uvm_km.c revision 1.32
1/*	$NetBSD: uvm_km.c,v 1.32 1999/09/12 01:17:36 chs 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#include "opt_uvmhist.h"
70
71/*
72 * uvm_km.c: handle kernel memory allocation and management
73 */
74
75/*
76 * overview of kernel memory management:
77 *
78 * the kernel virtual address space is mapped by "kernel_map."   kernel_map
79 * starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS.
80 * note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map).
81 *
82 * the kernel_map has several "submaps."   submaps can only appear in
83 * the kernel_map (user processes can't use them).   submaps "take over"
84 * the management of a sub-range of the kernel's address space.  submaps
85 * are typically allocated at boot time and are never released.   kernel
86 * virtual address space that is mapped by a submap is locked by the
87 * submap's lock -- not the kernel_map's lock.
88 *
89 * thus, the useful feature of submaps is that they allow us to break
90 * up the locking and protection of the kernel address space into smaller
91 * chunks.
92 *
93 * the vm system has several standard kernel submaps, including:
94 *   kmem_map => contains only wired kernel memory for the kernel
95 *		malloc.   *** access to kmem_map must be protected
96 *		by splimp() because we are allowed to call malloc()
97 *		at interrupt time ***
98 *   mb_map => memory for large mbufs,  *** protected by splimp ***
99 *   pager_map => used to map "buf" structures into kernel space
100 *   exec_map => used during exec to handle exec args
101 *   etc...
102 *
103 * the kernel allocates its private memory out of special uvm_objects whose
104 * reference count is set to UVM_OBJ_KERN (thus indicating that the objects
105 * are "special" and never die).   all kernel objects should be thought of
106 * as large, fixed-sized, sparsely populated uvm_objects.   each kernel
107 * object is equal to the size of kernel virtual address space (i.e. the
108 * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
109 *
110 * most kernel private memory lives in kernel_object.   the only exception
111 * to this is for memory that belongs to submaps that must be protected
112 * by splimp().    each of these submaps has their own private kernel
113 * object (e.g. kmem_object, mb_object).
114 *
115 * note that just because a kernel object spans the entire kernel virutal
116 * address space doesn't mean that it has to be mapped into the entire space.
117 * large chunks of a kernel object's space go unused either because
118 * that area of kernel VM is unmapped, or there is some other type of
119 * object mapped into that range (e.g. a vnode).    for submap's kernel
120 * objects, the only part of the object that can ever be populated is the
121 * offsets that are managed by the submap.
122 *
123 * note that the "offset" in a kernel object is always the kernel virtual
124 * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
125 * example:
126 *   suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
127 *   uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
128 *   kernel map].    if uvm_km_alloc returns virtual address 0xf8235000,
129 *   then that means that the page at offset 0x235000 in kernel_object is
130 *   mapped at 0xf8235000.
131 *
132 * note that the offsets in kmem_object and mb_object also follow this
133 * rule.   this means that the offsets for kmem_object must fall in the
134 * range of [vm_map_min(kmem_object) - vm_map_min(kernel_map)] to
135 * [vm_map_max(kmem_object) - vm_map_min(kernel_map)], so the offsets
136 * in those objects will typically not start at zero.
137 *
138 * kernel object have one other special property: when the kernel virtual
139 * memory mapping them is unmapped, the backing memory in the object is
140 * freed right away.   this is done with the uvm_km_pgremove() function.
141 * this has to be done because there is no backing store for kernel pages
142 * and no need to save them after they are no longer referenced.
143 */
144
145#include <sys/param.h>
146#include <sys/systm.h>
147#include <sys/proc.h>
148
149#include <vm/vm.h>
150#include <vm/vm_page.h>
151#include <vm/vm_kern.h>
152
153#include <uvm/uvm.h>
154
155/*
156 * global data structures
157 */
158
159vm_map_t kernel_map = NULL;
160
161struct vmi_list vmi_list;
162simple_lock_data_t vmi_list_slock;
163
164/*
165 * local data structues
166 */
167
168static struct vm_map		kernel_map_store;
169static struct uvm_object	kmem_object_store;
170static struct uvm_object	mb_object_store;
171
172/*
173 * All pager operations here are NULL, but the object must have
174 * a pager ops vector associated with it; various places assume
175 * it to be so.
176 */
177static struct uvm_pagerops	km_pager;
178
179/*
180 * uvm_km_init: init kernel maps and objects to reflect reality (i.e.
181 * KVM already allocated for text, data, bss, and static data structures).
182 *
183 * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
184 *    we assume that [min -> start] has already been allocated and that
185 *    "end" is the end.
186 */
187
188void
189uvm_km_init(start, end)
190	vaddr_t start, end;
191{
192	vaddr_t base = VM_MIN_KERNEL_ADDRESS;
193
194	/*
195	 * first, initialize the interrupt-safe map list.
196	 */
197	LIST_INIT(&vmi_list);
198	simple_lock_init(&vmi_list_slock);
199
200	/*
201	 * next, init kernel memory objects.
202	 */
203
204	/* kernel_object: for pageable anonymous kernel memory */
205	uvm.kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
206				 VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
207
208	/*
209	 * kmem_object: for use by the kernel malloc().  Memory is always
210	 * wired, and this object (and the kmem_map) can be accessed at
211	 * interrupt time.
212	 */
213	simple_lock_init(&kmem_object_store.vmobjlock);
214	kmem_object_store.pgops = &km_pager;
215	TAILQ_INIT(&kmem_object_store.memq);
216	kmem_object_store.uo_npages = 0;
217	/* we are special.  we never die */
218	kmem_object_store.uo_refs = UVM_OBJ_KERN_INTRSAFE;
219	uvmexp.kmem_object = &kmem_object_store;
220
221	/*
222	 * mb_object: for mbuf cluster pages on platforms which use the
223	 * mb_map.  Memory is always wired, and this object (and the mb_map)
224	 * can be accessed at interrupt time.
225	 */
226	simple_lock_init(&mb_object_store.vmobjlock);
227	mb_object_store.pgops = &km_pager;
228	TAILQ_INIT(&mb_object_store.memq);
229	mb_object_store.uo_npages = 0;
230	/* we are special.  we never die */
231	mb_object_store.uo_refs = UVM_OBJ_KERN_INTRSAFE;
232	uvmexp.mb_object = &mb_object_store;
233
234	/*
235	 * init the map and reserve allready allocated kernel space
236	 * before installing.
237	 */
238
239	uvm_map_setup(&kernel_map_store, base, end, VM_MAP_PAGEABLE);
240	kernel_map_store.pmap = pmap_kernel();
241	if (uvm_map(&kernel_map_store, &base, start - base, NULL,
242	    UVM_UNKNOWN_OFFSET, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL,
243	    UVM_INH_NONE, UVM_ADV_RANDOM,UVM_FLAG_FIXED)) != KERN_SUCCESS)
244		panic("uvm_km_init: could not reserve space for kernel");
245
246	/*
247	 * install!
248	 */
249
250	kernel_map = &kernel_map_store;
251}
252
253/*
254 * uvm_km_suballoc: allocate a submap in the kernel map.   once a submap
255 * is allocated all references to that area of VM must go through it.  this
256 * allows the locking of VAs in kernel_map to be broken up into regions.
257 *
258 * => if `fixed' is true, *min specifies where the region described
259 *      by the submap must start
260 * => if submap is non NULL we use that as the submap, otherwise we
261 *	alloc a new map
262 */
263struct vm_map *
264uvm_km_suballoc(map, min, max, size, flags, fixed, submap)
265	struct vm_map *map;
266	vaddr_t *min, *max;		/* OUT, OUT */
267	vsize_t size;
268	int flags;
269	boolean_t fixed;
270	struct vm_map *submap;
271{
272	int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
273
274	size = round_page(size);	/* round up to pagesize */
275
276	/*
277	 * first allocate a blank spot in the parent map
278	 */
279
280	if (uvm_map(map, min, size, NULL, UVM_UNKNOWN_OFFSET,
281	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
282	    UVM_ADV_RANDOM, mapflags)) != KERN_SUCCESS) {
283	       panic("uvm_km_suballoc: unable to allocate space in parent map");
284	}
285
286	/*
287	 * set VM bounds (min is filled in by uvm_map)
288	 */
289
290	*max = *min + size;
291
292	/*
293	 * add references to pmap and create or init the submap
294	 */
295
296	pmap_reference(vm_map_pmap(map));
297	if (submap == NULL) {
298		submap = uvm_map_create(vm_map_pmap(map), *min, *max, flags);
299		if (submap == NULL)
300			panic("uvm_km_suballoc: unable to create submap");
301	} else {
302		uvm_map_setup(submap, *min, *max, flags);
303		submap->pmap = vm_map_pmap(map);
304	}
305
306	/*
307	 * now let uvm_map_submap plug in it...
308	 */
309
310	if (uvm_map_submap(map, *min, *max, submap) != KERN_SUCCESS)
311		panic("uvm_km_suballoc: submap allocation failed");
312
313	return(submap);
314}
315
316/*
317 * uvm_km_pgremove: remove pages from a kernel uvm_object.
318 *
319 * => when you unmap a part of anonymous kernel memory you want to toss
320 *    the pages right away.    (this gets called from uvm_unmap_...).
321 */
322
323#define UKM_HASH_PENALTY 4      /* a guess */
324
325void
326uvm_km_pgremove(uobj, start, end)
327	struct uvm_object *uobj;
328	vaddr_t start, end;
329{
330	boolean_t by_list;
331	struct vm_page *pp, *ppnext;
332	vaddr_t curoff;
333	UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
334
335	simple_lock(&uobj->vmobjlock);		/* lock object */
336
337#ifdef DIAGNOSTIC
338	if (uobj->pgops != &aobj_pager)
339		panic("uvm_km_pgremove: object %p not an aobj", uobj);
340#endif
341
342	/* choose cheapest traversal */
343	by_list = (uobj->uo_npages <=
344	     ((end - start) >> PAGE_SHIFT) * UKM_HASH_PENALTY);
345
346	if (by_list)
347		goto loop_by_list;
348
349	/* by hash */
350
351	for (curoff = start ; curoff < end ; curoff += PAGE_SIZE) {
352		pp = uvm_pagelookup(uobj, curoff);
353		if (pp == NULL)
354			continue;
355
356		UVMHIST_LOG(maphist,"  page 0x%x, busy=%d", pp,
357		    pp->flags & PG_BUSY, 0, 0);
358
359		/* now do the actual work */
360		if (pp->flags & PG_BUSY) {
361			/* owner must check for this when done */
362			pp->flags |= PG_RELEASED;
363		} else {
364			/* free the swap slot... */
365			uao_dropswap(uobj, curoff >> PAGE_SHIFT);
366
367			/*
368			 * ...and free the page; note it may be on the
369			 * active or inactive queues.
370			 */
371			uvm_lock_pageq();
372			uvm_pagefree(pp);
373			uvm_unlock_pageq();
374		}
375		/* done */
376	}
377	simple_unlock(&uobj->vmobjlock);
378	return;
379
380loop_by_list:
381
382	for (pp = uobj->memq.tqh_first ; pp != NULL ; pp = ppnext) {
383		ppnext = pp->listq.tqe_next;
384		if (pp->offset < start || pp->offset >= end) {
385			continue;
386		}
387
388		UVMHIST_LOG(maphist,"  page 0x%x, busy=%d", pp,
389		    pp->flags & PG_BUSY, 0, 0);
390
391		/* now do the actual work */
392		if (pp->flags & PG_BUSY) {
393			/* owner must check for this when done */
394			pp->flags |= PG_RELEASED;
395		} else {
396			/* free the swap slot... */
397			uao_dropswap(uobj, pp->offset >> PAGE_SHIFT);
398
399			/*
400			 * ...and free the page; note it may be on the
401			 * active or inactive queues.
402			 */
403			uvm_lock_pageq();
404			uvm_pagefree(pp);
405			uvm_unlock_pageq();
406		}
407		/* done */
408	}
409	simple_unlock(&uobj->vmobjlock);
410	return;
411}
412
413
414/*
415 * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for "intrsafe"
416 *    objects
417 *
418 * => when you unmap a part of anonymous kernel memory you want to toss
419 *    the pages right away.    (this gets called from uvm_unmap_...).
420 * => none of the pages will ever be busy, and none of them will ever
421 *    be on the active or inactive queues (because these objects are
422 *    never allowed to "page").
423 */
424
425void
426uvm_km_pgremove_intrsafe(uobj, start, end)
427	struct uvm_object *uobj;
428	vaddr_t start, end;
429{
430	boolean_t by_list;
431	struct vm_page *pp, *ppnext;
432	vaddr_t curoff;
433	UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist);
434
435	simple_lock(&uobj->vmobjlock);		/* lock object */
436
437#ifdef DIAGNOSTIC
438	if (UVM_OBJ_IS_INTRSAFE_OBJECT(uobj) == 0)
439		panic("uvm_km_pgremove_intrsafe: object %p not intrsafe", uobj);
440#endif
441
442	/* choose cheapest traversal */
443	by_list = (uobj->uo_npages <=
444	     ((end - start) >> PAGE_SHIFT) * UKM_HASH_PENALTY);
445
446	if (by_list)
447		goto loop_by_list;
448
449	/* by hash */
450
451	for (curoff = start ; curoff < end ; curoff += PAGE_SIZE) {
452		pp = uvm_pagelookup(uobj, curoff);
453		if (pp == NULL)
454			continue;
455
456		UVMHIST_LOG(maphist,"  page 0x%x, busy=%d", pp,
457		    pp->flags & PG_BUSY, 0, 0);
458#ifdef DIAGNOSTIC
459		if (pp->flags & PG_BUSY)
460			panic("uvm_km_pgremove_intrsafe: busy page");
461		if (pp->pqflags & PQ_ACTIVE)
462			panic("uvm_km_pgremove_intrsafe: active page");
463		if (pp->pqflags & PQ_INACTIVE)
464			panic("uvm_km_pgremove_intrsafe: inactive page");
465#endif
466
467		/* free the page */
468		uvm_pagefree(pp);
469	}
470	simple_unlock(&uobj->vmobjlock);
471	return;
472
473loop_by_list:
474
475	for (pp = uobj->memq.tqh_first ; pp != NULL ; pp = ppnext) {
476		ppnext = pp->listq.tqe_next;
477		if (pp->offset < start || pp->offset >= end) {
478			continue;
479		}
480
481		UVMHIST_LOG(maphist,"  page 0x%x, busy=%d", pp,
482		    pp->flags & PG_BUSY, 0, 0);
483
484#ifdef DIAGNOSTIC
485		if (pp->flags & PG_BUSY)
486			panic("uvm_km_pgremove_intrsafe: busy page");
487		if (pp->pqflags & PQ_ACTIVE)
488			panic("uvm_km_pgremove_intrsafe: active page");
489		if (pp->pqflags & PQ_INACTIVE)
490			panic("uvm_km_pgremove_intrsafe: inactive page");
491#endif
492
493		/* free the page */
494		uvm_pagefree(pp);
495	}
496	simple_unlock(&uobj->vmobjlock);
497	return;
498}
499
500
501/*
502 * uvm_km_kmemalloc: lower level kernel memory allocator for malloc()
503 *
504 * => we map wired memory into the specified map using the obj passed in
505 * => NOTE: we can return NULL even if we can wait if there is not enough
506 *	free VM space in the map... caller should be prepared to handle
507 *	this case.
508 * => we return KVA of memory allocated
509 * => flags: NOWAIT, VALLOC - just allocate VA, TRYLOCK - fail if we can't
510 *	lock the map
511 */
512
513vaddr_t
514uvm_km_kmemalloc(map, obj, size, flags)
515	vm_map_t map;
516	struct uvm_object *obj;
517	vsize_t size;
518	int flags;
519{
520	vaddr_t kva, loopva;
521	vaddr_t offset;
522	struct vm_page *pg;
523	UVMHIST_FUNC("uvm_km_kmemalloc"); UVMHIST_CALLED(maphist);
524
525
526	UVMHIST_LOG(maphist,"  (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
527	map, obj, size, flags);
528#ifdef DIAGNOSTIC
529	/* sanity check */
530	if (vm_map_pmap(map) != pmap_kernel())
531		panic("uvm_km_kmemalloc: invalid map");
532#endif
533
534	/*
535	 * setup for call
536	 */
537
538	size = round_page(size);
539	kva = vm_map_min(map);	/* hint */
540
541	/*
542	 * allocate some virtual space
543	 */
544
545	if (uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
546	      UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
547			  UVM_ADV_RANDOM, (flags & UVM_KMF_TRYLOCK)))
548			!= KERN_SUCCESS) {
549		UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
550		return(0);
551	}
552
553	/*
554	 * if all we wanted was VA, return now
555	 */
556
557	if (flags & UVM_KMF_VALLOC) {
558		UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
559		return(kva);
560	}
561	/*
562	 * recover object offset from virtual address
563	 */
564
565	offset = kva - vm_map_min(kernel_map);
566	UVMHIST_LOG(maphist, "  kva=0x%x, offset=0x%x", kva, offset,0,0);
567
568	/*
569	 * now allocate and map in the memory... note that we are the only ones
570	 * whom should ever get a handle on this area of VM.
571	 */
572
573	loopva = kva;
574	while (size) {
575		simple_lock(&obj->vmobjlock);
576		pg = uvm_pagealloc(obj, offset, NULL, 0);
577		if (pg) {
578			pg->flags &= ~PG_BUSY;	/* new page */
579			UVM_PAGE_OWN(pg, NULL);
580		}
581		simple_unlock(&obj->vmobjlock);
582
583		/*
584		 * out of memory?
585		 */
586
587		if (pg == NULL) {
588			if (flags & UVM_KMF_NOWAIT) {
589				/* free everything! */
590				uvm_unmap(map, kva, kva + size);
591				return(0);
592			} else {
593				uvm_wait("km_getwait2");	/* sleep here */
594				continue;
595			}
596		}
597
598		/*
599		 * map it in: note that we call pmap_enter with the map and
600		 * object unlocked in case we are kmem_map/kmem_object
601		 * (because if pmap_enter wants to allocate out of kmem_object
602		 * it will need to lock it itself!)
603		 */
604		if (UVM_OBJ_IS_INTRSAFE_OBJECT(obj)) {
605			pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg),
606			    VM_PROT_ALL);
607		} else {
608			pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg),
609			    UVM_PROT_ALL, TRUE, VM_PROT_READ|VM_PROT_WRITE);
610		}
611		loopva += PAGE_SIZE;
612		offset += PAGE_SIZE;
613		size -= PAGE_SIZE;
614	}
615
616	UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
617	return(kva);
618}
619
620/*
621 * uvm_km_free: free an area of kernel memory
622 */
623
624void
625uvm_km_free(map, addr, size)
626	vm_map_t map;
627	vaddr_t addr;
628	vsize_t size;
629{
630
631	uvm_unmap(map, trunc_page(addr), round_page(addr+size));
632}
633
634/*
635 * uvm_km_free_wakeup: free an area of kernel memory and wake up
636 * anyone waiting for vm space.
637 *
638 * => XXX: "wanted" bit + unlock&wait on other end?
639 */
640
641void
642uvm_km_free_wakeup(map, addr, size)
643	vm_map_t map;
644	vaddr_t addr;
645	vsize_t size;
646{
647	vm_map_entry_t dead_entries;
648
649	vm_map_lock(map);
650	(void)uvm_unmap_remove(map, trunc_page(addr), round_page(addr+size),
651			 &dead_entries);
652	wakeup(map);
653	vm_map_unlock(map);
654
655	if (dead_entries != NULL)
656		uvm_unmap_detach(dead_entries, 0);
657}
658
659/*
660 * uvm_km_alloc1: allocate wired down memory in the kernel map.
661 *
662 * => we can sleep if needed
663 */
664
665vaddr_t
666uvm_km_alloc1(map, size, zeroit)
667	vm_map_t map;
668	vsize_t size;
669	boolean_t zeroit;
670{
671	vaddr_t kva, loopva, offset;
672	struct vm_page *pg;
673	UVMHIST_FUNC("uvm_km_alloc1"); UVMHIST_CALLED(maphist);
674
675	UVMHIST_LOG(maphist,"(map=0x%x, size=0x%x)", map, size,0,0);
676
677#ifdef DIAGNOSTIC
678	if (vm_map_pmap(map) != pmap_kernel())
679		panic("uvm_km_alloc1");
680#endif
681
682	size = round_page(size);
683	kva = vm_map_min(map);		/* hint */
684
685	/*
686	 * allocate some virtual space
687	 */
688
689	if (uvm_map(map, &kva, size, uvm.kernel_object, UVM_UNKNOWN_OFFSET,
690	      UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
691			  UVM_ADV_RANDOM, 0)) != KERN_SUCCESS) {
692		UVMHIST_LOG(maphist,"<- done (no VM)",0,0,0,0);
693		return(0);
694	}
695
696	/*
697	 * recover object offset from virtual address
698	 */
699
700	offset = kva - vm_map_min(kernel_map);
701	UVMHIST_LOG(maphist,"  kva=0x%x, offset=0x%x", kva, offset,0,0);
702
703	/*
704	 * now allocate the memory.  we must be careful about released pages.
705	 */
706
707	loopva = kva;
708	while (size) {
709		simple_lock(&uvm.kernel_object->vmobjlock);
710		pg = uvm_pagelookup(uvm.kernel_object, offset);
711
712		/*
713		 * if we found a page in an unallocated region, it must be
714		 * released
715		 */
716		if (pg) {
717			if ((pg->flags & PG_RELEASED) == 0)
718				panic("uvm_km_alloc1: non-released page");
719			pg->flags |= PG_WANTED;
720			UVM_UNLOCK_AND_WAIT(pg, &uvm.kernel_object->vmobjlock,
721			    FALSE, "km_alloc", 0);
722			continue;   /* retry */
723		}
724
725		/* allocate ram */
726		pg = uvm_pagealloc(uvm.kernel_object, offset, NULL, 0);
727		if (pg) {
728			pg->flags &= ~PG_BUSY;	/* new page */
729			UVM_PAGE_OWN(pg, NULL);
730		}
731		simple_unlock(&uvm.kernel_object->vmobjlock);
732		if (pg == NULL) {
733			uvm_wait("km_alloc1w");	/* wait for memory */
734			continue;
735		}
736
737		/*
738		 * map it in; note we're never called with an intrsafe
739		 * object, so we always use regular old pmap_enter().
740		 */
741		pmap_enter(map->pmap, loopva, VM_PAGE_TO_PHYS(pg),
742		    UVM_PROT_ALL, TRUE, VM_PROT_READ|VM_PROT_WRITE);
743
744		loopva += PAGE_SIZE;
745		offset += PAGE_SIZE;
746		size -= PAGE_SIZE;
747	}
748
749	/*
750	 * zero on request (note that "size" is now zero due to the above loop
751	 * so we need to subtract kva from loopva to reconstruct the size).
752	 */
753
754	if (zeroit)
755		memset((caddr_t)kva, 0, loopva - kva);
756
757	UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
758	return(kva);
759}
760
761/*
762 * uvm_km_valloc: allocate zero-fill memory in the kernel's address space
763 *
764 * => memory is not allocated until fault time
765 */
766
767vaddr_t
768uvm_km_valloc(map, size)
769	vm_map_t map;
770	vsize_t size;
771{
772	vaddr_t kva;
773	UVMHIST_FUNC("uvm_km_valloc"); UVMHIST_CALLED(maphist);
774
775	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x)", map, size, 0,0);
776
777#ifdef DIAGNOSTIC
778	if (vm_map_pmap(map) != pmap_kernel())
779		panic("uvm_km_valloc");
780#endif
781
782	size = round_page(size);
783	kva = vm_map_min(map);		/* hint */
784
785	/*
786	 * allocate some virtual space.  will be demand filled by kernel_object.
787	 */
788
789	if (uvm_map(map, &kva, size, uvm.kernel_object, UVM_UNKNOWN_OFFSET,
790	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
791	    UVM_ADV_RANDOM, 0)) != KERN_SUCCESS) {
792		UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
793		return(0);
794	}
795
796	UVMHIST_LOG(maphist, "<- done (kva=0x%x)", kva,0,0,0);
797	return(kva);
798}
799
800/*
801 * uvm_km_valloc_wait: allocate zero-fill memory in the kernel's address space
802 *
803 * => memory is not allocated until fault time
804 * => if no room in map, wait for space to free, unless requested size
805 *    is larger than map (in which case we return 0)
806 */
807
808vaddr_t
809uvm_km_valloc_wait(map, size)
810	vm_map_t map;
811	vsize_t size;
812{
813	vaddr_t kva;
814	UVMHIST_FUNC("uvm_km_valloc_wait"); UVMHIST_CALLED(maphist);
815
816	UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x)", map, size, 0,0);
817
818#ifdef DIAGNOSTIC
819	if (vm_map_pmap(map) != pmap_kernel())
820		panic("uvm_km_valloc_wait");
821#endif
822
823	size = round_page(size);
824	if (size > vm_map_max(map) - vm_map_min(map))
825		return(0);
826
827	while (1) {
828		kva = vm_map_min(map);		/* hint */
829
830		/*
831		 * allocate some virtual space.   will be demand filled
832		 * by kernel_object.
833		 */
834
835		if (uvm_map(map, &kva, size, uvm.kernel_object,
836		    UVM_UNKNOWN_OFFSET, UVM_MAPFLAG(UVM_PROT_ALL,
837		    UVM_PROT_ALL, UVM_INH_NONE, UVM_ADV_RANDOM, 0))
838		    == KERN_SUCCESS) {
839			UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
840			return(kva);
841		}
842
843		/*
844		 * failed.  sleep for a while (on map)
845		 */
846
847		UVMHIST_LOG(maphist,"<<<sleeping>>>",0,0,0,0);
848		tsleep((caddr_t)map, PVM, "vallocwait", 0);
849	}
850	/*NOTREACHED*/
851}
852
853/* Sanity; must specify both or none. */
854#if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \
855    (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE))
856#error Must specify MAP and UNMAP together.
857#endif
858
859/*
860 * uvm_km_alloc_poolpage: allocate a page for the pool allocator
861 *
862 * => if the pmap specifies an alternate mapping method, we use it.
863 */
864
865/* ARGSUSED */
866vaddr_t
867uvm_km_alloc_poolpage1(map, obj, waitok)
868	vm_map_t map;
869	struct uvm_object *obj;
870	boolean_t waitok;
871{
872#if defined(PMAP_MAP_POOLPAGE)
873	struct vm_page *pg;
874	vaddr_t va;
875
876 again:
877	pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
878	if (pg == NULL) {
879		if (waitok) {
880			uvm_wait("plpg");
881			goto again;
882		} else
883			return (0);
884	}
885	va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
886	if (va == 0)
887		uvm_pagefree(pg);
888	return (va);
889#else
890	vaddr_t va;
891	int s;
892
893	/*
894	 * NOTE: We may be called with a map that doens't require splimp
895	 * protection (e.g. kernel_map).  However, it does not hurt to
896	 * go to splimp in this case (since unprocted maps will never be
897	 * accessed in interrupt context).
898	 *
899	 * XXX We may want to consider changing the interface to this
900	 * XXX function.
901	 */
902
903	s = splimp();
904	va = uvm_km_kmemalloc(map, obj, PAGE_SIZE, waitok ? 0 : UVM_KMF_NOWAIT);
905	splx(s);
906	return (va);
907#endif /* PMAP_MAP_POOLPAGE */
908}
909
910/*
911 * uvm_km_free_poolpage: free a previously allocated pool page
912 *
913 * => if the pmap specifies an alternate unmapping method, we use it.
914 */
915
916/* ARGSUSED */
917void
918uvm_km_free_poolpage1(map, addr)
919	vm_map_t map;
920	vaddr_t addr;
921{
922#if defined(PMAP_UNMAP_POOLPAGE)
923	paddr_t pa;
924
925	pa = PMAP_UNMAP_POOLPAGE(addr);
926	uvm_pagefree(PHYS_TO_VM_PAGE(pa));
927#else
928	int s;
929
930	/*
931	 * NOTE: We may be called with a map that doens't require splimp
932	 * protection (e.g. kernel_map).  However, it does not hurt to
933	 * go to splimp in this case (since unprocted maps will never be
934	 * accessed in interrupt context).
935	 *
936	 * XXX We may want to consider changing the interface to this
937	 * XXX function.
938	 */
939
940	s = splimp();
941	uvm_km_free(map, addr, PAGE_SIZE);
942	splx(s);
943#endif /* PMAP_UNMAP_POOLPAGE */
944}
945