1 /*- 2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) 3 * 4 * Copyright (c) 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * The Mach Operating System project at Carnegie-Mellon University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94 35 * 36 * 37 * Copyright (c) 1987, 1990 Carnegie-Mellon University. 38 * All rights reserved. 39 * 40 * Permission to use, copy, modify and distribute this software and 41 * its documentation is hereby granted, provided that both the copyright 42 * notice and this permission notice appear in all copies of the 43 * software, derivative works or modified versions, and any portions 44 * thereof, and that both notices appear in supporting documentation. 45 * 46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" 47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND 48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. 49 * 50 * Carnegie Mellon requests users of this software to return to 51 * 52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU 53 * School of Computer Science 54 * Carnegie Mellon University 55 * Pittsburgh PA 15213-3890 56 * 57 * any improvements or extensions that they make and grant Carnegie the 58 * rights to redistribute these changes. 59 */ 60 61 #include <sys/cdefs.h> 62 __FBSDID("$FreeBSD$"); 63 64 #include "opt_vm.h" 65 #include "opt_kstack_pages.h" 66 #include "opt_kstack_max_pages.h" 67 #include "opt_kstack_usage_prof.h" 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/domainset.h> 72 #include <sys/limits.h> 73 #include <sys/lock.h> 74 #include <sys/malloc.h> 75 #include <sys/mutex.h> 76 #include <sys/proc.h> 77 #include <sys/racct.h> 78 #include <sys/resourcevar.h> 79 #include <sys/rwlock.h> 80 #include <sys/sched.h> 81 #include <sys/sf_buf.h> 82 #include <sys/shm.h> 83 #include <sys/smp.h> 84 #include <sys/vmmeter.h> 85 #include <sys/vmem.h> 86 #include <sys/sx.h> 87 #include <sys/sysctl.h> 88 #include <sys/eventhandler.h> 89 #include <sys/kernel.h> 90 #include <sys/ktr.h> 91 #include <sys/unistd.h> 92 93 #include <vm/uma.h> 94 #include <vm/vm.h> 95 #include <vm/vm_param.h> 96 #include <vm/pmap.h> 97 #include <vm/vm_domainset.h> 98 #include <vm/vm_map.h> 99 #include <vm/vm_page.h> 100 #include <vm/vm_pageout.h> 101 #include <vm/vm_object.h> 102 #include <vm/vm_kern.h> 103 #include <vm/vm_extern.h> 104 #include <vm/vm_pager.h> 105 #include <vm/swap_pager.h> 106 107 #include <machine/cpu.h> 108 109 /* 110 * MPSAFE 111 * 112 * WARNING! This code calls vm_map_check_protection() which only checks 113 * the associated vm_map_entry range. It does not determine whether the 114 * contents of the memory is actually readable or writable. In most cases 115 * just checking the vm_map_entry is sufficient within the kernel's address 116 * space. 117 */ 118 int 119 kernacc(void *addr, int len, int rw) 120 { 121 boolean_t rv; 122 vm_offset_t saddr, eaddr; 123 vm_prot_t prot; 124 125 KASSERT((rw & ~VM_PROT_ALL) == 0, 126 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 127 128 if ((vm_offset_t)addr + len > vm_map_max(kernel_map) || 129 (vm_offset_t)addr + len < (vm_offset_t)addr) 130 return (FALSE); 131 132 prot = rw; 133 saddr = trunc_page((vm_offset_t)addr); 134 eaddr = round_page((vm_offset_t)addr + len); 135 vm_map_lock_read(kernel_map); 136 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 137 vm_map_unlock_read(kernel_map); 138 return (rv == TRUE); 139 } 140 141 /* 142 * MPSAFE 143 * 144 * WARNING! This code calls vm_map_check_protection() which only checks 145 * the associated vm_map_entry range. It does not determine whether the 146 * contents of the memory is actually readable or writable. vmapbuf(), 147 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be 148 * used in conjunction with this call. 149 */ 150 int 151 useracc(void *addr, int len, int rw) 152 { 153 boolean_t rv; 154 vm_prot_t prot; 155 vm_map_t map; 156 157 KASSERT((rw & ~VM_PROT_ALL) == 0, 158 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 159 prot = rw; 160 map = &curproc->p_vmspace->vm_map; 161 if ((vm_offset_t)addr + len > vm_map_max(map) || 162 (vm_offset_t)addr + len < (vm_offset_t)addr) { 163 return (FALSE); 164 } 165 vm_map_lock_read(map); 166 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 167 round_page((vm_offset_t)addr + len), prot); 168 vm_map_unlock_read(map); 169 return (rv == TRUE); 170 } 171 172 int 173 vslock(void *addr, size_t len) 174 { 175 vm_offset_t end, last, start; 176 vm_size_t npages; 177 int error; 178 179 last = (vm_offset_t)addr + len; 180 start = trunc_page((vm_offset_t)addr); 181 end = round_page(last); 182 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) 183 return (EINVAL); 184 npages = atop(end - start); 185 if (npages > vm_page_max_user_wired) 186 return (ENOMEM); 187 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, 188 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 189 if (error == KERN_SUCCESS) { 190 curthread->td_vslock_sz += len; 191 return (0); 192 } 193 194 /* 195 * Return EFAULT on error to match copy{in,out}() behaviour 196 * rather than returning ENOMEM like mlock() would. 197 */ 198 return (EFAULT); 199 } 200 201 void 202 vsunlock(void *addr, size_t len) 203 { 204 205 /* Rely on the parameter sanity checks performed by vslock(). */ 206 MPASS(curthread->td_vslock_sz >= len); 207 curthread->td_vslock_sz -= len; 208 (void)vm_map_unwire(&curproc->p_vmspace->vm_map, 209 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), 210 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 211 } 212 213 /* 214 * Pin the page contained within the given object at the given offset. If the 215 * page is not resident, allocate and load it using the given object's pager. 216 * Return the pinned page if successful; otherwise, return NULL. 217 */ 218 static vm_page_t 219 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset) 220 { 221 vm_page_t m; 222 vm_pindex_t pindex; 223 224 pindex = OFF_TO_IDX(offset); 225 VM_OBJECT_WLOCK(object); 226 (void)vm_page_grab_valid(&m, object, pindex, 227 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 228 VM_OBJECT_WUNLOCK(object); 229 return (m); 230 } 231 232 /* 233 * Return a CPU private mapping to the page at the given offset within the 234 * given object. The page is pinned before it is mapped. 235 */ 236 struct sf_buf * 237 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset) 238 { 239 vm_page_t m; 240 241 m = vm_imgact_hold_page(object, offset); 242 if (m == NULL) 243 return (NULL); 244 sched_pin(); 245 return (sf_buf_alloc(m, SFB_CPUPRIVATE)); 246 } 247 248 /* 249 * Destroy the given CPU private mapping and unpin the page that it mapped. 250 */ 251 void 252 vm_imgact_unmap_page(struct sf_buf *sf) 253 { 254 vm_page_t m; 255 256 m = sf_buf_page(sf); 257 sf_buf_free(sf); 258 sched_unpin(); 259 vm_page_unwire(m, PQ_ACTIVE); 260 } 261 262 void 263 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz) 264 { 265 266 pmap_sync_icache(map->pmap, va, sz); 267 } 268 269 static uma_zone_t kstack_cache; 270 static int kstack_cache_size; 271 static int kstack_domain_iter; 272 273 static int 274 sysctl_kstack_cache_size(SYSCTL_HANDLER_ARGS) 275 { 276 int error, oldsize; 277 278 oldsize = kstack_cache_size; 279 error = sysctl_handle_int(oidp, arg1, arg2, req); 280 if (error == 0 && req->newptr && oldsize != kstack_cache_size) 281 uma_zone_set_maxcache(kstack_cache, kstack_cache_size); 282 return (error); 283 } 284 SYSCTL_PROC(_vm, OID_AUTO, kstack_cache_size, 285 CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &kstack_cache_size, 0, 286 sysctl_kstack_cache_size, "IU", "Maximum number of cached kernel stacks"); 287 288 /* 289 * Create the kernel stack (including pcb for i386) for a new thread. 290 * This routine directly affects the fork perf for a process and 291 * create performance for a thread. 292 */ 293 static vm_offset_t 294 vm_thread_stack_create(struct domainset *ds, vm_object_t *ksobjp, int pages) 295 { 296 vm_page_t ma[KSTACK_MAX_PAGES]; 297 vm_object_t ksobj; 298 vm_offset_t ks; 299 int i; 300 301 /* 302 * Allocate an object for the kstack. 303 */ 304 ksobj = vm_object_allocate(OBJT_DEFAULT, pages); 305 306 /* 307 * Get a kernel virtual address for this thread's kstack. 308 */ 309 #if defined(__mips__) 310 /* 311 * We need to align the kstack's mapped address to fit within 312 * a single TLB entry. 313 */ 314 if (vmem_xalloc(kernel_arena, (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE, 315 PAGE_SIZE * 2, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX, 316 M_BESTFIT | M_NOWAIT, &ks)) { 317 ks = 0; 318 } 319 #else 320 ks = kva_alloc((pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 321 #endif 322 if (ks == 0) { 323 printf("%s: kstack allocation failed\n", __func__); 324 vm_object_deallocate(ksobj); 325 return (0); 326 } 327 if (vm_ndomains > 1) { 328 ksobj->domain.dr_policy = ds; 329 ksobj->domain.dr_iter = 330 atomic_fetchadd_int(&kstack_domain_iter, 1); 331 } 332 333 if (KSTACK_GUARD_PAGES != 0) { 334 pmap_qremove(ks, KSTACK_GUARD_PAGES); 335 ks += KSTACK_GUARD_PAGES * PAGE_SIZE; 336 } 337 338 /* 339 * For the length of the stack, link in a real page of ram for each 340 * page of stack. 341 */ 342 VM_OBJECT_WLOCK(ksobj); 343 (void)vm_page_grab_pages(ksobj, 0, VM_ALLOC_NORMAL | VM_ALLOC_WIRED, 344 ma, pages); 345 for (i = 0; i < pages; i++) { 346 vm_page_valid(ma[i]); 347 vm_page_xunbusy(ma[i]); 348 } 349 VM_OBJECT_WUNLOCK(ksobj); 350 pmap_qenter(ks, ma, pages); 351 *ksobjp = ksobj; 352 353 return (ks); 354 } 355 356 static void 357 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages) 358 { 359 vm_page_t m; 360 int i; 361 362 pmap_qremove(ks, pages); 363 VM_OBJECT_WLOCK(ksobj); 364 for (i = 0; i < pages; i++) { 365 m = vm_page_lookup(ksobj, i); 366 if (m == NULL) 367 panic("%s: kstack already missing?", __func__); 368 vm_page_busy_acquire(m, 0); 369 vm_page_unwire_noq(m); 370 vm_page_free(m); 371 } 372 VM_OBJECT_WUNLOCK(ksobj); 373 vm_object_deallocate(ksobj); 374 kva_free(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), 375 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 376 } 377 378 /* 379 * Allocate the kernel stack for a new thread. 380 */ 381 int 382 vm_thread_new(struct thread *td, int pages) 383 { 384 vm_object_t ksobj; 385 vm_offset_t ks; 386 387 /* Bounds check */ 388 if (pages <= 1) 389 pages = kstack_pages; 390 else if (pages > KSTACK_MAX_PAGES) 391 pages = KSTACK_MAX_PAGES; 392 393 ks = 0; 394 ksobj = NULL; 395 if (pages == kstack_pages && kstack_cache != NULL) { 396 ks = (vm_offset_t)uma_zalloc(kstack_cache, M_NOWAIT); 397 if (ks != 0) 398 ksobj = PHYS_TO_VM_PAGE(pmap_kextract(ks))->object; 399 } 400 401 /* 402 * Ensure that kstack objects can draw pages from any memory 403 * domain. Otherwise a local memory shortage can block a process 404 * swap-in. 405 */ 406 if (ks == 0) 407 ks = vm_thread_stack_create(DOMAINSET_PREF(PCPU_GET(domain)), 408 &ksobj, pages); 409 if (ks == 0) 410 return (0); 411 td->td_kstack_obj = ksobj; 412 td->td_kstack = ks; 413 td->td_kstack_pages = pages; 414 return (1); 415 } 416 417 /* 418 * Dispose of a thread's kernel stack. 419 */ 420 void 421 vm_thread_dispose(struct thread *td) 422 { 423 vm_object_t ksobj; 424 vm_offset_t ks; 425 int pages; 426 427 pages = td->td_kstack_pages; 428 ksobj = td->td_kstack_obj; 429 ks = td->td_kstack; 430 td->td_kstack = 0; 431 td->td_kstack_pages = 0; 432 if (pages == kstack_pages) 433 uma_zfree(kstack_cache, (void *)ks); 434 else 435 vm_thread_stack_dispose(ksobj, ks, pages); 436 } 437 438 static int 439 kstack_import(void *arg, void **store, int cnt, int domain, int flags) 440 { 441 struct domainset *ds; 442 vm_object_t ksobj; 443 int i; 444 445 if (domain == UMA_ANYDOMAIN) 446 ds = DOMAINSET_RR(); 447 else 448 ds = DOMAINSET_PREF(domain); 449 450 for (i = 0; i < cnt; i++) { 451 store[i] = (void *)vm_thread_stack_create(ds, &ksobj, 452 kstack_pages); 453 if (store[i] == NULL) 454 break; 455 } 456 return (i); 457 } 458 459 static void 460 kstack_release(void *arg, void **store, int cnt) 461 { 462 vm_offset_t ks; 463 int i; 464 465 for (i = 0; i < cnt; i++) { 466 ks = (vm_offset_t)store[i]; 467 vm_thread_stack_dispose( 468 PHYS_TO_VM_PAGE(pmap_kextract(ks))->object, 469 ks, kstack_pages); 470 } 471 } 472 473 static void 474 kstack_cache_init(void *null) 475 { 476 kstack_cache = uma_zcache_create("kstack_cache", 477 kstack_pages * PAGE_SIZE, NULL, NULL, NULL, NULL, 478 kstack_import, kstack_release, NULL, 479 UMA_ZONE_FIRSTTOUCH); 480 kstack_cache_size = imax(128, mp_ncpus * 4); 481 uma_zone_set_maxcache(kstack_cache, kstack_cache_size); 482 } 483 484 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL); 485 486 #ifdef KSTACK_USAGE_PROF 487 /* 488 * Track maximum stack used by a thread in kernel. 489 */ 490 static int max_kstack_used; 491 492 SYSCTL_INT(_debug, OID_AUTO, max_kstack_used, CTLFLAG_RD, 493 &max_kstack_used, 0, 494 "Maxiumum stack depth used by a thread in kernel"); 495 496 void 497 intr_prof_stack_use(struct thread *td, struct trapframe *frame) 498 { 499 vm_offset_t stack_top; 500 vm_offset_t current; 501 int used, prev_used; 502 503 /* 504 * Testing for interrupted kernel mode isn't strictly 505 * needed. It optimizes the execution, since interrupts from 506 * usermode will have only the trap frame on the stack. 507 */ 508 if (TRAPF_USERMODE(frame)) 509 return; 510 511 stack_top = td->td_kstack + td->td_kstack_pages * PAGE_SIZE; 512 current = (vm_offset_t)(uintptr_t)&stack_top; 513 514 /* 515 * Try to detect if interrupt is using kernel thread stack. 516 * Hardware could use a dedicated stack for interrupt handling. 517 */ 518 if (stack_top <= current || current < td->td_kstack) 519 return; 520 521 used = stack_top - current; 522 for (;;) { 523 prev_used = max_kstack_used; 524 if (prev_used >= used) 525 break; 526 if (atomic_cmpset_int(&max_kstack_used, prev_used, used)) 527 break; 528 } 529 } 530 #endif /* KSTACK_USAGE_PROF */ 531 532 /* 533 * Implement fork's actions on an address space. 534 * Here we arrange for the address space to be copied or referenced, 535 * allocate a user struct (pcb and kernel stack), then call the 536 * machine-dependent layer to fill those in and make the new process 537 * ready to run. The new process is set up so that it returns directly 538 * to user mode to avoid stack copying and relocation problems. 539 */ 540 int 541 vm_forkproc(struct thread *td, struct proc *p2, struct thread *td2, 542 struct vmspace *vm2, int flags) 543 { 544 struct proc *p1 = td->td_proc; 545 struct domainset *dset; 546 int error; 547 548 if ((flags & RFPROC) == 0) { 549 /* 550 * Divorce the memory, if it is shared, essentially 551 * this changes shared memory amongst threads, into 552 * COW locally. 553 */ 554 if ((flags & RFMEM) == 0) { 555 if (p1->p_vmspace->vm_refcnt > 1) { 556 error = vmspace_unshare(p1); 557 if (error) 558 return (error); 559 } 560 } 561 cpu_fork(td, p2, td2, flags); 562 return (0); 563 } 564 565 if (flags & RFMEM) { 566 p2->p_vmspace = p1->p_vmspace; 567 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1); 568 } 569 dset = td2->td_domain.dr_policy; 570 while (vm_page_count_severe_set(&dset->ds_mask)) { 571 vm_wait_doms(&dset->ds_mask); 572 } 573 574 if ((flags & RFMEM) == 0) { 575 p2->p_vmspace = vm2; 576 if (p1->p_vmspace->vm_shm) 577 shmfork(p1, p2); 578 } 579 580 /* 581 * cpu_fork will copy and update the pcb, set up the kernel stack, 582 * and make the child ready to run. 583 */ 584 cpu_fork(td, p2, td2, flags); 585 return (0); 586 } 587 588 /* 589 * Called after process has been wait(2)'ed upon and is being reaped. 590 * The idea is to reclaim resources that we could not reclaim while 591 * the process was still executing. 592 */ 593 void 594 vm_waitproc(p) 595 struct proc *p; 596 { 597 598 vmspace_exitfree(p); /* and clean-out the vmspace */ 599 } 600 601 void 602 kick_proc0(void) 603 { 604 605 wakeup(&proc0); 606 } 607