1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2009 Hudson River Trading LLC
5 * Written by: John H. Baldwin <jhb@FreeBSD.org>
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
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.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30#include <sys/cdefs.h>
31__FBSDID("$FreeBSD$");
32
33/*
34 * This pager manages OBJT_SG objects.  These objects are backed by
35 * a scatter/gather list of physical address ranges.
36 */
37
38#include <sys/param.h>
39#include <sys/lock.h>
40#include <sys/mutex.h>
41#include <sys/rwlock.h>
42#include <sys/sglist.h>
43#include <sys/user.h>
44#include <sys/vmmeter.h>
45
46#include <vm/vm.h>
47#include <vm/vm_param.h>
48#include <vm/vm_object.h>
49#include <vm/vm_page.h>
50#include <vm/vm_pager.h>
51#include <vm/vm_phys.h>
52#include <vm/uma.h>
53
54static vm_object_t sg_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
55    vm_ooffset_t, struct ucred *);
56static void sg_pager_dealloc(vm_object_t);
57static int sg_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *);
58static void sg_pager_putpages(vm_object_t, vm_page_t *, int,
59		boolean_t, int *);
60static boolean_t sg_pager_haspage(vm_object_t, vm_pindex_t, int *,
61		int *);
62
63const struct pagerops sgpagerops = {
64	.pgo_kvme_type = KVME_TYPE_SG,
65	.pgo_alloc =	sg_pager_alloc,
66	.pgo_dealloc =	sg_pager_dealloc,
67	.pgo_getpages =	sg_pager_getpages,
68	.pgo_putpages =	sg_pager_putpages,
69	.pgo_haspage =	sg_pager_haspage,
70};
71
72static vm_object_t
73sg_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
74    vm_ooffset_t foff, struct ucred *cred)
75{
76	struct sglist *sg;
77	vm_object_t object;
78	vm_pindex_t npages, pindex;
79	int i;
80
81	/*
82	 * Offset should be page aligned.
83	 */
84	if (foff & PAGE_MASK)
85		return (NULL);
86
87	/*
88	 * The scatter/gather list must only include page-aligned
89	 * ranges.
90	 */
91	npages = 0;
92	sg = handle;
93	for (i = 0; i < sg->sg_nseg; i++) {
94		if ((sg->sg_segs[i].ss_paddr % PAGE_SIZE) != 0 ||
95		    (sg->sg_segs[i].ss_len % PAGE_SIZE) != 0)
96			return (NULL);
97		npages += sg->sg_segs[i].ss_len / PAGE_SIZE;
98	}
99
100	/*
101	 * The scatter/gather list has a fixed size.  Refuse requests
102	 * to map beyond that.
103	 */
104	size = round_page(size);
105	pindex = OFF_TO_IDX(foff) + OFF_TO_IDX(size);
106	if (pindex > npages || pindex < OFF_TO_IDX(foff) ||
107	    pindex < OFF_TO_IDX(size))
108		return (NULL);
109
110	/*
111	 * Allocate a new object and associate it with the
112	 * scatter/gather list.  It is ok for our purposes to have
113	 * multiple VM objects associated with the same scatter/gather
114	 * list because scatter/gather lists are static.  This is also
115	 * simpler than ensuring a unique object per scatter/gather
116	 * list.
117	 */
118	object = vm_object_allocate(OBJT_SG, npages);
119	object->handle = sglist_hold(sg);
120	TAILQ_INIT(&object->un_pager.sgp.sgp_pglist);
121	return (object);
122}
123
124static void
125sg_pager_dealloc(vm_object_t object)
126{
127	struct sglist *sg;
128	vm_page_t m;
129
130	/*
131	 * Free up our fake pages.
132	 */
133	while ((m = TAILQ_FIRST(&object->un_pager.sgp.sgp_pglist)) != 0) {
134		if (vm_page_busy_acquire(m, VM_ALLOC_WAITFAIL) == 0)
135			continue;
136		TAILQ_REMOVE(&object->un_pager.sgp.sgp_pglist, m, plinks.q);
137		vm_page_putfake(m);
138	}
139
140	sg = object->handle;
141	sglist_free(sg);
142	object->handle = NULL;
143	object->type = OBJT_DEAD;
144}
145
146static int
147sg_pager_getpages(vm_object_t object, vm_page_t *m, int count, int *rbehind,
148    int *rahead)
149{
150	struct sglist *sg;
151	vm_page_t m_paddr, page;
152	vm_pindex_t offset;
153	vm_paddr_t paddr;
154	vm_memattr_t memattr;
155	size_t space;
156	int i;
157
158	/* Since our haspage reports zero after/before, the count is 1. */
159	KASSERT(count == 1, ("%s: count %d", __func__, count));
160	/* Handle is stable while paging is in progress. */
161	sg = object->handle;
162	memattr = object->memattr;
163	offset = m[0]->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[0]->flags & PG_FICTITIOUS),
193	    ("backing page for SG is fake"));
194
195	/* Construct a new fake page. */
196	page = vm_page_getfake(paddr, memattr);
197	VM_OBJECT_WLOCK(object);
198	TAILQ_INSERT_TAIL(&object->un_pager.sgp.sgp_pglist, page, plinks.q);
199	vm_page_replace(page, object, offset, m[0]);
200	VM_OBJECT_WUNLOCK(object);
201	m[0] = page;
202	vm_page_valid(page);
203
204	if (rbehind)
205		*rbehind = 0;
206	if (rahead)
207		*rahead = 0;
208
209	return (VM_PAGER_OK);
210}
211
212static void
213sg_pager_putpages(vm_object_t object, vm_page_t *m, int count,
214    boolean_t sync, int *rtvals)
215{
216
217	panic("sg_pager_putpage called");
218}
219
220static boolean_t
221sg_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
222    int *after)
223{
224
225	if (before != NULL)
226		*before = 0;
227	if (after != NULL)
228		*after = 0;
229	return (TRUE);
230}
231