paging_target(struct vm_domain *vmd)
250 {
251 
252 	return (vmd->vmd_free_target - vmd->vmd_free_count);
253 }
254 
255 /*
256  * Returns TRUE if the pagedaemon needs to be woken up.
257  */
258 static inline int
259 vm_paging_needed(struct vm_domain *vmd, u_int free_count)
260 {
261 
262 	return (free_count < vmd->vmd_pageout_wakeup_thresh);
263 }
264 
265 /*
266  * Returns TRUE if the domain is below the min paging target.
267  */
268 static inline int
269 vm_paging_min(struct vm_domain *vmd)
270 {
271 
272         return (vmd->vmd_free_min > vmd->vmd_free_count);
273 }
274 
275 /*
276  * Returns TRUE if the domain is below the severe paging target.
277  */
278 static inline int
279 vm_paging_severe(struct vm_domain *vmd)
280 {
281 
282         return (vmd->vmd_free_severe > vmd->vmd_free_count);
283 }
284 
285 /*
286  * Return the number of pages we need to launder.
287  * A positive number indicates that we have a shortfall of clean pages.
288  */
289 static inline int
290 vm_laundry_target(struct vm_domain *vmd)
291 {
292 
293 	return (vm_paging_target(vmd));
294 }
295 
296 void pagedaemon_wakeup(int domain);
297 
298 static inline void
299 vm_domain_freecnt_inc(struct vm_domain *vmd, int adj)
300 {
301 	u_int old, new;
302 
303 	old = atomic_fetchadd_int(&vmd->vmd_free_count, adj);
304 	new = old + adj;
305 	/*
306 	 * Only update bitsets on transitions.  Notice we short-circuit the
307 	 * rest of the checks if we're above min already.
308 	 */
309 	if (old < vmd->vmd_free_min && (new >= vmd->vmd_free_min ||
310 	    (old < vmd->vmd_free_severe && new >= vmd->vmd_free_severe) ||
311 	    (old < vmd->vmd_pageout_free_min &&
312 	    new >= vmd->vmd_pageout_free_min)))
313 		vm_domain_clear(vmd);
314 }
315 
316 #endif	/* _KERNEL */
317 #endif				/* !_VM_PAGEQUEUE_ */
318