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