xref: /netbsd-src/sys/uvm/uvm_glue.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /*	$NetBSD: uvm_glue.c,v 1.146 2011/01/14 02:06:34 rmind Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * Copyright (c) 1991, 1993, The Regents of the University of California.
6  *
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * The Mach Operating System project at Carnegie-Mellon University.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by Charles D. Cranor,
23  *      Washington University, the University of California, Berkeley and
24  *      its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	@(#)vm_glue.c	8.6 (Berkeley) 1/5/94
42  * from: Id: uvm_glue.c,v 1.1.2.8 1998/02/07 01:16:54 chs Exp
43  *
44  *
45  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
46  * All rights reserved.
47  *
48  * Permission to use, copy, modify and distribute this software and
49  * its documentation is hereby granted, provided that both the copyright
50  * notice and this permission notice appear in all copies of the
51  * software, derivative works or modified versions, and any portions
52  * thereof, and that both notices appear in supporting documentation.
53  *
54  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
55  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
56  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
57  *
58  * Carnegie Mellon requests users of this software to return to
59  *
60  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
61  *  School of Computer Science
62  *  Carnegie Mellon University
63  *  Pittsburgh PA 15213-3890
64  *
65  * any improvements or extensions that they make and grant Carnegie the
66  * rights to redistribute these changes.
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: uvm_glue.c,v 1.146 2011/01/14 02:06:34 rmind Exp $");
71 
72 #include "opt_kgdb.h"
73 #include "opt_kstack.h"
74 #include "opt_uvmhist.h"
75 
76 /*
77  * uvm_glue.c: glue functions
78  */
79 
80 #include <sys/param.h>
81 #include <sys/kernel.h>
82 
83 #include <sys/systm.h>
84 #include <sys/proc.h>
85 #include <sys/resourcevar.h>
86 #include <sys/buf.h>
87 #include <sys/syncobj.h>
88 #include <sys/cpu.h>
89 #include <sys/atomic.h>
90 #include <sys/lwp.h>
91 
92 #include <uvm/uvm.h>
93 
94 /*
95  * XXXCDC: do these really belong here?
96  */
97 
98 /*
99  * uvm_kernacc: can the kernel access a region of memory
100  *
101  * - used only by /dev/kmem driver (mem.c)
102  */
103 
104 bool
105 uvm_kernacc(void *addr, size_t len, int rw)
106 {
107 	bool rv;
108 	vaddr_t saddr, eaddr;
109 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
110 
111 	saddr = trunc_page((vaddr_t)addr);
112 	eaddr = round_page((vaddr_t)addr + len);
113 	vm_map_lock_read(kernel_map);
114 	rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
115 	vm_map_unlock_read(kernel_map);
116 
117 	return(rv);
118 }
119 
120 #ifdef KGDB
121 /*
122  * Change protections on kernel pages from addr to addr+len
123  * (presumably so debugger can plant a breakpoint).
124  *
125  * We force the protection change at the pmap level.  If we were
126  * to use vm_map_protect a change to allow writing would be lazily-
127  * applied meaning we would still take a protection fault, something
128  * we really don't want to do.  It would also fragment the kernel
129  * map unnecessarily.  We cannot use pmap_protect since it also won't
130  * enforce a write-enable request.  Using pmap_enter is the only way
131  * we can ensure the change takes place properly.
132  */
133 void
134 uvm_chgkprot(void *addr, size_t len, int rw)
135 {
136 	vm_prot_t prot;
137 	paddr_t pa;
138 	vaddr_t sva, eva;
139 
140 	prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
141 	eva = round_page((vaddr_t)addr + len);
142 	for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
143 		/*
144 		 * Extract physical address for the page.
145 		 */
146 		if (pmap_extract(pmap_kernel(), sva, &pa) == false)
147 			panic("%s: invalid page", __func__);
148 		pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
149 	}
150 	pmap_update(pmap_kernel());
151 }
152 #endif
153 
154 /*
155  * uvm_vslock: wire user memory for I/O
156  *
157  * - called from physio and sys___sysctl
158  * - XXXCDC: consider nuking this (or making it a macro?)
159  */
160 
161 int
162 uvm_vslock(struct vmspace *vs, void *addr, size_t len, vm_prot_t access_type)
163 {
164 	struct vm_map *map;
165 	vaddr_t start, end;
166 	int error;
167 
168 	map = &vs->vm_map;
169 	start = trunc_page((vaddr_t)addr);
170 	end = round_page((vaddr_t)addr + len);
171 	error = uvm_fault_wire(map, start, end, access_type, 0);
172 	return error;
173 }
174 
175 /*
176  * uvm_vsunlock: unwire user memory wired by uvm_vslock()
177  *
178  * - called from physio and sys___sysctl
179  * - XXXCDC: consider nuking this (or making it a macro?)
180  */
181 
182 void
183 uvm_vsunlock(struct vmspace *vs, void *addr, size_t len)
184 {
185 	uvm_fault_unwire(&vs->vm_map, trunc_page((vaddr_t)addr),
186 		round_page((vaddr_t)addr + len));
187 }
188 
189 /*
190  * uvm_proc_fork: fork a virtual address space
191  *
192  * - the address space is copied as per parent map's inherit values
193  */
194 void
195 uvm_proc_fork(struct proc *p1, struct proc *p2, bool shared)
196 {
197 
198 	if (shared == true) {
199 		p2->p_vmspace = NULL;
200 		uvmspace_share(p1, p2);
201 	} else {
202 		p2->p_vmspace = uvmspace_fork(p1->p_vmspace);
203 	}
204 
205 	cpu_proc_fork(p1, p2);
206 }
207 
208 /*
209  * uvm_lwp_fork: fork a thread
210  *
211  * - a new PCB structure is allocated for the child process,
212  *	and filled in by MD layer
213  * - if specified, the child gets a new user stack described by
214  *	stack and stacksize
215  * - NOTE: the kernel stack may be at a different location in the child
216  *	process, and thus addresses of automatic variables may be invalid
217  *	after cpu_lwp_fork returns in the child process.  We do nothing here
218  *	after cpu_lwp_fork returns.
219  */
220 void
221 uvm_lwp_fork(struct lwp *l1, struct lwp *l2, void *stack, size_t stacksize,
222     void (*func)(void *), void *arg)
223 {
224 
225 	/* Fill stack with magic number. */
226 	kstack_setup_magic(l2);
227 
228 	/*
229 	 * cpu_lwp_fork() copy and update the pcb, and make the child ready
230  	 * to run.  If this is a normal user fork, the child will exit
231 	 * directly to user mode via child_return() on its first time
232 	 * slice and will not return here.  If this is a kernel thread,
233 	 * the specified entry point will be executed.
234 	 */
235 	cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
236 
237 	/* Inactive emap for new LWP. */
238 	l2->l_emap_gen = UVM_EMAP_INACTIVE;
239 }
240 
241 #ifndef USPACE_ALIGN
242 #define	USPACE_ALIGN	0
243 #endif
244 
245 static pool_cache_t uvm_uarea_cache;
246 
247 static void *
248 uarea_poolpage_alloc(struct pool *pp, int flags)
249 {
250 #if defined(PMAP_MAP_POOLPAGE)
251 	if (USPACE == PAGE_SIZE && USPACE_ALIGN == 0) {
252 		struct vm_page *pg;
253 		vaddr_t va;
254 
255 		pg = uvm_pagealloc(NULL, 0, NULL,
256 		   ((flags & PR_WAITOK) == 0 ? UVM_KMF_NOWAIT : 0));
257 		if (pg == NULL)
258 			return NULL;
259 		va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
260 		if (va == 0)
261 			uvm_pagefree(pg);
262 		return (void *)va;
263 	}
264 #endif
265 	return (void *)uvm_km_alloc(kernel_map, pp->pr_alloc->pa_pagesz,
266 	    USPACE_ALIGN, UVM_KMF_WIRED |
267 	    ((flags & PR_WAITOK) ? UVM_KMF_WAITVA :
268 	    (UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK)));
269 }
270 
271 static void
272 uarea_poolpage_free(struct pool *pp, void *addr)
273 {
274 #if defined(PMAP_MAP_POOLPAGE)
275 	if (USPACE == PAGE_SIZE && USPACE_ALIGN == 0) {
276 		paddr_t pa;
277 
278 		pa = PMAP_UNMAP_POOLPAGE((vaddr_t) addr);
279 		KASSERT(pa != 0);
280 		uvm_pagefree(PHYS_TO_VM_PAGE(pa));
281 		return;
282 	}
283 #endif
284 	uvm_km_free(kernel_map, (vaddr_t)addr, pp->pr_alloc->pa_pagesz,
285 	    UVM_KMF_WIRED);
286 }
287 
288 static struct pool_allocator uvm_uarea_allocator = {
289 	.pa_alloc = uarea_poolpage_alloc,
290 	.pa_free = uarea_poolpage_free,
291 	.pa_pagesz = USPACE,
292 };
293 
294 void
295 uvm_uarea_init(void)
296 {
297 	int flags = PR_NOTOUCH;
298 
299 	/*
300 	 * specify PR_NOALIGN unless the alignment provided by
301 	 * the backend (USPACE_ALIGN) is sufficient to provide
302 	 * pool page size (UPSACE) alignment.
303 	 */
304 
305 	if ((USPACE_ALIGN == 0 && USPACE != PAGE_SIZE) ||
306 	    (USPACE_ALIGN % USPACE) != 0) {
307 		flags |= PR_NOALIGN;
308 	}
309 
310 	uvm_uarea_cache = pool_cache_init(USPACE, USPACE_ALIGN, 0, flags,
311 	    "uarea", &uvm_uarea_allocator, IPL_NONE, NULL, NULL, NULL);
312 }
313 
314 /*
315  * uvm_uarea_alloc: allocate a u-area
316  */
317 
318 vaddr_t
319 uvm_uarea_alloc(void)
320 {
321 
322 	return (vaddr_t)pool_cache_get(uvm_uarea_cache, PR_WAITOK);
323 }
324 
325 /*
326  * uvm_uarea_free: free a u-area
327  */
328 
329 void
330 uvm_uarea_free(vaddr_t uaddr)
331 {
332 
333 	pool_cache_put(uvm_uarea_cache, (void *)uaddr);
334 }
335 
336 vaddr_t
337 uvm_lwp_getuarea(lwp_t *l)
338 {
339 
340 	return (vaddr_t)l->l_addr - UAREA_PCB_OFFSET;
341 }
342 
343 void
344 uvm_lwp_setuarea(lwp_t *l, vaddr_t addr)
345 {
346 
347 	l->l_addr = (void *)(addr + UAREA_PCB_OFFSET);
348 }
349 
350 /*
351  * uvm_proc_exit: exit a virtual address space
352  *
353  * - borrow proc0's address space because freeing the vmspace
354  *   of the dead process may block.
355  */
356 
357 void
358 uvm_proc_exit(struct proc *p)
359 {
360 	struct lwp *l = curlwp; /* XXX */
361 	struct vmspace *ovm;
362 
363 	KASSERT(p == l->l_proc);
364 	ovm = p->p_vmspace;
365 
366 	/*
367 	 * borrow proc0's address space.
368 	 */
369 	KPREEMPT_DISABLE(l);
370 	pmap_deactivate(l);
371 	p->p_vmspace = proc0.p_vmspace;
372 	pmap_activate(l);
373 	KPREEMPT_ENABLE(l);
374 
375 	uvmspace_free(ovm);
376 }
377 
378 void
379 uvm_lwp_exit(struct lwp *l)
380 {
381 	vaddr_t va = uvm_lwp_getuarea(l);
382 
383 	uvm_uarea_free(va);
384 #ifdef DIAGNOSTIC
385 	uvm_lwp_setuarea(l, (vaddr_t)NULL);
386 #endif
387 }
388 
389 /*
390  * uvm_init_limit: init per-process VM limits
391  *
392  * - called for process 0 and then inherited by all others.
393  */
394 
395 void
396 uvm_init_limits(struct proc *p)
397 {
398 
399 	/*
400 	 * Set up the initial limits on process VM.  Set the maximum
401 	 * resident set size to be all of (reasonably) available memory.
402 	 * This causes any single, large process to start random page
403 	 * replacement once it fills memory.
404 	 */
405 
406 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
407 	p->p_rlimit[RLIMIT_STACK].rlim_max = maxsmap;
408 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
409 	p->p_rlimit[RLIMIT_DATA].rlim_max = maxdmap;
410 	p->p_rlimit[RLIMIT_AS].rlim_cur = RLIM_INFINITY;
411 	p->p_rlimit[RLIMIT_AS].rlim_max = RLIM_INFINITY;
412 	p->p_rlimit[RLIMIT_RSS].rlim_cur = MIN(
413 	    VM_MAXUSER_ADDRESS, ctob((rlim_t)uvmexp.free));
414 }
415 
416 /*
417  * uvm_scheduler: process zero main loop.
418  */
419 
420 extern struct loadavg averunnable;
421 
422 void
423 uvm_scheduler(void)
424 {
425 	lwp_t *l = curlwp;
426 
427 	lwp_lock(l);
428 	l->l_priority = PRI_VM;
429 	l->l_class = SCHED_FIFO;
430 	lwp_unlock(l);
431 
432 	for (;;) {
433 		sched_pstats();
434 		(void)kpause("uvm", false, hz, NULL);
435 	}
436 }
437