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/asan.h> 72 #include <sys/domainset.h> 73 #include <sys/limits.h> 74 #include <sys/lock.h> 75 #include <sys/malloc.h> 76 #include <sys/msan.h> 77 #include <sys/mutex.h> 78 #include <sys/proc.h> 79 #include <sys/racct.h> 80 #include <sys/refcount.h> 81 #include <sys/resourcevar.h> 82 #include <sys/rwlock.h> 83 #include <sys/sched.h> 84 #include <sys/sf_buf.h> 85 #include <sys/shm.h> 86 #include <sys/smp.h> 87 #include <sys/vmmeter.h> 88 #include <sys/vmem.h> 89 #include <sys/sx.h> 90 #include <sys/sysctl.h> 91 #include <sys/kernel.h> 92 #include <sys/ktr.h> 93 #include <sys/unistd.h> 94 95 #include <vm/uma.h> 96 #include <vm/vm.h> 97 #include <vm/vm_param.h> 98 #include <vm/pmap.h> 99 #include <vm/vm_domainset.h> 100 #include <vm/vm_map.h> 101 #include <vm/vm_page.h> 102 #include <vm/vm_pageout.h> 103 #include <vm/vm_object.h> 104 #include <vm/vm_kern.h> 105 #include <vm/vm_extern.h> 106 #include <vm/vm_pager.h> 107 #include <vm/swap_pager.h> 108 109 #include <machine/cpu.h> 110 111 /* 112 * MPSAFE 113 * 114 * WARNING! This code calls vm_map_check_protection() which only checks 115 * the associated vm_map_entry range. It does not determine whether the 116 * contents of the memory is actually readable or writable. In most cases 117 * just checking the vm_map_entry is sufficient within the kernel's address 118 * space. 119 */ 120 int 121 kernacc(void *addr, int len, int rw) 122 { 123 boolean_t rv; 124 vm_offset_t saddr, eaddr; 125 vm_prot_t prot; 126 127 KASSERT((rw & ~VM_PROT_ALL) == 0, 128 ("illegal ``rw'' argument to kernacc (%x)\n", rw)); 129 130 if ((vm_offset_t)addr + len > vm_map_max(kernel_map) || 131 (vm_offset_t)addr + len < (vm_offset_t)addr) 132 return (FALSE); 133 134 prot = rw; 135 saddr = trunc_page((vm_offset_t)addr); 136 eaddr = round_page((vm_offset_t)addr + len); 137 vm_map_lock_read(kernel_map); 138 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); 139 vm_map_unlock_read(kernel_map); 140 return (rv == TRUE); 141 } 142 143 /* 144 * MPSAFE 145 * 146 * WARNING! This code calls vm_map_check_protection() which only checks 147 * the associated vm_map_entry range. It does not determine whether the 148 * contents of the memory is actually readable or writable. vmapbuf(), 149 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be 150 * used in conjunction with this call. 151 */ 152 int 153 useracc(void *addr, int len, int rw) 154 { 155 boolean_t rv; 156 vm_prot_t prot; 157 vm_map_t map; 158 159 KASSERT((rw & ~VM_PROT_ALL) == 0, 160 ("illegal ``rw'' argument to useracc (%x)\n", rw)); 161 prot = rw; 162 map = &curproc->p_vmspace->vm_map; 163 if ((vm_offset_t)addr + len > vm_map_max(map) || 164 (vm_offset_t)addr + len < (vm_offset_t)addr) { 165 return (FALSE); 166 } 167 vm_map_lock_read(map); 168 rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), 169 round_page((vm_offset_t)addr + len), prot); 170 vm_map_unlock_read(map); 171 return (rv == TRUE); 172 } 173 174 int 175 vslock(void *addr, size_t len) 176 { 177 vm_offset_t end, last, start; 178 vm_size_t npages; 179 int error; 180 181 last = (vm_offset_t)addr + len; 182 start = trunc_page((vm_offset_t)addr); 183 end = round_page(last); 184 if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) 185 return (EINVAL); 186 npages = atop(end - start); 187 if (npages > vm_page_max_user_wired) 188 return (ENOMEM); 189 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, 190 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 191 if (error == KERN_SUCCESS) { 192 curthread->td_vslock_sz += len; 193 return (0); 194 } 195 196 /* 197 * Return EFAULT on error to match copy{in,out}() behaviour 198 * rather than returning ENOMEM like mlock() would. 199 */ 200 return (EFAULT); 201 } 202 203 void 204 vsunlock(void *addr, size_t len) 205 { 206 207 /* Rely on the parameter sanity checks performed by vslock(). */ 208 MPASS(curthread->td_vslock_sz >= len); 209 curthread->td_vslock_sz -= len; 210 (void)vm_map_unwire(&curproc->p_vmspace->vm_map, 211 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), 212 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 213 } 214 215 /* 216 * Pin the page contained within the given object at the given offset. If the 217 * page is not resident, allocate and load it using the given object's pager. 218 * Return the pinned page if successful; otherwise, return NULL. 219 */ 220 static vm_page_t 221 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset) 222 { 223 vm_page_t m; 224 vm_pindex_t pindex; 225 226 pindex = OFF_TO_IDX(offset); 227 (void)vm_page_grab_valid_unlocked(&m, object, pindex, 228 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 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 vm_object_t kstack_object; 270 static uma_zone_t kstack_cache; 271 static int kstack_cache_size; 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 */ 291 static vm_offset_t 292 vm_thread_stack_create(struct domainset *ds, int pages) 293 { 294 vm_page_t ma[KSTACK_MAX_PAGES]; 295 vm_offset_t ks; 296 int i; 297 298 /* 299 * Get a kernel virtual address for this thread's kstack. 300 */ 301 ks = kva_alloc((pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 302 if (ks == 0) { 303 printf("%s: kstack allocation failed\n", __func__); 304 return (0); 305 } 306 307 if (KSTACK_GUARD_PAGES != 0) { 308 pmap_qremove(ks, KSTACK_GUARD_PAGES); 309 ks += KSTACK_GUARD_PAGES * PAGE_SIZE; 310 } 311 312 /* 313 * Allocate physical pages to back the stack. 314 */ 315 vm_thread_stack_back(ds, ks, ma, pages, VM_ALLOC_NORMAL); 316 for (i = 0; i < pages; i++) 317 vm_page_valid(ma[i]); 318 pmap_qenter(ks, ma, pages); 319 320 return (ks); 321 } 322 323 static void 324 vm_thread_stack_dispose(vm_offset_t ks, int pages) 325 { 326 vm_page_t m; 327 vm_pindex_t pindex; 328 int i; 329 330 pindex = atop(ks - VM_MIN_KERNEL_ADDRESS); 331 332 pmap_qremove(ks, pages); 333 VM_OBJECT_WLOCK(kstack_object); 334 for (i = 0; i < pages; i++) { 335 m = vm_page_lookup(kstack_object, pindex + i); 336 if (m == NULL) 337 panic("%s: kstack already missing?", __func__); 338 vm_page_xbusy_claim(m); 339 vm_page_unwire_noq(m); 340 vm_page_free(m); 341 } 342 VM_OBJECT_WUNLOCK(kstack_object); 343 kasan_mark((void *)ks, ptoa(pages), ptoa(pages), 0); 344 kva_free(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), 345 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE); 346 } 347 348 /* 349 * Allocate the kernel stack for a new thread. 350 */ 351 int 352 vm_thread_new(struct thread *td, int pages) 353 { 354 vm_offset_t ks; 355 356 /* Bounds check */ 357 if (pages <= 1) 358 pages = kstack_pages; 359 else if (pages > KSTACK_MAX_PAGES) 360 pages = KSTACK_MAX_PAGES; 361 362 ks = 0; 363 if (pages == kstack_pages && kstack_cache != NULL) 364 ks = (vm_offset_t)uma_zalloc(kstack_cache, M_NOWAIT); 365 366 /* 367 * Ensure that kstack objects can draw pages from any memory 368 * domain. Otherwise a local memory shortage can block a process 369 * swap-in. 370 */ 371 if (ks == 0) 372 ks = vm_thread_stack_create(DOMAINSET_PREF(PCPU_GET(domain)), 373 pages); 374 if (ks == 0) 375 return (0); 376 td->td_kstack = ks; 377 td->td_kstack_pages = pages; 378 kasan_mark((void *)ks, ptoa(pages), ptoa(pages), 0); 379 kmsan_mark((void *)ks, ptoa(pages), KMSAN_STATE_UNINIT); 380 return (1); 381 } 382 383 /* 384 * Dispose of a thread's kernel stack. 385 */ 386 void 387 vm_thread_dispose(struct thread *td) 388 { 389 vm_offset_t ks; 390 int pages; 391 392 pages = td->td_kstack_pages; 393 ks = td->td_kstack; 394 td->td_kstack = 0; 395 td->td_kstack_pages = 0; 396 kasan_mark((void *)ks, 0, ptoa(pages), KASAN_KSTACK_FREED); 397 if (pages == kstack_pages) 398 uma_zfree(kstack_cache, (void *)ks); 399 else 400 vm_thread_stack_dispose(ks, pages); 401 } 402 403 /* 404 * Allocate physical pages, following the specified NUMA policy, to back a 405 * kernel stack. 406 */ 407 void 408 vm_thread_stack_back(struct domainset *ds, vm_offset_t ks, vm_page_t ma[], 409 int npages, int req_class) 410 { 411 vm_pindex_t pindex; 412 int n; 413 414 pindex = atop(ks - VM_MIN_KERNEL_ADDRESS); 415 416 VM_OBJECT_WLOCK(kstack_object); 417 for (n = 0; n < npages;) { 418 if (vm_ndomains > 1) 419 kstack_object->domain.dr_policy = ds; 420 421 /* 422 * Use WAITFAIL to force a reset of the domain selection policy 423 * if we had to sleep for pages. 424 */ 425 n += vm_page_grab_pages(kstack_object, pindex + n, 426 req_class | VM_ALLOC_WIRED | VM_ALLOC_WAITFAIL, 427 &ma[n], npages - n); 428 } 429 VM_OBJECT_WUNLOCK(kstack_object); 430 } 431 432 static int 433 kstack_import(void *arg, void **store, int cnt, int domain, int flags) 434 { 435 struct domainset *ds; 436 int i; 437 438 if (domain == UMA_ANYDOMAIN) 439 ds = DOMAINSET_RR(); 440 else 441 ds = DOMAINSET_PREF(domain); 442 443 for (i = 0; i < cnt; i++) { 444 store[i] = (void *)vm_thread_stack_create(ds, kstack_pages); 445 if (store[i] == NULL) 446 break; 447 } 448 return (i); 449 } 450 451 static void 452 kstack_release(void *arg, void **store, int cnt) 453 { 454 vm_offset_t ks; 455 int i; 456 457 for (i = 0; i < cnt; i++) { 458 ks = (vm_offset_t)store[i]; 459 vm_thread_stack_dispose(ks, kstack_pages); 460 } 461 } 462 463 static void 464 kstack_cache_init(void *null) 465 { 466 kstack_object = vm_object_allocate(OBJT_SWAP, 467 atop(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS)); 468 kstack_cache = uma_zcache_create("kstack_cache", 469 kstack_pages * PAGE_SIZE, NULL, NULL, NULL, NULL, 470 kstack_import, kstack_release, NULL, 471 UMA_ZONE_FIRSTTOUCH); 472 kstack_cache_size = imax(128, mp_ncpus * 4); 473 uma_zone_set_maxcache(kstack_cache, kstack_cache_size); 474 } 475 SYSINIT(vm_kstacks, SI_SUB_KMEM, SI_ORDER_ANY, kstack_cache_init, NULL); 476 477 #ifdef KSTACK_USAGE_PROF 478 /* 479 * Track maximum stack used by a thread in kernel. 480 */ 481 static int max_kstack_used; 482 483 SYSCTL_INT(_debug, OID_AUTO, max_kstack_used, CTLFLAG_RD, 484 &max_kstack_used, 0, 485 "Maximum stack depth used by a thread in kernel"); 486 487 void 488 intr_prof_stack_use(struct thread *td, struct trapframe *frame) 489 { 490 vm_offset_t stack_top; 491 vm_offset_t current; 492 int used, prev_used; 493 494 /* 495 * Testing for interrupted kernel mode isn't strictly 496 * needed. It optimizes the execution, since interrupts from 497 * usermode will have only the trap frame on the stack. 498 */ 499 if (TRAPF_USERMODE(frame)) 500 return; 501 502 stack_top = td->td_kstack + td->td_kstack_pages * PAGE_SIZE; 503 current = (vm_offset_t)(uintptr_t)&stack_top; 504 505 /* 506 * Try to detect if interrupt is using kernel thread stack. 507 * Hardware could use a dedicated stack for interrupt handling. 508 */ 509 if (stack_top <= current || current < td->td_kstack) 510 return; 511 512 used = stack_top - current; 513 for (;;) { 514 prev_used = max_kstack_used; 515 if (prev_used >= used) 516 break; 517 if (atomic_cmpset_int(&max_kstack_used, prev_used, used)) 518 break; 519 } 520 } 521 #endif /* KSTACK_USAGE_PROF */ 522 523 /* 524 * Implement fork's actions on an address space. 525 * Here we arrange for the address space to be copied or referenced, 526 * allocate a user struct (pcb and kernel stack), then call the 527 * machine-dependent layer to fill those in and make the new process 528 * ready to run. The new process is set up so that it returns directly 529 * to user mode to avoid stack copying and relocation problems. 530 */ 531 int 532 vm_forkproc(struct thread *td, struct proc *p2, struct thread *td2, 533 struct vmspace *vm2, int flags) 534 { 535 struct proc *p1 = td->td_proc; 536 struct domainset *dset; 537 int error; 538 539 if ((flags & RFPROC) == 0) { 540 /* 541 * Divorce the memory, if it is shared, essentially 542 * this changes shared memory amongst threads, into 543 * COW locally. 544 */ 545 if ((flags & RFMEM) == 0) { 546 error = vmspace_unshare(p1); 547 if (error) 548 return (error); 549 } 550 cpu_fork(td, p2, td2, flags); 551 return (0); 552 } 553 554 if (flags & RFMEM) { 555 p2->p_vmspace = p1->p_vmspace; 556 refcount_acquire(&p1->p_vmspace->vm_refcnt); 557 } 558 dset = td2->td_domain.dr_policy; 559 while (vm_page_count_severe_set(&dset->ds_mask)) { 560 vm_wait_doms(&dset->ds_mask, 0); 561 } 562 563 if ((flags & RFMEM) == 0) { 564 p2->p_vmspace = vm2; 565 if (p1->p_vmspace->vm_shm) 566 shmfork(p1, p2); 567 } 568 569 /* 570 * cpu_fork will copy and update the pcb, set up the kernel stack, 571 * and make the child ready to run. 572 */ 573 cpu_fork(td, p2, td2, flags); 574 return (0); 575 } 576 577 /* 578 * Called after process has been wait(2)'ed upon and is being reaped. 579 * The idea is to reclaim resources that we could not reclaim while 580 * the process was still executing. 581 */ 582 void 583 vm_waitproc(p) 584 struct proc *p; 585 { 586 587 vmspace_exitfree(p); /* and clean-out the vmspace */ 588 } 589 590 void 591 kick_proc0(void) 592 { 593 594 wakeup(&proc0); 595 } 596