2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2017, Jeffrey Roberson <jeff@freebsd.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/bitset.h>
38 #include <sys/domainset.h>
41 #include <sys/mutex.h>
42 #include <sys/malloc.h>
43 #include <sys/vmmeter.h>
46 #include <vm/vm_param.h>
47 #include <vm/vm_domainset.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <vm/vm_phys.h>
54 * Iterators are written such that the first nowait pass has as short a
55 * codepath as possible to eliminate bloat from the allocator. It is
56 * assumed that most allocations are successful.
59 static int vm_domainset_default_stride = 64;
62 * Determine which policy is to be used for this allocation.
65 vm_domainset_iter_init(struct vm_domainset_iter *di, struct vm_object *obj,
68 struct domainset *domain;
71 * object policy takes precedence over thread policy. The policies
72 * are immutable and unsynchronized. Updates can race but pointer
73 * loads are assumed to be atomic.
75 if (obj != NULL && (domain = obj->domain.dr_policy) != NULL) {
76 di->di_domain = domain;
77 di->di_iter = &obj->domain.dr_iterator;
79 di->di_domain = curthread->td_domain.dr_policy;
80 di->di_iter = &curthread->td_domain.dr_iterator;
82 di->di_policy = di->di_domain->ds_policy;
83 if (di->di_policy == DOMAINSET_POLICY_INTERLEAVE) {
84 if (vm_object_reserv(obj)) {
86 * Color the pindex so we end up on the correct
87 * reservation boundary.
89 pindex += obj->pg_color;
90 pindex >>= VM_LEVEL_0_ORDER;
92 pindex /= vm_domainset_default_stride;
94 * Offset pindex so the first page of each object does
95 * not end up in domain 0.
98 pindex += (((uintptr_t)obj) / sizeof(*obj));
99 di->di_offset = pindex;
104 vm_domainset_iter_rr(struct vm_domainset_iter *di, int *domain)
107 *domain = di->di_domain->ds_order[
108 ++(*di->di_iter) % di->di_domain->ds_cnt];
112 vm_domainset_iter_prefer(struct vm_domainset_iter *di, int *domain)
117 d = di->di_domain->ds_order[
118 ++(*di->di_iter) % di->di_domain->ds_cnt];
119 } while (d == di->di_domain->ds_prefer);
124 vm_domainset_iter_interleave(struct vm_domainset_iter *di, int *domain)
128 d = di->di_offset % di->di_domain->ds_cnt;
130 *domain = di->di_domain->ds_order[d];
134 vm_domainset_iter_next(struct vm_domainset_iter *di, int *domain)
137 KASSERT(di->di_n > 0,
138 ("vm_domainset_iter_first: Invalid n %d", di->di_n));
139 switch (di->di_policy) {
140 case DOMAINSET_POLICY_FIRSTTOUCH:
142 * To prevent impossible allocations we convert an invalid
143 * first-touch to round-robin.
146 case DOMAINSET_POLICY_INTERLEAVE:
148 case DOMAINSET_POLICY_ROUNDROBIN:
149 vm_domainset_iter_rr(di, domain);
151 case DOMAINSET_POLICY_PREFER:
152 vm_domainset_iter_prefer(di, domain);
155 panic("vm_domainset_iter_first: Unknown policy %d",
158 KASSERT(*domain < vm_ndomains,
159 ("vm_domainset_iter_next: Invalid domain %d", *domain));
163 vm_domainset_iter_first(struct vm_domainset_iter *di, int *domain)
166 switch (di->di_policy) {
167 case DOMAINSET_POLICY_FIRSTTOUCH:
168 *domain = PCPU_GET(domain);
169 if (DOMAINSET_ISSET(*domain, &di->di_domain->ds_mask)) {
171 * Add an extra iteration because we will visit the
172 * current domain a second time in the rr iterator.
174 di->di_n = di->di_domain->ds_cnt + 1;
178 * To prevent impossible allocations we convert an invalid
179 * first-touch to round-robin.
182 case DOMAINSET_POLICY_ROUNDROBIN:
183 di->di_n = di->di_domain->ds_cnt;
184 vm_domainset_iter_rr(di, domain);
186 case DOMAINSET_POLICY_PREFER:
187 *domain = di->di_domain->ds_prefer;
188 di->di_n = di->di_domain->ds_cnt;
190 case DOMAINSET_POLICY_INTERLEAVE:
191 vm_domainset_iter_interleave(di, domain);
192 di->di_n = di->di_domain->ds_cnt;
195 panic("vm_domainset_iter_first: Unknown policy %d",
198 KASSERT(di->di_n > 0,
199 ("vm_domainset_iter_first: Invalid n %d", di->di_n));
200 KASSERT(*domain < vm_ndomains,
201 ("vm_domainset_iter_first: Invalid domain %d", *domain));
205 vm_domainset_iter_page_init(struct vm_domainset_iter *di, struct vm_object *obj,
206 vm_pindex_t pindex, int *domain, int *req)
209 vm_domainset_iter_init(di, obj, pindex);
211 *req = (di->di_flags & ~(VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL)) |
213 vm_domainset_iter_first(di, domain);
217 vm_domainset_iter_page(struct vm_domainset_iter *di, int *domain, int *req)
221 * If we exhausted all options with NOWAIT and did a WAITFAIL it
222 * is time to return an error to the caller.
224 if ((*req & VM_ALLOC_WAITFAIL) != 0)
227 /* If there are more domains to visit we run the iterator. */
228 if (--di->di_n != 0) {
229 vm_domainset_iter_next(di, domain);
233 /* If we visited all domains and this was a NOWAIT we return error. */
234 if ((di->di_flags & (VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL)) == 0)
238 * We have visited all domains with non-blocking allocations, try
239 * from the beginning with a blocking allocation.
241 vm_domainset_iter_first(di, domain);
249 vm_domainset_iter_malloc_init(struct vm_domainset_iter *di,
250 struct vm_object *obj, int *domain, int *flags)
253 vm_domainset_iter_init(di, obj, 0);
254 if (di->di_policy == DOMAINSET_POLICY_INTERLEAVE)
255 di->di_policy = DOMAINSET_POLICY_ROUNDROBIN;
256 di->di_flags = *flags;
257 *flags = (di->di_flags & ~M_WAITOK) | M_NOWAIT;
258 vm_domainset_iter_first(di, domain);
262 vm_domainset_iter_malloc(struct vm_domainset_iter *di, int *domain, int *flags)
265 /* If there are more domains to visit we run the iterator. */
266 if (--di->di_n != 0) {
267 vm_domainset_iter_next(di, domain);
271 /* If we visited all domains and this was a NOWAIT we return error. */
272 if ((di->di_flags & M_WAITOK) == 0)
276 * We have visited all domains with non-blocking allocations, try
277 * from the beginning with a blocking allocation.
279 vm_domainset_iter_first(di, domain);
280 *flags = di->di_flags;
287 vm_domainset_iter_page(struct vm_domainset_iter *di, int *domain, int *flags)
290 return (EJUSTRETURN);
294 vm_domainset_iter_page_init(struct vm_domainset_iter *di,
295 struct vm_object *obj, vm_pindex_t pindex, int *domain, int *flags)
302 vm_domainset_iter_malloc(struct vm_domainset_iter *di, int *domain, int *flags)
305 return (EJUSTRETURN);
309 vm_domainset_iter_malloc_init(struct vm_domainset_iter *di,
310 struct vm_object *obj, int *domain, int *flags)