uvm_km.c revision 1.107 1 1.107 matt /* $NetBSD: uvm_km.c,v 1.107 2011/01/04 08:26:33 matt 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 1.97 ad * malloc.
94 1.7 chuck * pager_map => used to map "buf" structures into kernel space
95 1.7 chuck * exec_map => used during exec to handle exec args
96 1.7 chuck * etc...
97 1.7 chuck *
98 1.7 chuck * the kernel allocates its private memory out of special uvm_objects whose
99 1.7 chuck * reference count is set to UVM_OBJ_KERN (thus indicating that the objects
100 1.7 chuck * are "special" and never die). all kernel objects should be thought of
101 1.47 chs * as large, fixed-sized, sparsely populated uvm_objects. each kernel
102 1.62 thorpej * object is equal to the size of kernel virtual address space (i.e. the
103 1.62 thorpej * value "VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS").
104 1.7 chuck *
105 1.101 pooka * note that just because a kernel object spans the entire kernel virtual
106 1.7 chuck * address space doesn't mean that it has to be mapped into the entire space.
107 1.47 chs * large chunks of a kernel object's space go unused either because
108 1.47 chs * that area of kernel VM is unmapped, or there is some other type of
109 1.7 chuck * object mapped into that range (e.g. a vnode). for submap's kernel
110 1.7 chuck * objects, the only part of the object that can ever be populated is the
111 1.7 chuck * offsets that are managed by the submap.
112 1.7 chuck *
113 1.7 chuck * note that the "offset" in a kernel object is always the kernel virtual
114 1.62 thorpej * address minus the VM_MIN_KERNEL_ADDRESS (aka vm_map_min(kernel_map)).
115 1.7 chuck * example:
116 1.62 thorpej * suppose VM_MIN_KERNEL_ADDRESS is 0xf8000000 and the kernel does a
117 1.7 chuck * uvm_km_alloc(kernel_map, PAGE_SIZE) [allocate 1 wired down page in the
118 1.7 chuck * kernel map]. if uvm_km_alloc returns virtual address 0xf8235000,
119 1.7 chuck * then that means that the page at offset 0x235000 in kernel_object is
120 1.47 chs * mapped at 0xf8235000.
121 1.7 chuck *
122 1.7 chuck * kernel object have one other special property: when the kernel virtual
123 1.7 chuck * memory mapping them is unmapped, the backing memory in the object is
124 1.7 chuck * freed right away. this is done with the uvm_km_pgremove() function.
125 1.7 chuck * this has to be done because there is no backing store for kernel pages
126 1.7 chuck * and no need to save them after they are no longer referenced.
127 1.7 chuck */
128 1.55 lukem
129 1.55 lukem #include <sys/cdefs.h>
130 1.107 matt __KERNEL_RCSID(0, "$NetBSD: uvm_km.c,v 1.107 2011/01/04 08:26:33 matt Exp $");
131 1.55 lukem
132 1.55 lukem #include "opt_uvmhist.h"
133 1.7 chuck
134 1.1 mrg #include <sys/param.h>
135 1.71 yamt #include <sys/malloc.h>
136 1.1 mrg #include <sys/systm.h>
137 1.1 mrg #include <sys/proc.h>
138 1.72 yamt #include <sys/pool.h>
139 1.1 mrg
140 1.1 mrg #include <uvm/uvm.h>
141 1.1 mrg
142 1.1 mrg /*
143 1.1 mrg * global data structures
144 1.1 mrg */
145 1.1 mrg
146 1.49 chs struct vm_map *kernel_map = NULL;
147 1.1 mrg
148 1.1 mrg /*
149 1.1 mrg * local data structues
150 1.1 mrg */
151 1.1 mrg
152 1.71 yamt static struct vm_map_kernel kernel_map_store;
153 1.70 yamt static struct vm_map_entry kernel_first_mapent_store;
154 1.1 mrg
155 1.72 yamt #if !defined(PMAP_MAP_POOLPAGE)
156 1.72 yamt
157 1.72 yamt /*
158 1.72 yamt * kva cache
159 1.72 yamt *
160 1.72 yamt * XXX maybe it's better to do this at the uvm_map layer.
161 1.72 yamt */
162 1.72 yamt
163 1.72 yamt #define KM_VACACHE_SIZE (32 * PAGE_SIZE) /* XXX tune */
164 1.72 yamt
165 1.72 yamt static void *km_vacache_alloc(struct pool *, int);
166 1.72 yamt static void km_vacache_free(struct pool *, void *);
167 1.72 yamt static void km_vacache_init(struct vm_map *, const char *, size_t);
168 1.72 yamt
169 1.72 yamt /* XXX */
170 1.72 yamt #define KM_VACACHE_POOL_TO_MAP(pp) \
171 1.72 yamt ((struct vm_map *)((char *)(pp) - \
172 1.72 yamt offsetof(struct vm_map_kernel, vmk_vacache)))
173 1.72 yamt
174 1.72 yamt static void *
175 1.72 yamt km_vacache_alloc(struct pool *pp, int flags)
176 1.72 yamt {
177 1.72 yamt vaddr_t va;
178 1.72 yamt size_t size;
179 1.72 yamt struct vm_map *map;
180 1.72 yamt size = pp->pr_alloc->pa_pagesz;
181 1.72 yamt
182 1.72 yamt map = KM_VACACHE_POOL_TO_MAP(pp);
183 1.72 yamt
184 1.73 yamt va = vm_map_min(map); /* hint */
185 1.72 yamt if (uvm_map(map, &va, size, NULL, UVM_UNKNOWN_OFFSET, size,
186 1.74 yamt UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
187 1.72 yamt UVM_ADV_RANDOM, UVM_FLAG_QUANTUM |
188 1.88 yamt ((flags & PR_WAITOK) ? UVM_FLAG_WAITVA :
189 1.88 yamt UVM_FLAG_TRYLOCK | UVM_FLAG_NOWAIT))))
190 1.72 yamt return NULL;
191 1.72 yamt
192 1.72 yamt return (void *)va;
193 1.72 yamt }
194 1.72 yamt
195 1.72 yamt static void
196 1.72 yamt km_vacache_free(struct pool *pp, void *v)
197 1.72 yamt {
198 1.72 yamt vaddr_t va = (vaddr_t)v;
199 1.72 yamt size_t size = pp->pr_alloc->pa_pagesz;
200 1.72 yamt struct vm_map *map;
201 1.72 yamt
202 1.72 yamt map = KM_VACACHE_POOL_TO_MAP(pp);
203 1.78 yamt uvm_unmap1(map, va, va + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
204 1.72 yamt }
205 1.72 yamt
206 1.72 yamt /*
207 1.72 yamt * km_vacache_init: initialize kva cache.
208 1.72 yamt */
209 1.72 yamt
210 1.72 yamt static void
211 1.72 yamt km_vacache_init(struct vm_map *map, const char *name, size_t size)
212 1.72 yamt {
213 1.72 yamt struct vm_map_kernel *vmk;
214 1.72 yamt struct pool *pp;
215 1.72 yamt struct pool_allocator *pa;
216 1.94 ad int ipl;
217 1.72 yamt
218 1.72 yamt KASSERT(VM_MAP_IS_KERNEL(map));
219 1.72 yamt KASSERT(size < (vm_map_max(map) - vm_map_min(map)) / 2); /* sanity */
220 1.72 yamt
221 1.97 ad
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 1.88 yamt pa->pa_backingmap = map;
230 1.88 yamt pa->pa_backingmapptr = NULL;
231 1.94 ad
232 1.94 ad if ((map->flags & VM_MAP_INTRSAFE) != 0)
233 1.94 ad ipl = IPL_VM;
234 1.94 ad else
235 1.94 ad ipl = IPL_NONE;
236 1.94 ad
237 1.94 ad pool_init(pp, PAGE_SIZE, 0, 0, PR_NOTOUCH | PR_RECURSIVE, name, pa,
238 1.94 ad ipl);
239 1.72 yamt }
240 1.72 yamt
241 1.72 yamt void
242 1.72 yamt uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
243 1.72 yamt {
244 1.72 yamt
245 1.72 yamt map->flags |= VM_MAP_VACACHE;
246 1.72 yamt if (size == 0)
247 1.72 yamt size = KM_VACACHE_SIZE;
248 1.72 yamt km_vacache_init(map, name, size);
249 1.72 yamt }
250 1.72 yamt
251 1.72 yamt #else /* !defined(PMAP_MAP_POOLPAGE) */
252 1.72 yamt
253 1.72 yamt void
254 1.92 yamt uvm_km_vacache_init(struct vm_map *map, const char *name, size_t size)
255 1.72 yamt {
256 1.72 yamt
257 1.72 yamt /* nothing */
258 1.72 yamt }
259 1.72 yamt
260 1.72 yamt #endif /* !defined(PMAP_MAP_POOLPAGE) */
261 1.72 yamt
262 1.88 yamt void
263 1.92 yamt uvm_km_va_drain(struct vm_map *map, uvm_flag_t flags)
264 1.88 yamt {
265 1.88 yamt struct vm_map_kernel *vmk = vm_map_to_kernel(map);
266 1.88 yamt
267 1.88 yamt callback_run_roundrobin(&vmk->vmk_reclaim_callback, NULL);
268 1.88 yamt }
269 1.88 yamt
270 1.1 mrg /*
271 1.1 mrg * uvm_km_init: init kernel maps and objects to reflect reality (i.e.
272 1.1 mrg * KVM already allocated for text, data, bss, and static data structures).
273 1.1 mrg *
274 1.62 thorpej * => KVM is defined by VM_MIN_KERNEL_ADDRESS/VM_MAX_KERNEL_ADDRESS.
275 1.82 christos * we assume that [vmin -> start] has already been allocated and that
276 1.62 thorpej * "end" is the end.
277 1.1 mrg */
278 1.1 mrg
279 1.8 mrg void
280 1.83 thorpej uvm_km_init(vaddr_t start, vaddr_t end)
281 1.1 mrg {
282 1.62 thorpej vaddr_t base = VM_MIN_KERNEL_ADDRESS;
283 1.27 thorpej
284 1.27 thorpej /*
285 1.27 thorpej * next, init kernel memory objects.
286 1.8 mrg */
287 1.1 mrg
288 1.8 mrg /* kernel_object: for pageable anonymous kernel memory */
289 1.34 chs uao_init();
290 1.95 ad uvm_kernel_object = uao_create(VM_MAX_KERNEL_ADDRESS -
291 1.62 thorpej VM_MIN_KERNEL_ADDRESS, UAO_FLAG_KERNOBJ);
292 1.1 mrg
293 1.24 thorpej /*
294 1.56 thorpej * init the map and reserve any space that might already
295 1.56 thorpej * have been allocated kernel space before installing.
296 1.8 mrg */
297 1.1 mrg
298 1.71 yamt uvm_map_setup_kernel(&kernel_map_store, base, end, VM_MAP_PAGEABLE);
299 1.71 yamt kernel_map_store.vmk_map.pmap = pmap_kernel();
300 1.70 yamt if (start != base) {
301 1.70 yamt int error;
302 1.70 yamt struct uvm_map_args args;
303 1.70 yamt
304 1.71 yamt error = uvm_map_prepare(&kernel_map_store.vmk_map,
305 1.71 yamt base, start - base,
306 1.70 yamt NULL, UVM_UNKNOWN_OFFSET, 0,
307 1.62 thorpej UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
308 1.70 yamt UVM_ADV_RANDOM, UVM_FLAG_FIXED), &args);
309 1.70 yamt if (!error) {
310 1.70 yamt kernel_first_mapent_store.flags =
311 1.70 yamt UVM_MAP_KERNEL | UVM_MAP_FIRST;
312 1.71 yamt error = uvm_map_enter(&kernel_map_store.vmk_map, &args,
313 1.70 yamt &kernel_first_mapent_store);
314 1.70 yamt }
315 1.70 yamt
316 1.70 yamt if (error)
317 1.70 yamt panic(
318 1.70 yamt "uvm_km_init: could not reserve space for kernel");
319 1.70 yamt }
320 1.47 chs
321 1.8 mrg /*
322 1.8 mrg * install!
323 1.8 mrg */
324 1.8 mrg
325 1.71 yamt kernel_map = &kernel_map_store.vmk_map;
326 1.72 yamt uvm_km_vacache_init(kernel_map, "kvakernel", 0);
327 1.1 mrg }
328 1.1 mrg
329 1.1 mrg /*
330 1.1 mrg * uvm_km_suballoc: allocate a submap in the kernel map. once a submap
331 1.1 mrg * is allocated all references to that area of VM must go through it. this
332 1.1 mrg * allows the locking of VAs in kernel_map to be broken up into regions.
333 1.1 mrg *
334 1.82 christos * => if `fixed' is true, *vmin specifies where the region described
335 1.5 thorpej * by the submap must start
336 1.1 mrg * => if submap is non NULL we use that as the submap, otherwise we
337 1.1 mrg * alloc a new map
338 1.1 mrg */
339 1.78 yamt
340 1.8 mrg struct vm_map *
341 1.83 thorpej uvm_km_suballoc(struct vm_map *map, vaddr_t *vmin /* IN/OUT */,
342 1.93 thorpej vaddr_t *vmax /* OUT */, vsize_t size, int flags, bool fixed,
343 1.83 thorpej struct vm_map_kernel *submap)
344 1.8 mrg {
345 1.8 mrg int mapflags = UVM_FLAG_NOMERGE | (fixed ? UVM_FLAG_FIXED : 0);
346 1.1 mrg
347 1.71 yamt KASSERT(vm_map_pmap(map) == pmap_kernel());
348 1.71 yamt
349 1.8 mrg size = round_page(size); /* round up to pagesize */
350 1.87 yamt size += uvm_mapent_overhead(size, flags);
351 1.1 mrg
352 1.8 mrg /*
353 1.8 mrg * first allocate a blank spot in the parent map
354 1.8 mrg */
355 1.8 mrg
356 1.82 christos if (uvm_map(map, vmin, size, NULL, UVM_UNKNOWN_OFFSET, 0,
357 1.8 mrg UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
358 1.43 chs UVM_ADV_RANDOM, mapflags)) != 0) {
359 1.8 mrg panic("uvm_km_suballoc: unable to allocate space in parent map");
360 1.8 mrg }
361 1.8 mrg
362 1.8 mrg /*
363 1.82 christos * set VM bounds (vmin is filled in by uvm_map)
364 1.8 mrg */
365 1.1 mrg
366 1.82 christos *vmax = *vmin + size;
367 1.5 thorpej
368 1.8 mrg /*
369 1.8 mrg * add references to pmap and create or init the submap
370 1.8 mrg */
371 1.1 mrg
372 1.8 mrg pmap_reference(vm_map_pmap(map));
373 1.8 mrg if (submap == NULL) {
374 1.71 yamt submap = malloc(sizeof(*submap), M_VMMAP, M_WAITOK);
375 1.8 mrg if (submap == NULL)
376 1.8 mrg panic("uvm_km_suballoc: unable to create submap");
377 1.8 mrg }
378 1.82 christos uvm_map_setup_kernel(submap, *vmin, *vmax, flags);
379 1.71 yamt submap->vmk_map.pmap = vm_map_pmap(map);
380 1.1 mrg
381 1.8 mrg /*
382 1.8 mrg * now let uvm_map_submap plug in it...
383 1.8 mrg */
384 1.1 mrg
385 1.82 christos if (uvm_map_submap(map, *vmin, *vmax, &submap->vmk_map) != 0)
386 1.8 mrg panic("uvm_km_suballoc: submap allocation failed");
387 1.1 mrg
388 1.71 yamt return(&submap->vmk_map);
389 1.1 mrg }
390 1.1 mrg
391 1.1 mrg /*
392 1.1 mrg * uvm_km_pgremove: remove pages from a kernel uvm_object.
393 1.1 mrg *
394 1.1 mrg * => when you unmap a part of anonymous kernel memory you want to toss
395 1.1 mrg * the pages right away. (this gets called from uvm_unmap_...).
396 1.1 mrg */
397 1.1 mrg
398 1.8 mrg void
399 1.83 thorpej uvm_km_pgremove(vaddr_t startva, vaddr_t endva)
400 1.1 mrg {
401 1.95 ad struct uvm_object * const uobj = uvm_kernel_object;
402 1.78 yamt const voff_t start = startva - vm_map_min(kernel_map);
403 1.78 yamt const voff_t end = endva - vm_map_min(kernel_map);
404 1.53 chs struct vm_page *pg;
405 1.52 chs voff_t curoff, nextoff;
406 1.53 chs int swpgonlydelta = 0;
407 1.8 mrg UVMHIST_FUNC("uvm_km_pgremove"); UVMHIST_CALLED(maphist);
408 1.1 mrg
409 1.78 yamt KASSERT(VM_MIN_KERNEL_ADDRESS <= startva);
410 1.78 yamt KASSERT(startva < endva);
411 1.86 yamt KASSERT(endva <= VM_MAX_KERNEL_ADDRESS);
412 1.78 yamt
413 1.97 ad mutex_enter(&uobj->vmobjlock);
414 1.3 chs
415 1.52 chs for (curoff = start; curoff < end; curoff = nextoff) {
416 1.52 chs nextoff = curoff + PAGE_SIZE;
417 1.52 chs pg = uvm_pagelookup(uobj, curoff);
418 1.53 chs if (pg != NULL && pg->flags & PG_BUSY) {
419 1.52 chs pg->flags |= PG_WANTED;
420 1.52 chs UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
421 1.52 chs "km_pgrm", 0);
422 1.97 ad mutex_enter(&uobj->vmobjlock);
423 1.52 chs nextoff = curoff;
424 1.8 mrg continue;
425 1.52 chs }
426 1.8 mrg
427 1.52 chs /*
428 1.52 chs * free the swap slot, then the page.
429 1.52 chs */
430 1.8 mrg
431 1.53 chs if (pg == NULL &&
432 1.64 pk uao_find_swslot(uobj, curoff >> PAGE_SHIFT) > 0) {
433 1.53 chs swpgonlydelta++;
434 1.53 chs }
435 1.52 chs uao_dropswap(uobj, curoff >> PAGE_SHIFT);
436 1.53 chs if (pg != NULL) {
437 1.97 ad mutex_enter(&uvm_pageqlock);
438 1.53 chs uvm_pagefree(pg);
439 1.97 ad mutex_exit(&uvm_pageqlock);
440 1.53 chs }
441 1.8 mrg }
442 1.97 ad mutex_exit(&uobj->vmobjlock);
443 1.8 mrg
444 1.54 chs if (swpgonlydelta > 0) {
445 1.95 ad mutex_enter(&uvm_swap_data_lock);
446 1.54 chs KASSERT(uvmexp.swpgonly >= swpgonlydelta);
447 1.54 chs uvmexp.swpgonly -= swpgonlydelta;
448 1.95 ad mutex_exit(&uvm_swap_data_lock);
449 1.54 chs }
450 1.24 thorpej }
451 1.24 thorpej
452 1.24 thorpej
453 1.24 thorpej /*
454 1.78 yamt * uvm_km_pgremove_intrsafe: like uvm_km_pgremove(), but for non object backed
455 1.78 yamt * regions.
456 1.24 thorpej *
457 1.24 thorpej * => when you unmap a part of anonymous kernel memory you want to toss
458 1.52 chs * the pages right away. (this is called from uvm_unmap_...).
459 1.24 thorpej * => none of the pages will ever be busy, and none of them will ever
460 1.52 chs * be on the active or inactive queues (because they have no object).
461 1.24 thorpej */
462 1.24 thorpej
463 1.24 thorpej void
464 1.102 ad uvm_km_pgremove_intrsafe(struct vm_map *map, vaddr_t start, vaddr_t end)
465 1.24 thorpej {
466 1.52 chs struct vm_page *pg;
467 1.52 chs paddr_t pa;
468 1.24 thorpej UVMHIST_FUNC("uvm_km_pgremove_intrsafe"); UVMHIST_CALLED(maphist);
469 1.24 thorpej
470 1.102 ad KASSERT(VM_MAP_IS_KERNEL(map));
471 1.102 ad KASSERT(vm_map_min(map) <= start);
472 1.78 yamt KASSERT(start < end);
473 1.102 ad KASSERT(end <= vm_map_max(map));
474 1.78 yamt
475 1.52 chs for (; start < end; start += PAGE_SIZE) {
476 1.52 chs if (!pmap_extract(pmap_kernel(), start, &pa)) {
477 1.24 thorpej continue;
478 1.40 chs }
479 1.52 chs pg = PHYS_TO_VM_PAGE(pa);
480 1.52 chs KASSERT(pg);
481 1.52 chs KASSERT(pg->uobject == NULL && pg->uanon == NULL);
482 1.52 chs uvm_pagefree(pg);
483 1.24 thorpej }
484 1.1 mrg }
485 1.1 mrg
486 1.78 yamt #if defined(DEBUG)
487 1.78 yamt void
488 1.102 ad uvm_km_check_empty(struct vm_map *map, vaddr_t start, vaddr_t end)
489 1.78 yamt {
490 1.102 ad struct vm_page *pg;
491 1.78 yamt vaddr_t va;
492 1.78 yamt paddr_t pa;
493 1.78 yamt
494 1.102 ad KDASSERT(VM_MAP_IS_KERNEL(map));
495 1.102 ad KDASSERT(vm_map_min(map) <= start);
496 1.78 yamt KDASSERT(start < end);
497 1.102 ad KDASSERT(end <= vm_map_max(map));
498 1.78 yamt
499 1.78 yamt for (va = start; va < end; va += PAGE_SIZE) {
500 1.78 yamt if (pmap_extract(pmap_kernel(), va, &pa)) {
501 1.81 simonb panic("uvm_km_check_empty: va %p has pa 0x%llx",
502 1.81 simonb (void *)va, (long long)pa);
503 1.78 yamt }
504 1.102 ad if ((map->flags & VM_MAP_INTRSAFE) == 0) {
505 1.97 ad mutex_enter(&uvm_kernel_object->vmobjlock);
506 1.96 ad pg = uvm_pagelookup(uvm_kernel_object,
507 1.78 yamt va - vm_map_min(kernel_map));
508 1.97 ad mutex_exit(&uvm_kernel_object->vmobjlock);
509 1.78 yamt if (pg) {
510 1.78 yamt panic("uvm_km_check_empty: "
511 1.78 yamt "has page hashed at %p", (const void *)va);
512 1.78 yamt }
513 1.78 yamt }
514 1.78 yamt }
515 1.78 yamt }
516 1.78 yamt #endif /* defined(DEBUG) */
517 1.1 mrg
518 1.1 mrg /*
519 1.78 yamt * uvm_km_alloc: allocate an area of kernel memory.
520 1.1 mrg *
521 1.78 yamt * => NOTE: we can return 0 even if we can wait if there is not enough
522 1.1 mrg * free VM space in the map... caller should be prepared to handle
523 1.1 mrg * this case.
524 1.1 mrg * => we return KVA of memory allocated
525 1.1 mrg */
526 1.1 mrg
527 1.14 eeh vaddr_t
528 1.83 thorpej uvm_km_alloc(struct vm_map *map, vsize_t size, vsize_t align, uvm_flag_t flags)
529 1.1 mrg {
530 1.14 eeh vaddr_t kva, loopva;
531 1.14 eeh vaddr_t offset;
532 1.44 thorpej vsize_t loopsize;
533 1.8 mrg struct vm_page *pg;
534 1.78 yamt struct uvm_object *obj;
535 1.78 yamt int pgaflags;
536 1.89 drochner vm_prot_t prot;
537 1.78 yamt UVMHIST_FUNC(__func__); UVMHIST_CALLED(maphist);
538 1.1 mrg
539 1.40 chs KASSERT(vm_map_pmap(map) == pmap_kernel());
540 1.78 yamt KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
541 1.78 yamt (flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
542 1.78 yamt (flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
543 1.1 mrg
544 1.8 mrg /*
545 1.8 mrg * setup for call
546 1.8 mrg */
547 1.8 mrg
548 1.78 yamt kva = vm_map_min(map); /* hint */
549 1.8 mrg size = round_page(size);
550 1.95 ad obj = (flags & UVM_KMF_PAGEABLE) ? uvm_kernel_object : NULL;
551 1.78 yamt UVMHIST_LOG(maphist," (map=0x%x, obj=0x%x, size=0x%x, flags=%d)",
552 1.78 yamt map, obj, size, flags);
553 1.1 mrg
554 1.8 mrg /*
555 1.8 mrg * allocate some virtual space
556 1.8 mrg */
557 1.8 mrg
558 1.78 yamt if (__predict_false(uvm_map(map, &kva, size, obj, UVM_UNKNOWN_OFFSET,
559 1.78 yamt align, UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL, UVM_INH_NONE,
560 1.78 yamt UVM_ADV_RANDOM,
561 1.78 yamt (flags & (UVM_KMF_TRYLOCK | UVM_KMF_NOWAIT | UVM_KMF_WAITVA))
562 1.78 yamt | UVM_FLAG_QUANTUM)) != 0)) {
563 1.8 mrg UVMHIST_LOG(maphist, "<- done (no VM)",0,0,0,0);
564 1.8 mrg return(0);
565 1.8 mrg }
566 1.8 mrg
567 1.8 mrg /*
568 1.8 mrg * if all we wanted was VA, return now
569 1.8 mrg */
570 1.8 mrg
571 1.78 yamt if (flags & (UVM_KMF_VAONLY | UVM_KMF_PAGEABLE)) {
572 1.8 mrg UVMHIST_LOG(maphist,"<- done valloc (kva=0x%x)", kva,0,0,0);
573 1.8 mrg return(kva);
574 1.8 mrg }
575 1.40 chs
576 1.8 mrg /*
577 1.8 mrg * recover object offset from virtual address
578 1.8 mrg */
579 1.8 mrg
580 1.8 mrg offset = kva - vm_map_min(kernel_map);
581 1.8 mrg UVMHIST_LOG(maphist, " kva=0x%x, offset=0x%x", kva, offset,0,0);
582 1.8 mrg
583 1.8 mrg /*
584 1.8 mrg * now allocate and map in the memory... note that we are the only ones
585 1.8 mrg * whom should ever get a handle on this area of VM.
586 1.8 mrg */
587 1.8 mrg
588 1.8 mrg loopva = kva;
589 1.44 thorpej loopsize = size;
590 1.78 yamt
591 1.107 matt pgaflags = UVM_FLAG_COLORMATCH;
592 1.103 ad if (flags & UVM_KMF_NOWAIT)
593 1.103 ad pgaflags |= UVM_PGA_USERESERVE;
594 1.78 yamt if (flags & UVM_KMF_ZERO)
595 1.78 yamt pgaflags |= UVM_PGA_ZERO;
596 1.89 drochner prot = VM_PROT_READ | VM_PROT_WRITE;
597 1.89 drochner if (flags & UVM_KMF_EXEC)
598 1.89 drochner prot |= VM_PROT_EXECUTE;
599 1.44 thorpej while (loopsize) {
600 1.78 yamt KASSERT(!pmap_extract(pmap_kernel(), loopva, NULL));
601 1.78 yamt
602 1.107 matt pg = uvm_pagealloc_strat(NULL, offset, NULL, pgaflags,
603 1.107 matt #ifdef UVM_KM_VMFREELIST
604 1.107 matt UVM_PGA_STRAT_ONLY, UVM_KM_VMFREELIST
605 1.107 matt #else
606 1.107 matt UVM_PGA_STRAT_NORMAL, 0
607 1.107 matt #endif
608 1.107 matt );
609 1.47 chs
610 1.8 mrg /*
611 1.8 mrg * out of memory?
612 1.8 mrg */
613 1.8 mrg
614 1.35 thorpej if (__predict_false(pg == NULL)) {
615 1.58 chs if ((flags & UVM_KMF_NOWAIT) ||
616 1.80 yamt ((flags & UVM_KMF_CANFAIL) && !uvm_reclaimable())) {
617 1.8 mrg /* free everything! */
618 1.78 yamt uvm_km_free(map, kva, size,
619 1.78 yamt flags & UVM_KMF_TYPEMASK);
620 1.58 chs return (0);
621 1.8 mrg } else {
622 1.8 mrg uvm_wait("km_getwait2"); /* sleep here */
623 1.8 mrg continue;
624 1.8 mrg }
625 1.8 mrg }
626 1.47 chs
627 1.78 yamt pg->flags &= ~PG_BUSY; /* new page */
628 1.78 yamt UVM_PAGE_OWN(pg, NULL);
629 1.78 yamt
630 1.8 mrg /*
631 1.52 chs * map it in
632 1.8 mrg */
633 1.40 chs
634 1.104 cegger pmap_kenter_pa(loopva, VM_PAGE_TO_PHYS(pg),
635 1.106 cegger prot, PMAP_KMPAGE);
636 1.8 mrg loopva += PAGE_SIZE;
637 1.8 mrg offset += PAGE_SIZE;
638 1.44 thorpej loopsize -= PAGE_SIZE;
639 1.8 mrg }
640 1.69 junyoung
641 1.51 chris pmap_update(pmap_kernel());
642 1.69 junyoung
643 1.8 mrg UVMHIST_LOG(maphist,"<- done (kva=0x%x)", kva,0,0,0);
644 1.8 mrg return(kva);
645 1.1 mrg }
646 1.1 mrg
647 1.1 mrg /*
648 1.1 mrg * uvm_km_free: free an area of kernel memory
649 1.1 mrg */
650 1.1 mrg
651 1.8 mrg void
652 1.83 thorpej uvm_km_free(struct vm_map *map, vaddr_t addr, vsize_t size, uvm_flag_t flags)
653 1.8 mrg {
654 1.1 mrg
655 1.78 yamt KASSERT((flags & UVM_KMF_TYPEMASK) == UVM_KMF_WIRED ||
656 1.78 yamt (flags & UVM_KMF_TYPEMASK) == UVM_KMF_PAGEABLE ||
657 1.78 yamt (flags & UVM_KMF_TYPEMASK) == UVM_KMF_VAONLY);
658 1.78 yamt KASSERT((addr & PAGE_MASK) == 0);
659 1.40 chs KASSERT(vm_map_pmap(map) == pmap_kernel());
660 1.1 mrg
661 1.8 mrg size = round_page(size);
662 1.1 mrg
663 1.78 yamt if (flags & UVM_KMF_PAGEABLE) {
664 1.78 yamt uvm_km_pgremove(addr, addr + size);
665 1.78 yamt pmap_remove(pmap_kernel(), addr, addr + size);
666 1.78 yamt } else if (flags & UVM_KMF_WIRED) {
667 1.102 ad uvm_km_pgremove_intrsafe(map, addr, addr + size);
668 1.78 yamt pmap_kremove(addr, size);
669 1.8 mrg }
670 1.99 yamt
671 1.99 yamt /*
672 1.99 yamt * uvm_unmap_remove calls pmap_update for us.
673 1.99 yamt */
674 1.8 mrg
675 1.78 yamt uvm_unmap1(map, addr, addr + size, UVM_FLAG_QUANTUM|UVM_FLAG_VAONLY);
676 1.66 pk }
677 1.66 pk
678 1.10 thorpej /* Sanity; must specify both or none. */
679 1.10 thorpej #if (defined(PMAP_MAP_POOLPAGE) || defined(PMAP_UNMAP_POOLPAGE)) && \
680 1.10 thorpej (!defined(PMAP_MAP_POOLPAGE) || !defined(PMAP_UNMAP_POOLPAGE))
681 1.10 thorpej #error Must specify MAP and UNMAP together.
682 1.10 thorpej #endif
683 1.10 thorpej
684 1.10 thorpej /*
685 1.10 thorpej * uvm_km_alloc_poolpage: allocate a page for the pool allocator
686 1.10 thorpej *
687 1.10 thorpej * => if the pmap specifies an alternate mapping method, we use it.
688 1.10 thorpej */
689 1.10 thorpej
690 1.11 thorpej /* ARGSUSED */
691 1.14 eeh vaddr_t
692 1.93 thorpej uvm_km_alloc_poolpage_cache(struct vm_map *map, bool waitok)
693 1.72 yamt {
694 1.72 yamt #if defined(PMAP_MAP_POOLPAGE)
695 1.78 yamt return uvm_km_alloc_poolpage(map, waitok);
696 1.72 yamt #else
697 1.72 yamt struct vm_page *pg;
698 1.72 yamt struct pool *pp = &vm_map_to_kernel(map)->vmk_vacache;
699 1.72 yamt vaddr_t va;
700 1.72 yamt
701 1.72 yamt if ((map->flags & VM_MAP_VACACHE) == 0)
702 1.78 yamt return uvm_km_alloc_poolpage(map, waitok);
703 1.72 yamt
704 1.72 yamt va = (vaddr_t)pool_get(pp, waitok ? PR_WAITOK : PR_NOWAIT);
705 1.72 yamt if (va == 0)
706 1.72 yamt return 0;
707 1.72 yamt KASSERT(!pmap_extract(pmap_kernel(), va, NULL));
708 1.72 yamt again:
709 1.103 ad pg = uvm_pagealloc(NULL, 0, NULL, waitok ? 0 : UVM_PGA_USERESERVE);
710 1.72 yamt if (__predict_false(pg == NULL)) {
711 1.72 yamt if (waitok) {
712 1.72 yamt uvm_wait("plpg");
713 1.72 yamt goto again;
714 1.72 yamt } else {
715 1.72 yamt pool_put(pp, (void *)va);
716 1.72 yamt return 0;
717 1.72 yamt }
718 1.72 yamt }
719 1.100 matt pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg),
720 1.106 cegger VM_PROT_READ|VM_PROT_WRITE, PMAP_KMPAGE);
721 1.72 yamt pmap_update(pmap_kernel());
722 1.72 yamt
723 1.72 yamt return va;
724 1.72 yamt #endif /* PMAP_MAP_POOLPAGE */
725 1.72 yamt }
726 1.72 yamt
727 1.72 yamt vaddr_t
728 1.93 thorpej uvm_km_alloc_poolpage(struct vm_map *map, bool waitok)
729 1.10 thorpej {
730 1.10 thorpej #if defined(PMAP_MAP_POOLPAGE)
731 1.10 thorpej struct vm_page *pg;
732 1.14 eeh vaddr_t va;
733 1.10 thorpej
734 1.107 matt
735 1.15 thorpej again:
736 1.107 matt #ifdef PMAP_ALLOC_POOLPAGE
737 1.107 matt pg = PMAP_ALLOC_POOLPAGE(waitok ? 0 : UVM_PGA_USERESERVE);
738 1.107 matt #else
739 1.103 ad pg = uvm_pagealloc(NULL, 0, NULL, waitok ? 0 : UVM_PGA_USERESERVE);
740 1.107 matt #endif
741 1.35 thorpej if (__predict_false(pg == NULL)) {
742 1.15 thorpej if (waitok) {
743 1.15 thorpej uvm_wait("plpg");
744 1.15 thorpej goto again;
745 1.15 thorpej } else
746 1.15 thorpej return (0);
747 1.15 thorpej }
748 1.10 thorpej va = PMAP_MAP_POOLPAGE(VM_PAGE_TO_PHYS(pg));
749 1.35 thorpej if (__predict_false(va == 0))
750 1.10 thorpej uvm_pagefree(pg);
751 1.10 thorpej return (va);
752 1.10 thorpej #else
753 1.14 eeh vaddr_t va;
754 1.16 thorpej
755 1.78 yamt va = uvm_km_alloc(map, PAGE_SIZE, 0,
756 1.78 yamt (waitok ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK) | UVM_KMF_WIRED);
757 1.10 thorpej return (va);
758 1.10 thorpej #endif /* PMAP_MAP_POOLPAGE */
759 1.10 thorpej }
760 1.10 thorpej
761 1.10 thorpej /*
762 1.10 thorpej * uvm_km_free_poolpage: free a previously allocated pool page
763 1.10 thorpej *
764 1.10 thorpej * => if the pmap specifies an alternate unmapping method, we use it.
765 1.10 thorpej */
766 1.10 thorpej
767 1.11 thorpej /* ARGSUSED */
768 1.10 thorpej void
769 1.83 thorpej uvm_km_free_poolpage_cache(struct vm_map *map, vaddr_t addr)
770 1.72 yamt {
771 1.72 yamt #if defined(PMAP_UNMAP_POOLPAGE)
772 1.78 yamt uvm_km_free_poolpage(map, addr);
773 1.72 yamt #else
774 1.72 yamt struct pool *pp;
775 1.72 yamt
776 1.72 yamt if ((map->flags & VM_MAP_VACACHE) == 0) {
777 1.78 yamt uvm_km_free_poolpage(map, addr);
778 1.72 yamt return;
779 1.72 yamt }
780 1.72 yamt
781 1.72 yamt KASSERT(pmap_extract(pmap_kernel(), addr, NULL));
782 1.102 ad uvm_km_pgremove_intrsafe(map, addr, addr + PAGE_SIZE);
783 1.72 yamt pmap_kremove(addr, PAGE_SIZE);
784 1.72 yamt #if defined(DEBUG)
785 1.72 yamt pmap_update(pmap_kernel());
786 1.72 yamt #endif
787 1.72 yamt KASSERT(!pmap_extract(pmap_kernel(), addr, NULL));
788 1.72 yamt pp = &vm_map_to_kernel(map)->vmk_vacache;
789 1.72 yamt pool_put(pp, (void *)addr);
790 1.72 yamt #endif
791 1.72 yamt }
792 1.72 yamt
793 1.72 yamt /* ARGSUSED */
794 1.72 yamt void
795 1.83 thorpej uvm_km_free_poolpage(struct vm_map *map, vaddr_t addr)
796 1.10 thorpej {
797 1.10 thorpej #if defined(PMAP_UNMAP_POOLPAGE)
798 1.14 eeh paddr_t pa;
799 1.10 thorpej
800 1.10 thorpej pa = PMAP_UNMAP_POOLPAGE(addr);
801 1.10 thorpej uvm_pagefree(PHYS_TO_VM_PAGE(pa));
802 1.10 thorpej #else
803 1.78 yamt uvm_km_free(map, addr, PAGE_SIZE, UVM_KMF_WIRED);
804 1.10 thorpej #endif /* PMAP_UNMAP_POOLPAGE */
805 1.1 mrg }
806