uvm_glue.c revision 1.97 1 1.97 chs /* $NetBSD: uvm_glue.c,v 1.97 2006/10/05 14:48:33 chs Exp $ */
2 1.1 mrg
3 1.48 chs /*
4 1.1 mrg * Copyright (c) 1997 Charles D. Cranor and Washington University.
5 1.48 chs * Copyright (c) 1991, 1993, The Regents of the University of California.
6 1.1 mrg *
7 1.1 mrg * All rights reserved.
8 1.1 mrg *
9 1.1 mrg * This code is derived from software contributed to Berkeley by
10 1.1 mrg * The Mach Operating System project at Carnegie-Mellon University.
11 1.1 mrg *
12 1.1 mrg * Redistribution and use in source and binary forms, with or without
13 1.1 mrg * modification, are permitted provided that the following conditions
14 1.1 mrg * are met:
15 1.1 mrg * 1. Redistributions of source code must retain the above copyright
16 1.1 mrg * notice, this list of conditions and the following disclaimer.
17 1.1 mrg * 2. Redistributions in binary form must reproduce the above copyright
18 1.1 mrg * notice, this list of conditions and the following disclaimer in the
19 1.1 mrg * documentation and/or other materials provided with the distribution.
20 1.1 mrg * 3. All advertising materials mentioning features or use of this software
21 1.1 mrg * must display the following acknowledgement:
22 1.1 mrg * This product includes software developed by Charles D. Cranor,
23 1.48 chs * Washington University, the University of California, Berkeley and
24 1.1 mrg * its contributors.
25 1.1 mrg * 4. Neither the name of the University nor the names of its contributors
26 1.1 mrg * may be used to endorse or promote products derived from this software
27 1.1 mrg * without specific prior written permission.
28 1.1 mrg *
29 1.1 mrg * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 1.1 mrg * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 1.1 mrg * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 1.1 mrg * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 1.1 mrg * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 1.1 mrg * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 1.1 mrg * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 1.1 mrg * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 1.1 mrg * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 1.1 mrg * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 1.1 mrg * SUCH DAMAGE.
40 1.1 mrg *
41 1.1 mrg * @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
42 1.4 mrg * from: Id: uvm_glue.c,v 1.1.2.8 1998/02/07 01:16:54 chs Exp
43 1.1 mrg *
44 1.1 mrg *
45 1.1 mrg * Copyright (c) 1987, 1990 Carnegie-Mellon University.
46 1.1 mrg * All rights reserved.
47 1.48 chs *
48 1.1 mrg * Permission to use, copy, modify and distribute this software and
49 1.1 mrg * its documentation is hereby granted, provided that both the copyright
50 1.1 mrg * notice and this permission notice appear in all copies of the
51 1.1 mrg * software, derivative works or modified versions, and any portions
52 1.1 mrg * thereof, and that both notices appear in supporting documentation.
53 1.48 chs *
54 1.48 chs * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
55 1.48 chs * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
56 1.1 mrg * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
57 1.48 chs *
58 1.1 mrg * Carnegie Mellon requests users of this software to return to
59 1.1 mrg *
60 1.1 mrg * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
61 1.1 mrg * School of Computer Science
62 1.1 mrg * Carnegie Mellon University
63 1.1 mrg * Pittsburgh PA 15213-3890
64 1.1 mrg *
65 1.1 mrg * any improvements or extensions that they make and grant Carnegie the
66 1.1 mrg * rights to redistribute these changes.
67 1.1 mrg */
68 1.55 lukem
69 1.55 lukem #include <sys/cdefs.h>
70 1.97 chs __KERNEL_RCSID(0, "$NetBSD: uvm_glue.c,v 1.97 2006/10/05 14:48:33 chs Exp $");
71 1.1 mrg
72 1.96 matt #include "opt_coredump.h"
73 1.49 lukem #include "opt_kgdb.h"
74 1.59 yamt #include "opt_kstack.h"
75 1.5 mrg #include "opt_uvmhist.h"
76 1.5 mrg
77 1.1 mrg /*
78 1.1 mrg * uvm_glue.c: glue functions
79 1.1 mrg */
80 1.1 mrg
81 1.1 mrg #include <sys/param.h>
82 1.1 mrg #include <sys/systm.h>
83 1.1 mrg #include <sys/proc.h>
84 1.1 mrg #include <sys/resourcevar.h>
85 1.1 mrg #include <sys/buf.h>
86 1.1 mrg #include <sys/user.h>
87 1.1 mrg
88 1.1 mrg #include <uvm/uvm.h>
89 1.1 mrg
90 1.1 mrg #include <machine/cpu.h>
91 1.1 mrg
92 1.1 mrg /*
93 1.1 mrg * local prototypes
94 1.1 mrg */
95 1.1 mrg
96 1.78 junyoung static void uvm_swapout(struct lwp *);
97 1.1 mrg
98 1.60 chs #define UVM_NUAREA_MAX 16
99 1.94 yamt static vaddr_t uvm_uareas;
100 1.94 yamt static int uvm_nuarea;
101 1.94 yamt static struct simplelock uvm_uareas_slock = SIMPLELOCK_INITIALIZER;
102 1.94 yamt #define UAREA_NEXTFREE(uarea) (*(vaddr_t *)(UAREA_TO_USER(uarea)))
103 1.60 chs
104 1.75 jdolecek static void uvm_uarea_free(vaddr_t);
105 1.75 jdolecek
106 1.1 mrg /*
107 1.1 mrg * XXXCDC: do these really belong here?
108 1.1 mrg */
109 1.1 mrg
110 1.28 thorpej /*
111 1.1 mrg * uvm_kernacc: can the kernel access a region of memory
112 1.1 mrg *
113 1.83 yamt * - used only by /dev/kmem driver (mem.c)
114 1.1 mrg */
115 1.1 mrg
116 1.6 mrg boolean_t
117 1.89 thorpej uvm_kernacc(caddr_t addr, size_t len, int rw)
118 1.6 mrg {
119 1.6 mrg boolean_t rv;
120 1.13 eeh vaddr_t saddr, eaddr;
121 1.6 mrg vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
122 1.6 mrg
123 1.31 kleink saddr = trunc_page((vaddr_t)addr);
124 1.43 chs eaddr = round_page((vaddr_t)addr + len);
125 1.6 mrg vm_map_lock_read(kernel_map);
126 1.6 mrg rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
127 1.6 mrg vm_map_unlock_read(kernel_map);
128 1.6 mrg
129 1.6 mrg return(rv);
130 1.1 mrg }
131 1.1 mrg
132 1.1 mrg #ifdef KGDB
133 1.1 mrg /*
134 1.1 mrg * Change protections on kernel pages from addr to addr+len
135 1.1 mrg * (presumably so debugger can plant a breakpoint).
136 1.1 mrg *
137 1.1 mrg * We force the protection change at the pmap level. If we were
138 1.1 mrg * to use vm_map_protect a change to allow writing would be lazily-
139 1.1 mrg * applied meaning we would still take a protection fault, something
140 1.1 mrg * we really don't want to do. It would also fragment the kernel
141 1.1 mrg * map unnecessarily. We cannot use pmap_protect since it also won't
142 1.1 mrg * enforce a write-enable request. Using pmap_enter is the only way
143 1.1 mrg * we can ensure the change takes place properly.
144 1.1 mrg */
145 1.6 mrg void
146 1.89 thorpej uvm_chgkprot(caddr_t addr, size_t len, int rw)
147 1.6 mrg {
148 1.6 mrg vm_prot_t prot;
149 1.13 eeh paddr_t pa;
150 1.13 eeh vaddr_t sva, eva;
151 1.6 mrg
152 1.6 mrg prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
153 1.31 kleink eva = round_page((vaddr_t)addr + len);
154 1.31 kleink for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
155 1.6 mrg /*
156 1.6 mrg * Extract physical address for the page.
157 1.6 mrg */
158 1.27 thorpej if (pmap_extract(pmap_kernel(), sva, &pa) == FALSE)
159 1.6 mrg panic("chgkprot: invalid page");
160 1.30 thorpej pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
161 1.6 mrg }
162 1.51 chris pmap_update(pmap_kernel());
163 1.1 mrg }
164 1.1 mrg #endif
165 1.1 mrg
166 1.1 mrg /*
167 1.52 chs * uvm_vslock: wire user memory for I/O
168 1.1 mrg *
169 1.1 mrg * - called from physio and sys___sysctl
170 1.1 mrg * - XXXCDC: consider nuking this (or making it a macro?)
171 1.1 mrg */
172 1.1 mrg
173 1.26 thorpej int
174 1.97 chs uvm_vslock(struct vmspace *vs, void *addr, size_t len, vm_prot_t access_type)
175 1.1 mrg {
176 1.50 chs struct vm_map *map;
177 1.26 thorpej vaddr_t start, end;
178 1.45 chs int error;
179 1.26 thorpej
180 1.97 chs map = &vs->vm_map;
181 1.31 kleink start = trunc_page((vaddr_t)addr);
182 1.31 kleink end = round_page((vaddr_t)addr + len);
183 1.93 drochner error = uvm_fault_wire(map, start, end, access_type, 0);
184 1.45 chs return error;
185 1.1 mrg }
186 1.1 mrg
187 1.1 mrg /*
188 1.52 chs * uvm_vsunlock: unwire user memory wired by uvm_vslock()
189 1.1 mrg *
190 1.1 mrg * - called from physio and sys___sysctl
191 1.1 mrg * - XXXCDC: consider nuking this (or making it a macro?)
192 1.1 mrg */
193 1.1 mrg
194 1.6 mrg void
195 1.97 chs uvm_vsunlock(struct vmspace *vs, void *addr, size_t len)
196 1.1 mrg {
197 1.97 chs uvm_fault_unwire(&vs->vm_map, trunc_page((vaddr_t)addr),
198 1.43 chs round_page((vaddr_t)addr + len));
199 1.1 mrg }
200 1.1 mrg
201 1.1 mrg /*
202 1.62 thorpej * uvm_proc_fork: fork a virtual address space
203 1.1 mrg *
204 1.1 mrg * - the address space is copied as per parent map's inherit values
205 1.62 thorpej */
206 1.62 thorpej void
207 1.89 thorpej uvm_proc_fork(struct proc *p1, struct proc *p2, boolean_t shared)
208 1.62 thorpej {
209 1.62 thorpej
210 1.62 thorpej if (shared == TRUE) {
211 1.62 thorpej p2->p_vmspace = NULL;
212 1.62 thorpej uvmspace_share(p1, p2);
213 1.62 thorpej } else {
214 1.62 thorpej p2->p_vmspace = uvmspace_fork(p1->p_vmspace);
215 1.62 thorpej }
216 1.62 thorpej
217 1.62 thorpej cpu_proc_fork(p1, p2);
218 1.62 thorpej }
219 1.62 thorpej
220 1.62 thorpej
221 1.62 thorpej /*
222 1.62 thorpej * uvm_lwp_fork: fork a thread
223 1.62 thorpej *
224 1.1 mrg * - a new "user" structure is allocated for the child process
225 1.1 mrg * [filled in by MD layer...]
226 1.20 thorpej * - if specified, the child gets a new user stack described by
227 1.20 thorpej * stack and stacksize
228 1.1 mrg * - NOTE: the kernel stack may be at a different location in the child
229 1.1 mrg * process, and thus addresses of automatic variables may be invalid
230 1.62 thorpej * after cpu_lwp_fork returns in the child process. We do nothing here
231 1.62 thorpej * after cpu_lwp_fork returns.
232 1.1 mrg * - XXXCDC: we need a way for this to return a failure value rather
233 1.1 mrg * than just hang
234 1.1 mrg */
235 1.6 mrg void
236 1.89 thorpej uvm_lwp_fork(struct lwp *l1, struct lwp *l2, void *stack, size_t stacksize,
237 1.89 thorpej void (*func)(void *), void *arg)
238 1.6 mrg {
239 1.45 chs int error;
240 1.6 mrg
241 1.6 mrg /*
242 1.7 thorpej * Wire down the U-area for the process, which contains the PCB
243 1.62 thorpej * and the kernel stack. Wired state is stored in l->l_flag's
244 1.62 thorpej * L_INMEM bit rather than in the vm_map_entry's wired count
245 1.61 chs * to prevent kernel_map fragmentation. If we reused a cached U-area,
246 1.62 thorpej * L_INMEM will already be set and we don't need to do anything.
247 1.21 thorpej *
248 1.61 chs * Note the kernel stack gets read/write accesses right off the bat.
249 1.6 mrg */
250 1.61 chs
251 1.62 thorpej if ((l2->l_flag & L_INMEM) == 0) {
252 1.94 yamt vaddr_t uarea = USER_TO_UAREA(l2->l_addr);
253 1.94 yamt
254 1.94 yamt error = uvm_fault_wire(kernel_map, uarea,
255 1.94 yamt uarea + USPACE, VM_PROT_READ | VM_PROT_WRITE, 0);
256 1.61 chs if (error)
257 1.62 thorpej panic("uvm_lwp_fork: uvm_fault_wire failed: %d", error);
258 1.67 scw #ifdef PMAP_UAREA
259 1.67 scw /* Tell the pmap this is a u-area mapping */
260 1.94 yamt PMAP_UAREA(uarea);
261 1.67 scw #endif
262 1.62 thorpej l2->l_flag |= L_INMEM;
263 1.61 chs }
264 1.59 yamt
265 1.59 yamt #ifdef KSTACK_CHECK_MAGIC
266 1.59 yamt /*
267 1.59 yamt * fill stack with magic number
268 1.59 yamt */
269 1.63 yamt kstack_setup_magic(l2);
270 1.59 yamt #endif
271 1.6 mrg
272 1.6 mrg /*
273 1.62 thorpej * cpu_lwp_fork() copy and update the pcb, and make the child ready
274 1.62 thorpej * to run. If this is a normal user fork, the child will exit
275 1.34 thorpej * directly to user mode via child_return() on its first time
276 1.34 thorpej * slice and will not return here. If this is a kernel thread,
277 1.34 thorpej * the specified entry point will be executed.
278 1.6 mrg */
279 1.62 thorpej cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
280 1.14 thorpej }
281 1.14 thorpej
282 1.14 thorpej /*
283 1.60 chs * uvm_uarea_alloc: allocate a u-area
284 1.60 chs */
285 1.60 chs
286 1.61 chs boolean_t
287 1.61 chs uvm_uarea_alloc(vaddr_t *uaddrp)
288 1.60 chs {
289 1.60 chs vaddr_t uaddr;
290 1.60 chs
291 1.60 chs #ifndef USPACE_ALIGN
292 1.60 chs #define USPACE_ALIGN 0
293 1.60 chs #endif
294 1.60 chs
295 1.62 thorpej simple_lock(&uvm_uareas_slock);
296 1.75 jdolecek if (uvm_nuarea > 0) {
297 1.94 yamt uaddr = uvm_uareas;
298 1.94 yamt uvm_uareas = UAREA_NEXTFREE(uaddr);
299 1.60 chs uvm_nuarea--;
300 1.62 thorpej simple_unlock(&uvm_uareas_slock);
301 1.61 chs *uaddrp = uaddr;
302 1.61 chs return TRUE;
303 1.60 chs } else {
304 1.62 thorpej simple_unlock(&uvm_uareas_slock);
305 1.84 yamt *uaddrp = uvm_km_alloc(kernel_map, USPACE, USPACE_ALIGN,
306 1.84 yamt UVM_KMF_PAGEABLE);
307 1.61 chs return FALSE;
308 1.60 chs }
309 1.60 chs }
310 1.60 chs
311 1.60 chs /*
312 1.75 jdolecek * uvm_uarea_free: free a u-area; never blocks
313 1.75 jdolecek */
314 1.75 jdolecek
315 1.92 perry static inline void
316 1.75 jdolecek uvm_uarea_free(vaddr_t uaddr)
317 1.75 jdolecek {
318 1.75 jdolecek simple_lock(&uvm_uareas_slock);
319 1.94 yamt UAREA_NEXTFREE(uaddr) = uvm_uareas;
320 1.94 yamt uvm_uareas = uaddr;
321 1.75 jdolecek uvm_nuarea++;
322 1.75 jdolecek simple_unlock(&uvm_uareas_slock);
323 1.75 jdolecek }
324 1.75 jdolecek
325 1.75 jdolecek /*
326 1.75 jdolecek * uvm_uarea_drain: return memory of u-areas over limit
327 1.75 jdolecek * back to system
328 1.60 chs */
329 1.60 chs
330 1.60 chs void
331 1.75 jdolecek uvm_uarea_drain(boolean_t empty)
332 1.60 chs {
333 1.75 jdolecek int leave = empty ? 0 : UVM_NUAREA_MAX;
334 1.75 jdolecek vaddr_t uaddr;
335 1.75 jdolecek
336 1.75 jdolecek if (uvm_nuarea <= leave)
337 1.75 jdolecek return;
338 1.60 chs
339 1.62 thorpej simple_lock(&uvm_uareas_slock);
340 1.75 jdolecek while(uvm_nuarea > leave) {
341 1.94 yamt uaddr = uvm_uareas;
342 1.94 yamt uvm_uareas = UAREA_NEXTFREE(uaddr);
343 1.75 jdolecek uvm_nuarea--;
344 1.62 thorpej simple_unlock(&uvm_uareas_slock);
345 1.84 yamt uvm_km_free(kernel_map, uaddr, USPACE, UVM_KMF_PAGEABLE);
346 1.75 jdolecek simple_lock(&uvm_uareas_slock);
347 1.60 chs }
348 1.75 jdolecek simple_unlock(&uvm_uareas_slock);
349 1.60 chs }
350 1.60 chs
351 1.60 chs /*
352 1.80 pk * uvm_exit: exit a virtual address space
353 1.80 pk *
354 1.80 pk * - the process passed to us is a dead (pre-zombie) process; we
355 1.80 pk * are running on a different context now (the reaper).
356 1.80 pk * - borrow proc0's address space because freeing the vmspace
357 1.80 pk * of the dead process may block.
358 1.80 pk */
359 1.80 pk
360 1.80 pk void
361 1.89 thorpej uvm_proc_exit(struct proc *p)
362 1.80 pk {
363 1.80 pk struct lwp *l = curlwp; /* XXX */
364 1.80 pk struct vmspace *ovm;
365 1.80 pk
366 1.80 pk KASSERT(p == l->l_proc);
367 1.80 pk ovm = p->p_vmspace;
368 1.80 pk
369 1.80 pk /*
370 1.80 pk * borrow proc0's address space.
371 1.80 pk */
372 1.80 pk pmap_deactivate(l);
373 1.80 pk p->p_vmspace = proc0.p_vmspace;
374 1.80 pk pmap_activate(l);
375 1.80 pk
376 1.80 pk uvmspace_free(ovm);
377 1.80 pk }
378 1.80 pk
379 1.80 pk void
380 1.80 pk uvm_lwp_exit(struct lwp *l)
381 1.80 pk {
382 1.94 yamt vaddr_t va = USER_TO_UAREA(l->l_addr);
383 1.80 pk
384 1.80 pk l->l_flag &= ~L_INMEM;
385 1.80 pk uvm_uarea_free(va);
386 1.80 pk l->l_addr = NULL;
387 1.80 pk }
388 1.80 pk
389 1.80 pk /*
390 1.1 mrg * uvm_init_limit: init per-process VM limits
391 1.1 mrg *
392 1.1 mrg * - called for process 0 and then inherited by all others.
393 1.1 mrg */
394 1.60 chs
395 1.6 mrg void
396 1.89 thorpej uvm_init_limits(struct proc *p)
397 1.6 mrg {
398 1.6 mrg
399 1.6 mrg /*
400 1.6 mrg * Set up the initial limits on process VM. Set the maximum
401 1.6 mrg * resident set size to be all of (reasonably) available memory.
402 1.6 mrg * This causes any single, large process to start random page
403 1.6 mrg * replacement once it fills memory.
404 1.6 mrg */
405 1.6 mrg
406 1.6 mrg p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
407 1.79 pk p->p_rlimit[RLIMIT_STACK].rlim_max = maxsmap;
408 1.6 mrg p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
409 1.79 pk p->p_rlimit[RLIMIT_DATA].rlim_max = maxdmap;
410 1.6 mrg p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free);
411 1.1 mrg }
412 1.1 mrg
413 1.1 mrg #ifdef DEBUG
414 1.1 mrg int enableswap = 1;
415 1.1 mrg int swapdebug = 0;
416 1.1 mrg #define SDB_FOLLOW 1
417 1.1 mrg #define SDB_SWAPIN 2
418 1.1 mrg #define SDB_SWAPOUT 4
419 1.1 mrg #endif
420 1.1 mrg
421 1.1 mrg /*
422 1.95 yamt * uvm_swapin: swap in an lwp's u-area.
423 1.1 mrg */
424 1.1 mrg
425 1.6 mrg void
426 1.89 thorpej uvm_swapin(struct lwp *l)
427 1.6 mrg {
428 1.13 eeh vaddr_t addr;
429 1.52 chs int s, error;
430 1.6 mrg
431 1.94 yamt addr = USER_TO_UAREA(l->l_addr);
432 1.62 thorpej /* make L_INMEM true */
433 1.93 drochner error = uvm_fault_wire(kernel_map, addr, addr + USPACE,
434 1.93 drochner VM_PROT_READ | VM_PROT_WRITE, 0);
435 1.52 chs if (error) {
436 1.52 chs panic("uvm_swapin: rewiring stack failed: %d", error);
437 1.52 chs }
438 1.6 mrg
439 1.6 mrg /*
440 1.6 mrg * Some architectures need to be notified when the user area has
441 1.6 mrg * moved to new physical page(s) (e.g. see mips/mips/vm_machdep.c).
442 1.6 mrg */
443 1.62 thorpej cpu_swapin(l);
444 1.41 enami SCHED_LOCK(s);
445 1.62 thorpej if (l->l_stat == LSRUN)
446 1.62 thorpej setrunqueue(l);
447 1.62 thorpej l->l_flag |= L_INMEM;
448 1.41 enami SCHED_UNLOCK(s);
449 1.62 thorpej l->l_swtime = 0;
450 1.6 mrg ++uvmexp.swapins;
451 1.1 mrg }
452 1.1 mrg
453 1.1 mrg /*
454 1.1 mrg * uvm_scheduler: process zero main loop
455 1.1 mrg *
456 1.1 mrg * - attempt to swapin every swaped-out, runnable process in order of
457 1.1 mrg * priority.
458 1.1 mrg * - if not enough memory, wake the pagedaemon and let it clear space.
459 1.1 mrg */
460 1.1 mrg
461 1.6 mrg void
462 1.89 thorpej uvm_scheduler(void)
463 1.1 mrg {
464 1.62 thorpej struct lwp *l, *ll;
465 1.32 augustss int pri;
466 1.6 mrg int ppri;
467 1.1 mrg
468 1.1 mrg loop:
469 1.1 mrg #ifdef DEBUG
470 1.6 mrg while (!enableswap)
471 1.43 chs tsleep(&proc0, PVM, "noswap", 0);
472 1.1 mrg #endif
473 1.62 thorpej ll = NULL; /* process to choose */
474 1.6 mrg ppri = INT_MIN; /* its priority */
475 1.29 thorpej proclist_lock_read();
476 1.6 mrg
477 1.62 thorpej LIST_FOREACH(l, &alllwp, l_list) {
478 1.6 mrg /* is it a runnable swapped out process? */
479 1.62 thorpej if (l->l_stat == LSRUN && (l->l_flag & L_INMEM) == 0) {
480 1.62 thorpej pri = l->l_swtime + l->l_slptime -
481 1.62 thorpej (l->l_proc->p_nice - NZERO) * 8;
482 1.6 mrg if (pri > ppri) { /* higher priority? remember it. */
483 1.62 thorpej ll = l;
484 1.6 mrg ppri = pri;
485 1.6 mrg }
486 1.6 mrg }
487 1.6 mrg }
488 1.39 sommerfe /*
489 1.39 sommerfe * XXXSMP: possible unlock/sleep race between here and the
490 1.39 sommerfe * "scheduler" tsleep below..
491 1.39 sommerfe */
492 1.28 thorpej proclist_unlock_read();
493 1.1 mrg
494 1.1 mrg #ifdef DEBUG
495 1.6 mrg if (swapdebug & SDB_FOLLOW)
496 1.62 thorpej printf("scheduler: running, procp %p pri %d\n", ll, ppri);
497 1.1 mrg #endif
498 1.6 mrg /*
499 1.6 mrg * Nothing to do, back to sleep
500 1.6 mrg */
501 1.62 thorpej if ((l = ll) == NULL) {
502 1.43 chs tsleep(&proc0, PVM, "scheduler", 0);
503 1.6 mrg goto loop;
504 1.6 mrg }
505 1.6 mrg
506 1.6 mrg /*
507 1.6 mrg * we have found swapped out process which we would like to bring
508 1.6 mrg * back in.
509 1.6 mrg *
510 1.6 mrg * XXX: this part is really bogus cuz we could deadlock on memory
511 1.6 mrg * despite our feeble check
512 1.6 mrg */
513 1.6 mrg if (uvmexp.free > atop(USPACE)) {
514 1.1 mrg #ifdef DEBUG
515 1.6 mrg if (swapdebug & SDB_SWAPIN)
516 1.6 mrg printf("swapin: pid %d(%s)@%p, pri %d free %d\n",
517 1.62 thorpej l->l_proc->p_pid, l->l_proc->p_comm, l->l_addr, ppri, uvmexp.free);
518 1.1 mrg #endif
519 1.62 thorpej uvm_swapin(l);
520 1.6 mrg goto loop;
521 1.6 mrg }
522 1.6 mrg /*
523 1.6 mrg * not enough memory, jab the pageout daemon and wait til the coast
524 1.6 mrg * is clear
525 1.6 mrg */
526 1.1 mrg #ifdef DEBUG
527 1.6 mrg if (swapdebug & SDB_FOLLOW)
528 1.6 mrg printf("scheduler: no room for pid %d(%s), free %d\n",
529 1.62 thorpej l->l_proc->p_pid, l->l_proc->p_comm, uvmexp.free);
530 1.1 mrg #endif
531 1.6 mrg uvm_wait("schedpwait");
532 1.1 mrg #ifdef DEBUG
533 1.6 mrg if (swapdebug & SDB_FOLLOW)
534 1.6 mrg printf("scheduler: room again, free %d\n", uvmexp.free);
535 1.1 mrg #endif
536 1.6 mrg goto loop;
537 1.1 mrg }
538 1.1 mrg
539 1.1 mrg /*
540 1.62 thorpej * swappable: is LWP "l" swappable?
541 1.1 mrg */
542 1.1 mrg
543 1.62 thorpej #define swappable(l) \
544 1.62 thorpej (((l)->l_flag & (L_INMEM)) && \
545 1.62 thorpej ((((l)->l_proc->p_flag) & (P_SYSTEM | P_WEXIT)) == 0) && \
546 1.62 thorpej (l)->l_holdcnt == 0)
547 1.1 mrg
548 1.1 mrg /*
549 1.1 mrg * swapout_threads: find threads that can be swapped and unwire their
550 1.1 mrg * u-areas.
551 1.1 mrg *
552 1.1 mrg * - called by the pagedaemon
553 1.1 mrg * - try and swap at least one processs
554 1.1 mrg * - processes that are sleeping or stopped for maxslp or more seconds
555 1.1 mrg * are swapped... otherwise the longest-sleeping or stopped process
556 1.1 mrg * is swapped, otherwise the longest resident process...
557 1.1 mrg */
558 1.60 chs
559 1.6 mrg void
560 1.89 thorpej uvm_swapout_threads(void)
561 1.1 mrg {
562 1.62 thorpej struct lwp *l;
563 1.62 thorpej struct lwp *outl, *outl2;
564 1.6 mrg int outpri, outpri2;
565 1.6 mrg int didswap = 0;
566 1.48 chs extern int maxslp;
567 1.6 mrg /* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */
568 1.1 mrg
569 1.1 mrg #ifdef DEBUG
570 1.6 mrg if (!enableswap)
571 1.6 mrg return;
572 1.1 mrg #endif
573 1.1 mrg
574 1.6 mrg /*
575 1.62 thorpej * outl/outpri : stop/sleep thread with largest sleeptime < maxslp
576 1.62 thorpej * outl2/outpri2: the longest resident thread (its swap time)
577 1.6 mrg */
578 1.62 thorpej outl = outl2 = NULL;
579 1.6 mrg outpri = outpri2 = 0;
580 1.29 thorpej proclist_lock_read();
581 1.62 thorpej LIST_FOREACH(l, &alllwp, l_list) {
582 1.81 yamt KASSERT(l->l_proc != NULL);
583 1.62 thorpej if (!swappable(l))
584 1.6 mrg continue;
585 1.62 thorpej switch (l->l_stat) {
586 1.68 cl case LSONPROC:
587 1.69 cl continue;
588 1.69 cl
589 1.62 thorpej case LSRUN:
590 1.62 thorpej if (l->l_swtime > outpri2) {
591 1.62 thorpej outl2 = l;
592 1.62 thorpej outpri2 = l->l_swtime;
593 1.6 mrg }
594 1.6 mrg continue;
595 1.48 chs
596 1.62 thorpej case LSSLEEP:
597 1.62 thorpej case LSSTOP:
598 1.62 thorpej if (l->l_slptime >= maxslp) {
599 1.62 thorpej uvm_swapout(l);
600 1.6 mrg didswap++;
601 1.62 thorpej } else if (l->l_slptime > outpri) {
602 1.62 thorpej outl = l;
603 1.62 thorpej outpri = l->l_slptime;
604 1.6 mrg }
605 1.6 mrg continue;
606 1.6 mrg }
607 1.6 mrg }
608 1.28 thorpej proclist_unlock_read();
609 1.6 mrg
610 1.6 mrg /*
611 1.6 mrg * If we didn't get rid of any real duds, toss out the next most
612 1.6 mrg * likely sleeping/stopped or running candidate. We only do this
613 1.6 mrg * if we are real low on memory since we don't gain much by doing
614 1.6 mrg * it (USPACE bytes).
615 1.6 mrg */
616 1.6 mrg if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE))) {
617 1.62 thorpej if ((l = outl) == NULL)
618 1.62 thorpej l = outl2;
619 1.1 mrg #ifdef DEBUG
620 1.6 mrg if (swapdebug & SDB_SWAPOUT)
621 1.62 thorpej printf("swapout_threads: no duds, try procp %p\n", l);
622 1.1 mrg #endif
623 1.62 thorpej if (l)
624 1.62 thorpej uvm_swapout(l);
625 1.6 mrg }
626 1.1 mrg }
627 1.1 mrg
628 1.1 mrg /*
629 1.62 thorpej * uvm_swapout: swap out lwp "l"
630 1.1 mrg *
631 1.48 chs * - currently "swapout" means "unwire U-area" and "pmap_collect()"
632 1.1 mrg * the pmap.
633 1.1 mrg * - XXXCDC: should deactivate all process' private anonymous memory
634 1.1 mrg */
635 1.1 mrg
636 1.6 mrg static void
637 1.89 thorpej uvm_swapout(struct lwp *l)
638 1.1 mrg {
639 1.13 eeh vaddr_t addr;
640 1.6 mrg int s;
641 1.62 thorpej struct proc *p = l->l_proc;
642 1.1 mrg
643 1.1 mrg #ifdef DEBUG
644 1.6 mrg if (swapdebug & SDB_SWAPOUT)
645 1.62 thorpej printf("swapout: lid %d.%d(%s)@%p, stat %x pri %d free %d\n",
646 1.62 thorpej p->p_pid, l->l_lid, p->p_comm, l->l_addr, l->l_stat,
647 1.62 thorpej l->l_slptime, uvmexp.free);
648 1.1 mrg #endif
649 1.1 mrg
650 1.6 mrg /*
651 1.6 mrg * Mark it as (potentially) swapped out.
652 1.6 mrg */
653 1.41 enami SCHED_LOCK(s);
654 1.69 cl if (l->l_stat == LSONPROC) {
655 1.69 cl KDASSERT(l->l_cpu != curcpu());
656 1.68 cl SCHED_UNLOCK(s);
657 1.68 cl return;
658 1.68 cl }
659 1.62 thorpej l->l_flag &= ~L_INMEM;
660 1.62 thorpej if (l->l_stat == LSRUN)
661 1.62 thorpej remrunqueue(l);
662 1.41 enami SCHED_UNLOCK(s);
663 1.62 thorpej l->l_swtime = 0;
664 1.53 chs p->p_stats->p_ru.ru_nswap++;
665 1.6 mrg ++uvmexp.swapouts;
666 1.68 cl
667 1.68 cl /*
668 1.68 cl * Do any machine-specific actions necessary before swapout.
669 1.68 cl * This can include saving floating point state, etc.
670 1.68 cl */
671 1.68 cl cpu_swapout(l);
672 1.43 chs
673 1.43 chs /*
674 1.43 chs * Unwire the to-be-swapped process's user struct and kernel stack.
675 1.43 chs */
676 1.94 yamt addr = USER_TO_UAREA(l->l_addr);
677 1.62 thorpej uvm_fault_unwire(kernel_map, addr, addr + USPACE); /* !L_INMEM */
678 1.43 chs pmap_collect(vm_map_pmap(&p->p_vmspace->vm_map));
679 1.1 mrg }
680 1.1 mrg
681 1.96 matt #ifdef COREDUMP
682 1.56 thorpej /*
683 1.56 thorpej * uvm_coredump_walkmap: walk a process's map for the purpose of dumping
684 1.56 thorpej * a core file.
685 1.56 thorpej */
686 1.56 thorpej
687 1.56 thorpej int
688 1.89 thorpej uvm_coredump_walkmap(struct proc *p, void *iocookie,
689 1.89 thorpej int (*func)(struct proc *, void *, struct uvm_coredump_state *),
690 1.89 thorpej void *cookie)
691 1.56 thorpej {
692 1.56 thorpej struct uvm_coredump_state state;
693 1.56 thorpej struct vmspace *vm = p->p_vmspace;
694 1.56 thorpej struct vm_map *map = &vm->vm_map;
695 1.56 thorpej struct vm_map_entry *entry;
696 1.56 thorpej int error;
697 1.56 thorpej
698 1.64 atatat entry = NULL;
699 1.64 atatat vm_map_lock_read(map);
700 1.87 matt state.end = 0;
701 1.64 atatat for (;;) {
702 1.64 atatat if (entry == NULL)
703 1.64 atatat entry = map->header.next;
704 1.64 atatat else if (!uvm_map_lookup_entry(map, state.end, &entry))
705 1.64 atatat entry = entry->next;
706 1.64 atatat if (entry == &map->header)
707 1.64 atatat break;
708 1.64 atatat
709 1.56 thorpej state.cookie = cookie;
710 1.86 matt if (state.end > entry->start) {
711 1.86 matt state.start = state.end;
712 1.86 matt } else {
713 1.86 matt state.start = entry->start;
714 1.86 matt }
715 1.86 matt state.realend = entry->end;
716 1.56 thorpej state.end = entry->end;
717 1.56 thorpej state.prot = entry->protection;
718 1.56 thorpej state.flags = 0;
719 1.56 thorpej
720 1.82 chs /*
721 1.82 chs * Dump the region unless one of the following is true:
722 1.82 chs *
723 1.82 chs * (1) the region has neither object nor amap behind it
724 1.82 chs * (ie. it has never been accessed).
725 1.82 chs *
726 1.82 chs * (2) the region has no amap and is read-only
727 1.82 chs * (eg. an executable text section).
728 1.82 chs *
729 1.82 chs * (3) the region's object is a device.
730 1.85 nathanw *
731 1.85 nathanw * (4) the region is unreadable by the process.
732 1.82 chs */
733 1.56 thorpej
734 1.82 chs KASSERT(!UVM_ET_ISSUBMAP(entry));
735 1.82 chs KASSERT(state.start < VM_MAXUSER_ADDRESS);
736 1.82 chs KASSERT(state.end <= VM_MAXUSER_ADDRESS);
737 1.82 chs if (entry->object.uvm_obj == NULL &&
738 1.82 chs entry->aref.ar_amap == NULL) {
739 1.86 matt state.realend = state.start;
740 1.86 matt } else if ((entry->protection & VM_PROT_WRITE) == 0 &&
741 1.82 chs entry->aref.ar_amap == NULL) {
742 1.86 matt state.realend = state.start;
743 1.86 matt } else if (entry->object.uvm_obj != NULL &&
744 1.82 chs UVM_OBJ_IS_DEVICE(entry->object.uvm_obj)) {
745 1.86 matt state.realend = state.start;
746 1.86 matt } else if ((entry->protection & VM_PROT_READ) == 0) {
747 1.86 matt state.realend = state.start;
748 1.86 matt } else {
749 1.86 matt if (state.start >= (vaddr_t)vm->vm_maxsaddr)
750 1.86 matt state.flags |= UVM_COREDUMP_STACK;
751 1.86 matt
752 1.86 matt /*
753 1.86 matt * If this an anonymous entry, only dump instantiated
754 1.86 matt * pages.
755 1.86 matt */
756 1.86 matt if (entry->object.uvm_obj == NULL) {
757 1.86 matt vaddr_t end;
758 1.86 matt
759 1.86 matt amap_lock(entry->aref.ar_amap);
760 1.86 matt for (end = state.start;
761 1.86 matt end < state.end; end += PAGE_SIZE) {
762 1.86 matt struct vm_anon *anon;
763 1.86 matt anon = amap_lookup(&entry->aref,
764 1.86 matt end - entry->start);
765 1.86 matt /*
766 1.86 matt * If we have already encountered an
767 1.86 matt * uninstantiated page, stop at the
768 1.86 matt * first instantied page.
769 1.86 matt */
770 1.86 matt if (anon != NULL &&
771 1.86 matt state.realend != state.end) {
772 1.86 matt state.end = end;
773 1.86 matt break;
774 1.86 matt }
775 1.86 matt
776 1.86 matt /*
777 1.86 matt * If this page is the first
778 1.86 matt * uninstantiated page, mark this as
779 1.86 matt * the real ending point. Continue to
780 1.86 matt * counting uninstantiated pages.
781 1.86 matt */
782 1.86 matt if (anon == NULL &&
783 1.86 matt state.realend == state.end) {
784 1.86 matt state.realend = end;
785 1.86 matt }
786 1.86 matt }
787 1.86 matt amap_unlock(entry->aref.ar_amap);
788 1.86 matt }
789 1.82 chs }
790 1.86 matt
791 1.56 thorpej
792 1.64 atatat vm_map_unlock_read(map);
793 1.88 matt error = (*func)(p, iocookie, &state);
794 1.56 thorpej if (error)
795 1.56 thorpej return (error);
796 1.64 atatat vm_map_lock_read(map);
797 1.56 thorpej }
798 1.64 atatat vm_map_unlock_read(map);
799 1.56 thorpej
800 1.56 thorpej return (0);
801 1.56 thorpej }
802 1.96 matt #endif /* COREDUMP */
803