The sched_setscheduler and pthread_setscheduler calls were inconsistent,
setting inconsistent priorities for their tasks. This was resulting in
deadlocks in the pthread_spin_lock code on single cpus.
Patch attached to fix
Also as future work these calls need proper configure.ac checking. If
anyone takes on that work, both pthread_setscheduler and
sched_setscheduler should be in the system libraries in order to do any
schedule priority changes (or deadlock occurs).
regards
-steve
Index: exec/coroipcs.c
===================================================================
--- exec/coroipcs.c (revision 1909)
+++ exec/coroipcs.c (working copy)
@@ -259,6 +259,13 @@
struct res_overlay res_overlay;
int send_ok;
+ if (api->sched_priority != 0) {
+ struct sched_param sched_param;
+
+ sched_param.sched_priority = api->sched_priority;
+ res = pthread_setschedparam (conn_info->thread, SCHED_RR, &sched_param);
+ }
+
for (;;) {
sop.sem_num = 0;
sop.sem_op = -1;
Index: exec/coroipcs.h
===================================================================
--- exec/coroipcs.h (revision 1909)
+++ exec/coroipcs.h (working copy)
@@ -45,6 +45,7 @@
struct coroipcs_init_state {
const char *socket_name;
+ int sched_priority;
void *(*malloc) (size_t size);
void (*free) (void *ptr);
void (*log_printf) (
Index: exec/timer.c
===================================================================
--- exec/timer.c (revision 1909)
+++ exec/timer.c (working copy)
@@ -90,6 +90,8 @@
static struct timerlist timers_timerlist;
+static int sched_priority = 0;
+
static void (*timer_serialize_lock_fn) (void);
static void (*timer_serialize_unlock_fn) (void);
@@ -107,11 +109,14 @@
unsigned long long timeout;
#if ! defined(TS_CLASS) && (defined(COROSYNC_BSD) || defined(COROSYNC_LINUX) || defined(COROSYNC_SOLARIS))
- struct sched_param sched_param;
int res;
- sched_param.sched_priority = 2;
- res = pthread_setschedparam (expiry_thread, SCHED_RR, &sched_param);
+ if (sched_priority != 0) {
+ struct sched_param sched_param;
+
+ sched_param.sched_priority = sched_priority;
+ res = pthread_setschedparam (expiry_thread, SCHED_RR, &sched_param);
+ }
#endif
pthread_mutex_unlock (&timer_mutex);
@@ -148,12 +153,14 @@
int corosync_timer_init (
void (*serialize_lock_fn) (void),
- void (*serialize_unlock_fn) (void))
+ void (*serialize_unlock_fn) (void),
+ int sched_priority_in)
{
int res;
timer_serialize_lock_fn = serialize_lock_fn;
timer_serialize_unlock_fn = serialize_unlock_fn;
+ sched_priority = sched_priority_in;
timerlist_init (&timers_timerlist);
Index: exec/main.c
===================================================================
--- exec/main.c (revision 1909)
+++ exec/main.c (working copy)
@@ -93,6 +93,8 @@
#define SERVER_BACKLOG 5
+static int sched_priority = 0;
+
static unsigned int service_count = 32;
#if defined(HAVE_PTHREAD_SPIN_LOCK)
@@ -364,16 +366,18 @@
struct sched_param sched_param;
int res;
- res = sched_get_priority_max (SCHED_RR);
- if (res != -1) {
- sched_param.sched_priority = 1;//res;
+ sched_priority = sched_get_priority_max (SCHED_RR);
+ if (sched_priority != -1) {
+ sched_param.sched_priority = sched_priority;
res = sched_setscheduler (0, SCHED_RR, &sched_param);
if (res == -1) {
log_printf (LOG_LEVEL_WARNING, "Could not set SCHED_RR at priority %d: %s\n",
sched_param.sched_priority, strerror (errno));
}
- } else
+ } else {
log_printf (LOG_LEVEL_WARNING, "Could not get maximum scheduler priority: %s\n", strerror (errno));
+ sched_priority = 0;
+ }
#else
log_printf(LOG_LEVEL_WARNING, "Scheduler priority left to default value (no OS support)\n");
#endif
@@ -727,7 +731,8 @@
corosync_timer_init (
serialize_lock,
- serialize_unlock);
+ serialize_unlock,
+ sched_priority);
log_printf (LOG_LEVEL_NOTICE, "Corosync Executive Service: started and ready to provide service.\n");
@@ -903,6 +908,8 @@
ipc_subsys_id = _logsys_subsys_create ("IPC", LOG_INFO);
+ ipc_init_state.sched_priority = sched_priority;
+
coroipcs_ipc_init (&ipc_init_state);
/*
Index: exec/timer.h
===================================================================
--- exec/timer.h (revision 1909)
+++ exec/timer.h (working copy)
@@ -39,7 +39,8 @@
extern void corosync_timer_init (
void (*serialize_lock) (void),
- void (*serialize_unlock) (void));
+ void (*serialize_unlock) (void),
+ int sched_priority);
extern int corosync_timer_add_duration (
unsigned long long nanoseconds_in_future,
Index: lib/coroipcc.c
===================================================================
--- lib/coroipcc.c (revision 1909)
+++ lib/coroipcc.c (working copy)
@@ -100,6 +100,10 @@
}
#endif
+#ifndef MSG_NOSIGNAL
+#define MSG_NOSIGNAL 0
+#endif
+
static int
coroipcc_send (
int s,
_______________________________________________
Openais mailing list
[email protected]
https://lists.linux-foundation.org/mailman/listinfo/openais
