xref: /netbsd-src/sys/arch/xen/x86/xen_shm_machdep.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*      $NetBSD: xen_shm_machdep.c,v 1.2 2007/11/22 16:17:06 bouyer 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  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *      This product includes software developed by Manuel Bouyer.
17  * 4. The name of the author may not be used to endorse or promote products
18  *    derived from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  *
31  */
32 
33 #include <sys/types.h>
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/malloc.h>
37 #include <sys/queue.h>
38 #include <sys/vmem.h>
39 #include <sys/kernel.h>
40 #include <uvm/uvm.h>
41 
42 #include <machine/pmap.h>
43 #include <xen/hypervisor.h>
44 #include <xen/xen.h>
45 #include <xen/evtchn.h>
46 #include <xen/xen_shm.h>
47 
48 /*
49  * Helper routines for the backend drivers. This implement the necessary
50  * functions to map a bunch of pages from foreign domains in our kernel VM
51  * space, do I/O to it, and unmap it.
52  *
53  * At boot time, we grap some kernel VM space that we'll use to map the foreign
54  * pages. We also maintain a virtual to machine mapping table to give back
55  * the appropriate address to bus_dma if requested.
56  * If no more VM space is available, we return an error. The caller can then
57  * register a callback which will be called when the required VM space is
58  * available.
59  */
60 
61 /* pointers to our VM space */
62 static vaddr_t xen_shm_base_address;
63 static u_long xen_shm_base_address_pg;
64 static vaddr_t xen_shm_end_address;
65 
66 /* Grab enouth VM space to map an entire vbd ring. */
67 #ifdef XEN3
68 /* Xen3 linux guests seems to eat more pages, gives enough for 10 vbd rings */
69 #define BLKIF_RING_SIZE __RING_SIZE((blkif_sring_t *)0, PAGE_SIZE)
70 #define XENSHM_NPAGES (BLKIF_RING_SIZE * (BLKIF_MAX_SEGMENTS_PER_REQUEST + 1) * 10)
71 #else
72 #define XENSHM_NPAGES (BLKIF_RING_SIZE * (BLKIF_MAX_SEGMENTS_PER_REQUEST + 1))
73 #endif
74 
75 static vsize_t xen_shm_size = (XENSHM_NPAGES * PAGE_SIZE);
76 
77 /* vm space management */
78 static vmem_t *xen_shm_arena;
79 
80 /* callbacks are registered in a FIFO list. */
81 
82 static SIMPLEQ_HEAD(xen_shm_callback_head, xen_shm_callback_entry)
83     xen_shm_callbacks;
84 struct xen_shm_callback_entry {
85 	SIMPLEQ_ENTRY(xen_shm_callback_entry) xshmc_entries;
86 	int (*xshmc_callback)(void *); /* our callback */
87 	void *xshmc_arg; /* cookie passed to the callback */
88 };
89 /* a pool of struct xen_shm_callback_entry */
90 static struct pool xen_shm_callback_pool;
91 
92 #ifdef DEBUG
93 /* for ratecheck(9) */
94 static struct timeval xen_shm_errintvl = { 60, 0 };  /* a minute, each */
95 #endif
96 
97 void
98 xen_shm_init()
99 {
100 	SIMPLEQ_INIT(&xen_shm_callbacks);
101 	pool_init(&xen_shm_callback_pool, sizeof(struct xen_shm_callback_entry),
102 	    0, 0, 0, "xshmc", NULL, IPL_VM);
103 	/* ensure we'll always get items */
104 	if (pool_prime(&xen_shm_callback_pool,
105 	    PAGE_SIZE / sizeof(struct xen_shm_callback_entry)) != 0) {
106 		panic("xen_shm_init can't prime pool");
107 	}
108 
109 	xen_shm_base_address = uvm_km_alloc(kernel_map, xen_shm_size, 0,
110 	    UVM_KMF_VAONLY);
111 	xen_shm_end_address = xen_shm_base_address + xen_shm_size;
112 	xen_shm_base_address_pg = xen_shm_base_address >> PAGE_SHIFT;
113 	if (xen_shm_base_address == 0) {
114 		panic("xen_shm_init no VM space");
115 	}
116 	xen_shm_arena = vmem_create("xen_shm",
117 	    xen_shm_base_address_pg,
118 	    (xen_shm_end_address >> PAGE_SHIFT) - 1 - xen_shm_base_address_pg,
119 	    1, NULL, NULL, NULL, 1, VM_NOSLEEP, IPL_VM);
120 	if (xen_shm_arena == NULL) {
121 		panic("xen_shm_init no arena");
122 	}
123 }
124 
125 int
126 #ifdef XEN3
127 xen_shm_map(int nentries, int domid, grant_ref_t *grefp, vaddr_t *vap,
128     grant_handle_t *handlep, int flags)
129 #else
130 xen_shm_map(paddr_t *ma, int nentries, int domid, vaddr_t *vap, int flags)
131 #endif
132 {
133 	int s, i;
134 	vaddr_t new_va;
135 	u_long new_va_pg;
136 #ifdef XEN3
137 	int err;
138 	gnttab_map_grant_ref_t op[XENSHM_MAX_PAGES_PER_REQUEST];
139 #else
140 	multicall_entry_t mcl[XENSHM_MAX_PAGES_PER_REQUEST];
141 	int remap_prot = PG_V | PG_RW | PG_U | PG_M;
142 #endif
143 
144 #ifdef DIAGNOSTIC
145 	if (nentries > XENSHM_MAX_PAGES_PER_REQUEST) {
146 		printf("xen_shm_map: %d entries\n", nentries);
147 		panic("xen_shm_map");
148 	}
149 #endif
150 	s = splvm(); /* splvm is the lowest level blocking disk and net IRQ */
151 	/*
152 	 * if a driver is waiting for ressources, don't try to allocate
153 	 * yet. This is to avoid a flood of small requests stalling large
154 	 * ones.
155 	 */
156 	if (__predict_false(SIMPLEQ_FIRST(&xen_shm_callbacks) != NULL) &&
157 	    (flags & XSHM_CALLBACK) == 0) {
158 #ifdef DEBUG
159 		static struct timeval lasttime;
160 #endif
161 		splx(s);
162 #ifdef DEBUG
163 		if (ratecheck(&lasttime, &xen_shm_errintvl))
164 			printf("xen_shm_map: ENOMEM1\n");
165 #endif
166 		return ENOMEM;
167 	}
168 	/* allocate the needed virtual space */
169 	new_va_pg = vmem_alloc(xen_shm_arena, nentries,
170 	    VM_INSTANTFIT | VM_NOSLEEP);
171 	if (new_va_pg == 0) {
172 #ifdef DEBUG
173 		static struct timeval lasttime;
174 #endif
175 		splx(s);
176 #ifdef DEBUG
177 		if (ratecheck(&lasttime, &xen_shm_errintvl))
178 			printf("xen_shm_map: ENOMEM\n");
179 #endif
180 		return ENOMEM;
181 	}
182 	splx(s);
183 
184 	new_va = new_va_pg << PAGE_SHIFT;
185 #ifdef XEN3
186 	for (i = 0; i < nentries; i++) {
187 		op[i].host_addr = new_va + i * PAGE_SIZE;
188 		op[i].dom = domid;
189 		op[i].ref = grefp[i];
190 		op[i].flags = GNTMAP_host_map |
191 		    ((flags & XSHM_RO) ? GNTMAP_readonly : 0);
192 	}
193 	err = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, op, nentries);
194 	if (__predict_false(err))
195 		panic("xen_shm_map: HYPERVISOR_grant_table_op failed");
196 	for (i = 0; i < nentries; i++) {
197 		if (__predict_false(op[i].status))
198 			return op[i].status;
199 		handlep[i] = op[i].handle;
200 	}
201 #else /* !XEN3 */
202 	for (i = 0; i < nentries; i++, new_va_pg++) {
203 		mcl[i].op = __HYPERVISOR_update_va_mapping_otherdomain;
204 		mcl[i].args[0] = new_va_pg;
205 		mcl[i].args[1] = ma[i] | remap_prot;
206 		mcl[i].args[2] = 0;
207 		mcl[i].args[3] = domid;
208 	}
209 	if (HYPERVISOR_multicall(mcl, nentries) != 0)
210 	    panic("xen_shm_map: HYPERVISOR_multicall");
211 
212 	for (i = 0; i < nentries; i++) {
213 		if ((mcl[i].args[5] != 0)) {
214 			printf("xen_shm_map: mcl[%d] failed\n", i);
215 			xen_shm_unmap(new_va, ma, nentries, domid);
216 			return EINVAL;
217 		}
218 	}
219 #endif /* !XEN3 */
220 	*vap = new_va;
221 	return 0;
222 }
223 
224 void
225 #ifdef XEN3
226 xen_shm_unmap(vaddr_t va, int nentries, grant_handle_t *handlep)
227 #else
228 xen_shm_unmap(vaddr_t va, paddr_t *pa, int nentries, int domid)
229 #endif
230 {
231 #ifdef XEN3
232 	gnttab_unmap_grant_ref_t op[XENSHM_MAX_PAGES_PER_REQUEST];
233 	int ret;
234 #else
235 	multicall_entry_t mcl[XENSHM_MAX_PAGES_PER_REQUEST];
236 #endif
237 	int i;
238 	int s;
239 	struct xen_shm_callback_entry *xshmc;
240 
241 #ifdef DIAGNOSTIC
242 	if (nentries > XENSHM_MAX_PAGES_PER_REQUEST) {
243 		printf("xen_shm_unmap: %d entries\n", nentries);
244 		panic("xen_shm_unmap");
245 	}
246 #endif
247 
248 #ifdef XEN3
249 	for (i = 0; i < nentries; i++) {
250 		op[i].host_addr = va + i * PAGE_SIZE;
251 		op[i].dev_bus_addr = 0;
252 		op[i].handle = handlep[i];
253 	}
254 	ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
255 	    op, nentries);
256 	if (__predict_false(ret))
257 		panic("xen_shm_unmap: unmap failed");
258 	va = va >> PAGE_SHIFT;
259 #else /* !XEN3 */
260 	va = va >> PAGE_SHIFT;
261 	for (i = 0; i < nentries; i++) {
262 		mcl[i].op = __HYPERVISOR_update_va_mapping;
263 		mcl[i].args[0] = va + i;
264 		mcl[i].args[1] = 0;
265 		mcl[i].args[2] = 0;
266 	}
267 	mcl[nentries - 1].args[2] = UVMF_FLUSH_TLB;
268 	if (HYPERVISOR_multicall(mcl, nentries) != 0)
269 		panic("xen_shm_unmap");
270 #endif /* !XEN3 */
271 	s = splvm(); /* splvm is the lowest level blocking disk and net IRQ */
272 	vmem_free(xen_shm_arena, va, nentries);
273 	while (__predict_false((xshmc = SIMPLEQ_FIRST(&xen_shm_callbacks))
274 	    != NULL)) {
275 		SIMPLEQ_REMOVE_HEAD(&xen_shm_callbacks, xshmc_entries);
276 		splx(s);
277 		if (xshmc->xshmc_callback(xshmc->xshmc_arg) == 0) {
278 			/* callback succeeded */
279 			s = splvm();
280 			pool_put(&xen_shm_callback_pool, xshmc);
281 		} else {
282 			/* callback failed, probably out of ressources */
283 			s = splvm();
284 			SIMPLEQ_INSERT_TAIL(&xen_shm_callbacks, xshmc,
285 					    xshmc_entries);
286 
287 			break;
288 		}
289 	}
290 	splx(s);
291 }
292 
293 int
294 xen_shm_callback(int (*callback)(void *), void *arg)
295 {
296 	struct xen_shm_callback_entry *xshmc;
297 	int s;
298 
299 	s = splvm();
300 	xshmc = pool_get(&xen_shm_callback_pool, PR_NOWAIT);
301 	if (xshmc == NULL) {
302 		splx(s);
303 		return ENOMEM;
304 	}
305 	xshmc->xshmc_arg = arg;
306 	xshmc->xshmc_callback = callback;
307 	SIMPLEQ_INSERT_TAIL(&xen_shm_callbacks, xshmc, xshmc_entries);
308 	splx(s);
309 	return 0;
310 }
311