]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2011 Citrix Ltd. | |
3 | * | |
4 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
5 | * the COPYING file in the top-level directory. | |
6 | * | |
7 | * Contributions after 2012-01-13 are licensed under the terms of the | |
8 | * GNU GPL, version 2 or (at your option) any later version. | |
9 | */ | |
10 | ||
11 | #include "config.h" | |
12 | ||
13 | #include <sys/resource.h> | |
14 | ||
15 | #include "hw/xen_backend.h" | |
16 | #include "sysemu/blockdev.h" | |
17 | #include "qemu/bitmap.h" | |
18 | ||
19 | #include <xen/hvm/params.h> | |
20 | #include <sys/mman.h> | |
21 | ||
22 | #include "sysemu/xen-mapcache.h" | |
23 | #include "trace.h" | |
24 | ||
25 | ||
26 | //#define MAPCACHE_DEBUG | |
27 | ||
28 | #ifdef MAPCACHE_DEBUG | |
29 | # define DPRINTF(fmt, ...) do { \ | |
30 | fprintf(stderr, "xen_mapcache: " fmt, ## __VA_ARGS__); \ | |
31 | } while (0) | |
32 | #else | |
33 | # define DPRINTF(fmt, ...) do { } while (0) | |
34 | #endif | |
35 | ||
36 | #if defined(__i386__) | |
37 | # define MCACHE_BUCKET_SHIFT 16 | |
38 | # define MCACHE_MAX_SIZE (1UL<<31) /* 2GB Cap */ | |
39 | #elif defined(__x86_64__) | |
40 | # define MCACHE_BUCKET_SHIFT 20 | |
41 | # define MCACHE_MAX_SIZE (1UL<<35) /* 32GB Cap */ | |
42 | #endif | |
43 | #define MCACHE_BUCKET_SIZE (1UL << MCACHE_BUCKET_SHIFT) | |
44 | ||
45 | /* This is the size of the virtual address space reserve to QEMU that will not | |
46 | * be use by MapCache. | |
47 | * From empirical tests I observed that qemu use 75MB more than the | |
48 | * max_mcache_size. | |
49 | */ | |
50 | #define NON_MCACHE_MEMORY_SIZE (80 * 1024 * 1024) | |
51 | ||
52 | #define mapcache_lock() ((void)0) | |
53 | #define mapcache_unlock() ((void)0) | |
54 | ||
55 | typedef struct MapCacheEntry { | |
56 | hwaddr paddr_index; | |
57 | uint8_t *vaddr_base; | |
58 | unsigned long *valid_mapping; | |
59 | uint8_t lock; | |
60 | hwaddr size; | |
61 | struct MapCacheEntry *next; | |
62 | } MapCacheEntry; | |
63 | ||
64 | typedef struct MapCacheRev { | |
65 | uint8_t *vaddr_req; | |
66 | hwaddr paddr_index; | |
67 | hwaddr size; | |
68 | QTAILQ_ENTRY(MapCacheRev) next; | |
69 | } MapCacheRev; | |
70 | ||
71 | typedef struct MapCache { | |
72 | MapCacheEntry *entry; | |
73 | unsigned long nr_buckets; | |
74 | QTAILQ_HEAD(map_cache_head, MapCacheRev) locked_entries; | |
75 | ||
76 | /* For most cases (>99.9%), the page address is the same. */ | |
77 | hwaddr last_address_index; | |
78 | uint8_t *last_address_vaddr; | |
79 | unsigned long max_mcache_size; | |
80 | unsigned int mcache_bucket_shift; | |
81 | ||
82 | phys_offset_to_gaddr_t phys_offset_to_gaddr; | |
83 | void *opaque; | |
84 | } MapCache; | |
85 | ||
86 | static MapCache *mapcache; | |
87 | ||
88 | static inline int test_bits(int nr, int size, const unsigned long *addr) | |
89 | { | |
90 | unsigned long res = find_next_zero_bit(addr, size + nr, nr); | |
91 | if (res >= nr + size) | |
92 | return 1; | |
93 | else | |
94 | return 0; | |
95 | } | |
96 | ||
97 | void xen_map_cache_init(phys_offset_to_gaddr_t f, void *opaque) | |
98 | { | |
99 | unsigned long size; | |
100 | struct rlimit rlimit_as; | |
101 | ||
102 | mapcache = g_malloc0(sizeof (MapCache)); | |
103 | ||
104 | mapcache->phys_offset_to_gaddr = f; | |
105 | mapcache->opaque = opaque; | |
106 | ||
107 | QTAILQ_INIT(&mapcache->locked_entries); | |
108 | mapcache->last_address_index = -1; | |
109 | ||
110 | if (geteuid() == 0) { | |
111 | rlimit_as.rlim_cur = RLIM_INFINITY; | |
112 | rlimit_as.rlim_max = RLIM_INFINITY; | |
113 | mapcache->max_mcache_size = MCACHE_MAX_SIZE; | |
114 | } else { | |
115 | getrlimit(RLIMIT_AS, &rlimit_as); | |
116 | rlimit_as.rlim_cur = rlimit_as.rlim_max; | |
117 | ||
118 | if (rlimit_as.rlim_max != RLIM_INFINITY) { | |
119 | fprintf(stderr, "Warning: QEMU's maximum size of virtual" | |
120 | " memory is not infinity.\n"); | |
121 | } | |
122 | if (rlimit_as.rlim_max < MCACHE_MAX_SIZE + NON_MCACHE_MEMORY_SIZE) { | |
123 | mapcache->max_mcache_size = rlimit_as.rlim_max - | |
124 | NON_MCACHE_MEMORY_SIZE; | |
125 | } else { | |
126 | mapcache->max_mcache_size = MCACHE_MAX_SIZE; | |
127 | } | |
128 | } | |
129 | ||
130 | setrlimit(RLIMIT_AS, &rlimit_as); | |
131 | ||
132 | mapcache->nr_buckets = | |
133 | (((mapcache->max_mcache_size >> XC_PAGE_SHIFT) + | |
134 | (1UL << (MCACHE_BUCKET_SHIFT - XC_PAGE_SHIFT)) - 1) >> | |
135 | (MCACHE_BUCKET_SHIFT - XC_PAGE_SHIFT)); | |
136 | ||
137 | size = mapcache->nr_buckets * sizeof (MapCacheEntry); | |
138 | size = (size + XC_PAGE_SIZE - 1) & ~(XC_PAGE_SIZE - 1); | |
139 | DPRINTF("%s, nr_buckets = %lx size %lu\n", __func__, | |
140 | mapcache->nr_buckets, size); | |
141 | mapcache->entry = g_malloc0(size); | |
142 | } | |
143 | ||
144 | static void xen_remap_bucket(MapCacheEntry *entry, | |
145 | hwaddr size, | |
146 | hwaddr address_index) | |
147 | { | |
148 | uint8_t *vaddr_base; | |
149 | xen_pfn_t *pfns; | |
150 | int *err; | |
151 | unsigned int i; | |
152 | hwaddr nb_pfn = size >> XC_PAGE_SHIFT; | |
153 | ||
154 | trace_xen_remap_bucket(address_index); | |
155 | ||
156 | pfns = g_malloc0(nb_pfn * sizeof (xen_pfn_t)); | |
157 | err = g_malloc0(nb_pfn * sizeof (int)); | |
158 | ||
159 | if (entry->vaddr_base != NULL) { | |
160 | if (munmap(entry->vaddr_base, entry->size) != 0) { | |
161 | perror("unmap fails"); | |
162 | exit(-1); | |
163 | } | |
164 | } | |
165 | if (entry->valid_mapping != NULL) { | |
166 | g_free(entry->valid_mapping); | |
167 | entry->valid_mapping = NULL; | |
168 | } | |
169 | ||
170 | for (i = 0; i < nb_pfn; i++) { | |
171 | pfns[i] = (address_index << (MCACHE_BUCKET_SHIFT-XC_PAGE_SHIFT)) + i; | |
172 | } | |
173 | ||
174 | vaddr_base = xc_map_foreign_bulk(xen_xc, xen_domid, PROT_READ|PROT_WRITE, | |
175 | pfns, err, nb_pfn); | |
176 | if (vaddr_base == NULL) { | |
177 | perror("xc_map_foreign_bulk"); | |
178 | exit(-1); | |
179 | } | |
180 | ||
181 | entry->vaddr_base = vaddr_base; | |
182 | entry->paddr_index = address_index; | |
183 | entry->size = size; | |
184 | entry->valid_mapping = (unsigned long *) g_malloc0(sizeof(unsigned long) * | |
185 | BITS_TO_LONGS(size >> XC_PAGE_SHIFT)); | |
186 | ||
187 | bitmap_zero(entry->valid_mapping, nb_pfn); | |
188 | for (i = 0; i < nb_pfn; i++) { | |
189 | if (!err[i]) { | |
190 | bitmap_set(entry->valid_mapping, i, 1); | |
191 | } | |
192 | } | |
193 | ||
194 | g_free(pfns); | |
195 | g_free(err); | |
196 | } | |
197 | ||
198 | uint8_t *xen_map_cache(hwaddr phys_addr, hwaddr size, | |
199 | uint8_t lock) | |
200 | { | |
201 | MapCacheEntry *entry, *pentry = NULL; | |
202 | hwaddr address_index; | |
203 | hwaddr address_offset; | |
204 | hwaddr __size = size; | |
205 | bool translated = false; | |
206 | ||
207 | tryagain: | |
208 | address_index = phys_addr >> MCACHE_BUCKET_SHIFT; | |
209 | address_offset = phys_addr & (MCACHE_BUCKET_SIZE - 1); | |
210 | ||
211 | trace_xen_map_cache(phys_addr); | |
212 | ||
213 | if (address_index == mapcache->last_address_index && !lock && !__size) { | |
214 | trace_xen_map_cache_return(mapcache->last_address_vaddr + address_offset); | |
215 | return mapcache->last_address_vaddr + address_offset; | |
216 | } | |
217 | ||
218 | /* size is always a multiple of MCACHE_BUCKET_SIZE */ | |
219 | if (size) { | |
220 | __size = size + address_offset; | |
221 | if (__size % MCACHE_BUCKET_SIZE) { | |
222 | __size += MCACHE_BUCKET_SIZE - (__size % MCACHE_BUCKET_SIZE); | |
223 | } | |
224 | } else { | |
225 | __size = MCACHE_BUCKET_SIZE; | |
226 | } | |
227 | ||
228 | entry = &mapcache->entry[address_index % mapcache->nr_buckets]; | |
229 | ||
230 | while (entry && entry->lock && entry->vaddr_base && | |
231 | (entry->paddr_index != address_index || entry->size != __size || | |
232 | !test_bits(address_offset >> XC_PAGE_SHIFT, size >> XC_PAGE_SHIFT, | |
233 | entry->valid_mapping))) { | |
234 | pentry = entry; | |
235 | entry = entry->next; | |
236 | } | |
237 | if (!entry) { | |
238 | entry = g_malloc0(sizeof (MapCacheEntry)); | |
239 | pentry->next = entry; | |
240 | xen_remap_bucket(entry, __size, address_index); | |
241 | } else if (!entry->lock) { | |
242 | if (!entry->vaddr_base || entry->paddr_index != address_index || | |
243 | entry->size != __size || | |
244 | !test_bits(address_offset >> XC_PAGE_SHIFT, size >> XC_PAGE_SHIFT, | |
245 | entry->valid_mapping)) { | |
246 | xen_remap_bucket(entry, __size, address_index); | |
247 | } | |
248 | } | |
249 | ||
250 | if(!test_bits(address_offset >> XC_PAGE_SHIFT, size >> XC_PAGE_SHIFT, | |
251 | entry->valid_mapping)) { | |
252 | mapcache->last_address_index = -1; | |
253 | if (!translated && mapcache->phys_offset_to_gaddr) { | |
254 | phys_addr = mapcache->phys_offset_to_gaddr(phys_addr, size, mapcache->opaque); | |
255 | translated = true; | |
256 | goto tryagain; | |
257 | } | |
258 | trace_xen_map_cache_return(NULL); | |
259 | return NULL; | |
260 | } | |
261 | ||
262 | mapcache->last_address_index = address_index; | |
263 | mapcache->last_address_vaddr = entry->vaddr_base; | |
264 | if (lock) { | |
265 | MapCacheRev *reventry = g_malloc0(sizeof(MapCacheRev)); | |
266 | entry->lock++; | |
267 | reventry->vaddr_req = mapcache->last_address_vaddr + address_offset; | |
268 | reventry->paddr_index = mapcache->last_address_index; | |
269 | reventry->size = entry->size; | |
270 | QTAILQ_INSERT_HEAD(&mapcache->locked_entries, reventry, next); | |
271 | } | |
272 | ||
273 | trace_xen_map_cache_return(mapcache->last_address_vaddr + address_offset); | |
274 | return mapcache->last_address_vaddr + address_offset; | |
275 | } | |
276 | ||
277 | ram_addr_t xen_ram_addr_from_mapcache(void *ptr) | |
278 | { | |
279 | MapCacheEntry *entry = NULL; | |
280 | MapCacheRev *reventry; | |
281 | hwaddr paddr_index; | |
282 | hwaddr size; | |
283 | int found = 0; | |
284 | ||
285 | QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) { | |
286 | if (reventry->vaddr_req == ptr) { | |
287 | paddr_index = reventry->paddr_index; | |
288 | size = reventry->size; | |
289 | found = 1; | |
290 | break; | |
291 | } | |
292 | } | |
293 | if (!found) { | |
294 | fprintf(stderr, "%s, could not find %p\n", __func__, ptr); | |
295 | QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) { | |
296 | DPRINTF(" "TARGET_FMT_plx" -> %p is present\n", reventry->paddr_index, | |
297 | reventry->vaddr_req); | |
298 | } | |
299 | abort(); | |
300 | return 0; | |
301 | } | |
302 | ||
303 | entry = &mapcache->entry[paddr_index % mapcache->nr_buckets]; | |
304 | while (entry && (entry->paddr_index != paddr_index || entry->size != size)) { | |
305 | entry = entry->next; | |
306 | } | |
307 | if (!entry) { | |
308 | DPRINTF("Trying to find address %p that is not in the mapcache!\n", ptr); | |
309 | return 0; | |
310 | } | |
311 | return (reventry->paddr_index << MCACHE_BUCKET_SHIFT) + | |
312 | ((unsigned long) ptr - (unsigned long) entry->vaddr_base); | |
313 | } | |
314 | ||
315 | void xen_invalidate_map_cache_entry(uint8_t *buffer) | |
316 | { | |
317 | MapCacheEntry *entry = NULL, *pentry = NULL; | |
318 | MapCacheRev *reventry; | |
319 | hwaddr paddr_index; | |
320 | hwaddr size; | |
321 | int found = 0; | |
322 | ||
323 | QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) { | |
324 | if (reventry->vaddr_req == buffer) { | |
325 | paddr_index = reventry->paddr_index; | |
326 | size = reventry->size; | |
327 | found = 1; | |
328 | break; | |
329 | } | |
330 | } | |
331 | if (!found) { | |
332 | DPRINTF("%s, could not find %p\n", __func__, buffer); | |
333 | QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) { | |
334 | DPRINTF(" "TARGET_FMT_plx" -> %p is present\n", reventry->paddr_index, reventry->vaddr_req); | |
335 | } | |
336 | return; | |
337 | } | |
338 | QTAILQ_REMOVE(&mapcache->locked_entries, reventry, next); | |
339 | g_free(reventry); | |
340 | ||
341 | if (mapcache->last_address_index == paddr_index) { | |
342 | mapcache->last_address_index = -1; | |
343 | mapcache->last_address_vaddr = NULL; | |
344 | } | |
345 | ||
346 | entry = &mapcache->entry[paddr_index % mapcache->nr_buckets]; | |
347 | while (entry && (entry->paddr_index != paddr_index || entry->size != size)) { | |
348 | pentry = entry; | |
349 | entry = entry->next; | |
350 | } | |
351 | if (!entry) { | |
352 | DPRINTF("Trying to unmap address %p that is not in the mapcache!\n", buffer); | |
353 | return; | |
354 | } | |
355 | entry->lock--; | |
356 | if (entry->lock > 0 || pentry == NULL) { | |
357 | return; | |
358 | } | |
359 | ||
360 | pentry->next = entry->next; | |
361 | if (munmap(entry->vaddr_base, entry->size) != 0) { | |
362 | perror("unmap fails"); | |
363 | exit(-1); | |
364 | } | |
365 | g_free(entry->valid_mapping); | |
366 | g_free(entry); | |
367 | } | |
368 | ||
369 | void xen_invalidate_map_cache(void) | |
370 | { | |
371 | unsigned long i; | |
372 | MapCacheRev *reventry; | |
373 | ||
374 | /* Flush pending AIO before destroying the mapcache */ | |
375 | bdrv_drain_all(); | |
376 | ||
377 | QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) { | |
378 | DPRINTF("There should be no locked mappings at this time, " | |
379 | "but "TARGET_FMT_plx" -> %p is present\n", | |
380 | reventry->paddr_index, reventry->vaddr_req); | |
381 | } | |
382 | ||
383 | mapcache_lock(); | |
384 | ||
385 | for (i = 0; i < mapcache->nr_buckets; i++) { | |
386 | MapCacheEntry *entry = &mapcache->entry[i]; | |
387 | ||
388 | if (entry->vaddr_base == NULL) { | |
389 | continue; | |
390 | } | |
391 | if (entry->lock > 0) { | |
392 | continue; | |
393 | } | |
394 | ||
395 | if (munmap(entry->vaddr_base, entry->size) != 0) { | |
396 | perror("unmap fails"); | |
397 | exit(-1); | |
398 | } | |
399 | ||
400 | entry->paddr_index = 0; | |
401 | entry->vaddr_base = NULL; | |
402 | entry->size = 0; | |
403 | g_free(entry->valid_mapping); | |
404 | entry->valid_mapping = NULL; | |
405 | } | |
406 | ||
407 | mapcache->last_address_index = -1; | |
408 | mapcache->last_address_vaddr = NULL; | |
409 | ||
410 | mapcache_unlock(); | |
411 | } |