Ignore bad quad and nest DTS

Correct DTS readings to be 8 bit with msb signed

Change-Id: Ib5d74535abd17696e9b30d63ccd28273ef4ddc61
CQ: SW410935
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/50807
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Reviewed-by: Christopher J. Cain <cjcain@us.ibm.com>
Reviewed-by: Andres A. Lugo-Reyes <aalugore@us.ibm.com>
Reviewed-by: William A. Bryan <wilbryan@us.ibm.com>
Reviewed-by: Martha Broyles <mbroyles@us.ibm.com>

3 files changed, 71 insertions, 20 deletions

diff --git a/src/occ_405/amec/amec_sensors_core.c b/src/occ_405/amec/amec_sensors_core.c
index b723366..2dbeb15 100755
--- a/src/occ_405/amec/amec_sensors_core.c
+++ b/src/occ_405/amec/amec_sensors_core.c
@@ -76,6 +76,7 @@ void amec_update_proc_core_sensors(uint8_t i_core)
   CoreData  *l_core_data_ptr;
   uint32_t  l_temp32 = 0;
   uint16_t  l_core_temp = 0;
+  uint16_t  l_quad_temp = 0;
   uint16_t  l_temp16 = 0;
   uint16_t  l_core_util = 0;
   uint16_t  l_core_freq = 0;
@@ -183,9 +184,11 @@ void amec_update_proc_core_sensors(uint8_t i_core)
 
     // Determine "core" temperature that will be returned in the poll for fan control
     // If there is at least 1 core online within the same quad use the quad temp else use the nest
-    if(QUAD_ONLINE(l_quad))
+    // verify quad temp is valid (not zero) this may be 0 if there were no valid quad DTS
+    l_quad_temp = AMECSENSOR_ARRAY_PTR(TEMPQ0, l_quad)->sample;
+    if( (QUAD_ONLINE(l_quad)) && l_quad_temp )
     {
-       l_core_temp = AMECSENSOR_ARRAY_PTR(TEMPQ0, l_quad)->sample;
+       l_core_temp = l_quad_temp;
     }
     else
     {
@@ -265,8 +268,6 @@ void amec_calc_dts_sensors(CoreData * i_core_data_ptr, uint8_t i_core)
 {
 #define DTS_PER_CORE     2
 #define QUAD_DTS_PER_CORE     2
-#define DTS_INVALID_MASK 0x0C00
-
 
     uint32_t      l_coreTemp = 0;
     uint8_t       k = 0;
@@ -297,12 +298,13 @@ void amec_calc_dts_sensors(CoreData * i_core_data_ptr, uint8_t i_core)
             //Check validity
             if (i_core_data_ptr->dts.core[k].fields.valid)
             {
-                l_coreDts[k] = i_core_data_ptr->dts.core[k].fields.reading;
+                // temperature is only 8 bits of reading field
+                l_coreDts[k] = (i_core_data_ptr->dts.core[k].fields.reading & 0xFF);
                 l_coreDtsCnt++;
 
-                //Hardware bug workaround:  Temperatures reaching 0 degrees C
-                //can show up as negative numbers.  To fix this, we discount
-                //values that have the 2 MSB's set.
+                //Hardware bug workaround:  Module test will detect bad DTS and write coefficients
+                //to force a reading of 0 or negative to indicate the DTS is bad.
+                //Throw out any DTS that is bad
                 if(((l_coreDts[k] & DTS_INVALID_MASK) == DTS_INVALID_MASK) ||
                     (l_coreDts[k] == 0))
                 {
@@ -338,13 +340,20 @@ void amec_calc_dts_sensors(CoreData * i_core_data_ptr, uint8_t i_core)
         // DTS values regardless of weight.
         for (k = 0; k < QUAD_DTS_PER_CORE; k++)
         {
-            if (i_core_data_ptr->dts.cache[k].fields.valid)
+            // temperature is only 8 bits of reading field
+            l_quadDtsTemp = (i_core_data_ptr->dts.cache[k].fields.reading & 0xFF);
+
+            if( (i_core_data_ptr->dts.cache[k].fields.valid) &&
+                ((l_quadDtsTemp & DTS_INVALID_MASK) != DTS_INVALID_MASK) &&
+                (l_quadDtsTemp != 0) )
             {
-                l_quadDts[k] = i_core_data_ptr->dts.cache[k].fields.reading;
+                l_quadDts[k] = l_quadDtsTemp;
                 l_quadDtsCnt++;
             }
         }
 
+        l_quadDtsTemp = 0;
+
         if(l_quadDtsCnt)
         {
             if (G_data_cnfg->thrm_thresh.proc_quad_weight)
diff --git a/src/occ_405/proc/proc_data.c b/src/occ_405/proc/proc_data.c
index 660cefd..f6694ef 100755
--- a/src/occ_405/proc/proc_data.c
+++ b/src/occ_405/proc/proc_data.c
@@ -513,10 +513,16 @@ CoreData * proc_get_bulk_core_data_ptr( const uint8_t i_occ_core_id )
 // End Function Specification
 void task_nest_dts( task_t * i_task )
 {
+#define MAX_NUM_NEST_DTS     3
+
     errlHndl_t  l_err = NULL;  // Error handler
     int         l_rc = 0;      // Return code
     ipc_nest_dts_parms_t * l_parms = (ipc_nest_dts_parms_t*)(G_nest_dts_gpe_req.cmd_data);
     uint16_t    l_avg = 0;
+    uint16_t    l_nestDtsTemp = 0;
+    uint8_t     l_nestDtsCnt = 0;  // Number of valid Nest DTSs
+    uint8_t     k = 0;
+    bool        l_nestDtsValid = FALSE;
     static bool L_scheduled = FALSE;
     static bool L_idle_trace = FALSE;
     static bool L_incomplete_trace = FALSE;
@@ -544,15 +550,47 @@ void task_nest_dts( task_t * i_task )
             if ((ASYNC_REQUEST_STATE_COMPLETE == G_nest_dts_gpe_req.request.completion_state) &&
                 (0 == G_nest_dts_parms.error.error))
             {
-                if (l_parms->data.sensor0.fields.valid &&
-                    l_parms->data.sensor1.fields.valid &&
-                    l_parms->data.sensor2.fields.valid)
+                for (k = 0; k < MAX_NUM_NEST_DTS; k++)
+                {
+                    // check valid and temperature for current nest DTS being processed
+                    if(k == 0)
+                    {
+                        l_nestDtsValid = l_parms->data.sensor0.fields.valid;
+                        // temperature is only 8 bits of reading field
+                        l_nestDtsTemp = (l_parms->data.sensor0.fields.reading & 0xFF);
+                    }
+                    else if(k == 1)
+                    {
+                        l_nestDtsValid = l_parms->data.sensor1.fields.valid;
+                        // temperature is only 8 bits of reading field
+                        l_nestDtsTemp = (l_parms->data.sensor1.fields.reading & 0xFF);
+                    }
+                    else if(k == 2)
+                    {
+                        l_nestDtsValid = l_parms->data.sensor2.fields.valid;
+                        // temperature is only 8 bits of reading field
+                        l_nestDtsTemp = (l_parms->data.sensor2.fields.reading & 0xFF);
+                    }
+                    else // shouldn't happen
+                    {
+                        break;
+                    }
+
+                    //Hardware bug workaround:  Module test will detect bad DTS and write coefficients
+                    //to force a reading of 0 or negative to indicate the DTS is bad.
+                    //Ignore any DTS that is not valid or marked bad
+                    if( (l_nestDtsValid) && ( (l_nestDtsTemp & DTS_INVALID_MASK) != DTS_INVALID_MASK) &&
+                        (l_nestDtsTemp != 0) )
+                    {
+                        l_avg += l_nestDtsTemp;
+                        l_nestDtsCnt++;
+                    }
+                } //for loop
+
+                if(l_nestDtsCnt)
                 {
-                    // Calculate the average of the 3 Nest DTS temps
-                    l_avg = l_parms->data.sensor0.fields.reading +
-                            l_parms->data.sensor1.fields.reading +
-                            l_parms->data.sensor2.fields.reading;
-                    l_avg = l_avg / 3;
+                    // Calculate the average of the valid Nest DTS temps
+                    l_avg = l_avg / l_nestDtsCnt;
 
                     // Mark the data as valid and update sensor
                     G_nest_dts_data_valid = TRUE;
@@ -560,10 +598,10 @@ void task_nest_dts( task_t * i_task )
                 }
                 else
                 {
-                    // Mark the data as invalid
+                    // No valid nest DTS, Mark the data as invalid
                     G_nest_dts_data_valid = FALSE;
                 }
-            }
+            } // if request completed without error
             else
             {
                 // Async request not finished, mark data invalid
diff --git a/src/occ_405/proc/proc_data.h b/src/occ_405/proc/proc_data.h
index f8b7870..afabbb1 100755
--- a/src/occ_405/proc/proc_data.h
+++ b/src/occ_405/proc/proc_data.h
@@ -46,6 +46,10 @@
 
 #define THREADS_PER_CORE    4
 
+// A dts should be considered invalid if it is returning a negative reading the dts reading is
+// actually only 8 bits (not 12 as the structure has) the signed bit is msb of the 8 bit reading
+#define DTS_INVALID_MASK    0x80
+
 #define NUM_CORE_DATA_BUFF       7
 #define NUM_CORE_DATA_DOUBLE_BUF 2
 #define NUM_CORE_DATA_EMPTY_BUF  1