xref: /netbsd-src/sys/arch/xen/x86/xen_shm_machdep.c (revision e6c7e151de239c49d2e38720a061ed9d1fa99309)
1 /*      $NetBSD: xen_shm_machdep.c,v 1.13 2019/01/27 02:08:39 pgoyette 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 #include <sys/cdefs.h>
28 __KERNEL_RCSID(0, "$NetBSD: xen_shm_machdep.c,v 1.13 2019/01/27 02:08:39 pgoyette Exp $");
29 
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/queue.h>
34 #include <sys/vmem.h>
35 #include <sys/kernel.h>
36 #include <uvm/uvm.h>
37 
38 #include <machine/pmap.h>
39 #include <xen/hypervisor.h>
40 #include <xen/xen.h>
41 #include <xen/evtchn.h>
42 #include <xen/xen_shm.h>
43 
44 /*
45  * Helper routines for the backend drivers. This implements the necessary
46  * functions to map a bunch of pages from foreign domains into our kernel VM
47  * space, do I/O to it, and unmap it.
48  *
49  * At boot time, we grab some kernel VM space that we'll use to map the foreign
50  * pages. We also maintain a virtual-to-machine mapping table to give back
51  * the appropriate address to bus_dma if requested.
52  *
53  * If no more VM space is available, we return an error. The caller can then
54  * register a callback which will be called when the required VM space is
55  * available.
56  */
57 
58 /* Grab enough VM space to map an entire vbd ring. */
59 /* Xen3 linux guests seems to eat more pages, gives enough for 10 vbd rings */
60 #define BLKIF_RING_SIZE __RING_SIZE((blkif_sring_t *)0, PAGE_SIZE)
61 #define XENSHM_NPAGES (BLKIF_RING_SIZE * (BLKIF_MAX_SEGMENTS_PER_REQUEST + 1) * 10)
62 
63 /* vm space management */
64 static vmem_t *xen_shm_arena __read_mostly;
65 
66 /* callbacks are registered in a FIFO list. */
67 static SIMPLEQ_HEAD(xen_shm_callback_head, xen_shm_callback_entry)
68     xen_shm_callbacks;
69 
70 struct xen_shm_callback_entry {
71 	SIMPLEQ_ENTRY(xen_shm_callback_entry) xshmc_entries;
72 	int (*xshmc_callback)(void *); /* our callback */
73 	void *xshmc_arg; /* cookie passed to the callback */
74 };
75 
76 /* a pool of struct xen_shm_callback_entry */
77 static struct pool xen_shm_callback_pool;
78 
79 #ifdef DEBUG
80 /* for ratecheck(9) */
81 static struct timeval xen_shm_errintvl = { 60, 0 };  /* a minute, each */
82 #endif
83 
84 void
85 xen_shm_init(void)
86 {
87 	vaddr_t xen_shm_base_address;
88 	vaddr_t xen_shm_end_address;
89 	u_long xen_shm_base_address_pg;
90 	vsize_t xen_shm_size;
91 
92 	SIMPLEQ_INIT(&xen_shm_callbacks);
93 	pool_init(&xen_shm_callback_pool, sizeof(struct xen_shm_callback_entry),
94 	    0, 0, 0, "xshmc", NULL, IPL_VM);
95 	/* ensure we'll always get items */
96 	if (pool_prime(&xen_shm_callback_pool,
97 	    PAGE_SIZE / sizeof(struct xen_shm_callback_entry)) != 0) {
98 		panic("xen_shm_init can't prime pool");
99 	}
100 
101 	xen_shm_size = (XENSHM_NPAGES * PAGE_SIZE);
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", xen_shm_base_address_pg,
111 	    (xen_shm_end_address >> PAGE_SHIFT) - 1 - xen_shm_base_address_pg,
112 	    1, NULL, NULL, NULL, 1, VM_NOSLEEP, IPL_VM);
113 	if (xen_shm_arena == NULL) {
114 		panic("xen_shm_init no arena");
115 	}
116 }
117 
118 int
119 xen_shm_map(int nentries, int domid, grant_ref_t *grefp, vaddr_t *vap,
120     grant_handle_t *handlep, int flags)
121 {
122 	gnttab_map_grant_ref_t op[XENSHM_MAX_PAGES_PER_REQUEST];
123 	vmem_addr_t new_va_pg;
124 	vaddr_t new_va;
125 	int ret, i, s;
126 
127 #ifdef DIAGNOSTIC
128 	if (nentries > XENSHM_MAX_PAGES_PER_REQUEST) {
129 		panic("xen_shm_map: %d entries", nentries);
130 	}
131 #endif
132 
133 	/* XXXSMP */
134 	s = splvm(); /* splvm is the lowest level blocking disk and net IRQ */
135 
136 	/*
137 	 * If a driver is waiting for resources, 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 		splx(s);
144 #ifdef DEBUG
145 		static struct timeval lasttime;
146 		if (ratecheck(&lasttime, &xen_shm_errintvl))
147 			printf("xen_shm_map: ENOMEM1\n");
148 #endif
149 		return ENOMEM;
150 	}
151 
152 	/* Allocate the needed virtual space. */
153 	if (vmem_alloc(xen_shm_arena, nentries,
154 	    VM_INSTANTFIT | VM_NOSLEEP, &new_va_pg) != 0) {
155 		splx(s);
156 #ifdef DEBUG
157 		static struct timeval lasttime;
158 		if (ratecheck(&lasttime, &xen_shm_errintvl))
159 			printf("xen_shm_map: ENOMEM\n");
160 #endif
161 		return ENOMEM;
162 	}
163 	splx(s);
164 
165 	new_va = new_va_pg << PAGE_SHIFT;
166 	for (i = 0; i < nentries; i++) {
167 		op[i].host_addr = new_va + i * PAGE_SIZE;
168 		op[i].dom = domid;
169 		op[i].ref = grefp[i];
170 		op[i].flags = GNTMAP_host_map |
171 		    ((flags & XSHM_RO) ? GNTMAP_readonly : 0);
172 	}
173 
174 	ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, op, nentries);
175 	if (__predict_false(ret)) {
176 		panic("xen_shm_map: HYPERVISOR_grant_table_op failed");
177 	}
178 
179 	for (i = 0; i < nentries; i++) {
180 		if (__predict_false(op[i].status))
181 			return op[i].status;
182 		handlep[i] = op[i].handle;
183 	}
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 	struct xen_shm_callback_entry *xshmc;
194 	int ret, i, s;
195 
196 #ifdef DIAGNOSTIC
197 	if (nentries > XENSHM_MAX_PAGES_PER_REQUEST) {
198 		panic("xen_shm_unmap: %d entries", nentries);
199 	}
200 #endif
201 
202 	for (i = 0; i < nentries; i++) {
203 		op[i].host_addr = va + i * PAGE_SIZE;
204 		op[i].dev_bus_addr = 0;
205 		op[i].handle = handlep[i];
206 	}
207 
208 	ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
209 	    op, nentries);
210 	if (__predict_false(ret)) {
211 		panic("xen_shm_unmap: unmap failed");
212 	}
213 
214 	va = va >> PAGE_SHIFT;
215 
216 	/* XXXSMP */
217 	s = splvm(); /* splvm is the lowest level blocking disk and net IRQ */
218 
219 	vmem_free(xen_shm_arena, va, nentries);
220 	while (__predict_false((xshmc = SIMPLEQ_FIRST(&xen_shm_callbacks))
221 	    != NULL)) {
222 		SIMPLEQ_REMOVE_HEAD(&xen_shm_callbacks, xshmc_entries);
223 		splx(s);
224 		if (xshmc->xshmc_callback(xshmc->xshmc_arg) == 0) {
225 			/* callback succeeded */
226 			s = splvm(); /* XXXSMP */
227 			pool_put(&xen_shm_callback_pool, xshmc);
228 		} else {
229 			/* callback failed, probably out of resources */
230 			s = splvm(); /* XXXSMP */
231 			SIMPLEQ_INSERT_TAIL(&xen_shm_callbacks, xshmc,
232 			    xshmc_entries);
233 			break;
234 		}
235 	}
236 
237 	splx(s);
238 }
239 
240 int
241 xen_shm_callback(int (*callback)(void *), void *arg)
242 {
243 	struct xen_shm_callback_entry *xshmc;
244 	int s;
245 
246 	s = splvm(); /* XXXSMP */
247 	xshmc = pool_get(&xen_shm_callback_pool, PR_NOWAIT);
248 	if (xshmc == NULL) {
249 		splx(s);
250 		return ENOMEM;
251 	}
252 	xshmc->xshmc_arg = arg;
253 	xshmc->xshmc_callback = callback;
254 	SIMPLEQ_INSERT_TAIL(&xen_shm_callbacks, xshmc, xshmc_entries);
255 	splx(s);
256 	return 0;
257 }
258