FSI SCOM Error Handling

Change-Id: I1538fbf386d5480e473f3f0b049492d494412624
RTC: 35064
Reviewed-on: http://gfw160.austin.ibm.com:8080/gerrit/5825
Tested-by: Jenkins Server
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>

4 files changed, 258 insertions, 109 deletions

diff --git a/src/usr/fsiscom/fsiscom.C b/src/usr/fsiscom/fsiscom.C
index d87a6f393..0c3e94361 100644
--- a/src/usr/fsiscom/fsiscom.C
+++ b/src/usr/fsiscom/fsiscom.C
@@ -28,14 +28,18 @@
 #include <trace/interface.H>
 #include <errl/errlentry.H>
 #include <errl/errlmanager.H>
+#include <errl/errludtarget.H>
+#include <xscom/piberror.H>
 #include <fsiscom/fsiscom_reasoncodes.H>
+#include <fsi/fsiif.H>
+#include <sys/time.h>
 #include "fsiscom.H"
 
 //Globals/Constants
 
 // Trace definition
 trace_desc_t* g_trac_fsiscom = NULL;
-TRAC_INIT(&g_trac_fsiscom, "FSISCOM", 2*KILOBYTE); //2K
+TRAC_INIT(&g_trac_fsiscom, FSISCOM_COMP_NAME, 2*KILOBYTE); //2K
 
 // Easy macro replace for unit testing
 //#define TRACUCOMP(args...)  TRACFCOMP(args)
@@ -54,8 +58,55 @@ union ioData6432
     };
 };
 
-//@fixme - not full tested due to simics instability.  Will full test when
-// enabling the scom test cases.
+///////////////////////////////////////////////////////////////////////////////
+///////////////////////////////////////////////////////////////////////////////
+void pib_error_handler( TARGETING::Target* i_target,
+                        errlHndl_t i_errlog,
+                        uint32_t i_status )
+{
+    //Add this target to the FFDC
+    ERRORLOG::ErrlUserDetailsTarget(i_target,"SCOM Target").addToLog(i_errlog);
+
+    //Add the callouts for the specific PCB/PIB error
+    uint32_t pib_error = i_status >> 12;
+    PIB::addFruCallouts( i_target,
+                         pib_error,
+                         i_errlog );
+
+    //Grab the PIB2OPB Status reg for a Resource Occupied error
+    if( pib_error == PIB::PIB_RESOURCE_OCCUPIED ) //piberr=001
+    {
+        FSI::getFsiFFDC( FSI::FFDC_PIB_FAIL,
+                         i_errlog,
+                         i_target );
+    }
+
+    //Recovery sequence from Markus
+    //  if SCOM fails and FSI Master displays "MasterTimeOut"
+    //     then 7,6  <covered by FSI driver>
+    //  else if SCOM fails and FSI2PIB Status shows PIB abort
+    //     then just perform unit reset (6) and wait 1 ms
+    //  else (PIB_abort='0' but PIB error is unequal 0)
+    //     then just perform unit reset (6) (wait not needed).
+    uint32_t l_command = 0;
+    size_t op_size = sizeof(uint32_t);
+    errlHndl_t l_err = DeviceFW::deviceOp( DeviceFW::WRITE,
+                                       i_target,
+                                       &l_command,
+                                       op_size,
+                                       DEVICE_FSI_ADDRESS(ENGINE_RESET_REG));
+    if(l_err)
+    {
+        TRACFCOMP( g_trac_fsiscom,
+                   ERR_MRK"Error resetting FSI : %.4X",
+                   ERRL_GETRC_SAFE(l_err) );
+        l_err->plid(i_errlog->plid());
+        errlCommit(l_err,FSISCOM_COMP_ID);
+    }
+
+    nanosleep( 0,NS_PER_MSEC ); //sleep for ms
+
+}
 
 ///////////////////////////////////////////////////////////////////////////////
 ///////////////////////////////////////////////////////////////////////////////
@@ -87,36 +138,41 @@ errlHndl_t fsiScomPerformOp(DeviceFW::OperationType i_opType,
              * @reasoncode   FSISCOM::RC_INVALID_LENGTH
              * @userdata1    SCOM Address
              * @userdata2    Data Length
-             * @devdesc      FSISCOM: fsiScomPerformOp> Invalid data length (!= 8 bytes)
+             * @devdesc      fsiScomPerformOp> Invalid data length (!= 8 bytes)
              */
             l_err = new ERRORLOG::ErrlEntry(ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                                             FSISCOM::MOD_FSISCOM_PERFORMOP,
                                             FSISCOM::RC_INVALID_LENGTH,
                                             l_scomAddr,
                                             TO_UINT64(io_buflen));
-            //@fixme: Need to callout target somehow.  Need to decide how to callout target and where
-            //  it should be done (this layer or somewhere higher in the call stack?)
+            l_err->addProcedureCallout( HWAS::EPUB_PRC_HB_CODE,
+                                        HWAS::SRCI_PRIORITY_LOW );
+            ERRORLOG::ErrlUserDetailsTarget(i_target,"SCOM Target").
+              addToLog(l_err);
             break;
         }
 
         if( (l_scomAddr & 0xFFFFFFFF80000000) != 0)
         {
-            TRACFCOMP( g_trac_fsiscom, ERR_MRK "fsiScomPerformOp> Address contains more than 31 bits : l_scomAddr=0x%.8x", l_scomAddr );
+            TRACFCOMP( g_trac_fsiscom, ERR_MRK "fsiScomPerformOp> Address contains more than 31 bits : l_scomAddr=0x%.16X", l_scomAddr );
             /*@
              * @errortype
              * @moduleid     FSISCOM::MOD_FSISCOM_PERFORMOP
              * @reasoncode   FSISCOM::RC_INVALID_ADDRESS
              * @userdata1    SCOM Address
              * @userdata2    0
-             * @devdesc      FSISCOM: fsiScomPerformOp> Address contains more than 31 bits.
+             * @devdesc      fsiScomPerformOp> Address contains
+             *               more than 31 bits.
              */
             l_err = new ERRORLOG::ErrlEntry(ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                                             FSISCOM::MOD_FSISCOM_PERFORMOP,
                                             FSISCOM::RC_INVALID_ADDRESS,
                                             l_scomAddr,
-                                            0);
-            //@fixme: Need to callout target somehow.  Need to decide how to callout target and where
-            //  it should be done (this layer or somewhere higher in the call stack?)
+                                            TARGETING::get_huid(i_target));
+            l_err->addProcedureCallout( HWAS::EPUB_PRC_HB_CODE,
+                                        HWAS::SRCI_PRIORITY_LOW );
+            ERRORLOG::ErrlUserDetailsTarget(i_target,"SCOM Target").
+              addToLog(l_err);
             break;
         }
 
@@ -183,12 +239,9 @@ errlHndl_t fsiScomPerformOp(DeviceFW::OperationType i_opType,
                 break;
             }
 
-            // atomic section <<
-            need_unlock = false;
-            mutex_unlock(l_mutex);
-
-            //bits 17-19 indicates PCB/PIB error
-            if(l_status & 0x00007000)
+            // Check the status reg for errors
+            if( (l_status & PIB_ERROR_BITS)      // PCB/PIB Errors
+                || (l_status & PIB_ABORT_BIT)  ) // PIB Abort
             {
                 TRACFCOMP( g_trac_fsiscom, ERR_MRK"fsiScomPerformOp:Write: PCB/PIB error received: l_status=0x%X)", l_status);
                 /*@
@@ -197,17 +250,18 @@ errlHndl_t fsiScomPerformOp(DeviceFW::OperationType i_opType,
                  * @reasoncode   FSISCOM::RC_WRITE_ERROR
                  * @userdata1    SCOM Addr
                  * @userdata2    SCOM Status Reg
-                 * @devdesc      fsiScomPerformOp> Error returned from SCOM Engine after write
+                 * @devdesc      fsiScomPerformOp> Error returned
+                 *               from SCOM Engine after write
                  */
-                l_err = new ERRORLOG::ErrlEntry(ERRORLOG::ERRL_SEV_UNRECOVERABLE,
-                                                FSISCOM::MOD_FSISCOM_PERFORMOP,
-                                                FSISCOM::RC_WRITE_ERROR,
-                                                l_scomAddr,
-                                                TO_UINT64(l_status));
+                l_err = new ERRORLOG::ErrlEntry(
+                                ERRORLOG::ERRL_SEV_UNRECOVERABLE,
+                                FSISCOM::MOD_FSISCOM_PERFORMOP,
+                                FSISCOM::RC_WRITE_ERROR,
+                                l_scomAddr,
+                                TO_UINT64(l_status));
 
-                //@fixme: Need to callout target somehow.  Need to decide how to callout target and where
-                //  it should be done (this layer or somewhere higher in the call stack?)
-                //@todo: May add recover actions later.  Currently undefined
+                // call common error handler to do callouts and recovery
+                pib_error_handler( i_target, l_err, l_status );
 
                 //Grab the PIB2OPB Status reg for a XSCOM Block error
                 if( (l_status & 0x00007000) == 0x00001000 ) //piberr=001
@@ -234,8 +288,9 @@ errlHndl_t fsiScomPerformOp(DeviceFW::OperationType i_opType,
                 break;
             }
 
-
-
+            // atomic section <<
+            need_unlock = false;
+            mutex_unlock(l_mutex);
         }
         else if(i_opType == DeviceFW::READ)
         {
@@ -269,8 +324,9 @@ errlHndl_t fsiScomPerformOp(DeviceFW::OperationType i_opType,
                 break;
             }
 
-           //bits 17-19 indicates PCB/PIB error
-            if((l_status & 0x00007000) != 0)
+            // Check the status reg for errors
+            if( (l_status & PIB_ERROR_BITS)      // PCB/PIB Errors
+                || (l_status & PIB_ABORT_BIT)  ) // PIB Abort
             {
                 TRACFCOMP( g_trac_fsiscom, ERR_MRK"fsiScomPerformOp:Read: PCB/PIB error received: l_status=0x%0.8X)", l_status);
 
@@ -288,9 +344,9 @@ errlHndl_t fsiScomPerformOp(DeviceFW::OperationType i_opType,
                                                 l_scomAddr,
                                                 TO_UINT64(l_status));
 
-                //@fixme: Need to callout target somehow.  Need to decide how to callout target and where
-                //  it should be done (this layer or somewhere higher in the call stack?)
-                //@todo: May add recover actions later.  Currently undefined
+                // call common error handler to do callouts and recovery
+                pib_error_handler( i_target, l_err, l_status );
+
                 break;
             }
 
@@ -332,18 +388,24 @@ errlHndl_t fsiScomPerformOp(DeviceFW::OperationType i_opType,
              * @errortype
              * @moduleid     FSISCOM::MOD_FSISCOM_PERFORMOP
              * @reasoncode   FSISCOM::RC_INVALID_OPTYPE
-             * @userdata1    Operation Type (i_opType) : 0=READ, 1=WRITE
-             * @userdata2    0
+             * @userdata1[0:31]    Operation Type (i_opType) : 0=READ, 1=WRITE
+             * @userdata1[32:64]   Input scom address
+             * @userdata2    Target HUID
              * @devdesc      fsiScomPerformOp> Unsupported Operation Type specified
              */
             l_err = new ERRORLOG::ErrlEntry(ERRORLOG::ERRL_SEV_UNRECOVERABLE,
                                             FSISCOM::MOD_FSISCOM_PERFORMOP,
                                             FSISCOM::RC_INVALID_OPTYPE,
-                                            TO_UINT64(i_opType),
-                                            0);
+                                            TWO_UINT32_TO_UINT64(i_opType,
+                                                                 l_scomAddr),
+                                            TARGETING::get_huid(i_target));
+            //Add this target to the FFDC
+            ERRORLOG::ErrlUserDetailsTarget(i_target,"SCOM Target").
+              addToLog(l_err);
+
+            l_err->addProcedureCallout(HWAS::EPUB_PRC_HB_CODE,
+                                       HWAS::SRCI_PRIORITY_HIGH);
 
-            //@fixme: Need to callout target somehow.  Need to decide how to callout target and where
-            //  it should be done (this layer or somewhere higher in the call stack?)
             break;
 
         }
diff --git a/src/usr/fsiscom/fsiscom.H b/src/usr/fsiscom/fsiscom.H
index 62ed12f27..0c17946b3 100644
--- a/src/usr/fsiscom/fsiscom.H
+++ b/src/usr/fsiscom/fsiscom.H
@@ -1,25 +1,25 @@
-//  IBM_PROLOG_BEGIN_TAG
-//  This is an automatically generated prolog.
-//
-//  $Source: src/usr/fsiscom/fsiscom.H $
-//
-//  IBM CONFIDENTIAL
-//
-//  COPYRIGHT International Business Machines Corp. 2011
-//
-//  p1
-//
-//  Object Code Only (OCO) source materials
-//  Licensed Internal Code Source Materials
-//  IBM HostBoot Licensed Internal Code
-//
-//  The source code for this program is not published or other-
-//  wise divested of its trade secrets, irrespective of what has
-//  been deposited with the U.S. Copyright Office.
-//
-//  Origin: 30
-//
-//  IBM_PROLOG_END
+/* IBM_PROLOG_BEGIN_TAG                                                   */
+/* This is an automatically generated prolog.                             */
+/*                                                                        */
+/* $Source: src/usr/fsiscom/fsiscom.H $                                   */
+/*                                                                        */
+/* IBM CONFIDENTIAL                                                       */
+/*                                                                        */
+/* COPYRIGHT International Business Machines Corp. 2011,2013              */
+/*                                                                        */
+/* p1                                                                     */
+/*                                                                        */
+/* Object Code Only (OCO) source materials                                */
+/* Licensed Internal Code Source Materials                                */
+/* IBM HostBoot Licensed Internal Code                                    */
+/*                                                                        */
+/* The source code for this program is not published or otherwise         */
+/* divested of its trade secrets, irrespective of what has been           */
+/* deposited with the U.S. Copyright Office.                              */
+/*                                                                        */
+/* Origin: 30                                                             */
+/*                                                                        */
+/* IBM_PROLOG_END_TAG                                                     */
 #ifndef __FSISCOM_H
 #define __FSISCOM_H
 
@@ -31,12 +31,18 @@ namespace FSISCOM
 {
 
   enum {
-    //FSI addresses are byte offsets, so need to multiply by 4 since each register is 4 bytes long.
-    //prefix with 0x10xx for engine offset
-    DATA0_REG   = 0x1000,  /* SCOM Data Register 0 (0x00) */
-    DATA1_REG   = 0x1004,  /* SCOM Data Register 1 (0x01) */
-    COMMAND_REG = 0x1008,  /* SCOM Command Register (0x02) */
-    STATUS_REG  = 0x101C,  /* STATUS Register (0x07) */
+    //FSI addresses are byte offsets, so need to multiply by 4
+    // since each register is 4 bytes long.
+    // prefix with 0x10xx for FSI2PIB engine offset
+    DATA0_REG         = 0x1000,  /* SCOM Data Register 0 (0x00) */
+    DATA1_REG         = 0x1004,  /* SCOM Data Register 1 (0x01) */
+    COMMAND_REG       = 0x1008,  /* SCOM Command Register (0x02) */
+    ENGINE_RESET_REG  = 0x1018,  /* Engine Reset Register (0x06) */
+    STATUS_REG        = 0x101C,  /* STATUS Register (0x07) */
+    PIB_RESET_REG     = 0x101C,  /* PIB Reset Register (0x07) */
+
+    PIB_ABORT_BIT     = 0x00100000, /* 12= PIB Abort */
+    PIB_ERROR_BITS    = 0x00007000, /* 17:19= PCB/PIB Errors */
   };
 
 /**
diff --git a/src/usr/fsiscom/test/fsiscomtest.H b/src/usr/fsiscom/test/fsiscomtest.H
index 5c570793b..16481987d 100644
--- a/src/usr/fsiscom/test/fsiscomtest.H
+++ b/src/usr/fsiscom/test/fsiscomtest.H
@@ -1,43 +1,124 @@
-//  IBM_PROLOG_BEGIN_TAG
-//  This is an automatically generated prolog.
-//
-//  $Source: src/usr/fsiscom/test/fsiscomtest.H $
-//
-//  IBM CONFIDENTIAL
-//
-//  COPYRIGHT International Business Machines Corp. 2011
-//
-//  p1
-//
-//  Object Code Only (OCO) source materials
-//  Licensed Internal Code Source Materials
-//  IBM HostBoot Licensed Internal Code
-//
-//  The source code for this program is not published or other-
-//  wise divested of its trade secrets, irrespective of what has
-//  been deposited with the U.S. Copyright Office.
-//
-//  Origin: 30
-//
-//  IBM_PROLOG_END
+/* IBM_PROLOG_BEGIN_TAG                                                   */
+/* This is an automatically generated prolog.                             */
+/*                                                                        */
+/* $Source: src/usr/fsiscom/test/fsiscomtest.H $                          */
+/*                                                                        */
+/* IBM CONFIDENTIAL                                                       */
+/*                                                                        */
+/* COPYRIGHT International Business Machines Corp. 2011,2013              */
+/*                                                                        */
+/* p1                                                                     */
+/*                                                                        */
+/* Object Code Only (OCO) source materials                                */
+/* Licensed Internal Code Source Materials                                */
+/* IBM HostBoot Licensed Internal Code                                    */
+/*                                                                        */
+/* The source code for this program is not published or otherwise         */
+/* divested of its trade secrets, irrespective of what has been           */
+/* deposited with the U.S. Copyright Office.                              */
+/*                                                                        */
+/* Origin: 30                                                             */
+/*                                                                        */
+/* IBM_PROLOG_END_TAG                                                     */
 #include <cxxtest/TestSuite.H>
 #include <errl/errlmanager.H>
 #include <errl/errlentry.H>
 #include <limits.h>
 #include <devicefw/driverif.H>
+#include <fsiscom/fsiscom_reasoncodes.H>
 
+extern trace_desc_t* g_trac_fsiscom;
 
 class FsiScomTest : public CxxTest::TestSuite
 {
   public:
 
     /**
-     * @brief test1
+     * @brief Test error path and recovery
      */
-    void test_fsiScom(void)
+    void test_Error(void)
     {
+        TRACFCOMP( g_trac_fsiscom, "FsiScomTest::test_Error> Start" );
 
-        //FSISCOM testing driving by SCOM device driver.  This allows for testing FSISCOM <-> XSCOM
-        //coherency.
+        //@fixme: RTC:80081
+        TRACFCOMP( g_trac_fsiscom, "FsiScomTest::test_Error> Disabling test until D892724 is in a build" );
+#if 0
+        uint64_t fails = 0;
+        uint64_t total = 0;
+        errlHndl_t l_err = NULL;
+        uint64_t regdata = 0;
+        size_t op_size = sizeof(uint64_t);
+
+        // Centaur target
+        TARGETING::EntityPath epath(TARGETING::EntityPath::PATH_PHYSICAL);
+        epath.addLast(TARGETING::TYPE_SYS,0);
+        epath.addLast(TARGETING::TYPE_NODE,0);
+        epath.addLast(TARGETING::TYPE_MEMBUF,4);
+        TARGETING::Target* fsi_target =
+          TARGETING::targetService().toTarget(epath);
+        if( fsi_target == NULL )
+        {
+            TRACFCOMP( g_trac_fsiscom, "FsiScomTest::test_Error> Target is NULL" );
+            TS_FAIL( "FsiScomTest::test_Error> Target is NULL" );
+        }
+
+        // Bad address read
+        total++;
+        l_err = DeviceFW::deviceOp(
+                          DeviceFW::READ,
+                          fsi_target,
+                          &regdata,
+                          op_size,
+                          DEVICE_FSISCOM_ADDRESS(0x11111111) );
+        if( l_err && (l_err->reasonCode() == FSISCOM::RC_READ_ERROR) )
+        {
+            delete l_err;
+        }
+        else
+        {
+            TRACFCOMP(g_trac_fsiscom, "FsiScomTest::test_Error> No/wrong error for bad address read, RC=%X", ERRL_GETRC_SAFE(l_err) );
+            TS_FAIL( "FsiScomTest::test_Error> No/wrong error for bad address read" );
+            fails++;
+            if( l_err ) { errlCommit(l_err,FSISCOM_COMP_ID); }
+        }
+
+        // Bad address write
+        total++;
+        l_err = DeviceFW::deviceOp(
+                          DeviceFW::WRITE,
+                          fsi_target,
+                          &regdata,
+                          op_size,
+                          DEVICE_FSISCOM_ADDRESS(0x00222222) );
+        if( l_err && (l_err->reasonCode() == FSISCOM::RC_WRITE_ERROR) )
+        {
+            delete l_err;
+        }
+        else
+        {
+            TRACFCOMP(g_trac_fsiscom, "FsiScomTest::test_Error> No/wrong error for bad address write, RC=%X", ERRL_GETRC_SAFE(l_err) );
+            TS_FAIL( "FsiScomTest::test_Error> No/wrong error for bad address write" );
+            fails++;
+            if( l_err ) { errlCommit(l_err,FSISCOM_COMP_ID); }
+        }
+
+        // Do a good scom to prove things are working again
+        total++;
+        l_err = DeviceFW::deviceOp(
+                          DeviceFW::READ,
+                          fsi_target,
+                          &regdata,
+                          op_size,
+                          DEVICE_FSISCOM_ADDRESS(0x02011403) );
+        if( l_err )
+        {
+            TRACFCOMP(g_trac_fsiscom, "FsiScomTest::test_Error> Error reading 0x02011403, RC=%X", ERRL_GETRC_SAFE(l_err) );
+            TS_FAIL( "FsiScomTest::test_Error> Error reading 0x02011403" );
+            fails++;
+            errlCommit(l_err,FSISCOM_COMP_ID);
+        }
+
+        TRACFCOMP( g_trac_fsiscom, "FsiScomTest::test_Error> %d/%d fails", fails, total );
+#endif
     };
 };
diff --git a/src/usr/fsiscom/test/makefile b/src/usr/fsiscom/test/makefile
index b6c001ea8..7753f9a1a 100644
--- a/src/usr/fsiscom/test/makefile
+++ b/src/usr/fsiscom/test/makefile
@@ -1,29 +1,29 @@
-#  IBM_PROLOG_BEGIN_TAG
-#  This is an automatically generated prolog.
+# IBM_PROLOG_BEGIN_TAG
+# This is an automatically generated prolog.
 #
-#  $Source: src/usr/fsiscom/test/makefile $
+# $Source: src/usr/fsiscom/test/makefile $
 #
-#  IBM CONFIDENTIAL
+# IBM CONFIDENTIAL
 #
-#  COPYRIGHT International Business Machines Corp. 2011
+# COPYRIGHT International Business Machines Corp. 2011,2013
 #
-#  p1
+# p1
 #
-#  Object Code Only (OCO) source materials
-#  Licensed Internal Code Source Materials
-#  IBM HostBoot Licensed Internal Code
+# Object Code Only (OCO) source materials
+# Licensed Internal Code Source Materials
+# IBM HostBoot Licensed Internal Code
 #
-#  The source code for this program is not published or other-
-#  wise divested of its trade secrets, irrespective of what has
-#  been deposited with the U.S. Copyright Office.
+# The source code for this program is not published or otherwise
+# divested of its trade secrets, irrespective of what has been
+# deposited with the U.S. Copyright Office.
 #
-#  Origin: 30
+# Origin: 30
 #
-#  IBM_PROLOG_END
+# IBM_PROLOG_END_TAG
 ROOTPATH = ../../../..
 
-#MODULE = testfsiscom  Disabled as we don't currently have any tests defined.
-#TESTS = *.H
+MODULE = testfsiscom
+TESTS = *.H
 
 
 include ${ROOTPATH}/config.mk