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 #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/asan.h>
70 #include <sys/domainset.h>
71 #include <sys/limits.h>
72 #include <sys/lock.h>
73 #include <sys/malloc.h>
74 #include <sys/mutex.h>
75 #include <sys/proc.h>
76 #include <sys/racct.h>
77 #include <sys/refcount.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/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
kernacc(void * addr,int len,int rw)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
useracc(void * addr,int len,int rw)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
vslock(void * addr,size_t len)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
vsunlock(void * addr,size_t len)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
vm_imgact_hold_page(vm_object_t object,vm_ooffset_t offset)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
223 pindex = OFF_TO_IDX(offset);
224 (void)vm_page_grab_valid_unlocked(&m, object, pindex,
225 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
226 return (m);
227 }
228
229 /*
230 * Return a CPU private mapping to the page at the given offset within the
231 * given object. The page is pinned before it is mapped.
232 */
233 struct sf_buf *
vm_imgact_map_page(vm_object_t object,vm_ooffset_t offset)234 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset)
235 {
236 vm_page_t m;
237
238 m = vm_imgact_hold_page(object, offset);
239 if (m == NULL)
240 return (NULL);
241 sched_pin();
242 return (sf_buf_alloc(m, SFB_CPUPRIVATE));
243 }
244
245 /*
246 * Destroy the given CPU private mapping and unpin the page that it mapped.
247 */
248 void
vm_imgact_unmap_page(struct sf_buf * sf)249 vm_imgact_unmap_page(struct sf_buf *sf)
250 {
251 vm_page_t m;
252
253 m = sf_buf_page(sf);
254 sf_buf_free(sf);
255 sched_unpin();
256 vm_page_unwire(m, PQ_ACTIVE);
257 }
258
259 void
vm_sync_icache(vm_map_t map,vm_offset_t va,vm_offset_t sz)260 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz)
261 {
262
263 pmap_sync_icache(map->pmap, va, sz);
264 }
265
266 vm_object_t kstack_object;
267 static uma_zone_t kstack_cache;
268 static int kstack_cache_size;
269
270 static int
sysctl_kstack_cache_size(SYSCTL_HANDLER_ARGS)271 sysctl_kstack_cache_size(SYSCTL_HANDLER_ARGS)
272 {
273 int error, oldsize;
274
275 oldsize = kstack_cache_size;
276 error = sysctl_handle_int(oidp, arg1, arg2, req);
277 if (error == 0 && req->newptr && oldsize != kstack_cache_size)
278 uma_zone_set_maxcache(kstack_cache, kstack_cache_size);
279 return (error);
280 }
281 SYSCTL_PROC(_vm, OID_AUTO, kstack_cache_size,
282 CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &kstack_cache_size, 0,
283 sysctl_kstack_cache_size, "IU", "Maximum number of cached kernel stacks");
284
285 /*
286 * Create the kernel stack (including pcb for i386) for a new thread.
287 */
288 static vm_offset_t
vm_thread_stack_create(struct domainset * ds,int pages)289 vm_thread_stack_create(struct domainset *ds, int pages)
290 {
291 vm_page_t ma[KSTACK_MAX_PAGES];
292 vm_offset_t ks;
293 int i;
294
295 /*
296 * Get a kernel virtual address for this thread's kstack.
297 */
298 #if defined(__mips__)
299 /*
300 * We need to align the kstack's mapped address to fit within
301 * a single TLB entry.
302 */
303 if (vmem_xalloc(kernel_arena, (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE,
304 PAGE_SIZE * 2, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX,
305 M_BESTFIT | M_NOWAIT, &ks)) {
306 ks = 0;
307 }
308 #else
309 ks = kva_alloc((pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
310 #endif
311 if (ks == 0) {
312 printf("%s: kstack allocation failed\n", __func__);
313 return (0);
314 }
315
316 if (KSTACK_GUARD_PAGES != 0) {
317 pmap_qremove(ks, KSTACK_GUARD_PAGES);
318 ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
319 }
320
321 /*
322 * Allocate physical pages to back the stack.
323 */
324 vm_thread_stack_back(ds, ks, ma, pages, VM_ALLOC_NORMAL);
325 for (i = 0; i < pages; i++)
326 vm_page_valid(ma[i]);
327 pmap_qenter(ks, ma, pages);
328
329 return (ks);
330 }
331
332 static void
vm_thread_stack_dispose(vm_offset_t ks,int pages)333 vm_thread_stack_dispose(vm_offset_t ks, int pages)
334 {
335 vm_page_t m;
336 vm_pindex_t pindex;
337 int i;
338
339 pindex = atop(ks - VM_MIN_KERNEL_ADDRESS);
340
341 pmap_qremove(ks, pages);
342 VM_OBJECT_WLOCK(kstack_object);
343 for (i = 0; i < pages; i++) {
344 m = vm_page_lookup(kstack_object, pindex + i);
345 if (m == NULL)
346 panic("%s: kstack already missing?", __func__);
347 vm_page_xbusy_claim(m);
348 vm_page_unwire_noq(m);
349 vm_page_free(m);
350 }
351 VM_OBJECT_WUNLOCK(kstack_object);
352 kasan_mark((void *)ks, ptoa(pages), ptoa(pages), 0);
353 kva_free(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE),
354 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
355 }
356
357 /*
358 * Allocate the kernel stack for a new thread.
359 */
360 int
vm_thread_new(struct thread * td,int pages)361 vm_thread_new(struct thread *td, int pages)
362 {
363 vm_offset_t ks;
364
365 /* Bounds check */
366 if (pages <= 1)
367 pages = kstack_pages;
368 else if (pages > KSTACK_MAX_PAGES)
369 pages = KSTACK_MAX_PAGES;
370
371 ks = 0;
372 if (pages == kstack_pages && kstack_cache != NULL)
373 ks = (vm_offset_t)uma_zalloc(kstack_cache, M_NOWAIT);
374
375 /*
376 * Ensure that kstack objects can draw pages from any memory
377 * domain. Otherwise a local memory shortage can block a process
378 * swap-in.
379 */
380 if (ks == 0)
381 ks = vm_thread_stack_create(DOMAINSET_PREF(PCPU_GET(domain)),
382 pages);
383 if (ks == 0)
384 return (0);
385 td->td_kstack = ks;
386 td->td_kstack_pages = pages;
387 kasan_mark((void *)ks, ptoa(pages), ptoa(pages), 0);
388 return (1);
389 }
390
391 /*
392 * Dispose of a thread's kernel stack.
393 */
394 void
vm_thread_dispose(struct thread * td)395 vm_thread_dispose(struct thread *td)
396 {
397 vm_offset_t ks;
398 int pages;
399
400 pages = td->td_kstack_pages;
401 ks = td->td_kstack;
402 td->td_kstack = 0;
403 td->td_kstack_pages = 0;
404 kasan_mark((void *)ks, 0, ptoa(pages), KASAN_KSTACK_FREED);
405 if (pages == kstack_pages)
406 uma_zfree(kstack_cache, (void *)ks);
407 else
408 vm_thread_stack_dispose(ks, pages);
409 }
410
411 /*
412 * Allocate physical pages, following the specified NUMA policy, to back a
413 * kernel stack.
414 */
415 void
vm_thread_stack_back(struct domainset * ds,vm_offset_t ks,vm_page_t ma[],int npages,int req_class)416 vm_thread_stack_back(struct domainset *ds, vm_offset_t ks, vm_page_t ma[],
417 int npages, int req_class)
418 {
419 vm_pindex_t pindex;
420 int n;
421
422 pindex = atop(ks - VM_MIN_KERNEL_ADDRESS);
423
424 VM_OBJECT_WLOCK(kstack_object);
425 for (n = 0; n < npages;) {
426 if (vm_ndomains > 1)
427 kstack_object->domain.dr_policy = ds;
428
429 /*
430 * Use WAITFAIL to force a reset of the domain selection policy
431 * if we had to sleep for pages.
432 */
433 n += vm_page_grab_pages(kstack_object, pindex + n,
434 req_class | VM_ALLOC_WIRED | VM_ALLOC_WAITFAIL,
435 &ma[n], npages - n);
436 }
437 VM_OBJECT_WUNLOCK(kstack_object);
438 }
439
440 static int
kstack_import(void * arg,void ** store,int cnt,int domain,int flags)441 kstack_import(void *arg, void **store, int cnt, int domain, int flags)
442 {
443 struct domainset *ds;
444 int i;
445
446 if (domain == UMA_ANYDOMAIN)
447 ds = DOMAINSET_RR();
448 else
449 ds = DOMAINSET_PREF(domain);
450
451 for (i = 0; i < cnt; i++) {
452 store[i] = (void *)vm_thread_stack_create(ds, kstack_pages);
453 if (store[i] == NULL)
454 break;
455 }
456 return (i);
457 }
458
459 static void
kstack_release(void * arg,void ** store,int cnt)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(ks, kstack_pages);
468 }
469 }
470
471 static void
kstack_cache_init(void * null)472 kstack_cache_init(void *null)
473 {
474 kstack_object = vm_object_allocate(OBJT_SWAP,
475 atop(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS));
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 SYSINIT(vm_kstacks, SI_SUB_KMEM, SI_ORDER_ANY, kstack_cache_init, NULL);
484
485 #ifdef KSTACK_USAGE_PROF
486 /*
487 * Track maximum stack used by a thread in kernel.
488 */
489 static int max_kstack_used;
490
491 SYSCTL_INT(_debug, OID_AUTO, max_kstack_used, CTLFLAG_RD,
492 &max_kstack_used, 0,
493 "Maximum stack depth used by a thread in kernel");
494
495 void
intr_prof_stack_use(struct thread * td,struct trapframe * frame)496 intr_prof_stack_use(struct thread *td, struct trapframe *frame)
497 {
498 vm_offset_t stack_top;
499 vm_offset_t current;
500 int used, prev_used;
501
502 /*
503 * Testing for interrupted kernel mode isn't strictly
504 * needed. It optimizes the execution, since interrupts from
505 * usermode will have only the trap frame on the stack.
506 */
507 if (TRAPF_USERMODE(frame))
508 return;
509
510 stack_top = td->td_kstack + td->td_kstack_pages * PAGE_SIZE;
511 current = (vm_offset_t)(uintptr_t)&stack_top;
512
513 /*
514 * Try to detect if interrupt is using kernel thread stack.
515 * Hardware could use a dedicated stack for interrupt handling.
516 */
517 if (stack_top <= current || current < td->td_kstack)
518 return;
519
520 used = stack_top - current;
521 for (;;) {
522 prev_used = max_kstack_used;
523 if (prev_used >= used)
524 break;
525 if (atomic_cmpset_int(&max_kstack_used, prev_used, used))
526 break;
527 }
528 }
529 #endif /* KSTACK_USAGE_PROF */
530
531 /*
532 * Implement fork's actions on an address space.
533 * Here we arrange for the address space to be copied or referenced,
534 * allocate a user struct (pcb and kernel stack), then call the
535 * machine-dependent layer to fill those in and make the new process
536 * ready to run. The new process is set up so that it returns directly
537 * to user mode to avoid stack copying and relocation problems.
538 */
539 int
vm_forkproc(struct thread * td,struct proc * p2,struct thread * td2,struct vmspace * vm2,int flags)540 vm_forkproc(struct thread *td, struct proc *p2, struct thread *td2,
541 struct vmspace *vm2, int flags)
542 {
543 struct proc *p1 = td->td_proc;
544 struct domainset *dset;
545 int error;
546
547 if ((flags & RFPROC) == 0) {
548 /*
549 * Divorce the memory, if it is shared, essentially
550 * this changes shared memory amongst threads, into
551 * COW locally.
552 */
553 if ((flags & RFMEM) == 0) {
554 error = vmspace_unshare(p1);
555 if (error)
556 return (error);
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 refcount_acquire(&p1->p_vmspace->vm_refcnt);
565 }
566 dset = td2->td_domain.dr_policy;
567 while (vm_page_count_severe_set(&dset->ds_mask)) {
568 vm_wait_doms(&dset->ds_mask, 0);
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
vm_waitproc(p)591 vm_waitproc(p)
592 struct proc *p;
593 {
594
595 vmspace_exitfree(p); /* and clean-out the vmspace */
596 }
597
598 void
kick_proc0(void)599 kick_proc0(void)
600 {
601
602 wakeup(&proc0);
603 }
604