sewardj | f98e1c0 | 2008-10-25 16:22:41 +0000 | [diff] [blame^] | 1 | |
| 2 | /*--------------------------------------------------------------------*/ |
| 3 | /*--- LibHB: a library for implementing and checking ---*/ |
| 4 | /*--- the happens-before relationship in concurrent programs. ---*/ |
| 5 | /*--- libhb_main.c ---*/ |
| 6 | /*--------------------------------------------------------------------*/ |
| 7 | |
| 8 | /* |
| 9 | This file is part of LibHB, a library for implementing and checking |
| 10 | the happens-before relationship in concurrent programs. |
| 11 | |
| 12 | Copyright (C) 2008-2008 OpenWorks Ltd |
| 13 | info@open-works.co.uk |
| 14 | |
| 15 | This program is free software; you can redistribute it and/or |
| 16 | modify it under the terms of the GNU General Public License as |
| 17 | published by the Free Software Foundation; either version 2 of the |
| 18 | License, or (at your option) any later version. |
| 19 | |
| 20 | This program is distributed in the hope that it will be useful, but |
| 21 | WITHOUT ANY WARRANTY; without even the implied warranty of |
| 22 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 23 | General Public License for more details. |
| 24 | |
| 25 | You should have received a copy of the GNU General Public License |
| 26 | along with this program; if not, write to the Free Software |
| 27 | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA |
| 28 | 02111-1307, USA. |
| 29 | |
| 30 | The GNU General Public License is contained in the file COPYING. |
| 31 | */ |
| 32 | |
| 33 | #include "pub_tool_basics.h" |
| 34 | #include "pub_tool_libcassert.h" |
| 35 | #include "pub_tool_libcbase.h" |
| 36 | #include "pub_tool_libcprint.h" |
| 37 | #include "pub_tool_mallocfree.h" |
| 38 | #include "pub_tool_wordfm.h" |
| 39 | #include "pub_tool_xarray.h" |
| 40 | #include "pub_tool_oset.h" |
| 41 | #include "pub_tool_threadstate.h" |
| 42 | #include "pub_tool_aspacemgr.h" |
| 43 | #include "pub_tool_execontext.h" |
| 44 | #include "pub_tool_errormgr.h" |
| 45 | |
| 46 | #include "hg_basics.h" |
| 47 | #include "hg_wordset.h" |
| 48 | #include "hg_lock_n_thread.h" |
| 49 | #include "hg_errors.h" |
| 50 | |
| 51 | #include "libhb.h" |
| 52 | |
| 53 | |
| 54 | /* fwds for |
| 55 | Globals needed by other parts of the library. These are set |
| 56 | once at startup and then never changed. */ |
| 57 | static void (*main_get_stacktrace)( Thr*, Addr*, UWord ) = NULL; |
| 58 | static struct _EC* (*main_stacktrace_to_EC)( Addr*, UWord ) = NULL; |
| 59 | static struct _EC* (*main_get_EC)( Thr* ) = NULL; |
| 60 | |
| 61 | ///////////////////////////////////////////////////////////////// |
| 62 | ///////////////////////////////////////////////////////////////// |
| 63 | // // |
| 64 | // // |
| 65 | // // |
| 66 | ///////////////////////////////////////////////////////////////// |
| 67 | ///////////////////////////////////////////////////////////////// |
| 68 | |
| 69 | |
| 70 | ///////////////////////////////////////////////////////////////// |
| 71 | ///////////////////////////////////////////////////////////////// |
| 72 | // // |
| 73 | // SECTION BEGIN compressed shadow memory // |
| 74 | // // |
| 75 | ///////////////////////////////////////////////////////////////// |
| 76 | ///////////////////////////////////////////////////////////////// |
| 77 | |
| 78 | #ifndef __HB_ZSM_H |
| 79 | #define __HB_ZSM_H |
| 80 | |
| 81 | typedef ULong SVal; |
| 82 | |
| 83 | /* This value has special significance to the implementation, and callers |
| 84 | may not store it in the shadow memory. */ |
| 85 | #define SVal_INVALID (3ULL << 62) |
| 86 | |
| 87 | /* This is the default value for shadow memory. Initially the shadow |
| 88 | memory contains no accessible areas and so all reads produce this |
| 89 | value. TODO: make this caller-defineable. */ |
| 90 | #define SVal_NOACCESS (2ULL << 62) |
| 91 | |
| 92 | /* Initialise the library. Once initialised, it will (or may) call |
| 93 | rcinc and rcdec in response to all the calls below, in order to |
| 94 | allow the user to do reference counting on the SVals stored herein. |
| 95 | It is important to understand, however, that due to internal |
| 96 | caching, the reference counts are in general inaccurate, and can be |
| 97 | both above or below the true reference count for an item. In |
| 98 | particular, the library may indicate that the reference count for |
| 99 | an item is zero, when in fact it is not. |
| 100 | |
| 101 | To make the reference counting exact and therefore non-pointless, |
| 102 | call zsm_flush_cache. Immediately after it returns, the reference |
| 103 | counts for all items, as deduced by the caller by observing calls |
| 104 | to rcinc and rcdec, will be correct, and so any items with a zero |
| 105 | reference count may be freed (or at least considered to be |
| 106 | unreferenced by this library). |
| 107 | */ |
| 108 | static void zsm_init ( void(*rcinc)(SVal), void(*rcdec)(SVal) ); |
| 109 | |
| 110 | static void zsm_set_range ( Addr, SizeT, SVal ); |
| 111 | static SVal zsm_read8 ( Addr ); |
| 112 | static void zsm_copy_range ( Addr, Addr, SizeT ); |
| 113 | static void zsm_flush_cache ( void ); |
| 114 | |
| 115 | #endif /* ! __HB_ZSM_H */ |
| 116 | |
| 117 | |
| 118 | /* For the shadow mem cache stuff we may want more intrusive |
| 119 | checks. Unfortunately there's no almost-zero-cost way to make them |
| 120 | selectable at run time. Hence set the #if 0 to #if 1 and |
| 121 | rebuild if you want them. */ |
| 122 | #if 0 |
| 123 | # define SCE_CACHELINE 1 /* do sanity-check CacheLine stuff */ |
| 124 | # define inline __attribute__((noinline)) |
| 125 | /* probably want to ditch -fomit-frame-pointer too */ |
| 126 | #else |
| 127 | # define SCE_CACHELINE 0 /* don't sanity-check CacheLine stuff */ |
| 128 | #endif |
| 129 | |
| 130 | /* For the SegmentID, SegmentSet and SVal stuff we may want more |
| 131 | intrusive checks. Again there's no zero cost way to do this. Set |
| 132 | the #if 0 to #if 1 and rebuild if you want them. */ |
| 133 | #if 0 |
| 134 | # define SCE_SVALS 1 /* sanity-check shadow value stuff */ |
| 135 | #else |
| 136 | # define SCE_SVALS 0 |
| 137 | #endif |
| 138 | |
| 139 | |
| 140 | /* Round a up to the next multiple of N. N must be a power of 2 */ |
| 141 | #define ROUNDUP(a, N) ((a + N - 1) & ~(N-1)) |
| 142 | /* Round a down to the next multiple of N. N must be a power of 2 */ |
| 143 | #define ROUNDDN(a, N) ((a) & ~(N-1)) |
| 144 | |
| 145 | |
| 146 | |
| 147 | /* ------ User-supplied RC functions ------ */ |
| 148 | static void(*rcinc)(SVal) = NULL; |
| 149 | static void(*rcdec)(SVal) = NULL; |
| 150 | |
| 151 | |
| 152 | /* ------ CacheLine ------ */ |
| 153 | |
| 154 | #define N_LINE_BITS 6 /* must be >= 3 */ |
| 155 | #define N_LINE_ARANGE (1 << N_LINE_BITS) |
| 156 | #define N_LINE_TREES (N_LINE_ARANGE >> 3) |
| 157 | |
| 158 | typedef |
| 159 | struct { |
| 160 | UShort descrs[N_LINE_TREES]; |
| 161 | SVal svals[N_LINE_ARANGE]; // == N_LINE_TREES * 8 |
| 162 | } |
| 163 | CacheLine; |
| 164 | |
| 165 | #define TREE_DESCR_16_0 (1<<0) |
| 166 | #define TREE_DESCR_32_0 (1<<1) |
| 167 | #define TREE_DESCR_16_1 (1<<2) |
| 168 | #define TREE_DESCR_64 (1<<3) |
| 169 | #define TREE_DESCR_16_2 (1<<4) |
| 170 | #define TREE_DESCR_32_1 (1<<5) |
| 171 | #define TREE_DESCR_16_3 (1<<6) |
| 172 | #define TREE_DESCR_8_0 (1<<7) |
| 173 | #define TREE_DESCR_8_1 (1<<8) |
| 174 | #define TREE_DESCR_8_2 (1<<9) |
| 175 | #define TREE_DESCR_8_3 (1<<10) |
| 176 | #define TREE_DESCR_8_4 (1<<11) |
| 177 | #define TREE_DESCR_8_5 (1<<12) |
| 178 | #define TREE_DESCR_8_6 (1<<13) |
| 179 | #define TREE_DESCR_8_7 (1<<14) |
| 180 | #define TREE_DESCR_DTY (1<<15) |
| 181 | |
| 182 | typedef |
| 183 | struct { |
| 184 | SVal dict[4]; /* can represent up to 4 diff values in the line */ |
| 185 | UChar ix2s[N_LINE_ARANGE/4]; /* array of N_LINE_ARANGE 2-bit |
| 186 | dict indexes */ |
| 187 | /* if dict[0] == SVal_INVALID then dict[1] is the index of the |
| 188 | LineF to use, and dict[2..] are also SVal_INVALID. */ |
| 189 | } |
| 190 | LineZ; /* compressed rep for a cache line */ |
| 191 | |
| 192 | typedef |
| 193 | struct { |
| 194 | Bool inUse; |
| 195 | SVal w64s[N_LINE_ARANGE]; |
| 196 | } |
| 197 | LineF; /* full rep for a cache line */ |
| 198 | |
| 199 | /* Shadow memory. |
| 200 | Primary map is a WordFM Addr SecMap*. |
| 201 | SecMaps cover some page-size-ish section of address space and hold |
| 202 | a compressed representation. |
| 203 | CacheLine-sized chunks of SecMaps are copied into a Cache, being |
| 204 | decompressed when moved into the cache and recompressed on the |
| 205 | way out. Because of this, the cache must operate as a writeback |
| 206 | cache, not a writethrough one. |
| 207 | |
| 208 | Each SecMap must hold a power-of-2 number of CacheLines. Hence |
| 209 | N_SECMAP_BITS must >= N_LINE_BITS. |
| 210 | */ |
| 211 | #define N_SECMAP_BITS 13 |
| 212 | #define N_SECMAP_ARANGE (1 << N_SECMAP_BITS) |
| 213 | |
| 214 | // # CacheLines held by a SecMap |
| 215 | #define N_SECMAP_ZLINES (N_SECMAP_ARANGE / N_LINE_ARANGE) |
| 216 | |
| 217 | /* The data in the SecMap is held in the array of LineZs. Each LineZ |
| 218 | either carries the required data directly, in a compressed |
| 219 | representation, or it holds (in .dict[0]) an index to the LineF in |
| 220 | .linesF that holds the full representation. |
| 221 | |
| 222 | Currently-unused LineF's have their .inUse bit set to zero. |
| 223 | Since each in-use LineF is referred to be exactly one LineZ, |
| 224 | the number of .linesZ[] that refer to .linesF should equal |
| 225 | the number of .linesF[] that have .inUse == True. |
| 226 | |
| 227 | RC obligations: the RCs presented to the user include exactly |
| 228 | the values in: |
| 229 | * direct Z reps, that is, ones for which .dict[0] != SVal_INVALID |
| 230 | * F reps that are in use (.inUse == True) |
| 231 | |
| 232 | Hence the following actions at the following transitions are required: |
| 233 | |
| 234 | F rep: .inUse==True -> .inUse==False -- rcdec_LineF |
| 235 | F rep: .inUse==False -> .inUse==True -- rcinc_LineF |
| 236 | Z rep: .dict[0] from other to SVal_INVALID -- rcdec_LineZ |
| 237 | Z rep: .dict[0] from SVal_INVALID to other -- rcinc_LineZ |
| 238 | */ |
| 239 | typedef |
| 240 | struct { |
| 241 | UInt magic; |
| 242 | LineZ linesZ[N_SECMAP_ZLINES]; |
| 243 | LineF* linesF; |
| 244 | UInt linesF_size; |
| 245 | } |
| 246 | SecMap; |
| 247 | |
| 248 | #define SecMap_MAGIC 0x571e58cbU |
| 249 | |
| 250 | static inline Bool is_sane_SecMap ( SecMap* sm ) { |
| 251 | return sm != NULL && sm->magic == SecMap_MAGIC; |
| 252 | } |
| 253 | |
| 254 | /* ------ Cache ------ */ |
| 255 | |
| 256 | #define N_WAY_BITS 16 |
| 257 | #define N_WAY_NENT (1 << N_WAY_BITS) |
| 258 | |
| 259 | /* Each tag is the address of the associated CacheLine, rounded down |
| 260 | to a CacheLine address boundary. A CacheLine size must be a power |
| 261 | of 2 and must be 8 or more. Hence an easy way to initialise the |
| 262 | cache so it is empty is to set all the tag values to any value % 8 |
| 263 | != 0, eg 1. This means all queries in the cache initially miss. |
| 264 | It does however require us to detect and not writeback, any line |
| 265 | with a bogus tag. */ |
| 266 | typedef |
| 267 | struct { |
| 268 | CacheLine lyns0[N_WAY_NENT]; |
| 269 | Addr tags0[N_WAY_NENT]; |
| 270 | } |
| 271 | Cache; |
| 272 | |
| 273 | static inline Bool is_valid_scache_tag ( Addr tag ) { |
| 274 | /* a valid tag should be naturally aligned to the start of |
| 275 | a CacheLine. */ |
| 276 | return 0 == (tag & (N_LINE_ARANGE - 1)); |
| 277 | } |
| 278 | |
| 279 | |
| 280 | /* --------- Primary data structures --------- */ |
| 281 | |
| 282 | /* Shadow memory primary map */ |
| 283 | static WordFM* map_shmem = NULL; /* WordFM Addr SecMap* */ |
| 284 | static Cache cache_shmem; |
| 285 | |
| 286 | |
| 287 | static UWord stats__secmaps_search = 0; // # SM finds |
| 288 | static UWord stats__secmaps_search_slow = 0; // # SM lookupFMs |
| 289 | static UWord stats__secmaps_allocd = 0; // # SecMaps issued |
| 290 | static UWord stats__secmap_ga_space_covered = 0; // # ga bytes covered |
| 291 | static UWord stats__secmap_linesZ_allocd = 0; // # LineZ's issued |
| 292 | static UWord stats__secmap_linesZ_bytes = 0; // .. using this much storage |
| 293 | static UWord stats__secmap_linesF_allocd = 0; // # LineF's issued |
| 294 | static UWord stats__secmap_linesF_bytes = 0; // .. using this much storage |
| 295 | static UWord stats__secmap_iterator_steppings = 0; // # calls to stepSMIter |
| 296 | static UWord stats__cache_Z_fetches = 0; // # Z lines fetched |
| 297 | static UWord stats__cache_Z_wbacks = 0; // # Z lines written back |
| 298 | static UWord stats__cache_F_fetches = 0; // # F lines fetched |
| 299 | static UWord stats__cache_F_wbacks = 0; // # F lines written back |
| 300 | static UWord stats__cache_invals = 0; // # cache invals |
| 301 | static UWord stats__cache_flushes = 0; // # cache flushes |
| 302 | static UWord stats__cache_totrefs = 0; // # total accesses |
| 303 | static UWord stats__cache_totmisses = 0; // # misses |
| 304 | static ULong stats__cache_make_New_arange = 0; // total arange made New |
| 305 | static ULong stats__cache_make_New_inZrep = 0; // arange New'd on Z reps |
| 306 | static UWord stats__cline_normalises = 0; // # calls to cacheline_normalise |
| 307 | static UWord stats__cline_read64s = 0; // # calls to s_m_read64 |
| 308 | static UWord stats__cline_read32s = 0; // # calls to s_m_read32 |
| 309 | static UWord stats__cline_read16s = 0; // # calls to s_m_read16 |
| 310 | static UWord stats__cline_read8s = 0; // # calls to s_m_read8 |
| 311 | static UWord stats__cline_write64s = 0; // # calls to s_m_write64 |
| 312 | static UWord stats__cline_write32s = 0; // # calls to s_m_write32 |
| 313 | static UWord stats__cline_write16s = 0; // # calls to s_m_write16 |
| 314 | static UWord stats__cline_write8s = 0; // # calls to s_m_write8 |
| 315 | static UWord stats__cline_set64s = 0; // # calls to s_m_set64 |
| 316 | static UWord stats__cline_set32s = 0; // # calls to s_m_set32 |
| 317 | static UWord stats__cline_set16s = 0; // # calls to s_m_set16 |
| 318 | static UWord stats__cline_set8s = 0; // # calls to s_m_set8 |
| 319 | static UWord stats__cline_get8s = 0; // # calls to s_m_get8 |
| 320 | static UWord stats__cline_copy8s = 0; // # calls to s_m_copy8 |
| 321 | static UWord stats__cline_64to32splits = 0; // # 64-bit accesses split |
| 322 | static UWord stats__cline_32to16splits = 0; // # 32-bit accesses split |
| 323 | static UWord stats__cline_16to8splits = 0; // # 16-bit accesses split |
| 324 | static UWord stats__cline_64to32pulldown = 0; // # calls to pulldown_to_32 |
| 325 | static UWord stats__cline_32to16pulldown = 0; // # calls to pulldown_to_16 |
| 326 | static UWord stats__cline_16to8pulldown = 0; // # calls to pulldown_to_8 |
| 327 | |
| 328 | static inline Addr shmem__round_to_SecMap_base ( Addr a ) { |
| 329 | return a & ~(N_SECMAP_ARANGE - 1); |
| 330 | } |
| 331 | static inline UWord shmem__get_SecMap_offset ( Addr a ) { |
| 332 | return a & (N_SECMAP_ARANGE - 1); |
| 333 | } |
| 334 | |
| 335 | |
| 336 | /*----------------------------------------------------------------*/ |
| 337 | /*--- map_shmem :: WordFM Addr SecMap ---*/ |
| 338 | /*--- shadow memory (low level handlers) (shmem__* fns) ---*/ |
| 339 | /*----------------------------------------------------------------*/ |
| 340 | |
| 341 | /*--------------- SecMap allocation --------------- */ |
| 342 | |
| 343 | static HChar* shmem__bigchunk_next = NULL; |
| 344 | static HChar* shmem__bigchunk_end1 = NULL; |
| 345 | |
| 346 | static void* shmem__bigchunk_alloc ( SizeT n ) |
| 347 | { |
| 348 | const SizeT sHMEM__BIGCHUNK_SIZE = 4096 * 256 * 4; |
| 349 | tl_assert(n > 0); |
| 350 | n = VG_ROUNDUP(n, 16); |
| 351 | tl_assert(shmem__bigchunk_next <= shmem__bigchunk_end1); |
| 352 | tl_assert(shmem__bigchunk_end1 - shmem__bigchunk_next |
| 353 | <= (SSizeT)sHMEM__BIGCHUNK_SIZE); |
| 354 | if (shmem__bigchunk_next + n > shmem__bigchunk_end1) { |
| 355 | if (0) |
| 356 | VG_(printf)("XXXXX bigchunk: abandoning %d bytes\n", |
| 357 | (Int)(shmem__bigchunk_end1 - shmem__bigchunk_next)); |
| 358 | shmem__bigchunk_next = VG_(am_shadow_alloc)( sHMEM__BIGCHUNK_SIZE ); |
| 359 | if (shmem__bigchunk_next == NULL) |
| 360 | VG_(out_of_memory_NORETURN)( |
| 361 | "helgrind:shmem__bigchunk_alloc", sHMEM__BIGCHUNK_SIZE ); |
| 362 | shmem__bigchunk_end1 = shmem__bigchunk_next + sHMEM__BIGCHUNK_SIZE; |
| 363 | } |
| 364 | tl_assert(shmem__bigchunk_next); |
| 365 | tl_assert( 0 == (((Addr)shmem__bigchunk_next) & (16-1)) ); |
| 366 | tl_assert(shmem__bigchunk_next + n <= shmem__bigchunk_end1); |
| 367 | shmem__bigchunk_next += n; |
| 368 | return shmem__bigchunk_next - n; |
| 369 | } |
| 370 | |
| 371 | static SecMap* shmem__alloc_SecMap ( void ) |
| 372 | { |
| 373 | Word i, j; |
| 374 | SecMap* sm = shmem__bigchunk_alloc( sizeof(SecMap) ); |
| 375 | if (0) VG_(printf)("alloc_SecMap %p\n",sm); |
| 376 | tl_assert(sm); |
| 377 | sm->magic = SecMap_MAGIC; |
| 378 | for (i = 0; i < N_SECMAP_ZLINES; i++) { |
| 379 | sm->linesZ[i].dict[0] = SVal_NOACCESS; |
| 380 | sm->linesZ[i].dict[1] = SVal_INVALID; |
| 381 | sm->linesZ[i].dict[2] = SVal_INVALID; |
| 382 | sm->linesZ[i].dict[3] = SVal_INVALID; |
| 383 | for (j = 0; j < N_LINE_ARANGE/4; j++) |
| 384 | sm->linesZ[i].ix2s[j] = 0; /* all reference dict[0] */ |
| 385 | } |
| 386 | sm->linesF = NULL; |
| 387 | sm->linesF_size = 0; |
| 388 | stats__secmaps_allocd++; |
| 389 | stats__secmap_ga_space_covered += N_SECMAP_ARANGE; |
| 390 | stats__secmap_linesZ_allocd += N_SECMAP_ZLINES; |
| 391 | stats__secmap_linesZ_bytes += N_SECMAP_ZLINES * sizeof(LineZ); |
| 392 | return sm; |
| 393 | } |
| 394 | |
| 395 | typedef struct { Addr gaKey; SecMap* sm; } SMCacheEnt; |
| 396 | static SMCacheEnt smCache[3] = { {1,NULL}, {1,NULL}, {1,NULL} }; |
| 397 | |
| 398 | static SecMap* shmem__find_SecMap ( Addr ga ) |
| 399 | { |
| 400 | SecMap* sm = NULL; |
| 401 | Addr gaKey = shmem__round_to_SecMap_base(ga); |
| 402 | // Cache |
| 403 | stats__secmaps_search++; |
| 404 | if (LIKELY(gaKey == smCache[0].gaKey)) |
| 405 | return smCache[0].sm; |
| 406 | if (LIKELY(gaKey == smCache[1].gaKey)) { |
| 407 | SMCacheEnt tmp = smCache[0]; |
| 408 | smCache[0] = smCache[1]; |
| 409 | smCache[1] = tmp; |
| 410 | return smCache[0].sm; |
| 411 | } |
| 412 | if (gaKey == smCache[2].gaKey) { |
| 413 | SMCacheEnt tmp = smCache[1]; |
| 414 | smCache[1] = smCache[2]; |
| 415 | smCache[2] = tmp; |
| 416 | return smCache[1].sm; |
| 417 | } |
| 418 | // end Cache |
| 419 | stats__secmaps_search_slow++; |
| 420 | if (VG_(lookupFM)( map_shmem, |
| 421 | NULL/*keyP*/, (UWord*)&sm, (UWord)gaKey )) { |
| 422 | tl_assert(sm != NULL); |
| 423 | smCache[2] = smCache[1]; |
| 424 | smCache[1] = smCache[0]; |
| 425 | smCache[0].gaKey = gaKey; |
| 426 | smCache[0].sm = sm; |
| 427 | } else { |
| 428 | tl_assert(sm == NULL); |
| 429 | } |
| 430 | return sm; |
| 431 | } |
| 432 | |
| 433 | static SecMap* shmem__find_or_alloc_SecMap ( Addr ga ) |
| 434 | { |
| 435 | SecMap* sm = shmem__find_SecMap ( ga ); |
| 436 | if (LIKELY(sm)) { |
| 437 | return sm; |
| 438 | } else { |
| 439 | /* create a new one */ |
| 440 | Addr gaKey = shmem__round_to_SecMap_base(ga); |
| 441 | sm = shmem__alloc_SecMap(); |
| 442 | tl_assert(sm); |
| 443 | VG_(addToFM)( map_shmem, (UWord)gaKey, (UWord)sm ); |
| 444 | return sm; |
| 445 | } |
| 446 | } |
| 447 | |
| 448 | |
| 449 | /* ------------ LineF and LineZ related ------------ */ |
| 450 | |
| 451 | static void rcinc_LineF ( LineF* lineF ) { |
| 452 | UWord i; |
| 453 | tl_assert(lineF->inUse); |
| 454 | for (i = 0; i < N_LINE_ARANGE; i++) |
| 455 | rcinc(lineF->w64s[i]); |
| 456 | } |
| 457 | |
| 458 | static void rcdec_LineF ( LineF* lineF ) { |
| 459 | UWord i; |
| 460 | tl_assert(lineF->inUse); |
| 461 | for (i = 0; i < N_LINE_ARANGE; i++) |
| 462 | rcdec(lineF->w64s[i]); |
| 463 | } |
| 464 | |
| 465 | static void rcinc_LineZ ( LineZ* lineZ ) { |
| 466 | tl_assert(lineZ->dict[0] != SVal_INVALID); |
| 467 | rcinc(lineZ->dict[0]); |
| 468 | if (lineZ->dict[1] != SVal_INVALID) rcinc(lineZ->dict[1]); |
| 469 | if (lineZ->dict[2] != SVal_INVALID) rcinc(lineZ->dict[2]); |
| 470 | if (lineZ->dict[3] != SVal_INVALID) rcinc(lineZ->dict[3]); |
| 471 | } |
| 472 | |
| 473 | static void rcdec_LineZ ( LineZ* lineZ ) { |
| 474 | tl_assert(lineZ->dict[0] != SVal_INVALID); |
| 475 | rcdec(lineZ->dict[0]); |
| 476 | if (lineZ->dict[1] != SVal_INVALID) rcdec(lineZ->dict[1]); |
| 477 | if (lineZ->dict[2] != SVal_INVALID) rcdec(lineZ->dict[2]); |
| 478 | if (lineZ->dict[3] != SVal_INVALID) rcdec(lineZ->dict[3]); |
| 479 | } |
| 480 | |
| 481 | inline |
| 482 | static void write_twobit_array ( UChar* arr, UWord ix, UWord b2 ) { |
| 483 | Word bix, shft, mask, prep; |
| 484 | tl_assert(ix >= 0); |
| 485 | bix = ix >> 2; |
| 486 | shft = 2 * (ix & 3); /* 0, 2, 4 or 6 */ |
| 487 | mask = 3 << shft; |
| 488 | prep = b2 << shft; |
| 489 | arr[bix] = (arr[bix] & ~mask) | prep; |
| 490 | } |
| 491 | |
| 492 | inline |
| 493 | static UWord read_twobit_array ( UChar* arr, UWord ix ) { |
| 494 | Word bix, shft; |
| 495 | tl_assert(ix >= 0); |
| 496 | bix = ix >> 2; |
| 497 | shft = 2 * (ix & 3); /* 0, 2, 4 or 6 */ |
| 498 | return (arr[bix] >> shft) & 3; |
| 499 | } |
| 500 | |
| 501 | /* Given address 'tag', find either the Z or F line containing relevant |
| 502 | data, so it can be read into the cache. |
| 503 | */ |
| 504 | static void find_ZF_for_reading ( /*OUT*/LineZ** zp, |
| 505 | /*OUT*/LineF** fp, Addr tag ) { |
| 506 | LineZ* lineZ; |
| 507 | LineF* lineF; |
| 508 | UWord zix; |
| 509 | SecMap* sm = shmem__find_or_alloc_SecMap(tag); |
| 510 | UWord smoff = shmem__get_SecMap_offset(tag); |
| 511 | /* since smoff is derived from a valid tag, it should be |
| 512 | cacheline-aligned. */ |
| 513 | tl_assert(0 == (smoff & (N_LINE_ARANGE - 1))); |
| 514 | zix = smoff >> N_LINE_BITS; |
| 515 | tl_assert(zix < N_SECMAP_ZLINES); |
| 516 | lineZ = &sm->linesZ[zix]; |
| 517 | lineF = NULL; |
| 518 | if (lineZ->dict[0] == SVal_INVALID) { |
| 519 | UInt fix = (UInt)lineZ->dict[1]; |
| 520 | tl_assert(sm->linesF); |
| 521 | tl_assert(sm->linesF_size > 0); |
| 522 | tl_assert(fix >= 0 && fix < sm->linesF_size); |
| 523 | lineF = &sm->linesF[fix]; |
| 524 | tl_assert(lineF->inUse); |
| 525 | lineZ = NULL; |
| 526 | } |
| 527 | *zp = lineZ; |
| 528 | *fp = lineF; |
| 529 | } |
| 530 | |
| 531 | /* Given address 'tag', return the relevant SecMap and the index of |
| 532 | the LineZ within it, in the expectation that the line is to be |
| 533 | overwritten. Regardless of whether 'tag' is currently associated |
| 534 | with a Z or F representation, to rcdec on the current |
| 535 | representation, in recognition of the fact that the contents are |
| 536 | just about to be overwritten. */ |
| 537 | static __attribute__((noinline)) |
| 538 | void find_Z_for_writing ( /*OUT*/SecMap** smp, |
| 539 | /*OUT*/Word* zixp, |
| 540 | Addr tag ) { |
| 541 | LineZ* lineZ; |
| 542 | LineF* lineF; |
| 543 | UWord zix; |
| 544 | SecMap* sm = shmem__find_or_alloc_SecMap(tag); |
| 545 | UWord smoff = shmem__get_SecMap_offset(tag); |
| 546 | /* since smoff is derived from a valid tag, it should be |
| 547 | cacheline-aligned. */ |
| 548 | tl_assert(0 == (smoff & (N_LINE_ARANGE - 1))); |
| 549 | zix = smoff >> N_LINE_BITS; |
| 550 | tl_assert(zix < N_SECMAP_ZLINES); |
| 551 | lineZ = &sm->linesZ[zix]; |
| 552 | lineF = NULL; |
| 553 | /* re RCs, we are freeing up this LineZ/LineF so that new data can |
| 554 | be parked in it. Hence have to rcdec it accordingly. */ |
| 555 | /* If lineZ has an associated lineF, free it up. */ |
| 556 | if (lineZ->dict[0] == SVal_INVALID) { |
| 557 | UInt fix = (UInt)lineZ->dict[1]; |
| 558 | tl_assert(sm->linesF); |
| 559 | tl_assert(sm->linesF_size > 0); |
| 560 | tl_assert(fix >= 0 && fix < sm->linesF_size); |
| 561 | lineF = &sm->linesF[fix]; |
| 562 | tl_assert(lineF->inUse); |
| 563 | rcdec_LineF(lineF); |
| 564 | lineF->inUse = False; |
| 565 | } else { |
| 566 | rcdec_LineZ(lineZ); |
| 567 | } |
| 568 | *smp = sm; |
| 569 | *zixp = zix; |
| 570 | } |
| 571 | |
| 572 | static __attribute__((noinline)) |
| 573 | void alloc_F_for_writing ( /*MOD*/SecMap* sm, /*OUT*/Word* fixp ) { |
| 574 | UInt i, new_size; |
| 575 | LineF* nyu; |
| 576 | |
| 577 | if (sm->linesF) { |
| 578 | tl_assert(sm->linesF_size > 0); |
| 579 | } else { |
| 580 | tl_assert(sm->linesF_size == 0); |
| 581 | } |
| 582 | |
| 583 | if (sm->linesF) { |
| 584 | for (i = 0; i < sm->linesF_size; i++) { |
| 585 | if (!sm->linesF[i].inUse) { |
| 586 | *fixp = (Word)i; |
| 587 | return; |
| 588 | } |
| 589 | } |
| 590 | } |
| 591 | |
| 592 | /* No free F line found. Expand existing array and try again. */ |
| 593 | new_size = sm->linesF_size==0 ? 1 : 2 * sm->linesF_size; |
| 594 | nyu = HG_(zalloc)( "libhb.aFfw.1 (LineF storage)", |
| 595 | new_size * sizeof(LineF) ); |
| 596 | tl_assert(nyu); |
| 597 | |
| 598 | stats__secmap_linesF_allocd += (new_size - sm->linesF_size); |
| 599 | stats__secmap_linesF_bytes += (new_size - sm->linesF_size) |
| 600 | * sizeof(LineF); |
| 601 | |
| 602 | if (0) |
| 603 | VG_(printf)("SM %p: expand F array from %d to %d\n", |
| 604 | sm, (Int)sm->linesF_size, new_size); |
| 605 | |
| 606 | for (i = 0; i < new_size; i++) |
| 607 | nyu[i].inUse = False; |
| 608 | |
| 609 | if (sm->linesF) { |
| 610 | for (i = 0; i < sm->linesF_size; i++) { |
| 611 | tl_assert(sm->linesF[i].inUse); |
| 612 | nyu[i] = sm->linesF[i]; |
| 613 | } |
| 614 | VG_(memset)(sm->linesF, 0, sm->linesF_size * sizeof(LineF) ); |
| 615 | HG_(free)(sm->linesF); |
| 616 | } |
| 617 | |
| 618 | sm->linesF = nyu; |
| 619 | sm->linesF_size = new_size; |
| 620 | |
| 621 | for (i = 0; i < sm->linesF_size; i++) { |
| 622 | if (!sm->linesF[i].inUse) { |
| 623 | *fixp = (Word)i; |
| 624 | return; |
| 625 | } |
| 626 | } |
| 627 | |
| 628 | /*NOTREACHED*/ |
| 629 | tl_assert(0); |
| 630 | } |
| 631 | |
| 632 | |
| 633 | /* ------------ CacheLine and implicit-tree related ------------ */ |
| 634 | |
| 635 | __attribute__((unused)) |
| 636 | static void pp_CacheLine ( CacheLine* cl ) { |
| 637 | Word i; |
| 638 | if (!cl) { |
| 639 | VG_(printf)("%s","pp_CacheLine(NULL)\n"); |
| 640 | return; |
| 641 | } |
| 642 | for (i = 0; i < N_LINE_TREES; i++) |
| 643 | VG_(printf)(" descr: %04lx\n", (UWord)cl->descrs[i]); |
| 644 | for (i = 0; i < N_LINE_ARANGE; i++) |
| 645 | VG_(printf)(" sval: %08lx\n", (UWord)cl->svals[i]); |
| 646 | } |
| 647 | |
| 648 | static UChar descr_to_validbits ( UShort descr ) |
| 649 | { |
| 650 | /* a.k.a Party Time for gcc's constant folder */ |
| 651 | # define DESCR(b8_7, b8_6, b8_5, b8_4, b8_3, b8_2, b8_1, b8_0, \ |
| 652 | b16_3, b32_1, b16_2, b64, b16_1, b32_0, b16_0) \ |
| 653 | ( (UShort) ( ( (b8_7) << 14) | ( (b8_6) << 13) | \ |
| 654 | ( (b8_5) << 12) | ( (b8_4) << 11) | \ |
| 655 | ( (b8_3) << 10) | ( (b8_2) << 9) | \ |
| 656 | ( (b8_1) << 8) | ( (b8_0) << 7) | \ |
| 657 | ( (b16_3) << 6) | ( (b32_1) << 5) | \ |
| 658 | ( (b16_2) << 4) | ( (b64) << 3) | \ |
| 659 | ( (b16_1) << 2) | ( (b32_0) << 1) | \ |
| 660 | ( (b16_0) << 0) ) ) |
| 661 | |
| 662 | # define BYTE(bit7, bit6, bit5, bit4, bit3, bit2, bit1, bit0) \ |
| 663 | ( (UChar) ( ( (bit7) << 7) | ( (bit6) << 6) | \ |
| 664 | ( (bit5) << 5) | ( (bit4) << 4) | \ |
| 665 | ( (bit3) << 3) | ( (bit2) << 2) | \ |
| 666 | ( (bit1) << 1) | ( (bit0) << 0) ) ) |
| 667 | |
| 668 | /* these should all get folded out at compile time */ |
| 669 | tl_assert(DESCR(1,0,0,0,0,0,0,0, 0,0,0, 0, 0,0,0) == TREE_DESCR_8_7); |
| 670 | tl_assert(DESCR(0,0,0,0,0,0,0,1, 0,0,0, 0, 0,0,0) == TREE_DESCR_8_0); |
| 671 | tl_assert(DESCR(0,0,0,0,0,0,0,0, 1,0,0, 0, 0,0,0) == TREE_DESCR_16_3); |
| 672 | tl_assert(DESCR(0,0,0,0,0,0,0,0, 0,1,0, 0, 0,0,0) == TREE_DESCR_32_1); |
| 673 | tl_assert(DESCR(0,0,0,0,0,0,0,0, 0,0,1, 0, 0,0,0) == TREE_DESCR_16_2); |
| 674 | tl_assert(DESCR(0,0,0,0,0,0,0,0, 0,0,0, 1, 0,0,0) == TREE_DESCR_64); |
| 675 | tl_assert(DESCR(0,0,0,0,0,0,0,0, 0,0,0, 0, 1,0,0) == TREE_DESCR_16_1); |
| 676 | tl_assert(DESCR(0,0,0,0,0,0,0,0, 0,0,0, 0, 0,1,0) == TREE_DESCR_32_0); |
| 677 | tl_assert(DESCR(0,0,0,0,0,0,0,0, 0,0,0, 0, 0,0,1) == TREE_DESCR_16_0); |
| 678 | |
| 679 | switch (descr) { |
| 680 | /* |
| 681 | +--------------------------------- TREE_DESCR_8_7 |
| 682 | | +------------------- TREE_DESCR_8_0 |
| 683 | | | +---------------- TREE_DESCR_16_3 |
| 684 | | | | +-------------- TREE_DESCR_32_1 |
| 685 | | | | | +------------ TREE_DESCR_16_2 |
| 686 | | | | | | +--------- TREE_DESCR_64 |
| 687 | | | | | | | +------ TREE_DESCR_16_1 |
| 688 | | | | | | | | +---- TREE_DESCR_32_0 |
| 689 | | | | | | | | | +-- TREE_DESCR_16_0 |
| 690 | | | | | | | | | | |
| 691 | | | | | | | | | | GRANULARITY, 7 -> 0 */ |
| 692 | case DESCR(1,1,1,1,1,1,1,1, 0,0,0, 0, 0,0,0): /* 8 8 8 8 8 8 8 8 */ |
| 693 | return BYTE(1,1,1,1,1,1,1,1); |
| 694 | case DESCR(1,1,0,0,1,1,1,1, 0,0,1, 0, 0,0,0): /* 8 8 16 8 8 8 8 */ |
| 695 | return BYTE(1,1,0,1,1,1,1,1); |
| 696 | case DESCR(0,0,1,1,1,1,1,1, 1,0,0, 0, 0,0,0): /* 16 8 8 8 8 8 8 */ |
| 697 | return BYTE(0,1,1,1,1,1,1,1); |
| 698 | case DESCR(0,0,0,0,1,1,1,1, 1,0,1, 0, 0,0,0): /* 16 16 8 8 8 8 */ |
| 699 | return BYTE(0,1,0,1,1,1,1,1); |
| 700 | |
| 701 | case DESCR(1,1,1,1,1,1,0,0, 0,0,0, 0, 0,0,1): /* 8 8 8 8 8 8 16 */ |
| 702 | return BYTE(1,1,1,1,1,1,0,1); |
| 703 | case DESCR(1,1,0,0,1,1,0,0, 0,0,1, 0, 0,0,1): /* 8 8 16 8 8 16 */ |
| 704 | return BYTE(1,1,0,1,1,1,0,1); |
| 705 | case DESCR(0,0,1,1,1,1,0,0, 1,0,0, 0, 0,0,1): /* 16 8 8 8 8 16 */ |
| 706 | return BYTE(0,1,1,1,1,1,0,1); |
| 707 | case DESCR(0,0,0,0,1,1,0,0, 1,0,1, 0, 0,0,1): /* 16 16 8 8 16 */ |
| 708 | return BYTE(0,1,0,1,1,1,0,1); |
| 709 | |
| 710 | case DESCR(1,1,1,1,0,0,1,1, 0,0,0, 0, 1,0,0): /* 8 8 8 8 16 8 8 */ |
| 711 | return BYTE(1,1,1,1,0,1,1,1); |
| 712 | case DESCR(1,1,0,0,0,0,1,1, 0,0,1, 0, 1,0,0): /* 8 8 16 16 8 8 */ |
| 713 | return BYTE(1,1,0,1,0,1,1,1); |
| 714 | case DESCR(0,0,1,1,0,0,1,1, 1,0,0, 0, 1,0,0): /* 16 8 8 16 8 8 */ |
| 715 | return BYTE(0,1,1,1,0,1,1,1); |
| 716 | case DESCR(0,0,0,0,0,0,1,1, 1,0,1, 0, 1,0,0): /* 16 16 16 8 8 */ |
| 717 | return BYTE(0,1,0,1,0,1,1,1); |
| 718 | |
| 719 | case DESCR(1,1,1,1,0,0,0,0, 0,0,0, 0, 1,0,1): /* 8 8 8 8 16 16 */ |
| 720 | return BYTE(1,1,1,1,0,1,0,1); |
| 721 | case DESCR(1,1,0,0,0,0,0,0, 0,0,1, 0, 1,0,1): /* 8 8 16 16 16 */ |
| 722 | return BYTE(1,1,0,1,0,1,0,1); |
| 723 | case DESCR(0,0,1,1,0,0,0,0, 1,0,0, 0, 1,0,1): /* 16 8 8 16 16 */ |
| 724 | return BYTE(0,1,1,1,0,1,0,1); |
| 725 | case DESCR(0,0,0,0,0,0,0,0, 1,0,1, 0, 1,0,1): /* 16 16 16 16 */ |
| 726 | return BYTE(0,1,0,1,0,1,0,1); |
| 727 | |
| 728 | case DESCR(0,0,0,0,1,1,1,1, 0,1,0, 0, 0,0,0): /* 32 8 8 8 8 */ |
| 729 | return BYTE(0,0,0,1,1,1,1,1); |
| 730 | case DESCR(0,0,0,0,1,1,0,0, 0,1,0, 0, 0,0,1): /* 32 8 8 16 */ |
| 731 | return BYTE(0,0,0,1,1,1,0,1); |
| 732 | case DESCR(0,0,0,0,0,0,1,1, 0,1,0, 0, 1,0,0): /* 32 16 8 8 */ |
| 733 | return BYTE(0,0,0,1,0,1,1,1); |
| 734 | case DESCR(0,0,0,0,0,0,0,0, 0,1,0, 0, 1,0,1): /* 32 16 16 */ |
| 735 | return BYTE(0,0,0,1,0,1,0,1); |
| 736 | |
| 737 | case DESCR(1,1,1,1,0,0,0,0, 0,0,0, 0, 0,1,0): /* 8 8 8 8 32 */ |
| 738 | return BYTE(1,1,1,1,0,0,0,1); |
| 739 | case DESCR(1,1,0,0,0,0,0,0, 0,0,1, 0, 0,1,0): /* 8 8 16 32 */ |
| 740 | return BYTE(1,1,0,1,0,0,0,1); |
| 741 | case DESCR(0,0,1,1,0,0,0,0, 1,0,0, 0, 0,1,0): /* 16 8 8 32 */ |
| 742 | return BYTE(0,1,1,1,0,0,0,1); |
| 743 | case DESCR(0,0,0,0,0,0,0,0, 1,0,1, 0, 0,1,0): /* 16 16 32 */ |
| 744 | return BYTE(0,1,0,1,0,0,0,1); |
| 745 | |
| 746 | case DESCR(0,0,0,0,0,0,0,0, 0,1,0, 0, 0,1,0): /* 32 32 */ |
| 747 | return BYTE(0,0,0,1,0,0,0,1); |
| 748 | |
| 749 | case DESCR(0,0,0,0,0,0,0,0, 0,0,0, 1, 0,0,0): /* 64 */ |
| 750 | return BYTE(0,0,0,0,0,0,0,1); |
| 751 | |
| 752 | default: return BYTE(0,0,0,0,0,0,0,0); |
| 753 | /* INVALID - any valid descr produces at least one |
| 754 | valid bit in tree[0..7]*/ |
| 755 | } |
| 756 | /* NOTREACHED*/ |
| 757 | tl_assert(0); |
| 758 | |
| 759 | # undef DESCR |
| 760 | # undef BYTE |
| 761 | } |
| 762 | |
| 763 | __attribute__((unused)) |
| 764 | static Bool is_sane_Descr ( UShort descr ) { |
| 765 | return descr_to_validbits(descr) != 0; |
| 766 | } |
| 767 | |
| 768 | static void sprintf_Descr ( /*OUT*/HChar* dst, UShort descr ) { |
| 769 | VG_(sprintf)(dst, |
| 770 | "%d%d%d%d%d%d%d%d %d%d%d %d %d%d%d", |
| 771 | (Int)((descr & TREE_DESCR_8_7) ? 1 : 0), |
| 772 | (Int)((descr & TREE_DESCR_8_6) ? 1 : 0), |
| 773 | (Int)((descr & TREE_DESCR_8_5) ? 1 : 0), |
| 774 | (Int)((descr & TREE_DESCR_8_4) ? 1 : 0), |
| 775 | (Int)((descr & TREE_DESCR_8_3) ? 1 : 0), |
| 776 | (Int)((descr & TREE_DESCR_8_2) ? 1 : 0), |
| 777 | (Int)((descr & TREE_DESCR_8_1) ? 1 : 0), |
| 778 | (Int)((descr & TREE_DESCR_8_0) ? 1 : 0), |
| 779 | (Int)((descr & TREE_DESCR_16_3) ? 1 : 0), |
| 780 | (Int)((descr & TREE_DESCR_32_1) ? 1 : 0), |
| 781 | (Int)((descr & TREE_DESCR_16_2) ? 1 : 0), |
| 782 | (Int)((descr & TREE_DESCR_64) ? 1 : 0), |
| 783 | (Int)((descr & TREE_DESCR_16_1) ? 1 : 0), |
| 784 | (Int)((descr & TREE_DESCR_32_0) ? 1 : 0), |
| 785 | (Int)((descr & TREE_DESCR_16_0) ? 1 : 0) |
| 786 | ); |
| 787 | } |
| 788 | static void sprintf_Byte ( /*OUT*/HChar* dst, UChar byte ) { |
| 789 | VG_(sprintf)(dst, "%d%d%d%d%d%d%d%d", |
| 790 | (Int)((byte & 128) ? 1 : 0), |
| 791 | (Int)((byte & 64) ? 1 : 0), |
| 792 | (Int)((byte & 32) ? 1 : 0), |
| 793 | (Int)((byte & 16) ? 1 : 0), |
| 794 | (Int)((byte & 8) ? 1 : 0), |
| 795 | (Int)((byte & 4) ? 1 : 0), |
| 796 | (Int)((byte & 2) ? 1 : 0), |
| 797 | (Int)((byte & 1) ? 1 : 0) |
| 798 | ); |
| 799 | } |
| 800 | |
| 801 | static Bool is_sane_Descr_and_Tree ( UShort descr, SVal* tree ) { |
| 802 | Word i; |
| 803 | UChar validbits = descr_to_validbits(descr); |
| 804 | HChar buf[128], buf2[128]; |
| 805 | if (validbits == 0) |
| 806 | goto bad; |
| 807 | for (i = 0; i < 8; i++) { |
| 808 | if (validbits & (1<<i)) { |
| 809 | if (tree[i] == SVal_INVALID) |
| 810 | goto bad; |
| 811 | } else { |
| 812 | if (tree[i] != SVal_INVALID) |
| 813 | goto bad; |
| 814 | } |
| 815 | } |
| 816 | return True; |
| 817 | bad: |
| 818 | sprintf_Descr( buf, descr ); |
| 819 | sprintf_Byte( buf2, validbits ); |
| 820 | VG_(printf)("%s","is_sane_Descr_and_Tree: bad tree {\n"); |
| 821 | VG_(printf)(" validbits 0x%02lx %s\n", (UWord)validbits, buf2); |
| 822 | VG_(printf)(" descr 0x%04lx %s\n", (UWord)descr, buf); |
| 823 | for (i = 0; i < 8; i++) |
| 824 | VG_(printf)(" [%ld] 0x%016llx\n", i, tree[i]); |
| 825 | VG_(printf)("%s","}\n"); |
| 826 | return 0; |
| 827 | } |
| 828 | |
| 829 | static Bool is_sane_CacheLine ( CacheLine* cl ) |
| 830 | { |
| 831 | Word tno, cloff; |
| 832 | |
| 833 | if (!cl) goto bad; |
| 834 | |
| 835 | for (tno = 0, cloff = 0; tno < N_LINE_TREES; tno++, cloff += 8) { |
| 836 | UShort descr = cl->descrs[tno]; |
| 837 | SVal* tree = &cl->svals[cloff]; |
| 838 | if (!is_sane_Descr_and_Tree(descr, tree)) |
| 839 | goto bad; |
| 840 | } |
| 841 | tl_assert(cloff == N_LINE_ARANGE); |
| 842 | return True; |
| 843 | bad: |
| 844 | pp_CacheLine(cl); |
| 845 | return False; |
| 846 | } |
| 847 | |
| 848 | static UShort normalise_tree ( /*MOD*/SVal* tree ) |
| 849 | { |
| 850 | UShort descr; |
| 851 | /* pre: incoming tree[0..7] does not have any invalid shvals, in |
| 852 | particular no zeroes. */ |
| 853 | if (UNLIKELY(tree[7] == SVal_INVALID || tree[6] == SVal_INVALID |
| 854 | || tree[5] == SVal_INVALID || tree[4] == SVal_INVALID |
| 855 | || tree[3] == SVal_INVALID || tree[2] == SVal_INVALID |
| 856 | || tree[1] == SVal_INVALID || tree[0] == SVal_INVALID)) |
| 857 | tl_assert(0); |
| 858 | |
| 859 | descr = TREE_DESCR_8_7 | TREE_DESCR_8_6 | TREE_DESCR_8_5 |
| 860 | | TREE_DESCR_8_4 | TREE_DESCR_8_3 | TREE_DESCR_8_2 |
| 861 | | TREE_DESCR_8_1 | TREE_DESCR_8_0; |
| 862 | /* build 16-bit layer */ |
| 863 | if (tree[1] == tree[0]) { |
| 864 | tree[1] = SVal_INVALID; |
| 865 | descr &= ~(TREE_DESCR_8_1 | TREE_DESCR_8_0); |
| 866 | descr |= TREE_DESCR_16_0; |
| 867 | } |
| 868 | if (tree[3] == tree[2]) { |
| 869 | tree[3] = SVal_INVALID; |
| 870 | descr &= ~(TREE_DESCR_8_3 | TREE_DESCR_8_2); |
| 871 | descr |= TREE_DESCR_16_1; |
| 872 | } |
| 873 | if (tree[5] == tree[4]) { |
| 874 | tree[5] = SVal_INVALID; |
| 875 | descr &= ~(TREE_DESCR_8_5 | TREE_DESCR_8_4); |
| 876 | descr |= TREE_DESCR_16_2; |
| 877 | } |
| 878 | if (tree[7] == tree[6]) { |
| 879 | tree[7] = SVal_INVALID; |
| 880 | descr &= ~(TREE_DESCR_8_7 | TREE_DESCR_8_6); |
| 881 | descr |= TREE_DESCR_16_3; |
| 882 | } |
| 883 | /* build 32-bit layer */ |
| 884 | if (tree[2] == tree[0] |
| 885 | && (descr & TREE_DESCR_16_1) && (descr & TREE_DESCR_16_0)) { |
| 886 | tree[2] = SVal_INVALID; /* [3,1] must already be SVal_INVALID */ |
| 887 | descr &= ~(TREE_DESCR_16_1 | TREE_DESCR_16_0); |
| 888 | descr |= TREE_DESCR_32_0; |
| 889 | } |
| 890 | if (tree[6] == tree[4] |
| 891 | && (descr & TREE_DESCR_16_3) && (descr & TREE_DESCR_16_2)) { |
| 892 | tree[6] = SVal_INVALID; /* [7,5] must already be SVal_INVALID */ |
| 893 | descr &= ~(TREE_DESCR_16_3 | TREE_DESCR_16_2); |
| 894 | descr |= TREE_DESCR_32_1; |
| 895 | } |
| 896 | /* build 64-bit layer */ |
| 897 | if (tree[4] == tree[0] |
| 898 | && (descr & TREE_DESCR_32_1) && (descr & TREE_DESCR_32_0)) { |
| 899 | tree[4] = SVal_INVALID; /* [7,6,5,3,2,1] must already be SVal_INVALID */ |
| 900 | descr &= ~(TREE_DESCR_32_1 | TREE_DESCR_32_0); |
| 901 | descr |= TREE_DESCR_64; |
| 902 | } |
| 903 | return descr; |
| 904 | } |
| 905 | |
| 906 | /* This takes a cacheline where all the data is at the leaves |
| 907 | (w8[..]) and builds a correctly normalised tree. */ |
| 908 | static void normalise_CacheLine ( /*MOD*/CacheLine* cl ) |
| 909 | { |
| 910 | Word tno, cloff; |
| 911 | for (tno = 0, cloff = 0; tno < N_LINE_TREES; tno++, cloff += 8) { |
| 912 | SVal* tree = &cl->svals[cloff]; |
| 913 | cl->descrs[tno] = normalise_tree( tree ); |
| 914 | } |
| 915 | tl_assert(cloff == N_LINE_ARANGE); |
| 916 | if (SCE_CACHELINE) |
| 917 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 918 | stats__cline_normalises++; |
| 919 | } |
| 920 | |
| 921 | |
| 922 | typedef struct { UChar count; SVal sval; } CountedSVal; |
| 923 | |
| 924 | static |
| 925 | void sequentialise_CacheLine ( /*OUT*/CountedSVal* dst, |
| 926 | /*OUT*/Word* dstUsedP, |
| 927 | Word nDst, CacheLine* src ) |
| 928 | { |
| 929 | Word tno, cloff, dstUsed; |
| 930 | |
| 931 | tl_assert(nDst == N_LINE_ARANGE); |
| 932 | dstUsed = 0; |
| 933 | |
| 934 | for (tno = 0, cloff = 0; tno < N_LINE_TREES; tno++, cloff += 8) { |
| 935 | UShort descr = src->descrs[tno]; |
| 936 | SVal* tree = &src->svals[cloff]; |
| 937 | |
| 938 | /* sequentialise the tree described by (descr,tree). */ |
| 939 | # define PUT(_n,_v) \ |
| 940 | do { dst[dstUsed ].count = (_n); \ |
| 941 | dst[dstUsed++].sval = (_v); \ |
| 942 | } while (0) |
| 943 | |
| 944 | /* byte 0 */ |
| 945 | if (descr & TREE_DESCR_64) PUT(8, tree[0]); else |
| 946 | if (descr & TREE_DESCR_32_0) PUT(4, tree[0]); else |
| 947 | if (descr & TREE_DESCR_16_0) PUT(2, tree[0]); else |
| 948 | if (descr & TREE_DESCR_8_0) PUT(1, tree[0]); |
| 949 | /* byte 1 */ |
| 950 | if (descr & TREE_DESCR_8_1) PUT(1, tree[1]); |
| 951 | /* byte 2 */ |
| 952 | if (descr & TREE_DESCR_16_1) PUT(2, tree[2]); else |
| 953 | if (descr & TREE_DESCR_8_2) PUT(1, tree[2]); |
| 954 | /* byte 3 */ |
| 955 | if (descr & TREE_DESCR_8_3) PUT(1, tree[3]); |
| 956 | /* byte 4 */ |
| 957 | if (descr & TREE_DESCR_32_1) PUT(4, tree[4]); else |
| 958 | if (descr & TREE_DESCR_16_2) PUT(2, tree[4]); else |
| 959 | if (descr & TREE_DESCR_8_4) PUT(1, tree[4]); |
| 960 | /* byte 5 */ |
| 961 | if (descr & TREE_DESCR_8_5) PUT(1, tree[5]); |
| 962 | /* byte 6 */ |
| 963 | if (descr & TREE_DESCR_16_3) PUT(2, tree[6]); else |
| 964 | if (descr & TREE_DESCR_8_6) PUT(1, tree[6]); |
| 965 | /* byte 7 */ |
| 966 | if (descr & TREE_DESCR_8_7) PUT(1, tree[7]); |
| 967 | |
| 968 | # undef PUT |
| 969 | /* END sequentialise the tree described by (descr,tree). */ |
| 970 | |
| 971 | } |
| 972 | tl_assert(cloff == N_LINE_ARANGE); |
| 973 | tl_assert(dstUsed <= nDst); |
| 974 | |
| 975 | *dstUsedP = dstUsed; |
| 976 | } |
| 977 | |
| 978 | /* Write the cacheline 'wix' to backing store. Where it ends up |
| 979 | is determined by its tag field. */ |
| 980 | static __attribute__((noinline)) void cacheline_wback ( UWord wix ) |
| 981 | { |
| 982 | Word i, j, k, m; |
| 983 | Addr tag; |
| 984 | SecMap* sm; |
| 985 | CacheLine* cl; |
| 986 | LineZ* lineZ; |
| 987 | LineF* lineF; |
| 988 | Word zix, fix, csvalsUsed; |
| 989 | CountedSVal csvals[N_LINE_ARANGE]; |
| 990 | SVal sv; |
| 991 | |
| 992 | if (0) |
| 993 | VG_(printf)("scache wback line %d\n", (Int)wix); |
| 994 | |
| 995 | tl_assert(wix >= 0 && wix < N_WAY_NENT); |
| 996 | |
| 997 | tag = cache_shmem.tags0[wix]; |
| 998 | cl = &cache_shmem.lyns0[wix]; |
| 999 | |
| 1000 | /* The cache line may have been invalidated; if so, ignore it. */ |
| 1001 | if (!is_valid_scache_tag(tag)) |
| 1002 | return; |
| 1003 | |
| 1004 | /* Where are we going to put it? */ |
| 1005 | sm = NULL; |
| 1006 | lineZ = NULL; |
| 1007 | lineF = NULL; |
| 1008 | zix = fix = -1; |
| 1009 | |
| 1010 | /* find the Z line to write in and rcdec it or the associated F |
| 1011 | line. */ |
| 1012 | find_Z_for_writing( &sm, &zix, tag ); |
| 1013 | |
| 1014 | tl_assert(sm); |
| 1015 | tl_assert(zix >= 0 && zix < N_SECMAP_ZLINES); |
| 1016 | lineZ = &sm->linesZ[zix]; |
| 1017 | |
| 1018 | /* Generate the data to be stored */ |
| 1019 | if (SCE_CACHELINE) |
| 1020 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 1021 | |
| 1022 | csvalsUsed = -1; |
| 1023 | sequentialise_CacheLine( csvals, &csvalsUsed, |
| 1024 | N_LINE_ARANGE, cl ); |
| 1025 | tl_assert(csvalsUsed >= 1 && csvalsUsed <= N_LINE_ARANGE); |
| 1026 | if (0) VG_(printf)("%lu ", csvalsUsed); |
| 1027 | |
| 1028 | lineZ->dict[0] = lineZ->dict[1] |
| 1029 | = lineZ->dict[2] = lineZ->dict[3] = SVal_INVALID; |
| 1030 | |
| 1031 | /* i indexes actual shadow values, k is cursor in csvals */ |
| 1032 | i = 0; |
| 1033 | for (k = 0; k < csvalsUsed; k++) { |
| 1034 | |
| 1035 | sv = csvals[k].sval; |
| 1036 | if (SCE_SVALS) |
| 1037 | tl_assert(csvals[k].count >= 1 && csvals[k].count <= 8); |
| 1038 | /* do we already have it? */ |
| 1039 | if (sv == lineZ->dict[0]) { j = 0; goto dict_ok; } |
| 1040 | if (sv == lineZ->dict[1]) { j = 1; goto dict_ok; } |
| 1041 | if (sv == lineZ->dict[2]) { j = 2; goto dict_ok; } |
| 1042 | if (sv == lineZ->dict[3]) { j = 3; goto dict_ok; } |
| 1043 | /* no. look for a free slot. */ |
| 1044 | if (SCE_SVALS) |
| 1045 | tl_assert(sv != SVal_INVALID); |
| 1046 | if (lineZ->dict[0] |
| 1047 | == SVal_INVALID) { lineZ->dict[0] = sv; j = 0; goto dict_ok; } |
| 1048 | if (lineZ->dict[1] |
| 1049 | == SVal_INVALID) { lineZ->dict[1] = sv; j = 1; goto dict_ok; } |
| 1050 | if (lineZ->dict[2] |
| 1051 | == SVal_INVALID) { lineZ->dict[2] = sv; j = 2; goto dict_ok; } |
| 1052 | if (lineZ->dict[3] |
| 1053 | == SVal_INVALID) { lineZ->dict[3] = sv; j = 3; goto dict_ok; } |
| 1054 | break; /* we'll have to use the f rep */ |
| 1055 | dict_ok: |
| 1056 | m = csvals[k].count; |
| 1057 | if (m == 8) { |
| 1058 | write_twobit_array( lineZ->ix2s, i+0, j ); |
| 1059 | write_twobit_array( lineZ->ix2s, i+1, j ); |
| 1060 | write_twobit_array( lineZ->ix2s, i+2, j ); |
| 1061 | write_twobit_array( lineZ->ix2s, i+3, j ); |
| 1062 | write_twobit_array( lineZ->ix2s, i+4, j ); |
| 1063 | write_twobit_array( lineZ->ix2s, i+5, j ); |
| 1064 | write_twobit_array( lineZ->ix2s, i+6, j ); |
| 1065 | write_twobit_array( lineZ->ix2s, i+7, j ); |
| 1066 | i += 8; |
| 1067 | } |
| 1068 | else if (m == 4) { |
| 1069 | write_twobit_array( lineZ->ix2s, i+0, j ); |
| 1070 | write_twobit_array( lineZ->ix2s, i+1, j ); |
| 1071 | write_twobit_array( lineZ->ix2s, i+2, j ); |
| 1072 | write_twobit_array( lineZ->ix2s, i+3, j ); |
| 1073 | i += 4; |
| 1074 | } |
| 1075 | else if (m == 1) { |
| 1076 | write_twobit_array( lineZ->ix2s, i+0, j ); |
| 1077 | i += 1; |
| 1078 | } |
| 1079 | else if (m == 2) { |
| 1080 | write_twobit_array( lineZ->ix2s, i+0, j ); |
| 1081 | write_twobit_array( lineZ->ix2s, i+1, j ); |
| 1082 | i += 2; |
| 1083 | } |
| 1084 | else { |
| 1085 | tl_assert(0); /* 8 4 2 or 1 are the only legitimate values for m */ |
| 1086 | } |
| 1087 | |
| 1088 | } |
| 1089 | |
| 1090 | if (LIKELY(i == N_LINE_ARANGE)) { |
| 1091 | /* Construction of the compressed representation was |
| 1092 | successful. */ |
| 1093 | rcinc_LineZ(lineZ); |
| 1094 | stats__cache_Z_wbacks++; |
| 1095 | } else { |
| 1096 | /* Cannot use the compressed(z) representation. Use the full(f) |
| 1097 | rep instead. */ |
| 1098 | tl_assert(i >= 0 && i < N_LINE_ARANGE); |
| 1099 | alloc_F_for_writing( sm, &fix ); |
| 1100 | tl_assert(sm->linesF); |
| 1101 | tl_assert(sm->linesF_size > 0); |
| 1102 | tl_assert(fix >= 0 && fix < (Word)sm->linesF_size); |
| 1103 | lineF = &sm->linesF[fix]; |
| 1104 | tl_assert(!lineF->inUse); |
| 1105 | lineZ->dict[0] = lineZ->dict[2] = lineZ->dict[3] = SVal_INVALID; |
| 1106 | lineZ->dict[1] = (SVal)fix; |
| 1107 | lineF->inUse = True; |
| 1108 | i = 0; |
| 1109 | for (k = 0; k < csvalsUsed; k++) { |
| 1110 | if (SCE_SVALS) |
| 1111 | tl_assert(csvals[k].count >= 1 && csvals[k].count <= 8); |
| 1112 | sv = csvals[k].sval; |
| 1113 | if (SCE_SVALS) |
| 1114 | tl_assert(sv != SVal_INVALID); |
| 1115 | for (m = csvals[k].count; m > 0; m--) { |
| 1116 | lineF->w64s[i] = sv; |
| 1117 | i++; |
| 1118 | } |
| 1119 | } |
| 1120 | tl_assert(i == N_LINE_ARANGE); |
| 1121 | rcinc_LineF(lineF); |
| 1122 | stats__cache_F_wbacks++; |
| 1123 | } |
| 1124 | |
| 1125 | //if (anyShared) |
| 1126 | // sm->mbHasShared = True; |
| 1127 | |
| 1128 | /* mb_tidy_one_cacheline(); */ |
| 1129 | } |
| 1130 | |
| 1131 | /* Fetch the cacheline 'wix' from the backing store. The tag |
| 1132 | associated with 'wix' is assumed to have already been filled in; |
| 1133 | hence that is used to determine where in the backing store to read |
| 1134 | from. */ |
| 1135 | static __attribute__((noinline)) void cacheline_fetch ( UWord wix ) |
| 1136 | { |
| 1137 | Word i; |
| 1138 | Addr tag; |
| 1139 | CacheLine* cl; |
| 1140 | LineZ* lineZ; |
| 1141 | LineF* lineF; |
| 1142 | |
| 1143 | if (0) |
| 1144 | VG_(printf)("scache fetch line %d\n", (Int)wix); |
| 1145 | |
| 1146 | tl_assert(wix >= 0 && wix < N_WAY_NENT); |
| 1147 | |
| 1148 | tag = cache_shmem.tags0[wix]; |
| 1149 | cl = &cache_shmem.lyns0[wix]; |
| 1150 | |
| 1151 | /* reject nonsense requests */ |
| 1152 | tl_assert(is_valid_scache_tag(tag)); |
| 1153 | |
| 1154 | lineZ = NULL; |
| 1155 | lineF = NULL; |
| 1156 | find_ZF_for_reading( &lineZ, &lineF, tag ); |
| 1157 | tl_assert( (lineZ && !lineF) || (!lineZ && lineF) ); |
| 1158 | |
| 1159 | /* expand the data into the bottom layer of the tree, then get |
| 1160 | cacheline_normalise to build the descriptor array. */ |
| 1161 | if (lineF) { |
| 1162 | tl_assert(lineF->inUse); |
| 1163 | for (i = 0; i < N_LINE_ARANGE; i++) { |
| 1164 | cl->svals[i] = lineF->w64s[i]; |
| 1165 | } |
| 1166 | stats__cache_F_fetches++; |
| 1167 | } else { |
| 1168 | for (i = 0; i < N_LINE_ARANGE; i++) { |
| 1169 | SVal sv; |
| 1170 | UWord ix = read_twobit_array( lineZ->ix2s, i ); |
| 1171 | /* correct, but expensive: tl_assert(ix >= 0 && ix <= 3); */ |
| 1172 | sv = lineZ->dict[ix]; |
| 1173 | tl_assert(sv != SVal_INVALID); |
| 1174 | cl->svals[i] = sv; |
| 1175 | } |
| 1176 | stats__cache_Z_fetches++; |
| 1177 | } |
| 1178 | normalise_CacheLine( cl ); |
| 1179 | } |
| 1180 | |
| 1181 | static void shmem__invalidate_scache ( void ) { |
| 1182 | Word wix; |
| 1183 | if (0) VG_(printf)("%s","scache inval\n"); |
| 1184 | tl_assert(!is_valid_scache_tag(1)); |
| 1185 | for (wix = 0; wix < N_WAY_NENT; wix++) { |
| 1186 | cache_shmem.tags0[wix] = 1/*INVALID*/; |
| 1187 | } |
| 1188 | stats__cache_invals++; |
| 1189 | } |
| 1190 | |
| 1191 | static void shmem__flush_and_invalidate_scache ( void ) { |
| 1192 | Word wix; |
| 1193 | Addr tag; |
| 1194 | if (0) VG_(printf)("%s","scache flush and invalidate\n"); |
| 1195 | tl_assert(!is_valid_scache_tag(1)); |
| 1196 | for (wix = 0; wix < N_WAY_NENT; wix++) { |
| 1197 | tag = cache_shmem.tags0[wix]; |
| 1198 | if (tag == 1/*INVALID*/) { |
| 1199 | /* already invalid; nothing to do */ |
| 1200 | } else { |
| 1201 | tl_assert(is_valid_scache_tag(tag)); |
| 1202 | cacheline_wback( wix ); |
| 1203 | } |
| 1204 | cache_shmem.tags0[wix] = 1/*INVALID*/; |
| 1205 | } |
| 1206 | stats__cache_flushes++; |
| 1207 | stats__cache_invals++; |
| 1208 | } |
| 1209 | |
| 1210 | |
| 1211 | static inline Bool aligned16 ( Addr a ) { |
| 1212 | return 0 == (a & 1); |
| 1213 | } |
| 1214 | static inline Bool aligned32 ( Addr a ) { |
| 1215 | return 0 == (a & 3); |
| 1216 | } |
| 1217 | static inline Bool aligned64 ( Addr a ) { |
| 1218 | return 0 == (a & 7); |
| 1219 | } |
| 1220 | static inline UWord get_cacheline_offset ( Addr a ) { |
| 1221 | return (UWord)(a & (N_LINE_ARANGE - 1)); |
| 1222 | } |
| 1223 | static inline Addr cacheline_ROUNDUP ( Addr a ) { |
| 1224 | return ROUNDUP(a, N_LINE_ARANGE); |
| 1225 | } |
| 1226 | static inline Addr cacheline_ROUNDDN ( Addr a ) { |
| 1227 | return ROUNDDN(a, N_LINE_ARANGE); |
| 1228 | } |
| 1229 | static inline UWord get_treeno ( Addr a ) { |
| 1230 | return get_cacheline_offset(a) >> 3; |
| 1231 | } |
| 1232 | static inline UWord get_tree_offset ( Addr a ) { |
| 1233 | return a & 7; |
| 1234 | } |
| 1235 | |
| 1236 | static __attribute__((noinline)) |
| 1237 | CacheLine* get_cacheline_MISS ( Addr a ); /* fwds */ |
| 1238 | static inline CacheLine* get_cacheline ( Addr a ) |
| 1239 | { |
| 1240 | /* tag is 'a' with the in-line offset masked out, |
| 1241 | eg a[31]..a[4] 0000 */ |
| 1242 | Addr tag = a & ~(N_LINE_ARANGE - 1); |
| 1243 | UWord wix = (a >> N_LINE_BITS) & (N_WAY_NENT - 1); |
| 1244 | stats__cache_totrefs++; |
| 1245 | if (LIKELY(tag == cache_shmem.tags0[wix])) { |
| 1246 | return &cache_shmem.lyns0[wix]; |
| 1247 | } else { |
| 1248 | return get_cacheline_MISS( a ); |
| 1249 | } |
| 1250 | } |
| 1251 | |
| 1252 | static __attribute__((noinline)) |
| 1253 | CacheLine* get_cacheline_MISS ( Addr a ) |
| 1254 | { |
| 1255 | /* tag is 'a' with the in-line offset masked out, |
| 1256 | eg a[31]..a[4] 0000 */ |
| 1257 | |
| 1258 | CacheLine* cl; |
| 1259 | Addr* tag_old_p; |
| 1260 | Addr tag = a & ~(N_LINE_ARANGE - 1); |
| 1261 | UWord wix = (a >> N_LINE_BITS) & (N_WAY_NENT - 1); |
| 1262 | |
| 1263 | tl_assert(tag != cache_shmem.tags0[wix]); |
| 1264 | |
| 1265 | /* Dump the old line into the backing store. */ |
| 1266 | stats__cache_totmisses++; |
| 1267 | |
| 1268 | cl = &cache_shmem.lyns0[wix]; |
| 1269 | tag_old_p = &cache_shmem.tags0[wix]; |
| 1270 | |
| 1271 | if (is_valid_scache_tag( *tag_old_p )) { |
| 1272 | /* EXPENSIVE and REDUNDANT: callee does it */ |
| 1273 | if (SCE_CACHELINE) |
| 1274 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 1275 | cacheline_wback( wix ); |
| 1276 | } |
| 1277 | /* and reload the new one */ |
| 1278 | *tag_old_p = tag; |
| 1279 | cacheline_fetch( wix ); |
| 1280 | if (SCE_CACHELINE) |
| 1281 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 1282 | return cl; |
| 1283 | } |
| 1284 | |
| 1285 | static UShort pulldown_to_32 ( /*MOD*/SVal* tree, UWord toff, UShort descr ) { |
| 1286 | stats__cline_64to32pulldown++; |
| 1287 | switch (toff) { |
| 1288 | case 0: case 4: |
| 1289 | tl_assert(descr & TREE_DESCR_64); |
| 1290 | tree[4] = tree[0]; |
| 1291 | descr &= ~TREE_DESCR_64; |
| 1292 | descr |= (TREE_DESCR_32_1 | TREE_DESCR_32_0); |
| 1293 | break; |
| 1294 | default: |
| 1295 | tl_assert(0); |
| 1296 | } |
| 1297 | return descr; |
| 1298 | } |
| 1299 | |
| 1300 | static UShort pulldown_to_16 ( /*MOD*/SVal* tree, UWord toff, UShort descr ) { |
| 1301 | stats__cline_32to16pulldown++; |
| 1302 | switch (toff) { |
| 1303 | case 0: case 2: |
| 1304 | if (!(descr & TREE_DESCR_32_0)) { |
| 1305 | descr = pulldown_to_32(tree, 0, descr); |
| 1306 | } |
| 1307 | tl_assert(descr & TREE_DESCR_32_0); |
| 1308 | tree[2] = tree[0]; |
| 1309 | descr &= ~TREE_DESCR_32_0; |
| 1310 | descr |= (TREE_DESCR_16_1 | TREE_DESCR_16_0); |
| 1311 | break; |
| 1312 | case 4: case 6: |
| 1313 | if (!(descr & TREE_DESCR_32_1)) { |
| 1314 | descr = pulldown_to_32(tree, 4, descr); |
| 1315 | } |
| 1316 | tl_assert(descr & TREE_DESCR_32_1); |
| 1317 | tree[6] = tree[4]; |
| 1318 | descr &= ~TREE_DESCR_32_1; |
| 1319 | descr |= (TREE_DESCR_16_3 | TREE_DESCR_16_2); |
| 1320 | break; |
| 1321 | default: |
| 1322 | tl_assert(0); |
| 1323 | } |
| 1324 | return descr; |
| 1325 | } |
| 1326 | |
| 1327 | static UShort pulldown_to_8 ( /*MOD*/SVal* tree, UWord toff, UShort descr ) { |
| 1328 | stats__cline_16to8pulldown++; |
| 1329 | switch (toff) { |
| 1330 | case 0: case 1: |
| 1331 | if (!(descr & TREE_DESCR_16_0)) { |
| 1332 | descr = pulldown_to_16(tree, 0, descr); |
| 1333 | } |
| 1334 | tl_assert(descr & TREE_DESCR_16_0); |
| 1335 | tree[1] = tree[0]; |
| 1336 | descr &= ~TREE_DESCR_16_0; |
| 1337 | descr |= (TREE_DESCR_8_1 | TREE_DESCR_8_0); |
| 1338 | break; |
| 1339 | case 2: case 3: |
| 1340 | if (!(descr & TREE_DESCR_16_1)) { |
| 1341 | descr = pulldown_to_16(tree, 2, descr); |
| 1342 | } |
| 1343 | tl_assert(descr & TREE_DESCR_16_1); |
| 1344 | tree[3] = tree[2]; |
| 1345 | descr &= ~TREE_DESCR_16_1; |
| 1346 | descr |= (TREE_DESCR_8_3 | TREE_DESCR_8_2); |
| 1347 | break; |
| 1348 | case 4: case 5: |
| 1349 | if (!(descr & TREE_DESCR_16_2)) { |
| 1350 | descr = pulldown_to_16(tree, 4, descr); |
| 1351 | } |
| 1352 | tl_assert(descr & TREE_DESCR_16_2); |
| 1353 | tree[5] = tree[4]; |
| 1354 | descr &= ~TREE_DESCR_16_2; |
| 1355 | descr |= (TREE_DESCR_8_5 | TREE_DESCR_8_4); |
| 1356 | break; |
| 1357 | case 6: case 7: |
| 1358 | if (!(descr & TREE_DESCR_16_3)) { |
| 1359 | descr = pulldown_to_16(tree, 6, descr); |
| 1360 | } |
| 1361 | tl_assert(descr & TREE_DESCR_16_3); |
| 1362 | tree[7] = tree[6]; |
| 1363 | descr &= ~TREE_DESCR_16_3; |
| 1364 | descr |= (TREE_DESCR_8_7 | TREE_DESCR_8_6); |
| 1365 | break; |
| 1366 | default: |
| 1367 | tl_assert(0); |
| 1368 | } |
| 1369 | return descr; |
| 1370 | } |
| 1371 | |
| 1372 | |
| 1373 | static UShort pullup_descr_to_16 ( UShort descr, UWord toff ) { |
| 1374 | UShort mask; |
| 1375 | switch (toff) { |
| 1376 | case 0: |
| 1377 | mask = TREE_DESCR_8_1 | TREE_DESCR_8_0; |
| 1378 | tl_assert( (descr & mask) == mask ); |
| 1379 | descr &= ~mask; |
| 1380 | descr |= TREE_DESCR_16_0; |
| 1381 | break; |
| 1382 | case 2: |
| 1383 | mask = TREE_DESCR_8_3 | TREE_DESCR_8_2; |
| 1384 | tl_assert( (descr & mask) == mask ); |
| 1385 | descr &= ~mask; |
| 1386 | descr |= TREE_DESCR_16_1; |
| 1387 | break; |
| 1388 | case 4: |
| 1389 | mask = TREE_DESCR_8_5 | TREE_DESCR_8_4; |
| 1390 | tl_assert( (descr & mask) == mask ); |
| 1391 | descr &= ~mask; |
| 1392 | descr |= TREE_DESCR_16_2; |
| 1393 | break; |
| 1394 | case 6: |
| 1395 | mask = TREE_DESCR_8_7 | TREE_DESCR_8_6; |
| 1396 | tl_assert( (descr & mask) == mask ); |
| 1397 | descr &= ~mask; |
| 1398 | descr |= TREE_DESCR_16_3; |
| 1399 | break; |
| 1400 | default: |
| 1401 | tl_assert(0); |
| 1402 | } |
| 1403 | return descr; |
| 1404 | } |
| 1405 | |
| 1406 | static UShort pullup_descr_to_32 ( UShort descr, UWord toff ) { |
| 1407 | UShort mask; |
| 1408 | switch (toff) { |
| 1409 | case 0: |
| 1410 | if (!(descr & TREE_DESCR_16_0)) |
| 1411 | descr = pullup_descr_to_16(descr, 0); |
| 1412 | if (!(descr & TREE_DESCR_16_1)) |
| 1413 | descr = pullup_descr_to_16(descr, 2); |
| 1414 | mask = TREE_DESCR_16_1 | TREE_DESCR_16_0; |
| 1415 | tl_assert( (descr & mask) == mask ); |
| 1416 | descr &= ~mask; |
| 1417 | descr |= TREE_DESCR_32_0; |
| 1418 | break; |
| 1419 | case 4: |
| 1420 | if (!(descr & TREE_DESCR_16_2)) |
| 1421 | descr = pullup_descr_to_16(descr, 4); |
| 1422 | if (!(descr & TREE_DESCR_16_3)) |
| 1423 | descr = pullup_descr_to_16(descr, 6); |
| 1424 | mask = TREE_DESCR_16_3 | TREE_DESCR_16_2; |
| 1425 | tl_assert( (descr & mask) == mask ); |
| 1426 | descr &= ~mask; |
| 1427 | descr |= TREE_DESCR_32_1; |
| 1428 | break; |
| 1429 | default: |
| 1430 | tl_assert(0); |
| 1431 | } |
| 1432 | return descr; |
| 1433 | } |
| 1434 | |
| 1435 | static Bool valid_value_is_above_me_32 ( UShort descr, UWord toff ) { |
| 1436 | switch (toff) { |
| 1437 | case 0: case 4: |
| 1438 | return 0 != (descr & TREE_DESCR_64); |
| 1439 | default: |
| 1440 | tl_assert(0); |
| 1441 | } |
| 1442 | } |
| 1443 | |
| 1444 | static Bool valid_value_is_below_me_16 ( UShort descr, UWord toff ) { |
| 1445 | switch (toff) { |
| 1446 | case 0: |
| 1447 | return 0 != (descr & (TREE_DESCR_8_1 | TREE_DESCR_8_0)); |
| 1448 | case 2: |
| 1449 | return 0 != (descr & (TREE_DESCR_8_3 | TREE_DESCR_8_2)); |
| 1450 | case 4: |
| 1451 | return 0 != (descr & (TREE_DESCR_8_5 | TREE_DESCR_8_4)); |
| 1452 | case 6: |
| 1453 | return 0 != (descr & (TREE_DESCR_8_7 | TREE_DESCR_8_6)); |
| 1454 | default: |
| 1455 | tl_assert(0); |
| 1456 | } |
| 1457 | } |
| 1458 | |
| 1459 | /* ------------ Cache management ------------ */ |
| 1460 | |
| 1461 | static void zsm_flush_cache ( void ) |
| 1462 | { |
| 1463 | shmem__flush_and_invalidate_scache(); |
| 1464 | } |
| 1465 | |
| 1466 | |
| 1467 | static void zsm_init ( void(*p_rcinc)(SVal), void(*p_rcdec)(SVal) ) |
| 1468 | { |
| 1469 | tl_assert( sizeof(UWord) == sizeof(Addr) ); |
| 1470 | |
| 1471 | rcinc = p_rcinc; |
| 1472 | rcdec = p_rcdec; |
| 1473 | |
| 1474 | tl_assert(map_shmem == NULL); |
| 1475 | map_shmem = VG_(newFM)( HG_(zalloc), "libhb.zsm_init.1 (map_shmem)", |
| 1476 | HG_(free), |
| 1477 | NULL/*unboxed UWord cmp*/); |
| 1478 | tl_assert(map_shmem != NULL); |
| 1479 | shmem__invalidate_scache(); |
| 1480 | |
| 1481 | /* a SecMap must contain an integral number of CacheLines */ |
| 1482 | tl_assert(0 == (N_SECMAP_ARANGE % N_LINE_ARANGE)); |
| 1483 | /* also ... a CacheLine holds an integral number of trees */ |
| 1484 | tl_assert(0 == (N_LINE_ARANGE % 8)); |
| 1485 | } |
| 1486 | |
| 1487 | ///////////////////////////////////////////////////////////////// |
| 1488 | ///////////////////////////////////////////////////////////////// |
| 1489 | // // |
| 1490 | // SECTION END compressed shadow memory // |
| 1491 | // // |
| 1492 | ///////////////////////////////////////////////////////////////// |
| 1493 | ///////////////////////////////////////////////////////////////// |
| 1494 | |
| 1495 | |
| 1496 | |
| 1497 | ///////////////////////////////////////////////////////////////// |
| 1498 | ///////////////////////////////////////////////////////////////// |
| 1499 | // // |
| 1500 | // SECTION BEGIN vts primitives // |
| 1501 | // // |
| 1502 | ///////////////////////////////////////////////////////////////// |
| 1503 | ///////////////////////////////////////////////////////////////// |
| 1504 | |
| 1505 | #ifndef __HB_VTS_H |
| 1506 | #define __HB_VTS_H |
| 1507 | |
| 1508 | /* VtsIDs can't exceed 30 bits, since they have to be packed into the |
| 1509 | lowest 30 bits of an SVal. */ |
| 1510 | typedef UInt VtsID; |
| 1511 | #define VtsID_INVALID 0xFFFFFFFF |
| 1512 | |
| 1513 | /* A VTS contains .ts, its vector clock, and also .id, a field to hold |
| 1514 | a backlink for the caller's convenience. Since we have no idea |
| 1515 | what to set that to in the library, it always gets set to |
| 1516 | VtsID_INVALID. */ |
| 1517 | typedef |
| 1518 | struct { |
| 1519 | VtsID id; |
| 1520 | XArray* ts; /* XArray* ScalarTS(abstract) */ |
| 1521 | } |
| 1522 | VTS; |
| 1523 | |
| 1524 | |
| 1525 | /* Create a new, empty VTS. */ |
| 1526 | VTS* VTS__new ( void ); |
| 1527 | |
| 1528 | /* Delete this VTS in its entirety. */ |
| 1529 | void VTS__delete ( VTS* vts ); |
| 1530 | |
| 1531 | /* Create a new singleton VTS. */ |
| 1532 | VTS* VTS__singleton ( Thr* thr, ULong tym ); |
| 1533 | |
| 1534 | /* Return a new VTS in which vts[me]++, so to speak. 'vts' itself is |
| 1535 | not modified. */ |
| 1536 | VTS* VTS__tick ( Thr* me, VTS* vts ); |
| 1537 | |
| 1538 | /* Return a new VTS constructed as the join (max) of the 2 args. |
| 1539 | Neither arg is modified. */ |
| 1540 | VTS* VTS__join ( VTS* a, VTS* b ); |
| 1541 | |
| 1542 | /* Compute the partial ordering relation of the two args. */ |
| 1543 | typedef |
| 1544 | enum { POrd_EQ=4, POrd_LT, POrd_GT, POrd_UN } |
| 1545 | POrd; |
| 1546 | |
| 1547 | POrd VTS__cmp ( VTS* a, VTS* b ); |
| 1548 | |
| 1549 | /* Compute an arbitrary structural (total) ordering on the two args, |
| 1550 | based on their VCs, so they can be looked up in a table, tree, etc. |
| 1551 | Returns -1, 0 or 1. */ |
| 1552 | Word VTS__cmp_structural ( VTS* a, VTS* b ); |
| 1553 | |
| 1554 | /* Debugging only. Display the given VTS in the buffer. */ |
| 1555 | void VTS__show ( HChar* buf, Int nBuf, VTS* vts ); |
| 1556 | |
| 1557 | /* Debugging only. Return vts[index], so to speak. */ |
| 1558 | ULong VTS__indexAt_SLOW ( VTS* vts, Thr* index ); |
| 1559 | |
| 1560 | #endif /* ! __HB_VTS_H */ |
| 1561 | |
| 1562 | |
| 1563 | /*--------------- to do with Vector Timestamps ---------------*/ |
| 1564 | |
| 1565 | /* Scalar Timestamp */ |
| 1566 | typedef |
| 1567 | struct { |
| 1568 | Thr* thr; |
| 1569 | ULong tym; |
| 1570 | } |
| 1571 | ScalarTS; |
| 1572 | |
| 1573 | |
| 1574 | static Bool is_sane_VTS ( VTS* vts ) |
| 1575 | { |
| 1576 | UWord i, n; |
| 1577 | ScalarTS *st1, *st2; |
| 1578 | if (!vts) return False; |
| 1579 | if (!vts->ts) return False; |
| 1580 | n = VG_(sizeXA)( vts->ts ); |
| 1581 | if (n >= 2) { |
| 1582 | for (i = 0; i < n-1; i++) { |
| 1583 | st1 = VG_(indexXA)( vts->ts, i ); |
| 1584 | st2 = VG_(indexXA)( vts->ts, i+1 ); |
| 1585 | if (st1->thr >= st2->thr) |
| 1586 | return False; |
| 1587 | if (st1->tym == 0 || st2->tym == 0) |
| 1588 | return False; |
| 1589 | } |
| 1590 | } |
| 1591 | return True; |
| 1592 | } |
| 1593 | |
| 1594 | |
| 1595 | /* Create a new, empty VTS. |
| 1596 | */ |
| 1597 | VTS* VTS__new ( void ) |
| 1598 | { |
| 1599 | VTS* vts; |
| 1600 | vts = HG_(zalloc)( "libhb.VTS__new.1", sizeof(VTS) ); |
| 1601 | tl_assert(vts); |
| 1602 | vts->id = VtsID_INVALID; |
| 1603 | vts->ts = VG_(newXA)( HG_(zalloc), "libhb.VTS__new.2", |
| 1604 | HG_(free), sizeof(ScalarTS) ); |
| 1605 | tl_assert(vts->ts); |
| 1606 | return vts; |
| 1607 | } |
| 1608 | |
| 1609 | |
| 1610 | /* Delete this VTS in its entirety. |
| 1611 | */ |
| 1612 | void VTS__delete ( VTS* vts ) |
| 1613 | { |
| 1614 | tl_assert(vts); |
| 1615 | tl_assert(vts->ts); |
| 1616 | VG_(deleteXA)( vts->ts ); |
| 1617 | HG_(free)(vts); |
| 1618 | } |
| 1619 | |
| 1620 | |
| 1621 | /* Create a new singleton VTS. |
| 1622 | */ |
| 1623 | VTS* VTS__singleton ( Thr* thr, ULong tym ) { |
| 1624 | ScalarTS st; |
| 1625 | VTS* vts; |
| 1626 | tl_assert(thr); |
| 1627 | tl_assert(tym >= 1); |
| 1628 | vts = VTS__new(); |
| 1629 | st.thr = thr; |
| 1630 | st.tym = tym; |
| 1631 | VG_(addToXA)( vts->ts, &st ); |
| 1632 | return vts; |
| 1633 | } |
| 1634 | |
| 1635 | |
| 1636 | /* Return a new VTS in which vts[me]++, so to speak. 'vts' itself is |
| 1637 | not modified. |
| 1638 | */ |
| 1639 | VTS* VTS__tick ( Thr* me, VTS* vts ) |
| 1640 | { |
| 1641 | ScalarTS* here = NULL; |
| 1642 | ScalarTS tmp; |
| 1643 | VTS* res; |
| 1644 | Word i, n; |
| 1645 | tl_assert(me); |
| 1646 | tl_assert(is_sane_VTS(vts)); |
| 1647 | //if (0) VG_(printf)("tick vts thrno %ld szin %d\n", |
| 1648 | // (Word)me->errmsg_index, (Int)VG_(sizeXA)(vts) ); |
| 1649 | res = VTS__new(); |
| 1650 | n = VG_(sizeXA)( vts->ts ); |
| 1651 | |
| 1652 | /* main loop doesn't handle zero-entry case correctly, so |
| 1653 | special-case it. */ |
| 1654 | if (n == 0) { |
| 1655 | tmp.thr = me; |
| 1656 | tmp.tym = 1; |
| 1657 | VG_(addToXA)( res->ts, &tmp ); |
| 1658 | tl_assert(is_sane_VTS(res)); |
| 1659 | return res; |
| 1660 | } |
| 1661 | |
| 1662 | for (i = 0; i < n; i++) { |
| 1663 | here = VG_(indexXA)( vts->ts, i ); |
| 1664 | if (me < here->thr) { |
| 1665 | /* We just went past 'me', without seeing it. */ |
| 1666 | tmp.thr = me; |
| 1667 | tmp.tym = 1; |
| 1668 | VG_(addToXA)( res->ts, &tmp ); |
| 1669 | tmp = *here; |
| 1670 | VG_(addToXA)( res->ts, &tmp ); |
| 1671 | i++; |
| 1672 | break; |
| 1673 | } |
| 1674 | else if (me == here->thr) { |
| 1675 | tmp = *here; |
| 1676 | tmp.tym++; |
| 1677 | VG_(addToXA)( res->ts, &tmp ); |
| 1678 | i++; |
| 1679 | break; |
| 1680 | } |
| 1681 | else /* me > here->thr */ { |
| 1682 | tmp = *here; |
| 1683 | VG_(addToXA)( res->ts, &tmp ); |
| 1684 | } |
| 1685 | } |
| 1686 | tl_assert(i >= 0 && i <= n); |
| 1687 | if (i == n && here && here->thr < me) { |
| 1688 | tmp.thr = me; |
| 1689 | tmp.tym = 1; |
| 1690 | VG_(addToXA)( res->ts, &tmp ); |
| 1691 | } else { |
| 1692 | for (/*keepgoing*/; i < n; i++) { |
| 1693 | here = VG_(indexXA)( vts->ts, i ); |
| 1694 | tmp = *here; |
| 1695 | VG_(addToXA)( res->ts, &tmp ); |
| 1696 | } |
| 1697 | } |
| 1698 | tl_assert(is_sane_VTS(res)); |
| 1699 | //if (0) VG_(printf)("tick vts thrno %ld szou %d\n", |
| 1700 | // (Word)me->errmsg_index, (Int)VG_(sizeXA)(res) ); |
| 1701 | return res; |
| 1702 | } |
| 1703 | |
| 1704 | |
| 1705 | /* Return a new VTS constructed as the join (max) of the 2 args. |
| 1706 | Neither arg is modified. |
| 1707 | */ |
| 1708 | VTS* VTS__join ( VTS* a, VTS* b ) |
| 1709 | { |
| 1710 | Word ia, ib, useda, usedb; |
| 1711 | ULong tyma, tymb, tymMax; |
| 1712 | Thr* thr; |
| 1713 | VTS* res; |
| 1714 | ScalarTS *tmpa, *tmpb; |
| 1715 | |
| 1716 | tl_assert(a && a->ts); |
| 1717 | tl_assert(b && b->ts); |
| 1718 | useda = VG_(sizeXA)( a->ts ); |
| 1719 | usedb = VG_(sizeXA)( b->ts ); |
| 1720 | |
| 1721 | res = VTS__new(); |
| 1722 | ia = ib = 0; |
| 1723 | |
| 1724 | while (1) { |
| 1725 | |
| 1726 | /* This logic is to enumerate triples (thr, tyma, tymb) drawn |
| 1727 | from a and b in order, where thr is the next Thr* |
| 1728 | occurring in either a or b, and tyma/b are the relevant |
| 1729 | scalar timestamps, taking into account implicit zeroes. */ |
| 1730 | tl_assert(ia >= 0 && ia <= useda); |
| 1731 | tl_assert(ib >= 0 && ib <= usedb); |
| 1732 | tmpa = tmpb = NULL; |
| 1733 | |
| 1734 | if (ia == useda && ib == usedb) { |
| 1735 | /* both empty - done */ |
| 1736 | break; |
| 1737 | } |
| 1738 | else |
| 1739 | if (ia == useda && ib != usedb) { |
| 1740 | /* a empty, use up b */ |
| 1741 | tmpb = VG_(indexXA)( b->ts, ib ); |
| 1742 | thr = tmpb->thr; |
| 1743 | tyma = 0; |
| 1744 | tymb = tmpb->tym; |
| 1745 | ib++; |
| 1746 | } |
| 1747 | else |
| 1748 | if (ia != useda && ib == usedb) { |
| 1749 | /* b empty, use up a */ |
| 1750 | tmpa = VG_(indexXA)( a->ts, ia ); |
| 1751 | thr = tmpa->thr; |
| 1752 | tyma = tmpa->tym; |
| 1753 | tymb = 0; |
| 1754 | ia++; |
| 1755 | } |
| 1756 | else { |
| 1757 | /* both not empty; extract lowest-Thr*'d triple */ |
| 1758 | tmpa = VG_(indexXA)( a->ts, ia ); |
| 1759 | tmpb = VG_(indexXA)( b->ts, ib ); |
| 1760 | if (tmpa->thr < tmpb->thr) { |
| 1761 | /* a has the lowest unconsidered Thr* */ |
| 1762 | thr = tmpa->thr; |
| 1763 | tyma = tmpa->tym; |
| 1764 | tymb = 0; |
| 1765 | ia++; |
| 1766 | } |
| 1767 | else |
| 1768 | if (tmpa->thr > tmpb->thr) { |
| 1769 | /* b has the lowest unconsidered Thr* */ |
| 1770 | thr = tmpb->thr; |
| 1771 | tyma = 0; |
| 1772 | tymb = tmpb->tym; |
| 1773 | ib++; |
| 1774 | } else { |
| 1775 | /* they both next mention the same Thr* */ |
| 1776 | tl_assert(tmpa->thr == tmpb->thr); |
| 1777 | thr = tmpa->thr; /* == tmpb->thr */ |
| 1778 | tyma = tmpa->tym; |
| 1779 | tymb = tmpb->tym; |
| 1780 | ia++; |
| 1781 | ib++; |
| 1782 | } |
| 1783 | } |
| 1784 | |
| 1785 | /* having laboriously determined (thr, tyma, tymb), do something |
| 1786 | useful with it. */ |
| 1787 | tymMax = tyma > tymb ? tyma : tymb; |
| 1788 | if (tymMax > 0) { |
| 1789 | ScalarTS st; |
| 1790 | st.thr = thr; |
| 1791 | st.tym = tymMax; |
| 1792 | VG_(addToXA)( res->ts, &st ); |
| 1793 | } |
| 1794 | |
| 1795 | } |
| 1796 | |
| 1797 | tl_assert(is_sane_VTS( res )); |
| 1798 | |
| 1799 | return res; |
| 1800 | } |
| 1801 | |
| 1802 | |
| 1803 | /* Compute the partial ordering relation of the two args. |
| 1804 | */ |
| 1805 | POrd VTS__cmp ( VTS* a, VTS* b ) |
| 1806 | { |
| 1807 | Word ia, ib, useda, usedb; |
| 1808 | ULong tyma, tymb; |
| 1809 | Thr* thr; |
| 1810 | ScalarTS *tmpa, *tmpb; |
| 1811 | |
| 1812 | Bool all_leq = True; |
| 1813 | Bool all_geq = True; |
| 1814 | |
| 1815 | tl_assert(a && a->ts); |
| 1816 | tl_assert(b && b->ts); |
| 1817 | useda = VG_(sizeXA)( a->ts ); |
| 1818 | usedb = VG_(sizeXA)( b->ts ); |
| 1819 | |
| 1820 | ia = ib = 0; |
| 1821 | |
| 1822 | while (1) { |
| 1823 | |
| 1824 | /* This logic is to enumerate triples (thr, tyma, tymb) drawn |
| 1825 | from a and b in order, where thr is the next Thr* |
| 1826 | occurring in either a or b, and tyma/b are the relevant |
| 1827 | scalar timestamps, taking into account implicit zeroes. */ |
| 1828 | tl_assert(ia >= 0 && ia <= useda); |
| 1829 | tl_assert(ib >= 0 && ib <= usedb); |
| 1830 | tmpa = tmpb = NULL; |
| 1831 | |
| 1832 | if (ia == useda && ib == usedb) { |
| 1833 | /* both empty - done */ |
| 1834 | break; |
| 1835 | } |
| 1836 | else |
| 1837 | if (ia == useda && ib != usedb) { |
| 1838 | /* a empty, use up b */ |
| 1839 | tmpb = VG_(indexXA)( b->ts, ib ); |
| 1840 | thr = tmpb->thr; |
| 1841 | tyma = 0; |
| 1842 | tymb = tmpb->tym; |
| 1843 | ib++; |
| 1844 | } |
| 1845 | else |
| 1846 | if (ia != useda && ib == usedb) { |
| 1847 | /* b empty, use up a */ |
| 1848 | tmpa = VG_(indexXA)( a->ts, ia ); |
| 1849 | thr = tmpa->thr; |
| 1850 | tyma = tmpa->tym; |
| 1851 | tymb = 0; |
| 1852 | ia++; |
| 1853 | } |
| 1854 | else { |
| 1855 | /* both not empty; extract lowest-Thr*'d triple */ |
| 1856 | tmpa = VG_(indexXA)( a->ts, ia ); |
| 1857 | tmpb = VG_(indexXA)( b->ts, ib ); |
| 1858 | if (tmpa->thr < tmpb->thr) { |
| 1859 | /* a has the lowest unconsidered Thr* */ |
| 1860 | thr = tmpa->thr; |
| 1861 | tyma = tmpa->tym; |
| 1862 | tymb = 0; |
| 1863 | ia++; |
| 1864 | } |
| 1865 | else |
| 1866 | if (tmpa->thr > tmpb->thr) { |
| 1867 | /* b has the lowest unconsidered Thr* */ |
| 1868 | thr = tmpb->thr; |
| 1869 | tyma = 0; |
| 1870 | tymb = tmpb->tym; |
| 1871 | ib++; |
| 1872 | } else { |
| 1873 | /* they both next mention the same Thr* */ |
| 1874 | tl_assert(tmpa->thr == tmpb->thr); |
| 1875 | thr = tmpa->thr; /* == tmpb->thr */ |
| 1876 | tyma = tmpa->tym; |
| 1877 | tymb = tmpb->tym; |
| 1878 | ia++; |
| 1879 | ib++; |
| 1880 | } |
| 1881 | } |
| 1882 | |
| 1883 | /* having laboriously determined (thr, tyma, tymb), do something |
| 1884 | useful with it. */ |
| 1885 | if (tyma < tymb) |
| 1886 | all_geq = False; |
| 1887 | if (tyma > tymb) |
| 1888 | all_leq = False; |
| 1889 | } |
| 1890 | |
| 1891 | if (all_leq && all_geq) |
| 1892 | return POrd_EQ; |
| 1893 | /* now we know they aren't equal, so either all_leq or all_geq or |
| 1894 | both are false. */ |
| 1895 | if (all_leq) |
| 1896 | return POrd_LT; |
| 1897 | if (all_geq) |
| 1898 | return POrd_GT; |
| 1899 | /* hmm, neither all_geq or all_leq. This means unordered. */ |
| 1900 | return POrd_UN; |
| 1901 | } |
| 1902 | |
| 1903 | |
| 1904 | /* Compute an arbitrary structural (total) ordering on the two args, |
| 1905 | based on their VCs, so they can be looked up in a table, tree, etc. |
| 1906 | Returns -1, 0 or 1. (really just 'deriving Ord' :-) |
| 1907 | */ |
| 1908 | Word VTS__cmp_structural ( VTS* a, VTS* b ) |
| 1909 | { |
| 1910 | /* We just need to generate an arbitrary total ordering based on |
| 1911 | a->ts and b->ts. Preferably do it in a way which comes across likely |
| 1912 | differences relatively quickly. */ |
| 1913 | Word i, useda, usedb; |
| 1914 | ScalarTS *tmpa, *tmpb; |
| 1915 | |
| 1916 | tl_assert(a && a->ts); |
| 1917 | tl_assert(b && b->ts); |
| 1918 | useda = VG_(sizeXA)( a->ts ); |
| 1919 | usedb = VG_(sizeXA)( b->ts ); |
| 1920 | |
| 1921 | if (useda < usedb) return -1; |
| 1922 | if (useda > usedb) return 1; |
| 1923 | |
| 1924 | /* Same length vectors, so let's step through them together. */ |
| 1925 | tl_assert(useda == usedb); |
| 1926 | for (i = 0; i < useda; i++) { |
| 1927 | tmpa = VG_(indexXA)( a->ts, i ); |
| 1928 | tmpb = VG_(indexXA)( b->ts, i ); |
| 1929 | if (tmpa->tym < tmpb->tym) return -1; |
| 1930 | if (tmpa->tym > tmpb->tym) return 1; |
| 1931 | if (tmpa->thr < tmpb->thr) return -1; |
| 1932 | if (tmpa->thr > tmpb->thr) return 1; |
| 1933 | } |
| 1934 | |
| 1935 | /* They're identical. */ |
| 1936 | return 0; |
| 1937 | } |
| 1938 | |
| 1939 | |
| 1940 | /* Debugging only. Display the given VTS in the buffer. |
| 1941 | */ |
| 1942 | void VTS__show ( HChar* buf, Int nBuf, VTS* vts ) { |
| 1943 | ScalarTS* st; |
| 1944 | HChar unit[64]; |
| 1945 | Word i, n; |
| 1946 | Int avail = nBuf; |
| 1947 | tl_assert(vts && vts->ts); |
| 1948 | tl_assert(nBuf > 16); |
| 1949 | buf[0] = '['; |
| 1950 | buf[1] = 0; |
| 1951 | n = VG_(sizeXA)( vts->ts ); |
| 1952 | for (i = 0; i < n; i++) { |
| 1953 | tl_assert(avail >= 40); |
| 1954 | st = VG_(indexXA)( vts->ts, i ); |
| 1955 | VG_(memset)(unit, 0, sizeof(unit)); |
| 1956 | VG_(sprintf)(unit, i < n-1 ? "%p:%lld " : "%p:%lld", |
| 1957 | st->thr, st->tym); |
| 1958 | if (avail < VG_(strlen)(unit) + 40/*let's say*/) { |
| 1959 | VG_(strcat)(buf, " ...]"); |
| 1960 | buf[nBuf-1] = 0; |
| 1961 | return; |
| 1962 | } |
| 1963 | VG_(strcat)(buf, unit); |
| 1964 | avail -= VG_(strlen)(unit); |
| 1965 | } |
| 1966 | VG_(strcat)(buf, "]"); |
| 1967 | buf[nBuf-1] = 0; |
| 1968 | } |
| 1969 | |
| 1970 | |
| 1971 | /* Debugging only. Return vts[index], so to speak. |
| 1972 | */ |
| 1973 | ULong VTS__indexAt_SLOW ( VTS* vts, Thr* idx ) { |
| 1974 | UWord i, n; |
| 1975 | tl_assert(vts && vts->ts); |
| 1976 | n = VG_(sizeXA)( vts->ts ); |
| 1977 | for (i = 0; i < n; i++) { |
| 1978 | ScalarTS* st = VG_(indexXA)( vts->ts, i ); |
| 1979 | if (st->thr == idx) |
| 1980 | return st->tym; |
| 1981 | } |
| 1982 | return 0; |
| 1983 | } |
| 1984 | |
| 1985 | |
| 1986 | ///////////////////////////////////////////////////////////////// |
| 1987 | ///////////////////////////////////////////////////////////////// |
| 1988 | // // |
| 1989 | // SECTION END vts primitives // |
| 1990 | // // |
| 1991 | ///////////////////////////////////////////////////////////////// |
| 1992 | ///////////////////////////////////////////////////////////////// |
| 1993 | |
| 1994 | |
| 1995 | |
| 1996 | ///////////////////////////////////////////////////////////////// |
| 1997 | ///////////////////////////////////////////////////////////////// |
| 1998 | // // |
| 1999 | // SECTION BEGIN main library // |
| 2000 | // // |
| 2001 | ///////////////////////////////////////////////////////////////// |
| 2002 | ///////////////////////////////////////////////////////////////// |
| 2003 | |
| 2004 | |
| 2005 | ///////////////////////////////////////////////////////// |
| 2006 | // // |
| 2007 | // VTS set // |
| 2008 | // // |
| 2009 | ///////////////////////////////////////////////////////// |
| 2010 | |
| 2011 | static WordFM* /* VTS* void void */ vts_set = NULL; |
| 2012 | |
| 2013 | static void vts_set_init ( void ) |
| 2014 | { |
| 2015 | tl_assert(!vts_set); |
| 2016 | vts_set = VG_(newFM)( HG_(zalloc), "libhb.vts_set_init.1", |
| 2017 | HG_(free), |
| 2018 | (Word(*)(UWord,UWord))VTS__cmp_structural ); |
| 2019 | tl_assert(vts_set); |
| 2020 | } |
| 2021 | |
| 2022 | /* Given a newly made VTS, look in vts_set to see if we already have |
| 2023 | an identical one. If yes, free up this one and return instead a |
| 2024 | pointer to the existing one. If no, add this one to the set and |
| 2025 | return the same pointer. Caller differentiates the two cases by |
| 2026 | comparing returned pointer with the supplied one (although that |
| 2027 | does require that the supplied VTS is not already in the set). |
| 2028 | */ |
| 2029 | static VTS* vts_set__find_and_dealloc__or_add ( VTS* cand ) |
| 2030 | { |
| 2031 | UWord keyW, valW; |
| 2032 | /* lookup cand (by value) */ |
| 2033 | if (VG_(lookupFM)( vts_set, &keyW, &valW, (UWord)cand )) { |
| 2034 | /* found it */ |
| 2035 | tl_assert(valW == 0); |
| 2036 | /* if this fails, cand (by ref) was already present (!) */ |
| 2037 | tl_assert(keyW != (UWord)cand); |
| 2038 | VTS__delete(cand); |
| 2039 | return (VTS*)keyW; |
| 2040 | } else { |
| 2041 | /* not present. Add and return pointer to same. */ |
| 2042 | VG_(addToFM)( vts_set, (UWord)cand, 0/*val is unused*/ ); |
| 2043 | return cand; |
| 2044 | } |
| 2045 | } |
| 2046 | |
| 2047 | |
| 2048 | ///////////////////////////////////////////////////////// |
| 2049 | // // |
| 2050 | // VTS table // |
| 2051 | // // |
| 2052 | ///////////////////////////////////////////////////////// |
| 2053 | |
| 2054 | static void VtsID__invalidate_caches ( void ); /* fwds */ |
| 2055 | |
| 2056 | /* A type to hold VTS table entries. Invariants: |
| 2057 | If .vts == NULL, then this entry is not in use, so: |
| 2058 | - .rc == 0 |
| 2059 | - this entry is on the freelist (unfortunately, does not imply |
| 2060 | any constraints on value for .nextfree) |
| 2061 | If .vts != NULL, then this entry is in use: |
| 2062 | - .vts is findable in vts_set |
| 2063 | - .vts->id == this entry number |
| 2064 | - no specific value for .rc (even 0 is OK) |
| 2065 | - this entry is not on freelist, so .nextfree == VtsID_INVALID |
| 2066 | */ |
| 2067 | typedef |
| 2068 | struct { |
| 2069 | VTS* vts; /* vts, in vts_set */ |
| 2070 | UWord rc; /* reference count - enough for entire aspace */ |
| 2071 | VtsID freelink; /* chain for free entries, VtsID_INVALID at end */ |
| 2072 | } |
| 2073 | VtsTE; |
| 2074 | |
| 2075 | /* The VTS table. */ |
| 2076 | static XArray* /* of VtsTE */ vts_tab = NULL; |
| 2077 | |
| 2078 | /* An index into the VTS table, indicating the start of the list of |
| 2079 | free (available for use) entries. If the list is empty, this is |
| 2080 | VtsID_INVALID. */ |
| 2081 | static VtsID vts_tab_freelist = VtsID_INVALID; |
| 2082 | |
| 2083 | /* Do a GC of vts_tab when the freelist becomes empty AND the size of |
| 2084 | vts_tab equals or exceeds this size. After GC, the value here is |
| 2085 | set appropriately so as to check for the next GC point. */ |
| 2086 | static Word vts_next_GC_at = 1000; |
| 2087 | |
| 2088 | static void vts_tab_init ( void ) |
| 2089 | { |
| 2090 | vts_tab |
| 2091 | = VG_(newXA)( HG_(zalloc), "libhb.vts_tab_init.1", |
| 2092 | HG_(free), sizeof(VtsTE) ); |
| 2093 | vts_tab_freelist |
| 2094 | = VtsID_INVALID; |
| 2095 | tl_assert(vts_tab); |
| 2096 | } |
| 2097 | |
| 2098 | /* Add ii to the free list, checking that it looks out-of-use. */ |
| 2099 | static void add_to_free_list ( VtsID ii ) |
| 2100 | { |
| 2101 | VtsTE* ie = VG_(indexXA)( vts_tab, ii ); |
| 2102 | tl_assert(ie->vts == NULL); |
| 2103 | tl_assert(ie->rc == 0); |
| 2104 | tl_assert(ie->freelink == VtsID_INVALID); |
| 2105 | ie->freelink = vts_tab_freelist; |
| 2106 | vts_tab_freelist = ii; |
| 2107 | } |
| 2108 | |
| 2109 | /* Get an entry from the free list. This will return VtsID_INVALID if |
| 2110 | the free list is empty. */ |
| 2111 | static VtsID get_from_free_list ( void ) |
| 2112 | { |
| 2113 | VtsID ii; |
| 2114 | VtsTE* ie; |
| 2115 | if (vts_tab_freelist == VtsID_INVALID) |
| 2116 | return VtsID_INVALID; |
| 2117 | ii = vts_tab_freelist; |
| 2118 | ie = VG_(indexXA)( vts_tab, ii ); |
| 2119 | tl_assert(ie->vts == NULL); |
| 2120 | tl_assert(ie->rc == 0); |
| 2121 | vts_tab_freelist = ie->freelink; |
| 2122 | return ii; |
| 2123 | } |
| 2124 | |
| 2125 | /* Produce a new VtsID that can be used, either by getting it from |
| 2126 | the freelist, or, if that is empty, by expanding vts_tab. */ |
| 2127 | static VtsID get_new_VtsID ( void ) |
| 2128 | { |
| 2129 | VtsID ii; |
| 2130 | VtsTE te; |
| 2131 | ii = get_from_free_list(); |
| 2132 | if (ii != VtsID_INVALID) |
| 2133 | return ii; |
| 2134 | te.vts = NULL; |
| 2135 | te.rc = 0; |
| 2136 | te.freelink = VtsID_INVALID; |
| 2137 | ii = (VtsID)VG_(addToXA)( vts_tab, &te ); |
| 2138 | return ii; |
| 2139 | } |
| 2140 | |
| 2141 | |
| 2142 | /* Indirect callback from lib_zsm. */ |
| 2143 | static void VtsID__rcinc ( VtsID ii ) |
| 2144 | { |
| 2145 | VtsTE* ie; |
| 2146 | /* VG_(indexXA) does a range check for us */ |
| 2147 | ie = VG_(indexXA)( vts_tab, ii ); |
| 2148 | tl_assert(ie->vts); /* else it's not in use */ |
| 2149 | tl_assert(ie->rc < ~0UL); /* else we can't continue */ |
| 2150 | tl_assert(ie->vts->id == ii); |
| 2151 | ie->rc++; |
| 2152 | } |
| 2153 | |
| 2154 | /* Indirect callback from lib_zsm. */ |
| 2155 | static void VtsID__rcdec ( VtsID ii ) |
| 2156 | { |
| 2157 | VtsTE* ie; |
| 2158 | /* VG_(indexXA) does a range check for us */ |
| 2159 | ie = VG_(indexXA)( vts_tab, ii ); |
| 2160 | tl_assert(ie->vts); /* else it's not in use */ |
| 2161 | tl_assert(ie->rc > 0); /* else RC snafu */ |
| 2162 | tl_assert(ie->vts->id == ii); |
| 2163 | ie->rc--; |
| 2164 | } |
| 2165 | |
| 2166 | |
| 2167 | /* Look up 'cand' in our collection of VTSs. If present, deallocate |
| 2168 | it and return the VtsID for the pre-existing version. If not |
| 2169 | present, add it to both vts_tab and vts_set, allocate a fresh VtsID |
| 2170 | for it, and return that. */ |
| 2171 | static VtsID vts_tab__find_and_dealloc__or_add ( VTS* cand ) |
| 2172 | { |
| 2173 | VTS* auld; |
| 2174 | tl_assert(cand->id == VtsID_INVALID); |
| 2175 | auld = vts_set__find_and_dealloc__or_add(cand); |
| 2176 | if (auld != cand) { |
| 2177 | /* We already have an Aulde one. Use that. */ |
| 2178 | VtsTE* ie; |
| 2179 | tl_assert(auld->id != VtsID_INVALID); |
| 2180 | ie = VG_(indexXA)( vts_tab, auld->id ); |
| 2181 | tl_assert(ie->vts == auld); |
| 2182 | return auld->id; |
| 2183 | } else { |
| 2184 | VtsID ii = get_new_VtsID(); |
| 2185 | VtsTE* ie = VG_(indexXA)( vts_tab, ii ); |
| 2186 | ie->vts = cand; |
| 2187 | ie->rc = 0; |
| 2188 | ie->freelink = VtsID_INVALID; |
| 2189 | cand->id = ii; |
| 2190 | return ii; |
| 2191 | } |
| 2192 | } |
| 2193 | |
| 2194 | |
| 2195 | static void show_vts_stats ( HChar* caller ) |
| 2196 | { |
| 2197 | UWord nSet, nTab, nLive; |
| 2198 | ULong totrc; |
| 2199 | UWord n, i; |
| 2200 | nSet = VG_(sizeFM)( vts_set ); |
| 2201 | nTab = VG_(sizeXA)( vts_tab ); |
| 2202 | totrc = 0; |
| 2203 | nLive = 0; |
| 2204 | n = VG_(sizeXA)( vts_tab ); |
| 2205 | for (i = 0; i < n; i++) { |
| 2206 | VtsTE* ie = VG_(indexXA)( vts_tab, i ); |
| 2207 | if (ie->vts) { |
| 2208 | nLive++; |
| 2209 | totrc += (ULong)ie->rc; |
| 2210 | } else { |
| 2211 | tl_assert(ie->rc == 0); |
| 2212 | } |
| 2213 | } |
| 2214 | VG_(printf)(" show_vts_stats %s\n", caller); |
| 2215 | VG_(printf)(" vts_tab size %4lu\n", nTab); |
| 2216 | VG_(printf)(" vts_tab live %4lu\n", nLive); |
| 2217 | VG_(printf)(" vts_set size %4lu\n", nSet); |
| 2218 | VG_(printf)(" total rc %4llu\n", totrc); |
| 2219 | } |
| 2220 | |
| 2221 | /* NOT TO BE CALLED FROM WITHIN libzsm. */ |
| 2222 | static void vts_tab__do_GC ( Bool show_stats ) |
| 2223 | { |
| 2224 | UWord i, nTab, nLive, nFreed; |
| 2225 | |
| 2226 | /* check this is actually necessary. */ |
| 2227 | tl_assert(vts_tab_freelist == VtsID_INVALID); |
| 2228 | |
| 2229 | /* empty the caches for partial order checks and binary joins. We |
| 2230 | could do better and prune out the entries to be deleted, but it |
| 2231 | ain't worth the hassle. */ |
| 2232 | VtsID__invalidate_caches(); |
| 2233 | |
| 2234 | /* First, make the reference counts up to date. */ |
| 2235 | zsm_flush_cache(); |
| 2236 | |
| 2237 | nTab = VG_(sizeXA)( vts_tab ); |
| 2238 | |
| 2239 | if (show_stats) { |
| 2240 | VG_(printf)("<<GC begins at vts_tab size %lu>>\n", nTab); |
| 2241 | show_vts_stats("before GC"); |
| 2242 | } |
| 2243 | |
| 2244 | /* Now we can inspect the entire vts_tab. Any entries |
| 2245 | with zero .rc fields are now no longer in use and can be |
| 2246 | free list, removed from vts_set, and deleted. */ |
| 2247 | nFreed = 0; |
| 2248 | for (i = 0; i < nTab; i++) { |
| 2249 | Bool present; |
| 2250 | UWord oldK = 0, oldV = 0; |
| 2251 | VtsTE* te = VG_(indexXA)( vts_tab, i ); |
| 2252 | if (te->vts == NULL) { |
| 2253 | tl_assert(te->rc == 0); |
| 2254 | continue; /* already on the free list (presumably) */ |
| 2255 | } |
| 2256 | if (te->rc > 0) |
| 2257 | continue; /* in use */ |
| 2258 | /* Ok, we got one we can free. */ |
| 2259 | tl_assert(te->vts->id == i); |
| 2260 | /* first, remove it from vts_set. */ |
| 2261 | present = VG_(delFromFM)( vts_set, |
| 2262 | &oldK, &oldV, (UWord)te->vts ); |
| 2263 | tl_assert(present); /* else it isn't in vts_set ?! */ |
| 2264 | tl_assert(oldV == 0); /* no info stored in vts_set val fields */ |
| 2265 | tl_assert(oldK == (UWord)te->vts); /* else what did delFromFM find?! */ |
| 2266 | /* now free the VTS itself */ |
| 2267 | VTS__delete(te->vts); |
| 2268 | te->vts = NULL; |
| 2269 | /* and finally put this entry on the free list */ |
| 2270 | tl_assert(te->freelink == VtsID_INVALID); /* can't already be on it */ |
| 2271 | add_to_free_list( i ); |
| 2272 | nFreed++; |
| 2273 | } |
| 2274 | |
| 2275 | /* Now figure out when the next GC should be. We'll allow the |
| 2276 | number of VTSs to double before GCing again. Except of course |
| 2277 | that since we can't (or, at least, don't) shrink vts_tab, we |
| 2278 | can't set the threshhold value smaller than it. */ |
| 2279 | tl_assert(nFreed <= nTab); |
| 2280 | nLive = nTab - nFreed; |
| 2281 | tl_assert(nLive >= 0 && nLive <= nTab); |
| 2282 | vts_next_GC_at = 2 * nLive; |
| 2283 | if (vts_next_GC_at < nTab) |
| 2284 | vts_next_GC_at = nTab; |
| 2285 | |
| 2286 | if (show_stats) { |
| 2287 | show_vts_stats("after GC"); |
| 2288 | VG_(printf)("<<GC ends, next gc at %ld>>\n", vts_next_GC_at); |
| 2289 | } |
| 2290 | |
| 2291 | if (1) { |
| 2292 | static UInt ctr = 0; |
| 2293 | tl_assert(nTab > 0); |
| 2294 | VG_(printf)("libhb: VTS GC: #%u old size %lu live %lu (%2llu%%)\n", |
| 2295 | ctr++, nTab, nLive, (100ULL * nLive) / nTab); |
| 2296 | } |
| 2297 | } |
| 2298 | |
| 2299 | |
| 2300 | ///////////////////////////////////////////////////////// |
| 2301 | // // |
| 2302 | // Vts IDs // |
| 2303 | // // |
| 2304 | ///////////////////////////////////////////////////////// |
| 2305 | |
| 2306 | ////////////////////////// |
| 2307 | static ULong stats__getOrdering_queries = 0; |
| 2308 | static ULong stats__getOrdering_misses = 0; |
| 2309 | static ULong stats__join2_queries = 0; |
| 2310 | static ULong stats__join2_misses = 0; |
| 2311 | |
| 2312 | static inline UInt ROL32 ( UInt w, Int n ) { |
| 2313 | w = (w << n) | (w >> (32-n)); |
| 2314 | return w; |
| 2315 | } |
| 2316 | static inline UInt hash_VtsIDs ( VtsID vi1, VtsID vi2, UInt nTab ) { |
| 2317 | UInt hash = ROL32(vi1,19) ^ ROL32(vi2,13); |
| 2318 | return hash % nTab; |
| 2319 | } |
| 2320 | |
| 2321 | #define N_GETORDERING_CACHE 1023 |
| 2322 | static |
| 2323 | struct { VtsID vi1; VtsID vi2; POrd ord; } |
| 2324 | getOrdering_cache[N_GETORDERING_CACHE]; |
| 2325 | |
| 2326 | #define N_JOIN2_CACHE 1023 |
| 2327 | static |
| 2328 | struct { VtsID vi1; VtsID vi2; VtsID res; } |
| 2329 | join2_cache[N_JOIN2_CACHE]; |
| 2330 | |
| 2331 | static void VtsID__invalidate_caches ( void ) { |
| 2332 | Int i; |
| 2333 | for (i = 0; i < N_GETORDERING_CACHE; i++) { |
| 2334 | getOrdering_cache[i].vi1 = VtsID_INVALID; |
| 2335 | getOrdering_cache[i].vi2 = VtsID_INVALID; |
| 2336 | getOrdering_cache[i].ord = 0; /* an invalid POrd value */ |
| 2337 | } |
| 2338 | for (i = 0; i < N_JOIN2_CACHE; i++) { |
| 2339 | join2_cache[i].vi1 = VtsID_INVALID; |
| 2340 | join2_cache[i].vi2 = VtsID_INVALID; |
| 2341 | join2_cache[i].res = VtsID_INVALID; |
| 2342 | } |
| 2343 | } |
| 2344 | ////////////////////////// |
| 2345 | |
| 2346 | static Bool VtsID__is_valid ( VtsID vi ) { |
| 2347 | VtsTE* ve; |
| 2348 | if (vi >= (VtsID)VG_(sizeXA)( vts_tab )) |
| 2349 | return False; |
| 2350 | ve = VG_(indexXA)( vts_tab, vi ); |
| 2351 | if (!ve->vts) |
| 2352 | return False; |
| 2353 | tl_assert(ve->vts->id == vi); |
| 2354 | return True; |
| 2355 | } |
| 2356 | |
| 2357 | static VTS* VtsID__to_VTS ( VtsID vi ) { |
| 2358 | VtsTE* te = VG_(indexXA)( vts_tab, vi ); |
| 2359 | tl_assert(te->vts); |
| 2360 | return te->vts; |
| 2361 | } |
| 2362 | |
| 2363 | static void VtsID__pp ( VtsID vi ) { |
| 2364 | HChar buf[100]; |
| 2365 | VTS* vts = VtsID__to_VTS(vi); |
| 2366 | VTS__show( buf, sizeof(buf)-1, vts ); |
| 2367 | buf[sizeof(buf)-1] = 0; |
| 2368 | VG_(printf)("%s", buf); |
| 2369 | } |
| 2370 | |
| 2371 | /* compute partial ordering relation of vi1 and vi2. */ |
| 2372 | __attribute__((noinline)) |
| 2373 | static POrd VtsID__getOrdering_WRK ( VtsID vi1, VtsID vi2 ) { |
| 2374 | UInt hash; |
| 2375 | POrd ord; |
| 2376 | VTS *v1, *v2; |
| 2377 | //if (vi1 == vi2) return POrd_EQ; |
| 2378 | tl_assert(vi1 != vi2); |
| 2379 | ////++ |
| 2380 | stats__getOrdering_queries++; |
| 2381 | hash = hash_VtsIDs(vi1, vi2, N_GETORDERING_CACHE); |
| 2382 | if (getOrdering_cache[hash].vi1 == vi1 |
| 2383 | && getOrdering_cache[hash].vi2 == vi2) |
| 2384 | return getOrdering_cache[hash].ord; |
| 2385 | stats__getOrdering_misses++; |
| 2386 | ////-- |
| 2387 | v1 = VtsID__to_VTS(vi1); |
| 2388 | v2 = VtsID__to_VTS(vi2); |
| 2389 | ord = VTS__cmp( v1, v2 ); |
| 2390 | ////++ |
| 2391 | getOrdering_cache[hash].vi1 = vi1; |
| 2392 | getOrdering_cache[hash].vi2 = vi2; |
| 2393 | getOrdering_cache[hash].ord = ord; |
| 2394 | ////-- |
| 2395 | return ord; |
| 2396 | } |
| 2397 | static inline POrd VtsID__getOrdering ( VtsID vi1, VtsID vi2 ) { |
| 2398 | return vi1 == vi2 ? POrd_EQ : VtsID__getOrdering_WRK(vi1, vi2); |
| 2399 | } |
| 2400 | |
| 2401 | /* compute binary join */ |
| 2402 | __attribute__((noinline)) |
| 2403 | static VtsID VtsID__join2_WRK ( VtsID vi1, VtsID vi2 ) { |
| 2404 | UInt hash; |
| 2405 | VtsID res; |
| 2406 | VTS *vts1, *vts2, *nyu; |
| 2407 | //if (vi1 == vi2) return vi1; |
| 2408 | tl_assert(vi1 != vi2); |
| 2409 | ////++ |
| 2410 | stats__join2_queries++; |
| 2411 | hash = hash_VtsIDs(vi1, vi2, N_JOIN2_CACHE); |
| 2412 | if (join2_cache[hash].vi1 == vi1 |
| 2413 | && join2_cache[hash].vi2 == vi2) |
| 2414 | return join2_cache[hash].res; |
| 2415 | stats__join2_misses++; |
| 2416 | ////-- |
| 2417 | vts1 = VtsID__to_VTS(vi1); |
| 2418 | vts2 = VtsID__to_VTS(vi2); |
| 2419 | nyu = VTS__join(vts1,vts2); |
| 2420 | res = vts_tab__find_and_dealloc__or_add(nyu); |
| 2421 | ////++ |
| 2422 | join2_cache[hash].vi1 = vi1; |
| 2423 | join2_cache[hash].vi2 = vi2; |
| 2424 | join2_cache[hash].res = res; |
| 2425 | ////-- |
| 2426 | return res; |
| 2427 | } |
| 2428 | static inline VtsID VtsID__join2 ( VtsID vi1, VtsID vi2 ) { |
| 2429 | return vi1 == vi2 ? vi1 : VtsID__join2_WRK(vi1, vi2); |
| 2430 | } |
| 2431 | |
| 2432 | /* create a singleton VTS, namely [thr:1] */ |
| 2433 | static VtsID VtsID__mk_Singleton ( Thr* thr, ULong tym ) { |
| 2434 | VTS* nyu = VTS__singleton(thr,tym); |
| 2435 | return vts_tab__find_and_dealloc__or_add(nyu); |
| 2436 | } |
| 2437 | |
| 2438 | /* tick operation, creates value 1 if specified index is absent */ |
| 2439 | static VtsID VtsID__tick ( VtsID vi, Thr* idx ) { |
| 2440 | VTS* vts = VtsID__to_VTS(vi); |
| 2441 | VTS* nyu = VTS__tick(idx,vts); |
| 2442 | return vts_tab__find_and_dealloc__or_add(nyu); |
| 2443 | } |
| 2444 | |
| 2445 | /* index into a VTS (only for assertions) */ |
| 2446 | static ULong VtsID__indexAt ( VtsID vi, Thr* idx ) { |
| 2447 | VTS* vts = VtsID__to_VTS(vi); |
| 2448 | return VTS__indexAt_SLOW( vts, idx ); |
| 2449 | } |
| 2450 | |
| 2451 | |
| 2452 | ///////////////////////////////////////////////////////// |
| 2453 | // // |
| 2454 | // Threads // |
| 2455 | // // |
| 2456 | ///////////////////////////////////////////////////////// |
| 2457 | |
| 2458 | struct _Thr { |
| 2459 | /* Current VTSs for this thread. They change as we go along. viR |
| 2460 | is the VTS to be used for reads, viW for writes. Usually they |
| 2461 | are the same, but can differ when we deal with reader-writer |
| 2462 | locks. It is always the case that VtsID__getOrdering(viW,viR) |
| 2463 | == POrd_LT or POrdEQ -- that is, viW must be the same, or |
| 2464 | lagging behind, viR. */ |
| 2465 | VtsID viR; |
| 2466 | VtsID viW; |
| 2467 | /* opaque (to us) data we hold on behalf of the library's user. */ |
| 2468 | void* opaque; |
| 2469 | }; |
| 2470 | |
| 2471 | static Thr* Thr__new ( void ) { |
| 2472 | Thr* thr = HG_(zalloc)( "libhb.Thr__new.1", sizeof(Thr) ); |
| 2473 | thr->viR = VtsID_INVALID; |
| 2474 | thr->viW = VtsID_INVALID; |
| 2475 | return thr; |
| 2476 | } |
| 2477 | |
| 2478 | |
| 2479 | ///////////////////////////////////////////////////////// |
| 2480 | // // |
| 2481 | // Shadow Values // |
| 2482 | // // |
| 2483 | ///////////////////////////////////////////////////////// |
| 2484 | |
| 2485 | // type SVal, SVal_INVALID and SVal_NOACCESS are defined by |
| 2486 | // hb_zsm.h. We have to do everything else here. |
| 2487 | |
| 2488 | /* SVal is 64 bit unsigned int. |
| 2489 | |
| 2490 | <---------30---------> <---------30---------> |
| 2491 | 00 X-----Rmin-VtsID-----X 00 X-----Wmin-VtsID-----X C(Rmin,Wmin) |
| 2492 | 01 X--------------------X XX X--------------------X E(rror) |
| 2493 | 10 X--------------------X XX X--------------------X A: SVal_NOACCESS |
| 2494 | 11 X--------------------X XX X--------------------X I: SVal_INVALID |
| 2495 | */ |
| 2496 | #define SVAL_TAGMASK (3ULL << 62) |
| 2497 | |
| 2498 | static inline Bool SVal__isC ( SVal s ) { |
| 2499 | return (0ULL << 62) == (s & SVAL_TAGMASK); |
| 2500 | } |
| 2501 | static inline SVal SVal__mkC ( VtsID rmini, VtsID wmini ) { |
| 2502 | //tl_assert(VtsID__is_valid(rmini)); |
| 2503 | //tl_assert(VtsID__is_valid(wmini)); |
| 2504 | return (((ULong)rmini) << 32) | ((ULong)wmini); |
| 2505 | } |
| 2506 | static inline VtsID SVal__unC_Rmin ( SVal s ) { |
| 2507 | tl_assert(SVal__isC(s)); |
| 2508 | return (VtsID)(s >> 32); |
| 2509 | } |
| 2510 | static inline VtsID SVal__unC_Wmin ( SVal s ) { |
| 2511 | tl_assert(SVal__isC(s)); |
| 2512 | return (VtsID)(s & 0xFFFFFFFFULL); |
| 2513 | } |
| 2514 | |
| 2515 | static Bool SVal__isE ( SVal s ) { |
| 2516 | return (1ULL << 62) == (s & SVAL_TAGMASK); |
| 2517 | } |
| 2518 | static SVal SVal__mkE ( void ) { |
| 2519 | return 1ULL << 62; |
| 2520 | } |
| 2521 | |
| 2522 | static Bool SVal__isA ( SVal s ) { |
| 2523 | return (2ULL << 62) == (s & SVAL_TAGMASK); |
| 2524 | } |
| 2525 | static SVal SVal__mkA ( void ) { |
| 2526 | return 2ULL << 62; |
| 2527 | } |
| 2528 | |
| 2529 | /* Direct callback from lib_zsm. */ |
| 2530 | static void SVal__rcinc ( SVal s ) { |
| 2531 | if (SVal__isC(s)) { |
| 2532 | VtsID__rcinc( SVal__unC_Rmin(s) ); |
| 2533 | VtsID__rcinc( SVal__unC_Wmin(s) ); |
| 2534 | } |
| 2535 | } |
| 2536 | |
| 2537 | /* Direct callback from lib_zsm. */ |
| 2538 | static void SVal__rcdec ( SVal s ) { |
| 2539 | if (SVal__isC(s)) { |
| 2540 | VtsID__rcdec( SVal__unC_Rmin(s) ); |
| 2541 | VtsID__rcdec( SVal__unC_Wmin(s) ); |
| 2542 | } |
| 2543 | } |
| 2544 | |
| 2545 | |
| 2546 | ///////////////////////////////////////////////////////// |
| 2547 | // // |
| 2548 | // Change-event map2 // |
| 2549 | // // |
| 2550 | ///////////////////////////////////////////////////////// |
| 2551 | |
| 2552 | #define EVENT_MAP_GC_AT (1 * 1000 * 1000) |
| 2553 | #define EVENT_MAP_GC_DISCARD_FRACTION 0.5 |
| 2554 | |
| 2555 | /* This is in two parts: |
| 2556 | |
| 2557 | 1. An OSet of RCECs. This is a set of reference-counted stack |
| 2558 | traces. When the reference count of a stack trace becomes zero, |
| 2559 | it is removed from the set and freed up. The intent is to have |
| 2560 | a set of stack traces which can be referred to from (2), but to |
| 2561 | only represent each one once. The set is indexed/searched by |
| 2562 | ordering on the stack trace vectors. |
| 2563 | |
| 2564 | 2. An OSet of OldRefs. These store information about each old ref |
| 2565 | that we need to record. It is indexed by address of the |
| 2566 | location for which the information is recorded. For LRU |
| 2567 | purposes, each OldRef also contains a generation number, |
| 2568 | indicating when it was most recently accessed. |
| 2569 | |
| 2570 | The important part of an OldRef is, however, its accs[] array. |
| 2571 | This is an array of N_OLDREF_ACCS pairs of Thr and a RCEC. This |
| 2572 | allows us to collect the last access-traceback by up to |
| 2573 | N_OLDREF_ACCS different threads for this location. The accs[] |
| 2574 | array is a MTF-array. If a pair falls off the end, that's too |
| 2575 | bad -- we will lose info about that thread's access to this |
| 2576 | location. |
| 2577 | |
| 2578 | When this OSet becomes too big, we can throw away the entries |
| 2579 | whose generation numbers are below some threshold; hence doing |
| 2580 | approximate LRU discarding. For each discarded OldRef we must |
| 2581 | of course decrement the reference count on the all RCECs it |
| 2582 | refers to, in order that entries from (1) eventually get |
| 2583 | discarded too. |
| 2584 | */ |
| 2585 | |
| 2586 | |
| 2587 | static UWord stats__ctxt_rcdec1 = 0; |
| 2588 | static UWord stats__ctxt_rcdec2 = 0; |
| 2589 | static UWord stats__ctxt_rcdec3 = 0; |
| 2590 | static UWord stats__ctxt_rcdec_calls = 0; |
| 2591 | static UWord stats__ctxt_rcdec_discards = 0; |
| 2592 | static UWord stats__ctxt_rcdec1_eq = 0; |
| 2593 | |
| 2594 | static UWord stats__ctxt_tab_curr = 0; |
| 2595 | static UWord stats__ctxt_tab_max = 0; |
| 2596 | |
| 2597 | static UWord stats__ctxt_tab_qs = 0; |
| 2598 | static UWord stats__ctxt_tab_cmps = 0; |
| 2599 | |
| 2600 | |
| 2601 | /////////////////////////////////////////////////////// |
| 2602 | //// Part (1): An OSet of RCECs |
| 2603 | /// |
| 2604 | |
| 2605 | #define N_FRAMES 8 |
| 2606 | |
| 2607 | // (UInt) `echo "Reference Counted Execution Context" | md5sum` |
| 2608 | #define RCEC_MAGIC 0xab88abb2UL |
| 2609 | |
| 2610 | //#define N_RCEC_TAB 98317 /* prime */ |
| 2611 | #define N_RCEC_TAB 196613 /* prime */ |
| 2612 | |
| 2613 | typedef |
| 2614 | struct _RCEC { |
| 2615 | struct _RCEC* next; |
| 2616 | UWord magic; |
| 2617 | UWord rc; |
| 2618 | UWord rcX; /* used for crosschecking */ |
| 2619 | UWord frames[1 + N_FRAMES]; /* first word is hash of all the rest */ |
| 2620 | } |
| 2621 | RCEC; |
| 2622 | |
| 2623 | static RCEC** contextTab = NULL; /* hash table of RCEC*s */ |
| 2624 | |
| 2625 | |
| 2626 | /* Gives an arbitrary total order on RCEC .frames fields */ |
| 2627 | static Word RCEC__cmp_by_frames ( RCEC* ec1, RCEC* ec2 ) { |
| 2628 | Word i; |
| 2629 | tl_assert(ec1 && ec1->magic == RCEC_MAGIC); |
| 2630 | tl_assert(ec2 && ec2->magic == RCEC_MAGIC); |
| 2631 | if (ec1->frames[0] < ec2->frames[0]) return -1; |
| 2632 | if (ec1->frames[0] > ec2->frames[0]) return 1; |
| 2633 | for (i = 1; i < 1 + N_FRAMES; i++) { |
| 2634 | if (ec1->frames[i] < ec2->frames[i]) return -1; |
| 2635 | if (ec1->frames[i] > ec2->frames[i]) return 1; |
| 2636 | } |
| 2637 | return 0; |
| 2638 | } |
| 2639 | |
| 2640 | |
| 2641 | /* Dec the ref of this RCEC. */ |
| 2642 | static void ctxt__rcdec ( RCEC* ec ) |
| 2643 | { |
| 2644 | stats__ctxt_rcdec_calls++; |
| 2645 | tl_assert(ec && ec->magic == RCEC_MAGIC); |
| 2646 | tl_assert(ec->rc > 0); |
| 2647 | ec->rc--; |
| 2648 | } |
| 2649 | |
| 2650 | static void ctxt__rcinc ( RCEC* ec ) |
| 2651 | { |
| 2652 | tl_assert(ec && ec->magic == RCEC_MAGIC); |
| 2653 | ec->rc++; |
| 2654 | } |
| 2655 | |
| 2656 | |
| 2657 | /* Find 'ec' in the RCEC list whose head pointer lives at 'headp' and |
| 2658 | move it one step closer the the front of the list, so as to make |
| 2659 | subsequent searches for it cheaper. */ |
| 2660 | static void move_RCEC_one_step_forward ( RCEC** headp, RCEC* ec ) |
| 2661 | { |
| 2662 | RCEC *ec0, *ec1, *ec2; |
| 2663 | if (ec == *headp) |
| 2664 | tl_assert(0); /* already at head of list */ |
| 2665 | tl_assert(ec != NULL); |
| 2666 | ec0 = *headp; |
| 2667 | ec1 = NULL; |
| 2668 | ec2 = NULL; |
| 2669 | while (True) { |
| 2670 | if (ec0 == NULL || ec0 == ec) break; |
| 2671 | ec2 = ec1; |
| 2672 | ec1 = ec0; |
| 2673 | ec0 = ec0->next; |
| 2674 | } |
| 2675 | tl_assert(ec0 == ec); |
| 2676 | if (ec0 != NULL && ec1 != NULL && ec2 != NULL) { |
| 2677 | RCEC* tmp; |
| 2678 | /* ec0 points to ec, ec1 to its predecessor, and ec2 to ec1's |
| 2679 | predecessor. Swap ec0 and ec1, that is, move ec0 one step |
| 2680 | closer to the start of the list. */ |
| 2681 | tl_assert(ec2->next == ec1); |
| 2682 | tl_assert(ec1->next == ec0); |
| 2683 | tmp = ec0->next; |
| 2684 | ec2->next = ec0; |
| 2685 | ec0->next = ec1; |
| 2686 | ec1->next = tmp; |
| 2687 | } |
| 2688 | else |
| 2689 | if (ec0 != NULL && ec1 != NULL && ec2 == NULL) { |
| 2690 | /* it's second in the list. */ |
| 2691 | tl_assert(*headp == ec1); |
| 2692 | tl_assert(ec1->next == ec0); |
| 2693 | ec1->next = ec0->next; |
| 2694 | ec0->next = ec1; |
| 2695 | *headp = ec0; |
| 2696 | } |
| 2697 | } |
| 2698 | |
| 2699 | |
| 2700 | /* Find the given RCEC in the tree, and return a pointer to it. Or, |
| 2701 | if not present, add the given one to the tree (by making a copy of |
| 2702 | it, so the caller can immediately deallocate the original) and |
| 2703 | return a pointer to the copy. The caller can safely have 'example' |
| 2704 | on its stack, since we will always return a pointer to a copy of |
| 2705 | it, not to the original. Note that the inserted node will have .rc |
| 2706 | of zero and so the caller must immediatly increment it. */ |
| 2707 | __attribute__((noinline)) |
| 2708 | static RCEC* ctxt__find_or_add ( RCEC* example ) |
| 2709 | { |
| 2710 | UWord hent; |
| 2711 | RCEC* copy; |
| 2712 | tl_assert(example && example->magic == RCEC_MAGIC); |
| 2713 | tl_assert(example->rc == 0); |
| 2714 | |
| 2715 | /* Search the hash table to see if we already have it. */ |
| 2716 | stats__ctxt_tab_qs++; |
| 2717 | hent = example->frames[0] % N_RCEC_TAB; |
| 2718 | copy = contextTab[hent]; |
| 2719 | while (1) { |
| 2720 | if (!copy) break; |
| 2721 | tl_assert(copy->magic == RCEC_MAGIC); |
| 2722 | stats__ctxt_tab_cmps++; |
| 2723 | if (0 == RCEC__cmp_by_frames(copy, example)) break; |
| 2724 | copy = copy->next; |
| 2725 | } |
| 2726 | |
| 2727 | if (copy) { |
| 2728 | tl_assert(copy != example); |
| 2729 | /* optimisation: if it's not at the head of its list, move 1 |
| 2730 | step fwds, to make future searches cheaper */ |
| 2731 | if (copy != contextTab[hent]) { |
| 2732 | move_RCEC_one_step_forward( &contextTab[hent], copy ); |
| 2733 | } |
| 2734 | } else { |
| 2735 | copy = HG_(zalloc)( "libhb.cfoa.1", sizeof(RCEC) ); |
| 2736 | tl_assert(copy != example); |
| 2737 | *copy = *example; |
| 2738 | copy->next = contextTab[hent]; |
| 2739 | contextTab[hent] = copy; |
| 2740 | stats__ctxt_tab_curr++; |
| 2741 | if (stats__ctxt_tab_curr > stats__ctxt_tab_max) |
| 2742 | stats__ctxt_tab_max = stats__ctxt_tab_curr; |
| 2743 | } |
| 2744 | return copy; |
| 2745 | } |
| 2746 | |
| 2747 | static inline UWord ROLW ( UWord w, Int n ) |
| 2748 | { |
| 2749 | Int bpw = 8 * sizeof(UWord); |
| 2750 | w = (w << n) | (w >> (bpw-n)); |
| 2751 | return w; |
| 2752 | } |
| 2753 | |
| 2754 | __attribute__((noinline)) |
| 2755 | static RCEC* get_RCEC ( Thr* thr ) |
| 2756 | { |
| 2757 | UWord hash, i; |
| 2758 | RCEC example; |
| 2759 | example.magic = RCEC_MAGIC; |
| 2760 | example.rc = 0; |
| 2761 | example.rcX = 0; |
| 2762 | main_get_stacktrace( thr, &example.frames[1], N_FRAMES ); |
| 2763 | hash = 0; |
| 2764 | for (i = 1; i < 1 + N_FRAMES; i++) { |
| 2765 | hash ^= example.frames[i]; |
| 2766 | hash = ROLW(hash, 19); |
| 2767 | } |
| 2768 | example.frames[0] = hash; |
| 2769 | return ctxt__find_or_add( &example ); |
| 2770 | } |
| 2771 | |
| 2772 | /////////////////////////////////////////////////////// |
| 2773 | //// Part (2): An OSet of OldRefs, that refer to (1) |
| 2774 | /// |
| 2775 | |
| 2776 | // (UInt) `echo "Old Reference Information" | md5sum` |
| 2777 | #define OldRef_MAGIC 0x30b1f075UL |
| 2778 | |
| 2779 | typedef struct { Thr* thr; RCEC* rcec; } Thr_n_RCEC; |
| 2780 | |
| 2781 | #define N_OLDREF_ACCS 3 |
| 2782 | |
| 2783 | typedef |
| 2784 | struct { |
| 2785 | Addr ea; |
| 2786 | UWord magic; |
| 2787 | UWord gen; /* when most recently accessed */ |
| 2788 | /* unused slots in this array have .thr == NULL */ |
| 2789 | Thr_n_RCEC accs[N_OLDREF_ACCS]; |
| 2790 | } |
| 2791 | OldRef; |
| 2792 | |
| 2793 | static Word OldRef__cmp_by_EA ( OldRef* r1, OldRef* r2 ) { |
| 2794 | tl_assert(r1 && r1->magic == OldRef_MAGIC); |
| 2795 | tl_assert(r2 && r2->magic == OldRef_MAGIC); |
| 2796 | if (r1->ea < r2->ea) return -1; |
| 2797 | if (r1->ea > r2->ea) return 1; |
| 2798 | return 0; |
| 2799 | } |
| 2800 | |
| 2801 | static OSet* oldrefTree = NULL; /* OSet* of OldRef */ |
| 2802 | static UWord oldrefGen = 0; /* current LRU generation # */ |
| 2803 | static UWord oldrefTreeN = 0; /* # elems in oldrefTree */ |
| 2804 | static UWord oldrefGenIncAt = 0; /* inc gen # when size hits this */ |
| 2805 | |
| 2806 | static void event_map_bind ( Addr a, Thr* thr ) |
| 2807 | { |
| 2808 | OldRef key, *ref; |
| 2809 | RCEC* here; |
| 2810 | Word i, j; |
| 2811 | |
| 2812 | key.ea = a; |
| 2813 | key.magic = OldRef_MAGIC; |
| 2814 | |
| 2815 | ref = VG_(OSetGen_Lookup)( oldrefTree, &key ); |
| 2816 | |
| 2817 | if (ref) { |
| 2818 | |
| 2819 | /* We already have a record for this address. We now need to |
| 2820 | see if we have a stack trace pertaining to this thread's |
| 2821 | access. */ |
| 2822 | tl_assert(ref->magic == OldRef_MAGIC); |
| 2823 | |
| 2824 | tl_assert(thr); |
| 2825 | for (i = 0; i < N_OLDREF_ACCS; i++) { |
| 2826 | if (ref->accs[i].thr == thr) |
| 2827 | break; |
| 2828 | } |
| 2829 | |
| 2830 | if (i < N_OLDREF_ACCS) { |
| 2831 | /* thread 'thr' has an entry at index 'i'. Update it. */ |
| 2832 | if (i > 0) { |
| 2833 | Thr_n_RCEC tmp = ref->accs[i-1]; |
| 2834 | ref->accs[i-1] = ref->accs[i]; |
| 2835 | ref->accs[i] = tmp; |
| 2836 | i--; |
| 2837 | } |
| 2838 | here = get_RCEC( thr ); |
| 2839 | if (here == ref->accs[i].rcec) stats__ctxt_rcdec1_eq++; |
| 2840 | ctxt__rcinc( here ); |
| 2841 | stats__ctxt_rcdec1++; |
| 2842 | ctxt__rcdec( ref->accs[i].rcec ); |
| 2843 | ref->accs[i].rcec = here; |
| 2844 | tl_assert(ref->accs[i].thr == thr); |
| 2845 | } else { |
| 2846 | here = get_RCEC( thr ); |
| 2847 | ctxt__rcinc( here ); |
| 2848 | /* No entry for this thread. Shuffle all of them down one |
| 2849 | slot, and put the new entry at the start of the array. */ |
| 2850 | if (ref->accs[N_OLDREF_ACCS-1].thr) { |
| 2851 | /* the last slot is in use. We must dec the rc on the |
| 2852 | associated rcec. */ |
| 2853 | tl_assert(ref->accs[N_OLDREF_ACCS-1].rcec); |
| 2854 | stats__ctxt_rcdec2++; |
| 2855 | ctxt__rcdec(ref->accs[N_OLDREF_ACCS-1].rcec); |
| 2856 | } else { |
| 2857 | tl_assert(!ref->accs[N_OLDREF_ACCS-1].rcec); |
| 2858 | } |
| 2859 | for (j = N_OLDREF_ACCS-1; j >= 1; j--) |
| 2860 | ref->accs[j] = ref->accs[j-1]; |
| 2861 | ref->accs[0].thr = thr; |
| 2862 | ref->accs[0].rcec = here; |
| 2863 | tl_assert(thr); /* thr==NULL is used to signify an empty slot, |
| 2864 | so we can't add a NULL thr. */ |
| 2865 | } |
| 2866 | |
| 2867 | ref->gen = oldrefGen; |
| 2868 | tl_assert(ref->ea == a); |
| 2869 | |
| 2870 | } else { |
| 2871 | |
| 2872 | /* We don't have a record for this address. Create a new one. */ |
| 2873 | if (oldrefTreeN >= oldrefGenIncAt) { |
| 2874 | oldrefGen++; |
| 2875 | oldrefGenIncAt = oldrefTreeN + 50000; |
| 2876 | if (0) VG_(printf)("oldrefTree: new gen %lu at size %lu\n", |
| 2877 | oldrefGen, oldrefTreeN ); |
| 2878 | } |
| 2879 | here = get_RCEC( thr ); |
| 2880 | ctxt__rcinc(here); |
| 2881 | ref = VG_(OSetGen_AllocNode)( oldrefTree, sizeof(OldRef) ); |
| 2882 | ref->magic = OldRef_MAGIC; |
| 2883 | ref->gen = oldrefGen; |
| 2884 | ref->ea = a; |
| 2885 | ref->accs[0].rcec = here; |
| 2886 | ref->accs[0].thr = thr; |
| 2887 | tl_assert(thr); /* thr==NULL is used to signify an empty slot, |
| 2888 | so we can't add a NULL thr. */ |
| 2889 | for (j = 1; j < N_OLDREF_ACCS; j++) { |
| 2890 | ref->accs[j].thr = NULL; |
| 2891 | ref->accs[j].rcec = NULL; |
| 2892 | } |
| 2893 | VG_(OSetGen_Insert)( oldrefTree, ref ); |
| 2894 | oldrefTreeN++; |
| 2895 | |
| 2896 | } |
| 2897 | } |
| 2898 | |
| 2899 | |
| 2900 | static |
| 2901 | Bool event_map_lookup ( /*OUT*/struct _EC** resEC, |
| 2902 | /*OUT*/Thr** resThr, |
| 2903 | Thr* thr_acc, Addr a ) |
| 2904 | { |
| 2905 | Word i; |
| 2906 | OldRef key, *ref; |
| 2907 | |
| 2908 | tl_assert(thr_acc); |
| 2909 | |
| 2910 | key.ea = a; |
| 2911 | key.magic = OldRef_MAGIC; |
| 2912 | |
| 2913 | ref = VG_(OSetGen_Lookup)( oldrefTree, &key ); |
| 2914 | if (ref) { |
| 2915 | tl_assert(ref->magic == OldRef_MAGIC); |
| 2916 | tl_assert(ref->accs[0].thr); /* first slot must always be used */ |
| 2917 | |
| 2918 | for (i = 0; i < N_OLDREF_ACCS; i++) { |
| 2919 | if (ref->accs[i].thr != NULL |
| 2920 | && ref->accs[i].thr != thr_acc) |
| 2921 | break; |
| 2922 | } |
| 2923 | /* If we didn't find an entry for some thread other than |
| 2924 | thr_acc, just return the entry for thread 0. It'll look |
| 2925 | pretty stupid to the user though. */ |
| 2926 | if (i == N_OLDREF_ACCS) |
| 2927 | i = 0; |
| 2928 | |
| 2929 | tl_assert(i >= 0 && i < N_OLDREF_ACCS); |
| 2930 | tl_assert(ref->accs[i].thr); |
| 2931 | tl_assert(ref->accs[i].rcec); |
| 2932 | tl_assert(ref->accs[i].rcec->magic == RCEC_MAGIC); |
| 2933 | |
| 2934 | *resEC = main_stacktrace_to_EC(&ref->accs[i].rcec->frames[1], N_FRAMES); |
| 2935 | *resThr = ref->accs[i].thr; |
| 2936 | return True; |
| 2937 | } else { |
| 2938 | return False; |
| 2939 | } |
| 2940 | } |
| 2941 | |
| 2942 | static void event_map_init ( void ) |
| 2943 | { |
| 2944 | Word i; |
| 2945 | tl_assert(!contextTab); |
| 2946 | contextTab = HG_(zalloc)( "libhb.event_map_init.1 (context table)", |
| 2947 | N_RCEC_TAB * sizeof(RCEC*) ); |
| 2948 | tl_assert(contextTab); |
| 2949 | for (i = 0; i < N_RCEC_TAB; i++) |
| 2950 | contextTab[i] = NULL; |
| 2951 | |
| 2952 | tl_assert(!oldrefTree); |
| 2953 | tl_assert(offsetof(OldRef,ea) == 0); /* prereq for unboxed cmps */ |
| 2954 | oldrefTree = VG_(OSetGen_Create)( |
| 2955 | offsetof(OldRef,ea), /* == 0 */ |
| 2956 | NULL, /* use unboxed cmp on OldRefs */ |
| 2957 | HG_(zalloc), "libhb.event_map_init.2 (oldref tree)", |
| 2958 | HG_(free) |
| 2959 | ); |
| 2960 | tl_assert(oldrefTree); |
| 2961 | |
| 2962 | oldrefGen = 0; |
| 2963 | oldrefGenIncAt = 0; |
| 2964 | oldrefTreeN = 0; |
| 2965 | } |
| 2966 | |
| 2967 | static void event_map__check_reference_counts ( Bool before ) |
| 2968 | { |
| 2969 | RCEC* rcec; |
| 2970 | OldRef* oldref; |
| 2971 | Word i; |
| 2972 | UWord nEnts = 0; |
| 2973 | |
| 2974 | /* Set the 'check' reference counts to zero. Also, optionally |
| 2975 | check that the real reference counts are non-zero. We allow |
| 2976 | these to fall to zero before a GC, but the GC must get rid of |
| 2977 | all those that are zero, hence none should be zero after a |
| 2978 | GC. */ |
| 2979 | for (i = 0; i < N_RCEC_TAB; i++) { |
| 2980 | for (rcec = contextTab[i]; rcec; rcec = rcec->next) { |
| 2981 | nEnts++; |
| 2982 | tl_assert(rcec); |
| 2983 | tl_assert(rcec->magic == RCEC_MAGIC); |
| 2984 | if (!before) |
| 2985 | tl_assert(rcec->rc > 0); |
| 2986 | rcec->rcX = 0; |
| 2987 | } |
| 2988 | } |
| 2989 | |
| 2990 | /* check that the stats are sane */ |
| 2991 | tl_assert(nEnts == stats__ctxt_tab_curr); |
| 2992 | tl_assert(stats__ctxt_tab_curr <= stats__ctxt_tab_max); |
| 2993 | |
| 2994 | /* visit all the referencing points, inc check ref counts */ |
| 2995 | VG_(OSetGen_ResetIter)( oldrefTree ); |
| 2996 | while ( (oldref = VG_(OSetGen_Next)( oldrefTree )) ) { |
| 2997 | tl_assert(oldref->magic == OldRef_MAGIC); |
| 2998 | for (i = 0; i < N_OLDREF_ACCS; i++) { |
| 2999 | if (oldref->accs[i].thr) { |
| 3000 | tl_assert(oldref->accs[i].rcec); |
| 3001 | tl_assert(oldref->accs[i].rcec->magic == RCEC_MAGIC); |
| 3002 | oldref->accs[i].rcec->rcX++; |
| 3003 | } else { |
| 3004 | tl_assert(!oldref->accs[i].rcec); |
| 3005 | } |
| 3006 | } |
| 3007 | } |
| 3008 | |
| 3009 | /* compare check ref counts with actual */ |
| 3010 | for (i = 0; i < N_RCEC_TAB; i++) { |
| 3011 | for (rcec = contextTab[i]; rcec; rcec = rcec->next) { |
| 3012 | tl_assert(rcec->rc == rcec->rcX); |
| 3013 | } |
| 3014 | } |
| 3015 | } |
| 3016 | |
| 3017 | static void event_map_maybe_GC ( void ) |
| 3018 | { |
| 3019 | OldRef* oldref; |
| 3020 | UWord keyW, valW, retained, maxGen; |
| 3021 | WordFM* genMap; |
| 3022 | XArray* refs2del; |
| 3023 | Word i, j, n2del; |
| 3024 | |
| 3025 | if (LIKELY(oldrefTreeN < EVENT_MAP_GC_AT)) |
| 3026 | return; |
| 3027 | |
| 3028 | if (0) |
| 3029 | VG_(printf)("libhb: event_map GC at size %lu\n", oldrefTreeN); |
| 3030 | |
| 3031 | /* Check our counting is sane */ |
| 3032 | tl_assert(oldrefTreeN == (UWord) VG_(OSetGen_Size)( oldrefTree )); |
| 3033 | |
| 3034 | /* Check the reference counts */ |
| 3035 | event_map__check_reference_counts( True/*before*/ ); |
| 3036 | |
| 3037 | /* Compute the distribution of generation values in the ref tree */ |
| 3038 | /* genMap :: generation-number -> count-of-nodes-with-that-number */ |
| 3039 | genMap = VG_(newFM)( HG_(zalloc), "libhb.emmG.1", |
| 3040 | HG_(free), NULL ); |
| 3041 | |
| 3042 | VG_(OSetGen_ResetIter)( oldrefTree ); |
| 3043 | while ( (oldref = VG_(OSetGen_Next)( oldrefTree )) ) { |
| 3044 | UWord key = oldref->gen; |
| 3045 | keyW = valW = 0; |
| 3046 | if (VG_(lookupFM)(genMap, &keyW, &valW, key )) { |
| 3047 | tl_assert(keyW == key); |
| 3048 | tl_assert(valW > 0); |
| 3049 | } |
| 3050 | /* now valW is the old count for generation 'key' */ |
| 3051 | VG_(addToFM)(genMap, key, valW+1); |
| 3052 | } |
| 3053 | |
| 3054 | tl_assert(VG_(sizeFM)(genMap) > 0); |
| 3055 | |
| 3056 | retained = oldrefTreeN; |
| 3057 | maxGen = 0; |
| 3058 | VG_(initIterFM)( genMap ); |
| 3059 | while (VG_(nextIterFM)( genMap, &keyW, &valW )) { |
| 3060 | tl_assert(keyW > 0); /* can't allow a generation # 0 */ |
| 3061 | if (0) VG_(printf)(" XXX: gen %lu has %lu\n", keyW, valW ); |
| 3062 | tl_assert(keyW >= maxGen); |
| 3063 | tl_assert(retained >= valW); |
| 3064 | if (retained - valW |
| 3065 | > (UWord)(EVENT_MAP_GC_AT * EVENT_MAP_GC_DISCARD_FRACTION)) { |
| 3066 | retained -= valW; |
| 3067 | maxGen = keyW; |
| 3068 | } else { |
| 3069 | break; |
| 3070 | } |
| 3071 | } |
| 3072 | VG_(doneIterFM)( genMap ); |
| 3073 | |
| 3074 | VG_(printf)( |
| 3075 | "libhb: EvM GC: delete generations %lu and below, " |
| 3076 | "retaining %lu entries\n", |
| 3077 | maxGen, retained ); |
| 3078 | |
| 3079 | VG_(deleteFM)( genMap, NULL, NULL ); |
| 3080 | |
| 3081 | /* If this fails, it means there's only one generation in the |
| 3082 | entire tree. So we're kind of in a bad situation, and need to |
| 3083 | do some stop-gap measure, such as randomly deleting half the |
| 3084 | entries. */ |
| 3085 | tl_assert(retained < oldrefTreeN); |
| 3086 | |
| 3087 | /* Now make up a big list of the oldrefTree entries we want to |
| 3088 | delete. We can't simultaneously traverse the tree and delete |
| 3089 | stuff from it, so first we need to copy them off somewhere |
| 3090 | else. (sigh) */ |
| 3091 | refs2del = VG_(newXA)( HG_(zalloc), "libhb.emmG.1", |
| 3092 | HG_(free), sizeof(OldRef*) ); |
| 3093 | |
| 3094 | VG_(OSetGen_ResetIter)( oldrefTree ); |
| 3095 | while ( (oldref = VG_(OSetGen_Next)( oldrefTree )) ) { |
| 3096 | tl_assert(oldref->magic == OldRef_MAGIC); |
| 3097 | if (oldref->gen <= maxGen) { |
| 3098 | VG_(addToXA)( refs2del, &oldref ); |
| 3099 | } |
| 3100 | } |
| 3101 | |
| 3102 | n2del = VG_(sizeXA)( refs2del ); |
| 3103 | tl_assert(n2del == (Word)(oldrefTreeN - retained)); |
| 3104 | |
| 3105 | if (0) VG_(printf)("%s","deleting entries\n"); |
| 3106 | for (i = 0; i < n2del; i++) { |
| 3107 | void* nd; |
| 3108 | OldRef* ref = *(OldRef**)VG_(indexXA)( refs2del, i ); |
| 3109 | tl_assert(ref); |
| 3110 | tl_assert(ref->magic == OldRef_MAGIC); |
| 3111 | for (j = 0; j < N_OLDREF_ACCS; j++) { |
| 3112 | if (ref->accs[j].rcec) { |
| 3113 | tl_assert(ref->accs[j].thr); |
| 3114 | stats__ctxt_rcdec3++; |
| 3115 | ctxt__rcdec( ref->accs[j].rcec ); |
| 3116 | } else { |
| 3117 | tl_assert(!ref->accs[j].thr); |
| 3118 | } |
| 3119 | } |
| 3120 | nd = VG_(OSetGen_Remove)( oldrefTree, ref ); |
| 3121 | VG_(OSetGen_FreeNode)( oldrefTree, nd ); |
| 3122 | } |
| 3123 | |
| 3124 | VG_(deleteXA)( refs2del ); |
| 3125 | |
| 3126 | tl_assert( VG_(OSetGen_Size)( oldrefTree ) == retained ); |
| 3127 | |
| 3128 | oldrefTreeN = retained; |
| 3129 | oldrefGenIncAt = oldrefTreeN; /* start new gen right away */ |
| 3130 | |
| 3131 | /* Throw away all RCECs with zero reference counts */ |
| 3132 | for (i = 0; i < N_RCEC_TAB; i++) { |
| 3133 | RCEC** pp = &contextTab[i]; |
| 3134 | RCEC* p = *pp; |
| 3135 | while (p) { |
| 3136 | if (p->rc == 0) { |
| 3137 | *pp = p->next; |
| 3138 | HG_(free)(p); |
| 3139 | p = *pp; |
| 3140 | tl_assert(stats__ctxt_tab_curr > 0); |
| 3141 | stats__ctxt_tab_curr--; |
| 3142 | } else { |
| 3143 | pp = &p->next; |
| 3144 | p = p->next; |
| 3145 | } |
| 3146 | } |
| 3147 | } |
| 3148 | |
| 3149 | /* Check the reference counts */ |
| 3150 | event_map__check_reference_counts( False/*after*/ ); |
| 3151 | |
| 3152 | //if (0) |
| 3153 | //VG_(printf)("XXXX final sizes: oldrefTree %ld, contextTree %ld\n\n", |
| 3154 | // VG_(OSetGen_Size)(oldrefTree), VG_(OSetGen_Size)(contextTree)); |
| 3155 | |
| 3156 | } |
| 3157 | |
| 3158 | |
| 3159 | ///////////////////////////////////////////////////////// |
| 3160 | // // |
| 3161 | // Core MSM // |
| 3162 | // // |
| 3163 | ///////////////////////////////////////////////////////// |
| 3164 | |
| 3165 | #define MSM_CONFACC 1 |
| 3166 | |
| 3167 | #define MSM_RACE2ERR 1 |
| 3168 | |
| 3169 | #define MSM_CHECK 0 |
| 3170 | |
| 3171 | static ULong stats__msm_read = 0; |
| 3172 | static ULong stats__msm_read_change = 0; |
| 3173 | static ULong stats__msm_write = 0; |
| 3174 | static ULong stats__msm_write_change = 0; |
| 3175 | |
| 3176 | __attribute__((noinline)) |
| 3177 | static void record_race_info ( Thr* acc_thr, |
| 3178 | Addr acc_addr, SizeT szB, Bool isWrite, |
| 3179 | SVal svOld, SVal svNew ) |
| 3180 | { |
| 3181 | Bool found; |
| 3182 | Thr* thrp = NULL; |
| 3183 | struct _EC* where = NULL; |
| 3184 | struct _EC* wherep = NULL; |
| 3185 | where = main_get_EC( acc_thr ); |
| 3186 | found = event_map_lookup( &wherep, &thrp, acc_thr, acc_addr ); |
| 3187 | if (found) { |
| 3188 | tl_assert(wherep); |
| 3189 | tl_assert(thrp); |
| 3190 | tl_assert(thrp->opaque); |
| 3191 | tl_assert(acc_thr->opaque); |
| 3192 | HG_(record_error_Race)( acc_thr->opaque, acc_addr, |
| 3193 | isWrite, szB, NULL/*mb_lastlock*/, |
| 3194 | wherep, thrp->opaque ); |
| 3195 | } else { |
| 3196 | tl_assert(!wherep); |
| 3197 | tl_assert(!thrp); |
| 3198 | tl_assert(acc_thr->opaque); |
| 3199 | HG_(record_error_Race)( acc_thr->opaque, acc_addr, |
| 3200 | isWrite, szB, NULL/*mb_lastlock*/, |
| 3201 | NULL, NULL ); |
| 3202 | } |
| 3203 | } |
| 3204 | |
| 3205 | static Bool is_sane_SVal_C ( SVal sv ) { |
| 3206 | POrd ord; |
| 3207 | if (!SVal__isC(sv)) return True; |
| 3208 | ord = VtsID__getOrdering( SVal__unC_Rmin(sv), SVal__unC_Wmin(sv) ); |
| 3209 | if (ord == POrd_EQ || ord == POrd_LT) return True; |
| 3210 | return False; |
| 3211 | } |
| 3212 | |
| 3213 | |
| 3214 | /* Compute new state following a read */ |
| 3215 | static inline SVal msm_read ( SVal svOld, |
| 3216 | /* The following are only needed for |
| 3217 | creating error reports. */ |
| 3218 | Thr* acc_thr, |
| 3219 | Addr acc_addr, SizeT szB ) |
| 3220 | { |
| 3221 | SVal svNew = SVal_INVALID; |
| 3222 | stats__msm_read++; |
| 3223 | |
| 3224 | /* Redundant sanity check on the constraints */ |
| 3225 | if (MSM_CHECK) { |
| 3226 | tl_assert(is_sane_SVal_C(svOld)); |
| 3227 | } |
| 3228 | |
| 3229 | if (SVal__isC(svOld)) { |
| 3230 | POrd ord; |
| 3231 | VtsID tviR = acc_thr->viR; |
| 3232 | VtsID tviW = acc_thr->viW; |
| 3233 | VtsID rmini = SVal__unC_Rmin(svOld); |
| 3234 | VtsID wmini = SVal__unC_Wmin(svOld); |
| 3235 | |
| 3236 | ord = VtsID__getOrdering(rmini,tviR); |
| 3237 | if (ord == POrd_EQ || ord == POrd_LT) { |
| 3238 | /* no race */ |
| 3239 | /* Note: RWLOCK subtlety: use tviW, not tviR */ |
| 3240 | svNew = SVal__mkC( rmini, VtsID__join2(wmini, tviW) ); |
| 3241 | goto out; |
| 3242 | } else { |
| 3243 | svNew = MSM_RACE2ERR |
| 3244 | ? SVal__mkE() |
| 3245 | : SVal__mkC( rmini, VtsID__join2(wmini,tviR) ); |
| 3246 | record_race_info( acc_thr, acc_addr, szB, False/*!isWrite*/, |
| 3247 | svOld, svNew ); |
| 3248 | goto out; |
| 3249 | } |
| 3250 | } |
| 3251 | if (SVal__isA(svOld)) { |
| 3252 | /* reading no-access memory (sigh); leave unchanged */ |
| 3253 | /* check for no pollution */ |
| 3254 | tl_assert(svOld == SVal_NOACCESS); |
| 3255 | svNew = SVal_NOACCESS; |
| 3256 | goto out; |
| 3257 | } |
| 3258 | if (SVal__isE(svOld)) { |
| 3259 | /* no race, location is already "in error" */ |
| 3260 | svNew = SVal__mkE(); |
| 3261 | goto out; |
| 3262 | } |
| 3263 | VG_(printf)("msm_read: bad svOld: 0x%016llx\n", svOld); |
| 3264 | tl_assert(0); |
| 3265 | |
| 3266 | out: |
| 3267 | if (MSM_CHECK) { |
| 3268 | tl_assert(is_sane_SVal_C(svNew)); |
| 3269 | } |
| 3270 | tl_assert(svNew != SVal_INVALID); |
| 3271 | if (svNew != svOld) { |
| 3272 | if (MSM_CONFACC && SVal__isC(svOld) && SVal__isC(svNew)) { |
| 3273 | event_map_bind( acc_addr, acc_thr ); |
| 3274 | stats__msm_read_change++; |
| 3275 | } |
| 3276 | } |
| 3277 | return svNew; |
| 3278 | } |
| 3279 | |
| 3280 | |
| 3281 | /* Compute new state following a write */ |
| 3282 | static inline SVal msm_write ( SVal svOld, |
| 3283 | /* The following are only needed for |
| 3284 | creating error reports. */ |
| 3285 | Thr* acc_thr, |
| 3286 | Addr acc_addr, SizeT szB ) |
| 3287 | { |
| 3288 | SVal svNew = SVal_INVALID; |
| 3289 | stats__msm_write++; |
| 3290 | |
| 3291 | /* Redundant sanity check on the constraints */ |
| 3292 | if (MSM_CHECK) { |
| 3293 | tl_assert(is_sane_SVal_C(svOld)); |
| 3294 | } |
| 3295 | |
| 3296 | if (SVal__isC(svOld)) { |
| 3297 | POrd ord; |
| 3298 | VtsID tviW = acc_thr->viW; |
| 3299 | VtsID wmini = SVal__unC_Wmin(svOld); |
| 3300 | |
| 3301 | ord = VtsID__getOrdering(wmini,tviW); |
| 3302 | if (ord == POrd_EQ || ord == POrd_LT) { |
| 3303 | /* no race */ |
| 3304 | svNew = SVal__mkC( tviW, tviW ); |
| 3305 | goto out; |
| 3306 | } else { |
| 3307 | VtsID rmini = SVal__unC_Rmin(svOld); |
| 3308 | svNew = MSM_RACE2ERR |
| 3309 | ? SVal__mkE() |
| 3310 | : SVal__mkC( VtsID__join2(rmini,tviW), |
| 3311 | VtsID__join2(wmini,tviW) ); |
| 3312 | record_race_info( acc_thr, acc_addr, szB, True/*isWrite*/, |
| 3313 | svOld, svNew ); |
| 3314 | goto out; |
| 3315 | } |
| 3316 | } |
| 3317 | if (SVal__isA(svOld)) { |
| 3318 | /* writing no-access memory (sigh); leave unchanged */ |
| 3319 | /* check for no pollution */ |
| 3320 | tl_assert(svOld == SVal_NOACCESS); |
| 3321 | svNew = SVal_NOACCESS; |
| 3322 | goto out; |
| 3323 | } |
| 3324 | if (SVal__isE(svOld)) { |
| 3325 | /* no race, location is already "in error" */ |
| 3326 | svNew = SVal__mkE(); |
| 3327 | goto out; |
| 3328 | } |
| 3329 | VG_(printf)("msm_write: bad svOld: 0x%016llx\n", svOld); |
| 3330 | tl_assert(0); |
| 3331 | |
| 3332 | out: |
| 3333 | if (MSM_CHECK) { |
| 3334 | tl_assert(is_sane_SVal_C(svNew)); |
| 3335 | } |
| 3336 | tl_assert(svNew != SVal_INVALID); |
| 3337 | if (svNew != svOld) { |
| 3338 | if (MSM_CONFACC && SVal__isC(svOld) && SVal__isC(svNew)) { |
| 3339 | event_map_bind( acc_addr, acc_thr ); |
| 3340 | stats__msm_write_change++; |
| 3341 | } |
| 3342 | } |
| 3343 | return svNew; |
| 3344 | } |
| 3345 | |
| 3346 | |
| 3347 | ///////////////////////////////////////////////////////// |
| 3348 | // // |
| 3349 | // Apply core MSM to specific memory locations // |
| 3350 | // // |
| 3351 | ///////////////////////////////////////////////////////// |
| 3352 | |
| 3353 | /*------------- ZSM accesses: 8 bit apply ------------- */ |
| 3354 | |
| 3355 | void zsm_apply8___msm_read ( Thr* thr, Addr a ) { |
| 3356 | CacheLine* cl; |
| 3357 | UWord cloff, tno, toff; |
| 3358 | SVal svOld, svNew; |
| 3359 | UShort descr; |
| 3360 | stats__cline_read8s++; |
| 3361 | cl = get_cacheline(a); |
| 3362 | cloff = get_cacheline_offset(a); |
| 3363 | tno = get_treeno(a); |
| 3364 | toff = get_tree_offset(a); /* == 0 .. 7 */ |
| 3365 | descr = cl->descrs[tno]; |
| 3366 | if (UNLIKELY( !(descr & (TREE_DESCR_8_0 << toff)) )) { |
| 3367 | SVal* tree = &cl->svals[tno << 3]; |
| 3368 | cl->descrs[tno] = pulldown_to_8(tree, toff, descr); |
| 3369 | if (SCE_CACHELINE) |
| 3370 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 3371 | } |
| 3372 | svOld = cl->svals[cloff]; |
| 3373 | svNew = msm_read( svOld, thr,a,1 ); |
| 3374 | tl_assert(svNew != SVal_INVALID); |
| 3375 | cl->svals[cloff] = svNew; |
| 3376 | } |
| 3377 | |
| 3378 | void zsm_apply8___msm_write ( Thr* thr, Addr a ) { |
| 3379 | CacheLine* cl; |
| 3380 | UWord cloff, tno, toff; |
| 3381 | SVal svOld, svNew; |
| 3382 | UShort descr; |
| 3383 | stats__cline_read8s++; |
| 3384 | cl = get_cacheline(a); |
| 3385 | cloff = get_cacheline_offset(a); |
| 3386 | tno = get_treeno(a); |
| 3387 | toff = get_tree_offset(a); /* == 0 .. 7 */ |
| 3388 | descr = cl->descrs[tno]; |
| 3389 | if (UNLIKELY( !(descr & (TREE_DESCR_8_0 << toff)) )) { |
| 3390 | SVal* tree = &cl->svals[tno << 3]; |
| 3391 | cl->descrs[tno] = pulldown_to_8(tree, toff, descr); |
| 3392 | if (SCE_CACHELINE) |
| 3393 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 3394 | } |
| 3395 | svOld = cl->svals[cloff]; |
| 3396 | svNew = msm_write( svOld, thr,a,1 ); |
| 3397 | tl_assert(svNew != SVal_INVALID); |
| 3398 | cl->svals[cloff] = svNew; |
| 3399 | } |
| 3400 | |
| 3401 | /*------------- ZSM accesses: 16 bit apply ------------- */ |
| 3402 | |
| 3403 | void zsm_apply16___msm_read ( Thr* thr, Addr a ) { |
| 3404 | CacheLine* cl; |
| 3405 | UWord cloff, tno, toff; |
| 3406 | SVal svOld, svNew; |
| 3407 | UShort descr; |
| 3408 | stats__cline_read16s++; |
| 3409 | if (UNLIKELY(!aligned16(a))) goto slowcase; |
| 3410 | cl = get_cacheline(a); |
| 3411 | cloff = get_cacheline_offset(a); |
| 3412 | tno = get_treeno(a); |
| 3413 | toff = get_tree_offset(a); /* == 0, 2, 4 or 6 */ |
| 3414 | descr = cl->descrs[tno]; |
| 3415 | if (UNLIKELY( !(descr & (TREE_DESCR_16_0 << toff)) )) { |
| 3416 | if (valid_value_is_below_me_16(descr, toff)) { |
| 3417 | goto slowcase; |
| 3418 | } else { |
| 3419 | SVal* tree = &cl->svals[tno << 3]; |
| 3420 | cl->descrs[tno] = pulldown_to_16(tree, toff, descr); |
| 3421 | } |
| 3422 | if (SCE_CACHELINE) |
| 3423 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 3424 | } |
| 3425 | svOld = cl->svals[cloff]; |
| 3426 | svNew = msm_read( svOld, thr,a,2 ); |
| 3427 | tl_assert(svNew != SVal_INVALID); |
| 3428 | cl->svals[cloff] = svNew; |
| 3429 | return; |
| 3430 | slowcase: /* misaligned, or must go further down the tree */ |
| 3431 | stats__cline_16to8splits++; |
| 3432 | zsm_apply8___msm_read( thr, a + 0 ); |
| 3433 | zsm_apply8___msm_read( thr, a + 1 ); |
| 3434 | } |
| 3435 | |
| 3436 | void zsm_apply16___msm_write ( Thr* thr, Addr a ) { |
| 3437 | CacheLine* cl; |
| 3438 | UWord cloff, tno, toff; |
| 3439 | SVal svOld, svNew; |
| 3440 | UShort descr; |
| 3441 | stats__cline_read16s++; |
| 3442 | if (UNLIKELY(!aligned16(a))) goto slowcase; |
| 3443 | cl = get_cacheline(a); |
| 3444 | cloff = get_cacheline_offset(a); |
| 3445 | tno = get_treeno(a); |
| 3446 | toff = get_tree_offset(a); /* == 0, 2, 4 or 6 */ |
| 3447 | descr = cl->descrs[tno]; |
| 3448 | if (UNLIKELY( !(descr & (TREE_DESCR_16_0 << toff)) )) { |
| 3449 | if (valid_value_is_below_me_16(descr, toff)) { |
| 3450 | goto slowcase; |
| 3451 | } else { |
| 3452 | SVal* tree = &cl->svals[tno << 3]; |
| 3453 | cl->descrs[tno] = pulldown_to_16(tree, toff, descr); |
| 3454 | } |
| 3455 | if (SCE_CACHELINE) |
| 3456 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 3457 | } |
| 3458 | svOld = cl->svals[cloff]; |
| 3459 | svNew = msm_write( svOld, thr,a,2 ); |
| 3460 | tl_assert(svNew != SVal_INVALID); |
| 3461 | cl->svals[cloff] = svNew; |
| 3462 | return; |
| 3463 | slowcase: /* misaligned, or must go further down the tree */ |
| 3464 | stats__cline_16to8splits++; |
| 3465 | zsm_apply8___msm_write( thr, a + 0 ); |
| 3466 | zsm_apply8___msm_write( thr, a + 1 ); |
| 3467 | } |
| 3468 | |
| 3469 | /*------------- ZSM accesses: 32 bit apply ------------- */ |
| 3470 | |
| 3471 | void zsm_apply32___msm_read ( Thr* thr, Addr a ) { |
| 3472 | CacheLine* cl; |
| 3473 | UWord cloff, tno, toff; |
| 3474 | SVal svOld, svNew; |
| 3475 | UShort descr; |
| 3476 | if (UNLIKELY(!aligned32(a))) goto slowcase; |
| 3477 | cl = get_cacheline(a); |
| 3478 | cloff = get_cacheline_offset(a); |
| 3479 | tno = get_treeno(a); |
| 3480 | toff = get_tree_offset(a); /* == 0 or 4 */ |
| 3481 | descr = cl->descrs[tno]; |
| 3482 | if (UNLIKELY( !(descr & (TREE_DESCR_32_0 << toff)) )) { |
| 3483 | if (valid_value_is_above_me_32(descr, toff)) { |
| 3484 | SVal* tree = &cl->svals[tno << 3]; |
| 3485 | cl->descrs[tno] = pulldown_to_32(tree, toff, descr); |
| 3486 | } else { |
| 3487 | goto slowcase; |
| 3488 | } |
| 3489 | if (SCE_CACHELINE) |
| 3490 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 3491 | } |
| 3492 | svOld = cl->svals[cloff]; |
| 3493 | svNew = msm_read( svOld, thr,a,4 ); |
| 3494 | tl_assert(svNew != SVal_INVALID); |
| 3495 | cl->svals[cloff] = svNew; |
| 3496 | return; |
| 3497 | slowcase: /* misaligned, or must go further down the tree */ |
| 3498 | stats__cline_32to16splits++; |
| 3499 | zsm_apply16___msm_read( thr, a + 0 ); |
| 3500 | zsm_apply16___msm_read( thr, a + 2 ); |
| 3501 | } |
| 3502 | |
| 3503 | void zsm_apply32___msm_write ( Thr* thr, Addr a ) { |
| 3504 | CacheLine* cl; |
| 3505 | UWord cloff, tno, toff; |
| 3506 | SVal svOld, svNew; |
| 3507 | UShort descr; |
| 3508 | if (UNLIKELY(!aligned32(a))) goto slowcase; |
| 3509 | cl = get_cacheline(a); |
| 3510 | cloff = get_cacheline_offset(a); |
| 3511 | tno = get_treeno(a); |
| 3512 | toff = get_tree_offset(a); /* == 0 or 4 */ |
| 3513 | descr = cl->descrs[tno]; |
| 3514 | if (UNLIKELY( !(descr & (TREE_DESCR_32_0 << toff)) )) { |
| 3515 | if (valid_value_is_above_me_32(descr, toff)) { |
| 3516 | SVal* tree = &cl->svals[tno << 3]; |
| 3517 | cl->descrs[tno] = pulldown_to_32(tree, toff, descr); |
| 3518 | } else { |
| 3519 | goto slowcase; |
| 3520 | } |
| 3521 | if (SCE_CACHELINE) |
| 3522 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 3523 | } |
| 3524 | svOld = cl->svals[cloff]; |
| 3525 | svNew = msm_write( svOld, thr,a,4 ); |
| 3526 | tl_assert(svNew != SVal_INVALID); |
| 3527 | cl->svals[cloff] = svNew; |
| 3528 | return; |
| 3529 | slowcase: /* misaligned, or must go further down the tree */ |
| 3530 | stats__cline_32to16splits++; |
| 3531 | zsm_apply16___msm_write( thr, a + 0 ); |
| 3532 | zsm_apply16___msm_write( thr, a + 2 ); |
| 3533 | } |
| 3534 | |
| 3535 | /*------------- ZSM accesses: 64 bit apply ------------- */ |
| 3536 | |
| 3537 | void zsm_apply64___msm_read ( Thr* thr, Addr a ) { |
| 3538 | CacheLine* cl; |
| 3539 | UWord cloff, tno, toff; |
| 3540 | SVal svOld, svNew; |
| 3541 | UShort descr; |
| 3542 | stats__cline_read64s++; |
| 3543 | if (UNLIKELY(!aligned64(a))) goto slowcase; |
| 3544 | cl = get_cacheline(a); |
| 3545 | cloff = get_cacheline_offset(a); |
| 3546 | tno = get_treeno(a); |
| 3547 | toff = get_tree_offset(a); /* == 0, unused */ |
| 3548 | descr = cl->descrs[tno]; |
| 3549 | if (UNLIKELY( !(descr & TREE_DESCR_64) )) { |
| 3550 | goto slowcase; |
| 3551 | } |
| 3552 | svOld = cl->svals[cloff]; |
| 3553 | svNew = msm_read( svOld, thr,a,8 ); |
| 3554 | tl_assert(svNew != SVal_INVALID); |
| 3555 | cl->svals[cloff] = svNew; |
| 3556 | return; |
| 3557 | slowcase: /* misaligned, or must go further down the tree */ |
| 3558 | stats__cline_64to32splits++; |
| 3559 | zsm_apply32___msm_read( thr, a + 0 ); |
| 3560 | zsm_apply32___msm_read( thr, a + 4 ); |
| 3561 | } |
| 3562 | |
| 3563 | void zsm_apply64___msm_write ( Thr* thr, Addr a ) { |
| 3564 | CacheLine* cl; |
| 3565 | UWord cloff, tno, toff; |
| 3566 | SVal svOld, svNew; |
| 3567 | UShort descr; |
| 3568 | stats__cline_read64s++; |
| 3569 | if (UNLIKELY(!aligned64(a))) goto slowcase; |
| 3570 | cl = get_cacheline(a); |
| 3571 | cloff = get_cacheline_offset(a); |
| 3572 | tno = get_treeno(a); |
| 3573 | toff = get_tree_offset(a); /* == 0, unused */ |
| 3574 | descr = cl->descrs[tno]; |
| 3575 | if (UNLIKELY( !(descr & TREE_DESCR_64) )) { |
| 3576 | goto slowcase; |
| 3577 | } |
| 3578 | svOld = cl->svals[cloff]; |
| 3579 | svNew = msm_write( svOld, thr,a,8 ); |
| 3580 | tl_assert(svNew != SVal_INVALID); |
| 3581 | cl->svals[cloff] = svNew; |
| 3582 | return; |
| 3583 | slowcase: /* misaligned, or must go further down the tree */ |
| 3584 | stats__cline_64to32splits++; |
| 3585 | zsm_apply32___msm_write( thr, a + 0 ); |
| 3586 | zsm_apply32___msm_write( thr, a + 4 ); |
| 3587 | } |
| 3588 | |
| 3589 | /*--------------- ZSM accesses: 8 bit write --------------- */ |
| 3590 | |
| 3591 | static |
| 3592 | void zsm_write8 ( Addr a, SVal svNew ) { |
| 3593 | CacheLine* cl; |
| 3594 | UWord cloff, tno, toff; |
| 3595 | UShort descr; |
| 3596 | stats__cline_set8s++; |
| 3597 | cl = get_cacheline(a); |
| 3598 | cloff = get_cacheline_offset(a); |
| 3599 | tno = get_treeno(a); |
| 3600 | toff = get_tree_offset(a); /* == 0 .. 7 */ |
| 3601 | descr = cl->descrs[tno]; |
| 3602 | if (UNLIKELY( !(descr & (TREE_DESCR_8_0 << toff)) )) { |
| 3603 | SVal* tree = &cl->svals[tno << 3]; |
| 3604 | cl->descrs[tno] = pulldown_to_8(tree, toff, descr); |
| 3605 | if (SCE_CACHELINE) |
| 3606 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 3607 | } |
| 3608 | tl_assert(svNew != SVal_INVALID); |
| 3609 | cl->svals[cloff] = svNew; |
| 3610 | } |
| 3611 | |
| 3612 | /*--------------- ZSM accesses: 16 bit write --------------- */ |
| 3613 | |
| 3614 | static |
| 3615 | void zsm_write16 ( Addr a, SVal svNew ) { |
| 3616 | CacheLine* cl; |
| 3617 | UWord cloff, tno, toff; |
| 3618 | UShort descr; |
| 3619 | stats__cline_set16s++; |
| 3620 | if (UNLIKELY(!aligned16(a))) goto slowcase; |
| 3621 | cl = get_cacheline(a); |
| 3622 | cloff = get_cacheline_offset(a); |
| 3623 | tno = get_treeno(a); |
| 3624 | toff = get_tree_offset(a); /* == 0, 2, 4 or 6 */ |
| 3625 | descr = cl->descrs[tno]; |
| 3626 | if (UNLIKELY( !(descr & (TREE_DESCR_16_0 << toff)) )) { |
| 3627 | if (valid_value_is_below_me_16(descr, toff)) { |
| 3628 | /* Writing at this level. Need to fix up 'descr'. */ |
| 3629 | cl->descrs[tno] = pullup_descr_to_16(descr, toff); |
| 3630 | /* At this point, the tree does not match cl->descr[tno] any |
| 3631 | more. The assignments below will fix it up. */ |
| 3632 | } else { |
| 3633 | /* We can't indiscriminately write on the w16 node as in the |
| 3634 | w64 case, as that might make the node inconsistent with |
| 3635 | its parent. So first, pull down to this level. */ |
| 3636 | SVal* tree = &cl->svals[tno << 3]; |
| 3637 | cl->descrs[tno] = pulldown_to_16(tree, toff, descr); |
| 3638 | if (SCE_CACHELINE) |
| 3639 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 3640 | } |
| 3641 | } |
| 3642 | tl_assert(svNew != SVal_INVALID); |
| 3643 | cl->svals[cloff + 0] = svNew; |
| 3644 | cl->svals[cloff + 1] = SVal_INVALID; |
| 3645 | return; |
| 3646 | slowcase: /* misaligned */ |
| 3647 | stats__cline_16to8splits++; |
| 3648 | zsm_write8( a + 0, svNew ); |
| 3649 | zsm_write8( a + 1, svNew ); |
| 3650 | } |
| 3651 | |
| 3652 | /*--------------- ZSM accesses: 32 bit write --------------- */ |
| 3653 | |
| 3654 | static |
| 3655 | void zsm_write32 ( Addr a, SVal svNew ) { |
| 3656 | CacheLine* cl; |
| 3657 | UWord cloff, tno, toff; |
| 3658 | UShort descr; |
| 3659 | stats__cline_set32s++; |
| 3660 | if (UNLIKELY(!aligned32(a))) goto slowcase; |
| 3661 | cl = get_cacheline(a); |
| 3662 | cloff = get_cacheline_offset(a); |
| 3663 | tno = get_treeno(a); |
| 3664 | toff = get_tree_offset(a); /* == 0 or 4 */ |
| 3665 | descr = cl->descrs[tno]; |
| 3666 | if (UNLIKELY( !(descr & (TREE_DESCR_32_0 << toff)) )) { |
| 3667 | if (valid_value_is_above_me_32(descr, toff)) { |
| 3668 | /* We can't indiscriminately write on the w32 node as in the |
| 3669 | w64 case, as that might make the node inconsistent with |
| 3670 | its parent. So first, pull down to this level. */ |
| 3671 | SVal* tree = &cl->svals[tno << 3]; |
| 3672 | cl->descrs[tno] = pulldown_to_32(tree, toff, descr); |
| 3673 | if (SCE_CACHELINE) |
| 3674 | tl_assert(is_sane_CacheLine(cl)); /* EXPENSIVE */ |
| 3675 | } else { |
| 3676 | /* Writing at this level. Need to fix up 'descr'. */ |
| 3677 | cl->descrs[tno] = pullup_descr_to_32(descr, toff); |
| 3678 | /* At this point, the tree does not match cl->descr[tno] any |
| 3679 | more. The assignments below will fix it up. */ |
| 3680 | } |
| 3681 | } |
| 3682 | tl_assert(svNew != SVal_INVALID); |
| 3683 | cl->svals[cloff + 0] = svNew; |
| 3684 | cl->svals[cloff + 1] = SVal_INVALID; |
| 3685 | cl->svals[cloff + 2] = SVal_INVALID; |
| 3686 | cl->svals[cloff + 3] = SVal_INVALID; |
| 3687 | return; |
| 3688 | slowcase: /* misaligned */ |
| 3689 | stats__cline_32to16splits++; |
| 3690 | zsm_write16( a + 0, svNew ); |
| 3691 | zsm_write16( a + 2, svNew ); |
| 3692 | } |
| 3693 | |
| 3694 | /*--------------- ZSM accesses: 64 bit write --------------- */ |
| 3695 | |
| 3696 | static |
| 3697 | void zsm_write64 ( Addr a, SVal svNew ) { |
| 3698 | CacheLine* cl; |
| 3699 | UWord cloff, tno, toff; |
| 3700 | stats__cline_set64s++; |
| 3701 | if (UNLIKELY(!aligned64(a))) goto slowcase; |
| 3702 | cl = get_cacheline(a); |
| 3703 | cloff = get_cacheline_offset(a); |
| 3704 | tno = get_treeno(a); |
| 3705 | toff = get_tree_offset(a); /* == 0 */ |
| 3706 | cl->descrs[tno] = TREE_DESCR_64; |
| 3707 | tl_assert(svNew != SVal_INVALID); |
| 3708 | cl->svals[cloff + 0] = svNew; |
| 3709 | cl->svals[cloff + 1] = SVal_INVALID; |
| 3710 | cl->svals[cloff + 2] = SVal_INVALID; |
| 3711 | cl->svals[cloff + 3] = SVal_INVALID; |
| 3712 | cl->svals[cloff + 4] = SVal_INVALID; |
| 3713 | cl->svals[cloff + 5] = SVal_INVALID; |
| 3714 | cl->svals[cloff + 6] = SVal_INVALID; |
| 3715 | cl->svals[cloff + 7] = SVal_INVALID; |
| 3716 | return; |
| 3717 | slowcase: /* misaligned */ |
| 3718 | stats__cline_64to32splits++; |
| 3719 | zsm_write32( a + 0, svNew ); |
| 3720 | zsm_write32( a + 4, svNew ); |
| 3721 | } |
| 3722 | |
| 3723 | /*------------- ZSM accesses: 8 bit read/copy ------------- */ |
| 3724 | |
| 3725 | static |
| 3726 | SVal zsm_read8 ( Addr a ) { |
| 3727 | CacheLine* cl; |
| 3728 | UWord cloff, tno, toff; |
| 3729 | UShort descr; |
| 3730 | stats__cline_get8s++; |
| 3731 | cl = get_cacheline(a); |
| 3732 | cloff = get_cacheline_offset(a); |
| 3733 | tno = get_treeno(a); |
| 3734 | toff = get_tree_offset(a); /* == 0 .. 7 */ |
| 3735 | descr = cl->descrs[tno]; |
| 3736 | if (UNLIKELY( !(descr & (TREE_DESCR_8_0 << toff)) )) { |
| 3737 | SVal* tree = &cl->svals[tno << 3]; |
| 3738 | cl->descrs[tno] = pulldown_to_8(tree, toff, descr); |
| 3739 | } |
| 3740 | return cl->svals[cloff]; |
| 3741 | } |
| 3742 | |
| 3743 | static void zsm_copy8 ( Addr src, Addr dst, Bool uu_normalise ) { |
| 3744 | SVal sv; |
| 3745 | stats__cline_copy8s++; |
| 3746 | sv = zsm_read8( src ); |
| 3747 | zsm_write8( dst, sv ); |
| 3748 | } |
| 3749 | |
| 3750 | /* ------------ Shadow memory range setting ops ------------ */ |
| 3751 | |
| 3752 | void zsm_apply_range___msm_read ( Thr* thr, |
| 3753 | Addr a, SizeT len ) |
| 3754 | { |
| 3755 | /* fast track a couple of common cases */ |
| 3756 | if (len == 4 && aligned32(a)) { |
| 3757 | zsm_apply32___msm_read( thr, a ); |
| 3758 | return; |
| 3759 | } |
| 3760 | if (len == 8 && aligned64(a)) { |
| 3761 | zsm_apply64___msm_read( thr, a ); |
| 3762 | return; |
| 3763 | } |
| 3764 | |
| 3765 | /* be completely general (but as efficient as possible) */ |
| 3766 | if (len == 0) return; |
| 3767 | |
| 3768 | if (!aligned16(a) && len >= 1) { |
| 3769 | zsm_apply8___msm_read( thr, a ); |
| 3770 | a += 1; |
| 3771 | len -= 1; |
| 3772 | tl_assert(aligned16(a)); |
| 3773 | } |
| 3774 | if (len == 0) return; |
| 3775 | |
| 3776 | if (!aligned32(a) && len >= 2) { |
| 3777 | zsm_apply16___msm_read( thr, a ); |
| 3778 | a += 2; |
| 3779 | len -= 2; |
| 3780 | tl_assert(aligned32(a)); |
| 3781 | } |
| 3782 | if (len == 0) return; |
| 3783 | |
| 3784 | if (!aligned64(a) && len >= 4) { |
| 3785 | zsm_apply32___msm_read( thr, a ); |
| 3786 | a += 4; |
| 3787 | len -= 4; |
| 3788 | tl_assert(aligned64(a)); |
| 3789 | } |
| 3790 | if (len == 0) return; |
| 3791 | |
| 3792 | if (len >= 8) { |
| 3793 | tl_assert(aligned64(a)); |
| 3794 | while (len >= 8) { |
| 3795 | zsm_apply64___msm_read( thr, a ); |
| 3796 | a += 8; |
| 3797 | len -= 8; |
| 3798 | } |
| 3799 | tl_assert(aligned64(a)); |
| 3800 | } |
| 3801 | if (len == 0) return; |
| 3802 | |
| 3803 | if (len >= 4) |
| 3804 | tl_assert(aligned32(a)); |
| 3805 | if (len >= 4) { |
| 3806 | zsm_apply32___msm_read( thr, a ); |
| 3807 | a += 4; |
| 3808 | len -= 4; |
| 3809 | } |
| 3810 | if (len == 0) return; |
| 3811 | |
| 3812 | if (len >= 2) |
| 3813 | tl_assert(aligned16(a)); |
| 3814 | if (len >= 2) { |
| 3815 | zsm_apply16___msm_read( thr, a ); |
| 3816 | a += 2; |
| 3817 | len -= 2; |
| 3818 | } |
| 3819 | if (len == 0) return; |
| 3820 | |
| 3821 | if (len >= 1) { |
| 3822 | zsm_apply8___msm_read( thr, a ); |
| 3823 | a += 1; |
| 3824 | len -= 1; |
| 3825 | } |
| 3826 | tl_assert(len == 0); |
| 3827 | } |
| 3828 | |
| 3829 | |
| 3830 | |
| 3831 | void zsm_apply_range___msm_write ( Thr* thr, |
| 3832 | Addr a, SizeT len ) |
| 3833 | { |
| 3834 | /* fast track a couple of common cases */ |
| 3835 | if (len == 4 && aligned32(a)) { |
| 3836 | zsm_apply32___msm_write( thr, a ); |
| 3837 | return; |
| 3838 | } |
| 3839 | if (len == 8 && aligned64(a)) { |
| 3840 | zsm_apply64___msm_write( thr, a ); |
| 3841 | return; |
| 3842 | } |
| 3843 | |
| 3844 | /* be completely general (but as efficient as possible) */ |
| 3845 | if (len == 0) return; |
| 3846 | |
| 3847 | if (!aligned16(a) && len >= 1) { |
| 3848 | zsm_apply8___msm_write( thr, a ); |
| 3849 | a += 1; |
| 3850 | len -= 1; |
| 3851 | tl_assert(aligned16(a)); |
| 3852 | } |
| 3853 | if (len == 0) return; |
| 3854 | |
| 3855 | if (!aligned32(a) && len >= 2) { |
| 3856 | zsm_apply16___msm_write( thr, a ); |
| 3857 | a += 2; |
| 3858 | len -= 2; |
| 3859 | tl_assert(aligned32(a)); |
| 3860 | } |
| 3861 | if (len == 0) return; |
| 3862 | |
| 3863 | if (!aligned64(a) && len >= 4) { |
| 3864 | zsm_apply32___msm_write( thr, a ); |
| 3865 | a += 4; |
| 3866 | len -= 4; |
| 3867 | tl_assert(aligned64(a)); |
| 3868 | } |
| 3869 | if (len == 0) return; |
| 3870 | |
| 3871 | if (len >= 8) { |
| 3872 | tl_assert(aligned64(a)); |
| 3873 | while (len >= 8) { |
| 3874 | zsm_apply64___msm_write( thr, a ); |
| 3875 | a += 8; |
| 3876 | len -= 8; |
| 3877 | } |
| 3878 | tl_assert(aligned64(a)); |
| 3879 | } |
| 3880 | if (len == 0) return; |
| 3881 | |
| 3882 | if (len >= 4) |
| 3883 | tl_assert(aligned32(a)); |
| 3884 | if (len >= 4) { |
| 3885 | zsm_apply32___msm_write( thr, a ); |
| 3886 | a += 4; |
| 3887 | len -= 4; |
| 3888 | } |
| 3889 | if (len == 0) return; |
| 3890 | |
| 3891 | if (len >= 2) |
| 3892 | tl_assert(aligned16(a)); |
| 3893 | if (len >= 2) { |
| 3894 | zsm_apply16___msm_write( thr, a ); |
| 3895 | a += 2; |
| 3896 | len -= 2; |
| 3897 | } |
| 3898 | if (len == 0) return; |
| 3899 | |
| 3900 | if (len >= 1) { |
| 3901 | zsm_apply8___msm_write( thr, a ); |
| 3902 | a += 1; |
| 3903 | len -= 1; |
| 3904 | } |
| 3905 | tl_assert(len == 0); |
| 3906 | } |
| 3907 | |
| 3908 | |
| 3909 | |
| 3910 | |
| 3911 | /* Block-copy states (needed for implementing realloc()). */ |
| 3912 | |
| 3913 | static void zsm_copy_range ( Addr src, Addr dst, SizeT len ) |
| 3914 | { |
| 3915 | SizeT i; |
| 3916 | if (len == 0) |
| 3917 | return; |
| 3918 | |
| 3919 | /* assert for non-overlappingness */ |
| 3920 | tl_assert(src+len <= dst || dst+len <= src); |
| 3921 | |
| 3922 | /* To be simple, just copy byte by byte. But so as not to wreck |
| 3923 | performance for later accesses to dst[0 .. len-1], normalise |
| 3924 | destination lines as we finish with them, and also normalise the |
| 3925 | line containing the first and last address. */ |
| 3926 | for (i = 0; i < len; i++) { |
| 3927 | Bool normalise |
| 3928 | = get_cacheline_offset( dst+i+1 ) == 0 /* last in line */ |
| 3929 | || i == 0 /* first in range */ |
| 3930 | || i == len-1; /* last in range */ |
| 3931 | zsm_copy8( src+i, dst+i, normalise ); |
| 3932 | } |
| 3933 | } |
| 3934 | |
| 3935 | |
| 3936 | /* For setting address ranges to a given value. Has considerable |
| 3937 | sophistication so as to avoid generating large numbers of pointless |
| 3938 | cache loads/writebacks for large ranges. */ |
| 3939 | |
| 3940 | /* Do small ranges in-cache, in the obvious way. */ |
| 3941 | static |
| 3942 | void zsm_set_range_SMALL ( Addr a, SizeT len, SVal svNew ) |
| 3943 | { |
| 3944 | /* fast track a couple of common cases */ |
| 3945 | if (len == 4 && aligned32(a)) { |
| 3946 | zsm_write32( a, svNew ); |
| 3947 | return; |
| 3948 | } |
| 3949 | if (len == 8 && aligned64(a)) { |
| 3950 | zsm_write64( a, svNew ); |
| 3951 | return; |
| 3952 | } |
| 3953 | |
| 3954 | /* be completely general (but as efficient as possible) */ |
| 3955 | if (len == 0) return; |
| 3956 | |
| 3957 | if (!aligned16(a) && len >= 1) { |
| 3958 | zsm_write8( a, svNew ); |
| 3959 | a += 1; |
| 3960 | len -= 1; |
| 3961 | tl_assert(aligned16(a)); |
| 3962 | } |
| 3963 | if (len == 0) return; |
| 3964 | |
| 3965 | if (!aligned32(a) && len >= 2) { |
| 3966 | zsm_write16( a, svNew ); |
| 3967 | a += 2; |
| 3968 | len -= 2; |
| 3969 | tl_assert(aligned32(a)); |
| 3970 | } |
| 3971 | if (len == 0) return; |
| 3972 | |
| 3973 | if (!aligned64(a) && len >= 4) { |
| 3974 | zsm_write32( a, svNew ); |
| 3975 | a += 4; |
| 3976 | len -= 4; |
| 3977 | tl_assert(aligned64(a)); |
| 3978 | } |
| 3979 | if (len == 0) return; |
| 3980 | |
| 3981 | if (len >= 8) { |
| 3982 | tl_assert(aligned64(a)); |
| 3983 | while (len >= 8) { |
| 3984 | zsm_write64( a, svNew ); |
| 3985 | a += 8; |
| 3986 | len -= 8; |
| 3987 | } |
| 3988 | tl_assert(aligned64(a)); |
| 3989 | } |
| 3990 | if (len == 0) return; |
| 3991 | |
| 3992 | if (len >= 4) |
| 3993 | tl_assert(aligned32(a)); |
| 3994 | if (len >= 4) { |
| 3995 | zsm_write32( a, svNew ); |
| 3996 | a += 4; |
| 3997 | len -= 4; |
| 3998 | } |
| 3999 | if (len == 0) return; |
| 4000 | |
| 4001 | if (len >= 2) |
| 4002 | tl_assert(aligned16(a)); |
| 4003 | if (len >= 2) { |
| 4004 | zsm_write16( a, svNew ); |
| 4005 | a += 2; |
| 4006 | len -= 2; |
| 4007 | } |
| 4008 | if (len == 0) return; |
| 4009 | |
| 4010 | if (len >= 1) { |
| 4011 | zsm_write8( a, svNew ); |
| 4012 | a += 1; |
| 4013 | len -= 1; |
| 4014 | } |
| 4015 | tl_assert(len == 0); |
| 4016 | } |
| 4017 | |
| 4018 | |
| 4019 | /* If we're doing a small range, hand off to zsm_set_range_SMALL. But |
| 4020 | for larger ranges, try to operate directly on the out-of-cache |
| 4021 | representation, rather than dragging lines into the cache, |
| 4022 | overwriting them, and forcing them out. This turns out to be an |
| 4023 | important performance optimisation. */ |
| 4024 | |
| 4025 | static void zsm_set_range ( Addr a, SizeT len, SVal svNew ) |
| 4026 | { |
| 4027 | tl_assert(svNew != SVal_INVALID); |
| 4028 | stats__cache_make_New_arange += (ULong)len; |
| 4029 | |
| 4030 | if (0 && len > 500) |
| 4031 | VG_(printf)("make New ( %#lx, %ld )\n", a, len ); |
| 4032 | |
| 4033 | if (0) { |
| 4034 | static UWord n_New_in_cache = 0; |
| 4035 | static UWord n_New_not_in_cache = 0; |
| 4036 | /* tag is 'a' with the in-line offset masked out, |
| 4037 | eg a[31]..a[4] 0000 */ |
| 4038 | Addr tag = a & ~(N_LINE_ARANGE - 1); |
| 4039 | UWord wix = (a >> N_LINE_BITS) & (N_WAY_NENT - 1); |
| 4040 | if (LIKELY(tag == cache_shmem.tags0[wix])) { |
| 4041 | n_New_in_cache++; |
| 4042 | } else { |
| 4043 | n_New_not_in_cache++; |
| 4044 | } |
| 4045 | if (0 == ((n_New_in_cache + n_New_not_in_cache) % 100000)) |
| 4046 | VG_(printf)("shadow_mem_make_New: IN %lu OUT %lu\n", |
| 4047 | n_New_in_cache, n_New_not_in_cache ); |
| 4048 | } |
| 4049 | |
| 4050 | if (LIKELY(len < 2 * N_LINE_ARANGE)) { |
| 4051 | zsm_set_range_SMALL( a, len, svNew ); |
| 4052 | } else { |
| 4053 | Addr before_start = a; |
| 4054 | Addr aligned_start = cacheline_ROUNDUP(a); |
| 4055 | Addr after_start = cacheline_ROUNDDN(a + len); |
| 4056 | UWord before_len = aligned_start - before_start; |
| 4057 | UWord aligned_len = after_start - aligned_start; |
| 4058 | UWord after_len = a + len - after_start; |
| 4059 | tl_assert(before_start <= aligned_start); |
| 4060 | tl_assert(aligned_start <= after_start); |
| 4061 | tl_assert(before_len < N_LINE_ARANGE); |
| 4062 | tl_assert(after_len < N_LINE_ARANGE); |
| 4063 | tl_assert(get_cacheline_offset(aligned_start) == 0); |
| 4064 | if (get_cacheline_offset(a) == 0) { |
| 4065 | tl_assert(before_len == 0); |
| 4066 | tl_assert(a == aligned_start); |
| 4067 | } |
| 4068 | if (get_cacheline_offset(a+len) == 0) { |
| 4069 | tl_assert(after_len == 0); |
| 4070 | tl_assert(after_start == a+len); |
| 4071 | } |
| 4072 | if (before_len > 0) { |
| 4073 | zsm_set_range_SMALL( before_start, before_len, svNew ); |
| 4074 | } |
| 4075 | if (after_len > 0) { |
| 4076 | zsm_set_range_SMALL( after_start, after_len, svNew ); |
| 4077 | } |
| 4078 | stats__cache_make_New_inZrep += (ULong)aligned_len; |
| 4079 | |
| 4080 | while (1) { |
| 4081 | Addr tag; |
| 4082 | UWord wix; |
| 4083 | if (aligned_start >= after_start) |
| 4084 | break; |
| 4085 | tl_assert(get_cacheline_offset(aligned_start) == 0); |
| 4086 | tag = aligned_start & ~(N_LINE_ARANGE - 1); |
| 4087 | wix = (aligned_start >> N_LINE_BITS) & (N_WAY_NENT - 1); |
| 4088 | if (tag == cache_shmem.tags0[wix]) { |
| 4089 | UWord i; |
| 4090 | for (i = 0; i < N_LINE_ARANGE / 8; i++) |
| 4091 | zsm_write64( aligned_start + i * 8, svNew ); |
| 4092 | } else { |
| 4093 | UWord i; |
| 4094 | Word zix; |
| 4095 | SecMap* sm; |
| 4096 | LineZ* lineZ; |
| 4097 | /* This line is not in the cache. Do not force it in; instead |
| 4098 | modify it in-place. */ |
| 4099 | /* find the Z line to write in and rcdec it or the |
| 4100 | associated F line. */ |
| 4101 | find_Z_for_writing( &sm, &zix, tag ); |
| 4102 | tl_assert(sm); |
| 4103 | tl_assert(zix >= 0 && zix < N_SECMAP_ZLINES); |
| 4104 | lineZ = &sm->linesZ[zix]; |
| 4105 | lineZ->dict[0] = svNew; |
| 4106 | lineZ->dict[1] = lineZ->dict[2] = lineZ->dict[3] = SVal_INVALID; |
| 4107 | for (i = 0; i < N_LINE_ARANGE/4; i++) |
| 4108 | lineZ->ix2s[i] = 0; /* all refer to dict[0] */ |
| 4109 | rcinc_LineZ(lineZ); |
| 4110 | } |
| 4111 | aligned_start += N_LINE_ARANGE; |
| 4112 | aligned_len -= N_LINE_ARANGE; |
| 4113 | } |
| 4114 | tl_assert(aligned_start == after_start); |
| 4115 | tl_assert(aligned_len == 0); |
| 4116 | } |
| 4117 | } |
| 4118 | |
| 4119 | |
| 4120 | ///////////////////////////////////////////////////////// |
| 4121 | // // |
| 4122 | // Synchronisation objects // |
| 4123 | // // |
| 4124 | ///////////////////////////////////////////////////////// |
| 4125 | |
| 4126 | // (UInt) `echo "Synchronisation object" | md5sum` |
| 4127 | #define SO_MAGIC 0x56b3c5b0U |
| 4128 | |
| 4129 | struct _SO { |
| 4130 | VtsID viR; /* r-clock of sender */ |
| 4131 | VtsID viW; /* w-clock of sender */ |
| 4132 | UInt magic; |
| 4133 | }; |
| 4134 | |
| 4135 | static SO* SO__Alloc ( void ) { |
| 4136 | SO* so = HG_(zalloc)( "libhb.SO__Alloc.1", sizeof(SO) ); |
| 4137 | so->viR = VtsID_INVALID; |
| 4138 | so->viW = VtsID_INVALID; |
| 4139 | so->magic = SO_MAGIC; |
| 4140 | return so; |
| 4141 | } |
| 4142 | static void SO__Dealloc ( SO* so ) { |
| 4143 | tl_assert(so); |
| 4144 | tl_assert(so->magic == SO_MAGIC); |
| 4145 | if (so->viR == VtsID_INVALID) { |
| 4146 | tl_assert(so->viW == VtsID_INVALID); |
| 4147 | } else { |
| 4148 | tl_assert(so->viW != VtsID_INVALID); |
| 4149 | VtsID__rcdec(so->viR); |
| 4150 | VtsID__rcdec(so->viW); |
| 4151 | } |
| 4152 | so->magic = 0; |
| 4153 | HG_(free)( so ); |
| 4154 | } |
| 4155 | |
| 4156 | |
| 4157 | ///////////////////////////////////////////////////////// |
| 4158 | // // |
| 4159 | // Top Level API // |
| 4160 | // // |
| 4161 | ///////////////////////////////////////////////////////// |
| 4162 | |
| 4163 | static void show_thread_state ( HChar* str, Thr* t ) |
| 4164 | { |
| 4165 | if (1) return; |
| 4166 | if (t->viR == t->viW) { |
| 4167 | VG_(printf)("thr \"%s\" %p has vi* %u==", str, t, t->viR ); |
| 4168 | VtsID__pp( t->viR ); |
| 4169 | VG_(printf)("%s","\n"); |
| 4170 | } else { |
| 4171 | VG_(printf)("thr \"%s\" %p has viR %u==", str, t, t->viR ); |
| 4172 | VtsID__pp( t->viR ); |
| 4173 | VG_(printf)(" viW %u==", t->viW); |
| 4174 | VtsID__pp( t->viW ); |
| 4175 | VG_(printf)("%s","\n"); |
| 4176 | } |
| 4177 | } |
| 4178 | |
| 4179 | |
| 4180 | Thr* libhb_init ( |
| 4181 | void (*get_stacktrace)( Thr*, Addr*, UWord ), |
| 4182 | struct _EC* (*stacktrace_to_EC)( Addr*, UWord ), |
| 4183 | struct _EC* (*get_EC)( Thr* ) |
| 4184 | ) |
| 4185 | { |
| 4186 | Thr* thr; |
| 4187 | VtsID vi; |
| 4188 | tl_assert(get_stacktrace); |
| 4189 | tl_assert(stacktrace_to_EC); |
| 4190 | tl_assert(get_EC); |
| 4191 | main_get_stacktrace = get_stacktrace; |
| 4192 | main_stacktrace_to_EC = stacktrace_to_EC; |
| 4193 | main_get_EC = get_EC; |
| 4194 | |
| 4195 | // No need to initialise hg_wordfm. |
| 4196 | // No need to initialise hg_wordset. |
| 4197 | |
| 4198 | vts_set_init(); |
| 4199 | vts_tab_init(); |
| 4200 | event_map_init(); |
| 4201 | VtsID__invalidate_caches(); |
| 4202 | |
| 4203 | // initialise shadow memory |
| 4204 | zsm_init( SVal__rcinc, SVal__rcdec ); |
| 4205 | |
| 4206 | thr = Thr__new(); |
| 4207 | vi = VtsID__mk_Singleton( thr, 1 ); |
| 4208 | thr->viR = vi; |
| 4209 | thr->viW = vi; |
| 4210 | VtsID__rcinc(thr->viR); |
| 4211 | VtsID__rcinc(thr->viW); |
| 4212 | |
| 4213 | show_thread_state(" root", thr); |
| 4214 | return thr; |
| 4215 | } |
| 4216 | |
| 4217 | Thr* libhb_create ( Thr* parent ) |
| 4218 | { |
| 4219 | /* The child's VTSs are copies of the parent's VTSs, but ticked at |
| 4220 | the child's index. Since the child's index is guaranteed |
| 4221 | unique, it has never been seen before, so the implicit value |
| 4222 | before the tick is zero and after that is one. */ |
| 4223 | Thr* child = Thr__new(); |
| 4224 | |
| 4225 | child->viR = VtsID__tick( parent->viR, child ); |
| 4226 | child->viW = VtsID__tick( parent->viW, child ); |
| 4227 | VtsID__rcinc(child->viR); |
| 4228 | VtsID__rcinc(child->viW); |
| 4229 | |
| 4230 | tl_assert(VtsID__indexAt( child->viR, child ) == 1); |
| 4231 | tl_assert(VtsID__indexAt( child->viW, child ) == 1); |
| 4232 | |
| 4233 | /* and the parent has to move along too */ |
| 4234 | VtsID__rcdec(parent->viR); |
| 4235 | VtsID__rcdec(parent->viW); |
| 4236 | parent->viR = VtsID__tick( parent->viR, parent ); |
| 4237 | parent->viW = VtsID__tick( parent->viW, parent ); |
| 4238 | VtsID__rcinc(parent->viR); |
| 4239 | VtsID__rcinc(parent->viW); |
| 4240 | |
| 4241 | show_thread_state(" child", child); |
| 4242 | show_thread_state("parent", parent); |
| 4243 | |
| 4244 | return child; |
| 4245 | } |
| 4246 | |
| 4247 | /* Shut down the library, and print stats (in fact that's _all_ |
| 4248 | this is for. */ |
| 4249 | void libhb_shutdown ( Bool show_stats ) |
| 4250 | { |
| 4251 | if (show_stats) { |
| 4252 | VG_(printf)("%s","<<< BEGIN libhb stats >>>\n"); |
| 4253 | VG_(printf)(" secmaps: %'10lu allocd (%'12lu g-a-range)\n", |
| 4254 | stats__secmaps_allocd, |
| 4255 | stats__secmap_ga_space_covered); |
| 4256 | VG_(printf)(" linesZ: %'10lu allocd (%'12lu bytes occupied)\n", |
| 4257 | stats__secmap_linesZ_allocd, |
| 4258 | stats__secmap_linesZ_bytes); |
| 4259 | VG_(printf)(" linesF: %'10lu allocd (%'12lu bytes occupied)\n", |
| 4260 | stats__secmap_linesF_allocd, |
| 4261 | stats__secmap_linesF_bytes); |
| 4262 | VG_(printf)(" secmaps: %'10lu iterator steppings\n", |
| 4263 | stats__secmap_iterator_steppings); |
| 4264 | VG_(printf)(" secmaps: %'10lu searches (%'12lu slow)\n", |
| 4265 | stats__secmaps_search, stats__secmaps_search_slow); |
| 4266 | |
| 4267 | VG_(printf)("%s","\n"); |
| 4268 | VG_(printf)(" cache: %'lu totrefs (%'lu misses)\n", |
| 4269 | stats__cache_totrefs, stats__cache_totmisses ); |
| 4270 | VG_(printf)(" cache: %'14lu Z-fetch, %'14lu F-fetch\n", |
| 4271 | stats__cache_Z_fetches, stats__cache_F_fetches ); |
| 4272 | VG_(printf)(" cache: %'14lu Z-wback, %'14lu F-wback\n", |
| 4273 | stats__cache_Z_wbacks, stats__cache_F_wbacks ); |
| 4274 | VG_(printf)(" cache: %'14lu invals, %'14lu flushes\n", |
| 4275 | stats__cache_invals, stats__cache_flushes ); |
| 4276 | VG_(printf)(" cache: %'14llu arange_New %'14llu direct-to-Zreps\n", |
| 4277 | stats__cache_make_New_arange, |
| 4278 | stats__cache_make_New_inZrep); |
| 4279 | |
| 4280 | VG_(printf)("%s","\n"); |
| 4281 | VG_(printf)(" cline: %'10lu normalises\n", |
| 4282 | stats__cline_normalises ); |
| 4283 | VG_(printf)(" cline: rds 8/4/2/1: %'13lu %'13lu %'13lu %'13lu\n", |
| 4284 | stats__cline_read64s, |
| 4285 | stats__cline_read32s, |
| 4286 | stats__cline_read16s, |
| 4287 | stats__cline_read8s ); |
| 4288 | VG_(printf)(" cline: wrs 8/4/2/1: %'13lu %'13lu %'13lu %'13lu\n", |
| 4289 | stats__cline_write64s, |
| 4290 | stats__cline_write32s, |
| 4291 | stats__cline_write16s, |
| 4292 | stats__cline_write8s ); |
| 4293 | VG_(printf)(" cline: sets 8/4/2/1: %'13lu %'13lu %'13lu %'13lu\n", |
| 4294 | stats__cline_set64s, |
| 4295 | stats__cline_set32s, |
| 4296 | stats__cline_set16s, |
| 4297 | stats__cline_set8s ); |
| 4298 | VG_(printf)(" cline: get1s %'lu, copy1s %'lu\n", |
| 4299 | stats__cline_get8s, stats__cline_copy8s ); |
| 4300 | VG_(printf)(" cline: splits: 8to4 %'12lu 4to2 %'12lu 2to1 %'12lu\n", |
| 4301 | stats__cline_64to32splits, |
| 4302 | stats__cline_32to16splits, |
| 4303 | stats__cline_16to8splits ); |
| 4304 | VG_(printf)(" cline: pulldowns: 8to4 %'12lu 4to2 %'12lu 2to1 %'12lu\n", |
| 4305 | stats__cline_64to32pulldown, |
| 4306 | stats__cline_32to16pulldown, |
| 4307 | stats__cline_16to8pulldown ); |
| 4308 | if (0) |
| 4309 | VG_(printf)(" cline: sizeof(CacheLineZ) %ld, covers %ld bytes of arange\n", |
| 4310 | (Word)sizeof(LineZ), (Word)N_LINE_ARANGE); |
| 4311 | |
| 4312 | VG_(printf)("%s","\n"); |
| 4313 | |
| 4314 | VG_(printf)(" libhb: %'13llu msm_read (%'llu changed)\n", |
| 4315 | stats__msm_read, stats__msm_read_change); |
| 4316 | VG_(printf)(" libhb: %'13llu msm_write (%'llu changed)\n", |
| 4317 | stats__msm_write, stats__msm_write_change); |
| 4318 | VG_(printf)(" libhb: %'13llu getOrd queries (%'llu misses)\n", |
| 4319 | stats__getOrdering_queries, stats__getOrdering_misses); |
| 4320 | VG_(printf)(" libhb: %'13llu join2 queries (%'llu misses)\n", |
| 4321 | stats__join2_queries, stats__join2_misses); |
| 4322 | |
| 4323 | VG_(printf)("%s","\n"); |
| 4324 | VG_(printf)( |
| 4325 | " libhb: %ld entries in vts_table (approximately %lu bytes)\n", |
| 4326 | VG_(sizeXA)( vts_tab ), VG_(sizeXA)( vts_tab ) * sizeof(VtsTE) |
| 4327 | ); |
| 4328 | VG_(printf)( " libhb: %lu entries in vts_set\n", |
| 4329 | VG_(sizeFM)( vts_set ) ); |
| 4330 | |
| 4331 | VG_(printf)("%s","\n"); |
| 4332 | VG_(printf)( " libhb: ctxt__rcdec: 1=%lu(%lu eq), 2=%lu, 3=%lu\n", |
| 4333 | stats__ctxt_rcdec1, stats__ctxt_rcdec1_eq, |
| 4334 | stats__ctxt_rcdec2, |
| 4335 | stats__ctxt_rcdec3 ); |
| 4336 | VG_(printf)( " libhb: ctxt__rcdec: calls %lu, discards %lu\n", |
| 4337 | stats__ctxt_rcdec_calls, stats__ctxt_rcdec_discards); |
| 4338 | VG_(printf)( " libhb: contextTab: %lu slots, %lu max ents\n", |
| 4339 | (UWord)N_RCEC_TAB, |
| 4340 | stats__ctxt_tab_curr ); |
| 4341 | VG_(printf)( " libhb: contextTab: %lu queries, %lu cmps\n", |
| 4342 | stats__ctxt_tab_qs, |
| 4343 | stats__ctxt_tab_cmps ); |
| 4344 | #if 0 |
| 4345 | VG_(printf)("sizeof(AvlNode) = %lu\n", sizeof(AvlNode)); |
| 4346 | VG_(printf)("sizeof(WordBag) = %lu\n", sizeof(WordBag)); |
| 4347 | VG_(printf)("sizeof(MaybeWord) = %lu\n", sizeof(MaybeWord)); |
| 4348 | VG_(printf)("sizeof(CacheLine) = %lu\n", sizeof(CacheLine)); |
| 4349 | VG_(printf)("sizeof(LineZ) = %lu\n", sizeof(LineZ)); |
| 4350 | VG_(printf)("sizeof(LineF) = %lu\n", sizeof(LineF)); |
| 4351 | VG_(printf)("sizeof(SecMap) = %lu\n", sizeof(SecMap)); |
| 4352 | VG_(printf)("sizeof(Cache) = %lu\n", sizeof(Cache)); |
| 4353 | VG_(printf)("sizeof(SMCacheEnt) = %lu\n", sizeof(SMCacheEnt)); |
| 4354 | VG_(printf)("sizeof(CountedSVal) = %lu\n", sizeof(CountedSVal)); |
| 4355 | VG_(printf)("sizeof(VTS) = %lu\n", sizeof(VTS)); |
| 4356 | VG_(printf)("sizeof(ScalarTS) = %lu\n", sizeof(ScalarTS)); |
| 4357 | VG_(printf)("sizeof(VtsTE) = %lu\n", sizeof(VtsTE)); |
| 4358 | VG_(printf)("sizeof(MSMInfo) = %lu\n", sizeof(MSMInfo)); |
| 4359 | |
| 4360 | VG_(printf)("sizeof(struct _XArray) = %lu\n", sizeof(struct _XArray)); |
| 4361 | VG_(printf)("sizeof(struct _WordFM) = %lu\n", sizeof(struct _WordFM)); |
| 4362 | VG_(printf)("sizeof(struct _Thr) = %lu\n", sizeof(struct _Thr)); |
| 4363 | VG_(printf)("sizeof(struct _SO) = %lu\n", sizeof(struct _SO)); |
| 4364 | #endif |
| 4365 | |
| 4366 | VG_(printf)("%s","<<< END libhb stats >>>\n"); |
| 4367 | VG_(printf)("%s","\n"); |
| 4368 | |
| 4369 | } |
| 4370 | } |
| 4371 | |
| 4372 | void libhb_async_exit ( Thr* thr ) |
| 4373 | { |
| 4374 | /* is there anything we need to do? */ |
| 4375 | } |
| 4376 | |
| 4377 | /* Both Segs and SOs point to VTSs. However, there is no sharing, so |
| 4378 | a Seg that points at a VTS is its one-and-only owner, and ditto for |
| 4379 | a SO that points at a VTS. */ |
| 4380 | |
| 4381 | SO* libhb_so_alloc ( void ) |
| 4382 | { |
| 4383 | return SO__Alloc(); |
| 4384 | } |
| 4385 | |
| 4386 | void libhb_so_dealloc ( SO* so ) |
| 4387 | { |
| 4388 | tl_assert(so); |
| 4389 | tl_assert(so->magic == SO_MAGIC); |
| 4390 | SO__Dealloc(so); |
| 4391 | } |
| 4392 | |
| 4393 | /* See comments in libhb.h for details on the meaning of |
| 4394 | strong vs weak sends and strong vs weak receives. */ |
| 4395 | void libhb_so_send ( Thr* thr, SO* so, Bool strong_send ) |
| 4396 | { |
| 4397 | /* Copy the VTSs from 'thr' into the sync object, and then move |
| 4398 | the thread along one step. */ |
| 4399 | |
| 4400 | tl_assert(so); |
| 4401 | tl_assert(so->magic == SO_MAGIC); |
| 4402 | |
| 4403 | /* stay sane .. a thread's read-clock must always lead or be the |
| 4404 | same as its write-clock */ |
| 4405 | { POrd ord = VtsID__getOrdering(thr->viW, thr->viR); |
| 4406 | tl_assert(ord == POrd_EQ || ord == POrd_LT); |
| 4407 | } |
| 4408 | |
| 4409 | /* since we're overwriting the VtsIDs in the SO, we need to drop |
| 4410 | any references made by the previous contents thereof */ |
| 4411 | if (so->viR == VtsID_INVALID) { |
| 4412 | tl_assert(so->viW == VtsID_INVALID); |
| 4413 | so->viR = thr->viR; |
| 4414 | so->viW = thr->viW; |
| 4415 | VtsID__rcinc(so->viR); |
| 4416 | VtsID__rcinc(so->viW); |
| 4417 | } else { |
| 4418 | /* In a strong send, we dump any previous VC in the SO and |
| 4419 | install the sending thread's VC instead. For a weak send we |
| 4420 | must join2 with what's already there. */ |
| 4421 | tl_assert(so->viW != VtsID_INVALID); |
| 4422 | VtsID__rcdec(so->viR); |
| 4423 | VtsID__rcdec(so->viW); |
| 4424 | so->viR = strong_send ? thr->viR : VtsID__join2( so->viR, thr->viR ); |
| 4425 | so->viW = strong_send ? thr->viW : VtsID__join2( so->viW, thr->viW ); |
| 4426 | VtsID__rcinc(so->viR); |
| 4427 | VtsID__rcinc(so->viW); |
| 4428 | } |
| 4429 | |
| 4430 | /* move both parent clocks along */ |
| 4431 | VtsID__rcdec(thr->viR); |
| 4432 | VtsID__rcdec(thr->viW); |
| 4433 | thr->viR = VtsID__tick( thr->viR, thr ); |
| 4434 | thr->viW = VtsID__tick( thr->viW, thr ); |
| 4435 | VtsID__rcinc(thr->viR); |
| 4436 | VtsID__rcinc(thr->viW); |
| 4437 | if (strong_send) |
| 4438 | show_thread_state("s-send", thr); |
| 4439 | else |
| 4440 | show_thread_state("w-send", thr); |
| 4441 | } |
| 4442 | |
| 4443 | void libhb_so_recv ( Thr* thr, SO* so, Bool strong_recv ) |
| 4444 | { |
| 4445 | tl_assert(so); |
| 4446 | tl_assert(so->magic == SO_MAGIC); |
| 4447 | |
| 4448 | if (so->viR != VtsID_INVALID) { |
| 4449 | tl_assert(so->viW != VtsID_INVALID); |
| 4450 | |
| 4451 | /* Weak receive (basically, an R-acquisition of a R-W lock). |
| 4452 | This advances the read-clock of the receiver, but not the |
| 4453 | write-clock. */ |
| 4454 | VtsID__rcdec(thr->viR); |
| 4455 | thr->viR = VtsID__join2( thr->viR, so->viR ); |
| 4456 | VtsID__rcinc(thr->viR); |
| 4457 | |
| 4458 | /* For a strong receive, we also advance the receiver's write |
| 4459 | clock, which means the receive as a whole is essentially |
| 4460 | equivalent to a W-acquisition of a R-W lock. */ |
| 4461 | if (strong_recv) { |
| 4462 | VtsID__rcdec(thr->viW); |
| 4463 | thr->viW = VtsID__join2( thr->viW, so->viW ); |
| 4464 | VtsID__rcinc(thr->viW); |
| 4465 | } |
| 4466 | |
| 4467 | if (strong_recv) |
| 4468 | show_thread_state("s-recv", thr); |
| 4469 | else |
| 4470 | show_thread_state("w-recv", thr); |
| 4471 | |
| 4472 | } else { |
| 4473 | tl_assert(so->viW == VtsID_INVALID); |
| 4474 | /* Deal with degenerate case: 'so' has no vts, so there has been |
| 4475 | no message posted to it. Just ignore this case. */ |
| 4476 | show_thread_state("d-recv", thr); |
| 4477 | } |
| 4478 | } |
| 4479 | |
| 4480 | Bool libhb_so_everSent ( SO* so ) |
| 4481 | { |
| 4482 | if (so->viR == VtsID_INVALID) { |
| 4483 | tl_assert(so->viW == VtsID_INVALID); |
| 4484 | return False; |
| 4485 | } else { |
| 4486 | tl_assert(so->viW != VtsID_INVALID); |
| 4487 | return True; |
| 4488 | } |
| 4489 | } |
| 4490 | |
| 4491 | #define XXX1 0 // 0x67a106c |
| 4492 | #define XXX2 0 |
| 4493 | |
| 4494 | static Bool TRACEME(Addr a, SizeT szB) { |
| 4495 | if (XXX1 && a <= XXX1 && XXX1 <= a+szB) return True; |
| 4496 | if (XXX2 && a <= XXX2 && XXX2 <= a+szB) return True; |
| 4497 | return False; |
| 4498 | } |
| 4499 | static void trace ( Thr* thr, Addr a, SizeT szB, HChar* s ) { |
| 4500 | SVal sv = zsm_read8(a); |
| 4501 | VG_(printf)("thr %p (%#lx,%lu) %s: 0x%016llx ", thr,a,szB,s,sv); |
| 4502 | show_thread_state("", thr); |
| 4503 | VG_(printf)("%s","\n"); |
| 4504 | } |
| 4505 | |
| 4506 | void libhb_range_new ( Thr* thr, Addr a, SizeT szB ) |
| 4507 | { |
| 4508 | SVal sv = SVal__mkC(thr->viW, thr->viW); |
| 4509 | tl_assert(is_sane_SVal_C(sv)); |
| 4510 | if(TRACEME(a,szB))trace(thr,a,szB,"nw-before"); |
| 4511 | zsm_set_range( a, szB, sv ); |
| 4512 | if(TRACEME(a,szB))trace(thr,a,szB,"nw-after "); |
| 4513 | } |
| 4514 | |
| 4515 | void libhb_range_noaccess ( Thr* thr, Addr a, SizeT szB ) |
| 4516 | { |
| 4517 | if(TRACEME(a,szB))trace(thr,a,szB,"NA-before"); |
| 4518 | zsm_set_range( a, szB, SVal__mkA() ); |
| 4519 | if(TRACEME(a,szB))trace(thr,a,szB,"NA-after "); |
| 4520 | } |
| 4521 | |
| 4522 | void* libhb_get_Thr_opaque ( Thr* thr ) { |
| 4523 | tl_assert(thr); |
| 4524 | return thr->opaque; |
| 4525 | } |
| 4526 | |
| 4527 | void libhb_set_Thr_opaque ( Thr* thr, void* v ) { |
| 4528 | tl_assert(thr); |
| 4529 | thr->opaque = v; |
| 4530 | } |
| 4531 | |
| 4532 | void libhb_copy_shadow_state ( Addr dst, Addr src, SizeT len ) |
| 4533 | { |
| 4534 | zsm_copy_range(dst, src, len); |
| 4535 | } |
| 4536 | |
| 4537 | void libhb_maybe_GC ( void ) |
| 4538 | { |
| 4539 | event_map_maybe_GC(); |
| 4540 | /* If there are still freelist entries available, no need for a |
| 4541 | GC. */ |
| 4542 | if (vts_tab_freelist != VtsID_INVALID) |
| 4543 | return; |
| 4544 | /* So all the table entries are full, and we're having to expand |
| 4545 | the table. But did we hit the threshhold point yet? */ |
| 4546 | if (VG_(sizeXA)( vts_tab ) < vts_next_GC_at) |
| 4547 | return; |
| 4548 | vts_tab__do_GC( False/*don't show stats*/ ); |
| 4549 | } |
| 4550 | |
| 4551 | |
| 4552 | ///////////////////////////////////////////////////////////////// |
| 4553 | ///////////////////////////////////////////////////////////////// |
| 4554 | // // |
| 4555 | // SECTION END main library // |
| 4556 | // // |
| 4557 | ///////////////////////////////////////////////////////////////// |
| 4558 | ///////////////////////////////////////////////////////////////// |
| 4559 | |
| 4560 | /*--------------------------------------------------------------------*/ |
| 4561 | /*--- end libhb_main.c ---*/ |
| 4562 | /*--------------------------------------------------------------------*/ |