PRD: Add additional FFDC to runtime NVDIMM analysis

Change-Id: I7472a883a27c30315c01f2722f98fe40d42c1070
CQ: SW476683
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/84181
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Benjamen G Tyner <ben.tyner@ibm.com>
Reviewed-by: Brian J Stegmiller <bjs@us.ibm.com>
Reviewed-by: Zane C Shelley <zshelle@us.ibm.com>
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/84182
Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com>
Tested-by: Jenkins OP HW <op-hw-jenkins+hostboot@us.ibm.com>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>

3 files changed, 74 insertions, 5 deletions

diff --git a/src/usr/diag/prdf/plat/mem/prdfP9Mca.C b/src/usr/diag/prdf/plat/mem/prdfP9Mca.C
index fc32c557b..fac29fce3 100644
--- a/src/usr/diag/prdf/plat/mem/prdfP9Mca.C
+++ b/src/usr/diag/prdf/plat/mem/prdfP9Mca.C
@@ -1359,7 +1359,7 @@ int32_t AnalyzeNvdimmHealthStatRegs( ExtensibleChip * i_chip,
             PRDF_ERR( PRDF_FUNC "Failed to get the global error log." );
             continue;
         }
-        PlatServices::nvdimmAddPage4Ffdc( dimm, mainErrl );
+        PlatServices::nvdimmAddFfdc( dimm, mainErrl );
 
         // De-assert the EVENT_N pin by setting bit 2 in NVDIMM_MGT_CMD1
         l_rc = __deassertEventN( dimm );
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices.C b/src/usr/diag/prdf/plat/prdfPlatServices.C
index e4b122ab5..491cc344d 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices.C
+++ b/src/usr/diag/prdf/plat/prdfPlatServices.C
@@ -59,6 +59,11 @@
 #include <p9c_mss_maint_cmds.H>
 #include <prdfParserUtils.H>
 #include <p9c_mss_rowRepairFuncs.H>
+#include <errl/errludlogregister.H>
+
+#ifdef CONFIG_NVDIMM
+#include <nvdimm.H>
+#endif
 
 using namespace TARGETING;
 
@@ -765,10 +770,75 @@ uint32_t nvdimmNotifyProtChange( TARGETING::TargetHandle_t i_target,
 
 }
 
-void nvdimmAddPage4Ffdc( TARGETING::TargetHandle_t i_nvdimm,
-                         errlHndl_t & io_errl  )
+void nvdimmAddFfdc( TARGETING::TargetHandle_t i_nvdimm, errlHndl_t & io_errl  )
 {
+    #define PRDF_FUNC "[PlatServices::nvdimmAddFfdc] "
+    // Add Page 4 Regs and Vendor Log using external Hostboot interfaces.
     NVDIMM::nvdimmAddPage4Regs( i_nvdimm, io_errl );
+    NVDIMM::nvdimmAddVendorLog( i_nvdimm, io_errl );
+
+    // Add PRD specific registers relevant to runtime NVDIMM analysis.
+    const uint16_t regList[] =
+    {
+        // Module health registers
+        NVDIMM::i2cReg::MODULE_HEALTH,
+        NVDIMM::i2cReg::MODULE_HEALTH_STATUS0,
+        NVDIMM::i2cReg::MODULE_HEALTH_STATUS1,
+
+        // Threshold status registers
+        NVDIMM::i2cReg::ERROR_THRESHOLD_STATUS,
+        NVDIMM::i2cReg::WARNING_THRESHOLD_STATUS,
+
+        // ES_TEMP registers
+        NVDIMM::i2cReg::ES_TEMP0,
+        NVDIMM::i2cReg::ES_TEMP1,
+        NVDIMM::i2cReg::ES_TEMP_WARNING_HIGH_THRESHOLD0,
+        NVDIMM::i2cReg::ES_TEMP_WARNING_HIGH_THRESHOLD1,
+        NVDIMM::i2cReg::ES_TEMP_WARNING_LOW_THRESHOLD0,
+        NVDIMM::i2cReg::ES_TEMP_WARNING_LOW_THRESHOLD1,
+
+        // NVM Lifetime registers
+        NVDIMM::i2cReg::NVM_LIFETIME,
+        NVDIMM::i2cReg::NVM_LIFETIME_ERROR_THRESHOLD,
+        NVDIMM::i2cReg::NVM_LIFETIME_WARNING_THRESHOLD,
+
+        // ES Lifetime registers
+        NVDIMM::i2cReg::ES_LIFETIME,
+        NVDIMM::i2cReg::ES_LIFETIME_ERROR_THRESHOLD,
+        NVDIMM::i2cReg::ES_LIFETIME_WARNING_THRESHOLD,
+
+        // Status registers
+        NVDIMM::i2cReg::ERASE_STATUS,
+        NVDIMM::i2cReg::ARM_STATUS,
+        NVDIMM::i2cReg::SET_EVENT_NOTIFICATION_STATUS,
+    };
+
+    ERRORLOG::ErrlUserDetailsLogRegister regUd( i_nvdimm );
+    for ( auto const & reg : regList )
+    {
+        // NVDIMM register size = 1 byte
+        size_t NVDIMM_SIZE = 1;
+
+        uint8_t data = 0;
+        errlHndl_t errl = deviceRead( i_nvdimm, &data, NVDIMM_SIZE,
+                                      DEVICE_NVDIMM_ADDRESS(reg) );
+        if ( errl )
+        {
+            PRDF_ERR( PRDF_FUNC "Failed to read register 0x%X on "
+                      "NVDIMM HUID: 0x%08x", reg, getHuid(i_nvdimm) );
+            // Don't commit, just delete the error and continue
+            delete errl; errl = nullptr;
+            continue;
+        }
+        // Only add registers that have non-zero data.
+        if ( 0 == data ) continue;
+
+        regUd.addDataBuffer( &data, sizeof(data), DEVICE_NVDIMM_ADDRESS(reg) );
+    }
+
+    regUd.addToLog( io_errl );
+
+    #undef PRDF_FUNC
 }
 
 #endif
diff --git a/src/usr/diag/prdf/plat/prdfPlatServices.H b/src/usr/diag/prdf/plat/prdfPlatServices.H
index 540742fd7..b99c20bed 100644
--- a/src/usr/diag/prdf/plat/prdfPlatServices.H
+++ b/src/usr/diag/prdf/plat/prdfPlatServices.H
@@ -189,8 +189,7 @@ uint32_t nvdimmNotifyProtChange( TARGETING::TargetHandle_t i_target,
  * @param i_nvdimm An nvdimm target
  * @param io_errl  Error log to add the FFDC to
  */
-void nvdimmAddPage4Ffdc( TARGETING::TargetHandle_t i_nvdimm,
-                         errlHndl_t & io_errl  );
+void nvdimmAddFfdc( TARGETING::TargetHandle_t i_nvdimm, errlHndl_t & io_errl  );
 
 #endif