xen-mapcache.c

   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 "blockdev.h"
  17 #include "bitmap.h"
  18
  19 #include <xen/hvm/params.h>
  20 #include <sys/mman.h>
  21
  22 #include "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     target_phys_addr_t paddr_index;
  57     uint8_t *vaddr_base;
  58     unsigned long *valid_mapping;
  59     uint8_t lock;
  60     target_phys_addr_t size;
  61     struct MapCacheEntry *next;
  62 } MapCacheEntry;
  63
  64 typedef struct MapCacheRev {
  65     uint8_t *vaddr_req;
  66     target_phys_addr_t paddr_index;
  67     target_phys_addr_t 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     target_phys_addr_t last_address_index;
  78     uint8_t *last_address_vaddr;
  79     unsigned long max_mcache_size;
  80     unsigned int mcache_bucket_shift;
  81 } MapCache;
  82
  83 static MapCache *mapcache;
  84
  85 static inline int test_bits(int nr, int size, const unsigned long *addr)
  86 {
  87     unsigned long res = find_next_zero_bit(addr, size + nr, nr);
  88     if (res >= nr + size)
  89         return 1;
  90     else
  91         return 0;
  92 }
  93
  94 void xen_map_cache_init(void)
  95 {
  96     unsigned long size;
  97     struct rlimit rlimit_as;
  98
  99     mapcache = g_malloc0(sizeof (MapCache));
 100
 101     QTAILQ_INIT(&mapcache->locked_entries);
 102     mapcache->last_address_index = -1;
 103
 104     if (geteuid() == 0) {
 105         rlimit_as.rlim_cur = RLIM_INFINITY;
 106         rlimit_as.rlim_max = RLIM_INFINITY;
 107         mapcache->max_mcache_size = MCACHE_MAX_SIZE;
 108     } else {
 109         getrlimit(RLIMIT_AS, &rlimit_as);
 110         rlimit_as.rlim_cur = rlimit_as.rlim_max;
 111
 112         if (rlimit_as.rlim_max != RLIM_INFINITY) {
 113             fprintf(stderr, "Warning: QEMU's maximum size of virtual"
 114                     " memory is not infinity.\n");
 115         }
 116         if (rlimit_as.rlim_max < MCACHE_MAX_SIZE + NON_MCACHE_MEMORY_SIZE) {
 117             mapcache->max_mcache_size = rlimit_as.rlim_max -
 118                 NON_MCACHE_MEMORY_SIZE;
 119         } else {
 120             mapcache->max_mcache_size = MCACHE_MAX_SIZE;
 121         }
 122     }
 123
 124     setrlimit(RLIMIT_AS, &rlimit_as);
 125
 126     mapcache->nr_buckets =
 127         (((mapcache->max_mcache_size >> XC_PAGE_SHIFT) +
 128           (1UL << (MCACHE_BUCKET_SHIFT - XC_PAGE_SHIFT)) - 1) >>
 129          (MCACHE_BUCKET_SHIFT - XC_PAGE_SHIFT));
 130
 131     size = mapcache->nr_buckets * sizeof (MapCacheEntry);
 132     size = (size + XC_PAGE_SIZE - 1) & ~(XC_PAGE_SIZE - 1);
 133     DPRINTF("%s, nr_buckets = %lx size %lu\n", __func__,
 134             mapcache->nr_buckets, size);
 135     mapcache->entry = g_malloc0(size);
 136 }
 137
 138 static void xen_remap_bucket(MapCacheEntry *entry,
 139                              target_phys_addr_t size,
 140                              target_phys_addr_t address_index)
 141 {
 142     uint8_t *vaddr_base;
 143     xen_pfn_t *pfns;
 144     int *err;
 145     unsigned int i;
 146     target_phys_addr_t nb_pfn = size >> XC_PAGE_SHIFT;
 147
 148     trace_xen_remap_bucket(address_index);
 149
 150     pfns = g_malloc0(nb_pfn * sizeof (xen_pfn_t));
 151     err = g_malloc0(nb_pfn * sizeof (int));
 152
 153     if (entry->vaddr_base != NULL) {
 154         if (munmap(entry->vaddr_base, entry->size) != 0) {
 155             perror("unmap fails");
 156             exit(-1);
 157         }
 158     }
 159     if (entry->valid_mapping != NULL) {
 160         g_free(entry->valid_mapping);
 161         entry->valid_mapping = NULL;
 162     }
 163
 164     for (i = 0; i < nb_pfn; i++) {
 165         pfns[i] = (address_index << (MCACHE_BUCKET_SHIFT-XC_PAGE_SHIFT)) + i;
 166     }
 167
 168     vaddr_base = xc_map_foreign_bulk(xen_xc, xen_domid, PROT_READ|PROT_WRITE,
 169                                      pfns, err, nb_pfn);
 170     if (vaddr_base == NULL) {
 171         perror("xc_map_foreign_bulk");
 172         exit(-1);
 173     }
 174
 175     entry->vaddr_base = vaddr_base;
 176     entry->paddr_index = address_index;
 177     entry->size = size;
 178     entry->valid_mapping = (unsigned long *) g_malloc0(sizeof(unsigned long) *
 179             BITS_TO_LONGS(size >> XC_PAGE_SHIFT));
 180
 181     bitmap_zero(entry->valid_mapping, nb_pfn);
 182     for (i = 0; i < nb_pfn; i++) {
 183         if (!err[i]) {
 184             bitmap_set(entry->valid_mapping, i, 1);
 185         }
 186     }
 187
 188     g_free(pfns);
 189     g_free(err);
 190 }
 191
 192 uint8_t *xen_map_cache(target_phys_addr_t phys_addr, target_phys_addr_t size,
 193                        uint8_t lock)
 194 {
 195     MapCacheEntry *entry, *pentry = NULL;
 196     target_phys_addr_t address_index  = phys_addr >> MCACHE_BUCKET_SHIFT;
 197     target_phys_addr_t address_offset = phys_addr & (MCACHE_BUCKET_SIZE - 1);
 198     target_phys_addr_t __size = size;
 199
 200     trace_xen_map_cache(phys_addr);
 201
 202     if (address_index == mapcache->last_address_index && !lock && !__size) {
 203         trace_xen_map_cache_return(mapcache->last_address_vaddr + address_offset);
 204         return mapcache->last_address_vaddr + address_offset;
 205     }
 206
 207     /* size is always a multiple of MCACHE_BUCKET_SIZE */
 208     if ((address_offset + (__size % MCACHE_BUCKET_SIZE)) > MCACHE_BUCKET_SIZE)
 209         __size += MCACHE_BUCKET_SIZE;
 210     if (__size % MCACHE_BUCKET_SIZE)
 211         __size += MCACHE_BUCKET_SIZE - (__size % MCACHE_BUCKET_SIZE);
 212     if (!__size)
 213         __size = MCACHE_BUCKET_SIZE;
 214
 215     entry = &mapcache->entry[address_index % mapcache->nr_buckets];
 216
 217     while (entry && entry->lock && entry->vaddr_base &&
 218             (entry->paddr_index != address_index || entry->size != __size ||
 219              !test_bits(address_offset >> XC_PAGE_SHIFT, size >> XC_PAGE_SHIFT,
 220                  entry->valid_mapping))) {
 221         pentry = entry;
 222         entry = entry->next;
 223     }
 224     if (!entry) {
 225         entry = g_malloc0(sizeof (MapCacheEntry));
 226         pentry->next = entry;
 227         xen_remap_bucket(entry, __size, address_index);
 228     } else if (!entry->lock) {
 229         if (!entry->vaddr_base || entry->paddr_index != address_index ||
 230                 entry->size != __size ||
 231                 !test_bits(address_offset >> XC_PAGE_SHIFT, size >> XC_PAGE_SHIFT,
 232                     entry->valid_mapping)) {
 233             xen_remap_bucket(entry, __size, address_index);
 234         }
 235     }
 236
 237     if(!test_bits(address_offset >> XC_PAGE_SHIFT, size >> XC_PAGE_SHIFT,
 238                 entry->valid_mapping)) {
 239         mapcache->last_address_index = -1;
 240         trace_xen_map_cache_return(NULL);
 241         return NULL;
 242     }
 243
 244     mapcache->last_address_index = address_index;
 245     mapcache->last_address_vaddr = entry->vaddr_base;
 246     if (lock) {
 247         MapCacheRev *reventry = g_malloc0(sizeof(MapCacheRev));
 248         entry->lock++;
 249         reventry->vaddr_req = mapcache->last_address_vaddr + address_offset;
 250         reventry->paddr_index = mapcache->last_address_index;
 251         reventry->size = entry->size;
 252         QTAILQ_INSERT_HEAD(&mapcache->locked_entries, reventry, next);
 253     }
 254
 255     trace_xen_map_cache_return(mapcache->last_address_vaddr + address_offset);
 256     return mapcache->last_address_vaddr + address_offset;
 257 }
 258
 259 ram_addr_t xen_ram_addr_from_mapcache(void *ptr)
 260 {
 261     MapCacheEntry *entry = NULL;
 262     MapCacheRev *reventry;
 263     target_phys_addr_t paddr_index;
 264     target_phys_addr_t size;
 265     int found = 0;
 266
 267     QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 268         if (reventry->vaddr_req == ptr) {
 269             paddr_index = reventry->paddr_index;
 270             size = reventry->size;
 271             found = 1;
 272             break;
 273         }
 274     }
 275     if (!found) {
 276         fprintf(stderr, "%s, could not find %p\n", __func__, ptr);
 277         QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 278             DPRINTF("   "TARGET_FMT_plx" -> %p is present\n", reventry->paddr_index,
 279                     reventry->vaddr_req);
 280         }
 281         abort();
 282         return 0;
 283     }
 284
 285     entry = &mapcache->entry[paddr_index % mapcache->nr_buckets];
 286     while (entry && (entry->paddr_index != paddr_index || entry->size != size)) {
 287         entry = entry->next;
 288     }
 289     if (!entry) {
 290         DPRINTF("Trying to find address %p that is not in the mapcache!\n", ptr);
 291         return 0;
 292     }
 293     return (reventry->paddr_index << MCACHE_BUCKET_SHIFT) +
 294         ((unsigned long) ptr - (unsigned long) entry->vaddr_base);
 295 }
 296
 297 void xen_invalidate_map_cache_entry(uint8_t *buffer)
 298 {
 299     MapCacheEntry *entry = NULL, *pentry = NULL;
 300     MapCacheRev *reventry;
 301     target_phys_addr_t paddr_index;
 302     target_phys_addr_t size;
 303     int found = 0;
 304
 305     if (mapcache->last_address_vaddr == buffer) {
 306         mapcache->last_address_index = -1;
 307     }
 308
 309     QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 310         if (reventry->vaddr_req == buffer) {
 311             paddr_index = reventry->paddr_index;
 312             size = reventry->size;
 313             found = 1;
 314             break;
 315         }
 316     }
 317     if (!found) {
 318         DPRINTF("%s, could not find %p\n", __func__, buffer);
 319         QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 320             DPRINTF("   "TARGET_FMT_plx" -> %p is present\n", reventry->paddr_index, reventry->vaddr_req);
 321         }
 322         return;
 323     }
 324     QTAILQ_REMOVE(&mapcache->locked_entries, reventry, next);
 325     g_free(reventry);
 326
 327     entry = &mapcache->entry[paddr_index % mapcache->nr_buckets];
 328     while (entry && (entry->paddr_index != paddr_index || entry->size != size)) {
 329         pentry = entry;
 330         entry = entry->next;
 331     }
 332     if (!entry) {
 333         DPRINTF("Trying to unmap address %p that is not in the mapcache!\n", buffer);
 334         return;
 335     }
 336     entry->lock--;
 337     if (entry->lock > 0 || pentry == NULL) {
 338         return;
 339     }
 340
 341     pentry->next = entry->next;
 342     if (munmap(entry->vaddr_base, entry->size) != 0) {
 343         perror("unmap fails");
 344         exit(-1);
 345     }
 346     g_free(entry->valid_mapping);
 347     g_free(entry);
 348 }
 349
 350 void xen_invalidate_map_cache(void)
 351 {
 352     unsigned long i;
 353     MapCacheRev *reventry;
 354
 355     /* Flush pending AIO before destroying the mapcache */
 356     bdrv_drain_all();
 357
 358     QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 359         DPRINTF("There should be no locked mappings at this time, "
 360                 "but "TARGET_FMT_plx" -> %p is present\n",
 361                 reventry->paddr_index, reventry->vaddr_req);
 362     }
 363
 364     mapcache_lock();
 365
 366     for (i = 0; i < mapcache->nr_buckets; i++) {
 367         MapCacheEntry *entry = &mapcache->entry[i];
 368
 369         if (entry->vaddr_base == NULL) {
 370             continue;
 371         }
 372
 373         if (munmap(entry->vaddr_base, entry->size) != 0) {
 374             perror("unmap fails");
 375             exit(-1);
 376         }
 377
 378         entry->paddr_index = 0;
 379         entry->vaddr_base = NULL;
 380         entry->size = 0;
 381         g_free(entry->valid_mapping);
 382         entry->valid_mapping = NULL;
 383     }
 384
 385     mapcache->last_address_index = -1;
 386     mapcache->last_address_vaddr = NULL;
 387
 388     mapcache_unlock();
 389 }