Commit | Line | Data |
---|---|---|
1c15e888 C |
1 | /* |
2 | * Copyright (c) 1982, 1986, 1989 Regents of the University of California. | |
3 | * All rights reserved. | |
4 | * | |
5 | * Redistribution is only permitted until one year after the first shipment | |
6 | * of 4.4BSD by the Regents. Otherwise, redistribution and use in source and | |
7 | * binary forms are permitted provided that: (1) source distributions retain | |
8 | * this entire copyright notice and comment, and (2) distributions including | |
9 | * binaries display the following acknowledgement: This product includes | |
10 | * software developed by the University of California, Berkeley and its | |
11 | * contributors'' in the documentation or other materials provided with the | |
12 | * distribution and in all advertising materials mentioning features or use | |
13 | * of this software. Neither the name of the University nor the names of | |
14 | * its contributors may be used to endorse or promote products derived from | |
15 | * this software without specific prior written permission. | |
16 | * THIS SOFTWARE IS PROVIDED AS IS'' AND WITHOUT ANY EXPRESS OR IMPLIED | |
17 | * WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF | |
18 | * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. | |
19 | * | |
20 | * @(#)kern_mman.c 7.18 (Berkeley) 6/30/90 | |
21 | */ | |
22 | ||
23 | #include "param.h" | |
24 | #include "systm.h" | |
25 | #include "map.h" | |
26 | #include "user.h" | |
27 | #include "proc.h" | |
28 | #include "buf.h" | |
29 | #include "vnode.h" | |
30 | #include "specdev.h" | |
31 | #include "seg.h" | |
32 | #include "acct.h" | |
33 | #include "wait.h" | |
34 | #include "vm.h" | |
35 | #include "text.h" | |
36 | #include "file.h" | |
37 | #include "vadvise.h" | |
38 | #include "cmap.h" | |
39 | #include "trace.h" | |
40 | #include "mman.h" | |
41 | #include "mapmem.h" | |
42 | #include "malloc.h" | |
43 | #include "conf.h" | |
44 | ||
45 | #include "machine/cpu.h" | |
46 | #include "machine/reg.h" | |
47 | #include "machine/psl.h" | |
48 | #include "machine/pte.h" | |
49 | #include "machine/mtpr.h" | |
50 | ||
51 | /* | |
52 | * The MMAP code here is temporary; it provides support | |
53 | * only for mmaping devices such as frame buffers. | |
54 | * All to be different next time... | |
55 | */ | |
56 | #ifndef MAPMEM | |
57 | #undef MMAP /* XXX */ | |
58 | #endif | |
59 | ||
60 | #ifdef MMAP | |
61 | struct mapmemops mmapops = { | |
62 | (int (*)())0, (int (*)())0, (int (*)())0, (int (*)())0 | |
63 | }; | |
64 | #endif | |
65 | ||
66 | /* ARGSUSED */ | |
67 | sbrk(p, uap, retval) | |
68 | struct proc *p; | |
69 | struct args { | |
70 | int incr; | |
71 | } *uap; | |
72 | int *retval; | |
73 | { | |
74 | ||
75 | /* Not yet implemented */ | |
76 | return (EOPNOTSUPP); | |
77 | } | |
78 | ||
79 | /* ARGSUSED */ | |
80 | sstk(p, uap, retval) | |
81 | struct proc *p; | |
82 | struct args { | |
83 | int incr; | |
84 | } *uap; | |
85 | int *retval; | |
86 | { | |
87 | ||
88 | /* Not yet implemented */ | |
89 | return (EOPNOTSUPP); | |
90 | } | |
91 | ||
92 | /* ARGSUSED */ | |
93 | getpagesize(p, uap, retval) | |
94 | struct proc *p; | |
95 | struct args *uap; | |
96 | int *retval; | |
97 | { | |
98 | ||
99 | *retval = NBPG * CLSIZE; | |
100 | return (0); | |
101 | } | |
102 | ||
103 | /* ARGSUSED */ | |
104 | smmap(p, uap, retval) | |
105 | register struct proc *p; | |
106 | register struct args { | |
107 | caddr_t addr; | |
108 | int len; | |
109 | int prot; | |
110 | int share; | |
111 | int fd; | |
112 | off_t pos; | |
113 | } *uap; | |
114 | int *retval; | |
115 | { | |
116 | #ifndef MMAP | |
117 | return (EOPNOTSUPP); | |
118 | #else | |
119 | register struct file *fp; | |
120 | struct mapmem *mp; | |
121 | struct vnode *vp; | |
122 | register struct pte *pte; | |
123 | struct pte *dpte; | |
124 | register int off; | |
125 | int error, fv, lv, pm, (*mapfun)(); | |
126 | dev_t dev; | |
127 | ||
128 | if (error = getvnode(u.u_ofile, uap->fd, &fp)) | |
129 | return (error); | |
130 | vp = (struct vnode *)fp->f_data; | |
131 | if (vp->v_type != VCHR) | |
132 | return (EINVAL); | |
133 | dev = vp->v_rdev; | |
134 | mapfun = cdevsw[major(dev)].d_mmap; | |
135 | if (mapfun == NULL) | |
136 | return (EINVAL); | |
137 | if (((int)uap->addr & CLOFSET) || (uap->pos & CLOFSET) || | |
138 | uap->len <= 0 || (uap->len & CLOFSET)) | |
139 | return (EINVAL); | |
140 | if ((uap->prot & PROT_WRITE) && (fp->f_flag&FWRITE) == 0) | |
141 | return (EINVAL); | |
142 | if ((uap->prot & PROT_READ) && (fp->f_flag&FREAD) == 0) | |
143 | return (EINVAL); | |
144 | if (uap->share != MAP_SHARED) | |
145 | return (EINVAL); | |
146 | for (off = 0; off < uap->len; off += NBPG) | |
147 | if ((*mapfun)(dev, uap->pos+off, uap->prot) == -1) | |
148 | return (EINVAL); /* Needs translation */ | |
149 | /* | |
150 | * Allocate a descriptor for this region and expand page | |
151 | * table to accomodate. | |
152 | */ | |
153 | if (uap->prot & PROT_WRITE) { | |
154 | pm = PG_UW|PG_FOD|PG_V; | |
155 | off = MM_RW; | |
156 | } else { | |
157 | pm = PG_URKR|PG_FOD|PG_V; | |
158 | off = MM_RO; | |
159 | } | |
160 | #if defined(hp300) | |
161 | pm |= PG_CI; | |
162 | off |= MM_CI; | |
163 | #endif | |
164 | error = mmalloc(p, uap->fd, &uap->addr, uap->len, off, &mmapops, &mp); | |
165 | if (error) | |
166 | return (error); | |
167 | /* | |
168 | * Now map it in. | |
169 | * Can't use mmmapin() because of args to map function. | |
170 | */ | |
171 | fv = btop(uap->addr); | |
172 | pte = vtopte(p, fv); | |
173 | dpte = dptopte(p, u.u_dsize); | |
174 | for (off = 0; off < uap->len; off += NBPG) { | |
175 | if ((off&CLOFSET) == 0 && pte < dpte) | |
176 | p->p_rssize -= vmemfree(pte, CLSIZE); | |
177 | *(int *)pte = pm; | |
178 | pte->pg_pfnum = (*mapfun)(dev, uap->pos+off, uap->prot); | |
179 | pte++; | |
180 | } | |
181 | newptes(vtopte(p, fv), fv, btoc(uap->len)); | |
182 | u.u_pofile[uap->fd] |= UF_MAPPED; | |
183 | return (0); | |
184 | #endif /* MMAP */ | |
185 | } | |
186 | ||
187 | /* ARGSUSED */ | |
188 | msync(p, uap, retval) | |
189 | struct proc *p; | |
190 | struct args { | |
191 | char *addr; | |
192 | int len; | |
193 | } *uap; | |
194 | int *retval; | |
195 | { | |
196 | ||
197 | /* Not yet implemented */ | |
198 | return (EOPNOTSUPP); | |
199 | } | |
200 | ||
201 | /* ARGSUSED */ | |
202 | munmap(p, uap, retval) | |
203 | register struct proc *p; | |
204 | register struct args { | |
205 | caddr_t addr; | |
206 | int len; | |
207 | } *uap; | |
208 | int *retval; | |
209 | { | |
210 | #ifndef MMAP | |
211 | return (EOPNOTSUPP); | |
212 | #else | |
213 | register struct mapmem *mp; | |
214 | register int fd; | |
215 | caddr_t eaddr; | |
216 | int error; | |
217 | ||
218 | if (((int)uap->addr & CLOFSET) || | |
219 | uap->len <= 0 || (uap->len & CLOFSET)) | |
220 | return (EINVAL); | |
221 | /* | |
222 | * Locate region mapping this range. If found, unmap it. | |
223 | */ | |
224 | eaddr = uap->addr + uap->len - 1; | |
225 | for (mp = u.u_mmap; mp; mp = mp->mm_next) | |
226 | if (mp->mm_ops == &mmapops && | |
227 | uap->addr >= mp->mm_uva && eaddr < mp->mm_uva+mp->mm_size) | |
228 | break; | |
229 | if (mp == MMNIL) | |
230 | return (EINVAL); | |
231 | fd = mp->mm_id; | |
232 | mmmapout(p, mp); | |
233 | error = mmfree(mp); | |
234 | /* | |
235 | * If no other range has this descriptor mapped, mark it as unmapped. | |
236 | */ | |
237 | for (mp = u.u_mmap; mp; mp = mp->mm_next) | |
238 | if (mp->mm_id == fd) | |
239 | break; | |
240 | if (mp == MMNIL) | |
241 | u.u_pofile[fd] &= ~UF_MAPPED; | |
242 | return (error); | |
243 | #endif /* MMAP */ | |
244 | } | |
245 | ||
246 | munmapfd(fd) | |
247 | int fd; | |
248 | { | |
249 | int error = 0; | |
250 | #ifdef MMAP | |
251 | struct proc *p = u.u_procp; /* XXX */ | |
252 | register struct mapmem *mp, **mpp; | |
253 | ||
254 | if (p->p_flag & SVFORK) | |
255 | return (0); | |
256 | mpp = &u.u_mmap; | |
257 | for (mp = *mpp; mp; mp = *mpp) { | |
258 | if (mp->mm_ops == &mmapops && mp->mm_id == fd) { | |
259 | mmmapout(p, mp); | |
260 | error = mmfree(mp); | |
261 | } else | |
262 | mpp = &mp->mm_next; | |
263 | } | |
264 | #endif | |
265 | u.u_pofile[fd] &= ~UF_MAPPED; | |
266 | return (error); | |
267 | } | |
268 | ||
269 | /* ARGSUSED */ | |
270 | mprotect(p, uap, retval) | |
271 | struct proc *p; | |
272 | struct args { | |
273 | char *addr; | |
274 | int len; | |
275 | int prot; | |
276 | } *uap; | |
277 | int *retval; | |
278 | { | |
279 | ||
280 | /* Not yet implemented */ | |
281 | return (EOPNOTSUPP); | |
282 | } | |
283 | ||
284 | /* ARGSUSED */ | |
285 | madvise(p, uap, retval) | |
286 | struct proc *p; | |
287 | struct args { | |
288 | char *addr; | |
289 | int len; | |
290 | int behav; | |
291 | } *uap; | |
292 | int *retval; | |
293 | { | |
294 | ||
295 | /* Not yet implemented */ | |
296 | return (EOPNOTSUPP); | |
297 | } | |
298 | ||
299 | /* ARGSUSED */ | |
300 | mincore(p, uap, retval) | |
301 | struct proc *p; | |
302 | struct args { | |
303 | char *addr; | |
304 | int len; | |
305 | char *vec; | |
306 | } *uap; | |
307 | int *retval; | |
308 | { | |
309 | ||
310 | /* Not yet implemented */ | |
311 | return (EOPNOTSUPP); | |
312 | } | |
313 | ||
314 | /* BEGIN DEFUNCT */ | |
315 | /* ARGSUSED */ | |
316 | obreak(p, uap, retval) | |
317 | struct proc *p; | |
318 | struct args { | |
319 | char *nsiz; | |
320 | } *uap; | |
321 | int *retval; | |
322 | { | |
323 | register segsz_t n, d, ds; | |
324 | int error; | |
325 | ||
326 | /* | |
327 | * set n to new data size | |
328 | */ | |
329 | n = btoc(uap->nsiz) - dptov(p, 0); | |
330 | if (n < 0) | |
331 | n = 0; | |
332 | /* | |
333 | * since we can't pass a -ve argument for the difference to chksize, | |
334 | * if d is negative, make ds equal to the final value and clear d. | |
335 | * keep the real difference in n for later use in expand. | |
336 | */ | |
337 | ds = u.u_dsize; | |
338 | if ((n = d = clrnd(n - u.u_dsize)) < 0) { | |
339 | ds += d; | |
340 | d = 0; | |
341 | } | |
342 | if (ctob(ds + d) > u.u_rlimit[RLIMIT_DATA].rlim_cur) | |
343 | return (ENOMEM); | |
344 | if (error = | |
345 | chksize((u_int)u.u_tsize, (u_int)ds, (u_int)d, (u_int)u.u_ssize)) | |
346 | return (error); | |
347 | #ifdef MAPMEM | |
348 | /* | |
349 | * If change would conflict with any mapped memory segment | |
350 | * return ENOMEM. | |
351 | */ | |
352 | if (u.u_mmap && n != 0) { | |
353 | caddr_t low, high; | |
354 | ||
355 | low = (caddr_t) ctob(dptov(p, ds)); | |
356 | high = low + ctob((n < 0) ? -n : n); | |
357 | if (mmclash(u.u_mmap, low, high)) | |
358 | return (ENOMEM); | |
359 | } | |
360 | #endif | |
361 | if (error = swpexpand(ds + d, u.u_ssize, &u.u_dmap, &u.u_smap)) | |
362 | return (error); | |
363 | if (p->p_mmsize && (p->p_mmsize -= n) < 0) | |
364 | p->p_mmsize = 0; | |
365 | expand((int)n, 0); | |
366 | return (0); | |
367 | } | |
368 | ||
369 | /* | |
370 | * Macros for clearing a page's reference bits. | |
371 | */ | |
372 | #ifdef REFBIT | |
373 | #if !defined(tahoe) | |
374 | #define uncache(pte) /* XXX */ | |
375 | #endif | |
376 | ||
377 | #define CLRREF(pte, c, p, i) { \ | |
378 | if (!isatpte(p, pte)) \ | |
379 | uncache(pte); \ | |
380 | if (pte->pg_u) { \ | |
381 | c = &cmap[pgtocm(pte->pg_pfnum)]; \ | |
382 | if (c->c_lock) \ | |
383 | continue; \ | |
384 | pte->pg_u = 0; \ | |
385 | if (anycl(pte, pg_m)) \ | |
386 | pte->pg_m = 1; \ | |
387 | distcl(pte); \ | |
388 | if (isatpte(p, pte)) \ | |
389 | distpte(p->p_textp, i, pte); \ | |
390 | } \ | |
391 | } | |
392 | #else | |
393 | #define CLRREF(pte, c, p, i) { \ | |
394 | c = &cmap[pgtocm(pte->pg_pfnum)]; \ | |
395 | if (c->c_lock) \ | |
396 | continue; \ | |
397 | pte->pg_v = 0; \ | |
398 | if (anycl(pte, pg_m)) \ | |
399 | pte->pg_m = 1; \ | |
400 | distcl(pte); \ | |
401 | if (isatpte(p, pte)) \ | |
402 | distpte(p->p_textp, i, pte); \ | |
403 | } | |
404 | #endif | |
405 | ||
406 | /* ARGSUSED */ | |
407 | ovadvise(rp, uap, retval) | |
408 | register struct proc *rp; | |
409 | struct args { | |
410 | int anom; | |
411 | } *uap; | |
412 | int *retval; | |
413 | { | |
414 | int oanom = rp->p_flag & SUANOM; | |
415 | register struct pte *pte; | |
416 | register struct cmap *c; | |
417 | register unsigned i; | |
418 | ||
419 | trace(TR_VADVISE, uap->anom, rp->p_pid); | |
420 | rp->p_flag &= ~(SSEQL|SUANOM); | |
421 | switch (uap->anom) { | |
422 | ||
423 | case VA_ANOM: | |
424 | rp->p_flag |= SUANOM; | |
425 | break; | |
426 | ||
427 | case VA_SEQL: | |
428 | rp->p_flag |= SSEQL; | |
429 | break; | |
430 | } | |
431 | if ((oanom && (rp->p_flag & SUANOM) == 0) || uap->anom == VA_FLUSH) { | |
432 | for (i = 0; i < rp->p_dsize; i += CLSIZE) { | |
433 | pte = dptopte(rp, i); | |
434 | #ifdef MAPMEM | |
435 | /* don't do mmap pages */ | |
436 | if (pte->pg_v && !pte->pg_fod) | |
437 | #else | |
438 | if (pte->pg_v) | |
439 | #endif | |
440 | CLRREF(pte, c, rp, i); | |
441 | } | |
442 | } | |
443 | if (uap->anom == VA_FLUSH) { /* invalidate all pages */ | |
444 | for (i = 1; i < rp->p_ssize; i += CLSIZE) { | |
445 | pte = sptopte(rp, i); | |
446 | if (pte->pg_v) | |
447 | CLRREF(pte, c, rp, i); | |
448 | } | |
449 | for (i = 0; i < rp->p_tsize; i += CLSIZE) { | |
450 | pte = tptopte(rp, i); | |
451 | if (pte->pg_v) | |
452 | CLRREF(pte, c, rp, i); | |
453 | } | |
454 | } | |
455 | #if defined(vax) || defined(tahoe) | |
456 | mtpr(TBIA, 0); | |
457 | #endif | |
458 | #if defined(hp300) | |
459 | TBIAU(); | |
460 | #endif | |
461 | #if defined(i386) | |
462 | tlbflush(); | |
463 | #endif | |
464 | return (0); | |
465 | } | |
466 | /* END DEFUNCT */ | |
467 | ||
468 | /* | |
469 | * Grow the stack to include the SP; true return if successful. | |
470 | * Clients do not care about the cause of the error. | |
471 | */ | |
472 | grow(sp) | |
473 | unsigned sp; | |
474 | { | |
475 | int si, error; | |
476 | ||
477 | if (sp >= USRSTACK-ctob(u.u_ssize)) | |
478 | return (0); | |
479 | si = clrnd(btoc((USRSTACK-sp)) - u.u_ssize + SINCR); | |
480 | if (ctob(si) > u.u_rlimit[RLIMIT_STACK].rlim_cur) | |
481 | return (0); | |
482 | if (error = chksize((u_int)u.u_tsize, (u_int)u.u_dsize, (u_int)0, | |
483 | (u_int)u.u_ssize+si)) | |
484 | return (0); | |
485 | if (error = swpexpand(u.u_dsize, u.u_ssize + si, &u.u_dmap, &u.u_smap)) | |
486 | return (0); | |
487 | expand(si, 1); | |
488 | return (1); | |
489 | } | |
490 | ||
491 | #ifdef MAPMEM | |
492 | ||
493 | /* | |
494 | * Called from vpassvm() after full context has been passed from fup to tup. | |
495 | * Always called in the context of the parent. NOTE: routines should NOT | |
496 | * destroy regions. | |
497 | */ | |
498 | mmvfork(fup, tup) | |
499 | struct user *fup, *tup; | |
500 | { | |
501 | register struct mapmem *mp; | |
502 | ||
503 | tup->u_mmap = fup->u_mmap; | |
504 | fup->u_mmap = (struct mapmem *) 0; | |
505 | for (mp = tup->u_mmap; mp; mp = mp->mm_next) | |
506 | if (mp->mm_ops->mm_vfork) | |
507 | (*mp->mm_ops->mm_vfork)(mp, fup, tup); | |
508 | } | |
509 | ||
510 | /* | |
511 | * Called from procdup() for both parent and child. If in parent | |
512 | * we need to duplicate mapped memory regions. In both parent and | |
513 | * child, we call object specific routine. | |
514 | */ | |
515 | mmfork(pup, cup) | |
516 | struct user *pup, *cup; | |
517 | { | |
518 | register struct mapmem *mp, **mpp; | |
519 | int error = 0; | |
520 | ||
521 | if (pup) { | |
522 | mmdup(pup, cup); | |
523 | for (mp = pup->u_mmap; mp; mp = mp->mm_next) | |
524 | if (mp->mm_ops->mm_fork) | |
525 | (*mp->mm_ops->mm_fork)(mp, 0); | |
526 | } else { | |
527 | mpp = &u.u_mmap; | |
528 | for (mp = *mpp; mp; mp = *mpp) { | |
529 | if (mp->mm_ops->mm_fork) | |
530 | (*mp->mm_ops->mm_fork)(mp, 1); | |
531 | if (*mpp == mp) | |
532 | mpp = &mp->mm_next; | |
533 | } | |
534 | error = mmexpand(u.u_procp); | |
535 | } | |
536 | return (error); | |
537 | } | |
538 | ||
539 | /* | |
540 | * Its not clear that having a seperate exec routine is useful since | |
541 | * exec frees the address space immediately afterwards. We probably | |
542 | * need a post-exec hook to reestablish any mappings that persist | |
543 | * across execs. | |
544 | */ | |
545 | mmexec(p) | |
546 | struct proc *p; | |
547 | { | |
548 | register struct mapmem *mp, **mpp; | |
549 | int error1, error = 0; | |
550 | ||
551 | mpp = &u.u_mmap; | |
552 | for (mp = *mpp; mp; mp = *mpp) { | |
553 | if (mp->mm_ops->mm_exec) | |
554 | error = (*mp->mm_ops->mm_exec)(mp); | |
555 | if (*mpp == mp) { | |
556 | *mpp = mp->mm_next; | |
557 | MMFREE(mp); | |
558 | } | |
559 | } | |
560 | if (error1 = mmexpand(p)) | |
561 | return (error1); | |
562 | if (p->p_mmsize) | |
563 | panic("mmexec"); | |
564 | return (error); | |
565 | } | |
566 | ||
567 | /* | |
568 | * Called from exit just before releasing address space. | |
569 | * We always reclaim resources regardless of what the object routine does. | |
570 | */ | |
571 | mmexit(p) | |
572 | struct proc *p; | |
573 | { | |
574 | register struct mapmem *mp, **mpp; | |
575 | int error1, error = 0; | |
576 | ||
577 | mpp = &u.u_mmap; | |
578 | for (mp = *mpp; mp; mp = *mpp) { | |
579 | if (mp->mm_ops->mm_exit) | |
580 | error = (*mp->mm_ops->mm_exit)(mp); | |
581 | if (*mpp == mp) { | |
582 | *mpp = mp->mm_next; | |
583 | MMFREE(mp); | |
584 | } | |
585 | } | |
586 | if (error1 = mmexpand(p)) | |
587 | return (error1); | |
588 | if (p->p_mmsize) | |
589 | panic("mmexit"); | |
590 | return (error); | |
591 | } | |
592 | ||
593 | /* | |
594 | * Called from core just before dumping process image to core file. | |
595 | * Used to unmap regions which cannot be dumped; e.g. a region mapping | |
596 | * hardware registers which are write-only or must be accessed as bytes. | |
597 | */ | |
598 | mmcore(p) | |
599 | struct proc *p; | |
600 | { | |
601 | register struct mapmem *mp, **mpp; | |
602 | int error = 0, error1, changed = 0; | |
603 | ||
604 | mpp = &u.u_mmap; | |
605 | for (mp = *mpp; mp; mp = *mpp) { | |
606 | if ((mp->mm_prot & MM_NOCORE) == 0) { | |
607 | mpp = &mp->mm_next; | |
608 | continue; | |
609 | } | |
610 | if (mp->mm_ops->mm_exit) | |
611 | error = (*mp->mm_ops->mm_exit)(mp); | |
612 | if (*mpp == mp) { | |
613 | *mpp = mp->mm_next; | |
614 | MMFREE(mp); | |
615 | } | |
616 | changed++; | |
617 | } | |
618 | if (changed && (error1 = mmexpand(p))) | |
619 | return (error1); | |
620 | return (error); | |
621 | } | |
622 | ||
623 | /* | |
624 | * Duplicate mapped memory regions in a forked process. | |
625 | * XXX child may wind up short a few regions if not enough resources. | |
626 | */ | |
627 | mmdup(pu, cu) | |
628 | struct user *pu, *cu; | |
629 | { | |
630 | register struct mapmem *pmp, *cmp; | |
631 | register struct pte *ppte, *cpte; | |
632 | register segsz_t count; | |
633 | ||
634 | /* | |
635 | * First duplicate the mmap chain | |
636 | */ | |
637 | MMALLOC(cu->u_mmap); | |
638 | pmp = pu->u_mmap; | |
639 | cmp = cu->u_mmap; | |
640 | while (pmp && cmp) { | |
641 | *cmp = *pmp; | |
642 | if (pmp->mm_next) | |
643 | MMALLOC(cmp->mm_next); | |
644 | pmp = pmp->mm_next; | |
645 | cmp = cmp->mm_next; | |
646 | } | |
647 | /* | |
648 | * Now duplicate user address space that vmdup() won't do | |
649 | * i.e. mapped regions outside of data segment. | |
650 | */ | |
651 | ppte = dptopte(pu->u_procp, pu->u_procp->p_dsize); | |
652 | cpte = dptopte(cu->u_procp, cu->u_procp->p_dsize); | |
653 | for (count = pu->u_procp->p_mmsize; count; count--) { | |
654 | if (ppte->pg_fod && ppte->pg_v) | |
655 | *(int *)cpte = *(int *)ppte; | |
656 | ppte++, cpte++; | |
657 | } | |
658 | cu->u_procp->p_flag |= SPTECHG; | |
659 | } | |
660 | ||
661 | mmalloc(p, id, uvap, size, prot, ops, mpp) | |
662 | struct proc *p; | |
663 | caddr_t *uvap; | |
664 | segsz_t size; | |
665 | struct mapmemops *ops; | |
666 | struct mapmem **mpp; | |
667 | { | |
668 | register struct mapmem *mp; | |
669 | register u_int uva; | |
670 | int error; | |
671 | ||
672 | /* | |
673 | * Validate size first | |
674 | */ | |
675 | if (size <= 0 || (size & CLOFSET)) | |
676 | return(EINVAL); | |
677 | /* | |
678 | * A uva of zero means to map at our discretion. | |
679 | * Our strategy is to place the segment at the max of: | |
680 | * - the current data + mapped memory size | |
681 | * - the default data size limit | |
682 | * (if it will fit within the MAXDSIZ limit) | |
683 | * If this is the first mapped memory region beyond the data | |
684 | * segment we round to a MMSEG boundary to allow for data | |
685 | * segment growth. | |
686 | */ | |
687 | uva = (u_int) *uvap; | |
688 | if (uva == 0) { | |
689 | register u_int uva2; | |
690 | ||
691 | uva = ctob(dptov(p, u.u_dsize + p->p_mmsize)); | |
692 | uva2 = ctob(dptov(p, btoc(DFLDSIZ))); | |
693 | uva2 = ((uva2 + (MMSEG-1)) & ~(MMSEG-1)); | |
694 | if (uva < uva2 && | |
695 | uva2 + size < ctob(dptov(p, btoc(MAXDSIZ)))) | |
696 | uva = uva2; | |
697 | else if (p->p_mmsize == 0) | |
698 | uva = ((uva + (MMSEG-1)) & ~(MMSEG-1)); | |
699 | } | |
700 | /* | |
701 | * Impose necessary constraints on address. | |
702 | */ | |
703 | if ((uva & CLOFSET) || uva < ctob(dptov(p, 0)) || | |
704 | uva+size >= ctob(sptov(p, u.u_ssize))) | |
705 | return (EINVAL); | |
706 | if (mmclash(u.u_mmap, (caddr_t)uva, (caddr_t)uva+size)) | |
707 | return (EINVAL); | |
708 | /* | |
709 | * Finally, allocate and initialize descriptor and expand | |
710 | * user address space as necessary. | |
711 | */ | |
712 | MMALLOC(mp); | |
713 | if (mp == MMNIL) | |
714 | return (ENOMEM); | |
715 | mp->mm_next = u.u_mmap; | |
716 | mp->mm_id = id; | |
717 | mp->mm_uva = (caddr_t) uva; | |
718 | mp->mm_size = size; | |
719 | mp->mm_prot = prot; | |
720 | mp->mm_ops = ops; | |
721 | u.u_mmap = mp; | |
722 | if (error = mmexpand(p)) { | |
723 | u.u_mmap = mp->mm_next; | |
724 | MMFREE(mp); | |
725 | return(error); | |
726 | } | |
727 | *uvap = (caddr_t) uva; | |
728 | *mpp = mp; | |
729 | return(0); | |
730 | } | |
731 | ||
732 | mmfree(p, mp) | |
733 | struct proc *p; | |
734 | register struct mapmem *mp; | |
735 | { | |
736 | register struct mapmem *cmp, **mpp; | |
737 | ||
738 | /* | |
739 | * Remove region from chain | |
740 | */ | |
741 | mpp = &u.u_mmap; | |
742 | for (cmp = *mpp; cmp; cmp = *mpp) { | |
743 | if (cmp == mp) | |
744 | break; | |
745 | mpp = &cmp->mm_next; | |
746 | } | |
747 | if (cmp == MMNIL) | |
748 | panic("mmfree"); | |
749 | *mpp = mp->mm_next; | |
750 | MMFREE(mp); | |
751 | return (mmexpand(p)); | |
752 | } | |
753 | ||
754 | mmmapin(p, mp, mapfunc) | |
755 | register struct proc *p; | |
756 | register struct mapmem *mp; | |
757 | int (*mapfunc)(); | |
758 | { | |
759 | register struct pte *pte; | |
760 | register int off; | |
761 | struct pte *dpte; | |
762 | int pm, fv, lv; | |
763 | ||
764 | /* | |
765 | * Verify that range can be mapped | |
766 | */ | |
767 | for (off = 0; off < mp->mm_size; off += NBPG) | |
768 | if ((*mapfunc)(mp, off) == -1) | |
769 | return (EINVAL); | |
770 | /* | |
771 | * Now verify that region is in range | |
772 | */ | |
773 | fv = btop(mp->mm_uva); | |
774 | lv = btop(mp->mm_uva + mp->mm_size - 1); | |
775 | if (fv < dptov(p, 0) || | |
776 | lv >= dptov(p, u.u_dsize + p->p_mmsize)) | |
777 | return (ENOMEM); | |
778 | /* | |
779 | * Finally, do the mapping. | |
780 | */ | |
781 | if (mp->mm_prot & MM_RO) | |
782 | pm = PG_URKR|PG_FOD|PG_V; | |
783 | else | |
784 | pm = PG_UW|PG_FOD|PG_V; | |
785 | #if defined(hp300) | |
786 | if (mp->mm_prot & MM_CI) | |
787 | pm |= PG_CI; | |
788 | #endif | |
789 | pte = vtopte(p, fv); | |
790 | dpte = dptopte(p, u.u_dsize); | |
791 | for (off = 0; off < mp->mm_size; off += NBPG) { | |
792 | if ((off&CLOFSET) == 0 && pte < dpte) | |
793 | p->p_rssize -= vmemfree(pte, CLSIZE); | |
794 | *(int *)pte = pm; | |
795 | pte->pg_pfnum = (*mapfunc)(mp, off); | |
796 | pte++; | |
797 | } | |
798 | newptes(vtopte(p, fv), (u_int)fv, (int)btoc(mp->mm_size)); | |
799 | return (0); | |
800 | } | |
801 | ||
802 | mmmapout(p, mp) | |
803 | register struct proc *p; | |
804 | register struct mapmem *mp; | |
805 | { | |
806 | register struct pte *pte; | |
807 | register int off; | |
808 | struct pte *dpte; | |
809 | int fv, lv; | |
810 | ||
811 | fv = btop(mp->mm_uva); | |
812 | lv = btop(mp->mm_uva + mp->mm_size - 1); | |
813 | if (fv < dptov(p, 0) || | |
814 | lv >= dptov(p, u.u_dsize + p->p_mmsize)) | |
815 | panic("mmmapout"); | |
816 | pte = vtopte(p, fv); | |
817 | dpte = dptopte(p, u.u_dsize); | |
818 | for (off = 0; off < mp->mm_size; off += NBPG) { | |
819 | if (pte < dpte) { | |
820 | if ((off & CLOFSET) == 0) | |
821 | p->p_rssize -= vmemfree(pte, CLSIZE); | |
822 | *(int *)pte = (PG_UW|PG_FOD); | |
823 | ((struct fpte *)pte)->pg_fileno = PG_FZERO; | |
824 | } else | |
825 | *(int *)pte = 0; | |
826 | pte++; | |
827 | } | |
828 | newptes(vtopte(p, fv), (u_int)fv, (int)btoc(mp->mm_size)); | |
829 | } | |
830 | ||
831 | mmexpand(p) | |
832 | struct proc *p; | |
833 | { | |
834 | register int szpt, change; | |
835 | caddr_t high; | |
836 | segsz_t nsize, oms; | |
837 | ||
838 | oms = p->p_mmsize; | |
839 | /* | |
840 | * Get new mmsize based on existing regions and use | |
841 | * that to calculate change in page table size. | |
842 | */ | |
843 | if (u.u_mmap) { | |
844 | mmrange(u.u_mmap, (caddr_t *)0, &high); | |
845 | nsize = btop(high) - dptov(p, u.u_dsize) + 1; | |
846 | if (nsize < 0) | |
847 | nsize = 0; | |
848 | } else | |
849 | nsize = 0; | |
850 | change = nsize - oms; | |
851 | if (change == 0) | |
852 | return(0); | |
853 | ||
854 | /* | |
855 | * Ensure data + mapped memory fits within maximum data limit. | |
856 | * This is possibly a little restrictive, but it helps keep | |
857 | * page table sizes down. | |
858 | */ | |
859 | if (change > 0 && | |
860 | (ctob(oms+change) > u.u_rlimit[RLIMIT_DATA].rlim_max || | |
861 | ctob(u.u_dsize+oms+change) > u.u_rlimit[RLIMIT_DATA].rlim_max)) | |
862 | return(ENOMEM); | |
863 | /* | |
864 | * Expand page table if necessary. | |
865 | * Note that ptexpand takes care of flushing the translation buffer. | |
866 | */ | |
867 | p->p_mmsize += change; | |
868 | #if defined(hp300) || defined(i386) | |
869 | szpt = ptsize(p) - u.u_pcb.pcb_szpt; | |
870 | if (szpt > 0) | |
871 | ptexpand(szpt, u.u_dsize, oms, u.u_ssize); | |
872 | setp0lr(u.u_pcb.pcb_p0lr + change); | |
873 | #endif | |
874 | #if defined(vax) || defined(tahoe) | |
875 | #if defined(vax) | |
876 | szpt = (u.u_pcb.pcb_p1br + (u.u_pcb.pcb_p1lr&~PME_CLR)) - | |
877 | (u.u_pcb.pcb_p0br + (u.u_pcb.pcb_p0lr&~AST_CLR)); | |
878 | #else | |
879 | szpt = (u.u_pcb.pcb_p2br + u.u_pcb.pcb_p2lr) - | |
880 | (u.u_pcb.pcb_p0br + u.u_pcb.pcb_p0lr); | |
881 | #endif | |
882 | if (change > szpt) | |
883 | ptexpand(clrnd(ctopt(change - szpt)), u.u_dsize, oms, u.u_ssize); | |
884 | /* | |
885 | * Clear new ptes. | |
886 | * We need to do this because there may be bogus (yet technically | |
887 | * valid) ptes above the old p0lr value. This can happen if the | |
888 | * data segment has shrunk in the past leaving such ptes behind. | |
889 | * There is no need to invalidate such ptes at that time since the | |
890 | * length register will prevent their use. We are safe on the HPs | |
891 | * because we do invalidate old ptes in setp0lr() when shrinking. | |
892 | */ | |
893 | if (change > 0) { | |
894 | struct pte *bpte; | |
895 | ||
896 | #if defined(vax) | |
897 | bpte = u.u_pcb.pcb_p0br + (u.u_pcb.pcb_p0lr&~AST_CLR); | |
898 | #else | |
899 | bpte = u.u_pcb.pcb_p0br + u.u_pcb.pcb_p0lr; | |
900 | #endif | |
901 | bzero((caddr_t)bpte, change * sizeof(struct pte)); | |
902 | mtpr(TBIA, 0); | |
903 | } | |
904 | /* avoid side-effects of setp0lr */ | |
905 | #if defined(vax) | |
906 | change += u.u_pcb.pcb_p0lr &~ AST_CLR; | |
907 | #else | |
908 | change += u.u_pcb.pcb_p0lr; | |
909 | #endif | |
910 | setp0lr(change); | |
911 | #endif | |
912 | return(0); | |
913 | } | |
914 | ||
915 | mmrange(mp, lap, hap) | |
916 | register struct mapmem *mp; | |
917 | caddr_t *lap, *hap; | |
918 | { | |
919 | register caddr_t low, high, top; | |
920 | ||
921 | low = high = 0; | |
922 | while (mp) { | |
923 | if (low == 0 || mp->mm_uva < low) | |
924 | low = mp->mm_uva; | |
925 | top = mp->mm_uva + mp->mm_size - 1; | |
926 | if (high == 0 || top > high) | |
927 | high = top; | |
928 | mp = mp->mm_next; | |
929 | } | |
930 | if (lap) | |
931 | *lap = low; | |
932 | if (hap) | |
933 | *hap = high; | |
934 | } | |
935 | ||
936 | mmclash(mp, la, ha) | |
937 | register struct mapmem *mp; | |
938 | caddr_t la, ha; | |
939 | { | |
940 | while (mp) { | |
941 | if (ha > mp->mm_uva && la < mp->mm_uva + mp->mm_size) | |
942 | return(1); | |
943 | mp = mp->mm_next; | |
944 | } | |
945 | return(0); | |
946 | } | |
947 | ||
948 | #endif /* MAPMEM */ |