HTMGT: Cache user power limit from BMC and add proc callout for 2AEx errors

Change-Id: Ib8918b03bfadd2948fe98e37b7e695c6df2ac7cb
CQ: SW414473
CQ: SW413111
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/55897
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: Sheldon Bailey <baileysh@us.ibm.com>
Reviewed-by: Martha Broyles <mbroyles@us.ibm.com>
Reviewed-by: Daniel M. Crowell <dcrowell@us.ibm.com>

4 files changed, 80 insertions, 14 deletions

diff --git a/src/usr/htmgt/htmgt_activate.C b/src/usr/htmgt/htmgt_activate.C
index b8dd0e862..8e38899d8 100644
--- a/src/usr/htmgt/htmgt_activate.C
+++ b/src/usr/htmgt/htmgt_activate.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2014,2017                        */
+/* Contributors Listed Below - COPYRIGHT 2014,2018                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -51,6 +51,7 @@ namespace HTMGT
     uint8_t * G_simicsHomerBuffer = NULL;
 #endif
 
+#define BMC_LIMIT_WAS_READ 0xFF000000
 
     // Wait for all OCCs to reach ready state
     errlHndl_t waitForOccReady()
@@ -199,15 +200,39 @@ namespace HTMGT
             err = SENSOR::getUserPowerLimit(limit, active);
             if (err)
             {
-                TMGT_ERR("sendOccUserPowerCap: Error getting user "
-                         "power limit");
-                break;
+                const uint32_t saved_power_limit =
+                    sys->getAttr<ATTR_HTMGT_SAVED_POWER_LIMIT>();
+                if (saved_power_limit & BMC_LIMIT_WAS_READ)
+                {
+                    // Attribute has been written since the power on,
+                    //     use the saved limit/active.
+                    limit = saved_power_limit & 0x0000FFFF;
+                    active = (saved_power_limit >> 16) & 0x00FF;
+                    TMGT_INF("SENSOR::getUserPowerLimit failed with rc=0x%04X. "
+                             "Using prior values: limit %dW, active: %c",
+                             err->reasonCode(), limit, active?'Y':'N');
+                }
+                else
+                {
+                    TMGT_ERR("sendOccUserPowerCap: Error getting user "
+                             "power limit from BMC, rc=0x%04X",
+                             err->reasonCode());
+                    break;
+                }
+
+            }
+            else
+            {
+                // Write attribute with power limit and state
+                TMGT_INF("SENSOR::getUserPowerLimit returned %dW, active: %c",
+                         limit, active?'Y':'N');
+                const uint32_t saved_limit = BMC_LIMIT_WAS_READ |
+                    ((active?0x01:0x00) << 16) | limit;
+                sys->setAttr<TARGETING::ATTR_HTMGT_SAVED_POWER_LIMIT>
+                    (saved_limit);
             }
 #endif
 
-            TMGT_INF("SENSOR::getUserPowerLimit returned %d, active = %d",
-                     limit, active);
-
             if (active)
             {
                 //Make sure this value is between the min & max allowed
@@ -216,18 +241,24 @@ namespace HTMGT
                 max = getMaxPowerCap(sys, is_redundant);
                 if ((limit != 0) && (limit < min))
                 {
-                    TMGT_INF("sendOccUserPowerCap:  User power cap %d is below"
+                    TMGT_INF("sendOccUserPowerCap:  User power cap %dW is below"
                              " the minimum of %d, clipping value",
                              limit, min);
                     limit = min;
                 }
                 else if (limit > max)
                 {
-                    TMGT_INF("sendOccUserPowerCap:  User power cap %d is above"
+                    TMGT_INF("sendOccUserPowerCap:  User power cap %dW is above"
                              " the maximum of %d, clipping value",
                              limit, min);
                     limit = max;
                 }
+                else if (limit == 0)
+                {
+                    TMGT_ERR("sendOccUserPowerCap: BMC is reporting that user "
+                             "cap is enabled, but the value is 0W!");
+                    active = false;
+                }
             }
             else
             {
@@ -243,7 +274,7 @@ namespace HTMGT
                 data[0] = limit >> 8;
                 data[1] = limit & 0xFF;
 
-                TMGT_INF("sendOccUserPowerCap:  Sending power cap %d to OCC %d",
+                TMGT_INF("sendOccUserPowerCap: Sending power cap %dW to OCC %d",
                          limit, occ->getInstance());
                 if (limit > 0)
                 {
@@ -257,7 +288,7 @@ namespace HTMGT
                 if (err)
                 {
                     TMGT_ERR("sendOccUserPowerCap: Failed sending command "
-                             "to OCC %d with rc = 0x%04X",
+                             "to OCC%d with rc=0x%04X",
                              occ->getInstance(), err->reasonCode());
                     break;
                 }
diff --git a/src/usr/htmgt/htmgt_occcmd.C b/src/usr/htmgt/htmgt_occcmd.C
index 05b5b1657..fbb380535 100644
--- a/src/usr/htmgt/htmgt_occcmd.C
+++ b/src/usr/htmgt/htmgt_occcmd.C
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2014,2017                        */
+/* Contributors Listed Below - COPYRIGHT 2014,2018                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -858,6 +858,26 @@ namespace HTMGT
                         iv_Occ->collectCheckpointScomData(l_excErr);
                     }
 
+                    // Add processor callout
+                    TARGETING::ConstTargetHandle_t procTarget =
+                        TARGETING::getParentChip(iv_Occ->getTarget());
+                    if (nullptr != procTarget)
+                    {
+                        const unsigned long huid =
+                            procTarget->getAttr<TARGETING::ATTR_HUID>();
+                        TMGT_ERR("handleOccException: Adding processor "
+                                 "callout (HUID=0x%0lX)", huid);
+                        l_excErr->addHwCallout(procTarget,
+                                               HWAS::SRCI_PRIORITY_MED,
+                                               HWAS::NO_DECONFIG,
+                                               HWAS::GARD_NULL);
+                    }
+                    else
+                    {
+                        TMGT_ERR("handleOccException: Unable to determine "
+                                 "parent chip to add callout");
+                    }
+
                     // Add OCC trace buffer to error log (ERR, IMP, INF)
                     iv_Occ->addOccTrace( l_excErr );
 
diff --git a/src/usr/targeting/common/xmltohb/attribute_types_openpower.xml b/src/usr/targeting/common/xmltohb/attribute_types_openpower.xml
index d7f25baec..c49d2f75a 100644
--- a/src/usr/targeting/common/xmltohb/attribute_types_openpower.xml
+++ b/src/usr/targeting/common/xmltohb/attribute_types_openpower.xml
@@ -5,7 +5,7 @@
 <!--                                                                        -->
 <!-- OpenPOWER HostBoot Project                                             -->
 <!--                                                                        -->
-<!-- Contributors Listed Below - COPYRIGHT 2016,2017                        -->
+<!-- Contributors Listed Below - COPYRIGHT 2016,2018                        -->
 <!-- [+] International Business Machines Corp.                              -->
 <!--                                                                        -->
 <!--                                                                        -->
@@ -1323,6 +1323,20 @@ ID for the sensor number returned with the elog. -->
     <readable/>
 </attribute>
 
+<attribute>
+    <id>HTMGT_SAVED_POWER_LIMIT</id>
+    <description>User power limit information from the BMC to be used if
+        communication with the BMC is lost after system is powered on.
+        Used internally by HTMGT.
+    </description>
+    <simpleType>
+        <uint32_t></uint32_t>
+    </simpleType>
+    <persistency>volatile-zeroed</persistency>
+    <readable/>
+    <writeable/>
+</attribute>
+
 <!-- end HTMGT attributes -->
 
 <attribute>
diff --git a/src/usr/targeting/common/xmltohb/target_types_openpower.xml b/src/usr/targeting/common/xmltohb/target_types_openpower.xml
index aaf06a358..83217b2da 100644
--- a/src/usr/targeting/common/xmltohb/target_types_openpower.xml
+++ b/src/usr/targeting/common/xmltohb/target_types_openpower.xml
@@ -5,7 +5,7 @@
 <!--                                                                        -->
 <!-- OpenPOWER HostBoot Project                                             -->
 <!--                                                                        -->
-<!-- Contributors Listed Below - COPYRIGHT 2016,2017                        -->
+<!-- Contributors Listed Below - COPYRIGHT 2016,2018                        -->
 <!-- [+] International Business Machines Corp.                              -->
 <!--                                                                        -->
 <!--                                                                        -->
@@ -102,6 +102,7 @@
     <attribute><id>PROC_MHZ_PER_WATT</id></attribute>
     <attribute><id>ALLOW_CALLHOME_ESELS_TO_BMC</id></attribute>
     <attribute><id>REPORT_THROTTLE_BELOW_NOMINAL</id></attribute>
+    <attribute><id>HTMGT_SAVED_POWER_LIMIT</id></attribute>
 </targetTypeExtension>
 
 <targetTypeExtension>