FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_glue.c
1 /*-
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
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 __FBSDID("$FreeBSD: releng/11.2/sys/vm/vm_glue.c 331842 2018-03-31 13:19:27Z kib $");
61
62 #include "opt_vm.h"
63 #include "opt_kstack_pages.h"
64 #include "opt_kstack_max_pages.h"
65 #include "opt_kstack_usage_prof.h"
66
67 #include <sys/param.h>
68 #include <sys/systm.h>
69 #include <sys/limits.h>
70 #include <sys/lock.h>
71 #include <sys/malloc.h>
72 #include <sys/mutex.h>
73 #include <sys/proc.h>
74 #include <sys/racct.h>
75 #include <sys/resourcevar.h>
76 #include <sys/rwlock.h>
77 #include <sys/sched.h>
78 #include <sys/sf_buf.h>
79 #include <sys/shm.h>
80 #include <sys/vmmeter.h>
81 #include <sys/vmem.h>
82 #include <sys/sx.h>
83 #include <sys/sysctl.h>
84 #include <sys/_kstack_cache.h>
85 #include <sys/eventhandler.h>
86 #include <sys/kernel.h>
87 #include <sys/ktr.h>
88 #include <sys/unistd.h>
89
90 #include <vm/vm.h>
91 #include <vm/vm_param.h>
92 #include <vm/pmap.h>
93 #include <vm/vm_map.h>
94 #include <vm/vm_page.h>
95 #include <vm/vm_pageout.h>
96 #include <vm/vm_object.h>
97 #include <vm/vm_kern.h>
98 #include <vm/vm_extern.h>
99 #include <vm/vm_pager.h>
100 #include <vm/swap_pager.h>
101
102 #include <machine/cpu.h>
103
104 /*
105 * MPSAFE
106 *
107 * WARNING! This code calls vm_map_check_protection() which only checks
108 * the associated vm_map_entry range. It does not determine whether the
109 * contents of the memory is actually readable or writable. In most cases
110 * just checking the vm_map_entry is sufficient within the kernel's address
111 * space.
112 */
113 int
114 kernacc(addr, len, rw)
115 void *addr;
116 int len, rw;
117 {
118 boolean_t rv;
119 vm_offset_t saddr, eaddr;
120 vm_prot_t prot;
121
122 KASSERT((rw & ~VM_PROT_ALL) == 0,
123 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
124
125 if ((vm_offset_t)addr + len > kernel_map->max_offset ||
126 (vm_offset_t)addr + len < (vm_offset_t)addr)
127 return (FALSE);
128
129 prot = rw;
130 saddr = trunc_page((vm_offset_t)addr);
131 eaddr = round_page((vm_offset_t)addr + len);
132 vm_map_lock_read(kernel_map);
133 rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot);
134 vm_map_unlock_read(kernel_map);
135 return (rv == TRUE);
136 }
137
138 /*
139 * MPSAFE
140 *
141 * WARNING! This code calls vm_map_check_protection() which only checks
142 * the associated vm_map_entry range. It does not determine whether the
143 * contents of the memory is actually readable or writable. vmapbuf(),
144 * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be
145 * used in conjunction with this call.
146 */
147 int
148 useracc(addr, len, rw)
149 void *addr;
150 int len, 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_wired)
185 return (ENOMEM);
186 #if 0
187 /*
188 * XXX - not yet
189 *
190 * The limit for transient usage of wired pages should be
191 * larger than for "permanent" wired pages (mlock()).
192 *
193 * Also, the sysctl code, which is the only present user
194 * of vslock(), does a hard loop on EAGAIN.
195 */
196 if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
197 return (EAGAIN);
198 #endif
199 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end,
200 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
201 if (error == KERN_SUCCESS) {
202 curthread->td_vslock_sz += len;
203 return (0);
204 }
205
206 /*
207 * Return EFAULT on error to match copy{in,out}() behaviour
208 * rather than returning ENOMEM like mlock() would.
209 */
210 return (EFAULT);
211 }
212
213 void
214 vsunlock(void *addr, size_t len)
215 {
216
217 /* Rely on the parameter sanity checks performed by vslock(). */
218 MPASS(curthread->td_vslock_sz >= len);
219 curthread->td_vslock_sz -= len;
220 (void)vm_map_unwire(&curproc->p_vmspace->vm_map,
221 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len),
222 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
223 }
224
225 /*
226 * Pin the page contained within the given object at the given offset. If the
227 * page is not resident, allocate and load it using the given object's pager.
228 * Return the pinned page if successful; otherwise, return NULL.
229 */
230 static vm_page_t
231 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset)
232 {
233 vm_page_t m;
234 vm_pindex_t pindex;
235 int rv;
236
237 VM_OBJECT_WLOCK(object);
238 pindex = OFF_TO_IDX(offset);
239 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
240 if (m->valid != VM_PAGE_BITS_ALL) {
241 vm_page_xbusy(m);
242 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL);
243 if (rv != VM_PAGER_OK) {
244 vm_page_lock(m);
245 vm_page_free(m);
246 vm_page_unlock(m);
247 m = NULL;
248 goto out;
249 }
250 vm_page_xunbusy(m);
251 }
252 vm_page_lock(m);
253 vm_page_hold(m);
254 vm_page_activate(m);
255 vm_page_unlock(m);
256 out:
257 VM_OBJECT_WUNLOCK(object);
258 return (m);
259 }
260
261 /*
262 * Return a CPU private mapping to the page at the given offset within the
263 * given object. The page is pinned before it is mapped.
264 */
265 struct sf_buf *
266 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset)
267 {
268 vm_page_t m;
269
270 m = vm_imgact_hold_page(object, offset);
271 if (m == NULL)
272 return (NULL);
273 sched_pin();
274 return (sf_buf_alloc(m, SFB_CPUPRIVATE));
275 }
276
277 /*
278 * Destroy the given CPU private mapping and unpin the page that it mapped.
279 */
280 void
281 vm_imgact_unmap_page(struct sf_buf *sf)
282 {
283 vm_page_t m;
284
285 m = sf_buf_page(sf);
286 sf_buf_free(sf);
287 sched_unpin();
288 vm_page_lock(m);
289 vm_page_unhold(m);
290 vm_page_unlock(m);
291 }
292
293 void
294 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz)
295 {
296
297 pmap_sync_icache(map->pmap, va, sz);
298 }
299
300 struct kstack_cache_entry *kstack_cache;
301 static int kstack_cache_size = 128;
302 static int kstacks;
303 static struct mtx kstack_cache_mtx;
304 MTX_SYSINIT(kstack_cache, &kstack_cache_mtx, "kstkch", MTX_DEF);
305
306 SYSCTL_INT(_vm, OID_AUTO, kstack_cache_size, CTLFLAG_RW, &kstack_cache_size, 0,
307 "");
308 SYSCTL_INT(_vm, OID_AUTO, kstacks, CTLFLAG_RD, &kstacks, 0,
309 "");
310
311 /*
312 * Create the kernel stack (including pcb for i386) for a new thread.
313 * This routine directly affects the fork perf for a process and
314 * create performance for a thread.
315 */
316 int
317 vm_thread_new(struct thread *td, int pages)
318 {
319 vm_object_t ksobj;
320 vm_offset_t ks;
321 vm_page_t ma[KSTACK_MAX_PAGES];
322 struct kstack_cache_entry *ks_ce;
323 int i;
324
325 /* Bounds check */
326 if (pages <= 1)
327 pages = kstack_pages;
328 else if (pages > KSTACK_MAX_PAGES)
329 pages = KSTACK_MAX_PAGES;
330
331 if (pages == kstack_pages) {
332 mtx_lock(&kstack_cache_mtx);
333 if (kstack_cache != NULL) {
334 ks_ce = kstack_cache;
335 kstack_cache = ks_ce->next_ks_entry;
336 mtx_unlock(&kstack_cache_mtx);
337
338 td->td_kstack_obj = ks_ce->ksobj;
339 td->td_kstack = (vm_offset_t)ks_ce;
340 td->td_kstack_pages = kstack_pages;
341 return (1);
342 }
343 mtx_unlock(&kstack_cache_mtx);
344 }
345
346 /*
347 * Allocate an object for the kstack.
348 */
349 ksobj = vm_object_allocate(OBJT_DEFAULT, pages);
350
351 /*
352 * Get a kernel virtual address for this thread's kstack.
353 */
354 #if defined(__mips__)
355 /*
356 * We need to align the kstack's mapped address to fit within
357 * a single TLB entry.
358 */
359 if (vmem_xalloc(kernel_arena, (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE,
360 PAGE_SIZE * 2, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX,
361 M_BESTFIT | M_NOWAIT, &ks)) {
362 ks = 0;
363 }
364 #else
365 ks = kva_alloc((pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
366 #endif
367 if (ks == 0) {
368 printf("vm_thread_new: kstack allocation failed\n");
369 vm_object_deallocate(ksobj);
370 return (0);
371 }
372
373 atomic_add_int(&kstacks, 1);
374 if (KSTACK_GUARD_PAGES != 0) {
375 pmap_qremove(ks, KSTACK_GUARD_PAGES);
376 ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
377 }
378 td->td_kstack_obj = ksobj;
379 td->td_kstack = ks;
380 /*
381 * Knowing the number of pages allocated is useful when you
382 * want to deallocate them.
383 */
384 td->td_kstack_pages = pages;
385 /*
386 * For the length of the stack, link in a real page of ram for each
387 * page of stack.
388 */
389 VM_OBJECT_WLOCK(ksobj);
390 (void)vm_page_grab_pages(ksobj, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
391 VM_ALLOC_WIRED, ma, pages);
392 for (i = 0; i < pages; i++)
393 ma[i]->valid = VM_PAGE_BITS_ALL;
394 VM_OBJECT_WUNLOCK(ksobj);
395 pmap_qenter(ks, ma, pages);
396 return (1);
397 }
398
399 static void
400 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages)
401 {
402 vm_page_t m;
403 int i;
404
405 atomic_add_int(&kstacks, -1);
406 pmap_qremove(ks, pages);
407 VM_OBJECT_WLOCK(ksobj);
408 for (i = 0; i < pages; i++) {
409 m = vm_page_lookup(ksobj, i);
410 if (m == NULL)
411 panic("vm_thread_dispose: kstack already missing?");
412 vm_page_lock(m);
413 vm_page_unwire(m, PQ_NONE);
414 vm_page_free(m);
415 vm_page_unlock(m);
416 }
417 VM_OBJECT_WUNLOCK(ksobj);
418 vm_object_deallocate(ksobj);
419 kva_free(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE),
420 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
421 }
422
423 /*
424 * Dispose of a thread's kernel stack.
425 */
426 void
427 vm_thread_dispose(struct thread *td)
428 {
429 vm_object_t ksobj;
430 vm_offset_t ks;
431 struct kstack_cache_entry *ks_ce;
432 int pages;
433
434 pages = td->td_kstack_pages;
435 ksobj = td->td_kstack_obj;
436 ks = td->td_kstack;
437 td->td_kstack = 0;
438 td->td_kstack_pages = 0;
439 if (pages == kstack_pages && kstacks <= kstack_cache_size) {
440 ks_ce = (struct kstack_cache_entry *)ks;
441 ks_ce->ksobj = ksobj;
442 mtx_lock(&kstack_cache_mtx);
443 ks_ce->next_ks_entry = kstack_cache;
444 kstack_cache = ks_ce;
445 mtx_unlock(&kstack_cache_mtx);
446 return;
447 }
448 vm_thread_stack_dispose(ksobj, ks, pages);
449 }
450
451 static void
452 vm_thread_stack_lowmem(void *nulll)
453 {
454 struct kstack_cache_entry *ks_ce, *ks_ce1;
455
456 mtx_lock(&kstack_cache_mtx);
457 ks_ce = kstack_cache;
458 kstack_cache = NULL;
459 mtx_unlock(&kstack_cache_mtx);
460
461 while (ks_ce != NULL) {
462 ks_ce1 = ks_ce;
463 ks_ce = ks_ce->next_ks_entry;
464
465 vm_thread_stack_dispose(ks_ce1->ksobj, (vm_offset_t)ks_ce1,
466 kstack_pages);
467 }
468 }
469
470 static void
471 kstack_cache_init(void *nulll)
472 {
473
474 EVENTHANDLER_REGISTER(vm_lowmem, vm_thread_stack_lowmem, NULL,
475 EVENTHANDLER_PRI_ANY);
476 }
477
478 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL);
479
480 #ifdef KSTACK_USAGE_PROF
481 /*
482 * Track maximum stack used by a thread in kernel.
483 */
484 static int max_kstack_used;
485
486 SYSCTL_INT(_debug, OID_AUTO, max_kstack_used, CTLFLAG_RD,
487 &max_kstack_used, 0,
488 "Maxiumum stack depth used by a thread in kernel");
489
490 void
491 intr_prof_stack_use(struct thread *td, struct trapframe *frame)
492 {
493 vm_offset_t stack_top;
494 vm_offset_t current;
495 int used, prev_used;
496
497 /*
498 * Testing for interrupted kernel mode isn't strictly
499 * needed. It optimizes the execution, since interrupts from
500 * usermode will have only the trap frame on the stack.
501 */
502 if (TRAPF_USERMODE(frame))
503 return;
504
505 stack_top = td->td_kstack + td->td_kstack_pages * PAGE_SIZE;
506 current = (vm_offset_t)(uintptr_t)&stack_top;
507
508 /*
509 * Try to detect if interrupt is using kernel thread stack.
510 * Hardware could use a dedicated stack for interrupt handling.
511 */
512 if (stack_top <= current || current < td->td_kstack)
513 return;
514
515 used = stack_top - current;
516 for (;;) {
517 prev_used = max_kstack_used;
518 if (prev_used >= used)
519 break;
520 if (atomic_cmpset_int(&max_kstack_used, prev_used, used))
521 break;
522 }
523 }
524 #endif /* KSTACK_USAGE_PROF */
525
526 /*
527 * Implement fork's actions on an address space.
528 * Here we arrange for the address space to be copied or referenced,
529 * allocate a user struct (pcb and kernel stack), then call the
530 * machine-dependent layer to fill those in and make the new process
531 * ready to run. The new process is set up so that it returns directly
532 * to user mode to avoid stack copying and relocation problems.
533 */
534 int
535 vm_forkproc(td, p2, td2, vm2, flags)
536 struct thread *td;
537 struct proc *p2;
538 struct thread *td2;
539 struct vmspace *vm2;
540 int flags;
541 {
542 struct proc *p1 = td->td_proc;
543 int error;
544
545 if ((flags & RFPROC) == 0) {
546 /*
547 * Divorce the memory, if it is shared, essentially
548 * this changes shared memory amongst threads, into
549 * COW locally.
550 */
551 if ((flags & RFMEM) == 0) {
552 if (p1->p_vmspace->vm_refcnt > 1) {
553 error = vmspace_unshare(p1);
554 if (error)
555 return (error);
556 }
557 }
558 cpu_fork(td, p2, td2, flags);
559 return (0);
560 }
561
562 if (flags & RFMEM) {
563 p2->p_vmspace = p1->p_vmspace;
564 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1);
565 }
566
567 while (vm_page_count_severe()) {
568 VM_WAIT;
569 }
570
571 if ((flags & RFMEM) == 0) {
572 p2->p_vmspace = vm2;
573 if (p1->p_vmspace->vm_shm)
574 shmfork(p1, p2);
575 }
576
577 /*
578 * cpu_fork will copy and update the pcb, set up the kernel stack,
579 * and make the child ready to run.
580 */
581 cpu_fork(td, p2, td2, flags);
582 return (0);
583 }
584
585 /*
586 * Called after process has been wait(2)'ed upon and is being reaped.
587 * The idea is to reclaim resources that we could not reclaim while
588 * the process was still executing.
589 */
590 void
591 vm_waitproc(p)
592 struct proc *p;
593 {
594
595 vmspace_exitfree(p); /* and clean-out the vmspace */
596 }
597
598 void
599 kick_proc0(void)
600 {
601
602 wakeup(&proc0);
603 }
Cache object: c2bf40251ff657571bfe8adaf737bd18
|