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 "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     MapCacheEntry *last_entry;
  78     unsigned long max_mcache_size;
  79     unsigned int mcache_bucket_shift;
  80
  81     phys_offset_to_gaddr_t phys_offset_to_gaddr;
  82     void *opaque;
  83 } MapCache;
  84
  85 static MapCache *mapcache;
  86
  87 static inline int test_bits(int nr, int size, const unsigned long *addr)
  88 {
  89     unsigned long res = find_next_zero_bit(addr, size + nr, nr);
  90     if (res >= nr + size)
  91         return 1;
  92     else
  93         return 0;
  94 }
  95
  96 void xen_map_cache_init(phys_offset_to_gaddr_t f, void *opaque)
  97 {
  98     unsigned long size;
  99     struct rlimit rlimit_as;
 100
 101     mapcache = g_malloc0(sizeof (MapCache));
 102
 103     mapcache->phys_offset_to_gaddr = f;
 104     mapcache->opaque = opaque;
 105
 106     QTAILQ_INIT(&mapcache->locked_entries);
 107
 108     if (geteuid() == 0) {
 109         rlimit_as.rlim_cur = RLIM_INFINITY;
 110         rlimit_as.rlim_max = RLIM_INFINITY;
 111         mapcache->max_mcache_size = MCACHE_MAX_SIZE;
 112     } else {
 113         getrlimit(RLIMIT_AS, &rlimit_as);
 114         rlimit_as.rlim_cur = rlimit_as.rlim_max;
 115
 116         if (rlimit_as.rlim_max != RLIM_INFINITY) {
 117             fprintf(stderr, "Warning: QEMU's maximum size of virtual"
 118                     " memory is not infinity.\n");
 119         }
 120         if (rlimit_as.rlim_max < MCACHE_MAX_SIZE + NON_MCACHE_MEMORY_SIZE) {
 121             mapcache->max_mcache_size = rlimit_as.rlim_max -
 122                 NON_MCACHE_MEMORY_SIZE;
 123         } else {
 124             mapcache->max_mcache_size = MCACHE_MAX_SIZE;
 125         }
 126     }
 127
 128     setrlimit(RLIMIT_AS, &rlimit_as);
 129
 130     mapcache->nr_buckets =
 131         (((mapcache->max_mcache_size >> XC_PAGE_SHIFT) +
 132           (1UL << (MCACHE_BUCKET_SHIFT - XC_PAGE_SHIFT)) - 1) >>
 133          (MCACHE_BUCKET_SHIFT - XC_PAGE_SHIFT));
 134
 135     size = mapcache->nr_buckets * sizeof (MapCacheEntry);
 136     size = (size + XC_PAGE_SIZE - 1) & ~(XC_PAGE_SIZE - 1);
 137     DPRINTF("%s, nr_buckets = %lx size %lu\n", __func__,
 138             mapcache->nr_buckets, size);
 139     mapcache->entry = g_malloc0(size);
 140 }
 141
 142 static void xen_remap_bucket(MapCacheEntry *entry,
 143                              hwaddr size,
 144                              hwaddr address_index)
 145 {
 146     uint8_t *vaddr_base;
 147     xen_pfn_t *pfns;
 148     int *err;
 149     unsigned int i;
 150     hwaddr nb_pfn = size >> XC_PAGE_SHIFT;
 151
 152     trace_xen_remap_bucket(address_index);
 153
 154     pfns = g_malloc0(nb_pfn * sizeof (xen_pfn_t));
 155     err = g_malloc0(nb_pfn * sizeof (int));
 156
 157     if (entry->vaddr_base != NULL) {
 158         if (munmap(entry->vaddr_base, entry->size) != 0) {
 159             perror("unmap fails");
 160             exit(-1);
 161         }
 162     }
 163     if (entry->valid_mapping != NULL) {
 164         g_free(entry->valid_mapping);
 165         entry->valid_mapping = NULL;
 166     }
 167
 168     for (i = 0; i < nb_pfn; i++) {
 169         pfns[i] = (address_index << (MCACHE_BUCKET_SHIFT-XC_PAGE_SHIFT)) + i;
 170     }
 171
 172     vaddr_base = xc_map_foreign_bulk(xen_xc, xen_domid, PROT_READ|PROT_WRITE,
 173                                      pfns, err, nb_pfn);
 174     if (vaddr_base == NULL) {
 175         perror("xc_map_foreign_bulk");
 176         exit(-1);
 177     }
 178
 179     entry->vaddr_base = vaddr_base;
 180     entry->paddr_index = address_index;
 181     entry->size = size;
 182     entry->valid_mapping = (unsigned long *) g_malloc0(sizeof(unsigned long) *
 183             BITS_TO_LONGS(size >> XC_PAGE_SHIFT));
 184
 185     bitmap_zero(entry->valid_mapping, nb_pfn);
 186     for (i = 0; i < nb_pfn; i++) {
 187         if (!err[i]) {
 188             bitmap_set(entry->valid_mapping, i, 1);
 189         }
 190     }
 191
 192     g_free(pfns);
 193     g_free(err);
 194 }
 195
 196 uint8_t *xen_map_cache(hwaddr phys_addr, hwaddr size,
 197                        uint8_t lock)
 198 {
 199     MapCacheEntry *entry, *pentry = NULL;
 200     hwaddr address_index;
 201     hwaddr address_offset;
 202     hwaddr __size = size;
 203     bool translated = false;
 204
 205 tryagain:
 206     address_index  = phys_addr >> MCACHE_BUCKET_SHIFT;
 207     address_offset = phys_addr & (MCACHE_BUCKET_SIZE - 1);
 208
 209     trace_xen_map_cache(phys_addr);
 210
 211     if (mapcache->last_entry != NULL &&
 212         mapcache->last_entry->paddr_index == address_index &&
 213         !lock && !__size) {
 214         trace_xen_map_cache_return(mapcache->last_entry->vaddr_base + address_offset);
 215         return mapcache->last_entry->vaddr_base + 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_entry = NULL;
 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_entry = entry;
 263     if (lock) {
 264         MapCacheRev *reventry = g_malloc0(sizeof(MapCacheRev));
 265         entry->lock++;
 266         reventry->vaddr_req = mapcache->last_entry->vaddr_base + address_offset;
 267         reventry->paddr_index = mapcache->last_entry->paddr_index;
 268         reventry->size = entry->size;
 269         QTAILQ_INSERT_HEAD(&mapcache->locked_entries, reventry, next);
 270     }
 271
 272     trace_xen_map_cache_return(mapcache->last_entry->vaddr_base + address_offset);
 273     return mapcache->last_entry->vaddr_base + address_offset;
 274 }
 275
 276 ram_addr_t xen_ram_addr_from_mapcache(void *ptr)
 277 {
 278     MapCacheEntry *entry = NULL;
 279     MapCacheRev *reventry;
 280     hwaddr paddr_index;
 281     hwaddr size;
 282     int found = 0;
 283
 284     QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 285         if (reventry->vaddr_req == ptr) {
 286             paddr_index = reventry->paddr_index;
 287             size = reventry->size;
 288             found = 1;
 289             break;
 290         }
 291     }
 292     if (!found) {
 293         fprintf(stderr, "%s, could not find %p\n", __func__, ptr);
 294         QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 295             DPRINTF("   "TARGET_FMT_plx" -> %p is present\n", reventry->paddr_index,
 296                     reventry->vaddr_req);
 297         }
 298         abort();
 299         return 0;
 300     }
 301
 302     entry = &mapcache->entry[paddr_index % mapcache->nr_buckets];
 303     while (entry && (entry->paddr_index != paddr_index || entry->size != size)) {
 304         entry = entry->next;
 305     }
 306     if (!entry) {
 307         DPRINTF("Trying to find address %p that is not in the mapcache!\n", ptr);
 308         return 0;
 309     }
 310     return (reventry->paddr_index << MCACHE_BUCKET_SHIFT) +
 311         ((unsigned long) ptr - (unsigned long) entry->vaddr_base);
 312 }
 313
 314 void xen_invalidate_map_cache_entry(uint8_t *buffer)
 315 {
 316     MapCacheEntry *entry = NULL, *pentry = NULL;
 317     MapCacheRev *reventry;
 318     hwaddr paddr_index;
 319     hwaddr size;
 320     int found = 0;
 321
 322     QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 323         if (reventry->vaddr_req == buffer) {
 324             paddr_index = reventry->paddr_index;
 325             size = reventry->size;
 326             found = 1;
 327             break;
 328         }
 329     }
 330     if (!found) {
 331         DPRINTF("%s, could not find %p\n", __func__, buffer);
 332         QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 333             DPRINTF("   "TARGET_FMT_plx" -> %p is present\n", reventry->paddr_index, reventry->vaddr_req);
 334         }
 335         return;
 336     }
 337     QTAILQ_REMOVE(&mapcache->locked_entries, reventry, next);
 338     g_free(reventry);
 339
 340     if (mapcache->last_entry != NULL &&
 341         mapcache->last_entry->paddr_index == paddr_index) {
 342         mapcache->last_entry = NULL;
 343     }
 344
 345     entry = &mapcache->entry[paddr_index % mapcache->nr_buckets];
 346     while (entry && (entry->paddr_index != paddr_index || entry->size != size)) {
 347         pentry = entry;
 348         entry = entry->next;
 349     }
 350     if (!entry) {
 351         DPRINTF("Trying to unmap address %p that is not in the mapcache!\n", buffer);
 352         return;
 353     }
 354     entry->lock--;
 355     if (entry->lock > 0 || pentry == NULL) {
 356         return;
 357     }
 358
 359     pentry->next = entry->next;
 360     if (munmap(entry->vaddr_base, entry->size) != 0) {
 361         perror("unmap fails");
 362         exit(-1);
 363     }
 364     g_free(entry->valid_mapping);
 365     g_free(entry);
 366 }
 367
 368 void xen_invalidate_map_cache(void)
 369 {
 370     unsigned long i;
 371     MapCacheRev *reventry;
 372
 373     /* Flush pending AIO before destroying the mapcache */
 374     bdrv_drain_all();
 375
 376     QTAILQ_FOREACH(reventry, &mapcache->locked_entries, next) {
 377         DPRINTF("There should be no locked mappings at this time, "
 378                 "but "TARGET_FMT_plx" -> %p is present\n",
 379                 reventry->paddr_index, reventry->vaddr_req);
 380     }
 381
 382     mapcache_lock();
 383
 384     for (i = 0; i < mapcache->nr_buckets; i++) {
 385         MapCacheEntry *entry = &mapcache->entry[i];
 386
 387         if (entry->vaddr_base == NULL) {
 388             continue;
 389         }
 390         if (entry->lock > 0) {
 391             continue;
 392         }
 393
 394         if (munmap(entry->vaddr_base, entry->size) != 0) {
 395             perror("unmap fails");
 396             exit(-1);
 397         }
 398
 399         entry->paddr_index = 0;
 400         entry->vaddr_base = NULL;
 401         entry->size = 0;
 402         g_free(entry->valid_mapping);
 403         entry->valid_mapping = NULL;
 404     }
 405
 406     mapcache->last_entry = NULL;
 407
 408     mapcache_unlock();
 409 }