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


  89  * routine which is responsible for processing reference and modify bits.
  90  *
  91  * Each segment is protected by it's containing address space lock.  To
  92  * access any field in the segment structure, the "as" must be locked.
  93  * If a segment field is to be modified, the address space lock must be
  94  * write locked.
  95  */
  96 
  97 typedef struct pcache_link {
  98         struct pcache_link      *p_lnext;
  99         struct pcache_link      *p_lprev;
 100 } pcache_link_t;
 101 
 102 typedef struct seg {
 103         caddr_t s_base;                 /* base virtual address */
 104         size_t  s_size;                 /* size in bytes */
 105         uint_t  s_szc;                  /* max page size code */
 106         uint_t  s_flags;                /* flags for segment, see below */
 107         struct  as *s_as;               /* containing address space */
 108         avl_node_t s_tree;              /* AVL tree links to segs in this as */
 109         struct  seg_ops *s_ops;         /* ops vector: see below */
 110         void *s_data;                   /* private data for instance */
 111         kmutex_t s_pmtx;                /* protects seg's pcache list */
 112         pcache_link_t s_phead;          /* head of seg's pcache list */
 113 } seg_t;
 114 
 115 #define S_PURGE         (0x01)          /* seg should be purged in as_gap() */
 116 
 117 struct  seg_ops {
 118         int     (*dup)(struct seg *, struct seg *);
 119         int     (*unmap)(struct seg *, caddr_t, size_t);
 120         void    (*free)(struct seg *);
 121         faultcode_t (*fault)(struct hat *, struct seg *, caddr_t, size_t,
 122             enum fault_type, enum seg_rw);
 123         faultcode_t (*faulta)(struct seg *, caddr_t);
 124         int     (*setprot)(struct seg *, caddr_t, size_t, uint_t);
 125         int     (*checkprot)(struct seg *, caddr_t, size_t, uint_t);
 126         int     (*kluster)(struct seg *, caddr_t, ssize_t);
 127         size_t  (*swapout)(struct seg *);
 128         int     (*sync)(struct seg *, caddr_t, size_t, int, uint_t);
 129         size_t  (*incore)(struct seg *, caddr_t, size_t, char *);
 130         int     (*lockop)(struct seg *, caddr_t, size_t, int, int, ulong_t *,
 131                         size_t);
 132         int     (*getprot)(struct seg *, caddr_t, size_t, uint_t *);
 133         u_offset_t      (*getoffset)(struct seg *, caddr_t);
 134         int     (*gettype)(struct seg *, caddr_t);
 135         int     (*getvp)(struct seg *, caddr_t, struct vnode **);
 136         int     (*advise)(struct seg *, caddr_t, size_t, uint_t);
 137         void    (*dump)(struct seg *);
 138         int     (*pagelock)(struct seg *, caddr_t, size_t, struct page ***,
 139                         enum lock_type, enum seg_rw);
 140         int     (*setpagesize)(struct seg *, caddr_t, size_t, uint_t);
 141         int     (*getmemid)(struct seg *, caddr_t, memid_t *);
 142         struct lgrp_mem_policy_info     *(*getpolicy)(struct seg *, caddr_t);
 143         int     (*capable)(struct seg *, segcapability_t);
 144         int     (*inherit)(struct seg *, caddr_t, size_t, uint_t);
 145 };
 146 
 147 #ifdef _KERNEL


 193  */
 194 #define SEGP_FORCE_WIRED        0x1     /* skip check against seg_pwindow */
 195 #define SEGP_AMP                0x2     /* anon map's pcache entry */
 196 #define SEGP_PSHIFT             0x4     /* addr pgsz shift for hash function */
 197 
 198 /*
 199  * Return values for seg_pinsert and seg_pinsert_check functions.
 200  */
 201 #define SEGP_SUCCESS            0       /* seg_pinsert() succeeded */
 202 #define SEGP_FAIL               1       /* seg_pinsert() failed */
 203 
 204 /* Page status bits for segop_incore */
 205 #define SEG_PAGE_INCORE         0x01    /* VA has a page backing it */
 206 #define SEG_PAGE_LOCKED         0x02    /* VA has a page that is locked */
 207 #define SEG_PAGE_HASCOW         0x04    /* VA has a page with a copy-on-write */
 208 #define SEG_PAGE_SOFTLOCK       0x08    /* VA has a page with softlock held */
 209 #define SEG_PAGE_VNODEBACKED    0x10    /* Segment is backed by a vnode */
 210 #define SEG_PAGE_ANON           0x20    /* VA has an anonymous page */
 211 #define SEG_PAGE_VNODE          0x40    /* VA has a vnode page backing it */
 212 
 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 #define seg_page(seg, addr) \
 244         (((uintptr_t)((addr) - (seg)->s_base)) >> PAGESHIFT)
 245 
 246 #define seg_pages(seg) \
 247         (((uintptr_t)((seg)->s_size + PAGEOFFSET)) >> PAGESHIFT)
 248 
 249 #define IE_NOMEM        -1      /* internal to seg layer */
 250 #define IE_RETRY        -2      /* internal to seg layer */
 251 #define IE_REATTACH     -3      /* internal to seg layer */
 252 
 253 /* Values for SEGOP_INHERIT */
 254 #define SEGP_INH_ZERO   0x01
 255 
 256 int seg_inherit_notsup(struct seg *, caddr_t, size_t, uint_t);
 257 
 258 /* Delay/retry factors for seg_p_mem_config_pre_del */
 259 #define SEGP_PREDEL_DELAY_FACTOR        4
 260 /*
 261  * As a workaround to being unable to purge the pagelock
 262  * cache during a DR delete memory operation, we use
 263  * a stall threshold that is twice the maximum seen
 264  * during testing.  This workaround will be removed
 265  * when a suitable fix is found.
 266  */
 267 #define SEGP_STALL_SECONDS      25
 268 #define SEGP_STALL_THRESHOLD \
 269         (SEGP_STALL_SECONDS * SEGP_PREDEL_DELAY_FACTOR)
 270 
 271 #ifdef VMDEBUG
 272 
 273 uint_t  seg_page(struct seg *, caddr_t);
 274 uint_t  seg_pages(struct seg *);
 275 
 276 #endif  /* VMDEBUG */
 277 
 278 boolean_t       seg_can_change_zones(struct seg *);
 279 size_t          seg_swresv(struct seg *);

























 280 
 281 #endif  /* _KERNEL */
 282 
 283 #ifdef  __cplusplus
 284 }
 285 #endif
 286 
 287 #endif  /* _VM_SEG_H */


  89  * routine which is responsible for processing reference and modify bits.
  90  *
  91  * Each segment is protected by it's containing address space lock.  To
  92  * access any field in the segment structure, the "as" must be locked.
  93  * If a segment field is to be modified, the address space lock must be
  94  * write locked.
  95  */
  96 
  97 typedef struct pcache_link {
  98         struct pcache_link      *p_lnext;
  99         struct pcache_link      *p_lprev;
 100 } pcache_link_t;
 101 
 102 typedef struct seg {
 103         caddr_t s_base;                 /* base virtual address */
 104         size_t  s_size;                 /* size in bytes */
 105         uint_t  s_szc;                  /* max page size code */
 106         uint_t  s_flags;                /* flags for segment, see below */
 107         struct  as *s_as;               /* containing address space */
 108         avl_node_t s_tree;              /* AVL tree links to segs in this as */
 109         const struct seg_ops *s_ops;    /* ops vector: see below */
 110         void *s_data;                   /* private data for instance */
 111         kmutex_t s_pmtx;                /* protects seg's pcache list */
 112         pcache_link_t s_phead;          /* head of seg's pcache list */
 113 } seg_t;
 114 
 115 #define S_PURGE         (0x01)          /* seg should be purged in as_gap() */
 116 
 117 struct  seg_ops {
 118         int     (*dup)(struct seg *, struct seg *);
 119         int     (*unmap)(struct seg *, caddr_t, size_t);
 120         void    (*free)(struct seg *);
 121         faultcode_t (*fault)(struct hat *, struct seg *, caddr_t, size_t,
 122             enum fault_type, enum seg_rw);
 123         faultcode_t (*faulta)(struct seg *, caddr_t);
 124         int     (*setprot)(struct seg *, caddr_t, size_t, uint_t);
 125         int     (*checkprot)(struct seg *, caddr_t, size_t, uint_t);
 126         int     (*kluster)(struct seg *, caddr_t, ssize_t);

 127         int     (*sync)(struct seg *, caddr_t, size_t, int, uint_t);
 128         size_t  (*incore)(struct seg *, caddr_t, size_t, char *);
 129         int     (*lockop)(struct seg *, caddr_t, size_t, int, int, ulong_t *,
 130                         size_t);
 131         int     (*getprot)(struct seg *, caddr_t, size_t, uint_t *);
 132         u_offset_t      (*getoffset)(struct seg *, caddr_t);
 133         int     (*gettype)(struct seg *, caddr_t);
 134         int     (*getvp)(struct seg *, caddr_t, struct vnode **);
 135         int     (*advise)(struct seg *, caddr_t, size_t, uint_t);
 136         void    (*dump)(struct seg *);
 137         int     (*pagelock)(struct seg *, caddr_t, size_t, struct page ***,
 138                         enum lock_type, enum seg_rw);
 139         int     (*setpagesize)(struct seg *, caddr_t, size_t, uint_t);
 140         int     (*getmemid)(struct seg *, caddr_t, memid_t *);
 141         struct lgrp_mem_policy_info     *(*getpolicy)(struct seg *, caddr_t);
 142         int     (*capable)(struct seg *, segcapability_t);
 143         int     (*inherit)(struct seg *, caddr_t, size_t, uint_t);
 144 };
 145 
 146 #ifdef _KERNEL


 192  */
 193 #define SEGP_FORCE_WIRED        0x1     /* skip check against seg_pwindow */
 194 #define SEGP_AMP                0x2     /* anon map's pcache entry */
 195 #define SEGP_PSHIFT             0x4     /* addr pgsz shift for hash function */
 196 
 197 /*
 198  * Return values for seg_pinsert and seg_pinsert_check functions.
 199  */
 200 #define SEGP_SUCCESS            0       /* seg_pinsert() succeeded */
 201 #define SEGP_FAIL               1       /* seg_pinsert() failed */
 202 
 203 /* Page status bits for segop_incore */
 204 #define SEG_PAGE_INCORE         0x01    /* VA has a page backing it */
 205 #define SEG_PAGE_LOCKED         0x02    /* VA has a page that is locked */
 206 #define SEG_PAGE_HASCOW         0x04    /* VA has a page with a copy-on-write */
 207 #define SEG_PAGE_SOFTLOCK       0x08    /* VA has a page with softlock held */
 208 #define SEG_PAGE_VNODEBACKED    0x10    /* Segment is backed by a vnode */
 209 #define SEG_PAGE_ANON           0x20    /* VA has an anonymous page */
 210 #define SEG_PAGE_VNODE          0x40    /* VA has a vnode page backing it */
 211 






























 212 #define seg_page(seg, addr) \
 213         (((uintptr_t)((addr) - (seg)->s_base)) >> PAGESHIFT)
 214 
 215 #define seg_pages(seg) \
 216         (((uintptr_t)((seg)->s_size + PAGEOFFSET)) >> PAGESHIFT)
 217 
 218 #define IE_NOMEM        -1      /* internal to seg layer */
 219 #define IE_RETRY        -2      /* internal to seg layer */
 220 #define IE_REATTACH     -3      /* internal to seg layer */
 221 
 222 /* Values for segop_inherit */
 223 #define SEGP_INH_ZERO   0x01
 224 


 225 /* Delay/retry factors for seg_p_mem_config_pre_del */
 226 #define SEGP_PREDEL_DELAY_FACTOR        4
 227 /*
 228  * As a workaround to being unable to purge the pagelock
 229  * cache during a DR delete memory operation, we use
 230  * a stall threshold that is twice the maximum seen
 231  * during testing.  This workaround will be removed
 232  * when a suitable fix is found.
 233  */
 234 #define SEGP_STALL_SECONDS      25
 235 #define SEGP_STALL_THRESHOLD \
 236         (SEGP_STALL_SECONDS * SEGP_PREDEL_DELAY_FACTOR)
 237 
 238 #ifdef VMDEBUG
 239 
 240 uint_t  seg_page(struct seg *, caddr_t);
 241 uint_t  seg_pages(struct seg *);
 242 
 243 #endif  /* VMDEBUG */
 244 
 245 boolean_t       seg_can_change_zones(struct seg *);
 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);
 272 
 273 #endif  /* _KERNEL */
 274 
 275 #ifdef  __cplusplus
 276 }
 277 #endif
 278 
 279 #endif  /* _VM_SEG_H */