1 /* $OpenBSD: uvm_glue.c,v 1.69 2014/12/15 20:38:22 tedu Exp $ */ 2 /* $NetBSD: uvm_glue.c,v 1.44 2001/02/06 19:54:44 eeh Exp $ */ 3 4 /* 5 * Copyright (c) 1997 Charles D. Cranor and Washington University. 6 * Copyright (c) 1991, 1993, The Regents of the University of California. 7 * 8 * All rights reserved. 9 * 10 * This code is derived from software contributed to Berkeley by 11 * The Mach Operating System project at Carnegie-Mellon University. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 38 * from: Id: uvm_glue.c,v 1.1.2.8 1998/02/07 01:16:54 chs Exp 39 * 40 * 41 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 42 * All rights reserved. 43 * 44 * Permission to use, copy, modify and distribute this software and 45 * its documentation is hereby granted, provided that both the copyright 46 * notice and this permission notice appear in all copies of the 47 * software, derivative works or modified versions, and any portions 48 * thereof, and that both notices appear in supporting documentation. 49 * 50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 53 * 54 * Carnegie Mellon requests users of this software to return to 55 * 56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 57 * School of Computer Science 58 * Carnegie Mellon University 59 * Pittsburgh PA 15213-3890 60 * 61 * any improvements or extensions that they make and grant Carnegie the 62 * rights to redistribute these changes. 63 */ 64 65 /* 66 * uvm_glue.c: glue functions 67 */ 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/proc.h> 72 #include <sys/resourcevar.h> 73 #include <sys/buf.h> 74 #include <sys/user.h> 75 #ifdef SYSVSHM 76 #include <sys/shm.h> 77 #endif 78 #include <sys/sched.h> 79 80 #include <uvm/uvm.h> 81 82 /* 83 * uvm_kernacc: can the kernel access a region of memory 84 * 85 * - called from malloc [DIAGNOSTIC], and /dev/kmem driver (mem.c) 86 */ 87 boolean_t 88 uvm_kernacc(caddr_t addr, size_t len, int rw) 89 { 90 boolean_t rv; 91 vaddr_t saddr, eaddr; 92 vm_prot_t prot = rw == B_READ ? PROT_READ : PROT_WRITE; 93 94 saddr = trunc_page((vaddr_t)addr); 95 eaddr = round_page((vaddr_t)addr + len); 96 vm_map_lock_read(kernel_map); 97 rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot); 98 vm_map_unlock_read(kernel_map); 99 100 return(rv); 101 } 102 103 #ifdef KGDB 104 /* 105 * Change protections on kernel pages from addr to addr+len 106 * (presumably so debugger can plant a breakpoint). 107 * 108 * We force the protection change at the pmap level. If we were 109 * to use vm_map_protect a change to allow writing would be lazily- 110 * applied meaning we would still take a protection fault, something 111 * we really don't want to do. It would also fragment the kernel 112 * map unnecessarily. We cannot use pmap_protect since it also won't 113 * enforce a write-enable request. Using pmap_enter is the only way 114 * we can ensure the change takes place properly. 115 */ 116 void 117 uvm_chgkprot(caddr_t addr, size_t len, int rw) 118 { 119 vm_prot_t prot; 120 paddr_t pa; 121 vaddr_t sva, eva; 122 123 prot = rw == B_READ ? PROT_READ : PROT_READ | PROT_WRITE; 124 eva = round_page((vaddr_t)addr + len); 125 for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) { 126 /* 127 * Extract physical address for the page. 128 * We use a cheezy hack to differentiate physical 129 * page 0 from an invalid mapping, not that it 130 * really matters... 131 */ 132 if (pmap_extract(pmap_kernel(), sva, &pa) == FALSE) 133 panic("chgkprot: invalid page"); 134 pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED); 135 } 136 pmap_update(pmap_kernel()); 137 } 138 #endif 139 140 /* 141 * uvm_vslock: wire user memory for I/O 142 * 143 * - called from physio and sys___sysctl 144 */ 145 146 int 147 uvm_vslock(struct proc *p, caddr_t addr, size_t len, vm_prot_t access_type) 148 { 149 struct vm_map *map; 150 vaddr_t start, end; 151 int rv; 152 153 map = &p->p_vmspace->vm_map; 154 start = trunc_page((vaddr_t)addr); 155 end = round_page((vaddr_t)addr + len); 156 if (end <= start) 157 return (EINVAL); 158 159 rv = uvm_fault_wire(map, start, end, access_type); 160 161 return (rv); 162 } 163 164 /* 165 * uvm_vsunlock: unwire user memory wired by uvm_vslock() 166 * 167 * - called from physio and sys___sysctl 168 */ 169 170 void 171 uvm_vsunlock(struct proc *p, caddr_t addr, size_t len) 172 { 173 vaddr_t start, end; 174 175 start = trunc_page((vaddr_t)addr); 176 end = round_page((vaddr_t)addr + len); 177 if (end <= start) 178 return; 179 180 uvm_fault_unwire(&p->p_vmspace->vm_map, start, end); 181 } 182 183 /* 184 * uvm_vslock_device: wire user memory, make sure it's device reachable 185 * and bounce if necessary. 186 * Always bounces for now. 187 */ 188 int 189 uvm_vslock_device(struct proc *p, void *addr, size_t len, 190 vm_prot_t access_type, void **retp) 191 { 192 struct vm_page *pg; 193 struct pglist pgl; 194 int npages; 195 vaddr_t start, end, off; 196 vaddr_t sva, va; 197 vsize_t sz; 198 int error, i; 199 200 start = trunc_page((vaddr_t)addr); 201 end = round_page((vaddr_t)addr + len); 202 sz = end - start; 203 off = (vaddr_t)addr - start; 204 if (end <= start) 205 return (EINVAL); 206 207 if ((error = uvm_fault_wire(&p->p_vmspace->vm_map, start, end, 208 access_type))) { 209 return (error); 210 } 211 212 npages = atop(sz); 213 for (i = 0; i < npages; i++) { 214 paddr_t pa; 215 216 if (!pmap_extract(p->p_vmspace->vm_map.pmap, 217 start + ptoa(i), &pa)) { 218 error = EFAULT; 219 goto out_unwire; 220 } 221 if (!PADDR_IS_DMA_REACHABLE(pa)) 222 break; 223 } 224 if (i == npages) { 225 *retp = NULL; 226 return (0); 227 } 228 229 if ((va = uvm_km_valloc(kernel_map, sz)) == 0) { 230 error = ENOMEM; 231 goto out_unwire; 232 } 233 sva = va; 234 235 TAILQ_INIT(&pgl); 236 error = uvm_pglistalloc(npages * PAGE_SIZE, dma_constraint.ucr_low, 237 dma_constraint.ucr_high, 0, 0, &pgl, npages, UVM_PLA_WAITOK); 238 if (error) 239 goto out_unmap; 240 241 while ((pg = TAILQ_FIRST(&pgl)) != NULL) { 242 TAILQ_REMOVE(&pgl, pg, pageq); 243 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), PROT_READ | PROT_WRITE); 244 va += PAGE_SIZE; 245 } 246 pmap_update(pmap_kernel()); 247 KASSERT(va == sva + sz); 248 *retp = (void *)(sva + off); 249 250 if ((error = copyin(addr, *retp, len)) == 0) 251 return 0; 252 253 uvm_km_pgremove_intrsafe(sva, sva + sz); 254 pmap_kremove(sva, sz); 255 pmap_update(pmap_kernel()); 256 out_unmap: 257 uvm_km_free(kernel_map, sva, sz); 258 out_unwire: 259 uvm_fault_unwire(&p->p_vmspace->vm_map, start, end); 260 return (error); 261 } 262 263 void 264 uvm_vsunlock_device(struct proc *p, void *addr, size_t len, void *map) 265 { 266 vaddr_t start, end; 267 vaddr_t kva; 268 vsize_t sz; 269 270 start = trunc_page((vaddr_t)addr); 271 end = round_page((vaddr_t)addr + len); 272 sz = end - start; 273 if (end <= start) 274 return; 275 276 if (map) 277 copyout(map, addr, len); 278 uvm_fault_unwire(&p->p_vmspace->vm_map, start, end); 279 280 if (!map) 281 return; 282 283 kva = trunc_page((vaddr_t)map); 284 uvm_km_pgremove_intrsafe(kva, kva + sz); 285 pmap_kremove(kva, sz); 286 pmap_update(pmap_kernel()); 287 uvm_km_free(kernel_map, kva, sz); 288 } 289 290 /* 291 * uvm_uarea_alloc: allocate the u-area for a new thread 292 */ 293 vaddr_t 294 uvm_uarea_alloc(void) 295 { 296 vaddr_t uaddr; 297 298 uaddr = uvm_km_kmemalloc_pla(kernel_map, uvm.kernel_object, USPACE, 299 USPACE_ALIGN, UVM_KMF_ZERO, 300 no_constraint.ucr_low, no_constraint.ucr_high, 301 0, 0, USPACE/PAGE_SIZE); 302 303 #ifdef PMAP_UAREA 304 /* Tell the pmap this is a u-area mapping */ 305 if (uaddr != 0) 306 PMAP_UAREA(uaddr); 307 #endif 308 309 return (uaddr); 310 } 311 312 /* 313 * uvm_uarea_free: free a dead thread's stack 314 * 315 * - the thread passed to us is a dead thread; we 316 * are running on a different context now (the reaper). 317 */ 318 void 319 uvm_uarea_free(struct proc *p) 320 { 321 uvm_km_free(kernel_map, (vaddr_t)p->p_addr, USPACE); 322 p->p_addr = NULL; 323 } 324 325 /* 326 * uvm_exit: exit a virtual address space 327 */ 328 void 329 uvm_exit(struct process *pr) 330 { 331 uvmspace_free(pr->ps_vmspace); 332 pr->ps_vmspace = NULL; 333 } 334 335 /* 336 * uvm_init_limit: init per-process VM limits 337 * 338 * - called for process 0 and then inherited by all others. 339 */ 340 void 341 uvm_init_limits(struct proc *p) 342 { 343 344 /* 345 * Set up the initial limits on process VM. Set the maximum 346 * resident set size to be all of (reasonably) available memory. 347 * This causes any single, large process to start random page 348 * replacement once it fills memory. 349 */ 350 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 351 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 352 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 353 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 354 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free); 355 } 356 357 #ifdef DEBUG 358 int enableswap = 1; 359 int swapdebug = 0; 360 #define SDB_FOLLOW 1 361 #define SDB_SWAPIN 2 362 #define SDB_SWAPOUT 4 363 #endif 364 365 366 /* 367 * swapout_threads: find threads that can be swapped 368 * 369 * - called by the pagedaemon 370 * - try and swap at least one processs 371 * - processes that are sleeping or stopped for maxslp or more seconds 372 * are swapped... otherwise the longest-sleeping or stopped process 373 * is swapped, otherwise the longest resident process... 374 */ 375 void 376 uvm_swapout_threads(void) 377 { 378 struct process *pr; 379 struct proc *p, *slpp; 380 struct process *outpr; 381 int outpri; 382 int didswap = 0; 383 extern int maxslp; 384 /* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */ 385 386 #ifdef DEBUG 387 if (!enableswap) 388 return; 389 #endif 390 391 /* 392 * outpr/outpri : stop/sleep process whose most active thread has 393 * the largest sleeptime < maxslp 394 */ 395 outpr = NULL; 396 outpri = 0; 397 LIST_FOREACH(pr, &allprocess, ps_list) { 398 if (pr->ps_flags & (PS_SYSTEM | PS_EXITING)) 399 continue; 400 401 /* 402 * slpp: the sleeping or stopped thread in pr with 403 * the smallest p_slptime 404 */ 405 slpp = NULL; 406 TAILQ_FOREACH(p, &pr->ps_threads, p_thr_link) { 407 switch (p->p_stat) { 408 case SRUN: 409 case SONPROC: 410 goto next_process; 411 412 case SSLEEP: 413 case SSTOP: 414 if (slpp == NULL || 415 slpp->p_slptime < p->p_slptime) 416 slpp = p; 417 continue; 418 } 419 } 420 421 if (slpp != NULL) { 422 if (slpp->p_slptime >= maxslp) { 423 pmap_collect(pr->ps_vmspace->vm_map.pmap); 424 didswap++; 425 } else if (slpp->p_slptime > outpri) { 426 outpr = pr; 427 outpri = slpp->p_slptime; 428 } 429 } 430 next_process: ; 431 } 432 433 /* 434 * If we didn't get rid of any real duds, toss out the next most 435 * likely sleeping/stopped or running candidate. We only do this 436 * if we are real low on memory since we don't gain much by doing 437 * it. 438 */ 439 if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE)) && 440 outpr != NULL) { 441 #ifdef DEBUG 442 if (swapdebug & SDB_SWAPOUT) 443 printf("swapout_threads: no duds, try procpr %p\n", 444 outpr); 445 #endif 446 pmap_collect(outpr->ps_vmspace->vm_map.pmap); 447 } 448 } 449 450 /* 451 * uvm_atopg: convert KVAs back to their page structures. 452 */ 453 struct vm_page * 454 uvm_atopg(vaddr_t kva) 455 { 456 struct vm_page *pg; 457 paddr_t pa; 458 boolean_t rv; 459 460 rv = pmap_extract(pmap_kernel(), kva, &pa); 461 KASSERT(rv); 462 pg = PHYS_TO_VM_PAGE(pa); 463 KASSERT(pg != NULL); 464 return (pg); 465 } 466 467 void 468 uvm_pause(void) 469 { 470 static unsigned int toggle; 471 if (toggle++ > 128) { 472 toggle = 0; 473 KERNEL_UNLOCK(); 474 KERNEL_LOCK(); 475 } 476 if (curcpu()->ci_schedstate.spc_schedflags & SPCF_SHOULDYIELD) 477 preempt(NULL); 478 } 479 480 #ifndef SMALL_KERNEL 481 int 482 fill_vmmap(struct process *pr, struct kinfo_vmentry *kve, 483 size_t *lenp) 484 { 485 struct vm_map *map; 486 487 if (pr != NULL) 488 map = &pr->ps_vmspace->vm_map; 489 else 490 map = kernel_map; 491 return uvm_map_fill_vmmap(map, kve, lenp); 492 } 493 #endif 494