Commit | Line | Data |
---|---|---|
d7e28ffe RR |
1 | /* Simple I/O model for guests, based on shared memory. |
2 | * Copyright (C) 2006 Rusty Russell IBM Corporation | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
17 | */ | |
18 | #include <linux/types.h> | |
19 | #include <linux/futex.h> | |
20 | #include <linux/jhash.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/highmem.h> | |
23 | #include <linux/uaccess.h> | |
24 | #include "lg.h" | |
25 | ||
26 | static struct list_head dma_hash[61]; | |
27 | ||
28 | void lguest_io_init(void) | |
29 | { | |
30 | unsigned int i; | |
31 | ||
32 | for (i = 0; i < ARRAY_SIZE(dma_hash); i++) | |
33 | INIT_LIST_HEAD(&dma_hash[i]); | |
34 | } | |
35 | ||
36 | /* FIXME: allow multi-page lengths. */ | |
37 | static int check_dma_list(struct lguest *lg, const struct lguest_dma *dma) | |
38 | { | |
39 | unsigned int i; | |
40 | ||
41 | for (i = 0; i < LGUEST_MAX_DMA_SECTIONS; i++) { | |
42 | if (!dma->len[i]) | |
43 | return 1; | |
44 | if (!lguest_address_ok(lg, dma->addr[i], dma->len[i])) | |
45 | goto kill; | |
46 | if (dma->len[i] > PAGE_SIZE) | |
47 | goto kill; | |
48 | /* We could do over a page, but is it worth it? */ | |
49 | if ((dma->addr[i] % PAGE_SIZE) + dma->len[i] > PAGE_SIZE) | |
50 | goto kill; | |
51 | } | |
52 | return 1; | |
53 | ||
54 | kill: | |
55 | kill_guest(lg, "bad DMA entry: %u@%#lx", dma->len[i], dma->addr[i]); | |
56 | return 0; | |
57 | } | |
58 | ||
59 | static unsigned int hash(const union futex_key *key) | |
60 | { | |
61 | return jhash2((u32*)&key->both.word, | |
62 | (sizeof(key->both.word)+sizeof(key->both.ptr))/4, | |
63 | key->both.offset) | |
64 | % ARRAY_SIZE(dma_hash); | |
65 | } | |
66 | ||
67 | static inline int key_eq(const union futex_key *a, const union futex_key *b) | |
68 | { | |
69 | return (a->both.word == b->both.word | |
70 | && a->both.ptr == b->both.ptr | |
71 | && a->both.offset == b->both.offset); | |
72 | } | |
73 | ||
74 | /* Must hold read lock on dmainfo owner's current->mm->mmap_sem */ | |
75 | static void unlink_dma(struct lguest_dma_info *dmainfo) | |
76 | { | |
77 | BUG_ON(!mutex_is_locked(&lguest_lock)); | |
78 | dmainfo->interrupt = 0; | |
79 | list_del(&dmainfo->list); | |
80 | drop_futex_key_refs(&dmainfo->key); | |
81 | } | |
82 | ||
83 | static int unbind_dma(struct lguest *lg, | |
84 | const union futex_key *key, | |
85 | unsigned long dmas) | |
86 | { | |
87 | int i, ret = 0; | |
88 | ||
89 | for (i = 0; i < LGUEST_MAX_DMA; i++) { | |
90 | if (key_eq(key, &lg->dma[i].key) && dmas == lg->dma[i].dmas) { | |
91 | unlink_dma(&lg->dma[i]); | |
92 | ret = 1; | |
93 | break; | |
94 | } | |
95 | } | |
96 | return ret; | |
97 | } | |
98 | ||
99 | int bind_dma(struct lguest *lg, | |
100 | unsigned long ukey, unsigned long dmas, u16 numdmas, u8 interrupt) | |
101 | { | |
102 | unsigned int i; | |
103 | int ret = 0; | |
104 | union futex_key key; | |
105 | struct rw_semaphore *fshared = ¤t->mm->mmap_sem; | |
106 | ||
107 | if (interrupt >= LGUEST_IRQS) | |
108 | return 0; | |
109 | ||
110 | mutex_lock(&lguest_lock); | |
111 | down_read(fshared); | |
112 | if (get_futex_key((u32 __user *)ukey, fshared, &key) != 0) { | |
113 | kill_guest(lg, "bad dma key %#lx", ukey); | |
114 | goto unlock; | |
115 | } | |
116 | get_futex_key_refs(&key); | |
117 | ||
118 | if (interrupt == 0) | |
119 | ret = unbind_dma(lg, &key, dmas); | |
120 | else { | |
121 | for (i = 0; i < LGUEST_MAX_DMA; i++) { | |
122 | if (lg->dma[i].interrupt) | |
123 | continue; | |
124 | ||
125 | lg->dma[i].dmas = dmas; | |
126 | lg->dma[i].num_dmas = numdmas; | |
127 | lg->dma[i].next_dma = 0; | |
128 | lg->dma[i].key = key; | |
129 | lg->dma[i].guestid = lg->guestid; | |
130 | lg->dma[i].interrupt = interrupt; | |
131 | list_add(&lg->dma[i].list, &dma_hash[hash(&key)]); | |
132 | ret = 1; | |
133 | goto unlock; | |
134 | } | |
135 | } | |
136 | drop_futex_key_refs(&key); | |
137 | unlock: | |
138 | up_read(fshared); | |
139 | mutex_unlock(&lguest_lock); | |
140 | return ret; | |
141 | } | |
142 | ||
143 | /* lgread from another guest */ | |
144 | static int lgread_other(struct lguest *lg, | |
145 | void *buf, u32 addr, unsigned bytes) | |
146 | { | |
147 | if (!lguest_address_ok(lg, addr, bytes) | |
148 | || access_process_vm(lg->tsk, addr, buf, bytes, 0) != bytes) { | |
149 | memset(buf, 0, bytes); | |
150 | kill_guest(lg, "bad address in registered DMA struct"); | |
151 | return 0; | |
152 | } | |
153 | return 1; | |
154 | } | |
155 | ||
156 | /* lgwrite to another guest */ | |
157 | static int lgwrite_other(struct lguest *lg, u32 addr, | |
158 | const void *buf, unsigned bytes) | |
159 | { | |
160 | if (!lguest_address_ok(lg, addr, bytes) | |
161 | || (access_process_vm(lg->tsk, addr, (void *)buf, bytes, 1) | |
162 | != bytes)) { | |
163 | kill_guest(lg, "bad address writing to registered DMA"); | |
164 | return 0; | |
165 | } | |
166 | return 1; | |
167 | } | |
168 | ||
169 | static u32 copy_data(struct lguest *srclg, | |
170 | const struct lguest_dma *src, | |
171 | const struct lguest_dma *dst, | |
172 | struct page *pages[]) | |
173 | { | |
174 | unsigned int totlen, si, di, srcoff, dstoff; | |
175 | void *maddr = NULL; | |
176 | ||
177 | totlen = 0; | |
178 | si = di = 0; | |
179 | srcoff = dstoff = 0; | |
180 | while (si < LGUEST_MAX_DMA_SECTIONS && src->len[si] | |
181 | && di < LGUEST_MAX_DMA_SECTIONS && dst->len[di]) { | |
182 | u32 len = min(src->len[si] - srcoff, dst->len[di] - dstoff); | |
183 | ||
184 | if (!maddr) | |
185 | maddr = kmap(pages[di]); | |
186 | ||
187 | /* FIXME: This is not completely portable, since | |
188 | archs do different things for copy_to_user_page. */ | |
189 | if (copy_from_user(maddr + (dst->addr[di] + dstoff)%PAGE_SIZE, | |
6d14bfe7 | 190 | (void __user *)src->addr[si], len) != 0) { |
d7e28ffe RR |
191 | kill_guest(srclg, "bad address in sending DMA"); |
192 | totlen = 0; | |
193 | break; | |
194 | } | |
195 | ||
196 | totlen += len; | |
197 | srcoff += len; | |
198 | dstoff += len; | |
199 | if (srcoff == src->len[si]) { | |
200 | si++; | |
201 | srcoff = 0; | |
202 | } | |
203 | if (dstoff == dst->len[di]) { | |
204 | kunmap(pages[di]); | |
205 | maddr = NULL; | |
206 | di++; | |
207 | dstoff = 0; | |
208 | } | |
209 | } | |
210 | ||
211 | if (maddr) | |
212 | kunmap(pages[di]); | |
213 | ||
214 | return totlen; | |
215 | } | |
216 | ||
217 | /* Src is us, ie. current. */ | |
218 | static u32 do_dma(struct lguest *srclg, const struct lguest_dma *src, | |
219 | struct lguest *dstlg, const struct lguest_dma *dst) | |
220 | { | |
221 | int i; | |
222 | u32 ret; | |
223 | struct page *pages[LGUEST_MAX_DMA_SECTIONS]; | |
224 | ||
225 | if (!check_dma_list(dstlg, dst) || !check_dma_list(srclg, src)) | |
226 | return 0; | |
227 | ||
228 | /* First get the destination pages */ | |
229 | for (i = 0; i < LGUEST_MAX_DMA_SECTIONS; i++) { | |
230 | if (dst->len[i] == 0) | |
231 | break; | |
232 | if (get_user_pages(dstlg->tsk, dstlg->mm, | |
233 | dst->addr[i], 1, 1, 1, pages+i, NULL) | |
234 | != 1) { | |
235 | kill_guest(dstlg, "Error mapping DMA pages"); | |
236 | ret = 0; | |
237 | goto drop_pages; | |
238 | } | |
239 | } | |
240 | ||
241 | /* Now copy until we run out of src or dst. */ | |
242 | ret = copy_data(srclg, src, dst, pages); | |
243 | ||
244 | drop_pages: | |
245 | while (--i >= 0) | |
246 | put_page(pages[i]); | |
247 | return ret; | |
248 | } | |
249 | ||
250 | static int dma_transfer(struct lguest *srclg, | |
251 | unsigned long udma, | |
252 | struct lguest_dma_info *dst) | |
253 | { | |
254 | struct lguest_dma dst_dma, src_dma; | |
255 | struct lguest *dstlg; | |
256 | u32 i, dma = 0; | |
257 | ||
258 | dstlg = &lguests[dst->guestid]; | |
259 | /* Get our dma list. */ | |
260 | lgread(srclg, &src_dma, udma, sizeof(src_dma)); | |
261 | ||
262 | /* We can't deadlock against them dmaing to us, because this | |
263 | * is all under the lguest_lock. */ | |
264 | down_read(&dstlg->mm->mmap_sem); | |
265 | ||
266 | for (i = 0; i < dst->num_dmas; i++) { | |
267 | dma = (dst->next_dma + i) % dst->num_dmas; | |
268 | if (!lgread_other(dstlg, &dst_dma, | |
269 | dst->dmas + dma * sizeof(struct lguest_dma), | |
270 | sizeof(dst_dma))) { | |
271 | goto fail; | |
272 | } | |
273 | if (!dst_dma.used_len) | |
274 | break; | |
275 | } | |
276 | if (i != dst->num_dmas) { | |
277 | unsigned long used_lenp; | |
278 | unsigned int ret; | |
279 | ||
280 | ret = do_dma(srclg, &src_dma, dstlg, &dst_dma); | |
281 | /* Put used length in src. */ | |
282 | lgwrite_u32(srclg, | |
283 | udma+offsetof(struct lguest_dma, used_len), ret); | |
284 | if (ret == 0 && src_dma.len[0] != 0) | |
285 | goto fail; | |
286 | ||
287 | /* Make sure destination sees contents before length. */ | |
288 | wmb(); | |
289 | used_lenp = dst->dmas | |
290 | + dma * sizeof(struct lguest_dma) | |
291 | + offsetof(struct lguest_dma, used_len); | |
292 | lgwrite_other(dstlg, used_lenp, &ret, sizeof(ret)); | |
293 | dst->next_dma++; | |
294 | } | |
295 | up_read(&dstlg->mm->mmap_sem); | |
296 | ||
297 | /* Do this last so dst doesn't simply sleep on lock. */ | |
298 | set_bit(dst->interrupt, dstlg->irqs_pending); | |
299 | wake_up_process(dstlg->tsk); | |
300 | return i == dst->num_dmas; | |
301 | ||
302 | fail: | |
303 | up_read(&dstlg->mm->mmap_sem); | |
304 | return 0; | |
305 | } | |
306 | ||
307 | void send_dma(struct lguest *lg, unsigned long ukey, unsigned long udma) | |
308 | { | |
309 | union futex_key key; | |
310 | int empty = 0; | |
311 | struct rw_semaphore *fshared = ¤t->mm->mmap_sem; | |
312 | ||
313 | again: | |
314 | mutex_lock(&lguest_lock); | |
315 | down_read(fshared); | |
316 | if (get_futex_key((u32 __user *)ukey, fshared, &key) != 0) { | |
317 | kill_guest(lg, "bad sending DMA key"); | |
318 | goto unlock; | |
319 | } | |
320 | /* Shared mapping? Look for other guests... */ | |
321 | if (key.shared.offset & 1) { | |
322 | struct lguest_dma_info *i; | |
323 | list_for_each_entry(i, &dma_hash[hash(&key)], list) { | |
324 | if (i->guestid == lg->guestid) | |
325 | continue; | |
326 | if (!key_eq(&key, &i->key)) | |
327 | continue; | |
328 | ||
329 | empty += dma_transfer(lg, udma, i); | |
330 | break; | |
331 | } | |
332 | if (empty == 1) { | |
333 | /* Give any recipients one chance to restock. */ | |
334 | up_read(¤t->mm->mmap_sem); | |
335 | mutex_unlock(&lguest_lock); | |
336 | empty++; | |
337 | goto again; | |
338 | } | |
339 | } else { | |
340 | /* Private mapping: tell our userspace. */ | |
341 | lg->dma_is_pending = 1; | |
342 | lg->pending_dma = udma; | |
343 | lg->pending_key = ukey; | |
344 | } | |
345 | unlock: | |
346 | up_read(fshared); | |
347 | mutex_unlock(&lguest_lock); | |
348 | } | |
349 | ||
350 | void release_all_dma(struct lguest *lg) | |
351 | { | |
352 | unsigned int i; | |
353 | ||
354 | BUG_ON(!mutex_is_locked(&lguest_lock)); | |
355 | ||
356 | down_read(&lg->mm->mmap_sem); | |
357 | for (i = 0; i < LGUEST_MAX_DMA; i++) { | |
358 | if (lg->dma[i].interrupt) | |
359 | unlink_dma(&lg->dma[i]); | |
360 | } | |
361 | up_read(&lg->mm->mmap_sem); | |
362 | } | |
363 | ||
364 | /* Userspace wants a dma buffer from this guest. */ | |
365 | unsigned long get_dma_buffer(struct lguest *lg, | |
366 | unsigned long ukey, unsigned long *interrupt) | |
367 | { | |
368 | unsigned long ret = 0; | |
369 | union futex_key key; | |
370 | struct lguest_dma_info *i; | |
371 | struct rw_semaphore *fshared = ¤t->mm->mmap_sem; | |
372 | ||
373 | mutex_lock(&lguest_lock); | |
374 | down_read(fshared); | |
375 | if (get_futex_key((u32 __user *)ukey, fshared, &key) != 0) { | |
376 | kill_guest(lg, "bad registered DMA buffer"); | |
377 | goto unlock; | |
378 | } | |
379 | list_for_each_entry(i, &dma_hash[hash(&key)], list) { | |
380 | if (key_eq(&key, &i->key) && i->guestid == lg->guestid) { | |
381 | unsigned int j; | |
382 | for (j = 0; j < i->num_dmas; j++) { | |
383 | struct lguest_dma dma; | |
384 | ||
385 | ret = i->dmas + j * sizeof(struct lguest_dma); | |
386 | lgread(lg, &dma, ret, sizeof(dma)); | |
387 | if (dma.used_len == 0) | |
388 | break; | |
389 | } | |
390 | *interrupt = i->interrupt; | |
391 | break; | |
392 | } | |
393 | } | |
394 | unlock: | |
395 | up_read(fshared); | |
396 | mutex_unlock(&lguest_lock); | |
397 | return ret; | |
398 | } | |
399 |