uvm_km.c revision 1.117
1/*	$NetBSD: uvm_km.c,v 1.117 2012/02/02 18:59:45 para 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. Neither the name of the University nor the names of its contributors
21 *    may be used to endorse or promote products derived from this software
22 *    without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 *	@(#)vm_kern.c   8.3 (Berkeley) 1/12/94
37 * from: Id: uvm_km.c,v 1.1.2.14 1998/02/06 05:19:27 chs Exp
38 *
39 *
40 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
41 * All rights reserved.
42 *
43 * Permission to use, copy, modify and distribute this software and
44 * its documentation is hereby granted, provided that both the copyright
45 * notice and this permission notice appear in all copies of the
46 * software, derivative works or modified versions, and any portions
47 * thereof, and that both notices appear in supporting documentation.
48 *
49 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
50 * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
51 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
52 *
53 * Carnegie Mellon requests users of this software to return to
54 *
55 *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
56 *  School of Computer Science
57 *  Carnegie Mellon University
58 *  Pittsburgh PA 15213-3890
59 *
60 * any improvements or extensions that they make and grant Carnegie the
61 * rights to redistribute these changes.
62 */
63
64/*
65 * uvm_km.c: handle kernel memory allocation and management
66 */
67
68/*
69 * overview of kernel memory management:
70 *
71 * the kernel virtual address space is mapped by "kernel_map."   kernel_map
72 * starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS.
73 * note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map).
74 *
75 * the kernel_map has several "submaps."   submaps can only appear in
76 * the kernel_map (user processes can't use them).   submaps "take over"
77 * the management of a sub-range of the kernel's address space.  submaps
78 * are typically allocated at boot time and are never released.   kernel
79 * virtual address space that is mapped by a submap is locked by the
80 * submap's lock -- not the kernel_map's lock.
81 *
82 * thus, the useful feature of submaps is that they allow us to break
83 * up the locking and protection of the kernel address space into smaller
84 * chunks.
85 *
86 * the vm system has several standard kernel submaps, including:
87 *   pager_map => used to map "buf" structures into kernel space
88 *   exec_map => used during exec to handle exec args
89 *   etc...
90 *
91 * the kernel allocates its private memory out of special uvm_objects whose
92 * reference count is set to UVM_OBJ_KERN (thus indicating that the objects
93 * are "special" and never die).   all kernel objects should be thought of
94 * as large, fixed-sized, sparsely populated uvm_objects.   each kernel
95 * object is equal to the size of kernel virtual address space (i.e. the
96 * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
97 *
98 * note that just because a kernel object spans the entire kernel virtual
99 * address space doesn't mean that it has to be mapped into the entire space.
100 * large chunks of a kernel object's space go unused either because
101 * that area of kernel VM is unmapped, or there is some other type of
102 * object mapped into that range (e.g. a vnode).    for submap's kernel
103 * objects, the only part of the object that can ever be populated is the
104 * offsets that are managed by the submap.
105 *
106 * note that the "offset" in a kernel object is always the kernel virtual
107 * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
108 * example:
109 *   suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
110 *   uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
111 *   kernel map].    if uvm_km_alloc returns virtual address 0xf8235000,
112 *   then that means that the page at offset 0x235000 in kernel_object is
113 *   mapped at 0xf8235000.
114 *
115 * kernel object have one other special property: when the kernel virtual
116 * memory mapping them is unmapped, the backing memory in the object is
117 * freed right away.   this is done with the uvm_km_pgremove() function.
118 * this has to be done because there is no backing store for kernel pages
119 * and no need to save them after they are no longer referenced.
120 */
121
122#include <sys/cdefs.h>
123__KERNEL_RCSID(0, "$NetBSD: uvm_km.c,v 1.117 2012/02/02 18:59:45 para Exp $");
124
125#include "opt_uvmhist.h"
126
127#include "opt_kmempages.h"
128
129#ifndef NKMEMPAGES
130#define NKMEMPAGES 0
131#endif
132
133/*
134 * Defaults for lower and upper-bounds for the kmem_arena page count.
135 * Can be overridden by kernel config options.
136 */
137#ifndef NKMEMPAGES_MIN
138#define NKMEMPAGES_MIN NKMEMPAGES_MIN_DEFAULT
139#endif
140
141#ifndef NKMEMPAGES_MAX
142#define NKMEMPAGES_MAX NKMEMPAGES_MAX_DEFAULT
143#endif
144
145
146#include <sys/param.h>
147#include <sys/systm.h>
148#include <sys/proc.h>
149#include <sys/pool.h>
150#include <sys/vmem.h>
151#include <sys/kmem.h>
152
153#include <uvm/uvm.h>
154
155/*
156 * global data structures
157 */
158
159struct vm_map *kernel_map = NULL;
160
161/*
162 * local data structues
163 */
164
165static struct vm_map		kernel_map_store;
166static struct vm_map_entry	kernel_image_mapent_store;
167static struct vm_map_entry	kernel_kmem_mapent_store;
168
169int nkmempages = 0;
170vaddr_t kmembase;
171vsize_t kmemsize;
172
173vmem_t *kmem_arena;
174vmem_t *kmem_va_arena;
175
176/*
177 * kmeminit_nkmempages: calculate the size of kmem_arena.
178 */
179void
180kmeminit_nkmempages(void)
181{
182	int npages;
183
184	if (nkmempages != 0) {
185		/*
186		 * It's already been set (by us being here before)
187		 * bail out now;
188		 */
189		return;
190	}
191
192	npages = physmem;
193
194	if (npages > NKMEMPAGES_MAX)
195		npages = NKMEMPAGES_MAX;
196
197	if (npages < NKMEMPAGES_MIN)
198		npages = NKMEMPAGES_MIN;
199
200	nkmempages = npages;
201}
202
203/*
204 * uvm_km_bootstrap: init kernel maps and objects to reflect reality (i.e.
205 * KVM already allocated for text, data, bss, and static data structures).
206 *
207 * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
208 *    we assume that [vmin -> start] has already been allocated and that
209 *    "end" is the end.
210 */
211
212void
213uvm_km_bootstrap(vaddr_t start, vaddr_t end)
214{
215	vaddr_t base = VM_MIN_KERNEL_ADDRESS;
216
217	kmeminit_nkmempages();
218	kmemsize = nkmempages * PAGE_SIZE;
219
220	/* kmemsize = MIN((((vsize_t)(end - start)) / 3),
221	    ((((vsize_t)uvmexp.npages) * PAGE_SIZE) / 2));
222	kmemsize = round_page(kmemsize); */
223
224	/*
225	 * next, init kernel memory objects.
226	 */
227
228	/* kernel_object: for pageable anonymous kernel memory */
229	uvm_kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
230				VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
231
232	/*
233	 * init the map and reserve any space that might already
234	 * have been allocated kernel space before installing.
235	 */
236
237	uvm_map_setup(&kernel_map_store, base, end, VM_MAP_PAGEABLE);
238	kernel_map_store.pmap = pmap_kernel();
239	if (start != base) {
240		int error;
241		struct uvm_map_args args;
242
243		error = uvm_map_prepare(&kernel_map_store,
244		    base, start - base,
245		    NULL, UVM_UNKNOWN_OFFSET, 0,
246		    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
247		    		UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args);
248		if (!error) {
249			kernel_image_mapent_store.flags =
250			    UVM_MAP_KERNEL | UVM_MAP_STATIC | UVM_MAP_NOMERGE;
251			error = uvm_map_enter(&kernel_map_store, &args,
252			    &kernel_image_mapent_store);
253		}
254
255		if (error)
256			panic(
257			    "uvm_km_bootstrap: could not reserve space for kernel");
258
259		kmembase = args.uma_start + args.uma_size;
260		error = uvm_map_prepare(&kernel_map_store,
261		    kmembase, kmemsize,
262		    NULL, UVM_UNKNOWN_OFFSET, 0,
263		    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
264		    		UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args);
265		if (!error) {
266			kernel_kmem_mapent_store.flags =
267			    UVM_MAP_KERNEL | UVM_MAP_STATIC | UVM_MAP_NOMERGE;
268			error = uvm_map_enter(&kernel_map_store, &args,
269			    &kernel_kmem_mapent_store);
270		}
271
272		if (error)
273			panic(
274			    "uvm_km_bootstrap: could not reserve kernel kmem");
275	} else {
276		kmembase = base;
277	}
278
279	/*
280	 * install!
281	 */
282
283	kernel_map = &kernel_map_store;
284
285	pool_subsystem_init();
286	vmem_bootstrap();
287
288	kmem_arena = vmem_create("kmem", kmembase, kmemsize, PAGE_SIZE,
289	    NULL, NULL, NULL,
290	    0, VM_NOSLEEP | VM_BOOTSTRAP, IPL_VM);
291
292	vmem_init(kmem_arena);
293
294	kmem_va_arena = vmem_create("kva", 0, 0, PAGE_SIZE,
295	    vmem_alloc, vmem_free, kmem_arena,
296	    16 * PAGE_SIZE, VM_NOSLEEP | VM_BOOTSTRAP, IPL_VM);
297}
298
299/*
300 * uvm_km_init: init the kernel maps virtual memory caches
301 * and start the pool/kmem allocator.
302 */
303void
304uvm_km_init(void)
305{
306
307	kmem_init();
308
309	kmeminit(); // killme
310}
311
312/*
313 * uvm_km_suballoc: allocate a submap in the kernel map.   once a submap
314 * is allocated all references to that area of VM must go through it.  this
315 * allows the locking of VAs in kernel_map to be broken up into regions.
316 *
317 * => if `fixed' is true, *vmin specifies where the region described
318 *   pager_map => used to map "buf" structures into kernel space
319 *      by the submap must start
320 * => if submap is non NULL we use that as the submap, otherwise we
321 *	alloc a new map
322 */
323
324struct vm_map *
325uvm_km_suballoc(struct vm_map *map, vaddr_t *vmin /* IN/OUT */,
326    vaddr_t *vmax /* OUT */, vsize_t size, int flags, bool fixed,
327    struct vm_map *submap)
328{
329	int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
330
331	KASSERT(vm_map_pmap(map) == pmap_kernel());
332
333	size = round_page(size);	/* round up to pagesize */
334
335	/*
336	 * first allocate a blank spot in the parent map
337	 */
338
339	if (uvm_map(map, vmin, size, NULL, UVM_UNKNOWN_OFFSET, 0,
340	    UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
341	    UVM_ADV_RANDOM, mapflags)) != 0) {
342	       panic("uvm_km_suballoc: unable to allocate space in parent map");
343	}
344
345	/*
346	 * set VM bounds (vmin is filled in by uvm_map)
347	 */
348
349	*vmax = *vmin + size;
350
351	/*
352	 * add references to pmap and create or init the submap
353	 */
354
355	pmap_reference(vm_map_pmap(map));
356	if (submap == NULL) {
357		submap = kmem_alloc(sizeof(*submap), KM_SLEEP);
358		if (submap == NULL)
359			panic("uvm_km_suballoc: unable to create submap");
360	}
361	uvm_map_setup(submap, *vmin, *vmax, flags);
362	submap->pmap = vm_map_pmap(map);
363
364	/*
365	 * now let uvm_map_submap plug in it...
366	 */
367
368	if (uvm_map_submap(map, *vmin, *vmax, submap) != 0)
369		panic("uvm_km_suballoc: submap allocation failed");
370
371	return(submap);
372}
373
374/*
375 * uvm_km_pgremove: remove pages from a kernel uvm_object and KVA.
376 */
377
378void
379uvm_km_pgremove(vaddr_t startva, vaddr_t endva)
380{
381	struct uvm_object * const uobj = uvm_kernel_object;
382	const voff_t start = startva - vm_map_min(kernel_map);
383	const voff_t end = endva - vm_map_min(kernel_map);
384	struct vm_page *pg;
385	voff_t curoff, nextoff;
386	int swpgonlydelta = 0;
387	UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
388
389	KASSERT(VM_MIN_KERNEL_ADDRESS <= startva);
390	KASSERT(startva < endva);
391	KASSERT(endva <= VM_MAX_KERNEL_ADDRESS);
392
393	mutex_enter(uobj->vmobjlock);
394	pmap_remove(pmap_kernel(), startva, endva);
395	for (curoff = start; curoff < end; curoff = nextoff) {
396		nextoff = curoff + PAGE_SIZE;
397		pg = uvm_pagelookup(uobj, curoff);
398		if (pg != NULL && pg->flags & PG_BUSY) {
399			pg->flags |= PG_WANTED;
400			UVM_UNLOCK_AND_WAIT(pg, uobj->vmobjlock, 0,
401				    "km_pgrm", 0);
402			mutex_enter(uobj->vmobjlock);
403			nextoff = curoff;
404			continue;
405		}
406
407		/*
408		 * free the swap slot, then the page.
409		 */
410
411		if (pg == NULL &&
412		    uao_find_swslot(uobj, curoff >> PAGE_SHIFT) > 0) {
413			swpgonlydelta++;
414		}
415		uao_dropswap(uobj, curoff >> PAGE_SHIFT);
416		if (pg != NULL) {
417			mutex_enter(&uvm_pageqlock);
418			uvm_pagefree(pg);
419			mutex_exit(&uvm_pageqlock);
420		}
421	}
422	mutex_exit(uobj->vmobjlock);
423
424	if (swpgonlydelta > 0) {
425		mutex_enter(&uvm_swap_data_lock);
426		KASSERT(uvmexp.swpgonly >= swpgonlydelta);
427		uvmexp.swpgonly -= swpgonlydelta;
428		mutex_exit(&uvm_swap_data_lock);
429	}
430}
431
432
433/*
434 * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for non object backed
435 *    regions.
436 *
437 * => when you unmap a part of anonymous kernel memory you want to toss
438 *    the pages right away.    (this is called from uvm_unmap_...).
439 * => none of the pages will ever be busy, and none of them will ever
440 *    be on the active or inactive queues (because they have no object).
441 */
442
443void
444uvm_km_pgremove_intrsafe(struct vm_map *map, vaddr_t start, vaddr_t end)
445{
446	struct vm_page *pg;
447	paddr_t pa;
448	UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist);
449
450	KASSERT(VM_MAP_IS_KERNEL(map));
451	KASSERT(vm_map_min(map) <= start);
452	KASSERT(start < end);
453	KASSERT(end <= vm_map_max(map));
454
455	for (; start < end; start += PAGE_SIZE) {
456		if (!pmap_extract(pmap_kernel(), start, &pa)) {
457			continue;
458		}
459		pg = PHYS_TO_VM_PAGE(pa);
460		KASSERT(pg);
461		KASSERT(pg->uobject == NULL && pg->uanon == NULL);
462		KASSERT((pg->flags & PG_BUSY) == 0);
463		uvm_pagefree(pg);
464	}
465}
466
467#if defined(DEBUG)
468void
469uvm_km_check_empty(struct vm_map *map, vaddr_t start, vaddr_t end)
470{
471	struct vm_page *pg;
472	vaddr_t va;
473	paddr_t pa;
474
475	KDASSERT(VM_MAP_IS_KERNEL(map));
476	KDASSERT(vm_map_min(map) <= start);
477	KDASSERT(start < end);
478	KDASSERT(end <= vm_map_max(map));
479
480	for (va = start; va < end; va += PAGE_SIZE) {
481		if (pmap_extract(pmap_kernel(), va, &pa)) {
482			panic("uvm_km_check_empty: va %p has pa 0x%llx",
483			    (void *)va, (long long)pa);
484		}
485		if ((map->flags & VM_MAP_INTRSAFE) == 0) {
486			mutex_enter(uvm_kernel_object->vmobjlock);
487			pg = uvm_pagelookup(uvm_kernel_object,
488			    va - vm_map_min(kernel_map));
489			mutex_exit(uvm_kernel_object->vmobjlock);
490			if (pg) {
491				panic("uvm_km_check_empty: "
492				    "has page hashed at %p", (const void *)va);
493			}
494		}
495	}
496}
497#endif /* defined(DEBUG) */
498
499/*
500 * uvm_km_alloc: allocate an area of kernel memory.
501 *
502 * => NOTE: we can return 0 even if we can wait if there is not enough
503 *	free VM space in the map... caller should be prepared to handle
504 *	this case.
505 * => we return KVA of memory allocated
506 */
507
508vaddr_t
509uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags)
510{
511	vaddr_t kva, loopva;
512	vaddr_t offset;
513	vsize_t loopsize;
514	struct vm_page *pg;
515	struct uvm_object *obj;
516	int pgaflags;
517	vm_prot_t prot;
518	UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist);
519
520	KASSERT(vm_map_pmap(map) == pmap_kernel());
521	KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
522		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
523		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
524	KASSERT((flags & UVM_KMF_VAONLY) != 0 || (flags & UVM_KMF_COLORMATCH) == 0);
525	KASSERT((flags & UVM_KMF_COLORMATCH) == 0 || (flags & UVM_KMF_VAONLY) != 0);
526
527	/*
528	 * setup for call
529	 */
530
531	kva = vm_map_min(map);	/* hint */
532	size = round_page(size);
533	obj = (flags & UVM_KMF_PAGEABLE) ? uvm_kernel_object : NULL;
534	UVMHIST_LOG(maphist,"  (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
535		    map, obj, size, flags);
536
537	/*
538	 * allocate some virtual space
539	 */
540
541	if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
542	    align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
543	    UVM_ADV_RANDOM,
544	    (flags & (UVM_KMF_TRYLOCK | UVM_KMF_NOWAIT | UVM_KMF_WAITVA
545	     | UVM_KMF_COLORMATCH)))) != 0)) {
546		UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
547		return(0);
548	}
549
550	/*
551	 * if all we wanted was VA, return now
552	 */
553
554	if (flags & (UVM_KMF_VAONLY | UVM_KMF_PAGEABLE)) {
555		UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
556		return(kva);
557	}
558
559	/*
560	 * recover object offset from virtual address
561	 */
562
563	offset = kva - vm_map_min(kernel_map);
564	UVMHIST_LOG(maphist, "  kva=0x%x, offset=0x%x", kva, offset,0,0);
565
566	/*
567	 * now allocate and map in the memory... note that we are the only ones
568	 * whom should ever get a handle on this area of VM.
569	 */
570
571	loopva = kva;
572	loopsize = size;
573
574	pgaflags = UVM_FLAG_COLORMATCH;
575	if (flags & UVM_KMF_NOWAIT)
576		pgaflags |= UVM_PGA_USERESERVE;
577	if (flags & UVM_KMF_ZERO)
578		pgaflags |= UVM_PGA_ZERO;
579	prot = VM_PROT_READ | VM_PROT_WRITE;
580	if (flags & UVM_KMF_EXEC)
581		prot |= VM_PROT_EXECUTE;
582	while (loopsize) {
583		KASSERTMSG(!pmap_extract(pmap_kernel(), loopva, NULL),
584		    "loopva=%#"PRIxVADDR, loopva);
585
586		pg = uvm_pagealloc_strat(NULL, offset, NULL, pgaflags,
587#ifdef UVM_KM_VMFREELIST
588		   UVM_PGA_STRAT_ONLY, UVM_KM_VMFREELIST
589#else
590		   UVM_PGA_STRAT_NORMAL, 0
591#endif
592		   );
593
594		/*
595		 * out of memory?
596		 */
597
598		if (__predict_false(pg == NULL)) {
599			if ((flags & UVM_KMF_NOWAIT) ||
600			    ((flags & UVM_KMF_CANFAIL) && !uvm_reclaimable())) {
601				/* free everything! */
602				uvm_km_free(map, kva, size,
603				    flags & UVM_KMF_TYPEMASK);
604				return (0);
605			} else {
606				uvm_wait("km_getwait2");	/* sleep here */
607				continue;
608			}
609		}
610
611		pg->flags &= ~PG_BUSY;	/* new page */
612		UVM_PAGE_OWN(pg, NULL);
613
614		/*
615		 * map it in
616		 */
617
618		pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg),
619		    prot, PMAP_KMPAGE);
620		loopva += PAGE_SIZE;
621		offset += PAGE_SIZE;
622		loopsize -= PAGE_SIZE;
623	}
624
625	pmap_update(pmap_kernel());
626
627	UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
628	return(kva);
629}
630
631/*
632 * uvm_km_free: free an area of kernel memory
633 */
634
635void
636uvm_km_free(struct vm_map *map, vaddr_t addr, vsize_t size, uvm_flag_t flags)
637{
638
639	KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
640		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
641		(flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
642	KASSERT((addr & PAGE_MASK) == 0);
643	KASSERT(vm_map_pmap(map) == pmap_kernel());
644
645	size = round_page(size);
646
647	if (flags & UVM_KMF_PAGEABLE) {
648		uvm_km_pgremove(addr, addr + size);
649	} else if (flags & UVM_KMF_WIRED) {
650		/*
651		 * Note: uvm_km_pgremove_intrsafe() extracts mapping, thus
652		 * remove it after.  See comment below about KVA visibility.
653		 */
654		uvm_km_pgremove_intrsafe(map, addr, addr + size);
655		pmap_kremove(addr, size);
656	}
657
658	/*
659	 * Note: uvm_unmap_remove() calls pmap_update() for us, before
660	 * KVA becomes globally available.
661	 */
662
663	uvm_unmap1(map, addr, addr + size, UVM_FLAG_VAONLY);
664}
665
666/* Sanity; must specify both or none. */
667#if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \
668    (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE))
669#error Must specify MAP and UNMAP together.
670#endif
671
672int
673uvm_km_kmem_alloc(vmem_t *vm, vmem_size_t size, vm_flag_t flags,
674    vmem_addr_t *addr)
675{
676	struct vm_page *pg;
677	vmem_addr_t va;
678	int rc;
679	vaddr_t loopva;
680	vsize_t loopsize;
681
682	size = round_page(size);
683
684#if defined(PMAP_MAP_POOLPAGE)
685	if (size == PAGE_SIZE) {
686again:
687#ifdef PMAP_ALLOC_POOLPAGE
688		pg = PMAP_ALLOC_POOLPAGE((flags & VM_SLEEP) ?
689		   0 : UVM_PGA_USERESERVE);
690#else
691		pg = uvm_pagealloc(NULL, 0, NULL,
692		   (flags & VM_SLEEP) ? 0 : UVM_PGA_USERESERVE);
693#endif /* PMAP_ALLOC_POOLPAGE */
694		if (__predict_false(pg == NULL)) {
695			if (flags & VM_SLEEP) {
696				uvm_wait("plpg");
697				goto again;
698			}
699		}
700		va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
701		if (__predict_false(va == 0)) {
702			uvm_pagefree(pg);
703			return ENOMEM;
704		}
705		*addr = va;
706		return 0;
707	}
708#endif /* PMAP_MAP_POOLPAGE */
709
710	rc = vmem_alloc(vm, size, flags, &va);
711	if (rc != 0)
712		return rc;
713
714	loopva = va;
715	loopsize = size;
716
717	while (loopsize) {
718		KASSERTMSG(!pmap_extract(pmap_kernel(), loopva, NULL),
719		    "loopva=%#"PRIxVADDR" loopsize=%#"PRIxVSIZE" vmem=%p",
720		    loopva, loopsize, vm);
721
722		pg = uvm_pagealloc(NULL, loopva, NULL,
723		    UVM_FLAG_COLORMATCH
724		    | ((flags & VM_SLEEP) ? 0 : UVM_PGA_USERESERVE));
725		if (__predict_false(pg == NULL)) {
726			if (flags & VM_SLEEP) {
727				uvm_wait("plpg");
728				continue;
729			} else {
730				uvm_km_pgremove_intrsafe(kernel_map, va,
731				    va + size);
732				pmap_kremove(va, size);
733				vmem_free(kmem_va_arena, va, size);
734				return ENOMEM;
735			}
736		}
737
738		pg->flags &= ~PG_BUSY;	/* new page */
739		UVM_PAGE_OWN(pg, NULL);
740		pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg),
741		    VM_PROT_READ|VM_PROT_WRITE, PMAP_KMPAGE);
742
743		loopva += PAGE_SIZE;
744		loopsize -= PAGE_SIZE;
745	}
746	pmap_update(pmap_kernel());
747
748	*addr = va;
749
750	return 0;
751}
752
753void
754uvm_km_kmem_free(vmem_t *vm, vmem_addr_t addr, size_t size)
755{
756
757	size = round_page(size);
758#if defined(PMAP_UNMAP_POOLPAGE)
759	if (size == PAGE_SIZE) {
760		paddr_t pa;
761
762		pa = PMAP_UNMAP_POOLPAGE(addr);
763		uvm_pagefree(PHYS_TO_VM_PAGE(pa));
764		return;
765	}
766#endif /* PMAP_UNMAP_POOLPAGE */
767	uvm_km_pgremove_intrsafe(kernel_map, addr, addr + size);
768	pmap_kremove(addr, size);
769	pmap_update(pmap_kernel());
770
771	vmem_free(vm, addr, size);
772}
773
774bool
775uvm_km_va_starved_p(void)
776{
777	vmem_size_t total;
778	vmem_size_t free;
779
780	total = vmem_size(kmem_arena, VMEM_ALLOC|VMEM_FREE);
781	free = vmem_size(kmem_arena, VMEM_FREE);
782
783	return (free < (total / 10));
784}
785
786