Added support for checking on the health status of an NVDIMM

* The check for the health status piggy backs off the current
  NVDIMM operation interface.
* Added the method doNvDimmCheckHealthStatus to perform the
  health check status of the individual NVDIMMs.
* Added wrapper methods nvDimmCheckHealthStatusOnSystem and
  nvdimmCheckHealthStatus that call the doNvDimmCheckHealthStatus
* Added an interface in the runtime commands to call make
  the check health status call.

Change-Id: Iefa1fcf5cb6a13c496fd776cdc34ade58ae0612b
CQ:SW469962
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/80589
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
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>
Reviewed-by: Corey V Swenson <cswenson@us.ibm.com>
Reviewed-by: Matt Derksen <mderkse1@us.ibm.com>
Reviewed-by: Daniel M Crowell <dcrowell@us.ibm.com>

3 files changed, 46 insertions, 1 deletions

diff --git a/src/include/runtime/interface.h b/src/include/runtime/interface.h
index 7d66f0ff6..9a0cb478e 100644
--- a/src/include/runtime/interface.h
+++ b/src/include/runtime/interface.h
@@ -579,6 +579,8 @@ typedef struct hostInterfaces
     //       error out.
     enum NVDIMM_Op_t: uint16_t
     {
+        /// The following operations pertain to arming/disarming
+        /// the NVDIMM
         // Disarm the NV logic such that the next save attempt is a NOOP
         HBRT_FW_NVDIMM_DISARM             = 0x0001,
         // Disable encryption on the NVDIMM and clear saved values from FW
@@ -589,6 +591,12 @@ typedef struct hostInterfaces
         HBRT_FW_NVDIMM_ENABLE_ENCRYPTION  = 0x0008,
         // Arm the NV logic
         HBRT_FW_NVDIMM_ARM                = 0x0010,
+
+        /// The following operation pertains to the Health of the NVDIMM
+        /// This operation can be performed with the arming/disarming
+        /// operation, these operation types are orthogonal to each other
+        // Manufacturing energy source(ES) health check request
+        HBRT_FW_MNFG_ES_HEALTH_CHECK      = 0x0020,
     };
 
     // NVDIMM (PHYP -> HBRT) message to request NVDIMM operation(s)
diff --git a/src/include/usr/isteps/nvdimm/nvdimm.H b/src/include/usr/isteps/nvdimm/nvdimm.H
index c07df0420..c1d6b334c 100644
--- a/src/include/usr/isteps/nvdimm/nvdimm.H
+++ b/src/include/usr/isteps/nvdimm/nvdimm.H
@@ -193,6 +193,39 @@ bool nvdimmArm(TARGETING::TargetHandleList &i_nvdimmTargetList);
  */
 bool nvdimmDisarm(TARGETING::TargetHandleList &i_nvdimmTargetList);
 
+/**
+ * @brief Check the health status of the individual NVDIMMs supplied in list
+ *
+ * @details The BPM will trigger the health check when power is applied at the
+ *          beginning of the IPL, with results ready to check about 20 mins
+ *          later. It is the caller's responsibility to ensure enough time has
+ *          passed to make this call.
+ *          Excerpt from the Jedec Standard, Byte Addressable Energy Backed
+ *          Interface of the interested flags (bits 0 .. 2).
+ *          ES_CMD_STATUS0
+ *              Bit 0 : Health Check in Progress
+ *              Bit 1 : Health Check Succeeded
+ *              Bit 2 : Health Check Failed
+ *
+ * @param[in] i_nvdimmTargetList - list of NVDIMMs to check the health of
+ *
+ * @return false if one or more NVDIMMs fail health check, else true
+ */
+bool nvDimmCheckHealthStatus(TARGETING::TargetHandleList &i_nvdimmTargetList);
+
+/**
+ * @brief A wrapper around the call to nvDimmCheckHealthStatus
+ *
+ * @details This will aggregate all the NVDIMMs of the system and pass
+ *          them to the call nvDimmCheckHealthStatus
+ *
+ * @see nvDimmCheckHealthStatus for more details
+ *
+ * @return false if one or more NVDIMMs fail health check, else true
+ */
+bool nvDimmCheckHealthStatusOnSystem();
+
+
 #endif
 
 /**
@@ -241,7 +274,6 @@ errlHndl_t notifyNvdimmProtectionChange(TARGETING::Target* i_target,
  * @param i_target     nvdimm target
  */
 void nvdimm_init(TARGETING::Target *i_nvdimm);
-
 }
 
 #endif  // NVDIMM_EXT_H__
diff --git a/src/include/usr/isteps/nvdimm/nvdimmreasoncodes.H b/src/include/usr/isteps/nvdimm/nvdimmreasoncodes.H
index 938361e54..8ed928a2d 100644
--- a/src/include/usr/isteps/nvdimm/nvdimmreasoncodes.H
+++ b/src/include/usr/isteps/nvdimm/nvdimmreasoncodes.H
@@ -98,6 +98,7 @@ enum nvdimmModuleId
     SET_ATTR_NVDIMM_ENCRYPTION_KEYS_FW      = 0x30,
     SEND_ATTR_NVDIMM_ARMED                  = 0x31,
     NVDIMM_FACTORY_RESET                    = 0x32,
+    NVDIMM_HEALTH_CHECK                     = 0x33,
 };
 
 /**
@@ -161,6 +162,10 @@ enum nvdimmReasonCode
     NVDIMM_ENCRYPTION_MAX_DARN_ERRORS       = NVDIMM_COMP_ID | 0x37,   // Darn random key gen reached max errors
     NVDIMM_ENCRYPTION_BAD_RANDOM_DATA       = NVDIMM_COMP_ID | 0x38,   // Generated key data not valid
     NVDIMM_CANNOT_MAKE_ATTRIBUTE            = NVDIMM_COMP_ID | 0x39,   // Cannot make Attribute
+    NVDIMM_HEALTH_CHECK_IN_PROGRESS_FAILURE = NVDIMM_COMP_ID | 0x3A,   // !< pertains to ES_CMD_STATUS0[0]; the health check in progress flag
+    NVDIMM_HEALTH_CHECK_REPORTED_FAILURE    = NVDIMM_COMP_ID | 0x3B,   // !< pertains to ES_CMD_STATUS0[2]; the health check reported a failure flag
+    NVDIMM_LIFETIME_MIN_REQ_NOT_MET         = NVDIMM_COMP_ID | 0x3C,   // !< pertains to ES_LIFETIME; BPM does not meet minimum requirement for a new BPM
+    NVDIMM_HEALTH_CHECK_NEVER_INITIATED     = NVDIMM_COMP_ID | 0x3D,   // !< A health check was never initiated at start of IPL
 };
 
 enum UserDetailsTypes