/**********************************************************************
*
* FUNCTION
* cs_psema
*
* DESCRIPTION
* Decrement a semaphore (wait or take operation)
*
* INPUTS
* handle - A pointer to a semaphore object filled in by the
* cs_sema_create() call.
*
* OUTPUTS
* Status_t - On success VSTATUS_OK is returned, otherwise
* the cause of the error.
*
* RESPONSIBLE ENGINEER:
* CS Developers.
*
* HISTORY
*
* NAME DATE REMARKS
* MGR 03/14/02 Initial creation of function.
* MGR 04/18/02 PR1728. vs_psema should return VSTATUS_AGAIN
* error code, if thread has been killed.
*
**********************************************************************/
Status_t
cs_psema_wait (Sema_t *handle, int timeout)
{
Status_t rc=VSTATUS_OK;
IB_ENTER (function, (unint)handle, 0U, 0U, 0U);
/* Validate handle */
if (handle == (Sema_t *) 0)
{
IB_LOG_ERROR0 ("NULL handle pointer");
IB_EXIT (function, VSTATUS_ILLPARM);
return VSTATUS_ILLPARM;
}
if (handle->magic != SEMAPHORE_MAGIC)
{
IB_LOG_ERRORX ("Invalid semaphore specified:", handle->magic);
IB_EXIT (function, VSTATUS_NXIO);
return VSTATUS_NXIO;
}
if (timeout == CS_SEMA_WAIT_FOREVER) {
return(cs_psema(handle));
} else {
#ifdef __VXWORKS__
if (ERROR == semTake(handle->semId, timeout*sysClkRateGet()))
{
IB_LOG_WARN0("Timed out trying to take semaphore");
rc = VSTATUS_NXIO;
} else {
handle->count--;
}
#elif defined (__LINUX__)
// LINUX USER SEMA
int i=0, rc=-1;
do {
rc = sem_trywait((sem_t *)&handle->osdSema);
++i;
(void)vs_thread_sleep(VTIMER_1S);
} while (i<timeout && rc);
if (rc != 0) {
IB_LOG_WARN0("Timed out trying to take semaphore");
rc = VSTATUS_NXIO;
}
#endif
}
IB_EXIT (function, rc);
return rc;
}
// This requires the caller to have invoked pm_initialize_config prior to this
// call.
// It completes PM initialization and then starts the actual PM.
void
unified_sm_pm(uint32_t argc, uint8_t ** argv)
{
mai_set_default_pkey(STL_DEFAULT_FM_PKEY);
if (pm_config.debug_rmpp) if3RmppDebugOn();
IB_LOG_INFO("Device: ", pm_config.hca);
IB_LOG_INFO("Port: ", pm_config.port);
pm_compute_pool_size(); // compute sizes for PM resources
if (!pm_config.start) {
IB_LOG_WARN0("PM not configured to start");
} else {
(void)sm_wait_ready(VIEO_PM_MOD_ID);
(void)pm_main();
}
}
Status_t
sm_check_Master() {
Status_t status;
STL_SM_INFO theirSmInfo;
STL_PORT_INFO portInfo;
uint8_t path[64];
static uint32_t fsmCheckMasterFailed=0; // count of fails to check master
static uint32_t fsmMultMaxFail = 1; // multiplier for sm_config.master_ping_max_fail
IB_ENTER(__func__, 0, 0, 0, 0);
(void)memset((void *)path, 0, 64);
if ((status = SM_Get_PortInfo(fd_sminfo, 1<<24, path, &portInfo)) != VSTATUS_OK) {
IB_LOG_ERRORRC("failed to get master SM Lid from my PortInfo, rc:", status);
// having a local problem
// reset count, must be healthy before we can consider becoming master
fsmCheckMasterFailed = 0;
goto stay_standby;
}
if (portInfo.LID == 0 || portInfo.MasterSMLID == 0
|| portInfo.PortStates.s.PortState != IB_PORT_ACTIVE
) {
if (smDebugPerf) {
if (portInfo.PortStates.s.PortState != IB_PORT_ACTIVE) {
IB_LOG_INFINI_INFO("our portInfo indicates state not active; portState=", (int)portInfo.PortStates.s.PortState);
} else
if (portInfo.MasterSMLID == 0) {
IB_LOG_INFINI_INFOX("our portInfo smLid is not set yet; Lid=", portInfo.LID);
}
}
// stay in standby until link comes up
if (portInfo.PortStates.s.PortState > IB_PORT_DOWN) {
IB_LOG_WARN0("Switching to DISCOVERY state; Local port uninitialized");
goto discovering;
}
goto stay_standby;
}
// make sure we aren't trying to talk to ourself during the handover window
if (portInfo.LID == portInfo.MasterSMLID) {
// we're talking to ourself. if a master SM doesn't come along and
// reprogram our SM LID in the timeout period, attempt a discovery.
// Since for large fabrics, the master can take some time to program
// our LID, use a higher upper limit for failure count to give the
// master enough time to program our SM LID.
fsmMultMaxFail = 2;
if (++fsmCheckMasterFailed >= fsmMultMaxFail * sm_config.master_ping_max_fail) {
IB_LOG_WARN0("Switching to DISCOVERY state; Timed out waiting for SM LID to get reprogrammed");
goto discovering;
}
if (smDebugPerf) {
IB_LOG_INFINI_INFOX("our portInfo smLid is not set yet; smLid=", portInfo.MasterSMLID);
}
goto stay_standby; // not yet at threshold
}
sm_topop->slid = portInfo.LID;
sm_topop->dlid = portInfo.MasterSMLID;
if ((status = SM_Get_SMInfo(fd_sminfo, 0, NULL, &theirSmInfo)) != VSTATUS_OK) {
if (++fsmCheckMasterFailed >= fsmMultMaxFail * sm_config.master_ping_max_fail) {
IB_LOG_WARNX("Switching to DISCOVERY state; Failed to get SmInfo from master SM at LID:", portInfo.MasterSMLID);
goto discovering;
} else {
IB_LOG_INFINI_INFO_FMT(__func__,
"failed to get SmInfo from master SM at LID[0x%X], retry count=%d",
portInfo.MasterSMLID, fsmCheckMasterFailed);
goto stay_standby; // not yet at threshold
}
}
sm_saw_another_sm = TRUE;
/*
* PR 105313 - restart results in 2 standby SMs
* Must check the state we get back to make sure master is still master
*/
if (theirSmInfo.u.s.SMStateCurrent != SM_STATE_MASTER) {
IB_LOG_WARN_FMT(__func__,
"SmInfo from SM at SMLID[0x%X] indicates SM is no longer master, switching to DISCOVERY state",
portInfo.MasterSMLID);
goto discovering;
}
// all OK, save data about master, reset fail count and threshold
sm_topop->sm_count = theirSmInfo.ActCount;
sm_topop->sm_key = theirSmInfo.SM_Key;
fsmCheckMasterFailed = 0;
fsmMultMaxFail = 1;
if (smDebugPerf) {
IB_LOG_INFINI_INFO_FMT(__func__,
"Master SM["FMT_U64"] at LID=0x%x has priority[%d] and smKey ["FMT_U64"]",
theirSmInfo.PortGUID, portInfo.MasterSMLID, theirSmInfo.u.s.Priority, theirSmInfo.SM_Key);
}
stay_standby:
status = VSTATUS_OK;
done:
IB_EXIT(__func__, status);
return(status);
discovering:
// reset count and threshold in case stay standby
fsmCheckMasterFailed = 0;
fsmMultMaxFail = 1;
status = VSTATUS_BAD;
goto done;
}
