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/vmmeter.h> 84 #include <sys/vmem.h> 85 #include <sys/sx.h> 86 #include <sys/sysctl.h> 87 #include <sys/eventhandler.h> 88 #include <sys/kernel.h> 89 #include <sys/ktr.h> 90 #include <sys/unistd.h> 91 92 #include <vm/uma.h> 93 #include <vm/vm.h> 94 #include <vm/vm_param.h> 95 #include <vm/pmap.h> 96 #include <vm/vm_domainset.h> 97 #include <vm/vm_map.h> 98 #include <vm/vm_page.h> 99 #include <vm/vm_pageout.h> 100 #include <vm/vm_object.h> 101 #include <vm/vm_kern.h> 102 #include <vm/vm_extern.h> 103 #include <vm/vm_pager.h> 104 #include <vm/swap_pager.h> 105 106 #include <machine/cpu.h> 107 108 /* 109 * MPSAFE 110 * 111 * WARNING! This code calls vm_map_check_protection() which only checks 112 * the associated vm_map_entry range. It does not determine whether the 113 * contents of the memory is actually readable or writable. In most cases 114 * just checking the vm_map_entry is sufficient within the kernel's address 115 * space. 116 */ 117 int 118 kernacc(void *addr, int len, int rw) 119 { 120 boolean_t rv; 121 vm_offset_t saddr, eaddr; 122 vm_prot_t prot; 123 124 KASSERT((rw & ~VM_PROT_ALL) == 0, 125 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 126 127 if ((vm_offset_t)addr + len > vm_map_max(kernel_map) || 128 (vm_offset_t)addr + len < (vm_offset_t)addr) 129 return (FALSE); 130 131 prot = rw; 132 saddr = trunc_page((vm_offset_t)addr); 133 eaddr = round_page((vm_offset_t)addr + len); 134 vm_map_lock_read(kernel_map); 135 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 136 vm_map_unlock_read(kernel_map); 137 return (rv == TRUE); 138 } 139 140 /* 141 * MPSAFE 142 * 143 * WARNING! This code calls vm_map_check_protection() which only checks 144 * the associated vm_map_entry range. It does not determine whether the 145 * contents of the memory is actually readable or writable. vmapbuf(), 146 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be 147 * used in conjunction with this call. 148 */ 149 int 150 useracc(void *addr, int len, int rw) 151 { 152 boolean_t rv; 153 vm_prot_t prot; 154 vm_map_t map; 155 156 KASSERT((rw & ~VM_PROT_ALL) == 0, 157 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 158 prot = rw; 159 map = &curproc->p_vmspace->vm_map; 160 if ((vm_offset_t)addr + len > vm_map_max(map) || 161 (vm_offset_t)addr + len < (vm_offset_t)addr) { 162 return (FALSE); 163 } 164 vm_map_lock_read(map); 165 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 166 round_page((vm_offset_t)addr + len), prot); 167 vm_map_unlock_read(map); 168 return (rv == TRUE); 169 } 170 171 int 172 vslock(void *addr, size_t len) 173 { 174 vm_offset_t end, last, start; 175 vm_size_t npages; 176 int error; 177 178 last = (vm_offset_t)addr + len; 179 start = trunc_page((vm_offset_t)addr); 180 end = round_page(last); 181 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) 182 return (EINVAL); 183 npages = atop(end - start); 184 if (npages > vm_page_max_user_wired) 185 return (ENOMEM); 186 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, 187 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 188 if (error == KERN_SUCCESS) { 189 curthread->td_vslock_sz += len; 190 return (0); 191 } 192 193 /* 194 * Return EFAULT on error to match copy{in,out}() behaviour 195 * rather than returning ENOMEM like mlock() would. 196 */ 197 return (EFAULT); 198 } 199 200 void 201 vsunlock(void *addr, size_t len) 202 { 203 204 /* Rely on the parameter sanity checks performed by vslock(). */ 205 MPASS(curthread->td_vslock_sz >= len); 206 curthread->td_vslock_sz -= len; 207 (void)vm_map_unwire(&curproc->p_vmspace->vm_map, 208 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), 209 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 210 } 211 212 /* 213 * Pin the page contained within the given object at the given offset. If the 214 * page is not resident, allocate and load it using the given object's pager. 215 * Return the pinned page if successful; otherwise, return NULL. 216 */ 217 static vm_page_t 218 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset) 219 { 220 vm_page_t m; 221 vm_pindex_t pindex; 222 int rv; 223 224 VM_OBJECT_WLOCK(object); 225 pindex = OFF_TO_IDX(offset); 226 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | 227 VM_ALLOC_WIRED); 228 if (m->valid != VM_PAGE_BITS_ALL) { 229 vm_page_xbusy(m); 230 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL); 231 if (rv != VM_PAGER_OK) { 232 vm_page_unwire_noq(m); 233 vm_page_free(m); 234 m = NULL; 235 goto out; 236 } 237 vm_page_xunbusy(m); 238 } 239 out: 240 VM_OBJECT_WUNLOCK(object); 241 return (m); 242 } 243 244 /* 245 * Return a CPU private mapping to the page at the given offset within the 246 * given object. The page is pinned before it is mapped. 247 */ 248 struct sf_buf * 249 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset) 250 { 251 vm_page_t m; 252 253 m = vm_imgact_hold_page(object, offset); 254 if (m == NULL) 255 return (NULL); 256 sched_pin(); 257 return (sf_buf_alloc(m, SFB_CPUPRIVATE)); 258 } 259 260 /* 261 * Destroy the given CPU private mapping and unpin the page that it mapped. 262 */ 263 void 264 vm_imgact_unmap_page(struct sf_buf *sf) 265 { 266 vm_page_t m; 267 268 m = sf_buf_page(sf); 269 sf_buf_free(sf); 270 sched_unpin(); 271 vm_page_unwire(m, PQ_ACTIVE); 272 } 273 274 void 275 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz) 276 { 277 278 pmap_sync_icache(map->pmap, va, sz); 279 } 280 281 static uma_zone_t kstack_cache; 282 static int kstack_cache_size = 128; 283 static int kstack_domain_iter; 284 285 static int 286 sysctl_kstack_cache_size(SYSCTL_HANDLER_ARGS) 287 { 288 int error, newsize; 289 290 newsize = kstack_cache_size; 291 error = sysctl_handle_int(oidp, &newsize, 0, req); 292 if (error == 0 && req->newptr && newsize != kstack_cache_size) 293 kstack_cache_size = 294 uma_zone_set_maxcache(kstack_cache, newsize); 295 return (error); 296 } 297 SYSCTL_PROC(_vm, OID_AUTO, kstack_cache_size, CTLTYPE_INT|CTLFLAG_RW, 298 &kstack_cache_size, 0, sysctl_kstack_cache_size, "IU", 299 "Maximum number of cached kernel stacks"); 300 301 /* 302 * Create the kernel stack (including pcb for i386) for a new thread. 303 * This routine directly affects the fork perf for a process and 304 * create performance for a thread. 305 */ 306 static vm_offset_t 307 vm_thread_stack_create(struct domainset *ds, vm_object_t *ksobjp, int pages) 308 { 309 vm_page_t ma[KSTACK_MAX_PAGES]; 310 vm_object_t ksobj; 311 vm_offset_t ks; 312 int i; 313 314 /* 315 * Allocate an object for the kstack. 316 */ 317 ksobj = vm_object_allocate(OBJT_DEFAULT, pages); 318 319 /* 320 * Get a kernel virtual address for this thread's kstack. 321 */ 322 #if defined(__mips__) 323 /* 324 * We need to align the kstack's mapped address to fit within 325 * a single TLB entry. 326 */ 327 if (vmem_xalloc(kernel_arena, (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE, 328 PAGE_SIZE * 2, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX, 329 M_BESTFIT | M_NOWAIT, &ks)) { 330 ks = 0; 331 } 332 #else 333 ks = kva_alloc((pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 334 #endif 335 if (ks == 0) { 336 printf("vm_thread_new: kstack allocation failed\n"); 337 vm_object_deallocate(ksobj); 338 return (0); 339 } 340 if (vm_ndomains > 1) { 341 ksobj->domain.dr_policy = ds; 342 ksobj->domain.dr_iter = 343 atomic_fetchadd_int(&kstack_domain_iter, 1); 344 } 345 346 if (KSTACK_GUARD_PAGES != 0) { 347 pmap_qremove(ks, KSTACK_GUARD_PAGES); 348 ks += KSTACK_GUARD_PAGES * PAGE_SIZE; 349 } 350 351 /* 352 * For the length of the stack, link in a real page of ram for each 353 * page of stack. 354 */ 355 VM_OBJECT_WLOCK(ksobj); 356 (void)vm_page_grab_pages(ksobj, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | 357 VM_ALLOC_WIRED, ma, pages); 358 for (i = 0; i < pages; i++) 359 ma[i]->valid = VM_PAGE_BITS_ALL; 360 VM_OBJECT_WUNLOCK(ksobj); 361 pmap_qenter(ks, ma, pages); 362 *ksobjp = ksobj; 363 364 return (ks); 365 } 366 367 static void 368 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages) 369 { 370 vm_page_t m; 371 int i; 372 373 pmap_qremove(ks, pages); 374 VM_OBJECT_WLOCK(ksobj); 375 for (i = 0; i < pages; i++) { 376 m = vm_page_lookup(ksobj, i); 377 if (m == NULL) 378 panic("vm_thread_dispose: kstack already missing?"); 379 vm_page_unwire_noq(m); 380 vm_page_free(m); 381 } 382 VM_OBJECT_WUNLOCK(ksobj); 383 vm_object_deallocate(ksobj); 384 kva_free(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), 385 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 386 } 387 388 /* 389 * Allocate the kernel stack for a new thread. 390 */ 391 int 392 vm_thread_new(struct thread *td, int pages) 393 { 394 vm_object_t ksobj; 395 vm_offset_t ks; 396 397 /* Bounds check */ 398 if (pages <= 1) 399 pages = kstack_pages; 400 else if (pages > KSTACK_MAX_PAGES) 401 pages = KSTACK_MAX_PAGES; 402 403 ks = 0; 404 ksobj = NULL; 405 if (pages == kstack_pages && kstack_cache != NULL) { 406 ks = (vm_offset_t)uma_zalloc(kstack_cache, M_NOWAIT); 407 if (ks != 0) 408 ksobj = PHYS_TO_VM_PAGE(pmap_kextract(ks))->object; 409 } 410 411 /* 412 * Ensure that kstack objects can draw pages from any memory 413 * domain. Otherwise a local memory shortage can block a process 414 * swap-in. 415 */ 416 if (ks == 0) 417 ks = vm_thread_stack_create(DOMAINSET_PREF(PCPU_GET(domain)), 418 &ksobj, pages); 419 if (ks == 0) 420 return (0); 421 td->td_kstack_obj = ksobj; 422 td->td_kstack = ks; 423 td->td_kstack_pages = pages; 424 return (1); 425 } 426 427 /* 428 * Dispose of a thread's kernel stack. 429 */ 430 void 431 vm_thread_dispose(struct thread *td) 432 { 433 vm_object_t ksobj; 434 vm_offset_t ks; 435 int pages; 436 437 pages = td->td_kstack_pages; 438 ksobj = td->td_kstack_obj; 439 ks = td->td_kstack; 440 td->td_kstack = 0; 441 td->td_kstack_pages = 0; 442 if (pages == kstack_pages) 443 uma_zfree(kstack_cache, (void *)ks); 444 else 445 vm_thread_stack_dispose(ksobj, ks, pages); 446 } 447 448 static int 449 kstack_import(void *arg, void **store, int cnt, int domain, int flags) 450 { 451 struct domainset *ds; 452 vm_object_t ksobj; 453 int i; 454 455 if (domain == UMA_ANYDOMAIN) 456 ds = DOMAINSET_RR(); 457 else 458 ds = DOMAINSET_PREF(domain); 459 460 for (i = 0; i < cnt; i++) { 461 store[i] = (void *)vm_thread_stack_create(ds, &ksobj, 462 kstack_pages); 463 if (store[i] == NULL) 464 break; 465 } 466 return (i); 467 } 468 469 static void 470 kstack_release(void *arg, void **store, int cnt) 471 { 472 vm_offset_t ks; 473 int i; 474 475 for (i = 0; i < cnt; i++) { 476 ks = (vm_offset_t)store[i]; 477 vm_thread_stack_dispose( 478 PHYS_TO_VM_PAGE(pmap_kextract(ks))->object, 479 ks, kstack_pages); 480 } 481 } 482 483 static void 484 kstack_cache_init(void *null) 485 { 486 kstack_cache = uma_zcache_create("kstack_cache", 487 kstack_pages * PAGE_SIZE, NULL, NULL, NULL, NULL, 488 kstack_import, kstack_release, NULL, 489 UMA_ZONE_NUMA|UMA_ZONE_MINBUCKET); 490 uma_zone_set_maxcache(kstack_cache, kstack_cache_size); 491 } 492 493 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL); 494 495 #ifdef KSTACK_USAGE_PROF 496 /* 497 * Track maximum stack used by a thread in kernel. 498 */ 499 static int max_kstack_used; 500 501 SYSCTL_INT(_debug, OID_AUTO, max_kstack_used, CTLFLAG_RD, 502 &max_kstack_used, 0, 503 "Maxiumum stack depth used by a thread in kernel"); 504 505 void 506 intr_prof_stack_use(struct thread *td, struct trapframe *frame) 507 { 508 vm_offset_t stack_top; 509 vm_offset_t current; 510 int used, prev_used; 511 512 /* 513 * Testing for interrupted kernel mode isn't strictly 514 * needed. It optimizes the execution, since interrupts from 515 * usermode will have only the trap frame on the stack. 516 */ 517 if (TRAPF_USERMODE(frame)) 518 return; 519 520 stack_top = td->td_kstack + td->td_kstack_pages * PAGE_SIZE; 521 current = (vm_offset_t)(uintptr_t)&stack_top; 522 523 /* 524 * Try to detect if interrupt is using kernel thread stack. 525 * Hardware could use a dedicated stack for interrupt handling. 526 */ 527 if (stack_top <= current || current < td->td_kstack) 528 return; 529 530 used = stack_top - current; 531 for (;;) { 532 prev_used = max_kstack_used; 533 if (prev_used >= used) 534 break; 535 if (atomic_cmpset_int(&max_kstack_used, prev_used, used)) 536 break; 537 } 538 } 539 #endif /* KSTACK_USAGE_PROF */ 540 541 /* 542 * Implement fork's actions on an address space. 543 * Here we arrange for the address space to be copied or referenced, 544 * allocate a user struct (pcb and kernel stack), then call the 545 * machine-dependent layer to fill those in and make the new process 546 * ready to run. The new process is set up so that it returns directly 547 * to user mode to avoid stack copying and relocation problems. 548 */ 549 int 550 vm_forkproc(struct thread *td, struct proc *p2, struct thread *td2, 551 struct vmspace *vm2, int flags) 552 { 553 struct proc *p1 = td->td_proc; 554 struct domainset *dset; 555 int error; 556 557 if ((flags & RFPROC) == 0) { 558 /* 559 * Divorce the memory, if it is shared, essentially 560 * this changes shared memory amongst threads, into 561 * COW locally. 562 */ 563 if ((flags & RFMEM) == 0) { 564 if (p1->p_vmspace->vm_refcnt > 1) { 565 error = vmspace_unshare(p1); 566 if (error) 567 return (error); 568 } 569 } 570 cpu_fork(td, p2, td2, flags); 571 return (0); 572 } 573 574 if (flags & RFMEM) { 575 p2->p_vmspace = p1->p_vmspace; 576 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1); 577 } 578 dset = td2->td_domain.dr_policy; 579 while (vm_page_count_severe_set(&dset->ds_mask)) { 580 vm_wait_doms(&dset->ds_mask); 581 } 582 583 if ((flags & RFMEM) == 0) { 584 p2->p_vmspace = vm2; 585 if (p1->p_vmspace->vm_shm) 586 shmfork(p1, p2); 587 } 588 589 /* 590 * cpu_fork will copy and update the pcb, set up the kernel stack, 591 * and make the child ready to run. 592 */ 593 cpu_fork(td, p2, td2, flags); 594 return (0); 595 } 596 597 /* 598 * Called after process has been wait(2)'ed upon and is being reaped. 599 * The idea is to reclaim resources that we could not reclaim while 600 * the process was still executing. 601 */ 602 void 603 vm_waitproc(p) 604 struct proc *p; 605 { 606 607 vmspace_exitfree(p); /* and clean-out the vmspace */ 608 } 609 610 void 611 kick_proc0(void) 612 { 613 614 wakeup(&proc0); 615 } 616