uvm_pglist.c revision 1.10 1 1.10 thorpej /* $NetBSD: uvm_pglist.c,v 1.10 2000/05/20 19:54:01 thorpej Exp $ */
2 1.2 thorpej
3 1.1 mrg /*-
4 1.1 mrg * Copyright (c) 1997 The NetBSD Foundation, Inc.
5 1.1 mrg * All rights reserved.
6 1.1 mrg *
7 1.1 mrg * This code is derived from software contributed to The NetBSD Foundation
8 1.1 mrg * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9 1.1 mrg * NASA Ames Research Center.
10 1.1 mrg *
11 1.1 mrg * Redistribution and use in source and binary forms, with or without
12 1.1 mrg * modification, are permitted provided that the following conditions
13 1.1 mrg * are met:
14 1.1 mrg * 1. Redistributions of source code must retain the above copyright
15 1.1 mrg * notice, this list of conditions and the following disclaimer.
16 1.1 mrg * 2. Redistributions in binary form must reproduce the above copyright
17 1.1 mrg * notice, this list of conditions and the following disclaimer in the
18 1.1 mrg * documentation and/or other materials provided with the distribution.
19 1.1 mrg * 3. All advertising materials mentioning features or use of this software
20 1.1 mrg * must display the following acknowledgement:
21 1.1 mrg * This product includes software developed by the NetBSD
22 1.1 mrg * Foundation, Inc. and its contributors.
23 1.1 mrg * 4. Neither the name of The NetBSD Foundation nor the names of its
24 1.1 mrg * contributors may be used to endorse or promote products derived
25 1.1 mrg * from this software without specific prior written permission.
26 1.1 mrg *
27 1.1 mrg * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28 1.1 mrg * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29 1.1 mrg * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30 1.1 mrg * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31 1.1 mrg * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 1.1 mrg * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 1.1 mrg * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 1.1 mrg * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 1.1 mrg * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 1.1 mrg * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 1.1 mrg * POSSIBILITY OF SUCH DAMAGE.
38 1.1 mrg */
39 1.1 mrg
40 1.1 mrg /*
41 1.1 mrg * uvm_pglist.c: pglist functions
42 1.1 mrg */
43 1.1 mrg
44 1.1 mrg #include <sys/param.h>
45 1.1 mrg #include <sys/systm.h>
46 1.1 mrg #include <sys/malloc.h>
47 1.1 mrg #include <sys/proc.h>
48 1.1 mrg
49 1.1 mrg #include <vm/vm.h>
50 1.1 mrg
51 1.1 mrg #include <uvm/uvm.h>
52 1.1 mrg
53 1.1 mrg #ifdef VM_PAGE_ALLOC_MEMORY_STATS
54 1.1 mrg #define STAT_INCR(v) (v)++
55 1.1 mrg #define STAT_DECR(v) do { \
56 1.1 mrg if ((v) == 0) \
57 1.1 mrg printf("%s:%d -- Already 0!\n", __FILE__, __LINE__); \
58 1.1 mrg else \
59 1.1 mrg (v)--; \
60 1.1 mrg } while (0)
61 1.1 mrg u_long uvm_pglistalloc_npages;
62 1.1 mrg #else
63 1.1 mrg #define STAT_INCR(v)
64 1.1 mrg #define STAT_DECR(v)
65 1.1 mrg #endif
66 1.1 mrg
67 1.1 mrg /*
68 1.1 mrg * uvm_pglistalloc: allocate a list of pages
69 1.1 mrg *
70 1.1 mrg * => allocated pages are placed at the tail of rlist. rlist is
71 1.1 mrg * assumed to be properly initialized by caller.
72 1.1 mrg * => returns 0 on success or errno on failure
73 1.1 mrg * => XXX: implementation allocates only a single segment, also
74 1.1 mrg * might be able to better advantage of vm_physeg[].
75 1.1 mrg * => doesn't take into account clean non-busy pages on inactive list
76 1.1 mrg * that could be used(?)
77 1.1 mrg * => params:
78 1.1 mrg * size the size of the allocation, rounded to page size.
79 1.1 mrg * low the low address of the allowed allocation range.
80 1.1 mrg * high the high address of the allowed allocation range.
81 1.1 mrg * alignment memory must be aligned to this power-of-two boundary.
82 1.1 mrg * boundary no segment in the allocation may cross this
83 1.1 mrg * power-of-two boundary (relative to zero).
84 1.1 mrg */
85 1.1 mrg
86 1.3 mrg int
87 1.3 mrg uvm_pglistalloc(size, low, high, alignment, boundary, rlist, nsegs, waitok)
88 1.6 eeh psize_t size;
89 1.6 eeh paddr_t low, high, alignment, boundary;
90 1.3 mrg struct pglist *rlist;
91 1.3 mrg int nsegs, waitok;
92 1.1 mrg {
93 1.6 eeh paddr_t try, idxpa, lastidxpa;
94 1.3 mrg int psi;
95 1.3 mrg struct vm_page *pgs;
96 1.9 thorpej int s, tryidx, idx, pgflidx, end, error, free_list;
97 1.3 mrg vm_page_t m;
98 1.3 mrg u_long pagemask;
99 1.1 mrg #ifdef DEBUG
100 1.3 mrg vm_page_t tp;
101 1.1 mrg #endif
102 1.1 mrg
103 1.1 mrg #ifdef DIAGNOSTIC
104 1.3 mrg if ((alignment & (alignment - 1)) != 0)
105 1.9 thorpej panic("uvm_pglistalloc: alignment must be power of 2");
106 1.1 mrg
107 1.3 mrg if ((boundary & (boundary - 1)) != 0)
108 1.9 thorpej panic("uvm_pglistalloc: boundary must be power of 2");
109 1.1 mrg #endif
110 1.3 mrg
111 1.1 mrg /*
112 1.3 mrg * Our allocations are always page granularity, so our alignment
113 1.3 mrg * must be, too.
114 1.1 mrg */
115 1.3 mrg if (alignment < PAGE_SIZE)
116 1.3 mrg alignment = PAGE_SIZE;
117 1.3 mrg
118 1.3 mrg size = round_page(size);
119 1.3 mrg try = roundup(low, alignment);
120 1.1 mrg
121 1.3 mrg if (boundary != 0 && boundary < size)
122 1.3 mrg return (EINVAL);
123 1.1 mrg
124 1.3 mrg pagemask = ~(boundary - 1);
125 1.1 mrg
126 1.3 mrg /* Default to "lose". */
127 1.3 mrg error = ENOMEM;
128 1.3 mrg
129 1.3 mrg /*
130 1.3 mrg * Block all memory allocation and lock the free list.
131 1.3 mrg */
132 1.7 thorpej s = uvm_lock_fpageq(); /* lock free page queue */
133 1.3 mrg
134 1.3 mrg /* Are there even any free pages? */
135 1.9 thorpej if (uvmexp.free <= (uvmexp.reserve_pagedaemon +
136 1.9 thorpej uvmexp.reserve_kernel))
137 1.3 mrg goto out;
138 1.3 mrg
139 1.3 mrg for (;; try += alignment) {
140 1.3 mrg if (try + size > high) {
141 1.3 mrg /*
142 1.3 mrg * We've run past the allowable range.
143 1.3 mrg */
144 1.3 mrg goto out;
145 1.3 mrg }
146 1.3 mrg
147 1.3 mrg /*
148 1.3 mrg * Make sure this is a managed physical page.
149 1.3 mrg */
150 1.3 mrg
151 1.3 mrg if ((psi = vm_physseg_find(atop(try), &idx)) == -1)
152 1.3 mrg continue; /* managed? */
153 1.3 mrg if (vm_physseg_find(atop(try + size), NULL) != psi)
154 1.3 mrg continue; /* end must be in this segment */
155 1.3 mrg
156 1.3 mrg tryidx = idx;
157 1.3 mrg end = idx + (size / PAGE_SIZE);
158 1.3 mrg pgs = vm_physmem[psi].pgs;
159 1.3 mrg
160 1.3 mrg /*
161 1.3 mrg * Found a suitable starting page. See of the range is free.
162 1.3 mrg */
163 1.3 mrg for (; idx < end; idx++) {
164 1.3 mrg if (VM_PAGE_IS_FREE(&pgs[idx]) == 0) {
165 1.3 mrg /*
166 1.3 mrg * Page not available.
167 1.3 mrg */
168 1.3 mrg break;
169 1.3 mrg }
170 1.3 mrg
171 1.3 mrg idxpa = VM_PAGE_TO_PHYS(&pgs[idx]);
172 1.3 mrg
173 1.3 mrg if (idx > tryidx) {
174 1.3 mrg lastidxpa = VM_PAGE_TO_PHYS(&pgs[idx - 1]);
175 1.3 mrg
176 1.3 mrg if ((lastidxpa + PAGE_SIZE) != idxpa) {
177 1.3 mrg /*
178 1.3 mrg * Region not contiguous.
179 1.3 mrg */
180 1.3 mrg break;
181 1.3 mrg }
182 1.3 mrg if (boundary != 0 &&
183 1.3 mrg ((lastidxpa ^ idxpa) & pagemask) != 0) {
184 1.3 mrg /*
185 1.3 mrg * Region crosses boundary.
186 1.3 mrg */
187 1.3 mrg break;
188 1.3 mrg }
189 1.3 mrg }
190 1.3 mrg }
191 1.3 mrg
192 1.3 mrg if (idx == end) {
193 1.3 mrg /*
194 1.3 mrg * Woo hoo! Found one.
195 1.3 mrg */
196 1.3 mrg break;
197 1.3 mrg }
198 1.1 mrg }
199 1.1 mrg
200 1.9 thorpej #if PGFL_NQUEUES != 2
201 1.9 thorpej #error uvm_pglistalloc needs to be updated
202 1.9 thorpej #endif
203 1.9 thorpej
204 1.3 mrg /*
205 1.3 mrg * we have a chunk of memory that conforms to the requested constraints.
206 1.3 mrg */
207 1.3 mrg idx = tryidx;
208 1.3 mrg while (idx < end) {
209 1.3 mrg m = &pgs[idx];
210 1.5 thorpej free_list = uvm_page_lookup_freelist(m);
211 1.9 thorpej pgflidx = (m->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN;
212 1.1 mrg #ifdef DEBUG
213 1.9 thorpej for (tp = TAILQ_FIRST(&uvm.page_free[
214 1.9 thorpej free_list].pgfl_queues[pgflidx]);
215 1.9 thorpej tp != NULL;
216 1.9 thorpej tp = TAILQ_NEXT(tp, pageq)) {
217 1.3 mrg if (tp == m)
218 1.3 mrg break;
219 1.3 mrg }
220 1.3 mrg if (tp == NULL)
221 1.3 mrg panic("uvm_pglistalloc: page not on freelist");
222 1.1 mrg #endif
223 1.9 thorpej TAILQ_REMOVE(&uvm.page_free[free_list].pgfl_queues[pgflidx],
224 1.9 thorpej m, pageq);
225 1.3 mrg uvmexp.free--;
226 1.9 thorpej if (m->flags & PG_ZERO)
227 1.9 thorpej uvmexp.zeropages--;
228 1.3 mrg m->flags = PG_CLEAN;
229 1.3 mrg m->pqflags = 0;
230 1.3 mrg m->uobject = NULL;
231 1.3 mrg m->uanon = NULL;
232 1.3 mrg m->wire_count = 0;
233 1.3 mrg m->loan_count = 0;
234 1.3 mrg TAILQ_INSERT_TAIL(rlist, m, pageq);
235 1.3 mrg idx++;
236 1.3 mrg STAT_INCR(uvm_pglistalloc_npages);
237 1.3 mrg }
238 1.3 mrg error = 0;
239 1.1 mrg
240 1.1 mrg out:
241 1.7 thorpej uvm_unlock_fpageq(s);
242 1.1 mrg
243 1.3 mrg /*
244 1.3 mrg * check to see if we need to generate some free pages waking
245 1.3 mrg * the pagedaemon.
246 1.3 mrg * XXX: we read uvm.free without locking
247 1.3 mrg */
248 1.3 mrg
249 1.3 mrg if (uvmexp.free < uvmexp.freemin ||
250 1.3 mrg (uvmexp.free < uvmexp.freetarg &&
251 1.3 mrg uvmexp.inactive < uvmexp.inactarg))
252 1.8 thorpej wakeup(&uvm.pagedaemon);
253 1.1 mrg
254 1.3 mrg return (error);
255 1.1 mrg }
256 1.1 mrg
257 1.1 mrg /*
258 1.1 mrg * uvm_pglistfree: free a list of pages
259 1.1 mrg *
260 1.1 mrg * => pages should already be unmapped
261 1.1 mrg */
262 1.1 mrg
263 1.3 mrg void
264 1.3 mrg uvm_pglistfree(list)
265 1.3 mrg struct pglist *list;
266 1.1 mrg {
267 1.3 mrg vm_page_t m;
268 1.3 mrg int s;
269 1.1 mrg
270 1.3 mrg /*
271 1.3 mrg * Block all memory allocation and lock the free list.
272 1.3 mrg */
273 1.7 thorpej s = uvm_lock_fpageq();
274 1.1 mrg
275 1.3 mrg while ((m = list->tqh_first) != NULL) {
276 1.1 mrg #ifdef DIAGNOSTIC
277 1.3 mrg if (m->pqflags & (PQ_ACTIVE|PQ_INACTIVE))
278 1.3 mrg panic("uvm_pglistfree: active/inactive page!");
279 1.1 mrg #endif
280 1.3 mrg TAILQ_REMOVE(list, m, pageq);
281 1.3 mrg m->pqflags = PQ_FREE;
282 1.9 thorpej TAILQ_INSERT_TAIL(&uvm.page_free[
283 1.9 thorpej uvm_page_lookup_freelist(m)].pgfl_queues[PGFL_UNKNOWN],
284 1.5 thorpej m, pageq);
285 1.3 mrg uvmexp.free++;
286 1.9 thorpej if (uvmexp.zeropages < UVM_PAGEZERO_TARGET)
287 1.9 thorpej uvm.page_idle_zero = vm_page_zero_enable;
288 1.3 mrg STAT_DECR(uvm_pglistalloc_npages);
289 1.3 mrg }
290 1.1 mrg
291 1.7 thorpej uvm_unlock_fpageq(s);
292 1.1 mrg }
293