xref: /netbsd-src/sys/uvm/uvm_km.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21)
1 /*	$NetBSD: uvm_km.c,v 1.91 2006/10/12 21:35:00 uwe 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_kern.c   8.3 (Berkeley) 1/12/94
42  * from: Id: uvm_km.c,v 1.1.2.14 1998/02/06 05:19:27 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 /*
70  * uvm_km.c: handle kernel memory allocation and management
71  */
72 
73 /*
74  * overview of kernel memory management:
75  *
76  * the kernel virtual address space is mapped by "kernel_map."   kernel_map
77  * starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS.
78  * note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map).
79  *
80  * the kernel_map has several "submaps."   submaps can only appear in
81  * the kernel_map (user processes can't use them).   submaps "take over"
82  * the management of a sub-range of the kernel's address space.  submaps
83  * are typically allocated at boot time and are never released.   kernel
84  * virtual address space that is mapped by a submap is locked by the
85  * submap's lock -- not the kernel_map's lock.
86  *
87  * thus, the useful feature of submaps is that they allow us to break
88  * up the locking and protection of the kernel address space into smaller
89  * chunks.
90  *
91  * the vm system has several standard kernel submaps, including:
92  *   kmem_map => contains only wired kernel memory for the kernel
93  *		malloc.   *** access to kmem_map must be protected
94  *		by splvm() because we are allowed to call malloc()
95  *		at interrupt time ***
96  *   mb_map => memory for large mbufs,  *** protected by splvm ***
97  *   pager_map => used to map "buf" structures into kernel space
98  *   exec_map => used during exec to handle exec args
99  *   etc...
100  *
101  * the kernel allocates its private memory out of special uvm_objects whose
102  * reference count is set to UVM_OBJ_KERN (thus indicating that the objects
103  * are "special" and never die).   all kernel objects should be thought of
104  * as large, fixed-sized, sparsely populated uvm_objects.   each kernel
105  * object is equal to the size of kernel virtual address space (i.e. the
106  * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
107  *
108  * note that just because a kernel object spans the entire kernel virutal
109  * address space doesn't mean that it has to be mapped into the entire space.
110  * large chunks of a kernel object's space go unused either because
111  * that area of kernel VM is unmapped, or there is some other type of
112  * object mapped into that range (e.g. a vnode).    for submap's kernel
113  * objects, the only part of the object that can ever be populated is the
114  * offsets that are managed by the submap.
115  *
116  * note that the "offset" in a kernel object is always the kernel virtual
117  * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
118  * example:
119  *   suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
120  *   uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
121  *   kernel map].    if uvm_km_alloc returns virtual address 0xf8235000,
122  *   then that means that the page at offset 0x235000 in kernel_object is
123  *   mapped at 0xf8235000.
124  *
125  * kernel object have one other special property: when the kernel virtual
126  * memory mapping them is unmapped, the backing memory in the object is
127  * freed right away.   this is done with the uvm_km_pgremove() function.
128  * this has to be done because there is no backing store for kernel pages
129  * and no need to save them after they are no longer referenced.
130  */
131 
132 #include <sys/cdefs.h>
133 __KERNEL_RCSID(0, "$NetBSD: uvm_km.c,v 1.91 2006/10/12 21:35:00 uwe Exp $");
134 
135 #include "opt_uvmhist.h"
136 
137 #include <sys/param.h>
138 #include <sys/malloc.h>
139 #include <sys/systm.h>
140 #include <sys/proc.h>
141 #include <sys/pool.h>
142 
143 #include <uvm/uvm.h>
144 
145 /*
146  * global data structures
147  */
148 
149 struct vm_map *kernel_map = NULL;
150 
151 /*
152  * local data structues
153  */
154 
155 static struct vm_map_kernel	kernel_map_store;
156 static struct vm_map_entry	kernel_first_mapent_store;
157 
158 #if !defined(PMAP_MAP_POOLPAGE)
159 
160 /*
161  * kva cache
162  *
163  * XXX maybe it's better to do this at the uvm_map layer.
164  */
165 
166 #define	KM_VACACHE_SIZE	(32 * PAGE_SIZE) /* XXX tune */
167 
168 static void *km_vacache_alloc(struct pool *, int);
169 static void km_vacache_free(struct pool *, void *);
170 static void km_vacache_init(struct vm_map *, const char *, size_t);
171 
172 /* XXX */
173 #define	KM_VACACHE_POOL_TO_MAP(pp) \
174 	((struct vm_map *)((char *)(pp) - \
175 	    offsetof(struct vm_map_kernel, vmk_vacache)))
176 
177 static void *
178 km_vacache_alloc(struct pool *pp, int flags)
179 {
180 	vaddr_t va;
181 	size_t size;
182 	struct vm_map *map;
183 	size = pp->pr_alloc->pa_pagesz;
184 
185 	map = KM_VACACHE_POOL_TO_MAP(pp);
186 
187 	va = vm_map_min(map); /* hint */
188 	if (uvm_map(map, &va, size, NULL, UVM_UNKNOWN_OFFSET, size,
189 	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
190 	    UVM_ADV_RANDOM, UVM_FLAG_QUANTUM |
191 	    ((flags & PR_WAITOK) ? UVM_FLAG_WAITVA :
192 	    UVM_FLAG_TRYLOCK | UVM_FLAG_NOWAIT))))
193 		return NULL;
194 
195 	return (void *)va;
196 }
197 
198 static void
199 km_vacache_free(struct pool *pp, void *v)
200 {
201 	vaddr_t va = (vaddr_t)v;
202 	size_t size = pp->pr_alloc->pa_pagesz;
203 	struct vm_map *map;
204 
205 	map = KM_VACACHE_POOL_TO_MAP(pp);
206 	uvm_unmap1(map, va, va + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
207 }
208 
209 /*
210  * km_vacache_init: initialize kva cache.
211  */
212 
213 static void
214 km_vacache_init(struct vm_map *map, const char *name, size_t size)
215 {
216 	struct vm_map_kernel *vmk;
217 	struct pool *pp;
218 	struct pool_allocator *pa;
219 
220 	KASSERT(VM_MAP_IS_KERNEL(map));
221 	KASSERT(size < (vm_map_max(map) - vm_map_min(map)) / 2); /* sanity */
222 
223 	vmk = vm_map_to_kernel(map);
224 	pp = &vmk->vmk_vacache;
225 	pa = &vmk->vmk_vacache_allocator;
226 	memset(pa, 0, sizeof(*pa));
227 	pa->pa_alloc = km_vacache_alloc;
228 	pa->pa_free = km_vacache_free;
229 	pa->pa_pagesz = (unsigned int)size;
230 	pa->pa_backingmap = map;
231 	pa->pa_backingmapptr = NULL;
232 	pool_init(pp, PAGE_SIZE, 0, 0, PR_NOTOUCH | PR_RECURSIVE, name, pa);
233 }
234 
235 void
236 uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
237 {
238 
239 	map->flags |= VM_MAP_VACACHE;
240 	if (size == 0)
241 		size = KM_VACACHE_SIZE;
242 	km_vacache_init(map, name, size);
243 }
244 
245 #else /* !defined(PMAP_MAP_POOLPAGE) */
246 
247 void
248 uvm_km_vacache_init(struct vm_map *map __unused, const char *name __unused,
249 		    size_t size __unused)
250 {
251 
252 	/* nothing */
253 }
254 
255 #endif /* !defined(PMAP_MAP_POOLPAGE) */
256 
257 void
258 uvm_km_va_drain(struct vm_map *map, uvm_flag_t flags __unused)
259 {
260 	struct vm_map_kernel *vmk = vm_map_to_kernel(map);
261 	const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
262 	int s = 0xdeadbeaf; /* XXX: gcc */
263 
264 	if (intrsafe) {
265 		s = splvm();
266 	}
267 	callback_run_roundrobin(&vmk->vmk_reclaim_callback, NULL);
268 	if (intrsafe) {
269 		splx(s);
270 	}
271 }
272 
273 /*
274  * uvm_km_init: init kernel maps and objects to reflect reality (i.e.
275  * KVM already allocated for text, data, bss, and static data structures).
276  *
277  * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
278  *    we assume that [vmin -> start] has already been allocated and that
279  *    "end" is the end.
280  */
281 
282 void
283 uvm_km_init(vaddr_t start, vaddr_t end)
284 {
285 	vaddr_t base = VM_MIN_KERNEL_ADDRESS;
286 
287 	/*
288 	 * next, init kernel memory objects.
289 	 */
290 
291 	/* kernel_object: for pageable anonymous kernel memory */
292 	uao_init();
293 	uvm.kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
294 				 VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
295 
296 	/*
297 	 * init the map and reserve any space that might already
298 	 * have been allocated kernel space before installing.
299 	 */
300 
301 	uvm_map_setup_kernel(&kernel_map_store, base, end, VM_MAP_PAGEABLE);
302 	kernel_map_store.vmk_map.pmap = pmap_kernel();
303 	if (start != base) {
304 		int error;
305 		struct uvm_map_args args;
306 
307 		error = uvm_map_prepare(&kernel_map_store.vmk_map,
308 		    base, start - base,
309 		    NULL, UVM_UNKNOWN_OFFSET, 0,
310 		    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
311 		    		UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args);
312 		if (!error) {
313 			kernel_first_mapent_store.flags =
314 			    UVM_MAP_KERNEL | UVM_MAP_FIRST;
315 			error = uvm_map_enter(&kernel_map_store.vmk_map, &args,
316 			    &kernel_first_mapent_store);
317 		}
318 
319 		if (error)
320 			panic(
321 			    "uvm_km_init: could not reserve space for kernel");
322 	}
323 
324 	/*
325 	 * install!
326 	 */
327 
328 	kernel_map = &kernel_map_store.vmk_map;
329 	uvm_km_vacache_init(kernel_map, "kvakernel", 0);
330 }
331 
332 /*
333  * uvm_km_suballoc: allocate a submap in the kernel map.   once a submap
334  * is allocated all references to that area of VM must go through it.  this
335  * allows the locking of VAs in kernel_map to be broken up into regions.
336  *
337  * => if `fixed' is true, *vmin specifies where the region described
338  *      by the submap must start
339  * => if submap is non NULL we use that as the submap, otherwise we
340  *	alloc a new map
341  */
342 
343 struct vm_map *
344 uvm_km_suballoc(struct vm_map *map, vaddr_t *vmin /* IN/OUT */,
345     vaddr_t *vmax /* OUT */, vsize_t size, int flags, boolean_t fixed,
346     struct vm_map_kernel *submap)
347 {
348 	int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
349 
350 	KASSERT(vm_map_pmap(map) == pmap_kernel());
351 
352 	size = round_page(size);	/* round up to pagesize */
353 	size += uvm_mapent_overhead(size, flags);
354 
355 	/*
356 	 * first allocate a blank spot in the parent map
357 	 */
358 
359 	if (uvm_map(map, vmin, size, NULL, UVM_UNKNOWN_OFFSET, 0,
360 	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
361 	    UVM_ADV_RANDOM, mapflags)) != 0) {
362 	       panic("uvm_km_suballoc: unable to allocate space in parent map");
363 	}
364 
365 	/*
366 	 * set VM bounds (vmin is filled in by uvm_map)
367 	 */
368 
369 	*vmax = *vmin + size;
370 
371 	/*
372 	 * add references to pmap and create or init the submap
373 	 */
374 
375 	pmap_reference(vm_map_pmap(map));
376 	if (submap == NULL) {
377 		submap = malloc(sizeof(*submap), M_VMMAP, M_WAITOK);
378 		if (submap == NULL)
379 			panic("uvm_km_suballoc: unable to create submap");
380 	}
381 	uvm_map_setup_kernel(submap, *vmin, *vmax, flags);
382 	submap->vmk_map.pmap = vm_map_pmap(map);
383 
384 	/*
385 	 * now let uvm_map_submap plug in it...
386 	 */
387 
388 	if (uvm_map_submap(map, *vmin, *vmax, &submap->vmk_map) != 0)
389 		panic("uvm_km_suballoc: submap allocation failed");
390 
391 	return(&submap->vmk_map);
392 }
393 
394 /*
395  * uvm_km_pgremove: remove pages from a kernel uvm_object.
396  *
397  * => when you unmap a part of anonymous kernel memory you want to toss
398  *    the pages right away.    (this gets called from uvm_unmap_...).
399  */
400 
401 void
402 uvm_km_pgremove(vaddr_t startva, vaddr_t endva)
403 {
404 	struct uvm_object * const uobj = uvm.kernel_object;
405 	const voff_t start = startva - vm_map_min(kernel_map);
406 	const voff_t end = endva - vm_map_min(kernel_map);
407 	struct vm_page *pg;
408 	voff_t curoff, nextoff;
409 	int swpgonlydelta = 0;
410 	UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
411 
412 	KASSERT(VM_MIN_KERNEL_ADDRESS <= startva);
413 	KASSERT(startva < endva);
414 	KASSERT(endva <= VM_MAX_KERNEL_ADDRESS);
415 
416 	simple_lock(&uobj->vmobjlock);
417 
418 	for (curoff = start; curoff < end; curoff = nextoff) {
419 		nextoff = curoff + PAGE_SIZE;
420 		pg = uvm_pagelookup(uobj, curoff);
421 		if (pg != NULL && pg->flags & PG_BUSY) {
422 			pg->flags |= PG_WANTED;
423 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
424 				    "km_pgrm", 0);
425 			simple_lock(&uobj->vmobjlock);
426 			nextoff = curoff;
427 			continue;
428 		}
429 
430 		/*
431 		 * free the swap slot, then the page.
432 		 */
433 
434 		if (pg == NULL &&
435 		    uao_find_swslot(uobj, curoff >> PAGE_SHIFT) > 0) {
436 			swpgonlydelta++;
437 		}
438 		uao_dropswap(uobj, curoff >> PAGE_SHIFT);
439 		if (pg != NULL) {
440 			uvm_lock_pageq();
441 			uvm_pagefree(pg);
442 			uvm_unlock_pageq();
443 		}
444 	}
445 	simple_unlock(&uobj->vmobjlock);
446 
447 	if (swpgonlydelta > 0) {
448 		simple_lock(&uvm.swap_data_lock);
449 		KASSERT(uvmexp.swpgonly >= swpgonlydelta);
450 		uvmexp.swpgonly -= swpgonlydelta;
451 		simple_unlock(&uvm.swap_data_lock);
452 	}
453 }
454 
455 
456 /*
457  * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for non object backed
458  *    regions.
459  *
460  * => when you unmap a part of anonymous kernel memory you want to toss
461  *    the pages right away.    (this is called from uvm_unmap_...).
462  * => none of the pages will ever be busy, and none of them will ever
463  *    be on the active or inactive queues (because they have no object).
464  */
465 
466 void
467 uvm_km_pgremove_intrsafe(vaddr_t start, vaddr_t end)
468 {
469 	struct vm_page *pg;
470 	paddr_t pa;
471 	UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist);
472 
473 	KASSERT(VM_MIN_KERNEL_ADDRESS <= start);
474 	KASSERT(start < end);
475 	KASSERT(end <= VM_MAX_KERNEL_ADDRESS);
476 
477 	for (; start < end; start += PAGE_SIZE) {
478 		if (!pmap_extract(pmap_kernel(), start, &pa)) {
479 			continue;
480 		}
481 		pg = PHYS_TO_VM_PAGE(pa);
482 		KASSERT(pg);
483 		KASSERT(pg->uobject == NULL && pg->uanon == NULL);
484 		uvm_pagefree(pg);
485 	}
486 }
487 
488 #if defined(DEBUG)
489 void
490 uvm_km_check_empty(vaddr_t start, vaddr_t end, boolean_t intrsafe)
491 {
492 	vaddr_t va;
493 	paddr_t pa;
494 
495 	KDASSERT(VM_MIN_KERNEL_ADDRESS <= start);
496 	KDASSERT(start < end);
497 	KDASSERT(end <= VM_MAX_KERNEL_ADDRESS);
498 
499 	for (va = start; va < end; va += PAGE_SIZE) {
500 		if (pmap_extract(pmap_kernel(), va, &pa)) {
501 			panic("uvm_km_check_empty: va %p has pa 0x%llx",
502 			    (void *)va, (long long)pa);
503 		}
504 		if (!intrsafe) {
505 			const struct vm_page *pg;
506 
507 			simple_lock(&uvm.kernel_object->vmobjlock);
508 			pg = uvm_pagelookup(uvm.kernel_object,
509 			    va - vm_map_min(kernel_map));
510 			simple_unlock(&uvm.kernel_object->vmobjlock);
511 			if (pg) {
512 				panic("uvm_km_check_empty: "
513 				    "has page hashed at %p", (const void *)va);
514 			}
515 		}
516 	}
517 }
518 #endif /* defined(DEBUG) */
519 
520 /*
521  * uvm_km_alloc: allocate an area of kernel memory.
522  *
523  * => NOTE: we can return 0 even if we can wait if there is not enough
524  *	free VM space in the map... caller should be prepared to handle
525  *	this case.
526  * => we return KVA of memory allocated
527  */
528 
529 vaddr_t
530 uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags)
531 {
532 	vaddr_t kva, loopva;
533 	vaddr_t offset;
534 	vsize_t loopsize;
535 	struct vm_page *pg;
536 	struct uvm_object *obj;
537 	int pgaflags;
538 	vm_prot_t prot;
539 	UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist);
540 
541 	KASSERT(vm_map_pmap(map) == pmap_kernel());
542 	KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
543 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
544 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
545 
546 	/*
547 	 * setup for call
548 	 */
549 
550 	kva = vm_map_min(map);	/* hint */
551 	size = round_page(size);
552 	obj = (flags & UVM_KMF_PAGEABLE) ? uvm.kernel_object : NULL;
553 	UVMHIST_LOG(maphist,"  (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
554 		    map, obj, size, flags);
555 
556 	/*
557 	 * allocate some virtual space
558 	 */
559 
560 	if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
561 	    align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
562 	    UVM_ADV_RANDOM,
563 	    (flags & (UVM_KMF_TRYLOCK | UVM_KMF_NOWAIT | UVM_KMF_WAITVA))
564 	    | UVM_FLAG_QUANTUM)) != 0)) {
565 		UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
566 		return(0);
567 	}
568 
569 	/*
570 	 * if all we wanted was VA, return now
571 	 */
572 
573 	if (flags & (UVM_KMF_VAONLY | UVM_KMF_PAGEABLE)) {
574 		UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
575 		return(kva);
576 	}
577 
578 	/*
579 	 * recover object offset from virtual address
580 	 */
581 
582 	offset = kva - vm_map_min(kernel_map);
583 	UVMHIST_LOG(maphist, "  kva=0x%x, offset=0x%x", kva, offset,0,0);
584 
585 	/*
586 	 * now allocate and map in the memory... note that we are the only ones
587 	 * whom should ever get a handle on this area of VM.
588 	 */
589 
590 	loopva = kva;
591 	loopsize = size;
592 
593 	pgaflags = UVM_PGA_USERESERVE;
594 	if (flags & UVM_KMF_ZERO)
595 		pgaflags |= UVM_PGA_ZERO;
596 	prot = VM_PROT_READ | VM_PROT_WRITE;
597 	if (flags & UVM_KMF_EXEC)
598 		prot |= VM_PROT_EXECUTE;
599 	while (loopsize) {
600 		KASSERT(!pmap_extract(pmap_kernel(), loopva, NULL));
601 
602 		pg = uvm_pagealloc(NULL, offset, NULL, pgaflags);
603 
604 		/*
605 		 * out of memory?
606 		 */
607 
608 		if (__predict_false(pg == NULL)) {
609 			if ((flags & UVM_KMF_NOWAIT) ||
610 			    ((flags & UVM_KMF_CANFAIL) && !uvm_reclaimable())) {
611 				/* free everything! */
612 				uvm_km_free(map, kva, size,
613 				    flags & UVM_KMF_TYPEMASK);
614 				return (0);
615 			} else {
616 				uvm_wait("km_getwait2");	/* sleep here */
617 				continue;
618 			}
619 		}
620 
621 		pg->flags &= ~PG_BUSY;	/* new page */
622 		UVM_PAGE_OWN(pg, NULL);
623 
624 		/*
625 		 * map it in
626 		 */
627 
628 		pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), prot);
629 		loopva += PAGE_SIZE;
630 		offset += PAGE_SIZE;
631 		loopsize -= PAGE_SIZE;
632 	}
633 
634        	pmap_update(pmap_kernel());
635 
636 	UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
637 	return(kva);
638 }
639 
640 /*
641  * uvm_km_free: free an area of kernel memory
642  */
643 
644 void
645 uvm_km_free(struct vm_map *map, vaddr_t addr, vsize_t size, uvm_flag_t flags)
646 {
647 
648 	KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
649 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
650 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
651 	KASSERT((addr & PAGE_MASK) == 0);
652 	KASSERT(vm_map_pmap(map) == pmap_kernel());
653 
654 	size = round_page(size);
655 
656 	if (flags & UVM_KMF_PAGEABLE) {
657 		uvm_km_pgremove(addr, addr + size);
658 		pmap_remove(pmap_kernel(), addr, addr + size);
659 	} else if (flags & UVM_KMF_WIRED) {
660 		uvm_km_pgremove_intrsafe(addr, addr + size);
661 		pmap_kremove(addr, size);
662 	}
663 
664 	uvm_unmap1(map, addr, addr + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
665 }
666 
667 /* Sanity; must specify both or none. */
668 #if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \
669     (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE))
670 #error Must specify MAP and UNMAP together.
671 #endif
672 
673 /*
674  * uvm_km_alloc_poolpage: allocate a page for the pool allocator
675  *
676  * => if the pmap specifies an alternate mapping method, we use it.
677  */
678 
679 /* ARGSUSED */
680 vaddr_t
681 uvm_km_alloc_poolpage_cache(struct vm_map *map, boolean_t waitok)
682 {
683 #if defined(PMAP_MAP_POOLPAGE)
684 	return uvm_km_alloc_poolpage(map, waitok);
685 #else
686 	struct vm_page *pg;
687 	struct pool *pp = &vm_map_to_kernel(map)->vmk_vacache;
688 	vaddr_t va;
689 	int s = 0xdeadbeaf; /* XXX: gcc */
690 	const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
691 
692 	if ((map->flags & VM_MAP_VACACHE) == 0)
693 		return uvm_km_alloc_poolpage(map, waitok);
694 
695 	if (intrsafe)
696 		s = splvm();
697 	va = (vaddr_t)pool_get(pp, waitok ? PR_WAITOK : PR_NOWAIT);
698 	if (intrsafe)
699 		splx(s);
700 	if (va == 0)
701 		return 0;
702 	KASSERT(!pmap_extract(pmap_kernel(), va, NULL));
703 again:
704 	pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
705 	if (__predict_false(pg == NULL)) {
706 		if (waitok) {
707 			uvm_wait("plpg");
708 			goto again;
709 		} else {
710 			if (intrsafe)
711 				s = splvm();
712 			pool_put(pp, (void *)va);
713 			if (intrsafe)
714 				splx(s);
715 			return 0;
716 		}
717 	}
718 	pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE);
719 	pmap_update(pmap_kernel());
720 
721 	return va;
722 #endif /* PMAP_MAP_POOLPAGE */
723 }
724 
725 vaddr_t
726 uvm_km_alloc_poolpage(struct vm_map *map, boolean_t waitok)
727 {
728 #if defined(PMAP_MAP_POOLPAGE)
729 	struct vm_page *pg;
730 	vaddr_t va;
731 
732  again:
733 	pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
734 	if (__predict_false(pg == NULL)) {
735 		if (waitok) {
736 			uvm_wait("plpg");
737 			goto again;
738 		} else
739 			return (0);
740 	}
741 	va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
742 	if (__predict_false(va == 0))
743 		uvm_pagefree(pg);
744 	return (va);
745 #else
746 	vaddr_t va;
747 	int s = 0xdeadbeaf; /* XXX: gcc */
748 	const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
749 
750 	if (intrsafe)
751 		s = splvm();
752 	va = uvm_km_alloc(map, PAGE_SIZE, 0,
753 	    (waitok ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK) | UVM_KMF_WIRED);
754 	if (intrsafe)
755 		splx(s);
756 	return (va);
757 #endif /* PMAP_MAP_POOLPAGE */
758 }
759 
760 /*
761  * uvm_km_free_poolpage: free a previously allocated pool page
762  *
763  * => if the pmap specifies an alternate unmapping method, we use it.
764  */
765 
766 /* ARGSUSED */
767 void
768 uvm_km_free_poolpage_cache(struct vm_map *map, vaddr_t addr)
769 {
770 #if defined(PMAP_UNMAP_POOLPAGE)
771 	uvm_km_free_poolpage(map, addr);
772 #else
773 	struct pool *pp;
774 	int s = 0xdeadbeaf; /* XXX: gcc */
775 	const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
776 
777 	if ((map->flags & VM_MAP_VACACHE) == 0) {
778 		uvm_km_free_poolpage(map, addr);
779 		return;
780 	}
781 
782 	KASSERT(pmap_extract(pmap_kernel(), addr, NULL));
783 	uvm_km_pgremove_intrsafe(addr, addr + PAGE_SIZE);
784 	pmap_kremove(addr, PAGE_SIZE);
785 #if defined(DEBUG)
786 	pmap_update(pmap_kernel());
787 #endif
788 	KASSERT(!pmap_extract(pmap_kernel(), addr, NULL));
789 	pp = &vm_map_to_kernel(map)->vmk_vacache;
790 	if (intrsafe)
791 		s = splvm();
792 	pool_put(pp, (void *)addr);
793 	if (intrsafe)
794 		splx(s);
795 #endif
796 }
797 
798 /* ARGSUSED */
799 void
800 uvm_km_free_poolpage(struct vm_map *map, vaddr_t addr)
801 {
802 #if defined(PMAP_UNMAP_POOLPAGE)
803 	paddr_t pa;
804 
805 	pa = PMAP_UNMAP_POOLPAGE(addr);
806 	uvm_pagefree(PHYS_TO_VM_PAGE(pa));
807 #else
808 	int s = 0xdeadbeaf; /* XXX: gcc */
809 	const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
810 
811 	if (intrsafe)
812 		s = splvm();
813 	uvm_km_free(map, addr, PAGE_SIZE, UVM_KMF_WIRED);
814 	if (intrsafe)
815 		splx(s);
816 #endif /* PMAP_UNMAP_POOLPAGE */
817 }
818