Print this page
const-ify make segment ops structures
There is no reason to keep the segment ops structures writable.
seg_inherit_notsup is redundant since segop_inherit checks for NULL properly
patch lower-case-segops
instead using SEGOP_* macros, define full-fledged segop_* functions
This will allow us to do some sanity checking or even implement stub
functionality in one place instead of duplicating it wherever these wrappers
are used.
remove whole-process swapping
Long before Unix supported paging, it used process swapping to reclaim
memory. The code is there and in theory it runs when we get *extremely* low
on memory. In practice, it never runs since the definition of low-on-memory
is antiquated. (XXX: define what antiquated means)
You can check the number of swapout/swapin events with kstats:
$ kstat -p ::vm:swapin ::vm:swapout
Split |
Close |
Expand all |
Collapse all |
--- old/usr/src/uts/common/vm/seg.h
+++ new/usr/src/uts/common/vm/seg.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 23 * Use is subject to license terms.
24 24 * Copyright (c) 2015, Joyent, Inc.
25 25 */
26 26
27 27 /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
28 28 /* All Rights Reserved */
29 29
30 30 /*
31 31 * University Copyright- Copyright (c) 1982, 1986, 1988
32 32 * The Regents of the University of California
33 33 * All Rights Reserved
34 34 *
35 35 * University Acknowledgment- Portions of this document are derived from
36 36 * software developed by the University of California, Berkeley, and its
37 37 * contributors.
38 38 */
39 39
40 40 #ifndef _VM_SEG_H
41 41 #define _VM_SEG_H
42 42
43 43 #include <sys/vnode.h>
44 44 #include <sys/avl.h>
45 45 #include <vm/seg_enum.h>
46 46 #include <vm/faultcode.h>
47 47 #include <vm/hat.h>
48 48
49 49 #ifdef __cplusplus
50 50 extern "C" {
51 51 #endif
52 52
53 53 /*
54 54 * VM - Segments.
55 55 */
56 56
57 57 struct anon_map;
58 58
59 59 /*
60 60 * kstat statistics for segment advise
61 61 */
62 62 typedef struct {
63 63 kstat_named_t MADV_FREE_hit;
64 64 kstat_named_t MADV_FREE_miss;
65 65 } segadvstat_t;
66 66
67 67 /*
68 68 * memory object ids
69 69 */
70 70 typedef struct memid { u_longlong_t val[2]; } memid_t;
71 71
72 72 /*
73 73 * An address space contains a set of segments, managed by drivers.
74 74 * Drivers support mapped devices, sharing, copy-on-write, etc.
75 75 *
76 76 * The seg structure contains a lock to prevent races, the base virtual
77 77 * address and size of the segment, a back pointer to the containing
78 78 * address space, pointers to maintain an AVL tree of segments in the
79 79 * same address space, and procedure and data hooks for the driver.
80 80 * The AVL tree of segments for the address space is sorted by
81 81 * ascending base addresses and overlapping segments are not allowed.
82 82 *
83 83 * After a segment is created, faults may occur on pages of the segment.
84 84 * When a fault occurs, the fault handling code must get the desired
85 85 * object and set up the hardware translation to the object. For some
86 86 * objects, the fault handling code also implements copy-on-write.
87 87 *
88 88 * When the hat wants to unload a translation, it can call the unload
89 89 * routine which is responsible for processing reference and modify bits.
90 90 *
91 91 * Each segment is protected by it's containing address space lock. To
92 92 * access any field in the segment structure, the "as" must be locked.
93 93 * If a segment field is to be modified, the address space lock must be
94 94 * write locked.
95 95 */
96 96
97 97 typedef struct pcache_link {
98 98 struct pcache_link *p_lnext;
↓ open down ↓ |
98 lines elided |
↑ open up ↑ |
99 99 struct pcache_link *p_lprev;
100 100 } pcache_link_t;
101 101
102 102 typedef struct seg {
103 103 caddr_t s_base; /* base virtual address */
104 104 size_t s_size; /* size in bytes */
105 105 uint_t s_szc; /* max page size code */
106 106 uint_t s_flags; /* flags for segment, see below */
107 107 struct as *s_as; /* containing address space */
108 108 avl_node_t s_tree; /* AVL tree links to segs in this as */
109 - struct seg_ops *s_ops; /* ops vector: see below */
109 + const struct seg_ops *s_ops; /* ops vector: see below */
110 110 void *s_data; /* private data for instance */
111 111 kmutex_t s_pmtx; /* protects seg's pcache list */
112 112 pcache_link_t s_phead; /* head of seg's pcache list */
113 113 } seg_t;
114 114
115 115 #define S_PURGE (0x01) /* seg should be purged in as_gap() */
116 116
117 117 struct seg_ops {
118 118 int (*dup)(struct seg *, struct seg *);
119 119 int (*unmap)(struct seg *, caddr_t, size_t);
120 120 void (*free)(struct seg *);
121 121 faultcode_t (*fault)(struct hat *, struct seg *, caddr_t, size_t,
122 122 enum fault_type, enum seg_rw);
123 123 faultcode_t (*faulta)(struct seg *, caddr_t);
124 124 int (*setprot)(struct seg *, caddr_t, size_t, uint_t);
125 125 int (*checkprot)(struct seg *, caddr_t, size_t, uint_t);
126 126 int (*kluster)(struct seg *, caddr_t, ssize_t);
127 - size_t (*swapout)(struct seg *);
128 127 int (*sync)(struct seg *, caddr_t, size_t, int, uint_t);
129 128 size_t (*incore)(struct seg *, caddr_t, size_t, char *);
130 129 int (*lockop)(struct seg *, caddr_t, size_t, int, int, ulong_t *,
131 130 size_t);
132 131 int (*getprot)(struct seg *, caddr_t, size_t, uint_t *);
133 132 u_offset_t (*getoffset)(struct seg *, caddr_t);
134 133 int (*gettype)(struct seg *, caddr_t);
135 134 int (*getvp)(struct seg *, caddr_t, struct vnode **);
136 135 int (*advise)(struct seg *, caddr_t, size_t, uint_t);
137 136 void (*dump)(struct seg *);
138 137 int (*pagelock)(struct seg *, caddr_t, size_t, struct page ***,
139 138 enum lock_type, enum seg_rw);
140 139 int (*setpagesize)(struct seg *, caddr_t, size_t, uint_t);
141 140 int (*getmemid)(struct seg *, caddr_t, memid_t *);
142 141 struct lgrp_mem_policy_info *(*getpolicy)(struct seg *, caddr_t);
143 142 int (*capable)(struct seg *, segcapability_t);
144 143 int (*inherit)(struct seg *, caddr_t, size_t, uint_t);
145 144 };
146 145
147 146 #ifdef _KERNEL
148 147
149 148 /*
150 149 * Generic segment operations
151 150 */
152 151 extern void seg_init(void);
153 152 extern struct seg *seg_alloc(struct as *as, caddr_t base, size_t size);
154 153 extern int seg_attach(struct as *as, caddr_t base, size_t size,
155 154 struct seg *seg);
156 155 extern void seg_unmap(struct seg *seg);
157 156 extern void seg_free(struct seg *seg);
158 157
159 158 /*
160 159 * functions for pagelock cache support
161 160 */
162 161 typedef int (*seg_preclaim_cbfunc_t)(void *, caddr_t, size_t,
163 162 struct page **, enum seg_rw, int);
164 163
165 164 extern struct page **seg_plookup(struct seg *seg, struct anon_map *amp,
166 165 caddr_t addr, size_t len, enum seg_rw rw, uint_t flags);
167 166 extern void seg_pinactive(struct seg *seg, struct anon_map *amp,
168 167 caddr_t addr, size_t len, struct page **pp, enum seg_rw rw,
169 168 uint_t flags, seg_preclaim_cbfunc_t callback);
170 169
171 170 extern void seg_ppurge(struct seg *seg, struct anon_map *amp,
172 171 uint_t flags);
173 172 extern void seg_ppurge_wiredpp(struct page **pp);
174 173
175 174 extern int seg_pinsert_check(struct seg *seg, struct anon_map *amp,
176 175 caddr_t addr, size_t len, uint_t flags);
177 176 extern int seg_pinsert(struct seg *seg, struct anon_map *amp,
178 177 caddr_t addr, size_t len, size_t wlen, struct page **pp, enum seg_rw rw,
179 178 uint_t flags, seg_preclaim_cbfunc_t callback);
180 179
181 180 extern void seg_pasync_thread(void);
182 181 extern void seg_preap(void);
183 182 extern int seg_p_disable(void);
184 183 extern void seg_p_enable(void);
185 184
186 185 extern segadvstat_t segadvstat;
187 186
188 187 /*
189 188 * Flags for pagelock cache support.
190 189 * Flags argument is passed as uint_t to pcache routines. upper 16 bits of
191 190 * the flags argument are reserved for alignment page shift when SEGP_PSHIFT
192 191 * is set.
193 192 */
194 193 #define SEGP_FORCE_WIRED 0x1 /* skip check against seg_pwindow */
195 194 #define SEGP_AMP 0x2 /* anon map's pcache entry */
196 195 #define SEGP_PSHIFT 0x4 /* addr pgsz shift for hash function */
197 196
198 197 /*
199 198 * Return values for seg_pinsert and seg_pinsert_check functions.
200 199 */
201 200 #define SEGP_SUCCESS 0 /* seg_pinsert() succeeded */
202 201 #define SEGP_FAIL 1 /* seg_pinsert() failed */
↓ open down ↓ |
65 lines elided |
↑ open up ↑ |
203 202
204 203 /* Page status bits for segop_incore */
205 204 #define SEG_PAGE_INCORE 0x01 /* VA has a page backing it */
206 205 #define SEG_PAGE_LOCKED 0x02 /* VA has a page that is locked */
207 206 #define SEG_PAGE_HASCOW 0x04 /* VA has a page with a copy-on-write */
208 207 #define SEG_PAGE_SOFTLOCK 0x08 /* VA has a page with softlock held */
209 208 #define SEG_PAGE_VNODEBACKED 0x10 /* Segment is backed by a vnode */
210 209 #define SEG_PAGE_ANON 0x20 /* VA has an anonymous page */
211 210 #define SEG_PAGE_VNODE 0x40 /* VA has a vnode page backing it */
212 211
213 -#define SEGOP_DUP(s, n) (*(s)->s_ops->dup)((s), (n))
214 -#define SEGOP_UNMAP(s, a, l) (*(s)->s_ops->unmap)((s), (a), (l))
215 -#define SEGOP_FREE(s) (*(s)->s_ops->free)((s))
216 -#define SEGOP_FAULT(h, s, a, l, t, rw) \
217 - (*(s)->s_ops->fault)((h), (s), (a), (l), (t), (rw))
218 -#define SEGOP_FAULTA(s, a) (*(s)->s_ops->faulta)((s), (a))
219 -#define SEGOP_SETPROT(s, a, l, p) (*(s)->s_ops->setprot)((s), (a), (l), (p))
220 -#define SEGOP_CHECKPROT(s, a, l, p) (*(s)->s_ops->checkprot)((s), (a), (l), (p))
221 -#define SEGOP_KLUSTER(s, a, d) (*(s)->s_ops->kluster)((s), (a), (d))
222 -#define SEGOP_SWAPOUT(s) (*(s)->s_ops->swapout)((s))
223 -#define SEGOP_SYNC(s, a, l, atr, f) \
224 - (*(s)->s_ops->sync)((s), (a), (l), (atr), (f))
225 -#define SEGOP_INCORE(s, a, l, v) (*(s)->s_ops->incore)((s), (a), (l), (v))
226 -#define SEGOP_LOCKOP(s, a, l, atr, op, b, p) \
227 - (*(s)->s_ops->lockop)((s), (a), (l), (atr), (op), (b), (p))
228 -#define SEGOP_GETPROT(s, a, l, p) (*(s)->s_ops->getprot)((s), (a), (l), (p))
229 -#define SEGOP_GETOFFSET(s, a) (*(s)->s_ops->getoffset)((s), (a))
230 -#define SEGOP_GETTYPE(s, a) (*(s)->s_ops->gettype)((s), (a))
231 -#define SEGOP_GETVP(s, a, vpp) (*(s)->s_ops->getvp)((s), (a), (vpp))
232 -#define SEGOP_ADVISE(s, a, l, b) (*(s)->s_ops->advise)((s), (a), (l), (b))
233 -#define SEGOP_DUMP(s) (*(s)->s_ops->dump)((s))
234 -#define SEGOP_PAGELOCK(s, a, l, p, t, rw) \
235 - (*(s)->s_ops->pagelock)((s), (a), (l), (p), (t), (rw))
236 -#define SEGOP_SETPAGESIZE(s, a, l, szc) \
237 - (*(s)->s_ops->setpagesize)((s), (a), (l), (szc))
238 -#define SEGOP_GETMEMID(s, a, mp) (*(s)->s_ops->getmemid)((s), (a), (mp))
239 -#define SEGOP_GETPOLICY(s, a) (*(s)->s_ops->getpolicy)((s), (a))
240 -#define SEGOP_CAPABLE(s, c) (*(s)->s_ops->capable)((s), (c))
241 -#define SEGOP_INHERIT(s, a, l, b) (*(s)->s_ops->inherit)((s), (a), (l), (b))
242 -
243 212 #define seg_page(seg, addr) \
244 213 (((uintptr_t)((addr) - (seg)->s_base)) >> PAGESHIFT)
245 214
246 215 #define seg_pages(seg) \
247 216 (((uintptr_t)((seg)->s_size + PAGEOFFSET)) >> PAGESHIFT)
248 217
249 218 #define IE_NOMEM -1 /* internal to seg layer */
250 219 #define IE_RETRY -2 /* internal to seg layer */
251 220 #define IE_REATTACH -3 /* internal to seg layer */
252 221
253 -/* Values for SEGOP_INHERIT */
222 +/* Values for segop_inherit */
254 223 #define SEGP_INH_ZERO 0x01
255 224
256 -int seg_inherit_notsup(struct seg *, caddr_t, size_t, uint_t);
257 -
258 225 /* Delay/retry factors for seg_p_mem_config_pre_del */
259 226 #define SEGP_PREDEL_DELAY_FACTOR 4
260 227 /*
261 228 * As a workaround to being unable to purge the pagelock
262 229 * cache during a DR delete memory operation, we use
263 230 * a stall threshold that is twice the maximum seen
264 231 * during testing. This workaround will be removed
265 232 * when a suitable fix is found.
266 233 */
267 234 #define SEGP_STALL_SECONDS 25
268 235 #define SEGP_STALL_THRESHOLD \
269 236 (SEGP_STALL_SECONDS * SEGP_PREDEL_DELAY_FACTOR)
↓ open down ↓ |
2 lines elided |
↑ open up ↑ |
270 237
271 238 #ifdef VMDEBUG
272 239
273 240 uint_t seg_page(struct seg *, caddr_t);
274 241 uint_t seg_pages(struct seg *);
275 242
276 243 #endif /* VMDEBUG */
277 244
278 245 boolean_t seg_can_change_zones(struct seg *);
279 246 size_t seg_swresv(struct seg *);
247 +
248 +/* segop wrappers */
249 +int segop_dup(struct seg *, struct seg *);
250 +int segop_unmap(struct seg *, caddr_t, size_t);
251 +void segop_free(struct seg *);
252 +faultcode_t segop_fault(struct hat *, struct seg *, caddr_t, size_t, enum fault_type, enum seg_rw);
253 +faultcode_t segop_faulta(struct seg *, caddr_t);
254 +int segop_setprot(struct seg *, caddr_t, size_t, uint_t);
255 +int segop_checkprot(struct seg *, caddr_t, size_t, uint_t);
256 +int segop_kluster(struct seg *, caddr_t, ssize_t);
257 +int segop_sync(struct seg *, caddr_t, size_t, int, uint_t);
258 +size_t segop_incore(struct seg *, caddr_t, size_t, char *);
259 +int segop_lockop(struct seg *, caddr_t, size_t, int, int, ulong_t *, size_t );
260 +int segop_getprot(struct seg *, caddr_t, size_t, uint_t *);
261 +u_offset_t segop_getoffset(struct seg *, caddr_t);
262 +int segop_gettype(struct seg *, caddr_t);
263 +int segop_getvp(struct seg *, caddr_t, struct vnode **);
264 +int segop_advise(struct seg *, caddr_t, size_t, uint_t);
265 +void segop_dump(struct seg *);
266 +int segop_pagelock(struct seg *, caddr_t, size_t, struct page ***, enum lock_type, enum seg_rw);
267 +int segop_setpagesize(struct seg *, caddr_t, size_t, uint_t);
268 +int segop_getmemid(struct seg *, caddr_t, memid_t *);
269 +struct lgrp_mem_policy_info *segop_getpolicy(struct seg *, caddr_t);
270 +int segop_capable(struct seg *, segcapability_t);
271 +int segop_inherit(struct seg *, caddr_t, size_t, uint_t);
280 272
281 273 #endif /* _KERNEL */
282 274
283 275 #ifdef __cplusplus
284 276 }
285 277 #endif
286 278
287 279 #endif /* _VM_SEG_H */
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX