xref: /freebsd/sys/vm/sg_pager.c (revision 9517e866259191fcd39434a97ad849a9b59b9b9f)
1 /*-
2  * Copyright (c) 2009 Advanced Computing Technologies LLC
3  * Written by: John H. Baldwin <jhb@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 /*
32  * This pager manages OBJT_SG objects.  These objects are backed by
33  * a scatter/gather list of physical address ranges.
34  */
35 
36 #include <sys/param.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/sglist.h>
40 #include <vm/vm.h>
41 #include <vm/vm_object.h>
42 #include <vm/vm_page.h>
43 #include <vm/vm_pager.h>
44 #include <vm/uma.h>
45 
46 static void sg_pager_init(void);
47 static vm_object_t sg_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
48     vm_ooffset_t, struct ucred *);
49 static void sg_pager_dealloc(vm_object_t);
50 static int sg_pager_getpages(vm_object_t, vm_page_t *, int, int);
51 static void sg_pager_putpages(vm_object_t, vm_page_t *, int,
52 		boolean_t, int *);
53 static boolean_t sg_pager_haspage(vm_object_t, vm_pindex_t, int *,
54 		int *);
55 
56 static uma_zone_t fakepg_zone;
57 
58 static vm_page_t sg_pager_getfake(vm_paddr_t, vm_memattr_t);
59 static void sg_pager_putfake(vm_page_t);
60 
61 struct pagerops sgpagerops = {
62 	.pgo_init =	sg_pager_init,
63 	.pgo_alloc =	sg_pager_alloc,
64 	.pgo_dealloc =	sg_pager_dealloc,
65 	.pgo_getpages =	sg_pager_getpages,
66 	.pgo_putpages =	sg_pager_putpages,
67 	.pgo_haspage =	sg_pager_haspage,
68 };
69 
70 static void
71 sg_pager_init(void)
72 {
73 
74 	fakepg_zone = uma_zcreate("SG fakepg", sizeof(struct vm_page),
75 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
76 	    UMA_ZONE_NOFREE|UMA_ZONE_VM);
77 }
78 
79 static vm_object_t
80 sg_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
81     vm_ooffset_t foff, struct ucred *cred)
82 {
83 	struct sglist *sg;
84 	vm_object_t object;
85 	vm_pindex_t npages, pindex;
86 	int i;
87 
88 	/*
89 	 * Offset should be page aligned.
90 	 */
91 	if (foff & PAGE_MASK)
92 		return (NULL);
93 
94 	/*
95 	 * The scatter/gather list must only include page-aligned
96 	 * ranges.
97 	 */
98 	npages = 0;
99 	sg = handle;
100 	for (i = 0; i < sg->sg_nseg; i++) {
101 		if ((sg->sg_segs[i].ss_paddr % PAGE_SIZE) != 0 ||
102 		    (sg->sg_segs[i].ss_len % PAGE_SIZE) != 0)
103 			return (NULL);
104 		npages += sg->sg_segs[i].ss_len / PAGE_SIZE;
105 	}
106 
107 	/*
108 	 * The scatter/gather list has a fixed size.  Refuse requests
109 	 * to map beyond that.
110 	 */
111 	size = round_page(size);
112 	pindex = OFF_TO_IDX(foff + size);
113 	if (pindex > npages)
114 		return (NULL);
115 
116 	/*
117 	 * Allocate a new object and associate it with the
118 	 * scatter/gather list.  It is ok for our purposes to have
119 	 * multiple VM objects associated with the same scatter/gather
120 	 * list because scatter/gather lists are static.  This is also
121 	 * simpler than ensuring a unique object per scatter/gather
122 	 * list.
123 	 */
124 	object = vm_object_allocate(OBJT_SG, npages);
125 	object->handle = sglist_hold(sg);
126 	TAILQ_INIT(&object->un_pager.sgp.sgp_pglist);
127 	return (object);
128 }
129 
130 static void
131 sg_pager_dealloc(vm_object_t object)
132 {
133 	struct sglist *sg;
134 	vm_page_t m;
135 
136 	/*
137 	 * Free up our fake pages.
138 	 */
139 	while ((m = TAILQ_FIRST(&object->un_pager.sgp.sgp_pglist)) != 0) {
140 		TAILQ_REMOVE(&object->un_pager.sgp.sgp_pglist, m, pageq);
141 		sg_pager_putfake(m);
142 	}
143 
144 	sg = object->handle;
145 	sglist_free(sg);
146 }
147 
148 static int
149 sg_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage)
150 {
151 	struct sglist *sg;
152 	vm_page_t m_paddr, page;
153 	vm_pindex_t offset;
154 	vm_paddr_t paddr;
155 	vm_memattr_t memattr;
156 	size_t space;
157 	int i;
158 
159 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
160 	sg = object->handle;
161 	memattr = object->memattr;
162 	VM_OBJECT_UNLOCK(object);
163 	offset = m[reqpage]->pindex;
164 
165 	/*
166 	 * Lookup the physical address of the requested page.  An initial
167 	 * value of '1' instead of '0' is used so we can assert that the
168 	 * page is found since '0' can be a valid page-aligned physical
169 	 * address.
170 	 */
171 	space = 0;
172 	paddr = 1;
173 	for (i = 0; i < sg->sg_nseg; i++) {
174 		if (space + sg->sg_segs[i].ss_len <= (offset * PAGE_SIZE)) {
175 			space += sg->sg_segs[i].ss_len;
176 			continue;
177 		}
178 		paddr = sg->sg_segs[i].ss_paddr + offset * PAGE_SIZE - space;
179 		break;
180 	}
181 	KASSERT(paddr != 1, ("invalid SG page index"));
182 
183 	/* If "paddr" is a real page, perform a sanity check on "memattr". */
184 	if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
185 	    pmap_page_get_memattr(m_paddr) != memattr) {
186 		memattr = pmap_page_get_memattr(m_paddr);
187 		printf(
188 	    "WARNING: A device driver has set \"memattr\" inconsistently.\n");
189 	}
190 
191 	/* Return a fake page for the requested page. */
192 	KASSERT(!(m[reqpage]->flags & PG_FICTITIOUS),
193 	    ("backing page for SG is fake"));
194 
195 	/* Construct a new fake page. */
196 	page = sg_pager_getfake(paddr, memattr);
197 	VM_OBJECT_LOCK(object);
198 	TAILQ_INSERT_TAIL(&object->un_pager.sgp.sgp_pglist, page, pageq);
199 
200 	/* Free the original pages and insert this fake page into the object. */
201 	vm_page_lock_queues();
202 	for (i = 0; i < count; i++)
203 		vm_page_free(m[i]);
204 	vm_page_unlock_queues();
205 	vm_page_insert(page, object, offset);
206 	m[reqpage] = page;
207 
208 	return (VM_PAGER_OK);
209 }
210 
211 static void
212 sg_pager_putpages(vm_object_t object, vm_page_t *m, int count,
213     boolean_t sync, int *rtvals)
214 {
215 
216 	panic("sg_pager_putpage called");
217 }
218 
219 static boolean_t
220 sg_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
221     int *after)
222 {
223 
224 	if (before != NULL)
225 		*before = 0;
226 	if (after != NULL)
227 		*after = 0;
228 	return (TRUE);
229 }
230 
231 /*
232  * Create a fictitious page with the specified physical address and memory
233  * attribute.  The memory attribute is the only the machine-dependent aspect
234  * of a fictitious page that must be initialized.
235  */
236 static vm_page_t
237 sg_pager_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
238 {
239 	vm_page_t m;
240 
241 	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
242 	m->phys_addr = paddr;
243 	/* Fictitious pages don't use "segind". */
244 	m->flags = PG_FICTITIOUS;
245 	/* Fictitious pages don't use "order" or "pool". */
246 	m->oflags = VPO_BUSY;
247 	m->wire_count = 1;
248 	pmap_page_set_memattr(m, memattr);
249 	return (m);
250 }
251 
252 static void
253 sg_pager_putfake(vm_page_t m)
254 {
255 
256 	if (!(m->flags & PG_FICTITIOUS))
257 		panic("sg_pager_putfake: bad page");
258 	uma_zfree(fakepg_zone, m);
259 }
260