freebsd-dev/libexec/rtld-elf/amd64/lockdflt.c
John Polstra e6f0183bff Remove the nanosleep calls from the spin loops in the locking code.
They provided little benefit (if any) and they caused some problems
in OpenOffice, at least in post-KSE -current and perhaps in other
environments too.  The nanosleep calls prevented the profiling timer
from advancing during the spinloops, thereby preventing the thread
scheduler from ever pre-empting the spinning thread.  Alexander
Kabaev diagnosed this problem, Martin Blapp helped with testing,
and Matt Dillon provided some helpful suggestions.

This is a short-term fix for a larger problem.  The use of spinlocking
isn't guaranteed to work in all cases.  For example, if the spinning
thread has higher priority than all other threads, it may never be
pre-empted, and the thread holding the lock may never progress far
enough to release the lock.  On the other hand, spinlocking is the
only locking that can work with an arbitrary unknown threads package.

I have some ideas for a much better fix in the longer term.  It
would eliminate all locking inside the dynamic linker by making it
safe for symbol lookups and lazy binding to proceed in parallel
with a call to dlopen or dlclose.  This means that the only mutual
exclusion needed would be to prevent multiple simultaneous calls
to dlopen and/or dlclose.  That mutual exclusion could be put into
the native pthreads library.  Applications using foreign threads
packages would have to make their own arrangements to ensure that
they did not have multiple threads in dlopen and/or dlclose -- a
reasonable requirement in my opinion.

MFC after:	3 days
2002-07-06 20:25:56 +00:00

277 lines
7.0 KiB
C

/*-
* Copyright 1999, 2000 John D. Polstra.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* $FreeBSD$
*/
/*
* Thread locking implementation for the dynamic linker.
*
* On 80486 and later CPUs we use the "simple, non-scalable
* reader-preference lock" from:
*
* J. M. Mellor-Crummey and M. L. Scott. "Scalable Reader-Writer
* Synchronization for Shared-Memory Multiprocessors." 3rd ACM Symp. on
* Principles and Practice of Parallel Programming, April 1991.
*
* In this algorithm the lock is a single word. Its low-order bit is
* set when a writer holds the lock. The remaining high-order bits
* contain a count of readers desiring the lock. The algorithm requires
* atomic "compare_and_store" and "add" operations.
*
* The "compare_and_store" operation requires the "cmpxchg" instruction
* on the x86. Unfortunately, the 80386 CPU does not support that
* instruction -- only the 80486 and later models support it. So on the
* 80386 we must use simple test-and-set exclusive locks instead. We
* determine which kind of lock to use by trying to execute a "cmpxchg"
* instruction and catching the SIGILL which results on the 80386.
*/
#include <setjmp.h>
#include <signal.h>
#include <stdlib.h>
#include <time.h>
#include "debug.h"
#include "rtld.h"
#define CACHE_LINE_SIZE 32
#define WAFLAG 0x1 /* A writer holds the lock */
#define RC_INCR 0x2 /* Adjusts count of readers desiring lock */
typedef struct Struct_Lock {
volatile int lock;
void *base;
} Lock;
static sigset_t fullsigmask, oldsigmask;
static inline int
cmpxchgl(int old, int new, volatile int *m)
{
int result;
__asm __volatile ("lock; cmpxchgl %2, %0"
: "+m"(*m), "=a"(result)
: "r"(new), "1"(old)
: "cc");
return result;
}
static inline int
xchgl(int v, volatile int *m)
{
int result;
__asm __volatile ("xchgl %0, %1"
: "=r"(result), "+m"(*m)
: "0"(v));
return result;
}
static void *
lock_create(void *context)
{
void *base;
char *p;
uintptr_t r;
Lock *l;
/*
* Arrange for the lock to occupy its own cache line. First, we
* optimistically allocate just a cache line, hoping that malloc
* will give us a well-aligned block of memory. If that doesn't
* work, we allocate a larger block and take a well-aligned cache
* line from it.
*/
base = xmalloc(CACHE_LINE_SIZE);
p = (char *)base;
if ((uintptr_t)p % CACHE_LINE_SIZE != 0) {
free(base);
base = xmalloc(2 * CACHE_LINE_SIZE);
p = (char *)base;
if ((r = (uintptr_t)p % CACHE_LINE_SIZE) != 0)
p += CACHE_LINE_SIZE - r;
}
l = (Lock *)p;
l->base = base;
l->lock = 0;
return l;
}
static void
lock_destroy(void *lock)
{
Lock *l = (Lock *)lock;
free(l->base);
}
/*
* Crude exclusive locks for the 80386, which does not support the
* cmpxchg instruction.
*/
static void
lock80386_acquire(void *lock)
{
Lock *l = (Lock *)lock;
sigset_t tmp_oldsigmask;
for ( ; ; ) {
sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask);
if (xchgl(1, &l->lock) == 0)
break;
sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL);
while (l->lock != 0)
; /* Spin */
}
oldsigmask = tmp_oldsigmask;
}
static void
lock80386_release(void *lock)
{
Lock *l = (Lock *)lock;
l->lock = 0;
sigprocmask(SIG_SETMASK, &oldsigmask, NULL);
}
/*
* Better reader/writer locks for the 80486 and later CPUs.
*/
static void
rlock_acquire(void *lock)
{
Lock *l = (Lock *)lock;
atomic_add_int(&l->lock, RC_INCR);
while (l->lock & WAFLAG)
; /* Spin */
}
static void
wlock_acquire(void *lock)
{
Lock *l = (Lock *)lock;
sigset_t tmp_oldsigmask;
for ( ; ; ) {
sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask);
if (cmpxchgl(0, WAFLAG, &l->lock) == 0)
break;
sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL);
}
oldsigmask = tmp_oldsigmask;
}
static void
rlock_release(void *lock)
{
Lock *l = (Lock *)lock;
atomic_add_int(&l->lock, -RC_INCR);
}
static void
wlock_release(void *lock)
{
Lock *l = (Lock *)lock;
atomic_add_int(&l->lock, -WAFLAG);
sigprocmask(SIG_SETMASK, &oldsigmask, NULL);
}
/*
* Code to determine at runtime whether the CPU supports the cmpxchg
* instruction. This instruction allows us to use locks that are more
* efficient, but it didn't exist on the 80386.
*/
static jmp_buf sigill_env;
static void
sigill(int sig)
{
longjmp(sigill_env, 1);
}
static int
cpu_supports_cmpxchg(void)
{
struct sigaction act, oact;
int result;
volatile int lock;
memset(&act, 0, sizeof act);
act.sa_handler = sigill;
sigemptyset(&act.sa_mask);
act.sa_flags = 0;
sigaction(SIGILL, &act, &oact);
if (setjmp(sigill_env) == 0) {
lock = 0;
cmpxchgl(0, 1, &lock);
result = 1;
} else
result = 0;
sigaction(SIGILL, &oact, NULL);
return result;
}
void
lockdflt_init(LockInfo *li)
{
li->context = NULL;
li->context_destroy = NULL;
li->lock_create = lock_create;
li->lock_destroy = lock_destroy;
if (cpu_supports_cmpxchg()) {
/* Use fast locks that require an 80486 or later. */
li->rlock_acquire = rlock_acquire;
li->wlock_acquire = wlock_acquire;
li->rlock_release = rlock_release;
li->wlock_release = wlock_release;
} else {
/* It's a cruddy old 80386. */
li->rlock_acquire = li->wlock_acquire = lock80386_acquire;
li->rlock_release = li->wlock_release = lock80386_release;
}
/*
* Construct a mask to block all signals except traps which might
* conceivably be generated within the dynamic linker itself.
*/
sigfillset(&fullsigmask);
sigdelset(&fullsigmask, SIGILL);
sigdelset(&fullsigmask, SIGTRAP);
sigdelset(&fullsigmask, SIGABRT);
sigdelset(&fullsigmask, SIGEMT);
sigdelset(&fullsigmask, SIGFPE);
sigdelset(&fullsigmask, SIGBUS);
sigdelset(&fullsigmask, SIGSEGV);
sigdelset(&fullsigmask, SIGSYS);
}