uvm_km.c revision 1.86 1 1.85 yamt /* $NetBSD: uvm_km.c,v 1.86 2006/04/05 21:56:24 yamt Exp $ */
2 1.1 mrg
3 1.47 chs /*
4 1.1 mrg * Copyright (c) 1997 Charles D. Cranor and Washington University.
5 1.47 chs * Copyright (c) 1991, 1993, The Regents of the University of California.
6 1.1 mrg *
7 1.1 mrg * All rights reserved.
8 1.1 mrg *
9 1.1 mrg * This code is derived from software contributed to Berkeley by
10 1.1 mrg * The Mach Operating System project at Carnegie-Mellon University.
11 1.1 mrg *
12 1.1 mrg * Redistribution and use in source and binary forms, with or without
13 1.1 mrg * modification, are permitted provided that the following conditions
14 1.1 mrg * are met:
15 1.1 mrg * 1. Redistributions of source code must retain the above copyright
16 1.1 mrg * notice, this list of conditions and the following disclaimer.
17 1.1 mrg * 2. Redistributions in binary form must reproduce the above copyright
18 1.1 mrg * notice, this list of conditions and the following disclaimer in the
19 1.1 mrg * documentation and/or other materials provided with the distribution.
20 1.1 mrg * 3. All advertising materials mentioning features or use of this software
21 1.1 mrg * must display the following acknowledgement:
22 1.1 mrg * This product includes software developed by Charles D. Cranor,
23 1.47 chs * Washington University, the University of California, Berkeley and
24 1.1 mrg * its contributors.
25 1.1 mrg * 4. Neither the name of the University nor the names of its contributors
26 1.1 mrg * may be used to endorse or promote products derived from this software
27 1.1 mrg * without specific prior written permission.
28 1.1 mrg *
29 1.1 mrg * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 1.1 mrg * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 1.1 mrg * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 1.1 mrg * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 1.1 mrg * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 1.1 mrg * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 1.1 mrg * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 1.1 mrg * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 1.1 mrg * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 1.1 mrg * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 1.1 mrg * SUCH DAMAGE.
40 1.1 mrg *
41 1.1 mrg * @(#)vm_kern.c 8.3 (Berkeley) 1/12/94
42 1.4 mrg * from: Id: uvm_km.c,v 1.1.2.14 1998/02/06 05:19:27 chs Exp
43 1.1 mrg *
44 1.1 mrg *
45 1.1 mrg * Copyright (c) 1987, 1990 Carnegie-Mellon University.
46 1.1 mrg * All rights reserved.
47 1.47 chs *
48 1.1 mrg * Permission to use, copy, modify and distribute this software and
49 1.1 mrg * its documentation is hereby granted, provided that both the copyright
50 1.1 mrg * notice and this permission notice appear in all copies of the
51 1.1 mrg * software, derivative works or modified versions, and any portions
52 1.1 mrg * thereof, and that both notices appear in supporting documentation.
53 1.47 chs *
54 1.47 chs * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
55 1.47 chs * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
56 1.1 mrg * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
57 1.47 chs *
58 1.1 mrg * Carnegie Mellon requests users of this software to return to
59 1.1 mrg *
60 1.1 mrg * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
61 1.1 mrg * School of Computer Science
62 1.1 mrg * Carnegie Mellon University
63 1.1 mrg * Pittsburgh PA 15213-3890
64 1.1 mrg *
65 1.1 mrg * any improvements or extensions that they make and grant Carnegie the
66 1.1 mrg * rights to redistribute these changes.
67 1.1 mrg */
68 1.6 mrg
69 1.1 mrg /*
70 1.1 mrg * uvm_km.c: handle kernel memory allocation and management
71 1.1 mrg */
72 1.1 mrg
73 1.7 chuck /*
74 1.7 chuck * overview of kernel memory management:
75 1.7 chuck *
76 1.7 chuck * the kernel virtual address space is mapped by "kernel_map." kernel_map
77 1.62 thorpej * starts at VM_MIN_KERNEL_ADDRESS and goes to VM_MAX_KERNEL_ADDRESS.
78 1.62 thorpej * note that VM_MIN_KERNEL_ADDRESS is equal to vm_map_min(kernel_map).
79 1.7 chuck *
80 1.47 chs * the kernel_map has several "submaps." submaps can only appear in
81 1.7 chuck * the kernel_map (user processes can't use them). submaps "take over"
82 1.7 chuck * the management of a sub-range of the kernel's address space. submaps
83 1.7 chuck * are typically allocated at boot time and are never released. kernel
84 1.47 chs * virtual address space that is mapped by a submap is locked by the
85 1.7 chuck * submap's lock -- not the kernel_map's lock.
86 1.7 chuck *
87 1.7 chuck * thus, the useful feature of submaps is that they allow us to break
88 1.7 chuck * up the locking and protection of the kernel address space into smaller
89 1.7 chuck * chunks.
90 1.7 chuck *
91 1.7 chuck * the vm system has several standard kernel submaps, including:
92 1.7 chuck * kmem_map => contains only wired kernel memory for the kernel
93 1.7 chuck * malloc. *** access to kmem_map must be protected
94 1.42 thorpej * by splvm() because we are allowed to call malloc()
95 1.7 chuck * at interrupt time ***
96 1.42 thorpej * mb_map => memory for large mbufs, *** protected by splvm ***
97 1.7 chuck * pager_map => used to map "buf" structures into kernel space
98 1.7 chuck * exec_map => used during exec to handle exec args
99 1.7 chuck * etc...
100 1.7 chuck *
101 1.7 chuck * the kernel allocates its private memory out of special uvm_objects whose
102 1.7 chuck * reference count is set to UVM_OBJ_KERN (thus indicating that the objects
103 1.7 chuck * are "special" and never die). all kernel objects should be thought of
104 1.47 chs * as large, fixed-sized, sparsely populated uvm_objects. each kernel
105 1.62 thorpej * object is equal to the size of kernel virtual address space (i.e. the
106 1.62 thorpej * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
107 1.7 chuck *
108 1.7 chuck * note that just because a kernel object spans the entire kernel virutal
109 1.7 chuck * address space doesn't mean that it has to be mapped into the entire space.
110 1.47 chs * large chunks of a kernel object's space go unused either because
111 1.47 chs * that area of kernel VM is unmapped, or there is some other type of
112 1.7 chuck * object mapped into that range (e.g. a vnode). for submap's kernel
113 1.7 chuck * objects, the only part of the object that can ever be populated is the
114 1.7 chuck * offsets that are managed by the submap.
115 1.7 chuck *
116 1.7 chuck * note that the "offset" in a kernel object is always the kernel virtual
117 1.62 thorpej * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
118 1.7 chuck * example:
119 1.62 thorpej * suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
120 1.7 chuck * uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
121 1.7 chuck * kernel map]. if uvm_km_alloc returns virtual address 0xf8235000,
122 1.7 chuck * then that means that the page at offset 0x235000 in kernel_object is
123 1.47 chs * mapped at 0xf8235000.
124 1.7 chuck *
125 1.7 chuck * kernel object have one other special property: when the kernel virtual
126 1.7 chuck * memory mapping them is unmapped, the backing memory in the object is
127 1.7 chuck * freed right away. this is done with the uvm_km_pgremove() function.
128 1.7 chuck * this has to be done because there is no backing store for kernel pages
129 1.7 chuck * and no need to save them after they are no longer referenced.
130 1.7 chuck */
131 1.55 lukem
132 1.55 lukem #include <sys/cdefs.h>
133 1.85 yamt __KERNEL_RCSID(0, "$NetBSD: uvm_km.c,v 1.86 2006/04/05 21:56:24 yamt Exp $");
134 1.55 lukem
135 1.55 lukem #include "opt_uvmhist.h"
136 1.7 chuck
137 1.1 mrg #include <sys/param.h>
138 1.71 yamt #include <sys/malloc.h>
139 1.1 mrg #include <sys/systm.h>
140 1.1 mrg #include <sys/proc.h>
141 1.72 yamt #include <sys/pool.h>
142 1.1 mrg
143 1.1 mrg #include <uvm/uvm.h>
144 1.1 mrg
145 1.1 mrg /*
146 1.1 mrg * global data structures
147 1.1 mrg */
148 1.1 mrg
149 1.49 chs struct vm_map *kernel_map = NULL;
150 1.1 mrg
151 1.1 mrg /*
152 1.1 mrg * local data structues
153 1.1 mrg */
154 1.1 mrg
155 1.71 yamt static struct vm_map_kernel kernel_map_store;
156 1.70 yamt static struct vm_map_entry kernel_first_mapent_store;
157 1.1 mrg
158 1.72 yamt #if !defined(PMAP_MAP_POOLPAGE)
159 1.72 yamt
160 1.72 yamt /*
161 1.72 yamt * kva cache
162 1.72 yamt *
163 1.72 yamt * XXX maybe it's better to do this at the uvm_map layer.
164 1.72 yamt */
165 1.72 yamt
166 1.72 yamt #define KM_VACACHE_SIZE (32 * PAGE_SIZE) /* XXX tune */
167 1.72 yamt
168 1.72 yamt static void *km_vacache_alloc(struct pool *, int);
169 1.72 yamt static void km_vacache_free(struct pool *, void *);
170 1.72 yamt static void km_vacache_init(struct vm_map *, const char *, size_t);
171 1.72 yamt
172 1.72 yamt /* XXX */
173 1.72 yamt #define KM_VACACHE_POOL_TO_MAP(pp) \
174 1.72 yamt ((struct vm_map *)((char *)(pp) - \
175 1.72 yamt offsetof(struct vm_map_kernel, vmk_vacache)))
176 1.72 yamt
177 1.72 yamt static void *
178 1.72 yamt km_vacache_alloc(struct pool *pp, int flags)
179 1.72 yamt {
180 1.72 yamt vaddr_t va;
181 1.72 yamt size_t size;
182 1.72 yamt struct vm_map *map;
183 1.72 yamt size = pp->pr_alloc->pa_pagesz;
184 1.72 yamt
185 1.72 yamt map = KM_VACACHE_POOL_TO_MAP(pp);
186 1.72 yamt
187 1.73 yamt va = vm_map_min(map); /* hint */
188 1.72 yamt if (uvm_map(map, &va, size, NULL, UVM_UNKNOWN_OFFSET, size,
189 1.74 yamt UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
190 1.72 yamt UVM_ADV_RANDOM, UVM_FLAG_QUANTUM |
191 1.72 yamt ((flags & PR_WAITOK) ? 0 : UVM_FLAG_TRYLOCK | UVM_FLAG_NOWAIT))))
192 1.72 yamt return NULL;
193 1.72 yamt
194 1.72 yamt return (void *)va;
195 1.72 yamt }
196 1.72 yamt
197 1.72 yamt static void
198 1.72 yamt km_vacache_free(struct pool *pp, void *v)
199 1.72 yamt {
200 1.72 yamt vaddr_t va = (vaddr_t)v;
201 1.72 yamt size_t size = pp->pr_alloc->pa_pagesz;
202 1.72 yamt struct vm_map *map;
203 1.72 yamt
204 1.72 yamt map = KM_VACACHE_POOL_TO_MAP(pp);
205 1.78 yamt uvm_unmap1(map, va, va + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
206 1.72 yamt }
207 1.72 yamt
208 1.72 yamt /*
209 1.72 yamt * km_vacache_init: initialize kva cache.
210 1.72 yamt */
211 1.72 yamt
212 1.72 yamt static void
213 1.72 yamt km_vacache_init(struct vm_map *map, const char *name, size_t size)
214 1.72 yamt {
215 1.72 yamt struct vm_map_kernel *vmk;
216 1.72 yamt struct pool *pp;
217 1.72 yamt struct pool_allocator *pa;
218 1.72 yamt
219 1.72 yamt KASSERT(VM_MAP_IS_KERNEL(map));
220 1.72 yamt KASSERT(size < (vm_map_max(map) - vm_map_min(map)) / 2); /* sanity */
221 1.72 yamt
222 1.72 yamt vmk = vm_map_to_kernel(map);
223 1.72 yamt pp = &vmk->vmk_vacache;
224 1.72 yamt pa = &vmk->vmk_vacache_allocator;
225 1.72 yamt memset(pa, 0, sizeof(*pa));
226 1.72 yamt pa->pa_alloc = km_vacache_alloc;
227 1.72 yamt pa->pa_free = km_vacache_free;
228 1.72 yamt pa->pa_pagesz = (unsigned int)size;
229 1.72 yamt pool_init(pp, PAGE_SIZE, 0, 0, PR_NOTOUCH | PR_RECURSIVE, name, pa);
230 1.72 yamt
231 1.72 yamt /* XXX for now.. */
232 1.72 yamt pool_sethiwat(pp, 0);
233 1.72 yamt }
234 1.72 yamt
235 1.72 yamt void
236 1.72 yamt uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
237 1.72 yamt {
238 1.72 yamt
239 1.72 yamt map->flags |= VM_MAP_VACACHE;
240 1.72 yamt if (size == 0)
241 1.72 yamt size = KM_VACACHE_SIZE;
242 1.72 yamt km_vacache_init(map, name, size);
243 1.72 yamt }
244 1.72 yamt
245 1.72 yamt #else /* !defined(PMAP_MAP_POOLPAGE) */
246 1.72 yamt
247 1.72 yamt void
248 1.72 yamt uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
249 1.72 yamt {
250 1.72 yamt
251 1.72 yamt /* nothing */
252 1.72 yamt }
253 1.72 yamt
254 1.72 yamt #endif /* !defined(PMAP_MAP_POOLPAGE) */
255 1.72 yamt
256 1.1 mrg /*
257 1.1 mrg * uvm_km_init: init kernel maps and objects to reflect reality (i.e.
258 1.1 mrg * KVM already allocated for text, data, bss, and static data structures).
259 1.1 mrg *
260 1.62 thorpej * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
261 1.82 christos * we assume that [vmin -> start] has already been allocated and that
262 1.62 thorpej * "end" is the end.
263 1.1 mrg */
264 1.1 mrg
265 1.8 mrg void
266 1.83 thorpej uvm_km_init(vaddr_t start, vaddr_t end)
267 1.1 mrg {
268 1.62 thorpej vaddr_t base = VM_MIN_KERNEL_ADDRESS;
269 1.27 thorpej
270 1.27 thorpej /*
271 1.27 thorpej * next, init kernel memory objects.
272 1.8 mrg */
273 1.1 mrg
274 1.8 mrg /* kernel_object: for pageable anonymous kernel memory */
275 1.34 chs uao_init();
276 1.62 thorpej uvm.kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
277 1.62 thorpej VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
278 1.1 mrg
279 1.24 thorpej /*
280 1.56 thorpej * init the map and reserve any space that might already
281 1.56 thorpej * have been allocated kernel space before installing.
282 1.8 mrg */
283 1.1 mrg
284 1.71 yamt uvm_map_setup_kernel(&kernel_map_store, base, end, VM_MAP_PAGEABLE);
285 1.71 yamt kernel_map_store.vmk_map.pmap = pmap_kernel();
286 1.70 yamt if (start != base) {
287 1.70 yamt int error;
288 1.70 yamt struct uvm_map_args args;
289 1.70 yamt
290 1.71 yamt error = uvm_map_prepare(&kernel_map_store.vmk_map,
291 1.71 yamt base, start - base,
292 1.70 yamt NULL, UVM_UNKNOWN_OFFSET, 0,
293 1.62 thorpej UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
294 1.70 yamt UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args);
295 1.70 yamt if (!error) {
296 1.70 yamt kernel_first_mapent_store.flags =
297 1.70 yamt UVM_MAP_KERNEL | UVM_MAP_FIRST;
298 1.71 yamt error = uvm_map_enter(&kernel_map_store.vmk_map, &args,
299 1.70 yamt &kernel_first_mapent_store);
300 1.70 yamt }
301 1.70 yamt
302 1.70 yamt if (error)
303 1.70 yamt panic(
304 1.70 yamt "uvm_km_init: could not reserve space for kernel");
305 1.70 yamt }
306 1.47 chs
307 1.8 mrg /*
308 1.8 mrg * install!
309 1.8 mrg */
310 1.8 mrg
311 1.71 yamt kernel_map = &kernel_map_store.vmk_map;
312 1.72 yamt uvm_km_vacache_init(kernel_map, "kvakernel", 0);
313 1.1 mrg }
314 1.1 mrg
315 1.1 mrg /*
316 1.1 mrg * uvm_km_suballoc: allocate a submap in the kernel map. once a submap
317 1.1 mrg * is allocated all references to that area of VM must go through it. this
318 1.1 mrg * allows the locking of VAs in kernel_map to be broken up into regions.
319 1.1 mrg *
320 1.82 christos * => if `fixed' is true, *vmin specifies where the region described
321 1.5 thorpej * by the submap must start
322 1.1 mrg * => if submap is non NULL we use that as the submap, otherwise we
323 1.1 mrg * alloc a new map
324 1.1 mrg */
325 1.78 yamt
326 1.8 mrg struct vm_map *
327 1.83 thorpej uvm_km_suballoc(struct vm_map *map, vaddr_t *vmin /* IN/OUT */,
328 1.83 thorpej vaddr_t *vmax /* OUT */, vsize_t size, int flags, boolean_t fixed,
329 1.83 thorpej struct vm_map_kernel *submap)
330 1.8 mrg {
331 1.8 mrg int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
332 1.1 mrg
333 1.71 yamt KASSERT(vm_map_pmap(map) == pmap_kernel());
334 1.71 yamt
335 1.8 mrg size = round_page(size); /* round up to pagesize */
336 1.1 mrg
337 1.8 mrg /*
338 1.8 mrg * first allocate a blank spot in the parent map
339 1.8 mrg */
340 1.8 mrg
341 1.82 christos if (uvm_map(map, vmin, size, NULL, UVM_UNKNOWN_OFFSET, 0,
342 1.8 mrg UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
343 1.43 chs UVM_ADV_RANDOM, mapflags)) != 0) {
344 1.8 mrg panic("uvm_km_suballoc: unable to allocate space in parent map");
345 1.8 mrg }
346 1.8 mrg
347 1.8 mrg /*
348 1.82 christos * set VM bounds (vmin is filled in by uvm_map)
349 1.8 mrg */
350 1.1 mrg
351 1.82 christos *vmax = *vmin + size;
352 1.5 thorpej
353 1.8 mrg /*
354 1.8 mrg * add references to pmap and create or init the submap
355 1.8 mrg */
356 1.1 mrg
357 1.8 mrg pmap_reference(vm_map_pmap(map));
358 1.8 mrg if (submap == NULL) {
359 1.71 yamt submap = malloc(sizeof(*submap), M_VMMAP, M_WAITOK);
360 1.8 mrg if (submap == NULL)
361 1.8 mrg panic("uvm_km_suballoc: unable to create submap");
362 1.8 mrg }
363 1.82 christos uvm_map_setup_kernel(submap, *vmin, *vmax, flags);
364 1.71 yamt submap->vmk_map.pmap = vm_map_pmap(map);
365 1.1 mrg
366 1.8 mrg /*
367 1.8 mrg * now let uvm_map_submap plug in it...
368 1.8 mrg */
369 1.1 mrg
370 1.82 christos if (uvm_map_submap(map, *vmin, *vmax, &submap->vmk_map) != 0)
371 1.8 mrg panic("uvm_km_suballoc: submap allocation failed");
372 1.1 mrg
373 1.71 yamt return(&submap->vmk_map);
374 1.1 mrg }
375 1.1 mrg
376 1.1 mrg /*
377 1.1 mrg * uvm_km_pgremove: remove pages from a kernel uvm_object.
378 1.1 mrg *
379 1.1 mrg * => when you unmap a part of anonymous kernel memory you want to toss
380 1.1 mrg * the pages right away. (this gets called from uvm_unmap_...).
381 1.1 mrg */
382 1.1 mrg
383 1.8 mrg void
384 1.83 thorpej uvm_km_pgremove(vaddr_t startva, vaddr_t endva)
385 1.1 mrg {
386 1.78 yamt struct uvm_object * const uobj = uvm.kernel_object;
387 1.78 yamt const voff_t start = startva - vm_map_min(kernel_map);
388 1.78 yamt const voff_t end = endva - vm_map_min(kernel_map);
389 1.53 chs struct vm_page *pg;
390 1.52 chs voff_t curoff, nextoff;
391 1.53 chs int swpgonlydelta = 0;
392 1.8 mrg UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
393 1.1 mrg
394 1.78 yamt KASSERT(VM_MIN_KERNEL_ADDRESS <= startva);
395 1.78 yamt KASSERT(startva < endva);
396 1.86 yamt KASSERT(endva <= VM_MAX_KERNEL_ADDRESS);
397 1.78 yamt
398 1.40 chs simple_lock(&uobj->vmobjlock);
399 1.3 chs
400 1.52 chs for (curoff = start; curoff < end; curoff = nextoff) {
401 1.52 chs nextoff = curoff + PAGE_SIZE;
402 1.52 chs pg = uvm_pagelookup(uobj, curoff);
403 1.53 chs if (pg != NULL && pg->flags & PG_BUSY) {
404 1.52 chs pg->flags |= PG_WANTED;
405 1.52 chs UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
406 1.52 chs "km_pgrm", 0);
407 1.52 chs simple_lock(&uobj->vmobjlock);
408 1.52 chs nextoff = curoff;
409 1.8 mrg continue;
410 1.52 chs }
411 1.8 mrg
412 1.52 chs /*
413 1.52 chs * free the swap slot, then the page.
414 1.52 chs */
415 1.8 mrg
416 1.53 chs if (pg == NULL &&
417 1.64 pk uao_find_swslot(uobj, curoff >> PAGE_SHIFT) > 0) {
418 1.53 chs swpgonlydelta++;
419 1.53 chs }
420 1.52 chs uao_dropswap(uobj, curoff >> PAGE_SHIFT);
421 1.53 chs if (pg != NULL) {
422 1.53 chs uvm_lock_pageq();
423 1.53 chs uvm_pagefree(pg);
424 1.53 chs uvm_unlock_pageq();
425 1.53 chs }
426 1.8 mrg }
427 1.8 mrg simple_unlock(&uobj->vmobjlock);
428 1.8 mrg
429 1.54 chs if (swpgonlydelta > 0) {
430 1.54 chs simple_lock(&uvm.swap_data_lock);
431 1.54 chs KASSERT(uvmexp.swpgonly >= swpgonlydelta);
432 1.54 chs uvmexp.swpgonly -= swpgonlydelta;
433 1.54 chs simple_unlock(&uvm.swap_data_lock);
434 1.54 chs }
435 1.24 thorpej }
436 1.24 thorpej
437 1.24 thorpej
438 1.24 thorpej /*
439 1.78 yamt * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for non object backed
440 1.78 yamt * regions.
441 1.24 thorpej *
442 1.24 thorpej * => when you unmap a part of anonymous kernel memory you want to toss
443 1.52 chs * the pages right away. (this is called from uvm_unmap_...).
444 1.24 thorpej * => none of the pages will ever be busy, and none of them will ever
445 1.52 chs * be on the active or inactive queues (because they have no object).
446 1.24 thorpej */
447 1.24 thorpej
448 1.24 thorpej void
449 1.83 thorpej uvm_km_pgremove_intrsafe(vaddr_t start, vaddr_t end)
450 1.24 thorpej {
451 1.52 chs struct vm_page *pg;
452 1.52 chs paddr_t pa;
453 1.24 thorpej UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist);
454 1.24 thorpej
455 1.78 yamt KASSERT(VM_MIN_KERNEL_ADDRESS <= start);
456 1.78 yamt KASSERT(start < end);
457 1.86 yamt KASSERT(end <= VM_MAX_KERNEL_ADDRESS);
458 1.78 yamt
459 1.52 chs for (; start < end; start += PAGE_SIZE) {
460 1.52 chs if (!pmap_extract(pmap_kernel(), start, &pa)) {
461 1.24 thorpej continue;
462 1.40 chs }
463 1.52 chs pg = PHYS_TO_VM_PAGE(pa);
464 1.52 chs KASSERT(pg);
465 1.52 chs KASSERT(pg->uobject == NULL && pg->uanon == NULL);
466 1.52 chs uvm_pagefree(pg);
467 1.24 thorpej }
468 1.1 mrg }
469 1.1 mrg
470 1.78 yamt #if defined(DEBUG)
471 1.78 yamt void
472 1.78 yamt uvm_km_check_empty(vaddr_t start, vaddr_t end, boolean_t intrsafe)
473 1.78 yamt {
474 1.78 yamt vaddr_t va;
475 1.78 yamt paddr_t pa;
476 1.78 yamt
477 1.78 yamt KDASSERT(VM_MIN_KERNEL_ADDRESS <= start);
478 1.78 yamt KDASSERT(start < end);
479 1.85 yamt KDASSERT(end <= VM_MAX_KERNEL_ADDRESS);
480 1.78 yamt
481 1.78 yamt for (va = start; va < end; va += PAGE_SIZE) {
482 1.78 yamt if (pmap_extract(pmap_kernel(), va, &pa)) {
483 1.81 simonb panic("uvm_km_check_empty: va %p has pa 0x%llx",
484 1.81 simonb (void *)va, (long long)pa);
485 1.78 yamt }
486 1.78 yamt if (!intrsafe) {
487 1.78 yamt const struct vm_page *pg;
488 1.78 yamt
489 1.78 yamt simple_lock(&uvm.kernel_object->vmobjlock);
490 1.78 yamt pg = uvm_pagelookup(uvm.kernel_object,
491 1.78 yamt va - vm_map_min(kernel_map));
492 1.78 yamt simple_unlock(&uvm.kernel_object->vmobjlock);
493 1.78 yamt if (pg) {
494 1.78 yamt panic("uvm_km_check_empty: "
495 1.78 yamt "has page hashed at %p", (const void *)va);
496 1.78 yamt }
497 1.78 yamt }
498 1.78 yamt }
499 1.78 yamt }
500 1.78 yamt #endif /* defined(DEBUG) */
501 1.1 mrg
502 1.1 mrg /*
503 1.78 yamt * uvm_km_alloc: allocate an area of kernel memory.
504 1.1 mrg *
505 1.78 yamt * => NOTE: we can return 0 even if we can wait if there is not enough
506 1.1 mrg * free VM space in the map... caller should be prepared to handle
507 1.1 mrg * this case.
508 1.1 mrg * => we return KVA of memory allocated
509 1.1 mrg */
510 1.1 mrg
511 1.14 eeh vaddr_t
512 1.83 thorpej uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags)
513 1.1 mrg {
514 1.14 eeh vaddr_t kva, loopva;
515 1.14 eeh vaddr_t offset;
516 1.44 thorpej vsize_t loopsize;
517 1.8 mrg struct vm_page *pg;
518 1.78 yamt struct uvm_object *obj;
519 1.78 yamt int pgaflags;
520 1.78 yamt UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist);
521 1.1 mrg
522 1.40 chs KASSERT(vm_map_pmap(map) == pmap_kernel());
523 1.78 yamt KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
524 1.78 yamt (flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
525 1.78 yamt (flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
526 1.1 mrg
527 1.8 mrg /*
528 1.8 mrg * setup for call
529 1.8 mrg */
530 1.8 mrg
531 1.78 yamt kva = vm_map_min(map); /* hint */
532 1.8 mrg size = round_page(size);
533 1.78 yamt obj = (flags & UVM_KMF_PAGEABLE) ? uvm.kernel_object : NULL;
534 1.78 yamt UVMHIST_LOG(maphist," (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
535 1.78 yamt map, obj, size, flags);
536 1.1 mrg
537 1.8 mrg /*
538 1.8 mrg * allocate some virtual space
539 1.8 mrg */
540 1.8 mrg
541 1.78 yamt if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
542 1.78 yamt align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
543 1.78 yamt UVM_ADV_RANDOM,
544 1.78 yamt (flags & (UVM_KMF_TRYLOCK | UVM_KMF_NOWAIT | UVM_KMF_WAITVA))
545 1.78 yamt | UVM_FLAG_QUANTUM)) != 0)) {
546 1.8 mrg UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
547 1.8 mrg return(0);
548 1.8 mrg }
549 1.8 mrg
550 1.8 mrg /*
551 1.8 mrg * if all we wanted was VA, return now
552 1.8 mrg */
553 1.8 mrg
554 1.78 yamt if (flags & (UVM_KMF_VAONLY | UVM_KMF_PAGEABLE)) {
555 1.8 mrg UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
556 1.8 mrg return(kva);
557 1.8 mrg }
558 1.40 chs
559 1.8 mrg /*
560 1.8 mrg * recover object offset from virtual address
561 1.8 mrg */
562 1.8 mrg
563 1.8 mrg offset = kva - vm_map_min(kernel_map);
564 1.8 mrg UVMHIST_LOG(maphist, " kva=0x%x, offset=0x%x", kva, offset,0,0);
565 1.8 mrg
566 1.8 mrg /*
567 1.8 mrg * now allocate and map in the memory... note that we are the only ones
568 1.8 mrg * whom should ever get a handle on this area of VM.
569 1.8 mrg */
570 1.8 mrg
571 1.8 mrg loopva = kva;
572 1.44 thorpej loopsize = size;
573 1.78 yamt
574 1.78 yamt pgaflags = UVM_PGA_USERESERVE;
575 1.78 yamt if (flags & UVM_KMF_ZERO)
576 1.78 yamt pgaflags |= UVM_PGA_ZERO;
577 1.44 thorpej while (loopsize) {
578 1.78 yamt KASSERT(!pmap_extract(pmap_kernel(), loopva, NULL));
579 1.78 yamt
580 1.78 yamt pg = uvm_pagealloc(NULL, offset, NULL, pgaflags);
581 1.47 chs
582 1.8 mrg /*
583 1.8 mrg * out of memory?
584 1.8 mrg */
585 1.8 mrg
586 1.35 thorpej if (__predict_false(pg == NULL)) {
587 1.58 chs if ((flags & UVM_KMF_NOWAIT) ||
588 1.80 yamt ((flags & UVM_KMF_CANFAIL) && !uvm_reclaimable())) {
589 1.8 mrg /* free everything! */
590 1.78 yamt uvm_km_free(map, kva, size,
591 1.78 yamt flags & UVM_KMF_TYPEMASK);
592 1.58 chs return (0);
593 1.8 mrg } else {
594 1.8 mrg uvm_wait("km_getwait2"); /* sleep here */
595 1.8 mrg continue;
596 1.8 mrg }
597 1.8 mrg }
598 1.47 chs
599 1.78 yamt pg->flags &= ~PG_BUSY; /* new page */
600 1.78 yamt UVM_PAGE_OWN(pg, NULL);
601 1.78 yamt
602 1.8 mrg /*
603 1.52 chs * map it in
604 1.8 mrg */
605 1.40 chs
606 1.78 yamt pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg),
607 1.78 yamt VM_PROT_READ | VM_PROT_WRITE);
608 1.8 mrg loopva += PAGE_SIZE;
609 1.8 mrg offset += PAGE_SIZE;
610 1.44 thorpej loopsize -= PAGE_SIZE;
611 1.8 mrg }
612 1.69 junyoung
613 1.51 chris pmap_update(pmap_kernel());
614 1.69 junyoung
615 1.8 mrg UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
616 1.8 mrg return(kva);
617 1.1 mrg }
618 1.1 mrg
619 1.1 mrg /*
620 1.1 mrg * uvm_km_free: free an area of kernel memory
621 1.1 mrg */
622 1.1 mrg
623 1.8 mrg void
624 1.83 thorpej uvm_km_free(struct vm_map *map, vaddr_t addr, vsize_t size, uvm_flag_t flags)
625 1.8 mrg {
626 1.1 mrg
627 1.78 yamt KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
628 1.78 yamt (flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
629 1.78 yamt (flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
630 1.78 yamt KASSERT((addr & PAGE_MASK) == 0);
631 1.40 chs KASSERT(vm_map_pmap(map) == pmap_kernel());
632 1.1 mrg
633 1.8 mrg size = round_page(size);
634 1.1 mrg
635 1.78 yamt if (flags & UVM_KMF_PAGEABLE) {
636 1.78 yamt uvm_km_pgremove(addr, addr + size);
637 1.78 yamt pmap_remove(pmap_kernel(), addr, addr + size);
638 1.78 yamt } else if (flags & UVM_KMF_WIRED) {
639 1.78 yamt uvm_km_pgremove_intrsafe(addr, addr + size);
640 1.78 yamt pmap_kremove(addr, size);
641 1.8 mrg }
642 1.8 mrg
643 1.78 yamt uvm_unmap1(map, addr, addr + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
644 1.66 pk }
645 1.66 pk
646 1.10 thorpej /* Sanity; must specify both or none. */
647 1.10 thorpej #if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \
648 1.10 thorpej (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE))
649 1.10 thorpej #error Must specify MAP and UNMAP together.
650 1.10 thorpej #endif
651 1.10 thorpej
652 1.10 thorpej /*
653 1.10 thorpej * uvm_km_alloc_poolpage: allocate a page for the pool allocator
654 1.10 thorpej *
655 1.10 thorpej * => if the pmap specifies an alternate mapping method, we use it.
656 1.10 thorpej */
657 1.10 thorpej
658 1.11 thorpej /* ARGSUSED */
659 1.14 eeh vaddr_t
660 1.83 thorpej uvm_km_alloc_poolpage_cache(struct vm_map *map, boolean_t waitok)
661 1.72 yamt {
662 1.72 yamt #if defined(PMAP_MAP_POOLPAGE)
663 1.78 yamt return uvm_km_alloc_poolpage(map, waitok);
664 1.72 yamt #else
665 1.72 yamt struct vm_page *pg;
666 1.72 yamt struct pool *pp = &vm_map_to_kernel(map)->vmk_vacache;
667 1.72 yamt vaddr_t va;
668 1.72 yamt int s = 0xdeadbeaf; /* XXX: gcc */
669 1.72 yamt const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
670 1.72 yamt
671 1.72 yamt if ((map->flags & VM_MAP_VACACHE) == 0)
672 1.78 yamt return uvm_km_alloc_poolpage(map, waitok);
673 1.72 yamt
674 1.72 yamt if (intrsafe)
675 1.72 yamt s = splvm();
676 1.72 yamt va = (vaddr_t)pool_get(pp, waitok ? PR_WAITOK : PR_NOWAIT);
677 1.72 yamt if (intrsafe)
678 1.72 yamt splx(s);
679 1.72 yamt if (va == 0)
680 1.72 yamt return 0;
681 1.72 yamt KASSERT(!pmap_extract(pmap_kernel(), va, NULL));
682 1.72 yamt again:
683 1.72 yamt pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
684 1.72 yamt if (__predict_false(pg == NULL)) {
685 1.72 yamt if (waitok) {
686 1.72 yamt uvm_wait("plpg");
687 1.72 yamt goto again;
688 1.72 yamt } else {
689 1.72 yamt if (intrsafe)
690 1.72 yamt s = splvm();
691 1.72 yamt pool_put(pp, (void *)va);
692 1.72 yamt if (intrsafe)
693 1.72 yamt splx(s);
694 1.72 yamt return 0;
695 1.72 yamt }
696 1.72 yamt }
697 1.79 yamt pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE);
698 1.72 yamt pmap_update(pmap_kernel());
699 1.72 yamt
700 1.72 yamt return va;
701 1.72 yamt #endif /* PMAP_MAP_POOLPAGE */
702 1.72 yamt }
703 1.72 yamt
704 1.72 yamt vaddr_t
705 1.83 thorpej uvm_km_alloc_poolpage(struct vm_map *map, boolean_t waitok)
706 1.10 thorpej {
707 1.10 thorpej #if defined(PMAP_MAP_POOLPAGE)
708 1.10 thorpej struct vm_page *pg;
709 1.14 eeh vaddr_t va;
710 1.10 thorpej
711 1.15 thorpej again:
712 1.29 chs pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_USERESERVE);
713 1.35 thorpej if (__predict_false(pg == NULL)) {
714 1.15 thorpej if (waitok) {
715 1.15 thorpej uvm_wait("plpg");
716 1.15 thorpej goto again;
717 1.15 thorpej } else
718 1.15 thorpej return (0);
719 1.15 thorpej }
720 1.10 thorpej va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
721 1.35 thorpej if (__predict_false(va == 0))
722 1.10 thorpej uvm_pagefree(pg);
723 1.10 thorpej return (va);
724 1.10 thorpej #else
725 1.14 eeh vaddr_t va;
726 1.72 yamt int s = 0xdeadbeaf; /* XXX: gcc */
727 1.72 yamt const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
728 1.16 thorpej
729 1.72 yamt if (intrsafe)
730 1.72 yamt s = splvm();
731 1.78 yamt va = uvm_km_alloc(map, PAGE_SIZE, 0,
732 1.78 yamt (waitok ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK) | UVM_KMF_WIRED);
733 1.72 yamt if (intrsafe)
734 1.72 yamt splx(s);
735 1.10 thorpej return (va);
736 1.10 thorpej #endif /* PMAP_MAP_POOLPAGE */
737 1.10 thorpej }
738 1.10 thorpej
739 1.10 thorpej /*
740 1.10 thorpej * uvm_km_free_poolpage: free a previously allocated pool page
741 1.10 thorpej *
742 1.10 thorpej * => if the pmap specifies an alternate unmapping method, we use it.
743 1.10 thorpej */
744 1.10 thorpej
745 1.11 thorpej /* ARGSUSED */
746 1.10 thorpej void
747 1.83 thorpej uvm_km_free_poolpage_cache(struct vm_map *map, vaddr_t addr)
748 1.72 yamt {
749 1.72 yamt #if defined(PMAP_UNMAP_POOLPAGE)
750 1.78 yamt uvm_km_free_poolpage(map, addr);
751 1.72 yamt #else
752 1.72 yamt struct pool *pp;
753 1.72 yamt int s = 0xdeadbeaf; /* XXX: gcc */
754 1.72 yamt const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
755 1.72 yamt
756 1.72 yamt if ((map->flags & VM_MAP_VACACHE) == 0) {
757 1.78 yamt uvm_km_free_poolpage(map, addr);
758 1.72 yamt return;
759 1.72 yamt }
760 1.72 yamt
761 1.72 yamt KASSERT(pmap_extract(pmap_kernel(), addr, NULL));
762 1.72 yamt uvm_km_pgremove_intrsafe(addr, addr + PAGE_SIZE);
763 1.72 yamt pmap_kremove(addr, PAGE_SIZE);
764 1.72 yamt #if defined(DEBUG)
765 1.72 yamt pmap_update(pmap_kernel());
766 1.72 yamt #endif
767 1.72 yamt KASSERT(!pmap_extract(pmap_kernel(), addr, NULL));
768 1.72 yamt pp = &vm_map_to_kernel(map)->vmk_vacache;
769 1.72 yamt if (intrsafe)
770 1.72 yamt s = splvm();
771 1.72 yamt pool_put(pp, (void *)addr);
772 1.72 yamt if (intrsafe)
773 1.72 yamt splx(s);
774 1.72 yamt #endif
775 1.72 yamt }
776 1.72 yamt
777 1.72 yamt /* ARGSUSED */
778 1.72 yamt void
779 1.83 thorpej uvm_km_free_poolpage(struct vm_map *map, vaddr_t addr)
780 1.10 thorpej {
781 1.10 thorpej #if defined(PMAP_UNMAP_POOLPAGE)
782 1.14 eeh paddr_t pa;
783 1.10 thorpej
784 1.10 thorpej pa = PMAP_UNMAP_POOLPAGE(addr);
785 1.10 thorpej uvm_pagefree(PHYS_TO_VM_PAGE(pa));
786 1.10 thorpej #else
787 1.72 yamt int s = 0xdeadbeaf; /* XXX: gcc */
788 1.72 yamt const boolean_t intrsafe = (map->flags & VM_MAP_INTRSAFE) != 0;
789 1.10 thorpej
790 1.72 yamt if (intrsafe)
791 1.72 yamt s = splvm();
792 1.78 yamt uvm_km_free(map, addr, PAGE_SIZE, UVM_KMF_WIRED);
793 1.72 yamt if (intrsafe)
794 1.72 yamt splx(s);
795 1.10 thorpej #endif /* PMAP_UNMAP_POOLPAGE */
796 1.1 mrg }
797