1 /*- 2 * Copyright 1999, 2000 John D. Polstra. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 15 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 16 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 17 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 18 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 19 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 20 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 21 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 22 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 23 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 24 * 25 * from: FreeBSD: src/libexec/rtld-elf/sparc64/lockdflt.c,v 1.3 2002/10/09 26 * $FreeBSD$ 27 */ 28 29 /* 30 * Thread locking implementation for the dynamic linker. 31 * 32 * We use the "simple, non-scalable reader-preference lock" from: 33 * 34 * J. M. Mellor-Crummey and M. L. Scott. "Scalable Reader-Writer 35 * Synchronization for Shared-Memory Multiprocessors." 3rd ACM Symp. on 36 * Principles and Practice of Parallel Programming, April 1991. 37 * 38 * In this algorithm the lock is a single word. Its low-order bit is 39 * set when a writer holds the lock. The remaining high-order bits 40 * contain a count of readers desiring the lock. The algorithm requires 41 * atomic "compare_and_store" and "add" operations, which we implement 42 * using assembly language sequences in "rtld_start.S". 43 */ 44 45 #include <signal.h> 46 #include <stdlib.h> 47 #include <time.h> 48 49 #include "debug.h" 50 #include "rtld.h" 51 #include "rtld_machdep.h" 52 53 #define WAFLAG 0x1 /* A writer holds the lock */ 54 #define RC_INCR 0x2 /* Adjusts count of readers desiring lock */ 55 56 typedef struct Struct_Lock { 57 volatile int lock; 58 void *base; 59 } Lock; 60 61 static sigset_t fullsigmask, oldsigmask; 62 static int thread_flag; 63 64 static void * 65 def_lock_create() 66 { 67 void *base; 68 char *p; 69 uintptr_t r; 70 Lock *l; 71 72 /* 73 * Arrange for the lock to occupy its own cache line. First, we 74 * optimistically allocate just a cache line, hoping that malloc 75 * will give us a well-aligned block of memory. If that doesn't 76 * work, we allocate a larger block and take a well-aligned cache 77 * line from it. 78 */ 79 base = xmalloc(CACHE_LINE_SIZE); 80 p = (char *)base; 81 if ((uintptr_t)p % CACHE_LINE_SIZE != 0) { 82 free(base); 83 base = xmalloc(2 * CACHE_LINE_SIZE); 84 p = (char *)base; 85 if ((r = (uintptr_t)p % CACHE_LINE_SIZE) != 0) 86 p += CACHE_LINE_SIZE - r; 87 } 88 l = (Lock *)p; 89 l->base = base; 90 l->lock = 0; 91 return l; 92 } 93 94 static void 95 def_lock_destroy(void *lock) 96 { 97 Lock *l = (Lock *)lock; 98 99 free(l->base); 100 } 101 102 static void 103 def_rlock_acquire(void *lock) 104 { 105 Lock *l = (Lock *)lock; 106 107 atomic_add_acq_int(&l->lock, RC_INCR); 108 while (l->lock & WAFLAG) 109 ; /* Spin */ 110 } 111 112 static void 113 def_wlock_acquire(void *lock) 114 { 115 Lock *l = (Lock *)lock; 116 sigset_t tmp_oldsigmask; 117 118 for ( ; ; ) { 119 sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask); 120 if (atomic_cmpset_acq_int(&l->lock, 0, WAFLAG)) 121 break; 122 sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL); 123 } 124 oldsigmask = tmp_oldsigmask; 125 } 126 127 static void 128 def_lock_release(void *lock) 129 { 130 Lock *l = (Lock *)lock; 131 132 if ((l->lock & WAFLAG) == 0) 133 atomic_add_rel_int(&l->lock, -RC_INCR); 134 else { 135 atomic_add_rel_int(&l->lock, -WAFLAG); 136 sigprocmask(SIG_SETMASK, &oldsigmask, NULL); 137 } 138 } 139 140 #if __i386__ 141 /* 142 * Import a crude exclusive lock implementation for i386 processors. 143 * This file will be removed once i386 support is deprecated in favor 144 * of i486+. 145 */ 146 #include "i386/lockdflt.c" 147 148 #endif 149 150 static int 151 def_thread_set_flag(int mask) 152 { 153 int old_val = thread_flag; 154 thread_flag |= mask; 155 return (old_val); 156 } 157 158 static int 159 def_thread_clr_flag(int mask) 160 { 161 int old_val = thread_flag; 162 thread_flag &= ~mask; 163 return (old_val); 164 } 165 166 /* 167 * Public interface exposed to the rest of the dynamic linker. 168 */ 169 static struct RtldLockInfo lockinfo; 170 static struct RtldLockInfo deflockinfo; 171 172 static __inline int 173 thread_mask_set(int mask) 174 { 175 return lockinfo.thread_set_flag(mask); 176 } 177 178 static __inline void 179 thread_mask_clear(int mask) 180 { 181 lockinfo.thread_clr_flag(mask); 182 } 183 184 #define RTLD_LOCK_CNT 2 185 struct rtld_lock { 186 void *handle; 187 int mask; 188 } rtld_locks[RTLD_LOCK_CNT]; 189 190 rtld_lock_t rtld_bind_lock = &rtld_locks[0]; 191 rtld_lock_t rtld_libc_lock = &rtld_locks[1]; 192 193 int 194 rlock_acquire(rtld_lock_t lock) 195 { 196 if (thread_mask_set(lock->mask)) { 197 dbg("rlock_acquire: recursed"); 198 return (0); 199 } 200 lockinfo.rlock_acquire(lock->handle); 201 return (1); 202 } 203 204 int 205 wlock_acquire(rtld_lock_t lock) 206 { 207 if (thread_mask_set(lock->mask)) { 208 dbg("wlock_acquire: recursed"); 209 return (0); 210 } 211 lockinfo.wlock_acquire(lock->handle); 212 return (1); 213 } 214 215 void 216 rlock_release(rtld_lock_t lock, int locked) 217 { 218 if (locked == 0) 219 return; 220 thread_mask_clear(lock->mask); 221 lockinfo.lock_release(lock->handle); 222 } 223 224 void 225 wlock_release(rtld_lock_t lock, int locked) 226 { 227 if (locked == 0) 228 return; 229 thread_mask_clear(lock->mask); 230 lockinfo.lock_release(lock->handle); 231 } 232 233 void 234 lockdflt_init() 235 { 236 int i; 237 238 deflockinfo.rtli_version = RTLI_VERSION; 239 deflockinfo.lock_create = def_lock_create; 240 deflockinfo.lock_destroy = def_lock_destroy; 241 deflockinfo.rlock_acquire = def_rlock_acquire; 242 deflockinfo.wlock_acquire = def_wlock_acquire; 243 deflockinfo.lock_release = def_lock_release; 244 deflockinfo.thread_set_flag = def_thread_set_flag; 245 deflockinfo.thread_clr_flag = def_thread_clr_flag; 246 deflockinfo.at_fork = NULL; 247 248 for (i = 0; i < RTLD_LOCK_CNT; i++) { 249 rtld_locks[i].mask = (1 << i); 250 rtld_locks[i].handle = NULL; 251 } 252 253 #if __i386__ 254 if (!cpu_supports_cmpxchg()) { 255 /* It's a cruddy old 80386. */ 256 deflockinfo.rlock_acquire = lock80386_acquire; 257 deflockinfo.wlock_acquire = lock80386_acquire; 258 deflockinfo.lock_release = lock80386_release; 259 } 260 #endif 261 262 memcpy(&lockinfo, &deflockinfo, sizeof(lockinfo)); 263 _rtld_thread_init(NULL); 264 /* 265 * Construct a mask to block all signals except traps which might 266 * conceivably be generated within the dynamic linker itself. 267 */ 268 sigfillset(&fullsigmask); 269 sigdelset(&fullsigmask, SIGILL); 270 sigdelset(&fullsigmask, SIGTRAP); 271 sigdelset(&fullsigmask, SIGABRT); 272 sigdelset(&fullsigmask, SIGEMT); 273 sigdelset(&fullsigmask, SIGFPE); 274 sigdelset(&fullsigmask, SIGBUS); 275 sigdelset(&fullsigmask, SIGSEGV); 276 sigdelset(&fullsigmask, SIGSYS); 277 } 278 279 /* 280 * Callback function to allow threads implementation to 281 * register their own locking primitives if the default 282 * one is not suitable. 283 * The current context should be the only context 284 * executing at the invocation time. 285 */ 286 void 287 _rtld_thread_init(struct RtldLockInfo *pli) 288 { 289 int flags, i; 290 void *locks[RTLD_LOCK_CNT]; 291 292 /* disable all locking while this function is running */ 293 flags = thread_mask_set(~0); 294 295 if (pli == NULL) 296 pli = &deflockinfo; 297 298 299 for (i = 0; i < RTLD_LOCK_CNT; i++) 300 if ((locks[i] = pli->lock_create()) == NULL) 301 break; 302 303 if (i < RTLD_LOCK_CNT) { 304 while (--i >= 0) 305 pli->lock_destroy(locks[i]); 306 abort(); 307 } 308 309 for (i = 0; i < RTLD_LOCK_CNT; i++) { 310 if (rtld_locks[i].handle == NULL) 311 continue; 312 if (flags & rtld_locks[i].mask) 313 lockinfo.lock_release(rtld_locks[i].handle); 314 lockinfo.lock_destroy(rtld_locks[i].handle); 315 } 316 317 for (i = 0; i < RTLD_LOCK_CNT; i++) { 318 rtld_locks[i].handle = locks[i]; 319 if (flags & rtld_locks[i].mask) 320 pli->wlock_acquire(rtld_locks[i].handle); 321 } 322 323 lockinfo.lock_create = pli->lock_create; 324 lockinfo.lock_destroy = pli->lock_destroy; 325 lockinfo.rlock_acquire = pli->rlock_acquire; 326 lockinfo.wlock_acquire = pli->wlock_acquire; 327 lockinfo.lock_release = pli->lock_release; 328 lockinfo.thread_set_flag = pli->thread_set_flag; 329 lockinfo.thread_clr_flag = pli->thread_clr_flag; 330 lockinfo.at_fork = pli->at_fork; 331 332 /* restore thread locking state, this time with new locks */ 333 thread_mask_clear(~0); 334 thread_mask_set(flags); 335 dbg("_rtld_thread_init: done"); 336 } 337