uvm_km.c revision 1.99
1/*	$NetBSD: uvm_km.c,v 1.99 2008/03/24 08:52:55 yamt 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 *   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 virutal
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 1.99 2008/03/24 08:52:55 yamt 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
147struct vm_map *kernel_map = NULL;
148
149/*
150 * local data structues
151 */
152
153static struct vm_map_kernel	kernel_map_store;
154static 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
166static void *km_vacache_alloc(struct pool *, int);
167static void km_vacache_free(struct pool *, void *);
168static 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
175static void *
176km_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
196static void
197km_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
211static void
212km_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
242void
243uvm_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
254void
255uvm_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
263void
264uvm_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
280void
281uvm_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
341struct vm_map *
342uvm_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
399void
400uvm_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
464void
465uvm_km_pgremove_intrsafe(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_MIN_KERNEL_ADDRESS <= start);
472	KASSERT(start < end);
473	KASSERT(end <= VM_MAX_KERNEL_ADDRESS);
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)
487void
488uvm_km_check_empty(vaddr_t start, vaddr_t end, bool intrsafe)
489{
490	vaddr_t va;
491	paddr_t pa;
492
493	KDASSERT(VM_MIN_KERNEL_ADDRESS <= start);
494	KDASSERT(start < end);
495	KDASSERT(end <= VM_MAX_KERNEL_ADDRESS);
496
497	for (va = start; va < end; va += PAGE_SIZE) {
498		if (pmap_extract(pmap_kernel(), va, &pa)) {
499			panic("uvm_km_check_empty: va %p has pa 0x%llx",
500			    (void *)va, (long long)pa);
501		}
502		if (!intrsafe) {
503			const struct vm_page *pg;
504
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
527vaddr_t
528uvm_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 = UVM_PGA_USERESERVE;
592	if (flags & UVM_KMF_ZERO)
593		pgaflags |= UVM_PGA_ZERO;
594	prot = VM_PROT_READ | VM_PROT_WRITE;
595	if (flags & UVM_KMF_EXEC)
596		prot |= VM_PROT_EXECUTE;
597	while (loopsize) {
598		KASSERT(!pmap_extract(pmap_kernel(), loopva, NULL));
599
600		pg = uvm_pagealloc(NULL, offset, NULL, pgaflags);
601
602		/*
603		 * out of memory?
604		 */
605
606		if (__predict_false(pg == NULL)) {
607			if ((flags & UVM_KMF_NOWAIT) ||
608			    ((flags & UVM_KMF_CANFAIL) && !uvm_reclaimable())) {
609				/* free everything! */
610				uvm_km_free(map, kva, size,
611				    flags & UVM_KMF_TYPEMASK);
612				return (0);
613			} else {
614				uvm_wait("km_getwait2");	/* sleep here */
615				continue;
616			}
617		}
618
619		pg->flags &= ~PG_BUSY;	/* new page */
620		UVM_PAGE_OWN(pg, NULL);
621
622		/*
623		 * map it in
624		 */
625
626		pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg), prot);
627		loopva += PAGE_SIZE;
628		offset += PAGE_SIZE;
629		loopsize -= PAGE_SIZE;
630	}
631
632       	pmap_update(pmap_kernel());
633
634	UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
635	return(kva);
636}
637
638/*
639 * uvm_km_free: free an area of kernel memory
640 */
641
642void
643uvm_km_free(struct vm_map *map, vaddr_t addr, vsize_t size, uvm_flag_t flags)
644{
645
646	KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
647		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
648		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
649	KASSERT((addr & PAGE_MASK) == 0);
650	KASSERT(vm_map_pmap(map) == pmap_kernel());
651
652	size = round_page(size);
653
654	if (flags & UVM_KMF_PAGEABLE) {
655		uvm_km_pgremove(addr, addr + size);
656		pmap_remove(pmap_kernel(), addr, addr + size);
657	} else if (flags & UVM_KMF_WIRED) {
658		uvm_km_pgremove_intrsafe(addr, addr + size);
659		pmap_kremove(addr, size);
660	}
661
662	/*
663	 * uvm_unmap_remove calls pmap_update for us.
664	 */
665
666	uvm_unmap1(map, addr, addr + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
667}
668
669/* Sanity; must specify both or none. */
670#if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \
671    (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE))
672#error Must specify MAP and UNMAP together.
673#endif
674
675/*
676 * uvm_km_alloc_poolpage: allocate a page for the pool allocator
677 *
678 * => if the pmap specifies an alternate mapping method, we use it.
679 */
680
681/* ARGSUSED */
682vaddr_t
683uvm_km_alloc_poolpage_cache(struct vm_map *map, bool waitok)
684{
685#if defined(PMAP_MAP_POOLPAGE)
686	return uvm_km_alloc_poolpage(map, waitok);
687#else
688	struct vm_page *pg;
689	struct pool *pp = &vm_map_to_kernel(map)->vmk_vacache;
690	vaddr_t va;
691
692	if ((map->flags & VM_MAP_VACACHE) == 0)
693		return uvm_km_alloc_poolpage(map, waitok);
694
695	va = (vaddr_t)pool_get(pp, waitok ? PR_WAITOK : PR_NOWAIT);
696	if (va == 0)
697		return 0;
698	KASSERT(!pmap_extract(pmap_kernel(), va, NULL));
699again:
700	pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
701	if (__predict_false(pg == NULL)) {
702		if (waitok) {
703			uvm_wait("plpg");
704			goto again;
705		} else {
706			pool_put(pp, (void *)va);
707			return 0;
708		}
709	}
710	pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE);
711	pmap_update(pmap_kernel());
712
713	return va;
714#endif /* PMAP_MAP_POOLPAGE */
715}
716
717vaddr_t
718uvm_km_alloc_poolpage(struct vm_map *map, bool waitok)
719{
720#if defined(PMAP_MAP_POOLPAGE)
721	struct vm_page *pg;
722	vaddr_t va;
723
724 again:
725	pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
726	if (__predict_false(pg == NULL)) {
727		if (waitok) {
728			uvm_wait("plpg");
729			goto again;
730		} else
731			return (0);
732	}
733	va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
734	if (__predict_false(va == 0))
735		uvm_pagefree(pg);
736	return (va);
737#else
738	vaddr_t va;
739
740	va = uvm_km_alloc(map, PAGE_SIZE, 0,
741	    (waitok ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK) | UVM_KMF_WIRED);
742	return (va);
743#endif /* PMAP_MAP_POOLPAGE */
744}
745
746/*
747 * uvm_km_free_poolpage: free a previously allocated pool page
748 *
749 * => if the pmap specifies an alternate unmapping method, we use it.
750 */
751
752/* ARGSUSED */
753void
754uvm_km_free_poolpage_cache(struct vm_map *map, vaddr_t addr)
755{
756#if defined(PMAP_UNMAP_POOLPAGE)
757	uvm_km_free_poolpage(map, addr);
758#else
759	struct pool *pp;
760
761	if ((map->flags & VM_MAP_VACACHE) == 0) {
762		uvm_km_free_poolpage(map, addr);
763		return;
764	}
765
766	KASSERT(pmap_extract(pmap_kernel(), addr, NULL));
767	uvm_km_pgremove_intrsafe(addr, addr + PAGE_SIZE);
768	pmap_kremove(addr, PAGE_SIZE);
769#if defined(DEBUG)
770	pmap_update(pmap_kernel());
771#endif
772	KASSERT(!pmap_extract(pmap_kernel(), addr, NULL));
773	pp = &vm_map_to_kernel(map)->vmk_vacache;
774	pool_put(pp, (void *)addr);
775#endif
776}
777
778/* ARGSUSED */
779void
780uvm_km_free_poolpage(struct vm_map *map, vaddr_t addr)
781{
782#if defined(PMAP_UNMAP_POOLPAGE)
783	paddr_t pa;
784
785	pa = PMAP_UNMAP_POOLPAGE(addr);
786	uvm_pagefree(PHYS_TO_VM_PAGE(pa));
787#else
788	uvm_km_free(map, addr, PAGE_SIZE, UVM_KMF_WIRED);
789#endif /* PMAP_UNMAP_POOLPAGE */
790}
791