///////////////////////////////////////////////////////////////////////////////
// ErrlManager::errlogMsgHndlr()
///////////////////////////////////////////////////////////////////////////////
void ErrlManager::errlogMsgHndlr ()
{
TRACFCOMP( g_trac_errl, ENTER_MRK "Enter ErrlManager::errlogMsgHndlr" );
while( 1 )
{
msg_t * theMsg = msg_wait( iv_msgQ );
TRACFCOMP( g_trac_errl, INFO_MRK"Got an error log Msg - Type: 0x%08x",
theMsg->type );
//Process message just received
switch( theMsg->type )
{
case ERRLOG_ACCESS_PNOR_TYPE:
{
// PNOR is up and running now.
setupPnorInfo();
//We are done with the msg
msg_free(theMsg);
// go back and wait for a next msg
break;
}
case ERRLOG_ACCESS_TARG_TYPE:
{
// TARGETING is up and running now.
// do we NOT need to send the error?
TARGETING::Target * sys = NULL;
TARGETING::targetService().getTopLevelTarget( sys );
TARGETING::SpFunctions spfn;
if (!(sys &&
sys->tryGetAttr<TARGETING::ATTR_SP_FUNCTIONS>(spfn) &&
spfn.baseServices))
{
iv_isSpBaseServices = false;
// if there are queued errors, clear the Mbox flag
// since they will never be sent, which will delete
// the errors that have been fully processed
ErrlListItr_t it = iv_errlList.begin();
while(it != iv_errlList.end())
{
// Mark MBOX processing complete
_clearFlag(*it, MBOX_FLAG);
_updateErrlListIter(it);
}
}
//We are done with the msg
msg_free(theMsg);
// go back and wait for a next msg
break;
}
case ERRLOG_ACCESS_MBOX_TYPE:
{
// MBOX is up and running now.
// do we need to send the errorlog
TARGETING::Target * sys = NULL;
TARGETING::targetService().getTopLevelTarget( sys );
TARGETING::SpFunctions spfn;
if (sys &&
sys->tryGetAttr<TARGETING::ATTR_SP_FUNCTIONS>(spfn) &&
spfn.mailboxEnabled)
{
iv_isMboxEnabled = true;
}
// if we're supposed to and can now send msgs, do it.
if (iv_isSpBaseServices && iv_isMboxEnabled)
{
// Register messageQ with Mailbox to receive message.
errlHndl_t l_err =
MBOX::msgq_register( MBOX::HB_ERROR_MSGQ,
iv_msgQ );
if( l_err )
{
TRACFCOMP(g_trac_errl, ERR_MRK "Msg queue already registered");
delete( l_err );
l_err = NULL;
//If we got an error then it means the message queue
//is registered with mailbox.
//This should not happen. So assert here.
assert(0);
}
// if errors came in before MBOX was ready,
// the errors would be on this list. send them now.
ErrlListItr_t it = iv_errlList.begin();
while(it != iv_errlList.end())
{
// Check if MBOX processing is needed
if (_isFlagSet(*it, MBOX_FLAG))
{
// send errlog
sendErrLogToFSP(it->first);
// Mark MBOX processing complete
_clearFlag(*it, MBOX_FLAG);
}
_updateErrlListIter(it);
}
}
else
{
// Delete errors that have been completely processed
ErrlListItr_t it = iv_errlList.begin();
while(it != iv_errlList.end())
{
// Mark MBOX processing complete
_clearFlag(*it, MBOX_FLAG);
_updateErrlListIter(it);
}
}
//We are done with the msg
msg_free(theMsg);
// go back and wait for a next msg
break;
}
case ERRLOG_ACCESS_IPMI_TYPE:
{
#ifdef CONFIG_BMC_IPMI
// IPMI is up and running now.
iv_isIpmiEnabled = true;
// if we can now send msgs, do it.
// if errors came in before IPMI was ready,
// the errors would be on this list. send them now.
ErrlListItr_t it = iv_errlList.begin();
while(it != iv_errlList.end())
{
// Check if IPMI processing is needed
if (_isFlagSet(*it, IPMI_FLAG))
{
// send errorlog
sendErrLogToBmc(it->first);
// Mark IPMI processing complete
_clearFlag(*it, IPMI_FLAG);
}
_updateErrlListIter(it);
}
#endif
//We are done with the msg
msg_free(theMsg);
// go back and wait for a next msg
break;
}
case ERRLOG_ACCESS_ERRLDISP_TYPE:
{
#ifdef CONFIG_CONSOLE_OUTPUT_ERRORDISPLAY
// Errldisplay now ready
iv_isErrlDisplayEnabled = true;
CONSOLE::displayf("ERRL",
"Dumping errors reported prior to registration");
// Display errlogs to errldisplay
ErrlListItr_t it = iv_errlList.begin();
while(it != iv_errlList.end())
{
// Check if ERRLDISP processing is needed
if (_isFlagSet(*it, ERRLDISP_FLAG))
{
ERRORLOGDISPLAY::errLogDisplay().msgDisplay
(it->first,
((it->first->reasonCode()) & 0xFF00));
// Mark ERRLDISP processing complete
_clearFlag(*it, ERRLDISP_FLAG);
}
_updateErrlListIter(it);
}
#endif
//We are done with the msg
msg_free(theMsg);
break;
}
case ERRLOG_NEEDS_TO_BE_COMMITTED_TYPE:
{
// Extract error log handle from the message. We need the
// error log handle to pass along
errlHndl_t l_err = (errlHndl_t) theMsg->extra_data;
// Ask the ErrlEntry to assign commit component, commit time
l_err->commit( (compId_t) theMsg->data[0] );
// Pair with all flags set to add to the errlList
ErrlFlagPair_t l_pair(l_err, ALL_FLAGS);
#ifdef CONFIG_CONSOLE_OUTPUT_ERRORDISPLAY
// Display errl to errldisplay
if (iv_isErrlDisplayEnabled)
{
ERRORLOGDISPLAY::errLogDisplay().msgDisplay
(l_err,
( (l_err->reasonCode()) & 0xFF00));
// Mark ERRLDISP processing complete on this error
_clearFlag(l_pair, ERRLDISP_FLAG);
}
#endif
//Save the error log to PNOR
bool l_savedToPnor = saveErrLogToPnor(l_err);
// Check if we actually saved the msg to PNOR
if (l_savedToPnor)
{
// Mark PNOR processing complete on this error
_clearFlag(l_pair, PNOR_FLAG);
}
#ifdef STORE_ERRL_IN_L3
//Write the error log to L3 memory
//useful ONLY for the hb-errl tool
saveErrLogEntry ( l_err );
#endif
//Try to send the error log if someone is there to receive
if (!iv_isSpBaseServices)
{
// Mark MBOX processing complete on this error
_clearFlag(l_pair, MBOX_FLAG);
}
else if (iv_isSpBaseServices && iv_isMboxEnabled)
{
sendErrLogToFSP(l_err);
// Mark MBOX processing complete on this error
_clearFlag(l_pair, MBOX_FLAG);
}
#ifdef CONFIG_BMC_IPMI
if (iv_isIpmiEnabled)
{
// convert to SEL/eSEL and send to BMC over IPMI
sendErrLogToBmc(l_err);
// Mark IPMI processing complete on this error
_clearFlag(l_pair, IPMI_FLAG);
}
#endif
//Ask the ErrlEntry to process any callouts
l_err->processCallout();
//Ask if it is a terminating log
if( l_err->isTerminateLog() )
{
TRACFCOMP( g_trac_errl, INFO_MRK
"Terminating error was committed"
" errlmanager is reqesting a shutdown.");
INITSERVICE::doShutdown(l_err->plid(), true);
TRACDCOMP( g_trac_errl,
INFO_MRK"shutdown in progress" );
}
// If l_errl has not been fully proccessed delete it
// otherwise add to list
if (l_pair.second == 0)
{
delete l_err;
l_err = NULL;
}
else
{
iv_errlList.push_back(l_pair);
}
//We are done with the msg
msg_free(theMsg);
// else go back and wait for a next msg
break;
}
case ERRLOG_COMMITTED_ACK_RESPONSE_TYPE:
{
//Hostboot must keep track and clean up hostboot error
//logs in PNOR after it is committed by FSP.
uint32_t l_tmpPlid = theMsg->data[0]>>32;
TRACFCOMP( g_trac_errl, INFO_MRK"ack: %.8x", l_tmpPlid);
bool didAck = ackErrLogInPnor(l_tmpPlid);
if (!didAck)
{
// couldn't find that errlog in PNOR, look in our
// errlMsgList - maybe it's there waiting
ErrlListItr_t it = std::find_if(iv_errlList.begin(),
iv_errlList.end(),
std::bind1st(ptr_fun(&compareEidToPlid)
,l_tmpPlid));
// Check if such errl was found
if (it != iv_errlList.end())
{
// We found the errlog
// Mark PNOR processing complete
_clearFlag(*it, PNOR_FLAG);
_updateErrlListIter(it);
}
}
msg_free(theMsg);
// We didn't have room before in PNOR to save an
// error log, so try now since we just ACKed one.
ErrlListItr_t it = std::find_if(iv_errlList.begin(),
iv_errlList.end(),
bind2nd(ptr_fun(_isFlagSet),
PNOR_FLAG));
// Check if such errl was found
if (it != iv_errlList.end())
{
bool l_savedToPnor = saveErrLogToPnor(it->first);
// check if we actually saved the msg to PNOR
if (l_savedToPnor)
{
// Mark PNOR processing complete
_clearFlag(*it, PNOR_FLAG);
_updateErrlListIter(it);
}
// else, still couldn't save it (for some reason) so
// it's still on the list.
}
break;
}
case ERRLOG_SHUTDOWN_TYPE:
TRACFCOMP( g_trac_errl, INFO_MRK "Shutdown event received" );
//Start shutdown process for error log
errlogShutdown();
// Respond that we are done shutting down.
msg_respond ( iv_msgQ, theMsg );
TRACFCOMP( g_trac_errl, INFO_MRK "Shutdown event processed" );
break;
default:
// Default Message
TRACFCOMP( g_trac_errl, ERR_MRK "Unexpected message type 0x%08x",
theMsg->type );
msg_free(theMsg);
break;
} // switch
}
//The errlogMsgHndlr should run all the time. It only
//exits when error log message thread is killed.
TRACFCOMP( g_trac_errl, EXIT_MRK "Exit ErrlManager::errlogMsgHndlr" );
return;
}
// ------------------------------------------------------------------
// setupPnorInfo
// ------------------------------------------------------------------
void ErrlManager::setupPnorInfo()
{
TRACFCOMP( g_trac_errl, ENTER_MRK"setupPnorInfo" );
do
{
// Get SPD PNOR section info from PNOR RP
PNOR::SectionInfo_t info;
errlHndl_t err = PNOR::getSectionInfo( PNOR::HB_ERRLOGS, info );
if (err)
{
TRACFCOMP( g_trac_errl, INFO_MRK"setupPnorInfo getSectionInfo failed");
assert(err == NULL);
break;
}
TRACFCOMP( g_trac_errl, INFO_MRK"setupPnorInfo sectionInfo id %d name \"%s\" size %d",
info.id, info.name, info.size );
// Set the globals appropriately
iv_pnorAddr = reinterpret_cast<char *> (info.vaddr);
iv_maxErrlInPnor = info.size / PNOR_ERROR_LENGTH;
TRACFCOMP( g_trac_errl, INFO_MRK"setupPnorInfo iv_pnorAddr %p maxErrlInPnor %d",
iv_pnorAddr, iv_maxErrlInPnor );
// initial value, in case PNOR is empty - start at this end slot
// so that our first save will increment and wrap correctly
iv_pnorOpenSlot = (iv_maxErrlInPnor - 1);
// walk thru memory, finding error logs and determine the highest ID
uint32_t l_maxId = 0;
for (uint32_t i = 0; i < iv_maxErrlInPnor; i++)
{
if (!isSlotEmpty(i))
{
uint32_t l_id = readEidFromFlattened(i);
// If id is not from HB (0x9XXXXXXX) grab plid instead
if ( (l_id & FIRST_BYTE_ERRLOG) != ERRLOG_PLID_BASE )
{
l_id = readPlidFromFlattened(i);
}
if (l_id > l_maxId )
{
l_maxId = l_id;
// set this - start at this 'max' slot so that our first
// save will increment correctly
iv_pnorOpenSlot = i;
}
// also check if it's ACKed or not
if (!isSlotACKed(i))
{
TRACFCOMP( g_trac_errl,
INFO_MRK"setupPnorInfo slot %d eid %.8X was not ACKed.",
i, l_id);
#ifdef CONFIG_BMC_IPMI
// for IPMI systems, unflatten to send down to the BMC
err = new ERRORLOG::ErrlEntry(
ERRORLOG::ERRL_SEV_UNRECOVERABLE, 0,0);
char *l_errlAddr = iv_pnorAddr + (PNOR_ERROR_LENGTH * i);
uint64_t rc = err->unflatten(l_errlAddr, PNOR_ERROR_LENGTH);
if (rc != 0)
{
// unflatten didn't work, nothing we can do
TRACFCOMP( g_trac_errl,
ERR_MRK"setupPnorInfo unflatten failed on slot %d eid %.8X.",
i, l_id);
}
else
{
if (iv_isIpmiEnabled)
{
// convert to SEL/eSEL and send to BMC over IPMI
sendErrLogToBmc(err);
delete err;
}
else
{
TRACFCOMP( g_trac_errl,
INFO_MRK"setupPnorInfo pushing slot %d eid %.8X to iv_errList.",
i, l_id);
// Pair with IPMI flag to add to the errlList
// so that it'll get sent down when IPMI is up
ErrlFlagPair_t l_pair(err, IPMI_FLAG);
iv_errlList.push_back(l_pair);
}
}
#else
// for FSP system, this shouldn't ever happen.
#endif
setACKInFlattened(i);
} // not ACKed
} // not empty
} // for
// bump the current eid to 1 past the max eid found
while (!__sync_bool_compare_and_swap(&iv_currLogId, iv_currLogId,
(iv_currLogId & ERRLOG_PLID_BASE_MASK) +
(l_maxId & ERRLOG_PLID_MASK) + 1));
TRACFCOMP( g_trac_errl, INFO_MRK"setupPnorInfo reseting LogId 0x%X", iv_currLogId);
// if error(s) came in before PNOR was ready,
// the error log(s) would be on this list. save now.
ErrlListItr_t it = iv_errlList.begin();
while(it != iv_errlList.end())
{
// Check if PNOR processing is needed
if (_isFlagSet(*it, PNOR_FLAG))
{
//ACK it if no one is there to receive
bool l_savedToPnor = saveErrLogToPnor(it->first);
// check if we actually saved the msg to PNOR
if (l_savedToPnor)
{
// Mark PNOR processing complete
_clearFlag(*it, PNOR_FLAG);
_updateErrlListIter(it);
}
else
{
// still couldn't save it (PNOR maybe full) so
// it's still on the list.
break; // get out of this while loop.
}
}
else
{
++it;
}
}
} while (0);
TRACFCOMP( g_trac_errl, EXIT_MRK"setupPnorInfo");
} // setupPnorInfo
// ------------------------------------------------------------------
// setupPnorInfo
// ------------------------------------------------------------------
void ErrlManager::setupPnorInfo()
{
TRACFCOMP( g_trac_errl, ENTER_MRK"setupPnorInfo" );
do
{
// Get SPD PNOR section info from PNOR RP
PNOR::SectionInfo_t info;
errlHndl_t err = PNOR::getSectionInfo( PNOR::HB_ERRLOGS, info );
if (err)
{
TRACFCOMP( g_trac_errl, INFO_MRK"setupPnorInfo getSectionInfo failed");
assert(err == NULL);
break;
}
TRACFCOMP( g_trac_errl, INFO_MRK"setupPnorInfo sectionInfo id %d name \"%s\" size %d",
info.id, info.name, info.size );
// Set the globals appropriately
iv_pnorAddr = reinterpret_cast<char *> (info.vaddr);
iv_maxErrlInPnor = info.size / PNOR_ERROR_LENGTH;
TRACFCOMP( g_trac_errl, INFO_MRK"setupPnorInfo iv_pnorAddr %p maxErrlInPnor %d",
iv_pnorAddr, iv_maxErrlInPnor );
// initial value, in case PNOR is empty - start at this end slot
// so that our first save will increment and wrap correctly
iv_pnorOpenSlot = (iv_maxErrlInPnor - 1);
// walk thru memory, finding error logs and determine the highest ID
uint32_t l_maxId = 0;
for (uint32_t i = 0; i < iv_maxErrlInPnor; i++)
{
if (!isSlotEmpty(i))
{
uint32_t l_id = readEidFromFlattened(i);
// If id is not from HB (0x9XXXXXXX) grab plid instead
if ( (l_id & FIRST_BYTE_ERRLOG) != ERRLOG_PLID_BASE )
{
l_id = readPlidFromFlattened(i);
}
if (l_id > l_maxId )
{
l_maxId = l_id;
// set this - start at this 'max' slot so that our first
// save will increment correctly
iv_pnorOpenSlot = i;
}
// also check if it's ACKed or not. and ACK it.
// for FSP system, this shouldn't ever happen.
// for non-FSP systems, this clears out all 'last IPL' logs
if (!isSlotACKed(i))
{
TRACFCOMP( g_trac_errl, INFO_MRK"setupPnorInfo slot %d eid %.8X was not ACKed.",
i, l_id);
setACKInFlattened(i);
} // not ACKed
} // not empty
} // for
// bump the current eid to 1 past the max eid found
while (!__sync_bool_compare_and_swap(&iv_currLogId, iv_currLogId,
(iv_currLogId & ERRLOG_PLID_BASE_MASK) +
(l_maxId & ERRLOG_PLID_MASK) + 1));
TRACFCOMP( g_trac_errl, INFO_MRK"setupPnorInfo reseting LogId 0x%X", iv_currLogId);
// if error(s) came in before PNOR was ready,
// the error log(s) would be on this list. save now.
ErrlListItr_t it = iv_errlList.begin();
while(it != iv_errlList.end())
{
// Check if PNOR processing is needed
if (_isPnorFlagSet(*it))
{
//ACK it if no one is there to receive
bool l_savedToPnor = saveErrLogToPnor(it->first);
// check if we actually saved the msg to PNOR
if (l_savedToPnor)
{
// Mark PNOR processing complete
_clearPnorFlag(*it);
_updateErrlListIter(it);
}
else
{
// still couldn't save it (PNOR maybe full) so
// it's still on the list.
break; // get out of this while loop.
}
}
else
{
++it;
}
}
} while (0);
TRACFCOMP( g_trac_errl, EXIT_MRK"setupPnorInfo");
} // setupPnorInfo
