Increment reboot count when gard records are written

Change-Id: Iaf9a4d4be19877e564ca1871564467441b472253
CQ:SW347568
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/23186
Reviewed-by: A. P. Williams III <iawillia@us.ibm.com>
Tested-by: Jenkins Server
Reviewed-by: William G. Hoffa <wghoffa@us.ibm.com>
Reviewed-by: Daniel M. Crowell <dcrowell@us.ibm.com>
Tested-by: Jenkins OP Build CI
Tested-by: Jenkins OP HW
Tested-by: FSP CI Jenkins
Reviewed-by: Matthew A. Ploetz <maploetz@us.ibm.com>
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/24340

6 files changed, 161 insertions, 4 deletions

diff --git a/src/include/usr/errl/errlmanager.H b/src/include/usr/errl/errlmanager.H
index 7112a85eb..1fbc7aa9d 100644
--- a/src/include/usr/errl/errlmanager.H
+++ b/src/include/usr/errl/errlmanager.H
@@ -176,6 +176,18 @@ public:
      static void errlAckErrorlog(uint32_t i_eid);
 
      /**
+      * @brief Calls flush error logs
+      *
+      * This is called by doIstep after every istep to ensure that the error
+      * logs have been cleared out before checking if a reconfigure loop
+      * is needed
+      *
+      * It is a static function because a module cannot call an interface on a
+      * singleton in another module
+      */
+      static void callFlushErrorLogs();
+
+     /**
       * @brief  Returns the HWAS ProcessCallout function pointer
       *
       * This is called by ErrlEntry::commit to get the HWAS ProcessCallout
@@ -242,6 +254,7 @@ private:
         ERRLOG_ACCESS_TARG_TYPE            = 0x00000036 | MBOX::FIRST_SECURE_MSG,
         ERRLOG_ACCESS_ERRLDISP_TYPE        = 0x00000037 | MBOX::FIRST_SECURE_MSG,
         ERRLOG_ACCESS_IPMI_TYPE            = 0x00000038 | MBOX::FIRST_SECURE_MSG,
+        ERRLOG_FLUSH_TYPE                  = 0x00000039 | MBOX::FIRST_SECURE_MSG,
     };
 
     /**
@@ -281,6 +294,15 @@ private:
      void sendResourcesMsg(errlManagerNeeds i_needs);
 
      /**
+      * @brief Flushes out the error log queue before calling to process
+      * deferred deconfigs
+      *
+      * called by static callFlushErrorLogs function.
+      */
+
+     void flushErrorLogs();
+
+     /**
       * @brief Sends msg to errlmanager telling which errlog to ack
       *
       * called by static errlAckErrorlog function.
diff --git a/src/include/usr/initservice/istepdispatcherif.H b/src/include/usr/initservice/istepdispatcherif.H
index bbec9480b..69cebf0a8 100644
--- a/src/include/usr/initservice/istepdispatcherif.H
+++ b/src/include/usr/initservice/istepdispatcherif.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2012,2015                        */
+/* Contributors Listed Below - COPYRIGHT 2012,2016                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -79,6 +79,18 @@ bool isShutdownRequested ( void );
  */
 void setAcceptIstepMessages ( bool i_accept );
 
+/**
+ * @brief This function will set a member class variable to state new gard
+ * records have been committed on this boot attempt.
+ *
+ * Gard records being created are an indication of forward progress.  Therefore
+ * if a gard record has been created we can issue a reboot that does not count
+ * towards the reboot count limit.
+ *
+ * @return None.
+ */
+void setNewGardRecord ( void );
+
 }
 
 #endif
diff --git a/src/usr/errl/errlmanager.C b/src/usr/errl/errlmanager.C
index ac0174713..d693d7a82 100644
--- a/src/usr/errl/errlmanager.C
+++ b/src/usr/errl/errlmanager.C
@@ -51,6 +51,7 @@
 #include <console/consoleif.H>
 #include <config.h>
 #include <functional>
+#include <hwas/common/deconfigGard.H>
 
 namespace ERRORLOG
 {
@@ -583,6 +584,13 @@ void ErrlManager::errlogMsgHndlr ()
                     }
                     break;
                 }
+            case ERRLOG_FLUSH_TYPE:
+                TRACFCOMP( g_trac_errl, INFO_MRK "Flush message received" );
+
+                // Since the errorlog is FIFO, all we need to do is respond
+                // to this message
+                msg_respond ( iv_msgQ, theMsg );
+                break;
             case ERRLOG_SHUTDOWN_TYPE:
                 TRACFCOMP( g_trac_errl, INFO_MRK "Shutdown event received" );
 
@@ -755,7 +763,42 @@ void ErrlManager::saveErrLogEntry( errlHndl_t& io_err )
 }
 #endif
 
+///////////////////////////////////////////////////////////////////////////////
+// ErrlManager::callFlushErrorLogs()
+///////////////////////////////////////////////////////////////////////////////
+void ErrlManager::callFlushErrorLogs()
+{
+    ERRORLOG::theErrlManager::instance().flushErrorLogs();
+    return;
+}
+///////////////////////////////////////////////////////////////////////////////
+// ErrlManager::flushErrorLogs()
+///////////////////////////////////////////////////////////////////////////////
+void ErrlManager::flushErrorLogs()
+{
+    TRACDCOMP( g_trac_errl, ENTER_MRK "ErrlManager::flushErrorLogs" );
+
+    // Create message to send to msg handler
+    msg_t *msg = NULL;
+    msg = msg_allocate();
+    msg->type = ERRLOG_FLUSH_TYPE;
+    do{
+        // Send message to msg handler, get msg back on l_RecvMsgQ
+        int rc = msg_sendrecv(iv_msgQ,msg);
 
+        if(rc)
+        {
+            TRACFCOMP(g_trac_errl, "Error sending error log flush message. "
+                "RC= %d",rc);
+            break;
+        }
+        // Msg_sendrecv doesn't return until the message has been responded
+        // to. So we do not need to check message type back.
+    }while(0);
+
+    TRACDCOMP( g_trac_errl, EXIT_MRK"ErrlManager::flushErrorLogs" );
+    return;
+}
 ///////////////////////////////////////////////////////////////////////////////
 ///////////////////////////////////////////////////////////////////////////////
 void ErrlManager::setHwasProcessCalloutFn(HWAS::processCalloutFn i_fn)
diff --git a/src/usr/hwas/hwasPlatDeconfigGard.C b/src/usr/hwas/hwasPlatDeconfigGard.C
index 0377cbd10..b98fa28d8 100644
--- a/src/usr/hwas/hwasPlatDeconfigGard.C
+++ b/src/usr/hwas/hwasPlatDeconfigGard.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2013,2015                        */
+/* Contributors Listed Below - COPYRIGHT 2013,2016                        */
 /* [+] Google Inc.                                                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
@@ -42,6 +42,7 @@
 #include <stdio.h>
 #include <sys/mm.h>
 #include <config.h>
+#include <initservice/istepdispatcherif.H>
 
 #include <pnor/pnorif.H>
 
@@ -380,6 +381,17 @@ errlHndl_t DeconfigGard::platCreateGardRecord(
         l_pRecord->iv_padding[5] = 0;
 
         _flush((void *)l_pRecord);
+
+        // We wrote a new gard record, we need to make sure to increment the
+        // reboot count so we can reconfigure and attempt to IPL
+        // Call setNewGardRecord in initservice.
+        #ifndef __HOSTBOOT_RUNTIME
+        #ifdef CONFIG_BMC_IPMI
+        HWAS_INF("New gard record committed, call INITSERVICE "
+            "::setNewGardRecord()");
+        INITSERVICE::setNewGardRecord();
+        #endif
+        #endif
     }
     while (0);
 
diff --git a/src/usr/initservice/istepdispatcher/istepdispatcher.C b/src/usr/initservice/istepdispatcher/istepdispatcher.C
index 88cafc7e7..cec5013ea 100644
--- a/src/usr/initservice/istepdispatcher/istepdispatcher.C
+++ b/src/usr/initservice/istepdispatcher/istepdispatcher.C
@@ -42,6 +42,7 @@
 #include <sys/task.h>                    //  tid_t, task_create, etc
 #include <sys/misc.h>                    //  cpu_all_winkle
 #include <errl/errlentry.H>              //  errlHndl_t
+#include <errl/errlmanager.H>
 #include <initservice/isteps_trace.H>    //  ISTEPS_TRACE buffer
 #include <initservice/initsvcudistep.H>  //  InitSvcUserDetailsIstep
 #include <initservice/taskargs.H>        //  TASK_ENTRY_MACRO
@@ -120,7 +121,8 @@ IStepDispatcher::IStepDispatcher() :
     iv_futureShutdown(false),
     iv_istepToCompleteBeforeShutdown(0),
     iv_substepToCompleteBeforeShutdown(0),
-    iv_acceptIstepMessages(true)
+    iv_acceptIstepMessages(true),
+    iv_newGardRecord(false)
 
 {
     mutex_init(&iv_bkPtMutex);
@@ -540,6 +542,39 @@ errlHndl_t IStepDispatcher::executeAllISteps()
                         else
                         {
                             #ifdef CONFIG_BMC_IPMI
+                            // Check the newGardRecord instance variable
+                            if(iv_newGardRecord)
+                            {
+                                // We have a new gard record committed. We need
+                                // to increment the reboot count
+                                uint16_t l_count = 0;
+                                SENSOR::RebootCountSensor l_sensor;
+                                // Read reboot count sensor
+                                err = l_sensor.getRebootCount(l_count);
+                                if (err)
+                                {
+                                    TRACFCOMP(g_trac_initsvc, ERR_MRK"executeAllISteps: getRebootCount failed");
+                                    break;
+                                }
+                                // Increment reboot count
+                                l_count++;
+                                err = l_sensor.setRebootCount(l_count);
+                                if (err)
+                                {
+                                    TRACFCOMP(g_trac_initsvc, ERR_MRK"executeAllISteps: setRebootCount failed");
+                                    break;
+                                }
+                            }
+                            else
+                            {
+                                // We are in a reconfig loop, but no new gard
+                                // records have been set. We will not increment
+                                // the reboot count.
+                                TRACFCOMP(g_trac_initsvc, "Reconfig loop needed "
+                                    "but no new gard records were commited. Do "
+                                    "not increment reboot count.");
+                            }
+
                             // @TODO RTC:124679 - Remove Once BMC Monitors
                             // Shutdown Attention
                             // Set Watchdog Timer before calling doShutdown()
@@ -782,6 +817,11 @@ errlHndl_t IStepDispatcher::doIstep(uint32_t i_istep,
         {
             // There was no PLD, process any deferred deconfig records (i.e.
             // actually do the deconfigures)
+            // We need to flush the errl buffer first
+            ERRORLOG::ErrlManager::callFlushErrorLogs();
+
+            // Regardless of the way the flush came back, we need to try to
+            // process the deferred deconfigs
             HWAS::theDeconfigGard().processDeferredDeconfig();
         }
 
@@ -1407,6 +1447,20 @@ void IStepDispatcher::iStepBreakPoint(uint32_t i_info)
     TRACFCOMP(g_trac_initsvc, EXIT_MRK"IStepDispatcher::handleBreakpointMsg");
     mutex_unlock(&iv_bkPtMutex);
 }
+// -----------------------------------------------------------------------------
+// IStepDispatcher::setNewGardRecord()
+// -----------------------------------------------------------------------------
+void IStepDispatcher::setNewGardRecord()
+{
+    TRACDCOMP(g_trac_initsvc, ENTER_MRK"IStepDispatcher::setNewGardRecord");
+
+    mutex_lock(&iv_mutex);
+    iv_newGardRecord = true;
+    mutex_unlock(&iv_mutex);
+
+    TRACDCOMP(g_trac_initsvc, EXIT_MRK"IStepDispatcher::setNewGardRecord");
+    return;
+}
 
 // ----------------------------------------------------------------------------
 // IStepDispatcher::isShutdownRequested()
@@ -1982,6 +2036,11 @@ void setAcceptIstepMessages(bool i_accept)
 {
     return IStepDispatcher::getTheInstance().setAcceptIstepMessages(i_accept);
 }
+void setNewGardRecord()
+{
+    return IStepDispatcher::getTheInstance().setNewGardRecord();
+}
+
 // ----------------------------------------------------------------------------
 // IStepDispatcher::getIstepInfo()
 // ----------------------------------------------------------------------------
diff --git a/src/usr/initservice/istepdispatcher/istepdispatcher.H b/src/usr/initservice/istepdispatcher/istepdispatcher.H
index 5d941963c..186639fa6 100644
--- a/src/usr/initservice/istepdispatcher/istepdispatcher.H
+++ b/src/usr/initservice/istepdispatcher/istepdispatcher.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2011,2015                        */
+/* Contributors Listed Below - COPYRIGHT 2011,2016                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -197,6 +197,12 @@ public:
 
     void setAcceptIstepMessages(bool accept);
 
+    /**
+     * @brief This function will set a boolean true which states a new gard
+     * record has been written.
+     */
+    void setNewGardRecord();
+
 protected:
 
     /**
@@ -415,6 +421,9 @@ private:
     //   accept istep messages or not
     bool iv_acceptIstepMessages;
 
+    // Instance variable to state if a new gard record was committed
+    bool iv_newGardRecord;
+
     // Message Queue for receiving message from SP or SPless user console
     msg_q_t iv_msgQ;