Allow subsequent PNOR operations after ECC errors

Change-Id: I4e66a383e088fa767849de3a1fb49800dd29a4d1
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/6965
Tested-by: Jenkins Server
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>

3 files changed, 43 insertions, 14 deletions

diff --git a/src/usr/pnor/pnorrp.C b/src/usr/pnor/pnorrp.C
index cfb7fbd07..0f212bce5 100644
--- a/src/usr/pnor/pnorrp.C
+++ b/src/usr/pnor/pnorrp.C
@@ -152,7 +152,6 @@ void* wait_for_message( void* unused )
 PnorRP::PnorRP()
 : iv_msgQ(NULL)
 ,iv_startupRC(0)
-,iv_shutdownUE(false)
 {
     TRACFCOMP(g_trac_pnor, "PnorRP::PnorRP> " );
 
@@ -580,12 +579,6 @@ void PnorRP::waitForMessage()
                 switch(message->type)
                 {
                     case( MSG_MM_RP_READ ):
-                        // do not allow reads in the shutdown path, only writes
-                        if( iv_shutdownUE )
-                        {
-                            status_rc = -EIO;
-                            break;
-                        }
                         l_errhdl = readFromDevice( dev_offset,
                                                    chip_select,
                                                    needs_ecc,
@@ -729,7 +722,6 @@ 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.
-                iv_shutdownUE = true;
                 o_fatalError = true;
                 INITSERVICE::doShutdown( PNOR::RC_ECC_UE, true );
             }
diff --git a/src/usr/pnor/pnorrp.H b/src/usr/pnor/pnorrp.H
index d246360b0..8c2fb2abc 100644
--- a/src/usr/pnor/pnorrp.H
+++ b/src/usr/pnor/pnorrp.H
@@ -149,12 +149,6 @@ class PnorRP
     std::map<uint64_t,FlashStats_t> iv_stats;
 
     /**
-     * Remember that we got a UE and we are shutting down
-     */
-    bool iv_shutdownUE;
-
-
-    /**
      * @brief Initialize the daemon, called by constructor
      */
     void initDaemon();
diff --git a/src/usr/pnor/test/pnorrptest.H b/src/usr/pnor/test/pnorrptest.H
index d48ec491c..71766fea0 100644
--- a/src/usr/pnor/test/pnorrptest.H
+++ b/src/usr/pnor/test/pnorrptest.H
@@ -248,6 +248,49 @@ class PnorRpTest : public CxxTest::TestSuite
             }
         }
 
+        //Enable this to test the UE handling, not turning on by default
+        // because it causes a shutdown
+#if 0 
+        TRACFCOMP(g_trac_pnor, "PnorRpTest::test_ECC: Attempting UE" );
+        // flush the page to make sure it gets out to the device
+        totals++;
+        rc = mm_remove_pages( RELEASE, dataptr, PAGESIZE );
+        if( rc )
+        {
+            TRACFCOMP( g_trac_pnor, "PnorRpTest::test_ECC> ERROR : error on RELEASE 2: rc=%X", rc );
+            TS_FAIL( "PnorRpTest::test_ECC> ERROR : error on RELEASE 2" );
+            fails++;
+        }
+        // generate data with CEs
+        chip_data_ptr = chip_data;
+        for (int i = 0; i < 9; i++)
+        {
+            memcpy( &data_ecc, chip_data_ptr, sizeof(uint64_t) );
+            uint64_t bad_data = data_ecc ^ (1ul << i) ^ (1ul << 9);
+            memcpy( chip_data_ptr, &bad_data, sizeof(uint64_t) );
+            chip_data_ptr += 9;
+        }
+        // write the bad data to the chip directly
+        totals++;
+        errhdl = deviceWrite(TARGETING::MASTER_PROCESSOR_CHIP_TARGET_SENTINEL,
+                             chip_data,
+                             l_size,
+                             DEVICE_PNOR_ADDRESS(0, TEST_PHYS_OFFSET));
+        if( errhdl )
+        {
+            TS_FAIL("PnorRpTest::test_ECC: PNORDD deviceWrite() 2 failed! Error committed.");
+            ERRORLOG::errlCommit(errhdl,PNOR_COMP_ID);
+            fails++;
+        }
+        uint32_t tmp2 = 0;
+        totals++;
+        memcpy( &tmp2, dataptr, sizeof(uint32_t) );
+        TRACFCOMP(g_trac_pnor, "PnorRpTest::test_ECC: UE all done - should never see this!!" );
+        fails++;
+        TS_FAIL("PnorRpTest::test_ECC: UE did not kill the task!.");
+#endif
+
+
         TRACFCOMP(g_trac_pnor, "PnorRpTest::test_ECC> %d/%d fails", fails, total );
     };