PNOR ECC error to reduce task-crash log sev.

Change-Id: I2d5d41ccb641bdde67df1d4556d57cd4cc8e1228
CQ: SW249695
Backport: release-fips810
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/9888
Tested-by: Jenkins Server
Reviewed-by: Douglas R. Gilbert <dgilbert@us.ibm.com>
Reviewed-by: Daniel M. Crowell <dcrowell@us.ibm.com>
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>

5 files changed, 37 insertions, 14 deletions

diff --git a/src/kernel/msghandler.C b/src/kernel/msghandler.C
index 4530ed6b6..3b21de98d 100644
--- a/src/kernel/msghandler.C
+++ b/src/kernel/msghandler.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* IBM CONFIDENTIAL                                                       */
 /*                                                                        */
-/* COPYRIGHT International Business Machines Corp. 2011,2013              */
+/* COPYRIGHT International Business Machines Corp. 2011,2014              */
 /*                                                                        */
 /* p1                                                                     */
 /*                                                                        */
@@ -56,6 +56,7 @@ void MessageHandler::sendMessage(msg_sys_types_t i_type, void* i_key,
         m->type = i_type;
         m->data[0] = reinterpret_cast<uint64_t>(i_key);
         m->data[1] = reinterpret_cast<uint64_t>(i_data);
+        m->extra_data = NULL;
         m->__reserved__async = 1;
 
         // Create pending response object.
@@ -191,7 +192,7 @@ int MessageHandler::recvMessage(msg_t* i_msg)
 
     while(task_t* end_task = endTaskList.remove())
     {
-        TaskManager::endTask(end_task, NULL, TASK_STATUS_CRASHED);
+        TaskManager::endTask(end_task, i_msg->extra_data, TASK_STATUS_CRASHED);
     }
 
     // Release memory for message (created from sendMsg).
diff --git a/src/usr/initservice/baseinitsvc/initservice.C b/src/usr/initservice/baseinitsvc/initservice.C
index fb5e1b5da..2b1db6b35 100644
--- a/src/usr/initservice/baseinitsvc/initservice.C
+++ b/src/usr/initservice/baseinitsvc/initservice.C
@@ -261,6 +261,16 @@ errlHndl_t InitService::startTask(
                                     INITSERVICE::BASE_INITSVC_MOD_ID,
                                     INITSERVICE::WAIT_TASK_FAILED,
                                     l_tidretrc, l_childsts, hbSwError);
+
+            // If the task crashed, then the l_childerrl is either NULL or
+            // contains an RC indicating that the issue causing the child
+            // to crash has already been reported.  Therefore, reduce the
+            // severity to informational.
+            if ((l_childsts == TASK_STATUS_CRASHED) &&
+                (NULL != l_childerrl))
+            {
+                l_errl->setSev(ERRORLOG::ERRL_SEV_INFORMATIONAL);
+            }
             break;
         } // endif tidretrc
 
@@ -380,6 +390,16 @@ errlHndl_t InitService::executeFn(
                                     INITSERVICE::WAIT_FN_FAILED,
                                     l_tidretrc, l_childsts, hbSwError);
 
+            // If the task crashed, then the l_childerrl is either NULL or
+            // contains an RC indicating that the issue causing the child
+            // to crash has already been reported.  Therefore, reduce the
+            // severity to informational.
+            if ((l_childsts == TASK_STATUS_CRASHED) &&
+                (NULL != l_childerrl))
+            {
+                l_errl->setSev(ERRORLOG::ERRL_SEV_INFORMATIONAL);
+            }
+
             TRACFCOMP(g_trac_initsvc,
                     "ERROR : task_wait_tid(0x%x). '%s', l_tidretrc=0x%x, l_childsts=0x%x",
                     l_tidlnchrc,
@@ -542,7 +562,7 @@ void InitService::init( void *io_ptr )
                 HWAS::RC_SYSAVAIL_INSUFFICIENT_HW);
             l_shutdownStatus = HWAS::RC_SYSAVAIL_INSUFFICIENT_HW;
         }
-        else 
+        else
         {
             // Set the shutdown status to be the plid to force a TI
             l_shutdownStatus = l_errl->plid();
diff --git a/src/usr/pnor/pnorrp.C b/src/usr/pnor/pnorrp.C
index e31499c7c..f9f1b5d62 100644
--- a/src/usr/pnor/pnorrp.C
+++ b/src/usr/pnor/pnorrp.C
@@ -368,7 +368,7 @@ errlHndl_t PnorRP::readTOC()
     TRACUCOMP(g_trac_pnor, "PnorRP::readTOC>" );
     errlHndl_t l_errhdl = NULL;
     uint8_t* tocBuffer = NULL;
-    bool fatal_error = false;
+    uint64_t fatal_error = 0;
     bool TOC_0_failed = false;
 
     do{
@@ -644,7 +644,7 @@ void PnorRP::waitForMessage()
     bool needs_ecc = false;
     int rc = 0;
     uint64_t status_rc = 0;
-    bool fatal_error = false;
+    uint64_t fatal_error = 0;
 
     while(1)
     {
@@ -675,7 +675,7 @@ void PnorRP::waitForMessage()
                                                    needs_ecc,
                                                    user_addr,
                                                    fatal_error );
-                        if( l_errhdl || fatal_error )
+                        if( l_errhdl || ( 0 != fatal_error ) )
                         {
                             status_rc = -EIO; /* I/O error */
                         }
@@ -741,8 +741,10 @@ void PnorRP::waitForMessage()
             /*  Expected Response:
              *      data[0] = virtual address requested
              *      data[1] = rc (0 or negative errno value)
+             *      extra_data = Specific reason code.
              */
             message->data[1] = status_rc;
+            message->extra_data = reinterpret_cast<void*>(fatal_error);
             rc = msg_respond( iv_msgQ, message );
             if( rc )
             {
@@ -764,12 +766,12 @@ errlHndl_t PnorRP::readFromDevice( uint64_t i_offset,
                                    uint64_t i_chip,
                                    bool i_ecc,
                                    void* o_dest,
-                                   bool& o_fatalError )
+                                   uint64_t& o_fatalError )
 {
     TRACUCOMP(g_trac_pnor, "PnorRP::readFromDevice> i_offset=0x%X, i_chip=%d", i_offset, i_chip );
     errlHndl_t l_errhdl = NULL;
     uint8_t* ecc_buffer = NULL;
-    o_fatalError = false;
+    o_fatalError = 0;
 
     do
     {
@@ -819,7 +821,7 @@ errlHndl_t PnorRP::readFromDevice( uint64_t i_offset,
                 // Also need to spawn a separate task to do the shutdown
                 //  so that the regular PNOR task can service the writes
                 //  that happen during shutdown.
-                o_fatalError = true;
+                o_fatalError = PNOR::RC_ECC_UE;
                 INITSERVICE::doShutdown( PNOR::RC_ECC_UE, true );
             }
             // found an error so we need to fix something
diff --git a/src/usr/pnor/pnorrp.H b/src/usr/pnor/pnorrp.H
index a1e463973..baafab9f3 100644
--- a/src/usr/pnor/pnorrp.H
+++ b/src/usr/pnor/pnorrp.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* IBM CONFIDENTIAL                                                       */
 /*                                                                        */
-/* COPYRIGHT International Business Machines Corp. 2011,2013              */
+/* COPYRIGHT International Business Machines Corp. 2011,2014              */
 /*                                                                        */
 /* p1                                                                     */
 /*                                                                        */
@@ -178,8 +178,8 @@ class PnorRP
      * @param[in] i_chip  Which PNOR chip
      * @param[in] i_ecc  true=apply ECC after reading
      * @param[out] o_dest  Buffer to copy data into
-     * @param[out] o_fatalError  true=fatal error encountered, but no
-     *                    log could be created
+     * @param[out] o_fatalError  non-zero=fatal error encountered, but no
+     *                    log could be created.  Value is the internal RC.
      *
      * @return Error from device
      */
@@ -187,7 +187,7 @@ class PnorRP
                                uint64_t i_chip,
                                bool i_ecc,
                                void* o_dest,
-                               bool& o_fatalError );
+                               uint64_t& o_fatalError );
 
     /**
      * @brief  Write 1 logical page of data to the PNOR device
diff --git a/src/usr/pnor/test/pnorrptest.H b/src/usr/pnor/test/pnorrptest.H
index 0ad5fc348..32f3c1dc2 100644
--- a/src/usr/pnor/test/pnorrptest.H
+++ b/src/usr/pnor/test/pnorrptest.H
@@ -508,7 +508,7 @@ class PnorRpTest : public CxxTest::TestSuite
     {
         TRACFCOMP(g_trac_pnor, "PnorRpTest::test_TOC Start" );
 
-        bool fatal_error = false;
+        uint64_t fatal_error = 0;
         uint64_t offset = 0;
         uint8_t* tocHeader = new uint8_t[PAGESIZE];
         uint8_t* tocEntry = new uint8_t[PAGESIZE];