xref: /freebsd/sys/vm/vm_pagequeue.h (revision 7bda9663949a80e4e56006369d6df8dc8eeb6cff)
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 } __aligned(CACHE_LINE_SIZE);
75 
76 #ifndef VM_BATCHQUEUE_SIZE
77 #define	VM_BATCHQUEUE_SIZE	7
78 #endif
79 
80 struct vm_batchqueue {
81 	vm_page_t	bq_pa[VM_BATCHQUEUE_SIZE];
82 	int		bq_cnt;
83 } __aligned(CACHE_LINE_SIZE);
84 
85 #include <vm/uma.h>
86 #include <sys/pidctrl.h>
87 struct sysctl_oid;
88 
89 /*
90  * One vm_domain per-numa domain.  Contains pagequeues, free page structures,
91  * and accounting.
92  *
93  * Lock Key:
94  * f	vmd_free_mtx
95  * p	vmd_pageout_mtx
96  * d	vm_domainset_lock
97  * a	atomic
98  * c	const after boot
99  * q	page queue lock
100 */
101 struct vm_domain {
102 	struct vm_pagequeue vmd_pagequeues[PQ_COUNT];
103 	struct mtx_padalign vmd_free_mtx;
104 	struct mtx_padalign vmd_pageout_mtx;
105 	uma_zone_t vmd_pgcache;		/* (c) page free cache. */
106 	struct vmem *vmd_kernel_arena;	/* (c) per-domain kva R/W arena. */
107 	struct vmem *vmd_kernel_rwx_arena; /* (c) per-domain kva R/W/X arena. */
108 	u_int vmd_domain;		/* (c) Domain number. */
109 	u_int vmd_page_count;		/* (c) Total page count. */
110 	long vmd_segs;			/* (c) bitmask of the segments */
111 	u_int __aligned(CACHE_LINE_SIZE) vmd_free_count; /* (a,f) free page count */
112 	u_int vmd_pageout_deficit;	/* (a) Estimated number of pages deficit */
113 	uint8_t vmd_pad[CACHE_LINE_SIZE - (sizeof(u_int) * 2)];
114 
115 	/* Paging control variables, used within single threaded page daemon. */
116 	struct pidctrl vmd_pid;		/* Pageout controller. */
117 	boolean_t vmd_oom;
118 	int vmd_oom_seq;
119 	int vmd_last_active_scan;
120 	struct vm_page vmd_markers[PQ_COUNT]; /* (q) markers for queue scans */
121 	struct vm_page vmd_inacthead; /* marker for LRU-defeating insertions */
122 	struct vm_page vmd_clock[2]; /* markers for active queue scan */
123 
124 	int vmd_pageout_wanted;		/* (a, p) pageout daemon wait channel */
125 	int vmd_pageout_pages_needed;	/* (d) page daemon waiting for pages? */
126 	bool vmd_minset;		/* (d) Are we in vm_min_domains? */
127 	bool vmd_severeset;		/* (d) Are we in vm_severe_domains? */
128 	enum {
129 		VM_LAUNDRY_IDLE = 0,
130 		VM_LAUNDRY_BACKGROUND,
131 		VM_LAUNDRY_SHORTFALL
132 	} vmd_laundry_request;
133 
134 	/* Paging thresholds and targets. */
135 	u_int vmd_clean_pages_freed;	/* (q) accumulator for laundry thread */
136 	u_int vmd_background_launder_target; /* (c) */
137 	u_int vmd_free_reserved;	/* (c) pages reserved for deadlock */
138 	u_int vmd_free_target;		/* (c) pages desired free */
139 	u_int vmd_free_min;		/* (c) pages desired free */
140 	u_int vmd_inactive_target;	/* (c) pages desired inactive */
141 	u_int vmd_pageout_free_min;	/* (c) min pages reserved for kernel */
142 	u_int vmd_pageout_wakeup_thresh;/* (c) min pages to wake pagedaemon */
143 	u_int vmd_interrupt_free_min;	/* (c) reserved pages for int code */
144 	u_int vmd_free_severe;		/* (c) severe page depletion point */
145 
146 	/* Name for sysctl etc. */
147 	struct sysctl_oid *vmd_oid;
148 	char vmd_name[sizeof(__XSTRING(MAXMEMDOM))];
149 } __aligned(CACHE_LINE_SIZE);
150 
151 extern struct vm_domain vm_dom[MAXMEMDOM];
152 
153 #define	VM_DOMAIN(n)	(&vm_dom[(n)])
154 
155 #define	vm_pagequeue_assert_locked(pq)	mtx_assert(&(pq)->pq_mutex, MA_OWNED)
156 #define	vm_pagequeue_lock(pq)		mtx_lock(&(pq)->pq_mutex)
157 #define	vm_pagequeue_lockptr(pq)	(&(pq)->pq_mutex)
158 #define	vm_pagequeue_trylock(pq)	mtx_trylock(&(pq)->pq_mutex)
159 #define	vm_pagequeue_unlock(pq)		mtx_unlock(&(pq)->pq_mutex)
160 
161 #define	vm_domain_free_assert_locked(n)					\
162 	    mtx_assert(vm_domain_free_lockptr((n)), MA_OWNED)
163 #define	vm_domain_free_assert_unlocked(n)				\
164 	    mtx_assert(vm_domain_free_lockptr((n)), MA_NOTOWNED)
165 #define	vm_domain_free_lock(d)						\
166 	    mtx_lock(vm_domain_free_lockptr((d)))
167 #define	vm_domain_free_lockptr(d)					\
168 	    (&(d)->vmd_free_mtx)
169 #define	vm_domain_free_trylock(d)					\
170 	    mtx_trylock(vm_domain_free_lockptr((d)))
171 #define	vm_domain_free_unlock(d)					\
172 	    mtx_unlock(vm_domain_free_lockptr((d)))
173 
174 #define	vm_domain_pageout_lockptr(d)					\
175 	    (&(d)->vmd_pageout_mtx)
176 #define	vm_domain_pageout_assert_locked(n)				\
177 	    mtx_assert(vm_domain_pageout_lockptr((n)), MA_OWNED)
178 #define	vm_domain_pageout_assert_unlocked(n)				\
179 	    mtx_assert(vm_domain_pageout_lockptr((n)), MA_NOTOWNED)
180 #define	vm_domain_pageout_lock(d)					\
181 	    mtx_lock(vm_domain_pageout_lockptr((d)))
182 #define	vm_domain_pageout_unlock(d)					\
183 	    mtx_unlock(vm_domain_pageout_lockptr((d)))
184 
185 static __inline void
186 vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend)
187 {
188 
189 	vm_pagequeue_assert_locked(pq);
190 	pq->pq_cnt += addend;
191 }
192 #define	vm_pagequeue_cnt_inc(pq)	vm_pagequeue_cnt_add((pq), 1)
193 #define	vm_pagequeue_cnt_dec(pq)	vm_pagequeue_cnt_add((pq), -1)
194 
195 static inline void
196 vm_batchqueue_init(struct vm_batchqueue *bq)
197 {
198 
199 	bq->bq_cnt = 0;
200 }
201 
202 static inline bool
203 vm_batchqueue_insert(struct vm_batchqueue *bq, vm_page_t m)
204 {
205 
206 	if (bq->bq_cnt < nitems(bq->bq_pa)) {
207 		bq->bq_pa[bq->bq_cnt++] = m;
208 		return (true);
209 	}
210 	return (false);
211 }
212 
213 static inline vm_page_t
214 vm_batchqueue_pop(struct vm_batchqueue *bq)
215 {
216 
217 	if (bq->bq_cnt == 0)
218 		return (NULL);
219 	return (bq->bq_pa[--bq->bq_cnt]);
220 }
221 
222 void vm_domain_set(struct vm_domain *vmd);
223 void vm_domain_clear(struct vm_domain *vmd);
224 int vm_domain_allocate(struct vm_domain *vmd, int req, int npages);
225 
226 /*
227  *      vm_pagequeue_domain:
228  *
229  *      Return the memory domain the page belongs to.
230  */
231 static inline struct vm_domain *
232 vm_pagequeue_domain(vm_page_t m)
233 {
234 
235 	return (VM_DOMAIN(vm_phys_domain(m)));
236 }
237 
238 /*
239  * Return the number of pages we need to free-up or cache
240  * A positive number indicates that we do not have enough free pages.
241  */
242 static inline int
243 vm_paging_target(struct vm_domain *vmd)
244 {
245 
246 	return (vmd->vmd_free_target - vmd->vmd_free_count);
247 }
248 
249 /*
250  * Returns TRUE if the pagedaemon needs to be woken up.
251  */
252 static inline int
253 vm_paging_needed(struct vm_domain *vmd, u_int free_count)
254 {
255 
256 	return (free_count < vmd->vmd_pageout_wakeup_thresh);
257 }
258 
259 /*
260  * Returns TRUE if the domain is below the min paging target.
261  */
262 static inline int
263 vm_paging_min(struct vm_domain *vmd)
264 {
265 
266         return (vmd->vmd_free_min > vmd->vmd_free_count);
267 }
268 
269 /*
270  * Returns TRUE if the domain is below the severe paging target.
271  */
272 static inline int
273 vm_paging_severe(struct vm_domain *vmd)
274 {
275 
276         return (vmd->vmd_free_severe > vmd->vmd_free_count);
277 }
278 
279 /*
280  * Return the number of pages we need to launder.
281  * A positive number indicates that we have a shortfall of clean pages.
282  */
283 static inline int
284 vm_laundry_target(struct vm_domain *vmd)
285 {
286 
287 	return (vm_paging_target(vmd));
288 }
289 
290 void pagedaemon_wakeup(int domain);
291 
292 static inline void
293 vm_domain_freecnt_inc(struct vm_domain *vmd, int adj)
294 {
295 	u_int old, new;
296 
297 	old = atomic_fetchadd_int(&vmd->vmd_free_count, adj);
298 	new = old + adj;
299 	/*
300 	 * Only update bitsets on transitions.  Notice we short-circuit the
301 	 * rest of the checks if we're above min already.
302 	 */
303 	if (old < vmd->vmd_free_min && (new >= vmd->vmd_free_min ||
304 	    (old < vmd->vmd_free_severe && new >= vmd->vmd_free_severe) ||
305 	    (old < vmd->vmd_pageout_free_min &&
306 	    new >= vmd->vmd_pageout_free_min)))
307 		vm_domain_clear(vmd);
308 }
309 
310 #endif	/* _KERNEL */
311 #endif				/* !_VM_PAGEQUEUE_ */
312