xref: /netbsd-src/sys/uvm/uvm_km.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: uvm_km.c,v 1.106 2010/05/14 05:02:06 cegger 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.
94  *   pager_map => used to map "buf" structures into kernel space
95  *   exec_map => used during exec to handle exec args
96  *   etc...
97  *
98  * the kernel allocates its private memory out of special uvm_objects whose
99  * reference count is set to UVM_OBJ_KERN (thus indicating that the objects
100  * are "special" and never die).   all kernel objects should be thought of
101  * as large, fixed-sized, sparsely populated uvm_objects.   each kernel
102  * object is equal to the size of kernel virtual address space (i.e. the
103  * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
104  *
105  * note that just because a kernel object spans the entire kernel virtual
106  * address space doesn't mean that it has to be mapped into the entire space.
107  * large chunks of a kernel object's space go unused either because
108  * that area of kernel VM is unmapped, or there is some other type of
109  * object mapped into that range (e.g. a vnode).    for submap's kernel
110  * objects, the only part of the object that can ever be populated is the
111  * offsets that are managed by the submap.
112  *
113  * note that the "offset" in a kernel object is always the kernel virtual
114  * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
115  * example:
116  *   suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
117  *   uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
118  *   kernel map].    if uvm_km_alloc returns virtual address 0xf8235000,
119  *   then that means that the page at offset 0x235000 in kernel_object is
120  *   mapped at 0xf8235000.
121  *
122  * kernel object have one other special property: when the kernel virtual
123  * memory mapping them is unmapped, the backing memory in the object is
124  * freed right away.   this is done with the uvm_km_pgremove() function.
125  * this has to be done because there is no backing store for kernel pages
126  * and no need to save them after they are no longer referenced.
127  */
128 
129 #include <sys/cdefs.h>
130 __KERNEL_RCSID(0, "$NetBSD: uvm_km.c,v 1.106 2010/05/14 05:02:06 cegger Exp $");
131 
132 #include "opt_uvmhist.h"
133 
134 #include <sys/param.h>
135 #include <sys/malloc.h>
136 #include <sys/systm.h>
137 #include <sys/proc.h>
138 #include <sys/pool.h>
139 
140 #include <uvm/uvm.h>
141 
142 /*
143  * global data structures
144  */
145 
146 struct vm_map *kernel_map = NULL;
147 
148 /*
149  * local data structues
150  */
151 
152 static struct vm_map_kernel	kernel_map_store;
153 static struct vm_map_entry	kernel_first_mapent_store;
154 
155 #if !defined(PMAP_MAP_POOLPAGE)
156 
157 /*
158  * kva cache
159  *
160  * XXX maybe it's better to do this at the uvm_map layer.
161  */
162 
163 #define	KM_VACACHE_SIZE	(32 * PAGE_SIZE) /* XXX tune */
164 
165 static void *km_vacache_alloc(struct pool *, int);
166 static void km_vacache_free(struct pool *, void *);
167 static void km_vacache_init(struct vm_map *, const char *, size_t);
168 
169 /* XXX */
170 #define	KM_VACACHE_POOL_TO_MAP(pp) \
171 	((struct vm_map *)((char *)(pp) - \
172 	    offsetof(struct vm_map_kernel, vmk_vacache)))
173 
174 static void *
175 km_vacache_alloc(struct pool *pp, int flags)
176 {
177 	vaddr_t va;
178 	size_t size;
179 	struct vm_map *map;
180 	size = pp->pr_alloc->pa_pagesz;
181 
182 	map = KM_VACACHE_POOL_TO_MAP(pp);
183 
184 	va = vm_map_min(map); /* hint */
185 	if (uvm_map(map, &va, size, NULL, UVM_UNKNOWN_OFFSET, size,
186 	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
187 	    UVM_ADV_RANDOM, UVM_FLAG_QUANTUM |
188 	    ((flags & PR_WAITOK) ? UVM_FLAG_WAITVA :
189 	    UVM_FLAG_TRYLOCK | UVM_FLAG_NOWAIT))))
190 		return NULL;
191 
192 	return (void *)va;
193 }
194 
195 static void
196 km_vacache_free(struct pool *pp, void *v)
197 {
198 	vaddr_t va = (vaddr_t)v;
199 	size_t size = pp->pr_alloc->pa_pagesz;
200 	struct vm_map *map;
201 
202 	map = KM_VACACHE_POOL_TO_MAP(pp);
203 	uvm_unmap1(map, va, va + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
204 }
205 
206 /*
207  * km_vacache_init: initialize kva cache.
208  */
209 
210 static void
211 km_vacache_init(struct vm_map *map, const char *name, size_t size)
212 {
213 	struct vm_map_kernel *vmk;
214 	struct pool *pp;
215 	struct pool_allocator *pa;
216 	int ipl;
217 
218 	KASSERT(VM_MAP_IS_KERNEL(map));
219 	KASSERT(size < (vm_map_max(map) - vm_map_min(map)) / 2); /* sanity */
220 
221 
222 	vmk = vm_map_to_kernel(map);
223 	pp = &vmk->vmk_vacache;
224 	pa = &vmk->vmk_vacache_allocator;
225 	memset(pa, 0, sizeof(*pa));
226 	pa->pa_alloc = km_vacache_alloc;
227 	pa->pa_free = km_vacache_free;
228 	pa->pa_pagesz = (unsigned int)size;
229 	pa->pa_backingmap = map;
230 	pa->pa_backingmapptr = NULL;
231 
232 	if ((map->flags & VM_MAP_INTRSAFE) != 0)
233 		ipl = IPL_VM;
234 	else
235 		ipl = IPL_NONE;
236 
237 	pool_init(pp, PAGE_SIZE, 0, 0, PR_NOTOUCH | PR_RECURSIVE, name, pa,
238 	    ipl);
239 }
240 
241 void
242 uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
243 {
244 
245 	map->flags |= VM_MAP_VACACHE;
246 	if (size == 0)
247 		size = KM_VACACHE_SIZE;
248 	km_vacache_init(map, name, size);
249 }
250 
251 #else /* !defined(PMAP_MAP_POOLPAGE) */
252 
253 void
254 uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
255 {
256 
257 	/* nothing */
258 }
259 
260 #endif /* !defined(PMAP_MAP_POOLPAGE) */
261 
262 void
263 uvm_km_va_drain(struct vm_map *map, uvm_flag_t flags)
264 {
265 	struct vm_map_kernel *vmk = vm_map_to_kernel(map);
266 
267 	callback_run_roundrobin(&vmk->vmk_reclaim_callback, NULL);
268 }
269 
270 /*
271  * uvm_km_init: init kernel maps and objects to reflect reality (i.e.
272  * KVM already allocated for text, data, bss, and static data structures).
273  *
274  * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
275  *    we assume that [vmin -> start] has already been allocated and that
276  *    "end" is the end.
277  */
278 
279 void
280 uvm_km_init(vaddr_t start, vaddr_t end)
281 {
282 	vaddr_t base = VM_MIN_KERNEL_ADDRESS;
283 
284 	/*
285 	 * next, init kernel memory objects.
286 	 */
287 
288 	/* kernel_object: for pageable anonymous kernel memory */
289 	uao_init();
290 	uvm_kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
291 				 VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
292 
293 	/*
294 	 * init the map and reserve any space that might already
295 	 * have been allocated kernel space before installing.
296 	 */
297 
298 	uvm_map_setup_kernel(&kernel_map_store, base, end, VM_MAP_PAGEABLE);
299 	kernel_map_store.vmk_map.pmap = pmap_kernel();
300 	if (start != base) {
301 		int error;
302 		struct uvm_map_args args;
303 
304 		error = uvm_map_prepare(&kernel_map_store.vmk_map,
305 		    base, start - base,
306 		    NULL, UVM_UNKNOWN_OFFSET, 0,
307 		    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
308 		    		UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args);
309 		if (!error) {
310 			kernel_first_mapent_store.flags =
311 			    UVM_MAP_KERNEL | UVM_MAP_FIRST;
312 			error = uvm_map_enter(&kernel_map_store.vmk_map, &args,
313 			    &kernel_first_mapent_store);
314 		}
315 
316 		if (error)
317 			panic(
318 			    "uvm_km_init: could not reserve space for kernel");
319 	}
320 
321 	/*
322 	 * install!
323 	 */
324 
325 	kernel_map = &kernel_map_store.vmk_map;
326 	uvm_km_vacache_init(kernel_map, "kvakernel", 0);
327 }
328 
329 /*
330  * uvm_km_suballoc: allocate a submap in the kernel map.   once a submap
331  * is allocated all references to that area of VM must go through it.  this
332  * allows the locking of VAs in kernel_map to be broken up into regions.
333  *
334  * => if `fixed' is true, *vmin specifies where the region described
335  *      by the submap must start
336  * => if submap is non NULL we use that as the submap, otherwise we
337  *	alloc a new map
338  */
339 
340 struct vm_map *
341 uvm_km_suballoc(struct vm_map *map, vaddr_t *vmin /* IN/OUT */,
342     vaddr_t *vmax /* OUT */, vsize_t size, int flags, bool fixed,
343     struct vm_map_kernel *submap)
344 {
345 	int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
346 
347 	KASSERT(vm_map_pmap(map) == pmap_kernel());
348 
349 	size = round_page(size);	/* round up to pagesize */
350 	size += uvm_mapent_overhead(size, flags);
351 
352 	/*
353 	 * first allocate a blank spot in the parent map
354 	 */
355 
356 	if (uvm_map(map, vmin, size, NULL, UVM_UNKNOWN_OFFSET, 0,
357 	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
358 	    UVM_ADV_RANDOM, mapflags)) != 0) {
359 	       panic("uvm_km_suballoc: unable to allocate space in parent map");
360 	}
361 
362 	/*
363 	 * set VM bounds (vmin is filled in by uvm_map)
364 	 */
365 
366 	*vmax = *vmin + size;
367 
368 	/*
369 	 * add references to pmap and create or init the submap
370 	 */
371 
372 	pmap_reference(vm_map_pmap(map));
373 	if (submap == NULL) {
374 		submap = malloc(sizeof(*submap), M_VMMAP, M_WAITOK);
375 		if (submap == NULL)
376 			panic("uvm_km_suballoc: unable to create submap");
377 	}
378 	uvm_map_setup_kernel(submap, *vmin, *vmax, flags);
379 	submap->vmk_map.pmap = vm_map_pmap(map);
380 
381 	/*
382 	 * now let uvm_map_submap plug in it...
383 	 */
384 
385 	if (uvm_map_submap(map, *vmin, *vmax, &submap->vmk_map) != 0)
386 		panic("uvm_km_suballoc: submap allocation failed");
387 
388 	return(&submap->vmk_map);
389 }
390 
391 /*
392  * uvm_km_pgremove: remove pages from a kernel uvm_object.
393  *
394  * => when you unmap a part of anonymous kernel memory you want to toss
395  *    the pages right away.    (this gets called from uvm_unmap_...).
396  */
397 
398 void
399 uvm_km_pgremove(vaddr_t startva, vaddr_t endva)
400 {
401 	struct uvm_object * const uobj = uvm_kernel_object;
402 	const voff_t start = startva - vm_map_min(kernel_map);
403 	const voff_t end = endva - vm_map_min(kernel_map);
404 	struct vm_page *pg;
405 	voff_t curoff, nextoff;
406 	int swpgonlydelta = 0;
407 	UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
408 
409 	KASSERT(VM_MIN_KERNEL_ADDRESS <= startva);
410 	KASSERT(startva < endva);
411 	KASSERT(endva <= VM_MAX_KERNEL_ADDRESS);
412 
413 	mutex_enter(&uobj->vmobjlock);
414 
415 	for (curoff = start; curoff < end; curoff = nextoff) {
416 		nextoff = curoff + PAGE_SIZE;
417 		pg = uvm_pagelookup(uobj, curoff);
418 		if (pg != NULL && pg->flags & PG_BUSY) {
419 			pg->flags |= PG_WANTED;
420 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
421 				    "km_pgrm", 0);
422 			mutex_enter(&uobj->vmobjlock);
423 			nextoff = curoff;
424 			continue;
425 		}
426 
427 		/*
428 		 * free the swap slot, then the page.
429 		 */
430 
431 		if (pg == NULL &&
432 		    uao_find_swslot(uobj, curoff >> PAGE_SHIFT) > 0) {
433 			swpgonlydelta++;
434 		}
435 		uao_dropswap(uobj, curoff >> PAGE_SHIFT);
436 		if (pg != NULL) {
437 			mutex_enter(&uvm_pageqlock);
438 			uvm_pagefree(pg);
439 			mutex_exit(&uvm_pageqlock);
440 		}
441 	}
442 	mutex_exit(&uobj->vmobjlock);
443 
444 	if (swpgonlydelta > 0) {
445 		mutex_enter(&uvm_swap_data_lock);
446 		KASSERT(uvmexp.swpgonly >= swpgonlydelta);
447 		uvmexp.swpgonly -= swpgonlydelta;
448 		mutex_exit(&uvm_swap_data_lock);
449 	}
450 }
451 
452 
453 /*
454  * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for non object backed
455  *    regions.
456  *
457  * => when you unmap a part of anonymous kernel memory you want to toss
458  *    the pages right away.    (this is called from uvm_unmap_...).
459  * => none of the pages will ever be busy, and none of them will ever
460  *    be on the active or inactive queues (because they have no object).
461  */
462 
463 void
464 uvm_km_pgremove_intrsafe(struct vm_map *map, vaddr_t start, vaddr_t end)
465 {
466 	struct vm_page *pg;
467 	paddr_t pa;
468 	UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist);
469 
470 	KASSERT(VM_MAP_IS_KERNEL(map));
471 	KASSERT(vm_map_min(map) <= start);
472 	KASSERT(start < end);
473 	KASSERT(end <= vm_map_max(map));
474 
475 	for (; start < end; start += PAGE_SIZE) {
476 		if (!pmap_extract(pmap_kernel(), start, &pa)) {
477 			continue;
478 		}
479 		pg = PHYS_TO_VM_PAGE(pa);
480 		KASSERT(pg);
481 		KASSERT(pg->uobject == NULL && pg->uanon == NULL);
482 		uvm_pagefree(pg);
483 	}
484 }
485 
486 #if defined(DEBUG)
487 void
488 uvm_km_check_empty(struct vm_map *map, vaddr_t start, vaddr_t end)
489 {
490 	struct vm_page *pg;
491 	vaddr_t va;
492 	paddr_t pa;
493 
494 	KDASSERT(VM_MAP_IS_KERNEL(map));
495 	KDASSERT(vm_map_min(map) <= start);
496 	KDASSERT(start < end);
497 	KDASSERT(end <= vm_map_max(map));
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 ((map->flags & VM_MAP_INTRSAFE) == 0) {
505 			mutex_enter(&uvm_kernel_object->vmobjlock);
506 			pg = uvm_pagelookup(uvm_kernel_object,
507 			    va - vm_map_min(kernel_map));
508 			mutex_exit(&uvm_kernel_object->vmobjlock);
509 			if (pg) {
510 				panic("uvm_km_check_empty: "
511 				    "has page hashed at %p", (const void *)va);
512 			}
513 		}
514 	}
515 }
516 #endif /* defined(DEBUG) */
517 
518 /*
519  * uvm_km_alloc: allocate an area of kernel memory.
520  *
521  * => NOTE: we can return 0 even if we can wait if there is not enough
522  *	free VM space in the map... caller should be prepared to handle
523  *	this case.
524  * => we return KVA of memory allocated
525  */
526 
527 vaddr_t
528 uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags)
529 {
530 	vaddr_t kva, loopva;
531 	vaddr_t offset;
532 	vsize_t loopsize;
533 	struct vm_page *pg;
534 	struct uvm_object *obj;
535 	int pgaflags;
536 	vm_prot_t prot;
537 	UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist);
538 
539 	KASSERT(vm_map_pmap(map) == pmap_kernel());
540 	KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
541 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
542 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
543 
544 	/*
545 	 * setup for call
546 	 */
547 
548 	kva = vm_map_min(map);	/* hint */
549 	size = round_page(size);
550 	obj = (flags & UVM_KMF_PAGEABLE) ? uvm_kernel_object : NULL;
551 	UVMHIST_LOG(maphist,"  (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
552 		    map, obj, size, flags);
553 
554 	/*
555 	 * allocate some virtual space
556 	 */
557 
558 	if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
559 	    align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
560 	    UVM_ADV_RANDOM,
561 	    (flags & (UVM_KMF_TRYLOCK | UVM_KMF_NOWAIT | UVM_KMF_WAITVA))
562 	    | UVM_FLAG_QUANTUM)) != 0)) {
563 		UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
564 		return(0);
565 	}
566 
567 	/*
568 	 * if all we wanted was VA, return now
569 	 */
570 
571 	if (flags & (UVM_KMF_VAONLY | UVM_KMF_PAGEABLE)) {
572 		UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
573 		return(kva);
574 	}
575 
576 	/*
577 	 * recover object offset from virtual address
578 	 */
579 
580 	offset = kva - vm_map_min(kernel_map);
581 	UVMHIST_LOG(maphist, "  kva=0x%x, offset=0x%x", kva, offset,0,0);
582 
583 	/*
584 	 * now allocate and map in the memory... note that we are the only ones
585 	 * whom should ever get a handle on this area of VM.
586 	 */
587 
588 	loopva = kva;
589 	loopsize = size;
590 
591 	pgaflags = 0;
592 	if (flags & UVM_KMF_NOWAIT)
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),
629 		    prot, PMAP_KMPAGE);
630 		loopva += PAGE_SIZE;
631 		offset += PAGE_SIZE;
632 		loopsize -= PAGE_SIZE;
633 	}
634 
635        	pmap_update(pmap_kernel());
636 
637 	UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
638 	return(kva);
639 }
640 
641 /*
642  * uvm_km_free: free an area of kernel memory
643  */
644 
645 void
646 uvm_km_free(struct vm_map *map, vaddr_t addr, vsize_t size, uvm_flag_t flags)
647 {
648 
649 	KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
650 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
651 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
652 	KASSERT((addr & PAGE_MASK) == 0);
653 	KASSERT(vm_map_pmap(map) == pmap_kernel());
654 
655 	size = round_page(size);
656 
657 	if (flags & UVM_KMF_PAGEABLE) {
658 		uvm_km_pgremove(addr, addr + size);
659 		pmap_remove(pmap_kernel(), addr, addr + size);
660 	} else if (flags & UVM_KMF_WIRED) {
661 		uvm_km_pgremove_intrsafe(map, addr, addr + size);
662 		pmap_kremove(addr, size);
663 	}
664 
665 	/*
666 	 * uvm_unmap_remove calls pmap_update for us.
667 	 */
668 
669 	uvm_unmap1(map, addr, addr + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
670 }
671 
672 /* Sanity; must specify both or none. */
673 #if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \
674     (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE))
675 #error Must specify MAP and UNMAP together.
676 #endif
677 
678 /*
679  * uvm_km_alloc_poolpage: allocate a page for the pool allocator
680  *
681  * => if the pmap specifies an alternate mapping method, we use it.
682  */
683 
684 /* ARGSUSED */
685 vaddr_t
686 uvm_km_alloc_poolpage_cache(struct vm_map *map, bool waitok)
687 {
688 #if defined(PMAP_MAP_POOLPAGE)
689 	return uvm_km_alloc_poolpage(map, waitok);
690 #else
691 	struct vm_page *pg;
692 	struct pool *pp = &vm_map_to_kernel(map)->vmk_vacache;
693 	vaddr_t va;
694 
695 	if ((map->flags & VM_MAP_VACACHE) == 0)
696 		return uvm_km_alloc_poolpage(map, waitok);
697 
698 	va = (vaddr_t)pool_get(pp, waitok ? PR_WAITOK : PR_NOWAIT);
699 	if (va == 0)
700 		return 0;
701 	KASSERT(!pmap_extract(pmap_kernel(), va, NULL));
702 again:
703 	pg = uvm_pagealloc(NULL, 0, NULL, waitok ? 0 : UVM_PGA_USERESERVE);
704 	if (__predict_false(pg == NULL)) {
705 		if (waitok) {
706 			uvm_wait("plpg");
707 			goto again;
708 		} else {
709 			pool_put(pp, (void *)va);
710 			return 0;
711 		}
712 	}
713 	pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg),
714 	    VM_PROT_READ|VM_PROT_WRITE, PMAP_KMPAGE);
715 	pmap_update(pmap_kernel());
716 
717 	return va;
718 #endif /* PMAP_MAP_POOLPAGE */
719 }
720 
721 vaddr_t
722 uvm_km_alloc_poolpage(struct vm_map *map, bool waitok)
723 {
724 #if defined(PMAP_MAP_POOLPAGE)
725 	struct vm_page *pg;
726 	vaddr_t va;
727 
728  again:
729 	pg = uvm_pagealloc(NULL, 0, NULL, waitok ? 0 : UVM_PGA_USERESERVE);
730 	if (__predict_false(pg == NULL)) {
731 		if (waitok) {
732 			uvm_wait("plpg");
733 			goto again;
734 		} else
735 			return (0);
736 	}
737 	va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
738 	if (__predict_false(va == 0))
739 		uvm_pagefree(pg);
740 	return (va);
741 #else
742 	vaddr_t va;
743 
744 	va = uvm_km_alloc(map, PAGE_SIZE, 0,
745 	    (waitok ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK) | UVM_KMF_WIRED);
746 	return (va);
747 #endif /* PMAP_MAP_POOLPAGE */
748 }
749 
750 /*
751  * uvm_km_free_poolpage: free a previously allocated pool page
752  *
753  * => if the pmap specifies an alternate unmapping method, we use it.
754  */
755 
756 /* ARGSUSED */
757 void
758 uvm_km_free_poolpage_cache(struct vm_map *map, vaddr_t addr)
759 {
760 #if defined(PMAP_UNMAP_POOLPAGE)
761 	uvm_km_free_poolpage(map, addr);
762 #else
763 	struct pool *pp;
764 
765 	if ((map->flags & VM_MAP_VACACHE) == 0) {
766 		uvm_km_free_poolpage(map, addr);
767 		return;
768 	}
769 
770 	KASSERT(pmap_extract(pmap_kernel(), addr, NULL));
771 	uvm_km_pgremove_intrsafe(map, addr, addr + PAGE_SIZE);
772 	pmap_kremove(addr, PAGE_SIZE);
773 #if defined(DEBUG)
774 	pmap_update(pmap_kernel());
775 #endif
776 	KASSERT(!pmap_extract(pmap_kernel(), addr, NULL));
777 	pp = &vm_map_to_kernel(map)->vmk_vacache;
778 	pool_put(pp, (void *)addr);
779 #endif
780 }
781 
782 /* ARGSUSED */
783 void
784 uvm_km_free_poolpage(struct vm_map *map, vaddr_t addr)
785 {
786 #if defined(PMAP_UNMAP_POOLPAGE)
787 	paddr_t pa;
788 
789 	pa = PMAP_UNMAP_POOLPAGE(addr);
790 	uvm_pagefree(PHYS_TO_VM_PAGE(pa));
791 #else
792 	uvm_km_free(map, addr, PAGE_SIZE, UVM_KMF_WIRED);
793 #endif /* PMAP_UNMAP_POOLPAGE */
794 }
795