libexec/rtld-elf/x86_64/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  */
  27
  28 /*
  29  * Thread locking implementation for the dynamic linker.
  30  *
  31  *   J. M. Mellor-Crummey and M. L. Scott. "Scalable Reader-Writer
  32  *   Synchronization for Shared-Memory Multiprocessors." 3rd ACM Symp. on
  33  *   Principles and Practice of Parallel Programming, April 1991.
  34  *
  35  * In this algorithm the lock is a single word.  Its low-order bit is
  36  * set when a writer holds the lock.  The remaining high-order bits
  37  * contain a count of readers desiring the lock.  The algorithm requires
  38  * atomic "compare_and_store" and "add" operations.
  39  */
  40
  41 #include <setjmp.h>
  42 #include <signal.h>
  43 #include <stdlib.h>
  44 #include <time.h>
  45
  46 #include "debug.h"
  47 #include "rtld.h"
  48
  49 #define WAFLAG          0x1     /* A writer holds the lock */
  50 #define RC_INCR         0x2     /* Adjusts count of readers desiring lock */
  51
  52 typedef struct Struct_Lock {
  53         volatile int lock;
  54         void *base;
  55 } Lock;
  56
  57 static sigset_t fullsigmask, oldsigmask;
  58
  59 static inline int
  60 cmpxchgl(int old, int new, volatile int *m)
  61 {
  62         int result;
  63
  64         __asm __volatile ("lock; cmpxchgl %2, %0"
  65             : "+m"(*m), "=a"(result)
  66             : "r"(new), "1"(old)
  67             : "cc");
  68
  69         return result;
  70 }
  71
  72 static void *
  73 lock_create(void *context)
  74 {
  75     void *base;
  76     char *p;
  77     uintptr_t r;
  78     Lock *l;
  79
  80     /*
  81      * Arrange for the lock to occupy its own cache line.  First, we
  82      * optimistically allocate just a cache line, hoping that malloc
  83      * will give us a well-aligned block of memory.  If that doesn't
  84      * work, we allocate a larger block and take a well-aligned cache
  85      * line from it.
  86      */
  87     base = xmalloc(CACHE_LINE_SIZE);
  88     p = (char *)base;
  89     if ((uintptr_t)p % CACHE_LINE_SIZE != 0) {
  90         free(base);
  91         base = xmalloc(2 * CACHE_LINE_SIZE);
  92         p = (char *)base;
  93         if ((r = (uintptr_t)p % CACHE_LINE_SIZE) != 0)
  94             p += CACHE_LINE_SIZE - r;
  95     }
  96     l = (Lock *)p;
  97     l->base = base;
  98     l->lock = 0;
  99     return l;
 100 }
 101
 102 static void
 103 lock_destroy(void *lock)
 104 {
 105     Lock *l = (Lock *)lock;
 106
 107     free(l->base);
 108 }
 109
 110 /*
 111  * Better reader/writer locks for the 80486 and later CPUs.
 112  */
 113 static void
 114 rlock_acquire(void *lock)
 115 {
 116     Lock *l = (Lock *)lock;
 117
 118     atomic_add_int(&l->lock, RC_INCR);
 119     while (l->lock & WAFLAG)
 120             ;   /* Spin */
 121 }
 122
 123 static void
 124 wlock_acquire(void *lock)
 125 {
 126     Lock *l = (Lock *)lock;
 127     sigset_t tmp_oldsigmask;
 128
 129     for ( ; ; ) {
 130         sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask);
 131         if (cmpxchgl(0, WAFLAG, &l->lock) == 0)
 132             break;
 133         sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL);
 134     }
 135     oldsigmask = tmp_oldsigmask;
 136 }
 137
 138 static void
 139 rlock_release(void *lock)
 140 {
 141     Lock *l = (Lock *)lock;
 142
 143     atomic_add_int(&l->lock, -RC_INCR);
 144 }
 145
 146 static void
 147 wlock_release(void *lock)
 148 {
 149     Lock *l = (Lock *)lock;
 150
 151     atomic_add_int(&l->lock, -WAFLAG);
 152     sigprocmask(SIG_SETMASK, &oldsigmask, NULL);
 153 }
 154
 155 void
 156 lockdflt_init(LockInfo *li)
 157 {
 158     li->context = NULL;
 159     li->context_destroy = NULL;
 160     li->lock_create = lock_create;
 161     li->lock_destroy = lock_destroy;
 162     li->rlock_acquire = rlock_acquire;
 163     li->wlock_acquire = wlock_acquire;
 164     li->rlock_release = rlock_release;
 165     li->wlock_release = wlock_release;
 166     /*
 167      * Construct a mask to block all signals except traps which might
 168      * conceivably be generated within the dynamic linker itself.
 169      */
 170     sigfillset(&fullsigmask);
 171     sigdelset(&fullsigmask, SIGILL);
 172     sigdelset(&fullsigmask, SIGTRAP);
 173     sigdelset(&fullsigmask, SIGABRT);
 174     sigdelset(&fullsigmask, SIGEMT);
 175     sigdelset(&fullsigmask, SIGFPE);
 176     sigdelset(&fullsigmask, SIGBUS);
 177     sigdelset(&fullsigmask, SIGSEGV);
 178     sigdelset(&fullsigmask, SIGSYS);
 179 }