1 /* $OpenBSD: uvm_glue.c,v 1.47 2007/05/26 20:26:51 pedro Exp $ */
2 /* $NetBSD: uvm_glue.c,v 1.44 2001/02/06 19:54:44 eeh Exp $ */
3
4 /*
5 * Copyright (c) 1997 Charles D. Cranor and Washington University.
6 * Copyright (c) 1991, 1993, The Regents of the University of California.
7 *
8 * All rights reserved.
9 *
10 * This code is derived from software contributed to Berkeley by
11 * The Mach Operating System project at Carnegie-Mellon University.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by Charles D. Cranor,
24 * Washington University, the University of California, Berkeley and
25 * its contributors.
26 * 4. Neither the name of the University nor the names of its contributors
27 * may be used to endorse or promote products derived from this software
28 * without specific prior written permission.
29 *
30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * SUCH DAMAGE.
41 *
42 * @(#)vm_glue.c 8.6 (Berkeley) 1/5/94
43 * from: Id: uvm_glue.c,v 1.1.2.8 1998/02/07 01:16:54 chs Exp
44 *
45 *
46 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
47 * All rights reserved.
48 *
49 * Permission to use, copy, modify and distribute this software and
50 * its documentation is hereby granted, provided that both the copyright
51 * notice and this permission notice appear in all copies of the
52 * software, derivative works or modified versions, and any portions
53 * thereof, and that both notices appear in supporting documentation.
54 *
55 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
56 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
57 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
58 *
59 * Carnegie Mellon requests users of this software to return to
60 *
61 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
62 * School of Computer Science
63 * Carnegie Mellon University
64 * Pittsburgh PA 15213-3890
65 *
66 * any improvements or extensions that they make and grant Carnegie the
67 * rights to redistribute these changes.
68 */
69
70 /*
71 * uvm_glue.c: glue functions
72 */
73
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/proc.h>
77 #include <sys/resourcevar.h>
78 #include <sys/buf.h>
79 #include <sys/user.h>
80 #ifdef SYSVSHM
81 #include <sys/shm.h>
82 #endif
83 #include <sys/sched.h>
84
85 #include <uvm/uvm.h>
86
87 #include <machine/cpu.h>
88
89 /*
90 * XXXCDC: do these really belong here?
91 */
92
93 int readbuffers = 0; /* allow KGDB to read kern buffer pool */
94 /* XXX: see uvm_kernacc */
95
96
97 /*
98 * uvm_kernacc: can the kernel access a region of memory
99 *
100 * - called from malloc [DIAGNOSTIC], and /dev/kmem driver (mem.c)
101 */
102
103 boolean_t
104 uvm_kernacc(addr, len, rw)
105 caddr_t addr;
106 size_t len;
107 int rw;
108 {
109 boolean_t rv;
110 vaddr_t saddr, eaddr;
111 vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
112
113 saddr = trunc_page((vaddr_t)addr);
114 eaddr = round_page((vaddr_t)addr + len);
115 vm_map_lock_read(kernel_map);
116 rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
117 vm_map_unlock_read(kernel_map);
118
119 return(rv);
120 }
121
122 #ifdef KGDB
123 /*
124 * Change protections on kernel pages from addr to addr+len
125 * (presumably so debugger can plant a breakpoint).
126 *
127 * We force the protection change at the pmap level. If we were
128 * to use vm_map_protect a change to allow writing would be lazily-
129 * applied meaning we would still take a protection fault, something
130 * we really don't want to do. It would also fragment the kernel
131 * map unnecessarily. We cannot use pmap_protect since it also won't
132 * enforce a write-enable request. Using pmap_enter is the only way
133 * we can ensure the change takes place properly.
134 */
135 void
136 uvm_chgkprot(addr, len, rw)
137 caddr_t addr;
138 size_t len;
139 int rw;
140 {
141 vm_prot_t prot;
142 paddr_t pa;
143 vaddr_t sva, eva;
144
145 prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
146 eva = round_page((vaddr_t)addr + len);
147 for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
148 /*
149 * Extract physical address for the page.
150 * We use a cheezy hack to differentiate physical
151 * page 0 from an invalid mapping, not that it
152 * really matters...
153 */
154 if (pmap_extract(pmap_kernel(), sva, &pa) == FALSE)
155 panic("chgkprot: invalid page");
156 pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
157 }
158 pmap_update(pmap_kernel());
159 }
160 #endif
161
162 /*
163 * uvm_vslock: wire user memory for I/O
164 *
165 * - called from physio and sys___sysctl
166 * - XXXCDC: consider nuking this (or making it a macro?)
167 */
168
169 int
170 uvm_vslock(p, addr, len, access_type)
171 struct proc *p;
172 caddr_t addr;
173 size_t len;
174 vm_prot_t access_type;
175 {
176 vm_map_t map;
177 vaddr_t start, end;
178 int rv;
179
180 map = &p->p_vmspace->vm_map;
181 start = trunc_page((vaddr_t)addr);
182 end = round_page((vaddr_t)addr + len);
183 if (end <= start)
184 return (EINVAL);
185
186 rv = uvm_fault_wire(map, start, end, access_type);
187
188 return (rv);
189 }
190
191 /*
192 * uvm_vsunlock: unwire user memory wired by uvm_vslock()
193 *
194 * - called from physio and sys___sysctl
195 * - XXXCDC: consider nuking this (or making it a macro?)
196 */
197
198 void
199 uvm_vsunlock(p, addr, len)
200 struct proc *p;
201 caddr_t addr;
202 size_t len;
203 {
204 vaddr_t start, end;
205
206 start = trunc_page((vaddr_t)addr);
207 end = round_page((vaddr_t)addr + len);
208 if (end <= start)
209 return;
210
211 uvm_fault_unwire(&p->p_vmspace->vm_map, start, end);
212 }
213
214 /*
215 * uvm_fork: fork a virtual address space
216 *
217 * - the address space is copied as per parent map's inherit values
218 * - a new "user" structure is allocated for the child process
219 * [filled in by MD layer...]
220 * - if specified, the child gets a new user stack described by
221 * stack and stacksize
222 * - NOTE: the kernel stack may be at a different location in the child
223 * process, and thus addresses of automatic variables may be invalid
224 * after cpu_fork returns in the child process. We do nothing here
225 * after cpu_fork returns.
226 * - XXXCDC: we need a way for this to return a failure value rather
227 * than just hang
228 */
229 void
230 uvm_fork(p1, p2, shared, stack, stacksize, func, arg)
231 struct proc *p1, *p2;
232 boolean_t shared;
233 void *stack;
234 size_t stacksize;
235 void (*func)(void *);
236 void *arg;
237 {
238 struct user *up = p2->p_addr;
239
240 if (shared == TRUE) {
241 p2->p_vmspace = NULL;
242 uvmspace_share(p1, p2); /* share vmspace */
243 } else
244 p2->p_vmspace = uvmspace_fork(p1->p_vmspace); /* fork vmspace */
245
246 #ifdef PMAP_UAREA
247 /* Tell the pmap this is a u-area mapping */
248 PMAP_UAREA((vaddr_t)up);
249 #endif
250
251 /*
252 * p_stats currently points at a field in the user struct. Copy
253 * parts of p_stats, and zero out the rest.
254 */
255 p2->p_stats = &up->u_stats;
256 memset(&up->u_stats.pstat_startzero, 0,
257 ((caddr_t)&up->u_stats.pstat_endzero -
258 (caddr_t)&up->u_stats.pstat_startzero));
259 memcpy(&up->u_stats.pstat_startcopy, &p1->p_stats->pstat_startcopy,
260 ((caddr_t)&up->u_stats.pstat_endcopy -
261 (caddr_t)&up->u_stats.pstat_startcopy));
262
263 /*
264 * cpu_fork() copy and update the pcb, and make the child ready
265 * to run. If this is a normal user fork, the child will exit
266 * directly to user mode via child_return() on its first time
267 * slice and will not return here. If this is a kernel thread,
268 * the specified entry point will be executed.
269 */
270 cpu_fork(p1, p2, stack, stacksize, func, arg);
271 }
272
273 /*
274 * uvm_exit: exit a virtual address space
275 *
276 * - the process passed to us is a dead (pre-zombie) process; we
277 * are running on a different context now (the reaper).
278 * - we must run in a separate thread because freeing the vmspace
279 * of the dead process may block.
280 */
281 void
282 uvm_exit(struct proc *p)
283 {
284 uvmspace_free(p->p_vmspace);
285 uvm_km_free(kernel_map, (vaddr_t)p->p_addr, USPACE);
286 p->p_addr = NULL;
287 }
288
289 /*
290 * uvm_init_limit: init per-process VM limits
291 *
292 * - called for process 0 and then inherited by all others.
293 */
294 void
295 uvm_init_limits(struct proc *p)
296 {
297
298 /*
299 * Set up the initial limits on process VM. Set the maximum
300 * resident set size to be all of (reasonably) available memory.
301 * This causes any single, large process to start random page
302 * replacement once it fills memory.
303 */
304
305 p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
306 p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
307 p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
308 p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
309 p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free);
310 }
311
312 #ifdef DEBUG
313 int enableswap = 1;
314 int swapdebug = 0;
315 #define SDB_FOLLOW 1
316 #define SDB_SWAPIN 2
317 #define SDB_SWAPOUT 4
318 #endif
319
320 /*
321 * uvm_scheduler: process zero main loop
322 *
323 * - if not enough memory, wake the pagedaemon and let it clear space.
324 */
325
326 void
327 uvm_scheduler(void)
328 {
329 /*
330 * Nothing to do, back to sleep
331 */
332 while (1)
333 tsleep(&proc0, PVM, "scheduler", 0);
334 }
335
336 /*
337 * swappable: is process "p" swappable?
338 */
339
340 #define swappable(p) (((p)->p_flag & (P_SYSTEM | P_WEXIT)) == 0)
341
342 /*
343 * swapout_threads: find threads that can be swapped
344 *
345 * - called by the pagedaemon
346 * - try and swap at least one processs
347 * - processes that are sleeping or stopped for maxslp or more seconds
348 * are swapped... otherwise the longest-sleeping or stopped process
349 * is swapped, otherwise the longest resident process...
350 */
351 void
352 uvm_swapout_threads(void)
353 {
354 struct proc *p;
355 struct proc *outp, *outp2;
356 int outpri, outpri2;
357 int didswap = 0;
358 extern int maxslp;
359 /* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */
360
361 #ifdef DEBUG
362 if (!enableswap)
363 return;
364 #endif
365
366 /*
367 * outp/outpri : stop/sleep process with largest sleeptime < maxslp
368 * outp2/outpri2: the longest resident process (its swap time)
369 */
370 outp = outp2 = NULL;
371 outpri = outpri2 = 0;
372 LIST_FOREACH(p, &allproc, p_list) {
373 if (!swappable(p))
374 continue;
375 switch (p->p_stat) {
376 case SRUN:
377 if (p->p_swtime > outpri2) {
378 outp2 = p;
379 outpri2 = p->p_swtime;
380 }
381 continue;
382
383 case SSLEEP:
384 case SSTOP:
385 if (p->p_slptime >= maxslp) {
386 pmap_collect(p->p_vmspace->vm_map.pmap);
387 didswap++;
388 } else if (p->p_slptime > outpri) {
389 outp = p;
390 outpri = p->p_slptime;
391 }
392 continue;
393 }
394 }
395
396 /*
397 * If we didn't get rid of any real duds, toss out the next most
398 * likely sleeping/stopped or running candidate. We only do this
399 * if we are real low on memory since we don't gain much by doing
400 * it.
401 */
402 if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE))) {
403 if ((p = outp) == NULL)
404 p = outp2;
405 #ifdef DEBUG
406 if (swapdebug & SDB_SWAPOUT)
407 printf("swapout_threads: no duds, try procp %p\n", p);
408 #endif
409 if (p)
410 pmap_collect(p->p_vmspace->vm_map.pmap);
411 }
412 }