uvm_pglist.c revision 1.10 1 /* $NetBSD: uvm_pglist.c,v 1.10 2000/05/20 19:54:01 thorpej Exp $ */
2
3 /*-
4 * Copyright (c) 1997 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9 * NASA Ames Research Center.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the NetBSD
22 * Foundation, Inc. and its contributors.
23 * 4. Neither the name of The NetBSD Foundation nor the names of its
24 * contributors may be used to endorse or promote products derived
25 * from this software without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 * POSSIBILITY OF SUCH DAMAGE.
38 */
39
40 /*
41 * uvm_pglist.c: pglist functions
42 */
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/malloc.h>
47 #include <sys/proc.h>
48
49 #include <vm/vm.h>
50
51 #include <uvm/uvm.h>
52
53 #ifdef VM_PAGE_ALLOC_MEMORY_STATS
54 #define STAT_INCR(v) (v)++
55 #define STAT_DECR(v) do { \
56 if ((v) == 0) \
57 printf("%s:%d -- Already 0!\n", __FILE__, __LINE__); \
58 else \
59 (v)--; \
60 } while (0)
61 u_long uvm_pglistalloc_npages;
62 #else
63 #define STAT_INCR(v)
64 #define STAT_DECR(v)
65 #endif
66
67 /*
68 * uvm_pglistalloc: allocate a list of pages
69 *
70 * => allocated pages are placed at the tail of rlist. rlist is
71 * assumed to be properly initialized by caller.
72 * => returns 0 on success or errno on failure
73 * => XXX: implementation allocates only a single segment, also
74 * might be able to better advantage of vm_physeg[].
75 * => doesn't take into account clean non-busy pages on inactive list
76 * that could be used(?)
77 * => params:
78 * size the size of the allocation, rounded to page size.
79 * low the low address of the allowed allocation range.
80 * high the high address of the allowed allocation range.
81 * alignment memory must be aligned to this power-of-two boundary.
82 * boundary no segment in the allocation may cross this
83 * power-of-two boundary (relative to zero).
84 */
85
86 int
87 uvm_pglistalloc(size, low, high, alignment, boundary, rlist, nsegs, waitok)
88 psize_t size;
89 paddr_t low, high, alignment, boundary;
90 struct pglist *rlist;
91 int nsegs, waitok;
92 {
93 paddr_t try, idxpa, lastidxpa;
94 int psi;
95 struct vm_page *pgs;
96 int s, tryidx, idx, pgflidx, end, error, free_list;
97 vm_page_t m;
98 u_long pagemask;
99 #ifdef DEBUG
100 vm_page_t tp;
101 #endif
102
103 #ifdef DIAGNOSTIC
104 if ((alignment & (alignment - 1)) != 0)
105 panic("uvm_pglistalloc: alignment must be power of 2");
106
107 if ((boundary & (boundary - 1)) != 0)
108 panic("uvm_pglistalloc: boundary must be power of 2");
109 #endif
110
111 /*
112 * Our allocations are always page granularity, so our alignment
113 * must be, too.
114 */
115 if (alignment < PAGE_SIZE)
116 alignment = PAGE_SIZE;
117
118 size = round_page(size);
119 try = roundup(low, alignment);
120
121 if (boundary != 0 && boundary < size)
122 return (EINVAL);
123
124 pagemask = ~(boundary - 1);
125
126 /* Default to "lose". */
127 error = ENOMEM;
128
129 /*
130 * Block all memory allocation and lock the free list.
131 */
132 s = uvm_lock_fpageq(); /* lock free page queue */
133
134 /* Are there even any free pages? */
135 if (uvmexp.free <= (uvmexp.reserve_pagedaemon +
136 uvmexp.reserve_kernel))
137 goto out;
138
139 for (;; try += alignment) {
140 if (try + size > high) {
141 /*
142 * We've run past the allowable range.
143 */
144 goto out;
145 }
146
147 /*
148 * Make sure this is a managed physical page.
149 */
150
151 if ((psi = vm_physseg_find(atop(try), &idx)) == -1)
152 continue; /* managed? */
153 if (vm_physseg_find(atop(try + size), NULL) != psi)
154 continue; /* end must be in this segment */
155
156 tryidx = idx;
157 end = idx + (size / PAGE_SIZE);
158 pgs = vm_physmem[psi].pgs;
159
160 /*
161 * Found a suitable starting page. See of the range is free.
162 */
163 for (; idx < end; idx++) {
164 if (VM_PAGE_IS_FREE(&pgs[idx]) == 0) {
165 /*
166 * Page not available.
167 */
168 break;
169 }
170
171 idxpa = VM_PAGE_TO_PHYS(&pgs[idx]);
172
173 if (idx > tryidx) {
174 lastidxpa = VM_PAGE_TO_PHYS(&pgs[idx - 1]);
175
176 if ((lastidxpa + PAGE_SIZE) != idxpa) {
177 /*
178 * Region not contiguous.
179 */
180 break;
181 }
182 if (boundary != 0 &&
183 ((lastidxpa ^ idxpa) & pagemask) != 0) {
184 /*
185 * Region crosses boundary.
186 */
187 break;
188 }
189 }
190 }
191
192 if (idx == end) {
193 /*
194 * Woo hoo! Found one.
195 */
196 break;
197 }
198 }
199
200 #if PGFL_NQUEUES != 2
201 #error uvm_pglistalloc needs to be updated
202 #endif
203
204 /*
205 * we have a chunk of memory that conforms to the requested constraints.
206 */
207 idx = tryidx;
208 while (idx < end) {
209 m = &pgs[idx];
210 free_list = uvm_page_lookup_freelist(m);
211 pgflidx = (m->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN;
212 #ifdef DEBUG
213 for (tp = TAILQ_FIRST(&uvm.page_free[
214 free_list].pgfl_queues[pgflidx]);
215 tp != NULL;
216 tp = TAILQ_NEXT(tp, pageq)) {
217 if (tp == m)
218 break;
219 }
220 if (tp == NULL)
221 panic("uvm_pglistalloc: page not on freelist");
222 #endif
223 TAILQ_REMOVE(&uvm.page_free[free_list].pgfl_queues[pgflidx],
224 m, pageq);
225 uvmexp.free--;
226 if (m->flags & PG_ZERO)
227 uvmexp.zeropages--;
228 m->flags = PG_CLEAN;
229 m->pqflags = 0;
230 m->uobject = NULL;
231 m->uanon = NULL;
232 m->wire_count = 0;
233 m->loan_count = 0;
234 TAILQ_INSERT_TAIL(rlist, m, pageq);
235 idx++;
236 STAT_INCR(uvm_pglistalloc_npages);
237 }
238 error = 0;
239
240 out:
241 uvm_unlock_fpageq(s);
242
243 /*
244 * check to see if we need to generate some free pages waking
245 * the pagedaemon.
246 * XXX: we read uvm.free without locking
247 */
248
249 if (uvmexp.free < uvmexp.freemin ||
250 (uvmexp.free < uvmexp.freetarg &&
251 uvmexp.inactive < uvmexp.inactarg))
252 wakeup(&uvm.pagedaemon);
253
254 return (error);
255 }
256
257 /*
258 * uvm_pglistfree: free a list of pages
259 *
260 * => pages should already be unmapped
261 */
262
263 void
264 uvm_pglistfree(list)
265 struct pglist *list;
266 {
267 vm_page_t m;
268 int s;
269
270 /*
271 * Block all memory allocation and lock the free list.
272 */
273 s = uvm_lock_fpageq();
274
275 while ((m = list->tqh_first) != NULL) {
276 #ifdef DIAGNOSTIC
277 if (m->pqflags & (PQ_ACTIVE|PQ_INACTIVE))
278 panic("uvm_pglistfree: active/inactive page!");
279 #endif
280 TAILQ_REMOVE(list, m, pageq);
281 m->pqflags = PQ_FREE;
282 TAILQ_INSERT_TAIL(&uvm.page_free[
283 uvm_page_lookup_freelist(m)].pgfl_queues[PGFL_UNKNOWN],
284 m, pageq);
285 uvmexp.free++;
286 if (uvmexp.zeropages < UVM_PAGEZERO_TARGET)
287 uvm.page_idle_zero = vm_page_zero_enable;
288 STAT_DECR(uvm_pglistalloc_npages);
289 }
290
291 uvm_unlock_fpageq(s);
292 }
293