Home | History | Annotate | Line # | Download | only in uvm
uvm_km.c revision 1.103
      1 /*	$NetBSD: uvm_km.c,v 1.103 2008/12/13 11:34:43 ad 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 (at) 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  *   mb_map => memory for large mbufs,
     95  *   pager_map => used to map "buf" structures into kernel space
     96  *   exec_map => used during exec to handle exec args
     97  *   etc...
     98  *
     99  * the kernel allocates its private memory out of special uvm_objects whose
    100  * reference count is set to UVM_OBJ_KERN (thus indicating that the objects
    101  * are "special" and never die).   all kernel objects should be thought of
    102  * as large, fixed-sized, sparsely populated uvm_objects.   each kernel
    103  * object is equal to the size of kernel virtual address space (i.e. the
    104  * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
    105  *
    106  * note that just because a kernel object spans the entire kernel virtual
    107  * address space doesn't mean that it has to be mapped into the entire space.
    108  * large chunks of a kernel object's space go unused either because
    109  * that area of kernel VM is unmapped, or there is some other type of
    110  * object mapped into that range (e.g. a vnode).    for submap's kernel
    111  * objects, the only part of the object that can ever be populated is the
    112  * offsets that are managed by the submap.
    113  *
    114  * note that the "offset" in a kernel object is always the kernel virtual
    115  * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
    116  * example:
    117  *   suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
    118  *   uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
    119  *   kernel map].    if uvm_km_alloc returns virtual address 0xf8235000,
    120  *   then that means that the page at offset 0x235000 in kernel_object is
    121  *   mapped at 0xf8235000.
    122  *
    123  * kernel object have one other special property: when the kernel virtual
    124  * memory mapping them is unmapped, the backing memory in the object is
    125  * freed right away.   this is done with the uvm_km_pgremove() function.
    126  * this has to be done because there is no backing store for kernel pages
    127  * and no need to save them after they are no longer referenced.
    128  */
    129 
    130 #include <sys/cdefs.h>
    131 __KERNEL_RCSID(0, "$NetBSD: uvm_km.c,v 1.103 2008/12/13 11:34:43 ad Exp $");
    132 
    133 #include "opt_uvmhist.h"
    134 
    135 #include <sys/param.h>
    136 #include <sys/malloc.h>
    137 #include <sys/systm.h>
    138 #include <sys/proc.h>
    139 #include <sys/pool.h>
    140 
    141 #include <uvm/uvm.h>
    142 
    143 /*
    144  * global data structures
    145  */
    146 
    147 struct vm_map *kernel_map = NULL;
    148 
    149 /*
    150  * local data structues
    151  */
    152 
    153 static struct vm_map_kernel	kernel_map_store;
    154 static struct vm_map_entry	kernel_first_mapent_store;
    155 
    156 #if !defined(PMAP_MAP_POOLPAGE)
    157 
    158 /*
    159  * kva cache
    160  *
    161  * XXX maybe it's better to do this at the uvm_map layer.
    162  */
    163 
    164 #define	KM_VACACHE_SIZE	(32 * PAGE_SIZE) /* XXX tune */
    165 
    166 static void *km_vacache_alloc(struct pool *, int);
    167 static void km_vacache_free(struct pool *, void *);
    168 static void km_vacache_init(struct vm_map *, const char *, size_t);
    169 
    170 /* XXX */
    171 #define	KM_VACACHE_POOL_TO_MAP(pp) \
    172 	((struct vm_map *)((char *)(pp) - \
    173 	    offsetof(struct vm_map_kernel, vmk_vacache)))
    174 
    175 static void *
    176 km_vacache_alloc(struct pool *pp, int flags)
    177 {
    178 	vaddr_t va;
    179 	size_t size;
    180 	struct vm_map *map;
    181 	size = pp->pr_alloc->pa_pagesz;
    182 
    183 	map = KM_VACACHE_POOL_TO_MAP(pp);
    184 
    185 	va = vm_map_min(map); /* hint */
    186 	if (uvm_map(map, &va, size, NULL, UVM_UNKNOWN_OFFSET, size,
    187 	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    188 	    UVM_ADV_RANDOM, UVM_FLAG_QUANTUM |
    189 	    ((flags & PR_WAITOK) ? UVM_FLAG_WAITVA :
    190 	    UVM_FLAG_TRYLOCK | UVM_FLAG_NOWAIT))))
    191 		return NULL;
    192 
    193 	return (void *)va;
    194 }
    195 
    196 static void
    197 km_vacache_free(struct pool *pp, void *v)
    198 {
    199 	vaddr_t va = (vaddr_t)v;
    200 	size_t size = pp->pr_alloc->pa_pagesz;
    201 	struct vm_map *map;
    202 
    203 	map = KM_VACACHE_POOL_TO_MAP(pp);
    204 	uvm_unmap1(map, va, va + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
    205 }
    206 
    207 /*
    208  * km_vacache_init: initialize kva cache.
    209  */
    210 
    211 static void
    212 km_vacache_init(struct vm_map *map, const char *name, size_t size)
    213 {
    214 	struct vm_map_kernel *vmk;
    215 	struct pool *pp;
    216 	struct pool_allocator *pa;
    217 	int ipl;
    218 
    219 	KASSERT(VM_MAP_IS_KERNEL(map));
    220 	KASSERT(size < (vm_map_max(map) - vm_map_min(map)) / 2); /* sanity */
    221 
    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 
    233 	if ((map->flags & VM_MAP_INTRSAFE) != 0)
    234 		ipl = IPL_VM;
    235 	else
    236 		ipl = IPL_NONE;
    237 
    238 	pool_init(pp, PAGE_SIZE, 0, 0, PR_NOTOUCH | PR_RECURSIVE, name, pa,
    239 	    ipl);
    240 }
    241 
    242 void
    243 uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
    244 {
    245 
    246 	map->flags |= VM_MAP_VACACHE;
    247 	if (size == 0)
    248 		size = KM_VACACHE_SIZE;
    249 	km_vacache_init(map, name, size);
    250 }
    251 
    252 #else /* !defined(PMAP_MAP_POOLPAGE) */
    253 
    254 void
    255 uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
    256 {
    257 
    258 	/* nothing */
    259 }
    260 
    261 #endif /* !defined(PMAP_MAP_POOLPAGE) */
    262 
    263 void
    264 uvm_km_va_drain(struct vm_map *map, uvm_flag_t flags)
    265 {
    266 	struct vm_map_kernel *vmk = vm_map_to_kernel(map);
    267 
    268 	callback_run_roundrobin(&vmk->vmk_reclaim_callback, NULL);
    269 }
    270 
    271 /*
    272  * uvm_km_init: init kernel maps and objects to reflect reality (i.e.
    273  * KVM already allocated for text, data, bss, and static data structures).
    274  *
    275  * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
    276  *    we assume that [vmin -> start] has already been allocated and that
    277  *    "end" is the end.
    278  */
    279 
    280 void
    281 uvm_km_init(vaddr_t start, vaddr_t end)
    282 {
    283 	vaddr_t base = VM_MIN_KERNEL_ADDRESS;
    284 
    285 	/*
    286 	 * next, init kernel memory objects.
    287 	 */
    288 
    289 	/* kernel_object: for pageable anonymous kernel memory */
    290 	uao_init();
    291 	uvm_kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
    292 				 VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
    293 
    294 	/*
    295 	 * init the map and reserve any space that might already
    296 	 * have been allocated kernel space before installing.
    297 	 */
    298 
    299 	uvm_map_setup_kernel(&kernel_map_store, base, end, VM_MAP_PAGEABLE);
    300 	kernel_map_store.vmk_map.pmap = pmap_kernel();
    301 	if (start != base) {
    302 		int error;
    303 		struct uvm_map_args args;
    304 
    305 		error = uvm_map_prepare(&kernel_map_store.vmk_map,
    306 		    base, start - base,
    307 		    NULL, UVM_UNKNOWN_OFFSET, 0,
    308 		    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    309 		    		UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args);
    310 		if (!error) {
    311 			kernel_first_mapent_store.flags =
    312 			    UVM_MAP_KERNEL | UVM_MAP_FIRST;
    313 			error = uvm_map_enter(&kernel_map_store.vmk_map, &args,
    314 			    &kernel_first_mapent_store);
    315 		}
    316 
    317 		if (error)
    318 			panic(
    319 			    "uvm_km_init: could not reserve space for kernel");
    320 	}
    321 
    322 	/*
    323 	 * install!
    324 	 */
    325 
    326 	kernel_map = &kernel_map_store.vmk_map;
    327 	uvm_km_vacache_init(kernel_map, "kvakernel", 0);
    328 }
    329 
    330 /*
    331  * uvm_km_suballoc: allocate a submap in the kernel map.   once a submap
    332  * is allocated all references to that area of VM must go through it.  this
    333  * allows the locking of VAs in kernel_map to be broken up into regions.
    334  *
    335  * => if `fixed' is true, *vmin specifies where the region described
    336  *      by the submap must start
    337  * => if submap is non NULL we use that as the submap, otherwise we
    338  *	alloc a new map
    339  */
    340 
    341 struct vm_map *
    342 uvm_km_suballoc(struct vm_map *map, vaddr_t *vmin /* IN/OUT */,
    343     vaddr_t *vmax /* OUT */, vsize_t size, int flags, bool fixed,
    344     struct vm_map_kernel *submap)
    345 {
    346 	int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
    347 
    348 	KASSERT(vm_map_pmap(map) == pmap_kernel());
    349 
    350 	size = round_page(size);	/* round up to pagesize */
    351 	size += uvm_mapent_overhead(size, flags);
    352 
    353 	/*
    354 	 * first allocate a blank spot in the parent map
    355 	 */
    356 
    357 	if (uvm_map(map, vmin, size, NULL, UVM_UNKNOWN_OFFSET, 0,
    358 	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    359 	    UVM_ADV_RANDOM, mapflags)) != 0) {
    360 	       panic("uvm_km_suballoc: unable to allocate space in parent map");
    361 	}
    362 
    363 	/*
    364 	 * set VM bounds (vmin is filled in by uvm_map)
    365 	 */
    366 
    367 	*vmax = *vmin + size;
    368 
    369 	/*
    370 	 * add references to pmap and create or init the submap
    371 	 */
    372 
    373 	pmap_reference(vm_map_pmap(map));
    374 	if (submap == NULL) {
    375 		submap = malloc(sizeof(*submap), M_VMMAP, M_WAITOK);
    376 		if (submap == NULL)
    377 			panic("uvm_km_suballoc: unable to create submap");
    378 	}
    379 	uvm_map_setup_kernel(submap, *vmin, *vmax, flags);
    380 	submap->vmk_map.pmap = vm_map_pmap(map);
    381 
    382 	/*
    383 	 * now let uvm_map_submap plug in it...
    384 	 */
    385 
    386 	if (uvm_map_submap(map, *vmin, *vmax, &submap->vmk_map) != 0)
    387 		panic("uvm_km_suballoc: submap allocation failed");
    388 
    389 	return(&submap->vmk_map);
    390 }
    391 
    392 /*
    393  * uvm_km_pgremove: remove pages from a kernel uvm_object.
    394  *
    395  * => when you unmap a part of anonymous kernel memory you want to toss
    396  *    the pages right away.    (this gets called from uvm_unmap_...).
    397  */
    398 
    399 void
    400 uvm_km_pgremove(vaddr_t startva, vaddr_t endva)
    401 {
    402 	struct uvm_object * const uobj = uvm_kernel_object;
    403 	const voff_t start = startva - vm_map_min(kernel_map);
    404 	const voff_t end = endva - vm_map_min(kernel_map);
    405 	struct vm_page *pg;
    406 	voff_t curoff, nextoff;
    407 	int swpgonlydelta = 0;
    408 	UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
    409 
    410 	KASSERT(VM_MIN_KERNEL_ADDRESS <= startva);
    411 	KASSERT(startva < endva);
    412 	KASSERT(endva <= VM_MAX_KERNEL_ADDRESS);
    413 
    414 	mutex_enter(&uobj->vmobjlock);
    415 
    416 	for (curoff = start; curoff < end; curoff = nextoff) {
    417 		nextoff = curoff + PAGE_SIZE;
    418 		pg = uvm_pagelookup(uobj, curoff);
    419 		if (pg != NULL && pg->flags & PG_BUSY) {
    420 			pg->flags |= PG_WANTED;
    421 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
    422 				    "km_pgrm", 0);
    423 			mutex_enter(&uobj->vmobjlock);
    424 			nextoff = curoff;
    425 			continue;
    426 		}
    427 
    428 		/*
    429 		 * free the swap slot, then the page.
    430 		 */
    431 
    432 		if (pg == NULL &&
    433 		    uao_find_swslot(uobj, curoff >> PAGE_SHIFT) > 0) {
    434 			swpgonlydelta++;
    435 		}
    436 		uao_dropswap(uobj, curoff >> PAGE_SHIFT);
    437 		if (pg != NULL) {
    438 			mutex_enter(&uvm_pageqlock);
    439 			uvm_pagefree(pg);
    440 			mutex_exit(&uvm_pageqlock);
    441 		}
    442 	}
    443 	mutex_exit(&uobj->vmobjlock);
    444 
    445 	if (swpgonlydelta > 0) {
    446 		mutex_enter(&uvm_swap_data_lock);
    447 		KASSERT(uvmexp.swpgonly >= swpgonlydelta);
    448 		uvmexp.swpgonly -= swpgonlydelta;
    449 		mutex_exit(&uvm_swap_data_lock);
    450 	}
    451 }
    452 
    453 
    454 /*
    455  * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for non object backed
    456  *    regions.
    457  *
    458  * => when you unmap a part of anonymous kernel memory you want to toss
    459  *    the pages right away.    (this is called from uvm_unmap_...).
    460  * => none of the pages will ever be busy, and none of them will ever
    461  *    be on the active or inactive queues (because they have no object).
    462  */
    463 
    464 void
    465 uvm_km_pgremove_intrsafe(struct vm_map *map, vaddr_t start, vaddr_t end)
    466 {
    467 	struct vm_page *pg;
    468 	paddr_t pa;
    469 	UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist);
    470 
    471 	KASSERT(VM_MAP_IS_KERNEL(map));
    472 	KASSERT(vm_map_min(map) <= start);
    473 	KASSERT(start < end);
    474 	KASSERT(end <= vm_map_max(map));
    475 
    476 	for (; start < end; start += PAGE_SIZE) {
    477 		if (!pmap_extract(pmap_kernel(), start, &pa)) {
    478 			continue;
    479 		}
    480 		pg = PHYS_TO_VM_PAGE(pa);
    481 		KASSERT(pg);
    482 		KASSERT(pg->uobject == NULL && pg->uanon == NULL);
    483 		uvm_pagefree(pg);
    484 	}
    485 }
    486 
    487 #if defined(DEBUG)
    488 void
    489 uvm_km_check_empty(struct vm_map *map, vaddr_t start, vaddr_t end)
    490 {
    491 	struct vm_page *pg;
    492 	vaddr_t va;
    493 	paddr_t pa;
    494 
    495 	KDASSERT(VM_MAP_IS_KERNEL(map));
    496 	KDASSERT(vm_map_min(map) <= start);
    497 	KDASSERT(start < end);
    498 	KDASSERT(end <= vm_map_max(map));
    499 
    500 	for (va = start; va < end; va += PAGE_SIZE) {
    501 		if (pmap_extract(pmap_kernel(), va, &pa)) {
    502 			panic("uvm_km_check_empty: va %p has pa 0x%llx",
    503 			    (void *)va, (long long)pa);
    504 		}
    505 		if ((map->flags & VM_MAP_INTRSAFE) == 0) {
    506 			mutex_enter(&uvm_kernel_object->vmobjlock);
    507 			pg = uvm_pagelookup(uvm_kernel_object,
    508 			    va - vm_map_min(kernel_map));
    509 			mutex_exit(&uvm_kernel_object->vmobjlock);
    510 			if (pg) {
    511 				panic("uvm_km_check_empty: "
    512 				    "has page hashed at %p", (const void *)va);
    513 			}
    514 		}
    515 	}
    516 }
    517 #endif /* defined(DEBUG) */
    518 
    519 /*
    520  * uvm_km_alloc: allocate an area of kernel memory.
    521  *
    522  * => NOTE: we can return 0 even if we can wait if there is not enough
    523  *	free VM space in the map... caller should be prepared to handle
    524  *	this case.
    525  * => we return KVA of memory allocated
    526  */
    527 
    528 vaddr_t
    529 uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags)
    530 {
    531 	vaddr_t kva, loopva;
    532 	vaddr_t offset;
    533 	vsize_t loopsize;
    534 	struct vm_page *pg;
    535 	struct uvm_object *obj;
    536 	int pgaflags;
    537 	vm_prot_t prot;
    538 	UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist);
    539 
    540 	KASSERT(vm_map_pmap(map) == pmap_kernel());
    541 	KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
    542 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
    543 		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
    544 
    545 	/*
    546 	 * setup for call
    547 	 */
    548 
    549 	kva = vm_map_min(map);	/* hint */
    550 	size = round_page(size);
    551 	obj = (flags & UVM_KMF_PAGEABLE) ? uvm_kernel_object : NULL;
    552 	UVMHIST_LOG(maphist,"  (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
    553 		    map, obj, size, flags);
    554 
    555 	/*
    556 	 * allocate some virtual space
    557 	 */
    558 
    559 	if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
    560 	    align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
    561 	    UVM_ADV_RANDOM,
    562 	    (flags & (UVM_KMF_TRYLOCK | UVM_KMF_NOWAIT | UVM_KMF_WAITVA))
    563 	    | UVM_FLAG_QUANTUM)) != 0)) {
    564 		UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
    565 		return(0);
    566 	}
    567 
    568 	/*
    569 	 * if all we wanted was VA, return now
    570 	 */
    571 
    572 	if (flags & (UVM_KMF_VAONLY | UVM_KMF_PAGEABLE)) {
    573 		UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
    574 		return(kva);
    575 	}
    576 
    577 	/*
    578 	 * recover object offset from virtual address
    579 	 */
    580 
    581 	offset = kva - vm_map_min(kernel_map);
    582 	UVMHIST_LOG(maphist, "  kva=0x%x, offset=0x%x", kva, offset,0,0);
    583 
    584 	/*
    585 	 * now allocate and map in the memory... note that we are the only ones
    586 	 * whom should ever get a handle on this area of VM.
    587 	 */
    588 
    589 	loopva = kva;
    590 	loopsize = size;
    591 
    592 	pgaflags = 0;
    593 	if (flags & UVM_KMF_NOWAIT)
    594 		pgaflags |= UVM_PGA_USERESERVE;
    595 	if (flags & UVM_KMF_ZERO)
    596 		pgaflags |= UVM_PGA_ZERO;
    597 	prot = VM_PROT_READ | VM_PROT_WRITE;
    598 	if (flags & UVM_KMF_EXEC)
    599 		prot |= VM_PROT_EXECUTE;
    600 	while (loopsize) {
    601 		KASSERT(!pmap_extract(pmap_kernel(), loopva, NULL));
    602 
    603 		pg = uvm_pagealloc(NULL, offset, NULL, pgaflags);
    604 
    605 		/*
    606 		 * out of memory?
    607 		 */
    608 
    609 		if (__predict_false(pg == NULL)) {
    610 			if ((flags & UVM_KMF_NOWAIT) ||
    611 			    ((flags & UVM_KMF_CANFAIL) && !uvm_reclaimable())) {
    612 				/* free everything! */
    613 				uvm_km_free(map, kva, size,
    614 				    flags & UVM_KMF_TYPEMASK);
    615 				return (0);
    616 			} else {
    617 				uvm_wait("km_getwait2");	/* sleep here */
    618 				continue;
    619 			}
    620 		}
    621 
    622 		pg->flags &= ~PG_BUSY;	/* new page */
    623 		UVM_PAGE_OWN(pg, NULL);
    624 
    625 		/*
    626 		 * map it in
    627 		 */
    628 
    629 		pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), 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