Initial 405 GPU support

Change-Id: I6e957ca1aa643d257274e99957df5b15ac8c889b
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/44254
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Reviewed-by: William A. Bryan <wilbryan@us.ibm.com>

2 files changed, 56 insertions, 5 deletions

diff --git a/src/occ_405/amec/amec_sensors_power.c b/src/occ_405/amec/amec_sensors_power.c
index 2233738..3820330 100755
--- a/src/occ_405/amec/amec_sensors_power.c
+++ b/src/occ_405/amec/amec_sensors_power.c
@@ -69,6 +69,7 @@ uint32_t G_curr_num_gpus_sys = 0;
 #define ADC_CONVERTED_VALUE(i_chan) \
     ((i_chan < MAX_APSS_ADC_CHANNELS) ? G_lastValidAdcValue[i_chan] : 0)
 
+extern bool    G_gpu_monitoring_allowed;
 extern uint8_t G_occ_interrupt_type;
 extern bool    G_vrm_thermal_monitoring;
 extern PWR_READING_TYPE  G_pwr_reading_type;
@@ -821,6 +822,11 @@ void amec_update_gpu_configuration(void)
         {
             G_gpu_config_done = TRUE;
             G_first_proc_gpu_config = l_valid_bitmask_proc;
+            if(G_first_proc_gpu_config)
+            {
+               // GPUs are present enable monitoring
+               G_gpu_monitoring_allowed = TRUE;
+            }
             G_first_sys_gpu_config = l_valid_bitmask_sys;
             G_first_num_gpus_sys = l_num_gpus_sys;
             TRAC_IMP("GPU presence detection completed. GPU configuration for this OCC: 0x%08X, total[%d]",
diff --git a/src/occ_405/amec/amec_sys.h b/src/occ_405/amec/amec_sys.h
index 74b2812..a45fb42 100755
--- a/src/occ_405/amec/amec_sys.h
+++ b/src/occ_405/amec/amec_sys.h
@@ -430,6 +430,53 @@ typedef struct
 } amec_quad_t;
 
 //-------------------------------------------------------------
+// GPU Structures
+//-------------------------------------------------------------
+
+typedef struct {
+    bool     disabled;  // GPU has been marked failed and no longer monitored
+    bool     readOnce;  // Comm has been established with GPU
+    bool     overtempError;  // Core OT error has been logged against GPU
+    bool     memOvertempError;  // Memory OT error has been logged against GPU
+    bool     checkDriverLoaded;   // Indicates if need to check if driver is loaded
+    bool     driverLoaded;   // Indicates if GPU driver is loaded
+    bool     checkMemTempSupport; // Indicates if need to check if mem monitoring is supported
+    bool     memTempSupported; // Indicates if memory temperature monitoring is supported
+    uint8_t  memErrorCount; // count of consecutive GPU mem temp read failures
+    uint8_t  errorCount;   // count of consecutive GPU core temp read failures
+} gpuStatus_t;
+
+typedef struct {
+    bool     check_pwr_limit; // Indicates if need to read power limits from GPU
+    bool     pwr_limits_read;   // Indicates if power limits were read i.e. have min/max
+    uint32_t gpu_min_pcap_mw; // Min GPU power limit in mW read from the GPU
+    uint32_t gpu_max_pcap_mw; // Max GPU power limit in mW read from the GPU
+    uint32_t gpu_desired_pcap_mw;  // AMEC determined pcap in mW to set
+    uint32_t gpu_requested_pcap_mw;  // Requested power cap in mW sent to GPU
+    uint32_t gpu_actual_pcap_mw;  // Actual power cap in mW read back from the GPU
+} gpuPcap_t;
+
+
+typedef struct
+{
+  //-----------------------------------
+  // Sensors
+  //-----------------------------------
+  sensor_t tempgpu;    // GPU core temperature
+  sensor_t tempgpumem; // GPU HBM temperature
+
+  //-----------------------------------
+  // Data
+  //-----------------------------------
+  // General Status of GPU
+  gpuStatus_t status;
+
+  // GPU Power Cap Information
+  gpuPcap_t pcap;
+
+} amec_gpu_t;
+
+//-------------------------------------------------------------
 // Proc Structure
 //-------------------------------------------------------------
 typedef struct
@@ -468,11 +515,6 @@ typedef struct
   // Nimbus DIMM Sensors
   sensor_t       tempdimm[NUM_DIMM_PORTS*NUM_DIMMS_PER_I2CPORT];
 
- // GPU Sensors
-  sensor_t tempgpu0;
-  sensor_t tempgpu1;
-  sensor_t tempgpu2;
-
   sensor_t curvdn;
   sensor_t pwrvdd;
   sensor_t pwrvdn;
@@ -607,6 +649,9 @@ typedef struct
   //     in the hopes of perhaps reusing some code from previous projects.
   amec_proc_t   proc[NUM_PROC_CHIPS_PER_OCC];
 
+  // GPU Data
+  amec_gpu_t   gpu[MAX_NUM_GPU_PER_DOMAIN];
+
   // OCC Firmware Data
   amec_fw_t     fw;