xref: /netbsd-src/sys/uvm/uvm_glue.c (revision 46f5119e40af2e51998f686b2fdcc76b5488f7f3)
1 /*	$NetBSD: uvm_glue.c,v 1.149 2011/02/18 10:43:52 drochner 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. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)vm_glue.c	8.6 (Berkeley) 1/5/94
37  * from: Id: uvm_glue.c,v 1.1.2.8 1998/02/07 01:16:54 chs Exp
38  *
39  *
40  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
41  * All rights reserved.
42  *
43  * Permission to use, copy, modify and distribute this software and
44  * its documentation is hereby granted, provided that both the copyright
45  * notice and this permission notice appear in all copies of the
46  * software, derivative works or modified versions, and any portions
47  * thereof, and that both notices appear in supporting documentation.
48  *
49  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
50  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
51  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52  *
53  * Carnegie Mellon requests users of this software to return to
54  *
55  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
56  *  School of Computer Science
57  *  Carnegie Mellon University
58  *  Pittsburgh PA 15213-3890
59  *
60  * any improvements or extensions that they make and grant Carnegie the
61  * rights to redistribute these changes.
62  */
63 
64 #include <sys/cdefs.h>
65 __KERNEL_RCSID(0, "$NetBSD: uvm_glue.c,v 1.149 2011/02/18 10:43:52 drochner Exp $");
66 
67 #include "opt_kgdb.h"
68 #include "opt_kstack.h"
69 #include "opt_uvmhist.h"
70 
71 /*
72  * uvm_glue.c: glue functions
73  */
74 
75 #include <sys/param.h>
76 #include <sys/kernel.h>
77 
78 #include <sys/systm.h>
79 #include <sys/proc.h>
80 #include <sys/resourcevar.h>
81 #include <sys/buf.h>
82 #include <sys/syncobj.h>
83 #include <sys/cpu.h>
84 #include <sys/atomic.h>
85 #include <sys/lwp.h>
86 
87 #include <uvm/uvm.h>
88 
89 /*
90  * XXXCDC: do these really belong here?
91  */
92 
93 /*
94  * uvm_kernacc: can the kernel access a region of memory
95  *
96  * - used only by /dev/kmem driver (mem.c)
97  */
98 
99 bool
100 uvm_kernacc(void *addr, size_t len, int rw)
101 {
102 	bool rv;
103 	vaddr_t saddr, eaddr;
104 	vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE;
105 
106 	saddr = trunc_page((vaddr_t)addr);
107 	eaddr = round_page((vaddr_t)addr + len);
108 	vm_map_lock_read(kernel_map);
109 	rv = uvm_map_checkprot(kernel_map, saddr, eaddr, prot);
110 	vm_map_unlock_read(kernel_map);
111 
112 	return(rv);
113 }
114 
115 #ifdef KGDB
116 /*
117  * Change protections on kernel pages from addr to addr+len
118  * (presumably so debugger can plant a breakpoint).
119  *
120  * We force the protection change at the pmap level.  If we were
121  * to use vm_map_protect a change to allow writing would be lazily-
122  * applied meaning we would still take a protection fault, something
123  * we really don't want to do.  It would also fragment the kernel
124  * map unnecessarily.  We cannot use pmap_protect since it also won't
125  * enforce a write-enable request.  Using pmap_enter is the only way
126  * we can ensure the change takes place properly.
127  */
128 void
129 uvm_chgkprot(void *addr, size_t len, int rw)
130 {
131 	vm_prot_t prot;
132 	paddr_t pa;
133 	vaddr_t sva, eva;
134 
135 	prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE;
136 	eva = round_page((vaddr_t)addr + len);
137 	for (sva = trunc_page((vaddr_t)addr); sva < eva; sva += PAGE_SIZE) {
138 		/*
139 		 * Extract physical address for the page.
140 		 */
141 		if (pmap_extract(pmap_kernel(), sva, &pa) == false)
142 			panic("%s: invalid page", __func__);
143 		pmap_enter(pmap_kernel(), sva, pa, prot, PMAP_WIRED);
144 	}
145 	pmap_update(pmap_kernel());
146 }
147 #endif
148 
149 /*
150  * uvm_vslock: wire user memory for I/O
151  *
152  * - called from physio and sys___sysctl
153  * - XXXCDC: consider nuking this (or making it a macro?)
154  */
155 
156 int
157 uvm_vslock(struct vmspace *vs, void *addr, size_t len, vm_prot_t access_type)
158 {
159 	struct vm_map *map;
160 	vaddr_t start, end;
161 	int error;
162 
163 	map = &vs->vm_map;
164 	start = trunc_page((vaddr_t)addr);
165 	end = round_page((vaddr_t)addr + len);
166 	error = uvm_fault_wire(map, start, end, access_type, 0);
167 	return error;
168 }
169 
170 /*
171  * uvm_vsunlock: unwire user memory wired by uvm_vslock()
172  *
173  * - called from physio and sys___sysctl
174  * - XXXCDC: consider nuking this (or making it a macro?)
175  */
176 
177 void
178 uvm_vsunlock(struct vmspace *vs, void *addr, size_t len)
179 {
180 	uvm_fault_unwire(&vs->vm_map, trunc_page((vaddr_t)addr),
181 		round_page((vaddr_t)addr + len));
182 }
183 
184 /*
185  * uvm_proc_fork: fork a virtual address space
186  *
187  * - the address space is copied as per parent map's inherit values
188  */
189 void
190 uvm_proc_fork(struct proc *p1, struct proc *p2, bool shared)
191 {
192 
193 	if (shared == true) {
194 		p2->p_vmspace = NULL;
195 		uvmspace_share(p1, p2);
196 	} else {
197 		p2->p_vmspace = uvmspace_fork(p1->p_vmspace);
198 	}
199 
200 	cpu_proc_fork(p1, p2);
201 }
202 
203 /*
204  * uvm_lwp_fork: fork a thread
205  *
206  * - a new PCB structure is allocated for the child process,
207  *	and filled in by MD layer
208  * - if specified, the child gets a new user stack described by
209  *	stack and stacksize
210  * - NOTE: the kernel stack may be at a different location in the child
211  *	process, and thus addresses of automatic variables may be invalid
212  *	after cpu_lwp_fork returns in the child process.  We do nothing here
213  *	after cpu_lwp_fork returns.
214  */
215 void
216 uvm_lwp_fork(struct lwp *l1, struct lwp *l2, void *stack, size_t stacksize,
217     void (*func)(void *), void *arg)
218 {
219 
220 	/* Fill stack with magic number. */
221 	kstack_setup_magic(l2);
222 
223 	/*
224 	 * cpu_lwp_fork() copy and update the pcb, and make the child ready
225  	 * to run.  If this is a normal user fork, the child will exit
226 	 * directly to user mode via child_return() on its first time
227 	 * slice and will not return here.  If this is a kernel thread,
228 	 * the specified entry point will be executed.
229 	 */
230 	cpu_lwp_fork(l1, l2, stack, stacksize, func, arg);
231 
232 	/* Inactive emap for new LWP. */
233 	l2->l_emap_gen = UVM_EMAP_INACTIVE;
234 }
235 
236 #ifndef USPACE_ALIGN
237 #define	USPACE_ALIGN	0
238 #endif
239 
240 static pool_cache_t uvm_uarea_cache;
241 #if defined(__HAVE_CPU_UAREA_ROUTINES)
242 static pool_cache_t uvm_uarea_system_cache;
243 #else
244 #define uvm_uarea_system_cache uvm_uarea_cache
245 #endif
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 #if defined(__HAVE_CPU_UAREA_ROUTINES)
266 	void *va = cpu_uarea_alloc(false);
267 	if (va)
268 		return (void *)va;
269 #endif
270 	return (void *)uvm_km_alloc(kernel_map, pp->pr_alloc->pa_pagesz,
271 	    USPACE_ALIGN, UVM_KMF_WIRED |
272 	    ((flags & PR_WAITOK) ? UVM_KMF_WAITVA :
273 	    (UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK)));
274 }
275 
276 static void
277 uarea_poolpage_free(struct pool *pp, void *addr)
278 {
279 #if defined(PMAP_MAP_POOLPAGE)
280 	if (USPACE == PAGE_SIZE && USPACE_ALIGN == 0) {
281 		paddr_t pa;
282 
283 		pa = PMAP_UNMAP_POOLPAGE((vaddr_t) addr);
284 		KASSERT(pa != 0);
285 		uvm_pagefree(PHYS_TO_VM_PAGE(pa));
286 		return;
287 	}
288 #endif
289 #if defined(__HAVE_CPU_UAREA_ROUTINES)
290 	if (cpu_uarea_free(addr))
291 		return;
292 #endif
293 	uvm_km_free(kernel_map, (vaddr_t)addr, pp->pr_alloc->pa_pagesz,
294 	    UVM_KMF_WIRED);
295 }
296 
297 static struct pool_allocator uvm_uarea_allocator = {
298 	.pa_alloc = uarea_poolpage_alloc,
299 	.pa_free = uarea_poolpage_free,
300 	.pa_pagesz = USPACE,
301 };
302 
303 #if defined(__HAVE_CPU_UAREA_ROUTINES)
304 static void *
305 uarea_system_poolpage_alloc(struct pool *pp, int flags)
306 {
307 	void * const va = cpu_uarea_alloc(true);
308 	KASSERT(va != NULL);
309 	return va;
310 }
311 
312 static void
313 uarea_system_poolpage_free(struct pool *pp, void *addr)
314 {
315 	if (!cpu_uarea_free(addr))
316 		panic("%s: failed to free uarea %p", __func__, addr);
317 }
318 
319 static struct pool_allocator uvm_uarea_system_allocator = {
320 	.pa_alloc = uarea_system_poolpage_alloc,
321 	.pa_free = uarea_system_poolpage_free,
322 	.pa_pagesz = USPACE,
323 };
324 #endif /* __HAVE_CPU_UAREA_ROUTINES */
325 
326 void
327 uvm_uarea_init(void)
328 {
329 	int flags = PR_NOTOUCH;
330 
331 	/*
332 	 * specify PR_NOALIGN unless the alignment provided by
333 	 * the backend (USPACE_ALIGN) is sufficient to provide
334 	 * pool page size (UPSACE) alignment.
335 	 */
336 
337 	if ((USPACE_ALIGN == 0 && USPACE != PAGE_SIZE) ||
338 	    (USPACE_ALIGN % USPACE) != 0) {
339 		flags |= PR_NOALIGN;
340 	}
341 
342 	uvm_uarea_cache = pool_cache_init(USPACE, USPACE_ALIGN, 0, flags,
343 	    "uarea", &uvm_uarea_allocator, IPL_NONE, NULL, NULL, NULL);
344 #if defined(__HAVE_CPU_UAREA_ROUTINES)
345 	uvm_uarea_system_cache = pool_cache_init(USPACE, USPACE_ALIGN,
346 	    0, flags, "uareasys", &uvm_uarea_system_allocator,
347 	    IPL_NONE, NULL, NULL, NULL);
348 #endif
349 }
350 
351 /*
352  * uvm_uarea_alloc: allocate a u-area
353  */
354 
355 vaddr_t
356 uvm_uarea_alloc(void)
357 {
358 
359 	return (vaddr_t)pool_cache_get(uvm_uarea_cache, PR_WAITOK);
360 }
361 
362 vaddr_t
363 uvm_uarea_system_alloc(void)
364 {
365 
366 	return (vaddr_t)pool_cache_get(uvm_uarea_system_cache, PR_WAITOK);
367 }
368 
369 /*
370  * uvm_uarea_free: free a u-area
371  */
372 
373 void
374 uvm_uarea_free(vaddr_t uaddr)
375 {
376 
377 	pool_cache_put(uvm_uarea_cache, (void *)uaddr);
378 }
379 
380 void
381 uvm_uarea_system_free(vaddr_t uaddr)
382 {
383 
384 	pool_cache_put(uvm_uarea_system_cache, (void *)uaddr);
385 }
386 
387 vaddr_t
388 uvm_lwp_getuarea(lwp_t *l)
389 {
390 
391 	return (vaddr_t)l->l_addr - UAREA_PCB_OFFSET;
392 }
393 
394 void
395 uvm_lwp_setuarea(lwp_t *l, vaddr_t addr)
396 {
397 
398 	l->l_addr = (void *)(addr + UAREA_PCB_OFFSET);
399 }
400 
401 /*
402  * uvm_proc_exit: exit a virtual address space
403  *
404  * - borrow proc0's address space because freeing the vmspace
405  *   of the dead process may block.
406  */
407 
408 void
409 uvm_proc_exit(struct proc *p)
410 {
411 	struct lwp *l = curlwp; /* XXX */
412 	struct vmspace *ovm;
413 
414 	KASSERT(p == l->l_proc);
415 	ovm = p->p_vmspace;
416 
417 	/*
418 	 * borrow proc0's address space.
419 	 */
420 	KPREEMPT_DISABLE(l);
421 	pmap_deactivate(l);
422 	p->p_vmspace = proc0.p_vmspace;
423 	pmap_activate(l);
424 	KPREEMPT_ENABLE(l);
425 
426 	uvmspace_free(ovm);
427 }
428 
429 void
430 uvm_lwp_exit(struct lwp *l)
431 {
432 	vaddr_t va = uvm_lwp_getuarea(l);
433 	bool system = (l->l_flag & LW_SYSTEM) != 0;
434 
435 	if (system)
436 		uvm_uarea_system_free(va);
437 	else
438 		uvm_uarea_free(va);
439 #ifdef DIAGNOSTIC
440 	uvm_lwp_setuarea(l, (vaddr_t)NULL);
441 #endif
442 }
443 
444 /*
445  * uvm_init_limit: init per-process VM limits
446  *
447  * - called for process 0 and then inherited by all others.
448  */
449 
450 void
451 uvm_init_limits(struct proc *p)
452 {
453 
454 	/*
455 	 * Set up the initial limits on process VM.  Set the maximum
456 	 * resident set size to be all of (reasonably) available memory.
457 	 * This causes any single, large process to start random page
458 	 * replacement once it fills memory.
459 	 */
460 
461 	p->p_rlimit[RLIMIT_STACK].rlim_cur = DFLSSIZ;
462 	p->p_rlimit[RLIMIT_STACK].rlim_max = maxsmap;
463 	p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ;
464 	p->p_rlimit[RLIMIT_DATA].rlim_max = maxdmap;
465 	p->p_rlimit[RLIMIT_AS].rlim_cur = RLIM_INFINITY;
466 	p->p_rlimit[RLIMIT_AS].rlim_max = RLIM_INFINITY;
467 	p->p_rlimit[RLIMIT_RSS].rlim_cur = MIN(
468 	    VM_MAXUSER_ADDRESS, ctob((rlim_t)uvmexp.free));
469 }
470 
471 /*
472  * uvm_scheduler: process zero main loop.
473  */
474 
475 extern struct loadavg averunnable;
476 
477 void
478 uvm_scheduler(void)
479 {
480 	lwp_t *l = curlwp;
481 
482 	lwp_lock(l);
483 	l->l_priority = PRI_VM;
484 	l->l_class = SCHED_FIFO;
485 	lwp_unlock(l);
486 
487 	for (;;) {
488 		sched_pstats();
489 		(void)kpause("uvm", false, hz, NULL);
490 	}
491 }
492