xref: /netbsd-src/sys/arch/xen/x86/xen_shm_machdep.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*      $NetBSD: xen_shm_machdep.c,v 1.8 2010/03/28 20:46:18 snj Exp $      */
2 
3 /*
4  * Copyright (c) 2006 Manuel Bouyer.
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 ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  */
27 
28 #include <sys/cdefs.h>
29 __KERNEL_RCSID(0, "$NetBSD: xen_shm_machdep.c,v 1.8 2010/03/28 20:46:18 snj Exp $");
30 
31 
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/queue.h>
36 #include <sys/vmem.h>
37 #include <sys/kernel.h>
38 #include <uvm/uvm.h>
39 
40 #include <machine/pmap.h>
41 #include <xen/hypervisor.h>
42 #include <xen/xen.h>
43 #include <xen/evtchn.h>
44 #include <xen/xen_shm.h>
45 
46 /*
47  * Helper routines for the backend drivers. This implement the necessary
48  * functions to map a bunch of pages from foreign domains in our kernel VM
49  * space, do I/O to it, and unmap it.
50  *
51  * At boot time, we grap some kernel VM space that we'll use to map the foreign
52  * pages. We also maintain a virtual to machine mapping table to give back
53  * the appropriate address to bus_dma if requested.
54  * If no more VM space is available, we return an error. The caller can then
55  * register a callback which will be called when the required VM space is
56  * available.
57  */
58 
59 /* pointers to our VM space */
60 static vaddr_t xen_shm_base_address;
61 static u_long xen_shm_base_address_pg;
62 static vaddr_t xen_shm_end_address;
63 
64 /* Grab enough VM space to map an entire vbd ring. */
65 /* Xen3 linux guests seems to eat more pages, gives enough for 10 vbd rings */
66 #define BLKIF_RING_SIZE __RING_SIZE((blkif_sring_t *)0, PAGE_SIZE)
67 #define XENSHM_NPAGES (BLKIF_RING_SIZE * (BLKIF_MAX_SEGMENTS_PER_REQUEST + 1) * 10)
68 
69 static vsize_t xen_shm_size = (XENSHM_NPAGES * PAGE_SIZE);
70 
71 /* vm space management */
72 static vmem_t *xen_shm_arena;
73 
74 /* callbacks are registered in a FIFO list. */
75 
76 static SIMPLEQ_HEAD(xen_shm_callback_head, xen_shm_callback_entry)
77     xen_shm_callbacks;
78 struct xen_shm_callback_entry {
79 	SIMPLEQ_ENTRY(xen_shm_callback_entry) xshmc_entries;
80 	int (*xshmc_callback)(void *); /* our callback */
81 	void *xshmc_arg; /* cookie passed to the callback */
82 };
83 /* a pool of struct xen_shm_callback_entry */
84 static struct pool xen_shm_callback_pool;
85 
86 #ifdef DEBUG
87 /* for ratecheck(9) */
88 static struct timeval xen_shm_errintvl = { 60, 0 };  /* a minute, each */
89 #endif
90 
91 void
92 xen_shm_init(void)
93 {
94 	SIMPLEQ_INIT(&xen_shm_callbacks);
95 	pool_init(&xen_shm_callback_pool, sizeof(struct xen_shm_callback_entry),
96 	    0, 0, 0, "xshmc", NULL, IPL_VM);
97 	/* ensure we'll always get items */
98 	if (pool_prime(&xen_shm_callback_pool,
99 	    PAGE_SIZE / sizeof(struct xen_shm_callback_entry)) != 0) {
100 		panic("xen_shm_init can't prime pool");
101 	}
102 
103 	xen_shm_base_address = uvm_km_alloc(kernel_map, xen_shm_size, 0,
104 	    UVM_KMF_VAONLY);
105 	xen_shm_end_address = xen_shm_base_address + xen_shm_size;
106 	xen_shm_base_address_pg = xen_shm_base_address >> PAGE_SHIFT;
107 	if (xen_shm_base_address == 0) {
108 		panic("xen_shm_init no VM space");
109 	}
110 	xen_shm_arena = vmem_create("xen_shm",
111 	    xen_shm_base_address_pg,
112 	    (xen_shm_end_address >> PAGE_SHIFT) - 1 - xen_shm_base_address_pg,
113 	    1, NULL, NULL, NULL, 1, VM_NOSLEEP, IPL_VM);
114 	if (xen_shm_arena == NULL) {
115 		panic("xen_shm_init no arena");
116 	}
117 }
118 
119 int
120 xen_shm_map(int nentries, int domid, grant_ref_t *grefp, vaddr_t *vap,
121     grant_handle_t *handlep, int flags)
122 {
123 	int s, i;
124 	vaddr_t new_va;
125 	u_long new_va_pg;
126 	int err;
127 	gnttab_map_grant_ref_t op[XENSHM_MAX_PAGES_PER_REQUEST];
128 
129 #ifdef DIAGNOSTIC
130 	if (nentries > XENSHM_MAX_PAGES_PER_REQUEST) {
131 		printf("xen_shm_map: %d entries\n", nentries);
132 		panic("xen_shm_map");
133 	}
134 #endif
135 	s = splvm(); /* splvm is the lowest level blocking disk and net IRQ */
136 	/*
137 	 * if a driver is waiting for ressources, don't try to allocate
138 	 * yet. This is to avoid a flood of small requests stalling large
139 	 * ones.
140 	 */
141 	if (__predict_false(SIMPLEQ_FIRST(&xen_shm_callbacks) != NULL) &&
142 	    (flags & XSHM_CALLBACK) == 0) {
143 #ifdef DEBUG
144 		static struct timeval lasttime;
145 #endif
146 		splx(s);
147 #ifdef DEBUG
148 		if (ratecheck(&lasttime, &xen_shm_errintvl))
149 			printf("xen_shm_map: ENOMEM1\n");
150 #endif
151 		return ENOMEM;
152 	}
153 	/* allocate the needed virtual space */
154 	new_va_pg = vmem_alloc(xen_shm_arena, nentries,
155 	    VM_INSTANTFIT | VM_NOSLEEP);
156 	if (new_va_pg == 0) {
157 #ifdef DEBUG
158 		static struct timeval lasttime;
159 #endif
160 		splx(s);
161 #ifdef DEBUG
162 		if (ratecheck(&lasttime, &xen_shm_errintvl))
163 			printf("xen_shm_map: ENOMEM\n");
164 #endif
165 		return ENOMEM;
166 	}
167 	splx(s);
168 
169 	new_va = new_va_pg << PAGE_SHIFT;
170 	for (i = 0; i < nentries; i++) {
171 		op[i].host_addr = new_va + i * PAGE_SIZE;
172 		op[i].dom = domid;
173 		op[i].ref = grefp[i];
174 		op[i].flags = GNTMAP_host_map |
175 		    ((flags & XSHM_RO) ? GNTMAP_readonly : 0);
176 	}
177 	err = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, op, nentries);
178 	if (__predict_false(err))
179 		panic("xen_shm_map: HYPERVISOR_grant_table_op failed");
180 	for (i = 0; i < nentries; i++) {
181 		if (__predict_false(op[i].status))
182 			return op[i].status;
183 		handlep[i] = op[i].handle;
184 	}
185 	*vap = new_va;
186 	return 0;
187 }
188 
189 void
190 xen_shm_unmap(vaddr_t va, int nentries, grant_handle_t *handlep)
191 {
192 	gnttab_unmap_grant_ref_t op[XENSHM_MAX_PAGES_PER_REQUEST];
193 	int ret;
194 	int i;
195 	int s;
196 	struct xen_shm_callback_entry *xshmc;
197 
198 #ifdef DIAGNOSTIC
199 	if (nentries > XENSHM_MAX_PAGES_PER_REQUEST) {
200 		printf("xen_shm_unmap: %d entries\n", nentries);
201 		panic("xen_shm_unmap");
202 	}
203 #endif
204 
205 	for (i = 0; i < nentries; i++) {
206 		op[i].host_addr = va + i * PAGE_SIZE;
207 		op[i].dev_bus_addr = 0;
208 		op[i].handle = handlep[i];
209 	}
210 	ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
211 	    op, nentries);
212 	if (__predict_false(ret))
213 		panic("xen_shm_unmap: unmap failed");
214 	va = va >> PAGE_SHIFT;
215 	s = splvm(); /* splvm is the lowest level blocking disk and net IRQ */
216 	vmem_free(xen_shm_arena, va, nentries);
217 	while (__predict_false((xshmc = SIMPLEQ_FIRST(&xen_shm_callbacks))
218 	    != NULL)) {
219 		SIMPLEQ_REMOVE_HEAD(&xen_shm_callbacks, xshmc_entries);
220 		splx(s);
221 		if (xshmc->xshmc_callback(xshmc->xshmc_arg) == 0) {
222 			/* callback succeeded */
223 			s = splvm();
224 			pool_put(&xen_shm_callback_pool, xshmc);
225 		} else {
226 			/* callback failed, probably out of ressources */
227 			s = splvm();
228 			SIMPLEQ_INSERT_TAIL(&xen_shm_callbacks, xshmc,
229 					    xshmc_entries);
230 
231 			break;
232 		}
233 	}
234 	splx(s);
235 }
236 
237 int
238 xen_shm_callback(int (*callback)(void *), void *arg)
239 {
240 	struct xen_shm_callback_entry *xshmc;
241 	int s;
242 
243 	s = splvm();
244 	xshmc = pool_get(&xen_shm_callback_pool, PR_NOWAIT);
245 	if (xshmc == NULL) {
246 		splx(s);
247 		return ENOMEM;
248 	}
249 	xshmc->xshmc_arg = arg;
250 	xshmc->xshmc_callback = callback;
251 	SIMPLEQ_INSERT_TAIL(&xen_shm_callbacks, xshmc, xshmc_entries);
252 	splx(s);
253 	return 0;
254 }
255