1 /* $OpenBSD: uvm_glue.c,v 1.80 2021/03/26 13:40:05 mpi 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 /* 104 * uvm_vslock: wire user memory for I/O 105 * 106 * - called from sys_sysctl 107 */ 108 int 109 uvm_vslock(struct proc *p, caddr_t addr, size_t len, vm_prot_t access_type) 110 { 111 struct vm_map *map = &p->p_vmspace->vm_map; 112 vaddr_t start, end; 113 114 start = trunc_page((vaddr_t)addr); 115 end = round_page((vaddr_t)addr + len); 116 if (end <= start) 117 return (EINVAL); 118 119 return uvm_fault_wire(map, start, end, access_type); 120 } 121 122 /* 123 * uvm_vsunlock: unwire user memory wired by uvm_vslock() 124 * 125 * - called from sys_sysctl 126 */ 127 void 128 uvm_vsunlock(struct proc *p, caddr_t addr, size_t len) 129 { 130 vaddr_t start, end; 131 132 start = trunc_page((vaddr_t)addr); 133 end = round_page((vaddr_t)addr + len); 134 KASSERT(end > start); 135 136 uvm_fault_unwire(&p->p_vmspace->vm_map, start, end); 137 } 138 139 /* 140 * uvm_vslock_device: wire user memory, make sure it's device reachable 141 * and bounce if necessary. 142 * 143 * - called from physio 144 */ 145 int 146 uvm_vslock_device(struct proc *p, void *addr, size_t len, 147 vm_prot_t access_type, void **retp) 148 { 149 struct vm_map *map = &p->p_vmspace->vm_map; 150 struct vm_page *pg; 151 struct pglist pgl; 152 int npages; 153 vaddr_t start, end, off; 154 vaddr_t sva, va; 155 vsize_t sz; 156 int error, mapv, i; 157 158 start = trunc_page((vaddr_t)addr); 159 end = round_page((vaddr_t)addr + len); 160 sz = end - start; 161 off = (vaddr_t)addr - start; 162 if (end <= start) 163 return (EINVAL); 164 165 vm_map_lock_read(map); 166 retry: 167 mapv = map->timestamp; 168 vm_map_unlock_read(map); 169 170 if ((error = uvm_fault_wire(map, start, end, access_type))) 171 return (error); 172 173 vm_map_lock_read(map); 174 if (mapv != map->timestamp) 175 goto retry; 176 177 npages = atop(sz); 178 for (i = 0; i < npages; i++) { 179 paddr_t pa; 180 181 if (!pmap_extract(map->pmap, start + ptoa(i), &pa)) { 182 error = EFAULT; 183 goto out_unwire; 184 } 185 if (!PADDR_IS_DMA_REACHABLE(pa)) 186 break; 187 } 188 if (i == npages) { 189 *retp = NULL; 190 return (0); 191 } 192 193 if ((va = uvm_km_valloc(kernel_map, sz)) == 0) { 194 error = ENOMEM; 195 goto out_unwire; 196 } 197 sva = va; 198 199 TAILQ_INIT(&pgl); 200 error = uvm_pglistalloc(npages * PAGE_SIZE, dma_constraint.ucr_low, 201 dma_constraint.ucr_high, 0, 0, &pgl, npages, UVM_PLA_WAITOK); 202 if (error) 203 goto out_unmap; 204 205 while ((pg = TAILQ_FIRST(&pgl)) != NULL) { 206 TAILQ_REMOVE(&pgl, pg, pageq); 207 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), PROT_READ | PROT_WRITE); 208 va += PAGE_SIZE; 209 } 210 pmap_update(pmap_kernel()); 211 KASSERT(va == sva + sz); 212 *retp = (void *)(sva + off); 213 214 if ((error = copyin(addr, *retp, len)) == 0) 215 return 0; 216 217 uvm_km_pgremove_intrsafe(sva, sva + sz); 218 pmap_kremove(sva, sz); 219 pmap_update(pmap_kernel()); 220 out_unmap: 221 uvm_km_free(kernel_map, sva, sz); 222 out_unwire: 223 uvm_fault_unwire_locked(map, start, end); 224 vm_map_unlock_read(map); 225 return (error); 226 } 227 228 /* 229 * uvm_vsunlock_device: unwire user memory wired by uvm_vslock_device() 230 * 231 * - called from physio 232 */ 233 void 234 uvm_vsunlock_device(struct proc *p, void *addr, size_t len, void *map) 235 { 236 vaddr_t start, end; 237 vaddr_t kva; 238 vsize_t sz; 239 240 start = trunc_page((vaddr_t)addr); 241 end = round_page((vaddr_t)addr + len); 242 KASSERT(end > start); 243 sz = end - start; 244 245 if (map) 246 copyout(map, addr, len); 247 248 uvm_fault_unwire_locked(&p->p_vmspace->vm_map, start, end); 249 vm_map_unlock_read(&p->p_vmspace->vm_map); 250 251 if (!map) 252 return; 253 254 kva = trunc_page((vaddr_t)map); 255 uvm_km_pgremove_intrsafe(kva, kva + sz); 256 pmap_kremove(kva, sz); 257 pmap_update(pmap_kernel()); 258 uvm_km_free(kernel_map, kva, sz); 259 } 260 261 /* 262 * uvm_uarea_alloc: allocate the u-area for a new thread 263 */ 264 vaddr_t 265 uvm_uarea_alloc(void) 266 { 267 vaddr_t uaddr; 268 269 uaddr = uvm_km_kmemalloc_pla(kernel_map, uvm.kernel_object, USPACE, 270 USPACE_ALIGN, UVM_KMF_ZERO, 271 no_constraint.ucr_low, no_constraint.ucr_high, 272 0, 0, USPACE/PAGE_SIZE); 273 274 return (uaddr); 275 } 276 277 /* 278 * uvm_uarea_free: free a dead thread's stack 279 * 280 * - the thread passed to us is a dead thread; we 281 * are running on a different context now (the reaper). 282 */ 283 void 284 uvm_uarea_free(struct proc *p) 285 { 286 uvm_km_free(kernel_map, (vaddr_t)p->p_addr, USPACE); 287 p->p_addr = NULL; 288 } 289 290 /* 291 * uvm_exit: exit a virtual address space 292 */ 293 void 294 uvm_exit(struct process *pr) 295 { 296 struct vmspace *vm = pr->ps_vmspace; 297 298 pr->ps_vmspace = NULL; 299 uvmspace_free(vm); 300 } 301 302 /* 303 * uvm_init_limit: init per-process VM limits 304 * 305 * - called for process 0 and then inherited by all others. 306 */ 307 void 308 uvm_init_limits(struct plimit *limit0) 309 { 310 /* 311 * Set up the initial limits on process VM. Set the maximum 312 * resident set size to be all of (reasonably) available memory. 313 * This causes any single, large process to start random page 314 * replacement once it fills memory. 315 */ 316 limit0->pl_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ; 317 limit0->pl_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; 318 limit0->pl_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; 319 limit0->pl_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; 320 limit0->pl_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free); 321 } 322 323 #ifdef DEBUG 324 int enableswap = 1; 325 int swapdebug = 0; 326 #define SDB_FOLLOW 1 327 #define SDB_SWAPIN 2 328 #define SDB_SWAPOUT 4 329 #endif 330 331 332 /* 333 * swapout_threads: find threads that can be swapped 334 * 335 * - called by the pagedaemon 336 * - try and swap at least one process 337 * - processes that are sleeping or stopped for maxslp or more seconds 338 * are swapped... otherwise the longest-sleeping or stopped process 339 * is swapped, otherwise the longest resident process... 340 */ 341 void 342 uvm_swapout_threads(void) 343 { 344 struct process *pr; 345 struct proc *p, *slpp; 346 struct process *outpr; 347 int outpri; 348 int didswap = 0; 349 extern int maxslp; 350 /* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */ 351 352 #ifdef DEBUG 353 if (!enableswap) 354 return; 355 #endif 356 357 /* 358 * outpr/outpri : stop/sleep process whose most active thread has 359 * the largest sleeptime < maxslp 360 */ 361 outpr = NULL; 362 outpri = 0; 363 LIST_FOREACH(pr, &allprocess, ps_list) { 364 if (pr->ps_flags & (PS_SYSTEM | PS_EXITING)) 365 continue; 366 367 /* 368 * slpp: the sleeping or stopped thread in pr with 369 * the smallest p_slptime 370 */ 371 slpp = NULL; 372 TAILQ_FOREACH(p, &pr->ps_threads, p_thr_link) { 373 switch (p->p_stat) { 374 case SRUN: 375 case SONPROC: 376 goto next_process; 377 378 case SSLEEP: 379 case SSTOP: 380 if (slpp == NULL || 381 slpp->p_slptime < p->p_slptime) 382 slpp = p; 383 continue; 384 } 385 } 386 387 if (slpp != NULL) { 388 if (slpp->p_slptime >= maxslp) { 389 pmap_collect(pr->ps_vmspace->vm_map.pmap); 390 didswap++; 391 } else if (slpp->p_slptime > outpri) { 392 outpr = pr; 393 outpri = slpp->p_slptime; 394 } 395 } 396 next_process: ; 397 } 398 399 /* 400 * If we didn't get rid of any real duds, toss out the next most 401 * likely sleeping/stopped or running candidate. We only do this 402 * if we are real low on memory since we don't gain much by doing 403 * it. 404 */ 405 if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE)) && 406 outpr != NULL) { 407 #ifdef DEBUG 408 if (swapdebug & SDB_SWAPOUT) 409 printf("swapout_threads: no duds, try procpr %p\n", 410 outpr); 411 #endif 412 pmap_collect(outpr->ps_vmspace->vm_map.pmap); 413 } 414 } 415 416 /* 417 * uvm_atopg: convert KVAs back to their page structures. 418 */ 419 struct vm_page * 420 uvm_atopg(vaddr_t kva) 421 { 422 struct vm_page *pg; 423 paddr_t pa; 424 boolean_t rv; 425 426 rv = pmap_extract(pmap_kernel(), kva, &pa); 427 KASSERT(rv); 428 pg = PHYS_TO_VM_PAGE(pa); 429 KASSERT(pg != NULL); 430 return (pg); 431 } 432 433 void 434 uvm_pause(void) 435 { 436 static unsigned int toggle; 437 if (toggle++ > 128) { 438 toggle = 0; 439 KERNEL_UNLOCK(); 440 KERNEL_LOCK(); 441 } 442 sched_pause(preempt); 443 } 444 445 #ifndef SMALL_KERNEL 446 int 447 fill_vmmap(struct process *pr, struct kinfo_vmentry *kve, 448 size_t *lenp) 449 { 450 struct vm_map *map; 451 452 if (pr != NULL) 453 map = &pr->ps_vmspace->vm_map; 454 else 455 map = kernel_map; 456 return uvm_map_fill_vmmap(map, kve, lenp); 457 } 458 #endif 459