2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
32 * from: @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
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.
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.
48 * Carnegie Mellon requests users of this software to return to
50 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
51 * School of Computer Science
52 * Carnegie Mellon University
53 * Pittsburgh PA 15213-3890
55 * any improvements or extensions that they make and grant Carnegie the
56 * rights to redistribute these changes.
59 #include <sys/cdefs.h>
60 __FBSDID("$FreeBSD$");
63 #include "opt_kstack_pages.h"
64 #include "opt_kstack_max_pages.h"
65 #include "opt_kstack_usage_prof.h"
67 #include <sys/param.h>
68 #include <sys/systm.h>
69 #include <sys/limits.h>
71 #include <sys/malloc.h>
72 #include <sys/mutex.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>
80 #include <sys/vmmeter.h>
83 #include <sys/sysctl.h>
84 #include <sys/_kstack_cache.h>
85 #include <sys/eventhandler.h>
86 #include <sys/kernel.h>
88 #include <sys/unistd.h>
91 #include <vm/vm_param.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>
102 #include <machine/cpu.h>
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
114 kernacc(addr, len, rw)
119 vm_offset_t saddr, eaddr;
122 KASSERT((rw & ~VM_PROT_ALL) == 0,
123 ("illegal ``rw'' argument to kernacc (%x)\n", rw));
125 if ((vm_offset_t)addr + len > kernel_map->max_offset ||
126 (vm_offset_t)addr + len < (vm_offset_t)addr)
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);
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.
148 useracc(addr, len, rw)
156 KASSERT((rw & ~VM_PROT_ALL) == 0,
157 ("illegal ``rw'' argument to useracc (%x)\n", 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) {
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);
172 vslock(void *addr, size_t len)
174 vm_offset_t end, last, start;
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)
183 npages = atop(end - start);
184 if (npages > vm_page_max_wired)
190 * The limit for transient usage of wired pages should be
191 * larger than for "permanent" wired pages (mlock()).
193 * Also, the sysctl code, which is the only present user
194 * of vslock(), does a hard loop on EAGAIN.
196 if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
199 error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end,
200 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
202 * Return EFAULT on error to match copy{in,out}() behaviour
203 * rather than returning ENOMEM like mlock() would.
205 return (error == KERN_SUCCESS ? 0 : EFAULT);
209 vsunlock(void *addr, size_t len)
212 /* Rely on the parameter sanity checks performed by vslock(). */
213 (void)vm_map_unwire(&curproc->p_vmspace->vm_map,
214 trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len),
215 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
219 * Pin the page contained within the given object at the given offset. If the
220 * page is not resident, allocate and load it using the given object's pager.
221 * Return the pinned page if successful; otherwise, return NULL.
224 vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset)
230 VM_OBJECT_WLOCK(object);
231 pindex = OFF_TO_IDX(offset);
232 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
233 if (m->valid != VM_PAGE_BITS_ALL) {
235 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL);
236 if (rv != VM_PAGER_OK) {
250 VM_OBJECT_WUNLOCK(object);
255 * Return a CPU private mapping to the page at the given offset within the
256 * given object. The page is pinned before it is mapped.
259 vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset)
263 m = vm_imgact_hold_page(object, offset);
267 return (sf_buf_alloc(m, SFB_CPUPRIVATE));
271 * Destroy the given CPU private mapping and unpin the page that it mapped.
274 vm_imgact_unmap_page(struct sf_buf *sf)
287 vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz)
290 pmap_sync_icache(map->pmap, va, sz);
293 struct kstack_cache_entry *kstack_cache;
294 static int kstack_cache_size = 128;
296 static struct mtx kstack_cache_mtx;
297 MTX_SYSINIT(kstack_cache, &kstack_cache_mtx, "kstkch", MTX_DEF);
299 SYSCTL_INT(_vm, OID_AUTO, kstack_cache_size, CTLFLAG_RW, &kstack_cache_size, 0,
301 SYSCTL_INT(_vm, OID_AUTO, kstacks, CTLFLAG_RD, &kstacks, 0,
305 * Create the kernel stack (including pcb for i386) for a new thread.
306 * This routine directly affects the fork perf for a process and
307 * create performance for a thread.
310 vm_thread_new(struct thread *td, int pages)
314 vm_page_t ma[KSTACK_MAX_PAGES];
315 struct kstack_cache_entry *ks_ce;
320 pages = kstack_pages;
321 else if (pages > KSTACK_MAX_PAGES)
322 pages = KSTACK_MAX_PAGES;
324 if (pages == kstack_pages) {
325 mtx_lock(&kstack_cache_mtx);
326 if (kstack_cache != NULL) {
327 ks_ce = kstack_cache;
328 kstack_cache = ks_ce->next_ks_entry;
329 mtx_unlock(&kstack_cache_mtx);
331 td->td_kstack_obj = ks_ce->ksobj;
332 td->td_kstack = (vm_offset_t)ks_ce;
333 td->td_kstack_pages = kstack_pages;
336 mtx_unlock(&kstack_cache_mtx);
340 * Allocate an object for the kstack.
342 ksobj = vm_object_allocate(OBJT_DEFAULT, pages);
345 * Get a kernel virtual address for this thread's kstack.
347 #if defined(__mips__)
349 * We need to align the kstack's mapped address to fit within
350 * a single TLB entry.
352 if (vmem_xalloc(kernel_arena, (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE,
353 PAGE_SIZE * 2, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX,
354 M_BESTFIT | M_NOWAIT, &ks)) {
358 ks = kva_alloc((pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
361 printf("vm_thread_new: kstack allocation failed\n");
362 vm_object_deallocate(ksobj);
366 atomic_add_int(&kstacks, 1);
367 if (KSTACK_GUARD_PAGES != 0) {
368 pmap_qremove(ks, KSTACK_GUARD_PAGES);
369 ks += KSTACK_GUARD_PAGES * PAGE_SIZE;
371 td->td_kstack_obj = ksobj;
374 * Knowing the number of pages allocated is useful when you
375 * want to deallocate them.
377 td->td_kstack_pages = pages;
379 * For the length of the stack, link in a real page of ram for each
382 VM_OBJECT_WLOCK(ksobj);
383 (void)vm_page_grab_pages(ksobj, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
384 VM_ALLOC_WIRED, ma, pages);
385 for (i = 0; i < pages; i++)
386 ma[i]->valid = VM_PAGE_BITS_ALL;
387 VM_OBJECT_WUNLOCK(ksobj);
388 pmap_qenter(ks, ma, pages);
393 vm_thread_stack_dispose(vm_object_t ksobj, vm_offset_t ks, int pages)
398 atomic_add_int(&kstacks, -1);
399 pmap_qremove(ks, pages);
400 VM_OBJECT_WLOCK(ksobj);
401 for (i = 0; i < pages; i++) {
402 m = vm_page_lookup(ksobj, i);
404 panic("vm_thread_dispose: kstack already missing?");
406 vm_page_unwire(m, PQ_NONE);
410 VM_OBJECT_WUNLOCK(ksobj);
411 vm_object_deallocate(ksobj);
412 kva_free(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE),
413 (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
417 * Dispose of a thread's kernel stack.
420 vm_thread_dispose(struct thread *td)
424 struct kstack_cache_entry *ks_ce;
427 pages = td->td_kstack_pages;
428 ksobj = td->td_kstack_obj;
431 td->td_kstack_pages = 0;
432 if (pages == kstack_pages && kstacks <= kstack_cache_size) {
433 ks_ce = (struct kstack_cache_entry *)ks;
434 ks_ce->ksobj = ksobj;
435 mtx_lock(&kstack_cache_mtx);
436 ks_ce->next_ks_entry = kstack_cache;
437 kstack_cache = ks_ce;
438 mtx_unlock(&kstack_cache_mtx);
441 vm_thread_stack_dispose(ksobj, ks, pages);
445 vm_thread_stack_lowmem(void *nulll)
447 struct kstack_cache_entry *ks_ce, *ks_ce1;
449 mtx_lock(&kstack_cache_mtx);
450 ks_ce = kstack_cache;
452 mtx_unlock(&kstack_cache_mtx);
454 while (ks_ce != NULL) {
456 ks_ce = ks_ce->next_ks_entry;
458 vm_thread_stack_dispose(ks_ce1->ksobj, (vm_offset_t)ks_ce1,
464 kstack_cache_init(void *nulll)
467 EVENTHANDLER_REGISTER(vm_lowmem, vm_thread_stack_lowmem, NULL,
468 EVENTHANDLER_PRI_ANY);
471 SYSINIT(vm_kstacks, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY, kstack_cache_init, NULL);
473 #ifdef KSTACK_USAGE_PROF
475 * Track maximum stack used by a thread in kernel.
477 static int max_kstack_used;
479 SYSCTL_INT(_debug, OID_AUTO, max_kstack_used, CTLFLAG_RD,
481 "Maxiumum stack depth used by a thread in kernel");
484 intr_prof_stack_use(struct thread *td, struct trapframe *frame)
486 vm_offset_t stack_top;
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.
495 if (TRAPF_USERMODE(frame))
498 stack_top = td->td_kstack + td->td_kstack_pages * PAGE_SIZE;
499 current = (vm_offset_t)(uintptr_t)&stack_top;
502 * Try to detect if interrupt is using kernel thread stack.
503 * Hardware could use a dedicated stack for interrupt handling.
505 if (stack_top <= current || current < td->td_kstack)
508 used = stack_top - current;
510 prev_used = max_kstack_used;
511 if (prev_used >= used)
513 if (atomic_cmpset_int(&max_kstack_used, prev_used, used))
517 #endif /* KSTACK_USAGE_PROF */
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.
528 vm_forkproc(td, p2, td2, vm2, flags)
535 struct proc *p1 = td->td_proc;
538 if ((flags & RFPROC) == 0) {
540 * Divorce the memory, if it is shared, essentially
541 * this changes shared memory amongst threads, into
544 if ((flags & RFMEM) == 0) {
545 if (p1->p_vmspace->vm_refcnt > 1) {
546 error = vmspace_unshare(p1);
551 cpu_fork(td, p2, td2, flags);
556 p2->p_vmspace = p1->p_vmspace;
557 atomic_add_int(&p1->p_vmspace->vm_refcnt, 1);
560 while (vm_page_count_severe()) {
564 if ((flags & RFMEM) == 0) {
566 if (p1->p_vmspace->vm_shm)
571 * cpu_fork will copy and update the pcb, set up the kernel stack,
572 * and make the child ready to run.
574 cpu_fork(td, p2, td2, flags);
579 * Called after process has been wait(2)'ed upon and is being reaped.
580 * The idea is to reclaim resources that we could not reclaim while
581 * the process was still executing.
588 vmspace_exitfree(p); /* and clean-out the vmspace */