Annotation of qemu/xen-mapcache.c, revision 1.1.1.2
1.1 root 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: */
8:
9: #include "config.h"
10:
11: #include <sys/resource.h>
12:
13: #include "hw/xen_backend.h"
14: #include "blockdev.h"
15: #include "bitmap.h"
16:
17: #include <xen/hvm/params.h>
18: #include <sys/mman.h>
19:
20: #include "xen-mapcache.h"
21: #include "trace.h"
22:
23:
24: //#define MAPCACHE_DEBUG
25:
26: #ifdef MAPCACHE_DEBUG
27: # define DPRINTF(fmt, ...) do { \
28: fprintf(stderr, "xen_mapcache: " fmt, ## __VA_ARGS__); \
29: } while (0)
30: #else
31: # define DPRINTF(fmt, ...) do { } while (0)
32: #endif
33:
34: #if defined(__i386__)
35: # define MCACHE_BUCKET_SHIFT 16
36: # define MCACHE_MAX_SIZE (1UL<<31) /* 2GB Cap */
37: #elif defined(__x86_64__)
38: # define MCACHE_BUCKET_SHIFT 20
39: # define MCACHE_MAX_SIZE (1UL<<35) /* 32GB Cap */
40: #endif
41: #define MCACHE_BUCKET_SIZE (1UL << MCACHE_BUCKET_SHIFT)
42:
1.1.1.2 ! root 43: /* This is the size of the virtual address space reserve to QEMU that will not
! 44: * be use by MapCache.
! 45: * From empirical tests I observed that qemu use 75MB more than the
! 46: * max_mcache_size.
! 47: */
! 48: #define NON_MCACHE_MEMORY_SIZE (80 * 1024 * 1024)
! 49:
1.1 root 50: #define mapcache_lock() ((void)0)
51: #define mapcache_unlock() ((void)0)
52:
53: typedef struct MapCacheEntry {
54: target_phys_addr_t paddr_index;
55: uint8_t *vaddr_base;
56: unsigned long *valid_mapping;
57: uint8_t lock;
58: target_phys_addr_t size;
59: struct MapCacheEntry *next;
60: } MapCacheEntry;
61:
62: typedef struct MapCacheRev {
63: uint8_t *vaddr_req;
64: target_phys_addr_t paddr_index;
65: target_phys_addr_t size;
66: QTAILQ_ENTRY(MapCacheRev) next;
67: } MapCacheRev;
68:
69: typedef struct MapCache {
70: MapCacheEntry *entry;
71: unsigned long nr_buckets;
72: QTAILQ_HEAD(map_cache_head, MapCacheRev) locked_entries;
73:
74: /* For most cases (>99.9%), the page address is the same. */
75: target_phys_addr_t last_address_index;
76: uint8_t *last_address_vaddr;
77: unsigned long max_mcache_size;
78: unsigned int mcache_bucket_shift;
79: } MapCache;
80:
81: static MapCache *mapcache;
82:
83: static inline int test_bits(int nr, int size, const unsigned long *addr)
84: {
85: unsigned long res = find_next_zero_bit(addr, size + nr, nr);
86: if (res >= nr + size)
87: return 1;
88: else
89: return 0;
90: }
91:
92: void xen_map_cache_init(void)
93: {
94: unsigned long size;
95: struct rlimit rlimit_as;
96:
1.1.1.2 ! root 97: mapcache = g_malloc0(sizeof (MapCache));
1.1 root 98:
99: QTAILQ_INIT(&mapcache->locked_entries);
100: mapcache->last_address_index = -1;
101:
1.1.1.2 ! root 102: if (geteuid() == 0) {
! 103: rlimit_as.rlim_cur = RLIM_INFINITY;
! 104: rlimit_as.rlim_max = RLIM_INFINITY;
! 105: mapcache->max_mcache_size = MCACHE_MAX_SIZE;
1.1 root 106: } else {
1.1.1.2 ! root 107: getrlimit(RLIMIT_AS, &rlimit_as);
! 108: rlimit_as.rlim_cur = rlimit_as.rlim_max;
! 109:
! 110: if (rlimit_as.rlim_max != RLIM_INFINITY) {
! 111: fprintf(stderr, "Warning: QEMU's maximum size of virtual"
! 112: " memory is not infinity.\n");
! 113: }
! 114: if (rlimit_as.rlim_max < MCACHE_MAX_SIZE + NON_MCACHE_MEMORY_SIZE) {
! 115: mapcache->max_mcache_size = rlimit_as.rlim_max -
! 116: NON_MCACHE_MEMORY_SIZE;
! 117: } else {
! 118: mapcache->max_mcache_size = MCACHE_MAX_SIZE;
! 119: }
1.1 root 120: }
121:
122: setrlimit(RLIMIT_AS, &rlimit_as);
123:
124: mapcache->nr_buckets =
125: (((mapcache->max_mcache_size >> XC_PAGE_SHIFT) +
126: (1UL << (MCACHE_BUCKET_SHIFT - XC_PAGE_SHIFT)) - 1) >>
127: (MCACHE_BUCKET_SHIFT - XC_PAGE_SHIFT));
128:
129: size = mapcache->nr_buckets * sizeof (MapCacheEntry);
130: size = (size + XC_PAGE_SIZE - 1) & ~(XC_PAGE_SIZE - 1);
131: DPRINTF("%s, nr_buckets = %lx size %lu\n", __func__,
132: mapcache->nr_buckets, size);
1.1.1.2 ! root 133: mapcache->entry = g_malloc0(size);
1.1 root 134: }
135:
136: static void xen_remap_bucket(MapCacheEntry *entry,
137: target_phys_addr_t size,
138: target_phys_addr_t address_index)
139: {
140: uint8_t *vaddr_base;
141: xen_pfn_t *pfns;
142: int *err;
143: unsigned int i;
144: target_phys_addr_t nb_pfn = size >> XC_PAGE_SHIFT;
145:
146: trace_xen_remap_bucket(address_index);
147:
1.1.1.2 ! root 148: pfns = g_malloc0(nb_pfn * sizeof (xen_pfn_t));
! 149: err = g_malloc0(nb_pfn * sizeof (int));
1.1 root 150:
151: if (entry->vaddr_base != NULL) {
152: if (munmap(entry->vaddr_base, entry->size) != 0) {
153: perror("unmap fails");
154: exit(-1);
155: }
156: }
157: if (entry->valid_mapping != NULL) {
1.1.1.2 ! root 158: g_free(entry->valid_mapping);
1.1 root 159: entry->valid_mapping = NULL;
160: }
161:
162: for (i = 0; i < nb_pfn; i++) {
163: pfns[i] = (address_index << (MCACHE_BUCKET_SHIFT-XC_PAGE_SHIFT)) + i;
164: }
165:
166: vaddr_base = xc_map_foreign_bulk(xen_xc, xen_domid, PROT_READ|PROT_WRITE,
167: pfns, err, nb_pfn);
168: if (vaddr_base == NULL) {
169: perror("xc_map_foreign_bulk");
170: exit(-1);
171: }
172:
173: entry->vaddr_base = vaddr_base;
174: entry->paddr_index = address_index;
175: entry->size = size;
1.1.1.2 ! root 176: entry->valid_mapping = (unsigned long *) g_malloc0(sizeof(unsigned long) *
1.1 root 177: BITS_TO_LONGS(size >> XC_PAGE_SHIFT));
178:
179: bitmap_zero(entry->valid_mapping, nb_pfn);
180: for (i = 0; i < nb_pfn; i++) {
181: if (!err[i]) {
182: bitmap_set(entry->valid_mapping, i, 1);
183: }
184: }
185:
1.1.1.2 ! root 186: g_free(pfns);
! 187: g_free(err);
1.1 root 188: }
189:
190: uint8_t *xen_map_cache(target_phys_addr_t phys_addr, target_phys_addr_t size,
191: uint8_t lock)
192: {
193: MapCacheEntry *entry, *pentry = NULL;
194: target_phys_addr_t address_index = phys_addr >> MCACHE_BUCKET_SHIFT;
195: target_phys_addr_t address_offset = phys_addr & (MCACHE_BUCKET_SIZE - 1);
196: target_phys_addr_t __size = size;
197:
198: trace_xen_map_cache(phys_addr);
199:
200: if (address_index == mapcache->last_address_index && !lock && !__size) {
201: trace_xen_map_cache_return(mapcache->last_address_vaddr + address_offset);
202: return mapcache->last_address_vaddr + address_offset;
203: }
204:
205: /* size is always a multiple of MCACHE_BUCKET_SIZE */
206: if ((address_offset + (__size % MCACHE_BUCKET_SIZE)) > MCACHE_BUCKET_SIZE)
207: __size += MCACHE_BUCKET_SIZE;
208: if (__size % MCACHE_BUCKET_SIZE)
209: __size += MCACHE_BUCKET_SIZE - (__size % MCACHE_BUCKET_SIZE);
210: if (!__size)
211: __size = MCACHE_BUCKET_SIZE;
212:
213: entry = &mapcache->entry[address_index % mapcache->nr_buckets];
214:
215: while (entry && entry->lock && entry->vaddr_base &&
216: (entry->paddr_index != address_index || entry->size != __size ||
217: !test_bits(address_offset >> XC_PAGE_SHIFT, size >> XC_PAGE_SHIFT,
218: entry->valid_mapping))) {
219: pentry = entry;
220: entry = entry->next;
221: }
222: if (!entry) {
1.1.1.2 ! root 223: entry = g_malloc0(sizeof (MapCacheEntry));
1.1 root 224: pentry->next = entry;
225: xen_remap_bucket(entry, __size, address_index);
226: } else if (!entry->lock) {
227: if (!entry->vaddr_base || entry->paddr_index != address_index ||
228: entry->size != __size ||
229: !test_bits(address_offset >> XC_PAGE_SHIFT, size >> XC_PAGE_SHIFT,
230: entry->valid_mapping)) {
231: xen_remap_bucket(entry, __size, address_index);
232: }
233: }
234:
235: if(!test_bits(address_offset >> XC_PAGE_SHIFT, size >> XC_PAGE_SHIFT,
236: entry->valid_mapping)) {
237: mapcache->last_address_index = -1;
238: trace_xen_map_cache_return(NULL);
239: return NULL;
240: }
241:
242: mapcache->last_address_index = address_index;
243: mapcache->last_address_vaddr = entry->vaddr_base;
244: if (lock) {
1.1.1.2 ! root 245: MapCacheRev *reventry = g_malloc0(sizeof(MapCacheRev));
1.1 root 246: entry->lock++;
247: reventry->vaddr_req = mapcache->last_address_vaddr + address_offset;
248: reventry->paddr_index = mapcache->last_address_index;
249: reventry->size = entry->size;
250: QTAILQ_INSERT_HEAD(&mapcache->locked_entries, reventry, next);
251: }
252:
253: trace_xen_map_cache_return(mapcache->last_address_vaddr + address_offset);
254: return mapcache->last_address_vaddr + address_offset;
255: }
256:
257: ram_addr_t xen_ram_addr_from_mapcache(void *ptr)
258: {
1.1.1.2 ! root 259: MapCacheEntry *entry = NULL;
1.1 root 260: MapCacheRev *reventry;
261: target_phys_addr_t paddr_index;
262: target_phys_addr_t size;
263: int found = 0;
264:
265: QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
266: if (reventry->vaddr_req == ptr) {
267: paddr_index = reventry->paddr_index;
268: size = reventry->size;
269: found = 1;
270: break;
271: }
272: }
273: if (!found) {
274: fprintf(stderr, "%s, could not find %p\n", __func__, ptr);
275: QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
276: DPRINTF(" "TARGET_FMT_plx" -> %p is present\n", reventry->paddr_index,
277: reventry->vaddr_req);
278: }
279: abort();
280: return 0;
281: }
282:
283: entry = &mapcache->entry[paddr_index % mapcache->nr_buckets];
284: while (entry && (entry->paddr_index != paddr_index || entry->size != size)) {
285: entry = entry->next;
286: }
287: if (!entry) {
288: DPRINTF("Trying to find address %p that is not in the mapcache!\n", ptr);
289: return 0;
290: }
291: return (reventry->paddr_index << MCACHE_BUCKET_SHIFT) +
292: ((unsigned long) ptr - (unsigned long) entry->vaddr_base);
293: }
294:
295: void xen_invalidate_map_cache_entry(uint8_t *buffer)
296: {
297: MapCacheEntry *entry = NULL, *pentry = NULL;
298: MapCacheRev *reventry;
299: target_phys_addr_t paddr_index;
300: target_phys_addr_t size;
301: int found = 0;
302:
303: if (mapcache->last_address_vaddr == buffer) {
304: mapcache->last_address_index = -1;
305: }
306:
307: QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
308: if (reventry->vaddr_req == buffer) {
309: paddr_index = reventry->paddr_index;
310: size = reventry->size;
311: found = 1;
312: break;
313: }
314: }
315: if (!found) {
316: DPRINTF("%s, could not find %p\n", __func__, buffer);
317: QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
318: DPRINTF(" "TARGET_FMT_plx" -> %p is present\n", reventry->paddr_index, reventry->vaddr_req);
319: }
320: return;
321: }
322: QTAILQ_REMOVE(&mapcache->locked_entries, reventry, next);
1.1.1.2 ! root 323: g_free(reventry);
1.1 root 324:
325: entry = &mapcache->entry[paddr_index % mapcache->nr_buckets];
326: while (entry && (entry->paddr_index != paddr_index || entry->size != size)) {
327: pentry = entry;
328: entry = entry->next;
329: }
330: if (!entry) {
331: DPRINTF("Trying to unmap address %p that is not in the mapcache!\n", buffer);
332: return;
333: }
334: entry->lock--;
335: if (entry->lock > 0 || pentry == NULL) {
336: return;
337: }
338:
339: pentry->next = entry->next;
340: if (munmap(entry->vaddr_base, entry->size) != 0) {
341: perror("unmap fails");
342: exit(-1);
343: }
1.1.1.2 ! root 344: g_free(entry->valid_mapping);
! 345: g_free(entry);
1.1 root 346: }
347:
348: void xen_invalidate_map_cache(void)
349: {
350: unsigned long i;
351: MapCacheRev *reventry;
352:
353: /* Flush pending AIO before destroying the mapcache */
354: qemu_aio_flush();
355:
356: QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
357: DPRINTF("There should be no locked mappings at this time, "
358: "but "TARGET_FMT_plx" -> %p is present\n",
359: reventry->paddr_index, reventry->vaddr_req);
360: }
361:
362: mapcache_lock();
363:
364: for (i = 0; i < mapcache->nr_buckets; i++) {
365: MapCacheEntry *entry = &mapcache->entry[i];
366:
367: if (entry->vaddr_base == NULL) {
368: continue;
369: }
370:
371: if (munmap(entry->vaddr_base, entry->size) != 0) {
372: perror("unmap fails");
373: exit(-1);
374: }
375:
376: entry->paddr_index = 0;
377: entry->vaddr_base = NULL;
378: entry->size = 0;
1.1.1.2 ! root 379: g_free(entry->valid_mapping);
1.1 root 380: entry->valid_mapping = NULL;
381: }
382:
383: mapcache->last_address_index = -1;
384: mapcache->last_address_vaddr = NULL;
385:
386: mapcache_unlock();
387: }
unix.superglobalmegacorp.com