xref: /illumos-gate/usr/src/uts/common/vm/seg.h (revision 284ce987)
17c478bd9Sstevel@tonic-gate /*
27c478bd9Sstevel@tonic-gate  * CDDL HEADER START
37c478bd9Sstevel@tonic-gate  *
47c478bd9Sstevel@tonic-gate  * The contents of this file are subject to the terms of the
50209230bSgjelinek  * Common Development and Distribution License (the "License").
60209230bSgjelinek  * You may not use this file except in compliance with the License.
77c478bd9Sstevel@tonic-gate  *
87c478bd9Sstevel@tonic-gate  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
97c478bd9Sstevel@tonic-gate  * or http://www.opensolaris.org/os/licensing.
107c478bd9Sstevel@tonic-gate  * See the License for the specific language governing permissions
117c478bd9Sstevel@tonic-gate  * and limitations under the License.
127c478bd9Sstevel@tonic-gate  *
137c478bd9Sstevel@tonic-gate  * When distributing Covered Code, include this CDDL HEADER in each
147c478bd9Sstevel@tonic-gate  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
157c478bd9Sstevel@tonic-gate  * If applicable, add the following below this CDDL HEADER, with the
167c478bd9Sstevel@tonic-gate  * fields enclosed by brackets "[]" replaced with your own identifying
177c478bd9Sstevel@tonic-gate  * information: Portions Copyright [yyyy] [name of copyright owner]
187c478bd9Sstevel@tonic-gate  *
197c478bd9Sstevel@tonic-gate  * CDDL HEADER END
207c478bd9Sstevel@tonic-gate  */
217c478bd9Sstevel@tonic-gate /*
22a98e9dbfSaguzovsk  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
237c478bd9Sstevel@tonic-gate  * Use is subject to license terms.
24*284ce987SPatrick Mooney  * Copyright 2017 Joyent, Inc.
257c478bd9Sstevel@tonic-gate  */
277c478bd9Sstevel@tonic-gate /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
28b4203d75SMarcel Telka /*	  All Rights Reserved	*/
307c478bd9Sstevel@tonic-gate /*
317c478bd9Sstevel@tonic-gate  * University Copyright- Copyright (c) 1982, 1986, 1988
327c478bd9Sstevel@tonic-gate  * The Regents of the University of California
337c478bd9Sstevel@tonic-gate  * All Rights Reserved
347c478bd9Sstevel@tonic-gate  *
357c478bd9Sstevel@tonic-gate  * University Acknowledgment- Portions of this document are derived from
367c478bd9Sstevel@tonic-gate  * software developed by the University of California, Berkeley, and its
377c478bd9Sstevel@tonic-gate  * contributors.
387c478bd9Sstevel@tonic-gate  */
407c478bd9Sstevel@tonic-gate #ifndef	_VM_SEG_H
417c478bd9Sstevel@tonic-gate #define	_VM_SEG_H
437c478bd9Sstevel@tonic-gate #include <sys/vnode.h>
447c478bd9Sstevel@tonic-gate #include <sys/avl.h>
457c478bd9Sstevel@tonic-gate #include <vm/seg_enum.h>
467c478bd9Sstevel@tonic-gate #include <vm/faultcode.h>
477c478bd9Sstevel@tonic-gate #include <vm/hat.h>
497c478bd9Sstevel@tonic-gate #ifdef	__cplusplus
507c478bd9Sstevel@tonic-gate extern "C" {
517c478bd9Sstevel@tonic-gate #endif
537c478bd9Sstevel@tonic-gate /*
547c478bd9Sstevel@tonic-gate  * VM - Segments.
557c478bd9Sstevel@tonic-gate  */
57a98e9dbfSaguzovsk struct anon_map;
597c478bd9Sstevel@tonic-gate /*
607c478bd9Sstevel@tonic-gate  * kstat statistics for segment advise
617c478bd9Sstevel@tonic-gate  */
627c478bd9Sstevel@tonic-gate typedef struct {
637c478bd9Sstevel@tonic-gate 	kstat_named_t MADV_FREE_hit;
647c478bd9Sstevel@tonic-gate 	kstat_named_t MADV_FREE_miss;
657c478bd9Sstevel@tonic-gate } segadvstat_t;
677c478bd9Sstevel@tonic-gate /*
687c478bd9Sstevel@tonic-gate  * memory object ids
697c478bd9Sstevel@tonic-gate  */
707c478bd9Sstevel@tonic-gate typedef struct memid { u_longlong_t val[2]; } memid_t;
727c478bd9Sstevel@tonic-gate /*
737c478bd9Sstevel@tonic-gate  * An address space contains a set of segments, managed by drivers.
747c478bd9Sstevel@tonic-gate  * Drivers support mapped devices, sharing, copy-on-write, etc.
757c478bd9Sstevel@tonic-gate  *
767c478bd9Sstevel@tonic-gate  * The seg structure contains a lock to prevent races, the base virtual
777c478bd9Sstevel@tonic-gate  * address and size of the segment, a back pointer to the containing
787c478bd9Sstevel@tonic-gate  * address space, pointers to maintain an AVL tree of segments in the
797c478bd9Sstevel@tonic-gate  * same address space, and procedure and data hooks for the driver.
807c478bd9Sstevel@tonic-gate  * The AVL tree of segments for the address space is sorted by
817c478bd9Sstevel@tonic-gate  * ascending base addresses and overlapping segments are not allowed.
827c478bd9Sstevel@tonic-gate  *
837c478bd9Sstevel@tonic-gate  * After a segment is created, faults may occur on pages of the segment.
847c478bd9Sstevel@tonic-gate  * When a fault occurs, the fault handling code must get the desired
857c478bd9Sstevel@tonic-gate  * object and set up the hardware translation to the object.  For some
867c478bd9Sstevel@tonic-gate  * objects, the fault handling code also implements copy-on-write.
877c478bd9Sstevel@tonic-gate  *
887c478bd9Sstevel@tonic-gate  * When the hat wants to unload a translation, it can call the unload
897c478bd9Sstevel@tonic-gate  * routine which is responsible for processing reference and modify bits.
907c478bd9Sstevel@tonic-gate  *
917c478bd9Sstevel@tonic-gate  * Each segment is protected by it's containing address space lock.  To
927c478bd9Sstevel@tonic-gate  * access any field in the segment structure, the "as" must be locked.
937c478bd9Sstevel@tonic-gate  * If a segment field is to be modified, the address space lock must be
947c478bd9Sstevel@tonic-gate  * write locked.
957c478bd9Sstevel@tonic-gate  */
97a98e9dbfSaguzovsk typedef struct pcache_link {
98a98e9dbfSaguzovsk 	struct pcache_link	*p_lnext;
99a98e9dbfSaguzovsk 	struct pcache_link	*p_lprev;
100a98e9dbfSaguzovsk } pcache_link_t;
102a98e9dbfSaguzovsk typedef struct seg {
1037c478bd9Sstevel@tonic-gate 	caddr_t	s_base;			/* base virtual address */
1047c478bd9Sstevel@tonic-gate 	size_t	s_size;			/* size in bytes */
1057c478bd9Sstevel@tonic-gate 	uint_t	s_szc;			/* max page size code */
1067c478bd9Sstevel@tonic-gate 	uint_t	s_flags;		/* flags for segment, see below */
1077c478bd9Sstevel@tonic-gate 	struct	as *s_as;		/* containing address space */
1087c478bd9Sstevel@tonic-gate 	avl_node_t s_tree;		/* AVL tree links to segs in this as */
1097c478bd9Sstevel@tonic-gate 	struct	seg_ops *s_ops;		/* ops vector: see below */
1107c478bd9Sstevel@tonic-gate 	void *s_data;			/* private data for instance */
111a98e9dbfSaguzovsk 	kmutex_t s_pmtx;		/* protects seg's pcache list */
112a98e9dbfSaguzovsk 	pcache_link_t s_phead;		/* head of seg's pcache list */
113a98e9dbfSaguzovsk } seg_t;
1157c478bd9Sstevel@tonic-gate #define	S_PURGE		(0x01)		/* seg should be purged in as_gap() */
116*284ce987SPatrick Mooney #define	S_HOLE		(0x02)		/* seg represents hole in AS */
1187c478bd9Sstevel@tonic-gate struct	seg_ops {
1197c478bd9Sstevel@tonic-gate 	int	(*dup)(struct seg *, struct seg *);
1207c478bd9Sstevel@tonic-gate 	int	(*unmap)(struct seg *, caddr_t, size_t);
1217c478bd9Sstevel@tonic-gate 	void	(*free)(struct seg *);
1227c478bd9Sstevel@tonic-gate 	faultcode_t (*fault)(struct hat *, struct seg *, caddr_t, size_t,
1237c478bd9Sstevel@tonic-gate 	    enum fault_type, enum seg_rw);
1247c478bd9Sstevel@tonic-gate 	faultcode_t (*faulta)(struct seg *, caddr_t);
1257c478bd9Sstevel@tonic-gate 	int	(*setprot)(struct seg *, caddr_t, size_t, uint_t);
1267c478bd9Sstevel@tonic-gate 	int	(*checkprot)(struct seg *, caddr_t, size_t, uint_t);
1277c478bd9Sstevel@tonic-gate 	int	(*kluster)(struct seg *, caddr_t, ssize_t);
1287c478bd9Sstevel@tonic-gate 	size_t	(*swapout)(struct seg *);
1297c478bd9Sstevel@tonic-gate 	int	(*sync)(struct seg *, caddr_t, size_t, int, uint_t);
1307c478bd9Sstevel@tonic-gate 	size_t	(*incore)(struct seg *, caddr_t, size_t, char *);
1317c478bd9Sstevel@tonic-gate 	int	(*lockop)(struct seg *, caddr_t, size_t, int, int, ulong_t *,
1327c478bd9Sstevel@tonic-gate 			size_t);
1337c478bd9Sstevel@tonic-gate 	int	(*getprot)(struct seg *, caddr_t, size_t, uint_t *);
1347c478bd9Sstevel@tonic-gate 	u_offset_t	(*getoffset)(struct seg *, caddr_t);
1357c478bd9Sstevel@tonic-gate 	int	(*gettype)(struct seg *, caddr_t);
1367c478bd9Sstevel@tonic-gate 	int	(*getvp)(struct seg *, caddr_t, struct vnode **);
1377c478bd9Sstevel@tonic-gate 	int	(*advise)(struct seg *, caddr_t, size_t, uint_t);
1387c478bd9Sstevel@tonic-gate 	void	(*dump)(struct seg *);
1397c478bd9Sstevel@tonic-gate 	int	(*pagelock)(struct seg *, caddr_t, size_t, struct page ***,
1407c478bd9Sstevel@tonic-gate 			enum lock_type, enum seg_rw);
1417c478bd9Sstevel@tonic-gate 	int	(*setpagesize)(struct seg *, caddr_t, size_t, uint_t);
1427c478bd9Sstevel@tonic-gate 	int	(*getmemid)(struct seg *, caddr_t, memid_t *);
1437c478bd9Sstevel@tonic-gate 	struct lgrp_mem_policy_info	*(*getpolicy)(struct seg *, caddr_t);
1441bd5c35fSelowe 	int	(*capable)(struct seg *, segcapability_t);
1459d12795fSRobert Mustacchi 	int	(*inherit)(struct seg *, caddr_t, size_t, uint_t);
1467c478bd9Sstevel@tonic-gate };
1487c478bd9Sstevel@tonic-gate #ifdef _KERNEL
1507c478bd9Sstevel@tonic-gate /*
1517c478bd9Sstevel@tonic-gate  * Generic segment operations
1527c478bd9Sstevel@tonic-gate  */
1537c478bd9Sstevel@tonic-gate extern	void	seg_init(void);
1547c478bd9Sstevel@tonic-gate extern	struct	seg *seg_alloc(struct as *as, caddr_t base, size_t size);
1557c478bd9Sstevel@tonic-gate extern	int	seg_attach(struct as *as, caddr_t base, size_t size,
1567c478bd9Sstevel@tonic-gate 			struct seg *seg);
1577c478bd9Sstevel@tonic-gate extern	void	seg_unmap(struct seg *seg);
1587c478bd9Sstevel@tonic-gate extern	void	seg_free(struct seg *seg);
1607c478bd9Sstevel@tonic-gate /*
1617c478bd9Sstevel@tonic-gate  * functions for pagelock cache support
1627c478bd9Sstevel@tonic-gate  */
163a98e9dbfSaguzovsk typedef	int (*seg_preclaim_cbfunc_t)(void *, caddr_t, size_t,
164a98e9dbfSaguzovsk     struct page **, enum seg_rw, int);
166a98e9dbfSaguzovsk extern	struct	page **seg_plookup(struct seg *seg, struct anon_map *amp,
167a98e9dbfSaguzovsk     caddr_t addr, size_t len, enum seg_rw rw, uint_t flags);
168a98e9dbfSaguzovsk extern	void	seg_pinactive(struct seg *seg, struct anon_map *amp,
169a98e9dbfSaguzovsk     caddr_t addr, size_t len, struct page **pp, enum seg_rw rw,
170a98e9dbfSaguzovsk     uint_t flags, seg_preclaim_cbfunc_t callback);
172a98e9dbfSaguzovsk extern	void	seg_ppurge(struct seg *seg, struct anon_map *amp,
173a98e9dbfSaguzovsk     uint_t flags);
174a98e9dbfSaguzovsk extern	void	seg_ppurge_wiredpp(struct page **pp);
176a98e9dbfSaguzovsk extern	int	seg_pinsert_check(struct seg *seg, struct anon_map *amp,
177a98e9dbfSaguzovsk     caddr_t addr, size_t len, uint_t flags);
178a98e9dbfSaguzovsk extern	int	seg_pinsert(struct seg *seg, struct anon_map *amp,
179a98e9dbfSaguzovsk     caddr_t addr, size_t len, size_t wlen, struct page **pp, enum seg_rw rw,
180a98e9dbfSaguzovsk     uint_t flags, seg_preclaim_cbfunc_t callback);
1827c478bd9Sstevel@tonic-gate extern	void	seg_pasync_thread(void);
1837c478bd9Sstevel@tonic-gate extern	void	seg_preap(void);
184cee1d74bSjfrank extern	int	seg_p_disable(void);
185cee1d74bSjfrank extern	void	seg_p_enable(void);
187a98e9dbfSaguzovsk extern	segadvstat_t	segadvstat;
1897c478bd9Sstevel@tonic-gate /*
190a98e9dbfSaguzovsk  * Flags for pagelock cache support.
191a98e9dbfSaguzovsk  * Flags argument is passed as uint_t to pcache routines.  upper 16 bits of
192a98e9dbfSaguzovsk  * the flags argument are reserved for alignment page shift when SEGP_PSHIFT
193a98e9dbfSaguzovsk  * is set.
1947c478bd9Sstevel@tonic-gate  */
195a98e9dbfSaguzovsk #define	SEGP_FORCE_WIRED	0x1	/* skip check against seg_pwindow */
196a98e9dbfSaguzovsk #define	SEGP_AMP		0x2	/* anon map's pcache entry */
197a98e9dbfSaguzovsk #define	SEGP_PSHIFT		0x4	/* addr pgsz shift for hash function */
1997c478bd9Sstevel@tonic-gate /*
2007c478bd9Sstevel@tonic-gate  * Return values for seg_pinsert and seg_pinsert_check functions.
2017c478bd9Sstevel@tonic-gate  */
2027c478bd9Sstevel@tonic-gate #define	SEGP_SUCCESS		0	/* seg_pinsert() succeeded */
2037c478bd9Sstevel@tonic-gate #define	SEGP_FAIL		1	/* seg_pinsert() failed */
2057c478bd9Sstevel@tonic-gate /* Page status bits for segop_incore */
2067c478bd9Sstevel@tonic-gate #define	SEG_PAGE_INCORE		0x01	/* VA has a page backing it */
2077c478bd9Sstevel@tonic-gate #define	SEG_PAGE_LOCKED		0x02	/* VA has a page that is locked */
2087c478bd9Sstevel@tonic-gate #define	SEG_PAGE_HASCOW		0x04	/* VA has a page with a copy-on-write */
2097c478bd9Sstevel@tonic-gate #define	SEG_PAGE_SOFTLOCK	0x08	/* VA has a page with softlock held */
2107c478bd9Sstevel@tonic-gate #define	SEG_PAGE_VNODEBACKED	0x10	/* Segment is backed by a vnode */
2117c478bd9Sstevel@tonic-gate #define	SEG_PAGE_ANON		0x20	/* VA has an anonymous page */
2127c478bd9Sstevel@tonic-gate #define	SEG_PAGE_VNODE		0x40	/* VA has a vnode page backing it */
2147c478bd9Sstevel@tonic-gate #define	SEGOP_DUP(s, n)		    (*(s)->s_ops->dup)((s), (n))
2157c478bd9Sstevel@tonic-gate #define	SEGOP_UNMAP(s, a, l)	    (*(s)->s_ops->unmap)((s), (a), (l))
2167c478bd9Sstevel@tonic-gate #define	SEGOP_FREE(s)		    (*(s)->s_ops->free)((s))
2177c478bd9Sstevel@tonic-gate #define	SEGOP_FAULT(h, s, a, l, t, rw) \
2187c478bd9Sstevel@tonic-gate 		(*(s)->s_ops->fault)((h), (s), (a), (l), (t), (rw))
2197c478bd9Sstevel@tonic-gate #define	SEGOP_FAULTA(s, a)	    (*(s)->s_ops->faulta)((s), (a))
2207c478bd9Sstevel@tonic-gate #define	SEGOP_SETPROT(s, a, l, p)   (*(s)->s_ops->setprot)((s), (a), (l), (p))
2217c478bd9Sstevel@tonic-gate #define	SEGOP_CHECKPROT(s, a, l, p) (*(s)->s_ops->checkprot)((s), (a), (l), (p))
2227c478bd9Sstevel@tonic-gate #define	SEGOP_KLUSTER(s, a, d)	    (*(s)->s_ops->kluster)((s), (a), (d))
2237c478bd9Sstevel@tonic-gate #define	SEGOP_SWAPOUT(s)	    (*(s)->s_ops->swapout)((s))
2247c478bd9Sstevel@tonic-gate #define	SEGOP_SYNC(s, a, l, atr, f) \
2257c478bd9Sstevel@tonic-gate 		(*(s)->s_ops->sync)((s), (a), (l), (atr), (f))
2267c478bd9Sstevel@tonic-gate #define	SEGOP_INCORE(s, a, l, v)    (*(s)->s_ops->incore)((s), (a), (l), (v))
2277c478bd9Sstevel@tonic-gate #define	SEGOP_LOCKOP(s, a, l, atr, op, b, p) \
2287c478bd9Sstevel@tonic-gate 		(*(s)->s_ops->lockop)((s), (a), (l), (atr), (op), (b), (p))
2297c478bd9Sstevel@tonic-gate #define	SEGOP_GETPROT(s, a, l, p)   (*(s)->s_ops->getprot)((s), (a), (l), (p))
2307c478bd9Sstevel@tonic-gate #define	SEGOP_GETOFFSET(s, a)	    (*(s)->s_ops->getoffset)((s), (a))
2317c478bd9Sstevel@tonic-gate #define	SEGOP_GETTYPE(s, a)	    (*(s)->s_ops->gettype)((s), (a))
2327c478bd9Sstevel@tonic-gate #define	SEGOP_GETVP(s, a, vpp)	    (*(s)->s_ops->getvp)((s), (a), (vpp))
2337c478bd9Sstevel@tonic-gate #define	SEGOP_ADVISE(s, a, l, b)    (*(s)->s_ops->advise)((s), (a), (l), (b))
2347c478bd9Sstevel@tonic-gate #define	SEGOP_DUMP(s)		    (*(s)->s_ops->dump)((s))
2357c478bd9Sstevel@tonic-gate #define	SEGOP_PAGELOCK(s, a, l, p, t, rw) \
2367c478bd9Sstevel@tonic-gate 		(*(s)->s_ops->pagelock)((s), (a), (l), (p), (t), (rw))
2377c478bd9Sstevel@tonic-gate #define	SEGOP_SETPAGESIZE(s, a, l, szc) \
2387c478bd9Sstevel@tonic-gate 		(*(s)->s_ops->setpagesize)((s), (a), (l), (szc))
2397c478bd9Sstevel@tonic-gate #define	SEGOP_GETMEMID(s, a, mp)    (*(s)->s_ops->getmemid)((s), (a), (mp))
2407c478bd9Sstevel@tonic-gate #define	SEGOP_GETPOLICY(s, a)	    (*(s)->s_ops->getpolicy)((s), (a))
2411bd5c35fSelowe #define	SEGOP_CAPABLE(s, c)	    (*(s)->s_ops->capable)((s), (c))
2429d12795fSRobert Mustacchi #define	SEGOP_INHERIT(s, a, l, b)   (*(s)->s_ops->inherit)((s), (a), (l), (b))
2447c478bd9Sstevel@tonic-gate #define	seg_page(seg, addr) \
2457c478bd9Sstevel@tonic-gate 	(((uintptr_t)((addr) - (seg)->s_base)) >> PAGESHIFT)
2477c478bd9Sstevel@tonic-gate #define	seg_pages(seg) \
2487c478bd9Sstevel@tonic-gate 	(((uintptr_t)((seg)->s_size + PAGEOFFSET)) >> PAGESHIFT)
2507c478bd9Sstevel@tonic-gate #define	IE_NOMEM	-1	/* internal to seg layer */
2517c478bd9Sstevel@tonic-gate #define	IE_RETRY	-2	/* internal to seg layer */
2527c478bd9Sstevel@tonic-gate #define	IE_REATTACH	-3	/* internal to seg layer */
2549d12795fSRobert Mustacchi /* Values for SEGOP_INHERIT */
2559d12795fSRobert Mustacchi #define	SEGP_INH_ZERO	0x01
2569d12795fSRobert Mustacchi 
2579d12795fSRobert Mustacchi int seg_inherit_notsup(struct seg *, caddr_t, size_t, uint_t);
2589d12795fSRobert Mustacchi 
2597c478bd9Sstevel@tonic-gate /* Delay/retry factors for seg_p_mem_config_pre_del */
2607c478bd9Sstevel@tonic-gate #define	SEGP_PREDEL_DELAY_FACTOR	4
2617c478bd9Sstevel@tonic-gate /*
2627c478bd9Sstevel@tonic-gate  * As a workaround to being unable to purge the pagelock
2637c478bd9Sstevel@tonic-gate  * cache during a DR delete memory operation, we use
2647c478bd9Sstevel@tonic-gate  * a stall threshold that is twice the maximum seen
2657c478bd9Sstevel@tonic-gate  * during testing.  This workaround will be removed
2667c478bd9Sstevel@tonic-gate  * when a suitable fix is found.
2677c478bd9Sstevel@tonic-gate  */
2687c478bd9Sstevel@tonic-gate #define	SEGP_STALL_SECONDS	25
2697c478bd9Sstevel@tonic-gate #define	SEGP_STALL_THRESHOLD \
2707c478bd9Sstevel@tonic-gate 	(SEGP_STALL_SECONDS * SEGP_PREDEL_DELAY_FACTOR)
2727c478bd9Sstevel@tonic-gate #ifdef VMDEBUG
2747c478bd9Sstevel@tonic-gate uint_t	seg_page(struct seg *, caddr_t);
2757c478bd9Sstevel@tonic-gate uint_t	seg_pages(struct seg *);
2777c478bd9Sstevel@tonic-gate #endif	/* VMDEBUG */
2790209230bSgjelinek boolean_t	seg_can_change_zones(struct seg *);
2800209230bSgjelinek size_t		seg_swresv(struct seg *);
2827c478bd9Sstevel@tonic-gate #endif	/* _KERNEL */
2847c478bd9Sstevel@tonic-gate #ifdef	__cplusplus
2857c478bd9Sstevel@tonic-gate }
2867c478bd9Sstevel@tonic-gate #endif
2887c478bd9Sstevel@tonic-gate #endif	/* _VM_SEG_H */