Implement an HBRT interface to log a gard event from PHYP/OPAL

Added a firmware notify interface, gard_event_t, to accept a
gard event message, from PHYP/OPAL, to log that event.

Change-Id: I9bcf684f0850c9a07ab7d46635aa07a2c1e9917c
RTC: 210201
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/82199
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Reviewed-by: Matt Derksen <mderkse1@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: Christian R Geddes <crgeddes@us.ibm.com>
Reviewed-by: Daniel M Crowell <dcrowell@us.ibm.com>

4 files changed, 191 insertions, 14 deletions

diff --git a/src/include/runtime/README.md b/src/include/runtime/README.md
index 8a4e7298a..68b7382be 100755
--- a/src/include/runtime/README.md
+++ b/src/include/runtime/README.md
@@ -2,7 +2,7 @@
 How to create an HBRT to FW request message interface
     0) If passing an HBRT to FSP via MBOX or receiving a firmware notify message,
        then use instruction 'generic_hbrt_fsp_message.H::GenericFspMboxMessage_t'
-       below.
+       and/or 'How to create an HBRT Firmware Notify message' below.
     1) The biggest part will be defining the interface.  Inspect the current
        interfaces (req_hcode_update, error_log, etc) for inspiration.
     2) Once an interface has been designed, add it to the anonymous
@@ -35,13 +35,14 @@ How to create an HBRT to FW request message interface
         1) The biggest part will be defining the interface.  Inspect the current
            interfaces (AttributeSetter_t, SingleScomOpHbrtFspData_t,
            TargetDeconfigHbrtFspData_t, etc) for inspiration.
-        2) Once an interface has been designed, add the structure to this file
-           with the other interfaces.
+        2) Once an interface has been designed, add the structure to the file,
+           generic_hbrt_fsp_message.H, among the other interfaces.
         3) Create an MBOX message queue enum for the interface and add to:
            /hostboot/src/include/usr/mbox/mbox_queues.H::queue_id_t
            see current message queues for example
         4) Add a new message type for the interface to:
-           enum generic_hbrt_fsp_message.H::GENERIC_FSP_MBOX_MESSAGE_MSG_TYPE.
+           enum GenericFspMboxMessage_t::GENERIC_FSP_MBOX_MESSAGE_MSG_TYPE in
+           file generic_hbrt_fsp_message.H.
         5) How to use the new interface to pass a message
            a) Make sure g_hostInterfaces and g_hostInterfaces->firmware_request
               are not NULL.
diff --git a/src/include/runtime/interface.h b/src/include/runtime/interface.h
index 9a0cb478e..152226d54 100644
--- a/src/include/runtime/interface.h
+++ b/src/include/runtime/interface.h
@@ -552,17 +552,18 @@ typedef struct hostInterfaces
 
     enum  // hbrt_fw_msg::io_type             the struct associated with io_type
     {
-       HBRT_FW_MSG_TYPE_REQ_NOP = 0,
-       HBRT_FW_MSG_TYPE_RESP_NOP = 1,           // struct resp_generic
-       HBRT_FW_MSG_TYPE_RESP_GENERIC = 2,       // struct resp_generic
-       HBRT_FW_MSG_TYPE_REQ_HCODE_UPDATE = 3,   // struct req_hcode_update
-       HBRT_FW_MSG_HBRT_FSP_REQ = 4,            // struct GenericFspMboxMessage_t
-       HBRT_FW_MSG_TYPE_ERROR_LOG = 5,          // struct error_log
-       HBRT_FW_MSG_HBRT_FSP_RESP = 6,           // struct GenericFspMboxMessage_t
-       HBRT_FW_MSG_TYPE_I2C_LOCK = 7,           // struct req_i2c_lock
-       HBRT_FW_MSG_TYPE_SBE_STATE = 8,          // struct sbe_state
-       HBRT_FW_MSG_TYPE_NVDIMM_PROTECTION = 9,  // struct nvdimm_protection_state
+       HBRT_FW_MSG_TYPE_REQ_NOP           =  0,
+       HBRT_FW_MSG_TYPE_RESP_NOP          =  1, // struct resp_generic
+       HBRT_FW_MSG_TYPE_RESP_GENERIC      =  2, // struct resp_generic
+       HBRT_FW_MSG_TYPE_REQ_HCODE_UPDATE  =  3, // struct req_hcode_update
+       HBRT_FW_MSG_HBRT_FSP_REQ           =  4, // struct GenericFspMboxMessage_t
+       HBRT_FW_MSG_TYPE_ERROR_LOG         =  5, // struct error_log
+       HBRT_FW_MSG_HBRT_FSP_RESP          =  6, // struct GenericFspMboxMessage_t
+       HBRT_FW_MSG_TYPE_I2C_LOCK          =  7, // struct req_i2c_lock
+       HBRT_FW_MSG_TYPE_SBE_STATE         =  8, // struct sbe_state
+       HBRT_FW_MSG_TYPE_NVDIMM_PROTECTION =  9, // struct nvdimm_protection_state
        HBRT_FW_MSG_TYPE_NVDIMM_OPERATION  = 10, // struct nvdimm_operation_t
+       HBRT_FW_MSG_TYPE_GARD_EVENT        = 11, // struct gard_event_t
     };
 
     // NVDIMM protection state enum
@@ -611,6 +612,39 @@ typedef struct hostInterfaces
                              // see @note associated with NVDIMM_Op_t above
     } __attribute__ ((packed));
 
+    // Gard event error type
+    // @note This needs to stay in sync with the FSP Mailbox specification for
+    //       command : Gard-able Error Detected - cmd 0xCE, s/c 0x63, mod 01
+    enum GARD_ERROR_t: uint32_t
+    {
+        HBRT_GARD_ERROR_UNKNOWN                  = 0x0000,
+        HBRT_GARD_ERROR_COMPUTATION_TEST_FAILURE = 0x0001,
+        HBRT_GARD_ERROR_SLB                      = 0x0002,
+        HBRT_GARD_ERROR_CHIP_TOD_FAILURE         = 0x0003,
+        HBRT_GARD_ERROR_TIMEFAC_FAILURE          = 0x0004,
+        HBRT_GARD_ERROR_PROC_RECOVERY_THRESHOLD  = 0x0005,
+        HBRT_GARD_ERROR_NX                       = 0x0008,
+        HBRT_GARD_ERROR_SLW                      = 0x0009,
+        HBRT_GARD_ERROR_CAPP_UNIT                = 0x000A,
+
+        // Mark the end of the gard error types.
+        // This is not valid, just a marker
+        HBRT_GARD_ERROR_LAST,
+    };
+
+    // Gard event (PHYP/OPAL -> HBRT)
+    struct gard_event_t
+    {
+        GARD_ERROR_t i_error_type;     // Gard event error type enum
+        uint32_t     i_procId;         // Processor ID for
+                                       // error types 0x0001 to 0x0005
+                                       // Chip ID for
+                                       // error types 0x0008 to 0x000A
+        uint32_t     i_plid;           // Platform log identifier
+        uint16_t     i_sub_unit_mask;  // Currently not being used
+        uint16_t     i_recovery_level; // Currently not being used
+    } __attribute__ ((packed));
+
     struct hbrt_fw_msg   // define struct hbrt_fw_msg
     {
        hbrt_fw_msg() { req_hcode_update = { 0 }; };  // ctor
@@ -685,6 +719,10 @@ typedef struct hostInterfaces
           // io_type set to HBRT_FW_MSG_TYPE_NVDIMM_OPERATION
           struct nvdimm_operation_t nvdimm_operation;
 
+          // This struct is sent from PHYP/OPAL to HBRT with
+          // io_type set to HBRT_FW_MSG_TYPE_GARD_EVENT
+          struct gard_event_t gard_event;
+
           // This struct is sent from HBRT with
           // io_type set to HBRT_FW_MSG_HBRT_FSP_REQ or
           // HBRT_FW_MSG_HBRT_FSP_RESP
diff --git a/src/include/usr/runtime/runtime_reasoncodes.H b/src/include/usr/runtime/runtime_reasoncodes.H
index 5cfdce5c2..b54f29751 100644
--- a/src/include/usr/runtime/runtime_reasoncodes.H
+++ b/src/include/usr/runtime/runtime_reasoncodes.H
@@ -147,6 +147,8 @@ namespace RUNTIME
         RC_NO_SPACE_FOR_ATTRIBUTE_SERIALIZATION = RUNTIME_COMP_ID | 0x47,
         RC_CANNOT_MAKE_ATTRIBUTE       = RUNTIME_COMP_ID | 0x48,
         RT_NO_OMI_TARGET_FOUND         = RUNTIME_COMP_ID | 0x49,
+        RC_LOG_GARD_EVENT_UNKNOWN_ERROR_TYPE   = RUNTIME_COMP_ID | 0x4A,
+        RC_LOG_GARD_EVENT              = RUNTIME_COMP_ID | 0x4B,
     };
 
     enum UserDetailsTypes
diff --git a/src/usr/util/runtime/rt_fwnotify.C b/src/usr/util/runtime/rt_fwnotify.C
index a96b24bba..e9ebabe6d 100644
--- a/src/usr/util/runtime/rt_fwnotify.C
+++ b/src/usr/util/runtime/rt_fwnotify.C
@@ -653,6 +653,117 @@ int doNvDimmOperation(const hostInterfaces::nvdimm_operation_t& i_nvDimmOp)
 }
 
 /**
+ *  @brief Log the gard event from PHYP/OPAL
+ *
+ *  @param[in] i_gardEvent - The details of the gard event
+ *                           @see hostInterfaces::gard_event_t for more info
+ *
+ **/
+void logGardEvent(const hostInterfaces::gard_event_t& i_gardEvent)
+{
+    // Trace input components
+    TRACFCOMP(g_trac_runtime,
+              ENTER_MRK"logGardEvent: Gard Event Data: "
+                       "error type(0x%.8X), processor ID(0x%.8X), "
+                       "PLID(0x%.8X), sub unit mask(0x.%4X), "
+                       "recovery level(0x.%4X)",
+                       i_gardEvent.i_error_type,
+                       i_gardEvent.i_procId,
+                       i_gardEvent.i_plid,
+                       i_gardEvent.i_sub_unit_mask,
+                       i_gardEvent.i_recovery_level);
+
+    errlHndl_t l_err{nullptr};
+
+    do
+    {
+        // Make sure the error type is valid, if not, log it
+        if ((i_gardEvent.i_error_type == hostInterfaces::HBRT_GARD_ERROR_UNKNOWN )   ||
+            (i_gardEvent.i_error_type >= hostInterfaces::HBRT_GARD_ERROR_LAST) )
+        {
+            TRACFCOMP(g_trac_runtime, "logGardEvent: ERROR: unknown/invalid "
+                                      "error type 0x%.8X",
+                                      i_gardEvent.i_error_type);
+
+            /* @
+             * @errortype
+             * @severity         ERRL_SEV_PREDICTIVE
+             * @moduleid         MOD_RT_FIRMWARE_NOTIFY
+             * @reasoncode       RC_LOG_GARD_EVENT_UNKNOWN_ERROR_TYPE
+             * @userdata1[0:31]  GARD error type
+             * @userdata1[32:63] Processor ID
+             * @userdata2[0:31]  Sub unit mask
+             * @userdata2[32:63] Recovery level
+             * @devdesc          Unknown/invalid error type
+             * @custdesc         Internal firmware error
+             */
+            l_err = new ErrlEntry( ERRL_SEV_PREDICTIVE,
+                                   MOD_RT_FIRMWARE_NOTIFY,
+                                   RC_LOG_GARD_EVENT_UNKNOWN_ERROR_TYPE,
+                                   TWO_UINT32_TO_UINT64(
+                                        i_gardEvent.i_error_type,
+                                        i_gardEvent.i_procId),
+                                   TWO_UINT32_TO_UINT64(
+                                        i_gardEvent.i_sub_unit_mask,
+                                        i_gardEvent.i_recovery_level),
+                                   ErrlEntry::ADD_SW_CALLOUT);
+            break;
+        }
+
+
+        // Get the Target associated with processor ID
+        TARGETING::TargetHandle_t l_procTarget{nullptr};
+        l_err = RT_TARG::getHbTarget(i_gardEvent.i_procId, l_procTarget);
+        if (l_err)
+        {
+            TRACFCOMP(g_trac_runtime, "logGardEvent: Error getting "
+                                      "HB Target from processor ID 0x%0X, "
+                                      "exiting ...",
+                                      i_gardEvent.i_procId);
+            break;
+        }
+
+        // Log the GARD event
+        /* @
+         * @errortype
+         * @severity         ERRL_SEV_PREDICTIVE
+         * @moduleid         MOD_RT_FIRMWARE_NOTIFY
+         * @reasoncode       RC_LOG_GARD_EVENT
+         * @userdata1[0:31]  GARD error type
+         * @userdata1[32:63] Processor ID
+         * @userdata2[0:31]  Sub unit mask
+         * @userdata2[32:63] Recovery level
+         * @devdesc          Gard event from Opal/Phyp
+         * @custdesc         Hardware error detected at runtime
+         */
+        l_err = new ErrlEntry( ERRL_SEV_PREDICTIVE,
+                               MOD_RT_FIRMWARE_NOTIFY,
+                               RC_LOG_GARD_EVENT,
+                               TWO_UINT32_TO_UINT64(
+                                    i_gardEvent.i_error_type,
+                                    i_gardEvent.i_procId),
+                               TWO_UINT32_TO_UINT64(
+                                    i_gardEvent.i_sub_unit_mask,
+                                    i_gardEvent.i_recovery_level));
+
+        // Set the PLID to the given gard event PLID if it exist
+        if (i_gardEvent.i_plid)
+        {
+            l_err->plid(i_gardEvent.i_plid);
+        }
+
+        // Do the actual gard
+        l_err->addHwCallout( l_procTarget, HWAS::SRCI_PRIORITY_MED,
+                             HWAS::NO_DECONFIG, HWAS::GARD_PHYP);
+    } while(0);
+
+    // Commit any error log that occurred.
+    errlCommit(l_err, RUNTIME_COMP_ID);
+
+    TRACFCOMP(g_trac_runtime, EXIT_MRK"logGardEvent")
+}
+
+/**
  * @see  src/include/runtime/interface.h for definition of call
  *
  */
@@ -764,6 +875,31 @@ void firmware_notify( uint64_t i_len, void *i_data )
             } // END case hostInterfaces::HBRT_FW_MSG_TYPE_NVDIMM_OPERATION:
             break;
 
+            case hostInterfaces::HBRT_FW_MSG_TYPE_GARD_EVENT:
+            {
+                uint64_t l_minMsgSize = hostInterfaces::HBRT_FW_MSG_BASE_SIZE +
+                                sizeof(hostInterfaces::hbrt_fw_msg::gard_event);
+                if (i_len < l_minMsgSize)
+                {
+                    l_badMessage = true;
+
+                    TRACFCOMP(g_trac_runtime, ERR_MRK"firmware_notify: "
+                     "Received message HBRT_FW_MSG_TYPE_GARD_EVENT, "
+                     "but size of message data(%d) is not adequate for a "
+                     "complete message of this type, with size requirement of "
+                     "%d", i_len, l_minMsgSize );
+
+                    // Pack user data 1 with the message input type, the only
+                    // data that can be safely retrieved
+                    l_userData1 = l_hbrt_fw_msg->io_type;
+
+                    break;
+                }
+
+                logGardEvent(l_hbrt_fw_msg->gard_event);
+            } // END case hostInterfaces::HBRT_FW_MSG_TYPE_GARD_EVENT:
+            break;
+
             default:
             {
                 l_badMessage = true;