Hi Michal & Krzysztof,
This is a replacement to the previous mqueues-4.00 libraries patch
that takes into account that the Opteron and i386 syscall numbers
must be different.
Regards,
Joe
patchfile: mqueues-4.00-lib-a1.patch
diff -ura mqueues-4.00/library/src/mqueue.c mqueues-4.00-new/library/src/mqueue.c
--- mqueues-4.00/library/src/mqueue.c 2003-11-08 09:20:56.000000000 -0500
+++ mqueues-4.00-new/library/src/mqueue.c 2003-11-20 11:03:51.000000000 -0500
@@ -32,43 +32,55 @@
#include "../include/mqueue.h"
#include "config.h"
-#ifdef SIGEVTHREAD_SUPPORT
-#include <pthread.h>
-#endif
+#ifndef __NR_mq_open
+#ifdef __i386__
#define __NR_mq_open 274
+#elif __x86_64__
+#define __NR_mq_open 237
+#else
+#error unsupported architecture
+#endif
+
#define __NR_mq_unlink (__NR_mq_open+1)
#define __NR_mq_timedsend (__NR_mq_open+2)
#define __NR_mq_timedreceive (__NR_mq_open+3)
#define __NR_mq_notify (__NR_mq_open+4)
#define __NR_mq_getattr (__NR_mq_open+5)
#define __NR_mq_setattr (__NR_mq_open+6)
-
-#define ERRNO_OK_SIGNAL 0
-#define ERRNO_OK_THREAD 1
-#define ERRNO_REMOVE_THREAD 2
+#endif
#ifdef SIGEVTHREAD_SUPPORT
+#include <pthread.h>
+
+/*
+ * Force any references in this file to the following externs to
+ * define to NULL if global definitions cannot be found at link
+ * edit time.
+ */
+#pragma weak pthread_create
+#pragma weak pthread_sigmask
+#pragma weak pthread_barrier_init
+#pragma weak pthread_barrier_wait
+#pragma weak pthread_attr_init
+#pragma weak pthread_attr_setdetachstate
+
struct thread_communication_data
{
- struct sigevent notification; /* first field must be copied as it can */
- mqd_t mqdes; /* be used after mq_notify exits */
- int *result; /* and it can use its internal data */
+ struct sigevent notification; /* must be a copy of, not a ptr to */
+ mqd_t mqdes;
+ int *result;
pthread_barrier_t *barrier;
};
-#endif
+#endif /* SIGEVTHREAD_SUPPORT */
-_syscall5(int, mq_timedsend, mqd_t, mqdes, const char*, msg_ptr,
- size_t, msg_len, unsigned int, msg_prio, const struct timespec*, abs_timeout);
-_syscall5(ssize_t, mq_timedreceive, mqd_t, mqdes, char*, msg_ptr,
- size_t, msg_len, unsigned int*, msg_prio, const struct timespec*, abs_timeout);
-_syscall2(int, mq_getattr, mqd_t, mqdes, struct mq_attr*, mqstat);
-_syscall3(int, mq_setattr, mqd_t, mqdes, const struct mq_attr*, mqstat,
- struct mq_attr*, omqstat);
+#define ERRNO_OK_SIGNAL 0
+#define ERRNO_OK_THREAD 1
+#define ERRNO_REMOVE_THREAD 2
/* check if name is a valid path */
-inline int is_valid_path(const char *name)
+static inline int is_valid_path(const char *name)
{
int len;
@@ -83,7 +95,7 @@
goto err_einval;
return 0;
-
+
err_einval:
errno = EINVAL;
return -1;
@@ -92,12 +104,63 @@
return -1;
}
-mqd_t mq_open(const char *name, int oflag, /* mode_t mode,
- struct mq_attr *attr */ ...)
+/*
+ * kernel interfaces. We use glibc's syscall(3) instead of the macros
+ * _syscall1, _syscall2, etc, as the macros generate compilation errors
+ * when mqueue.c is built as a dynamic shared library.
+ */
+static inline mqd_t __mq_open(const char *name,
+ int oflag, mode_t mode, struct mq_attr* attr)
+{
+ return syscall(__NR_mq_open, name, oflag, mode, attr);
+}
+
+static inline int __mq_unlink(const char *name)
+{
+ return syscall(__NR_mq_unlink, name);
+}
+
+static inline int __mq_notify(mqd_t mqdes, const struct sigevent *notification)
+{
+ return syscall(__NR_mq_notify, mqdes, notification);
+}
+
+static inline int __mq_getattr(mqd_t mqdes, struct mq_attr *mqstat)
+{
+ return syscall(__NR_mq_getattr, mqdes, mqstat);
+}
+
+static inline int __mq_setattr( mqd_t mqdes, const struct mq_attr *mqstat,
+ struct mq_attr *omqstat)
+{
+ return syscall(__NR_mq_setattr, mqdes, mqstat, omqstat);
+}
+
+static inline int __mq_timedsend(mqd_t mqdes, const char *msg_ptr,
+ size_t msg_len, unsigned int msg_prio,
+ const struct timespec *abs_timeout)
+{
+ return syscall(__NR_mq_timedsend, mqdes, msg_ptr, msg_len,
+ msg_prio, abs_timeout);
+}
+
+static inline ssize_t __mq_timedreceive(mqd_t mqdes, char *msg_ptr,
+ size_t msg_len, unsigned int *msg_prio,
+ const struct timespec *abs_timeout)
+{
+ return syscall(__NR_mq_timedreceive, mqdes, msg_ptr, msg_len,
+ msg_prio, abs_timeout);
+}
+
+/*
+ * application-visible wrappers around the kernel interfaces
+ */
+
+mqd_t mq_open(const char *name, int oflag, ...)
{
-unsigned long mode;
-struct mq_attr *attr;
-va_list ap;
+ unsigned long mode;
+ struct mq_attr *attr;
+ va_list ap;
va_start(ap, oflag);
mode = va_arg(ap, unsigned long);
@@ -107,10 +170,8 @@
if (is_valid_path(name) < 0)
return (mqd_t)-1;
- static inline _syscall4(mqd_t, mq_open, const char *, name, int, oflag, mode_t, mode, struct mq_attr*, attr);
-
/* omit leading slash */
- return mq_open(name + 1, oflag, mode, attr);
+ return __mq_open(name + 1, oflag, mode, attr);
}
int mq_close(mqd_t mqdes)
@@ -123,51 +184,77 @@
if (is_valid_path(name) < 0)
return -1;
- static inline _syscall1(int, mq_unlink, const char*, name);
-
- return mq_unlink(name+1);
+ return __mq_unlink(name+1);
}
-int mq_send(mqd_t mqdes, const char *msg_ptr, size_t msg_len, unsigned int msg_prio)
+int mq_send(mqd_t mqdes, const char *msg_ptr, size_t msg_len,
+ unsigned int msg_prio)
{
- return mq_timedsend(mqdes, msg_ptr, msg_len, msg_prio, NULL);
+ return __mq_timedsend(mqdes, msg_ptr, msg_len, msg_prio, NULL);
}
-ssize_t mq_receive(mqd_t mqdes, char *msg_ptr, size_t msg_len, unsigned int *msg_prio)
+ssize_t mq_receive(mqd_t mqdes, char *msg_ptr, size_t msg_len,
+ unsigned int *msg_prio)
{
- return mq_timedreceive(mqdes, msg_ptr, msg_len, msg_prio, NULL);
+ return __mq_timedreceive(mqdes, msg_ptr, msg_len, msg_prio, NULL);
+}
+
+int mq_timedsend(mqd_t mqdes, const char *msg_ptr, size_t msg_len,
+ unsigned int msg_prio,
+ const struct timespec *abs_timeout)
+{
+ return __mq_timedsend(mqdes, msg_ptr, msg_len, msg_prio, abs_timeout);
+}
+
+ssize_t mq_timedreceive(mqd_t mqdes, char *msg_ptr, size_t msg_len,
+ unsigned int *msg_prio,
+ const struct timespec *abs_timeout)
+{
+ return __mq_timedreceive(mqdes, msg_ptr, msg_len, msg_prio,
+ abs_timeout);
+}
+
+int mq_getattr(mqd_t mqdes, struct mq_attr *mqstat)
+{
+ return __mq_getattr(mqdes, mqstat);
+}
+
+int mq_setattr( mqd_t mqdes, const struct mq_attr *mqstat,
+ struct mq_attr *omqstat)
+{
+ return __mq_setattr(mqdes, mqstat, omqstat);
}
#ifdef SIGEVTHREAD_SUPPORT
static void *notification_handler(void * raw_arg)
{
- struct thread_communication_data arg = *(struct thread_communication_data *)raw_arg;
+ struct thread_communication_data arg =
+ *(struct thread_communication_data *)raw_arg;
siginfo_t info;
- sigset_t sigset;
+ sigset_t signalset;
arg.notification.sigev_signo = SIGUSR1;
- sigfillset(&sigset);
- pthread_sigmask(SIG_BLOCK, &sigset, NULL);
+ sigfillset(&signalset);
+ pthread_sigmask(SIG_BLOCK, &signalset, NULL);
/* set up kernel notification */
- static inline _syscall2(int, mq_notify, mqd_t, _mqdes, const struct sigevent*, notification);
if (mq_notify(arg.mqdes, &(arg.notification))) {
*(arg.result) = errno;
pthread_barrier_wait(arg.barrier);
- return;
+ return NULL;
}
/* allow parent to continue */
*(arg.result) = 0;
pthread_barrier_wait(arg.barrier);
- sigemptyset(&sigset);
- sigaddset(&sigset, arg.notification.sigev_signo);
+ sigemptyset(&signalset);
+ sigaddset(&signalset, arg.notification.sigev_signo);
/* wait for kernel signal */
do {
- sigwaitinfo(&sigset, &info);
+ sigwaitinfo(&signalset, &info);
/* if si_errno == ERRNO_OK_SIGNAL then this signal was sent
* from kernel to whole process as an effect of SIGEV_SIGNAL
* notification
@@ -180,19 +267,20 @@
if (info.si_errno == ERRNO_OK_THREAD)
(arg.notification.sigev_notify_function)(info._sifields._rt.si_sigval);
+ return NULL;
}
#endif
int mq_notify(mqd_t mqdes, const struct sigevent *notification)
{
- struct sigevent n, *ptr;
+ struct sigevent n;
+ const struct sigevent *ptr;
#ifdef SIGEVTHREAD_SUPPORT
pthread_t handler_thread;
pthread_attr_t thread_attr;
pthread_barrier_t barrier;
struct thread_communication_data thread_data;
#endif
- int result;
ptr = &n;
@@ -200,15 +288,21 @@
n.sigev_notify = SIGEV_NONE;
n.sigev_notify_attributes = NULL;
} else
- ptr = (struct sigevent *) notification;
+ ptr = (const struct sigevent *) notification;
if (ptr->sigev_notify == SIGEV_THREAD) {
#ifdef SIGEVTHREAD_SUPPORT
+ int result;
+ if (&pthread_create == NULL) {
+ errno = ENOSYS;
+ return -1;
+ }
pthread_barrier_init(&barrier, NULL, 2);
if (ptr->sigev_notify_attributes == NULL) {
pthread_attr_init(&thread_attr);
- pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
+ pthread_attr_setdetachstate(&thread_attr,
+ PTHREAD_CREATE_DETACHED);
n.sigev_notify_attributes = &thread_attr;
}
@@ -217,8 +311,9 @@
thread_data.mqdes = mqdes;
thread_data.result = &result;
- if (pthread_create(&handler_thread, ptr->sigev_notify_attributes,
- notification_handler, &thread_data)) {
+ if (pthread_create(&handler_thread,
+ ptr->sigev_notify_attributes,
+ notification_handler, &thread_data)) {
errno = EAGAIN;
return -1;
}
@@ -228,13 +323,11 @@
return -1;
}
#else
- fprintf(stderr, "SIGEV_THREAD not supported!\n");
errno = ENOSYS;
return -1;
#endif
} else {
- static inline _syscall2(int, mq_notify, mqd_t, mqdes, const struct sigevent*, notification);
- return mq_notify(mqdes, ptr);
+ return __mq_notify(mqdes, ptr);
}
return 0;
Joe Korty <[email protected]> writes:
> +#ifndef __NR_mq_open
>
> +#ifdef __i386__
> #define __NR_mq_open 274
> +#elif __x86_64__
> +#define __NR_mq_open 237
> +#else
FYI. I already allocated 237 for something else on x86-64. The patch
is guaranteed to get broken.
-Andi
On Thu, Nov 20, 2003 at 06:15:25PM +0100, Andi Kleen wrote:
> Joe Korty <[email protected]> writes:
> > +#ifndef __NR_mq_open
> >
> > +#ifdef __i386__
> > #define __NR_mq_open 274
> > +#elif __x86_64__
> > +#define __NR_mq_open 237
> > +#else
>
> FYI. I already allocated 237 for something else on x86-64. The patch
> is guaranteed to get broken.
Define an offical set for Opteron and I'll use it.
Joe