xref: /openbsd-src/sys/uvm/uvm_glue.c (revision a28daedfc357b214be5c701aa8ba8adb29a7f1c2)
1 /*	$OpenBSD: uvm_glue.c,v 1.49 2009/03/20 15:19:04 oga 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(caddr_t addr, size_t len, int rw)
105 {
106 	boolean_t rv;
107 	vaddr_t saddr, eaddr;
108 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
109 
110 	saddr = trunc_page((vaddr_t)addr);
111 	eaddr = round_page((vaddr_t)addr + len);
112 	vm_map_lock_read(kernel_map);
113 	rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
114 	vm_map_unlock_read(kernel_map);
115 
116 	return(rv);
117 }
118 
119 #ifdef KGDB
120 /*
121  * Change protections on kernel pages from addr to addr+len
122  * (presumably so debugger can plant a breakpoint).
123  *
124  * We force the protection change at the pmap level.  If we were
125  * to use vm_map_protect a change to allow writing would be lazily-
126  * applied meaning we would still take a protection fault, something
127  * we really don't want to do.  It would also fragment the kernel
128  * map unnecessarily.  We cannot use pmap_protect since it also won't
129  * enforce a write-enable request.  Using pmap_enter is the only way
130  * we can ensure the change takes place properly.
131  */
132 void
133 uvm_chgkprot(caddr_t addr, size_t len, int rw)
134 {
135 	vm_prot_t prot;
136 	paddr_t pa;
137 	vaddr_t sva, eva;
138 
139 	prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
140 	eva = round_page((vaddr_t)addr + len);
141 	for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
142 		/*
143 		 * Extract physical address for the page.
144 		 * We use a cheezy hack to differentiate physical
145 		 * page 0 from an invalid mapping, not that it
146 		 * really matters...
147 		 */
148 		if (pmap_extract(pmap_kernel(), sva, &pa) == FALSE)
149 			panic("chgkprot: invalid page");
150 		pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
151 	}
152 	pmap_update(pmap_kernel());
153 }
154 #endif
155 
156 /*
157  * uvm_vslock: wire user memory for I/O
158  *
159  * - called from physio and sys___sysctl
160  * - XXXCDC: consider nuking this (or making it a macro?)
161  */
162 
163 int
164 uvm_vslock(struct proc *p, caddr_t addr, size_t len, vm_prot_t access_type)
165 {
166 	vm_map_t map;
167 	vaddr_t start, end;
168 	int rv;
169 
170 	map = &p->p_vmspace->vm_map;
171 	start = trunc_page((vaddr_t)addr);
172 	end = round_page((vaddr_t)addr + len);
173 	if (end <= start)
174 		return (EINVAL);
175 
176 	rv = uvm_fault_wire(map, start, end, access_type);
177 
178 	return (rv);
179 }
180 
181 /*
182  * uvm_vsunlock: unwire user memory wired by uvm_vslock()
183  *
184  * - called from physio and sys___sysctl
185  * - XXXCDC: consider nuking this (or making it a macro?)
186  */
187 
188 void
189 uvm_vsunlock(struct proc *p, caddr_t addr, size_t len)
190 {
191 	vaddr_t start, end;
192 
193 	start = trunc_page((vaddr_t)addr);
194 	end = round_page((vaddr_t)addr + len);
195 	if (end <= start)
196 		return;
197 
198 	uvm_fault_unwire(&p->p_vmspace->vm_map, start, end);
199 }
200 
201 /*
202  * uvm_fork: fork a virtual address space
203  *
204  * - the address space is copied as per parent map's inherit values
205  * - a new "user" structure is allocated for the child process
206  *	[filled in by MD layer...]
207  * - if specified, the child gets a new user stack described by
208  *	stack and stacksize
209  * - NOTE: the kernel stack may be at a different location in the child
210  *	process, and thus addresses of automatic variables may be invalid
211  *	after cpu_fork returns in the child process.  We do nothing here
212  *	after cpu_fork returns.
213  * - XXXCDC: we need a way for this to return a failure value rather
214  *   than just hang
215  */
216 void
217 uvm_fork(p1, p2, shared, stack, stacksize, func, arg)
218 	struct proc *p1, *p2;
219 	boolean_t shared;
220 	void *stack;
221 	size_t stacksize;
222 	void (*func)(void *);
223 	void *arg;
224 {
225 	struct user *up = p2->p_addr;
226 
227 	if (shared == TRUE) {
228 		p2->p_vmspace = NULL;
229 		uvmspace_share(p1, p2);			/* share vmspace */
230 	} else
231 		p2->p_vmspace = uvmspace_fork(p1->p_vmspace); /* fork vmspace */
232 
233 #ifdef PMAP_UAREA
234 	/* Tell the pmap this is a u-area mapping */
235 	PMAP_UAREA((vaddr_t)up);
236 #endif
237 
238 	/*
239 	 * p_stats currently points at a field in the user struct.  Copy
240 	 * parts of p_stats, and zero out the rest.
241 	 */
242 	p2->p_stats = &up->u_stats;
243 	memset(&up->u_stats.pstat_startzero, 0,
244 	       ((caddr_t)&up->u_stats.pstat_endzero -
245 		(caddr_t)&up->u_stats.pstat_startzero));
246 	memcpy(&up->u_stats.pstat_startcopy, &p1->p_stats->pstat_startcopy,
247 	       ((caddr_t)&up->u_stats.pstat_endcopy -
248 		(caddr_t)&up->u_stats.pstat_startcopy));
249 
250 	/*
251 	 * cpu_fork() copy and update the pcb, and make the child ready
252 	 * to run.  If this is a normal user fork, the child will exit
253 	 * directly to user mode via child_return() on its first time
254 	 * slice and will not return here.  If this is a kernel thread,
255 	 * the specified entry point will be executed.
256 	 */
257 	cpu_fork(p1, p2, stack, stacksize, func, arg);
258 }
259 
260 /*
261  * uvm_exit: exit a virtual address space
262  *
263  * - the process passed to us is a dead (pre-zombie) process; we
264  *   are running on a different context now (the reaper).
265  * - we must run in a separate thread because freeing the vmspace
266  *   of the dead process may block.
267  */
268 void
269 uvm_exit(struct proc *p)
270 {
271 	uvmspace_free(p->p_vmspace);
272 	p->p_vmspace = NULL;
273 	uvm_km_free(kernel_map, (vaddr_t)p->p_addr, USPACE);
274 	p->p_addr = NULL;
275 }
276 
277 /*
278  * uvm_init_limit: init per-process VM limits
279  *
280  * - called for process 0 and then inherited by all others.
281  */
282 void
283 uvm_init_limits(struct proc *p)
284 {
285 
286 	/*
287 	 * Set up the initial limits on process VM.  Set the maximum
288 	 * resident set size to be all of (reasonably) available memory.
289 	 * This causes any single, large process to start random page
290 	 * replacement once it fills memory.
291 	 */
292 
293 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
294 	p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ;
295 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
296 	p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ;
297 	p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(uvmexp.free);
298 }
299 
300 #ifdef DEBUG
301 int	enableswap = 1;
302 int	swapdebug = 0;
303 #define	SDB_FOLLOW	1
304 #define SDB_SWAPIN	2
305 #define SDB_SWAPOUT	4
306 #endif
307 
308 /*
309  * uvm_scheduler: process zero main loop
310  *
311  * - if not enough memory, wake the pagedaemon and let it clear space.
312  */
313 
314 void
315 uvm_scheduler(void)
316 {
317 	/*
318 	 * Nothing to do, back to sleep
319 	 */
320 	while (1)
321 		tsleep(&proc0, PVM, "scheduler", 0);
322 }
323 
324 /*
325  * swappable: is process "p" swappable?
326  */
327 
328 #define	swappable(p) (((p)->p_flag & (P_SYSTEM | P_WEXIT)) == 0)
329 
330 /*
331  * swapout_threads: find threads that can be swapped
332  *
333  * - called by the pagedaemon
334  * - try and swap at least one processs
335  * - processes that are sleeping or stopped for maxslp or more seconds
336  *   are swapped... otherwise the longest-sleeping or stopped process
337  *   is swapped, otherwise the longest resident process...
338  */
339 void
340 uvm_swapout_threads(void)
341 {
342 	struct proc *p;
343 	struct proc *outp, *outp2;
344 	int outpri, outpri2;
345 	int didswap = 0;
346 	extern int maxslp;
347 	/* XXXCDC: should move off to uvmexp. or uvm., also in uvm_meter */
348 
349 #ifdef DEBUG
350 	if (!enableswap)
351 		return;
352 #endif
353 
354 	/*
355 	 * outp/outpri  : stop/sleep process with largest sleeptime < maxslp
356 	 * outp2/outpri2: the longest resident process (its swap time)
357 	 */
358 	outp = outp2 = NULL;
359 	outpri = outpri2 = 0;
360 	LIST_FOREACH(p, &allproc, p_list) {
361 		if (!swappable(p))
362 			continue;
363 		switch (p->p_stat) {
364 		case SRUN:
365 			if (p->p_swtime > outpri2) {
366 				outp2 = p;
367 				outpri2 = p->p_swtime;
368 			}
369 			continue;
370 
371 		case SSLEEP:
372 		case SSTOP:
373 			if (p->p_slptime >= maxslp) {
374 				pmap_collect(p->p_vmspace->vm_map.pmap);
375 				didswap++;
376 			} else if (p->p_slptime > outpri) {
377 				outp = p;
378 				outpri = p->p_slptime;
379 			}
380 			continue;
381 		}
382 	}
383 
384 	/*
385 	 * If we didn't get rid of any real duds, toss out the next most
386 	 * likely sleeping/stopped or running candidate.  We only do this
387 	 * if we are real low on memory since we don't gain much by doing
388 	 * it.
389 	 */
390 	if (didswap == 0 && uvmexp.free <= atop(round_page(USPACE))) {
391 		if ((p = outp) == NULL)
392 			p = outp2;
393 #ifdef DEBUG
394 		if (swapdebug & SDB_SWAPOUT)
395 			printf("swapout_threads: no duds, try procp %p\n", p);
396 #endif
397 		if (p)
398 			pmap_collect(p->p_vmspace->vm_map.pmap);
399 	}
400 }
401