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