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 2006 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
27 /* All Rights Reserved */
28
29 /*
30 * Portions of this source code were derived from Berkeley 4.3 BSD
31 * under license from the Regents of the University of California.
32 */
33
34 /*
35 * VM - segment for non-faulting loads.
36 */
37
38 #include <sys/types.h>
39 #include <sys/t_lock.h>
40 #include <sys/param.h>
41 #include <sys/mman.h>
42 #include <sys/errno.h>
43 #include <sys/kmem.h>
44 #include <sys/cmn_err.h>
45 #include <sys/vnode.h>
46 #include <sys/proc.h>
47 #include <sys/conf.h>
48 #include <sys/debug.h>
49 #include <sys/archsystm.h>
50 #include <sys/lgrp.h>
51
52 #include <vm/page.h>
53 #include <vm/hat.h>
54 #include <vm/as.h>
55 #include <vm/seg.h>
56 #include <vm/vpage.h>
57
58 /*
59 * Private seg op routines.
60 */
61 static int segnf_dup(struct seg *seg, struct seg *newseg);
62 static int segnf_unmap(struct seg *seg, caddr_t addr, size_t len);
63 static void segnf_free(struct seg *seg);
64 static faultcode_t segnf_nomap(void);
65 static int segnf_setprot(struct seg *seg, caddr_t addr,
66 size_t len, uint_t prot);
67 static int segnf_checkprot(struct seg *seg, caddr_t addr,
68 size_t len, uint_t prot);
69 static int segnf_nop(void);
70 static int segnf_getprot(struct seg *seg, caddr_t addr,
71 size_t len, uint_t *protv);
72 static u_offset_t segnf_getoffset(struct seg *seg, caddr_t addr);
73 static int segnf_gettype(struct seg *seg, caddr_t addr);
74 static int segnf_getvp(struct seg *seg, caddr_t addr, struct vnode **vpp);
75 static void segnf_dump(struct seg *seg);
76 static int segnf_pagelock(struct seg *seg, caddr_t addr, size_t len,
77 struct page ***ppp, enum lock_type type, enum seg_rw rw);
78 static int segnf_setpagesize(struct seg *seg, caddr_t addr, size_t len,
79 uint_t szc);
80
81
82 struct seg_ops segnf_ops = {
83 .dup = segnf_dup,
84 .unmap = segnf_unmap,
85 .free = segnf_free,
86 .fault = (faultcode_t (*)(struct hat *, struct seg *, caddr_t,
87 size_t, enum fault_type, enum seg_rw))segnf_nomap,
88 .faulta = (faultcode_t (*)(struct seg *, caddr_t)) segnf_nomap,
89 .setprot = segnf_setprot,
90 .checkprot = segnf_checkprot,
91 .sync = (int (*)(struct seg *, caddr_t, size_t, int, uint_t))
92 segnf_nop,
93 .incore = (size_t (*)(struct seg *, caddr_t, size_t, char *))
94 segnf_nop,
95 .lockop = (int (*)(struct seg *, caddr_t, size_t, int, int,
96 ulong_t *, size_t))segnf_nop,
97 .getprot = segnf_getprot,
98 .getoffset = segnf_getoffset,
99 .gettype = segnf_gettype,
100 .getvp = segnf_getvp,
101 .advise = (int (*)(struct seg *, caddr_t, size_t, uint_t))
102 segnf_nop,
103 .dump = segnf_dump,
104 .pagelock = segnf_pagelock,
105 .setpagesize = segnf_setpagesize,
106 };
107
108 /*
109 * vnode and page for the page of zeros we use for the nf mappings.
110 */
111 static kmutex_t segnf_lock;
112 static struct vnode nfvp;
113 static struct page **nfpp;
114
115 #define addr_to_vcolor(addr) \
116 (shm_alignment) ? \
117 ((int)(((uintptr_t)(addr) & (shm_alignment - 1)) >> PAGESHIFT)) : 0
118
119 /*
120 * We try to limit the number of Non-fault segments created.
121 * Non fault segments are created to optimize sparc V9 code which uses
122 * the sparc nonfaulting load ASI (ASI_PRIMARY_NOFAULT).
123 *
124 * There are several reasons why creating too many non-fault segments
125 * could cause problems.
126 *
127 * First, excessive allocation of kernel resources for the seg
128 * structures and the HAT data to map the zero pages.
129 *
130 * Secondly, creating nofault segments actually uses up user virtual
131 * address space. This makes it unavailable for subsequent mmap(0, ...)
132 * calls which use as_gap() to find empty va regions. Creation of too
133 * many nofault segments could thus interfere with the ability of the
134 * runtime linker to load a shared object.
135 */
136 #define MAXSEGFORNF (10000)
137 #define MAXNFSEARCH (5)
138
139
140 /*
141 * Must be called from startup()
142 */
143 void
144 segnf_init()
145 {
146 mutex_init(&segnf_lock, NULL, MUTEX_DEFAULT, NULL);
147 }
148
149
150 /*
151 * Create a no-fault segment.
152 *
153 * The no-fault segment is not technically necessary, as the code in
154 * nfload() in trap.c will emulate the SPARC instruction and load
155 * a value of zero in the destination register.
156 *
157 * However, this code tries to put a page of zero's at the nofault address
158 * so that subsequent non-faulting loads to the same page will not
159 * trap with a tlb miss.
160 *
161 * In order to help limit the number of segments we merge adjacent nofault
162 * segments into a single segment. If we get a large number of segments
163 * we'll also try to delete a random other nf segment.
164 */
165 /* ARGSUSED */
166 int
167 segnf_create(struct seg *seg, void *argsp)
168 {
169 uint_t prot;
170 pgcnt_t vacpgs;
171 u_offset_t off = 0;
172 caddr_t vaddr = NULL;
173 int i, color;
174 struct seg *s1;
175 struct seg *s2;
176 size_t size;
177 struct as *as = seg->s_as;
178
179 ASSERT(as && AS_WRITE_HELD(as, &as->a_lock));
180
181 /*
182 * Need a page per virtual color or just 1 if no vac.
183 */
184 mutex_enter(&segnf_lock);
185 if (nfpp == NULL) {
186 struct seg kseg;
187
188 vacpgs = 1;
189 if (shm_alignment > PAGESIZE) {
190 vacpgs = shm_alignment >> PAGESHIFT;
191 }
192
193 nfpp = kmem_alloc(sizeof (*nfpp) * vacpgs, KM_SLEEP);
194
195 kseg.s_as = &kas;
196 for (i = 0; i < vacpgs; i++, off += PAGESIZE,
197 vaddr += PAGESIZE) {
198 nfpp[i] = page_create_va(&nfvp, off, PAGESIZE,
199 PG_WAIT | PG_NORELOC, &kseg, vaddr);
200 page_io_unlock(nfpp[i]);
201 page_downgrade(nfpp[i]);
202 pagezero(nfpp[i], 0, PAGESIZE);
203 }
204 }
205 mutex_exit(&segnf_lock);
206
207 hat_map(as->a_hat, seg->s_base, seg->s_size, HAT_MAP);
208
209 /*
210 * s_data can't be NULL because of ASSERTS in the common vm code.
211 */
212 seg->s_ops = &segnf_ops;
213 seg->s_data = seg;
214 seg->s_flags |= S_PURGE;
215
216 mutex_enter(&as->a_contents);
217 as->a_flags |= AS_NEEDSPURGE;
218 mutex_exit(&as->a_contents);
219
220 prot = PROT_READ;
221 color = addr_to_vcolor(seg->s_base);
222 if (as != &kas)
223 prot |= PROT_USER;
224 hat_memload(as->a_hat, seg->s_base, nfpp[color],
225 prot | HAT_NOFAULT, HAT_LOAD);
226
227 /*
228 * At this point see if we can concatenate a segment to
229 * a non-fault segment immediately before and/or after it.
230 */
231 if ((s1 = AS_SEGPREV(as, seg)) != NULL &&
232 s1->s_ops == &segnf_ops &&
233 s1->s_base + s1->s_size == seg->s_base) {
234 size = s1->s_size;
235 seg_free(s1);
236 seg->s_base -= size;
237 seg->s_size += size;
238 }
239
240 if ((s2 = AS_SEGNEXT(as, seg)) != NULL &&
241 s2->s_ops == &segnf_ops &&
242 seg->s_base + seg->s_size == s2->s_base) {
243 size = s2->s_size;
244 seg_free(s2);
245 seg->s_size += size;
246 }
247
248 /*
249 * if we already have a lot of segments, try to delete some other
250 * nofault segment to reduce the probability of uncontrolled segment
251 * creation.
252 *
253 * the code looks around quickly (no more than MAXNFSEARCH segments
254 * each way) for another NF segment and then deletes it.
255 */
256 if (avl_numnodes(&as->a_segtree) > MAXSEGFORNF) {
257 size = 0;
258 s2 = NULL;
259 s1 = AS_SEGPREV(as, seg);
260 while (size++ < MAXNFSEARCH && s1 != NULL) {
261 if (s1->s_ops == &segnf_ops)
262 s2 = s1;
263 s1 = AS_SEGPREV(s1->s_as, seg);
264 }
265 if (s2 == NULL) {
266 s1 = AS_SEGNEXT(as, seg);
267 while (size-- > 0 && s1 != NULL) {
268 if (s1->s_ops == &segnf_ops)
269 s2 = s1;
270 s1 = AS_SEGNEXT(as, seg);
271 }
272 }
273 if (s2 != NULL)
274 seg_unmap(s2);
275 }
276
277 return (0);
278 }
279
280 /*
281 * Never really need "No fault" segments, so they aren't dup'd.
282 */
283 /* ARGSUSED */
284 static int
285 segnf_dup(struct seg *seg, struct seg *newseg)
286 {
287 panic("segnf_dup");
288 return (0);
289 }
290
291 /*
292 * Split a segment at addr for length len.
293 */
294 static int
295 segnf_unmap(struct seg *seg, caddr_t addr, size_t len)
296 {
297 ASSERT(seg->s_as && AS_WRITE_HELD(seg->s_as, &seg->s_as->a_lock));
298
299 /*
300 * Check for bad sizes.
301 */
302 if (addr < seg->s_base || addr + len > seg->s_base + seg->s_size ||
303 (len & PAGEOFFSET) || ((uintptr_t)addr & PAGEOFFSET)) {
304 cmn_err(CE_PANIC, "segnf_unmap: bad unmap size");
305 }
306
307 /*
308 * Unload any hardware translations in the range to be taken out.
309 */
310 hat_unload(seg->s_as->a_hat, addr, len, HAT_UNLOAD_UNMAP);
311
312 if (addr == seg->s_base && len == seg->s_size) {
313 /*
314 * Freeing entire segment.
315 */
316 seg_free(seg);
317 } else if (addr == seg->s_base) {
318 /*
319 * Freeing the beginning of the segment.
320 */
321 seg->s_base += len;
322 seg->s_size -= len;
323 } else if (addr + len == seg->s_base + seg->s_size) {
324 /*
325 * Freeing the end of the segment.
326 */
327 seg->s_size -= len;
328 } else {
329 /*
330 * The section to go is in the middle of the segment, so we
331 * have to cut it into two segments. We shrink the existing
332 * "seg" at the low end, and create "nseg" for the high end.
333 */
334 caddr_t nbase = addr + len;
335 size_t nsize = (seg->s_base + seg->s_size) - nbase;
336 struct seg *nseg;
337
338 /*
339 * Trim down "seg" before trying to stick "nseg" into the as.
340 */
341 seg->s_size = addr - seg->s_base;
342 nseg = seg_alloc(seg->s_as, nbase, nsize);
343 if (nseg == NULL)
344 cmn_err(CE_PANIC, "segnf_unmap: seg_alloc failed");
345
346 /*
347 * s_data can't be NULL because of ASSERTs in common VM code.
348 */
349 nseg->s_ops = seg->s_ops;
350 nseg->s_data = nseg;
351 nseg->s_flags |= S_PURGE;
352 mutex_enter(&seg->s_as->a_contents);
353 seg->s_as->a_flags |= AS_NEEDSPURGE;
354 mutex_exit(&seg->s_as->a_contents);
355 }
356
357 return (0);
358 }
359
360 /*
361 * Free a segment.
362 */
363 static void
364 segnf_free(struct seg *seg)
365 {
366 ASSERT(seg->s_as && AS_WRITE_HELD(seg->s_as, &seg->s_as->a_lock));
367 }
368
369 /*
370 * No faults allowed on segnf.
371 */
372 static faultcode_t
373 segnf_nomap(void)
374 {
375 return (FC_NOMAP);
376 }
377
378 /* ARGSUSED */
379 static int
380 segnf_setprot(struct seg *seg, caddr_t addr, size_t len, uint_t prot)
381 {
382 ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
383 return (EACCES);
384 }
385
386 /* ARGSUSED */
387 static int
388 segnf_checkprot(struct seg *seg, caddr_t addr, size_t len, uint_t prot)
389 {
390 uint_t sprot;
391 ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
392
393 sprot = seg->s_as == &kas ? PROT_READ : PROT_READ|PROT_USER;
394 return ((prot & sprot) == prot ? 0 : EACCES);
395 }
396
397 static int
398 segnf_nop(void)
399 {
400 return (0);
401 }
402
403 static int
404 segnf_getprot(struct seg *seg, caddr_t addr, size_t len, uint_t *protv)
405 {
406 size_t pgno = seg_page(seg, addr + len) - seg_page(seg, addr) + 1;
407 size_t p;
408 ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
409
410 for (p = 0; p < pgno; ++p)
411 protv[p] = PROT_READ;
412 return (0);
413 }
414
415 /* ARGSUSED */
416 static u_offset_t
417 segnf_getoffset(struct seg *seg, caddr_t addr)
418 {
419 ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
420
421 return ((u_offset_t)0);
422 }
423
424 /* ARGSUSED */
425 static int
426 segnf_gettype(struct seg *seg, caddr_t addr)
427 {
428 ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
429
430 return (MAP_SHARED);
431 }
432
433 /* ARGSUSED */
434 static int
435 segnf_getvp(struct seg *seg, caddr_t addr, struct vnode **vpp)
436 {
437 ASSERT(seg->s_as && AS_LOCK_HELD(seg->s_as, &seg->s_as->a_lock));
438
439 *vpp = &nfvp;
440 return (0);
441 }
442
443 /*
444 * segnf pages are not dumped, so we just return
445 */
446 /* ARGSUSED */
447 static void
448 segnf_dump(struct seg *seg)
449 {}
450
451 /*ARGSUSED*/
452 static int
453 segnf_pagelock(struct seg *seg, caddr_t addr, size_t len,
454 struct page ***ppp, enum lock_type type, enum seg_rw rw)
455 {
456 return (ENOTSUP);
457 }
458
459 /*ARGSUSED*/
460 static int
461 segnf_setpagesize(struct seg *seg, caddr_t addr, size_t len,
462 uint_t szc)
463 {
464 return (ENOTSUP);
465 }