libexec/rtld-elf/amd64/lockdflt.c

   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  * $FreeBSD: src/libexec/rtld-elf/i386/lockdflt.c,v 1.5.2.4 2002/07/11 23:52:32 jdp Exp $
  26  * $DragonFly: src/libexec/rtld-elf/amd64/lockdflt.c,v 1.1 2006/07/27 00:40:35 corecode Exp $
  27  */
  28
  29 /*
  30  * Thread locking implementation for the dynamic linker.
  31  *
  32  *   J. M. Mellor-Crummey and M. L. Scott. "Scalable Reader-Writer
  33  *   Synchronization for Shared-Memory Multiprocessors." 3rd ACM Symp. on
  34  *   Principles and Practice of Parallel Programming, April 1991.
  35  *
  36  * In this algorithm the lock is a single word.  Its low-order bit is
  37  * set when a writer holds the lock.  The remaining high-order bits
  38  * contain a count of readers desiring the lock.  The algorithm requires
  39  * atomic "compare_and_store" and "add" operations.
  40  */
  41
  42 #include <setjmp.h>
  43 #include <signal.h>
  44 #include <stdlib.h>
  45 #include <time.h>
  46
  47 #include "debug.h"
  48 #include "rtld.h"
  49
  50 #define WAFLAG          0x1     /* A writer holds the lock */
  51 #define RC_INCR         0x2     /* Adjusts count of readers desiring lock */
  52
  53 typedef struct Struct_Lock {
  54         volatile int lock;
  55         void *base;
  56 } Lock;
  57
  58 static sigset_t fullsigmask, oldsigmask;
  59
  60 static inline int
  61 cmpxchgl(int old, int new, volatile int *m)
  62 {
  63         int result;
  64
  65         __asm __volatile ("lock; cmpxchgl %2, %0"
  66             : "+m"(*m), "=a"(result)
  67             : "r"(new), "1"(old)
  68             : "cc");
  69
  70         return result;
  71 }
  72
  73 static void *
  74 lock_create(void *context)
  75 {
  76     void *base;
  77     char *p;
  78     uintptr_t r;
  79     Lock *l;
  80
  81     /*
  82      * Arrange for the lock to occupy its own cache line.  First, we
  83      * optimistically allocate just a cache line, hoping that malloc
  84      * will give us a well-aligned block of memory.  If that doesn't
  85      * work, we allocate a larger block and take a well-aligned cache
  86      * line from it.
  87      */
  88     base = xmalloc(CACHE_LINE_SIZE);
  89     p = (char *)base;
  90     if ((uintptr_t)p % CACHE_LINE_SIZE != 0) {
  91         free(base);
  92         base = xmalloc(2 * CACHE_LINE_SIZE);
  93         p = (char *)base;
  94         if ((r = (uintptr_t)p % CACHE_LINE_SIZE) != 0)
  95             p += CACHE_LINE_SIZE - r;
  96     }
  97     l = (Lock *)p;
  98     l->base = base;
  99     l->lock = 0;
 100     return l;
 101 }
 102
 103 static void
 104 lock_destroy(void *lock)
 105 {
 106     Lock *l = (Lock *)lock;
 107
 108     free(l->base);
 109 }
 110
 111 /*
 112  * Better reader/writer locks for the 80486 and later CPUs.
 113  */
 114 static void
 115 rlock_acquire(void *lock)
 116 {
 117     Lock *l = (Lock *)lock;
 118
 119     atomic_add_int(&l->lock, RC_INCR);
 120     while (l->lock & WAFLAG)
 121             ;   /* Spin */
 122 }
 123
 124 static void
 125 wlock_acquire(void *lock)
 126 {
 127     Lock *l = (Lock *)lock;
 128     sigset_t tmp_oldsigmask;
 129
 130     for ( ; ; ) {
 131         sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask);
 132         if (cmpxchgl(0, WAFLAG, &l->lock) == 0)
 133             break;
 134         sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL);
 135     }
 136     oldsigmask = tmp_oldsigmask;
 137 }
 138
 139 static void
 140 rlock_release(void *lock)
 141 {
 142     Lock *l = (Lock *)lock;
 143
 144     atomic_add_int(&l->lock, -RC_INCR);
 145 }
 146
 147 static void
 148 wlock_release(void *lock)
 149 {
 150     Lock *l = (Lock *)lock;
 151
 152     atomic_add_int(&l->lock, -WAFLAG);
 153     sigprocmask(SIG_SETMASK, &oldsigmask, NULL);
 154 }
 155
 156 void
 157 lockdflt_init(LockInfo *li)
 158 {
 159     li->context = NULL;
 160     li->context_destroy = NULL;
 161     li->lock_create = lock_create;
 162     li->lock_destroy = lock_destroy;
 163     li->rlock_acquire = rlock_acquire;
 164     li->wlock_acquire = wlock_acquire;
 165     li->rlock_release = rlock_release;
 166     li->wlock_release = wlock_release;
 167     /*
 168      * Construct a mask to block all signals except traps which might
 169      * conceivably be generated within the dynamic linker itself.
 170      */
 171     sigfillset(&fullsigmask);
 172     sigdelset(&fullsigmask, SIGILL);
 173     sigdelset(&fullsigmask, SIGTRAP);
 174     sigdelset(&fullsigmask, SIGABRT);
 175     sigdelset(&fullsigmask, SIGEMT);
 176     sigdelset(&fullsigmask, SIGFPE);
 177     sigdelset(&fullsigmask, SIGBUS);
 178     sigdelset(&fullsigmask, SIGSEGV);
 179     sigdelset(&fullsigmask, SIGSYS);
 180 }