xref: /freebsd/sys/vm/vm_pagequeue.h (revision 52f72944b8f5abb2386eae924357dee8aea17d5b)
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 #include <sys/pidctrl.h>
77 struct sysctl_oid;
78 
79 /*
80  * One vm_domain per-numa domain.  Contains pagequeues, free page structures,
81  * and accounting.
82  *
83  * Lock Key:
84  * f   vmd_free_mtx
85  * p   vmd_pageout_mtx
86  * d   vm_domainset_lock
87  * a   atomic
88  * c   const after boot
89 */
90 struct vm_domain {
91 	struct vm_pagequeue vmd_pagequeues[PQ_COUNT];
92 	struct mtx_padalign vmd_free_mtx;
93 	struct mtx_padalign vmd_pageout_mtx;
94 	struct vmem *vmd_kernel_arena;	/* (c) per-domain kva arena. */
95 	u_int vmd_domain;		/* (c) Domain number. */
96 	u_int vmd_page_count;		/* (c) Total page count. */
97 	long vmd_segs;			/* (c) bitmask of the segments */
98 	u_int __aligned(CACHE_LINE_SIZE) vmd_free_count; /* (a,f) free page count */
99 	u_int vmd_pageout_deficit;	/* (a) Estimated number of pages deficit */
100 	uint8_t vmd_pad[CACHE_LINE_SIZE - (sizeof(u_int) * 2)];
101 
102 	/* Paging control variables, used within single threaded page daemon. */
103 	struct pidctrl vmd_pid;		/* Pageout controller. */
104 	boolean_t vmd_oom;
105 	int vmd_oom_seq;
106 	int vmd_last_active_scan;
107 	struct vm_page vmd_laundry_marker;
108 	struct vm_page vmd_marker; /* marker for pagedaemon private use */
109 	struct vm_page vmd_inacthead; /* marker for LRU-defeating insertions */
110 
111 	int vmd_pageout_wanted;		/* (a, p) pageout daemon wait channel */
112 	int vmd_pageout_pages_needed;	/* (d) page daemon waiting for pages? */
113 	bool vmd_minset;		/* (d) Are we in vm_min_domains? */
114 	bool vmd_severeset;		/* (d) Are we in vm_severe_domains? */
115 	int vmd_inactq_scans;
116 	enum {
117 		VM_LAUNDRY_IDLE = 0,
118 		VM_LAUNDRY_BACKGROUND,
119 		VM_LAUNDRY_SHORTFALL
120 	} vmd_laundry_request;
121 
122 	/* Paging thresholds. */
123 	u_int vmd_background_launder_target;
124 	u_int vmd_free_reserved;	/* (c) pages reserved for deadlock */
125 	u_int vmd_free_target;		/* (c) pages desired free */
126 	u_int vmd_free_min;		/* (c) pages desired free */
127 	u_int vmd_inactive_target;	/* (c) pages desired inactive */
128 	u_int vmd_pageout_free_min;	/* (c) min pages reserved for kernel */
129 	u_int vmd_pageout_wakeup_thresh;/* (c) min pages to wake pagedaemon */
130 	u_int vmd_interrupt_free_min;	/* (c) reserved pages for int code */
131 	u_int vmd_free_severe;		/* (c) severe page depletion point */
132 
133 	/* Name for sysctl etc. */
134 	struct sysctl_oid *vmd_oid;
135 	char vmd_name[sizeof(__XSTRING(MAXMEMDOM))];
136 } __aligned(CACHE_LINE_SIZE);
137 
138 extern struct vm_domain vm_dom[MAXMEMDOM];
139 
140 #define	VM_DOMAIN(n)	(&vm_dom[(n)])
141 
142 #define	vm_pagequeue_assert_locked(pq)	mtx_assert(&(pq)->pq_mutex, MA_OWNED)
143 #define	vm_pagequeue_lock(pq)		mtx_lock(&(pq)->pq_mutex)
144 #define	vm_pagequeue_lockptr(pq)	(&(pq)->pq_mutex)
145 #define	vm_pagequeue_unlock(pq)		mtx_unlock(&(pq)->pq_mutex)
146 
147 #define	vm_domain_free_assert_locked(n)					\
148 	    mtx_assert(vm_domain_free_lockptr((n)), MA_OWNED)
149 #define	vm_domain_free_assert_unlocked(n)				\
150 	    mtx_assert(vm_domain_free_lockptr((n)), MA_NOTOWNED)
151 #define	vm_domain_free_lock(d)						\
152 	    mtx_lock(vm_domain_free_lockptr((d)))
153 #define	vm_domain_free_lockptr(d)					\
154 	    (&(d)->vmd_free_mtx)
155 #define	vm_domain_free_unlock(d)					\
156 	    mtx_unlock(vm_domain_free_lockptr((d)))
157 
158 #define	vm_domain_pageout_lockptr(d)					\
159 	    (&(d)->vmd_pageout_mtx)
160 #define	vm_domain_pageout_assert_locked(n)				\
161 	    mtx_assert(vm_domain_pageout_lockptr((n)), MA_OWNED)
162 #define	vm_domain_pageout_assert_unlocked(n)				\
163 	    mtx_assert(vm_domain_pageout_lockptr((n)), MA_NOTOWNED)
164 #define	vm_domain_pageout_lock(d)					\
165 	    mtx_lock(vm_domain_pageout_lockptr((d)))
166 #define	vm_domain_pageout_unlock(d)					\
167 	    mtx_unlock(vm_domain_pageout_lockptr((d)))
168 
169 static __inline void
170 vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend)
171 {
172 
173 #ifdef notyet
174 	vm_pagequeue_assert_locked(pq);
175 #endif
176 	pq->pq_cnt += addend;
177 }
178 #define	vm_pagequeue_cnt_inc(pq)	vm_pagequeue_cnt_add((pq), 1)
179 #define	vm_pagequeue_cnt_dec(pq)	vm_pagequeue_cnt_add((pq), -1)
180 
181 void vm_domain_set(struct vm_domain *vmd);
182 void vm_domain_clear(struct vm_domain *vmd);
183 int vm_domain_allocate(struct vm_domain *vmd, int req, int npages);
184 
185 /*
186  *      vm_pagequeue_domain:
187  *
188  *      Return the memory domain the page belongs to.
189  */
190 static inline struct vm_domain *
191 vm_pagequeue_domain(vm_page_t m)
192 {
193 
194 	return (VM_DOMAIN(vm_phys_domain(m)));
195 }
196 
197 /*
198  * Return the number of pages we need to free-up or cache
199  * A positive number indicates that we do not have enough free pages.
200  */
201 static inline int
202 vm_paging_target(struct vm_domain *vmd)
203 {
204 
205 	return (vmd->vmd_free_target - vmd->vmd_free_count);
206 }
207 
208 /*
209  * Returns TRUE if the pagedaemon needs to be woken up.
210  */
211 static inline int
212 vm_paging_needed(struct vm_domain *vmd, u_int free_count)
213 {
214 
215 	return (free_count < vmd->vmd_pageout_wakeup_thresh);
216 }
217 
218 /*
219  * Returns TRUE if the domain is below the min paging target.
220  */
221 static inline int
222 vm_paging_min(struct vm_domain *vmd)
223 {
224 
225         return (vmd->vmd_free_min > vmd->vmd_free_count);
226 }
227 
228 /*
229  * Returns TRUE if the domain is below the severe paging target.
230  */
231 static inline int
232 vm_paging_severe(struct vm_domain *vmd)
233 {
234 
235         return (vmd->vmd_free_severe > vmd->vmd_free_count);
236 }
237 
238 /*
239  * Return the number of pages we need to launder.
240  * A positive number indicates that we have a shortfall of clean pages.
241  */
242 static inline int
243 vm_laundry_target(struct vm_domain *vmd)
244 {
245 
246 	return (vm_paging_target(vmd));
247 }
248 
249 void pagedaemon_wakeup(int domain);
250 
251 static inline void
252 vm_domain_freecnt_inc(struct vm_domain *vmd, int adj)
253 {
254 	u_int old, new;
255 
256 	old = atomic_fetchadd_int(&vmd->vmd_free_count, adj);
257 	new = old + adj;
258 	/*
259 	 * Only update bitsets on transitions.  Notice we short-circuit the
260 	 * rest of the checks if we're above min already.
261 	 */
262 	if (old < vmd->vmd_free_min && (new >= vmd->vmd_free_min ||
263 	    (old < vmd->vmd_free_severe && new >= vmd->vmd_free_severe) ||
264 	    (old < vmd->vmd_pageout_free_min &&
265 	    new >= vmd->vmd_pageout_free_min)))
266 		vm_domain_clear(vmd);
267 }
268 
269 #endif	/* _KERNEL */
270 #endif				/* !_VM_PAGEQUEUE_ */
271