xref: /titanic_50/usr/src/uts/common/vm/seg.h (revision 67e3a03ed4a2813074d36330f062ed6e593a4937)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 /*
30  * University Copyright- Copyright (c) 1982, 1986, 1988
31  * The Regents of the University of California
32  * All Rights Reserved
33  *
34  * University Acknowledgment- Portions of this document are derived from
35  * software developed by the University of California, Berkeley, and its
36  * contributors.
37  */
38 
39 #ifndef	_VM_SEG_H
40 #define	_VM_SEG_H
41 
42 #pragma ident	"%Z%%M%	%I%	%E% SMI"
43 
44 #include <sys/vnode.h>
45 #include <sys/avl.h>
46 #include <vm/seg_enum.h>
47 #include <vm/faultcode.h>
48 #include <vm/hat.h>
49 
50 #ifdef	__cplusplus
51 extern "C" {
52 #endif
53 
54 /*
55  * VM - Segments.
56  */
57 
58 /*
59  * kstat statistics for segment advise
60  */
61 typedef struct {
62 	kstat_named_t MADV_FREE_hit;
63 	kstat_named_t MADV_FREE_miss;
64 } segadvstat_t;
65 
66 /*
67  * memory object ids
68  */
69 typedef struct memid { u_longlong_t val[2]; } memid_t;
70 
71 /*
72  * An address space contains a set of segments, managed by drivers.
73  * Drivers support mapped devices, sharing, copy-on-write, etc.
74  *
75  * The seg structure contains a lock to prevent races, the base virtual
76  * address and size of the segment, a back pointer to the containing
77  * address space, pointers to maintain an AVL tree of segments in the
78  * same address space, and procedure and data hooks for the driver.
79  * The AVL tree of segments for the address space is sorted by
80  * ascending base addresses and overlapping segments are not allowed.
81  *
82  * After a segment is created, faults may occur on pages of the segment.
83  * When a fault occurs, the fault handling code must get the desired
84  * object and set up the hardware translation to the object.  For some
85  * objects, the fault handling code also implements copy-on-write.
86  *
87  * When the hat wants to unload a translation, it can call the unload
88  * routine which is responsible for processing reference and modify bits.
89  *
90  * Each segment is protected by it's containing address space lock.  To
91  * access any field in the segment structure, the "as" must be locked.
92  * If a segment field is to be modified, the address space lock must be
93  * write locked.
94  */
95 
96 struct seg {
97 	caddr_t	s_base;			/* base virtual address */
98 	size_t	s_size;			/* size in bytes */
99 	uint_t	s_szc;			/* max page size code */
100 	uint_t	s_flags;		/* flags for segment, see below */
101 	struct	as *s_as;		/* containing address space */
102 	avl_node_t s_tree;		/* AVL tree links to segs in this as */
103 	struct	seg_ops *s_ops;		/* ops vector: see below */
104 	void *s_data;			/* private data for instance */
105 };
106 
107 #define	S_PURGE		(0x01)		/* seg should be purged in as_gap() */
108 
109 struct	seg_ops {
110 	int	(*dup)(struct seg *, struct seg *);
111 	int	(*unmap)(struct seg *, caddr_t, size_t);
112 	void	(*free)(struct seg *);
113 	faultcode_t (*fault)(struct hat *, struct seg *, caddr_t, size_t,
114 	    enum fault_type, enum seg_rw);
115 	faultcode_t (*faulta)(struct seg *, caddr_t);
116 	int	(*setprot)(struct seg *, caddr_t, size_t, uint_t);
117 	int	(*checkprot)(struct seg *, caddr_t, size_t, uint_t);
118 	int	(*kluster)(struct seg *, caddr_t, ssize_t);
119 	size_t	(*swapout)(struct seg *);
120 	int	(*sync)(struct seg *, caddr_t, size_t, int, uint_t);
121 	size_t	(*incore)(struct seg *, caddr_t, size_t, char *);
122 	int	(*lockop)(struct seg *, caddr_t, size_t, int, int, ulong_t *,
123 			size_t);
124 	int	(*getprot)(struct seg *, caddr_t, size_t, uint_t *);
125 	u_offset_t	(*getoffset)(struct seg *, caddr_t);
126 	int	(*gettype)(struct seg *, caddr_t);
127 	int	(*getvp)(struct seg *, caddr_t, struct vnode **);
128 	int	(*advise)(struct seg *, caddr_t, size_t, uint_t);
129 	void	(*dump)(struct seg *);
130 	int	(*pagelock)(struct seg *, caddr_t, size_t, struct page ***,
131 			enum lock_type, enum seg_rw);
132 	int	(*setpagesize)(struct seg *, caddr_t, size_t, uint_t);
133 	int	(*getmemid)(struct seg *, caddr_t, memid_t *);
134 	struct lgrp_mem_policy_info	*(*getpolicy)(struct seg *, caddr_t);
135 	int	(*capable)(struct seg *, segcapability_t);
136 };
137 
138 #ifdef _KERNEL
139 /*
140  * Generic segment operations
141  */
142 extern	void	seg_init(void);
143 extern	struct	seg *seg_alloc(struct as *as, caddr_t base, size_t size);
144 extern	int	seg_attach(struct as *as, caddr_t base, size_t size,
145 			struct seg *seg);
146 extern	void	seg_unmap(struct seg *seg);
147 extern	void	seg_free(struct seg *seg);
148 
149 /*
150  * functions for pagelock cache support
151  */
152 extern	void	seg_ppurge(struct seg *seg);
153 extern	void	seg_ppurge_seg(int (*callback)());
154 extern	void	seg_pinactive(struct seg *seg, caddr_t addr, size_t len,
155 			struct page **pp, enum seg_rw rw, int (*callback)());
156 extern	int	seg_pinsert_check(struct seg *seg, size_t len, uint_t flags);
157 extern	int	seg_pinsert(struct seg *seg, caddr_t addr, size_t len,
158 			struct page **pp, enum seg_rw rw, uint_t flags,
159 			int (*callback)());
160 extern	struct	page **seg_plookup(struct seg *seg, caddr_t addr,
161 			size_t len, enum seg_rw rw);
162 extern	void	seg_pasync_thread(void);
163 extern	void	seg_preap(void);
164 extern	int	seg_p_disable(void);
165 extern	void	seg_p_enable(void);
166 
167 extern	int	seg_preapahead;
168 extern	segadvstat_t  segadvstat;
169 /*
170  * Flags for pagelock cache support
171  */
172 #define	SEGP_ASYNC_FLUSH	0x1	/* flushed by async thread */
173 #define	SEGP_FORCE_WIRED	0x2	/* skip check against seg_pwindow */
174 
175 /*
176  * Return values for seg_pinsert and seg_pinsert_check functions.
177  */
178 #define	SEGP_SUCCESS		0	/* seg_pinsert() succeeded */
179 #define	SEGP_FAIL		1	/* seg_pinsert() failed */
180 
181 /* Page status bits for segop_incore */
182 #define	SEG_PAGE_INCORE		0x01	/* VA has a page backing it */
183 #define	SEG_PAGE_LOCKED		0x02	/* VA has a page that is locked */
184 #define	SEG_PAGE_HASCOW		0x04	/* VA has a page with a copy-on-write */
185 #define	SEG_PAGE_SOFTLOCK	0x08	/* VA has a page with softlock held */
186 #define	SEG_PAGE_VNODEBACKED	0x10	/* Segment is backed by a vnode */
187 #define	SEG_PAGE_ANON		0x20	/* VA has an anonymous page */
188 #define	SEG_PAGE_VNODE		0x40	/* VA has a vnode page backing it */
189 
190 #define	SEGOP_DUP(s, n)		    (*(s)->s_ops->dup)((s), (n))
191 #define	SEGOP_UNMAP(s, a, l)	    (*(s)->s_ops->unmap)((s), (a), (l))
192 #define	SEGOP_FREE(s)		    (*(s)->s_ops->free)((s))
193 #define	SEGOP_FAULT(h, s, a, l, t, rw) \
194 		(*(s)->s_ops->fault)((h), (s), (a), (l), (t), (rw))
195 #define	SEGOP_FAULTA(s, a)	    (*(s)->s_ops->faulta)((s), (a))
196 #define	SEGOP_SETPROT(s, a, l, p)   (*(s)->s_ops->setprot)((s), (a), (l), (p))
197 #define	SEGOP_CHECKPROT(s, a, l, p) (*(s)->s_ops->checkprot)((s), (a), (l), (p))
198 #define	SEGOP_KLUSTER(s, a, d)	    (*(s)->s_ops->kluster)((s), (a), (d))
199 #define	SEGOP_SWAPOUT(s)	    (*(s)->s_ops->swapout)((s))
200 #define	SEGOP_SYNC(s, a, l, atr, f) \
201 		(*(s)->s_ops->sync)((s), (a), (l), (atr), (f))
202 #define	SEGOP_INCORE(s, a, l, v)    (*(s)->s_ops->incore)((s), (a), (l), (v))
203 #define	SEGOP_LOCKOP(s, a, l, atr, op, b, p) \
204 		(*(s)->s_ops->lockop)((s), (a), (l), (atr), (op), (b), (p))
205 #define	SEGOP_GETPROT(s, a, l, p)   (*(s)->s_ops->getprot)((s), (a), (l), (p))
206 #define	SEGOP_GETOFFSET(s, a)	    (*(s)->s_ops->getoffset)((s), (a))
207 #define	SEGOP_GETTYPE(s, a)	    (*(s)->s_ops->gettype)((s), (a))
208 #define	SEGOP_GETVP(s, a, vpp)	    (*(s)->s_ops->getvp)((s), (a), (vpp))
209 #define	SEGOP_ADVISE(s, a, l, b)    (*(s)->s_ops->advise)((s), (a), (l), (b))
210 #define	SEGOP_DUMP(s)		    (*(s)->s_ops->dump)((s))
211 #define	SEGOP_PAGELOCK(s, a, l, p, t, rw) \
212 		(*(s)->s_ops->pagelock)((s), (a), (l), (p), (t), (rw))
213 #define	SEGOP_SETPAGESIZE(s, a, l, szc) \
214 		(*(s)->s_ops->setpagesize)((s), (a), (l), (szc))
215 #define	SEGOP_GETMEMID(s, a, mp)    (*(s)->s_ops->getmemid)((s), (a), (mp))
216 #define	SEGOP_GETPOLICY(s, a)	    (*(s)->s_ops->getpolicy)((s), (a))
217 #define	SEGOP_CAPABLE(s, c)	    (*(s)->s_ops->capable)((s), (c))
218 
219 #define	seg_page(seg, addr) \
220 	(((uintptr_t)((addr) - (seg)->s_base)) >> PAGESHIFT)
221 
222 #define	seg_pages(seg) \
223 	(((uintptr_t)((seg)->s_size + PAGEOFFSET)) >> PAGESHIFT)
224 
225 #define	IE_NOMEM	-1	/* internal to seg layer */
226 #define	IE_RETRY	-2	/* internal to seg layer */
227 #define	IE_REATTACH	-3	/* internal to seg layer */
228 
229 /* Delay/retry factors for seg_p_mem_config_pre_del */
230 #define	SEGP_PREDEL_DELAY_FACTOR	4
231 /*
232  * As a workaround to being unable to purge the pagelock
233  * cache during a DR delete memory operation, we use
234  * a stall threshold that is twice the maximum seen
235  * during testing.  This workaround will be removed
236  * when a suitable fix is found.
237  */
238 #define	SEGP_STALL_SECONDS	25
239 #define	SEGP_STALL_THRESHOLD \
240 	(SEGP_STALL_SECONDS * SEGP_PREDEL_DELAY_FACTOR)
241 
242 #ifdef VMDEBUG
243 
244 uint_t	seg_page(struct seg *, caddr_t);
245 uint_t	seg_pages(struct seg *);
246 
247 #endif	/* VMDEBUG */
248 
249 boolean_t	seg_can_change_zones(struct seg *);
250 size_t		seg_swresv(struct seg *);
251 
252 #endif	/* _KERNEL */
253 
254 #ifdef	__cplusplus
255 }
256 #endif
257 
258 #endif	/* _VM_SEG_H */
259