Fix issue where false errorlogs are reported.

Change-Id: Ifa5e6812001bf83de19b5b68d5b98a7acdc1dd06
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/17694
Tested-by: FSP CI Jenkins
Reviewed-by: Guillermo J. Silva <guilsilv@us.ibm.com>
Tested-by: Guillermo J. Silva <guilsilv@us.ibm.com>

3 files changed, 40 insertions, 3 deletions

diff --git a/src/occ/cmdh/cmdh_fsp_cmds.c b/src/occ/cmdh/cmdh_fsp_cmds.c
index d15decb..bb1f12a 100755
--- a/src/occ/cmdh/cmdh_fsp_cmds.c
+++ b/src/occ/cmdh/cmdh_fsp_cmds.c
@@ -167,6 +167,19 @@ errlHndl_t cmdh_tmgt_poll (const cmdh_fsp_cmd_t * i_cmd_ptr,
             l_poll_rsp->errl_address   = getErrlOCIAddrByID(l_poll_rsp->errl_id);
             l_poll_rsp->errl_length    = getErrlLengthByID(l_poll_rsp->errl_id);
 
+            //If errl_id is not 0, then neither address or length should be zero.
+            //This should not happen, but if it does tmgt will create an error log that
+            //includes the data at the errl slot address given.
+            //NOTE: One cause for a false errlog id is corruption of data in one errl slot
+            //      due to writing data greater than the size of the previous slot.  For
+            //      example writing the CallHome errorlog (3kb) into a regular sized (2kb) slot.
+            if ( (l_poll_rsp->errl_id != 0) &&
+                 ((l_poll_rsp->errl_address == 0) || (l_poll_rsp->errl_length == 0)))
+            {
+                TRAC_ERR("An error ID has been sent via poll but the address or size is 0. "
+                         "ErrlId:0x%X, sz:0x%X, address:0x%X.",
+                         l_poll_rsp->errl_id, l_poll_rsp->errl_length, l_poll_rsp->errl_address);
+            }
             l_poll_rsp->data_length[0] = CONVERT_UINT16_UINT8_HIGH(CMDH_POLL_RESP_LEN_V0);
             l_poll_rsp->data_length[1] = CONVERT_UINT16_UINT8_LOW(CMDH_POLL_RESP_LEN_V0);
             l_rc                       = ERRL_RC_SUCCESS;
@@ -248,6 +261,22 @@ ERRL_RC cmdh_poll_v10(cmdh_fsp_rsp_t * o_rsp_ptr)
     l_poll_rsp->errl_address    = getErrlOCIAddrByID(l_poll_rsp->errl_id);
     // Byte 13 - 14:
     l_poll_rsp->errl_length     = getErrlLengthByID(l_poll_rsp->errl_id);
+
+            //If errl_id is not 0, then neither address or length should be zero.
+            //This should not happen, but if it does tmgt will create an error log that
+            //includes the data at the errl slot address given that can be used for debug.
+            //NOTE: One cause for a false errlog id is corruption of data in one errl slot
+            //      due to writing data greater than the size of the previous slot.  For
+            //      example writing the CallHome errorlog (3kb) into a regular sized (2kb) slot.
+            //      Make sure to verify the order of the memory allocation for the errl slots.
+    if ( (l_poll_rsp->errl_id != 0) &&
+         ((l_poll_rsp->errl_address == 0) || (l_poll_rsp->errl_length == 0)))
+    {
+        TRAC_ERR("An error ID has been sent via poll but the address or size is 0. "
+                 "ErrlId:0x%X, sz:0x%X, address:0x%X.",
+                 l_poll_rsp->errl_id, l_poll_rsp->errl_length, l_poll_rsp->errl_address);
+    }
+
     // Byte 15 - 16: reserved.
     // Byte 17 - 32 (16 bytes): OCC level
     memcpy( (void *) l_poll_rsp->occ_level, (void *) &G_occ_buildname[0], 16);
@@ -260,6 +289,7 @@ ERRL_RC cmdh_poll_v10(cmdh_fsp_rsp_t * o_rsp_ptr)
     // Byte 40:
     l_poll_rsp->sensor_dblock_version = 0x01;  //Currently only 0x01 is supported.
 
+    //l_rsp_index is used as an index into o_rsp_ptr
     uint16_t l_rsp_index = CMDH_POLL_RESP_LEN_V10;
 
     ////////////////////
diff --git a/src/occ/errl/errl.c b/src/occ/errl/errl.c
index ca0a279..cd75621 100755
--- a/src/occ/errl/errl.c
+++ b/src/occ/errl/errl.c
@@ -79,6 +79,8 @@ uint8_t getErrSlotNumAndErrId(
     uint8_t     l_rc = ERRL_INVALID_SLOT;
     uint32_t    l_mask = ERRL_SLOT_MASK_DEFAULT;
 
+    //Use severity to determine what slots are available for the given
+    //type of errorlog severity.
     switch ( i_severity )
     {
         case ERRL_SEV_INFORMATIONAL:
@@ -100,16 +102,18 @@ uint8_t getErrSlotNumAndErrId(
     if ( l_mask != ERRL_SLOT_MASK_DEFAULT )
     {
         // 1.  Find an available slot
+        //     l_slotBitWord represents the available slots given the severity type.
         uint8_t             l_slot = ERRL_INVALID_SLOT;
         uint32_t            l_slotBitWord = ~(G_occErrSlotBits | l_mask);
         SsxMachineContext   l_ctx;
 
-        // 2. use assembly cntlzw to get slot & (disable/enable interrupts)
+        // 2. use assembly cntlzw (count leading zeros) to get available slot based on
+        // severity type, and (disable/enable interrupts)
         ssx_critical_section_enter(SSX_NONCRITICAL, &l_ctx);
         __asm__ __volatile__ ( "cntlzw %0, %1;" : "=r" (l_slot) : "r" (l_slotBitWord));
         ssx_critical_section_exit(&l_ctx);
 
-        // slot is valid
+        // A slot is available and valid
         if ( l_slot < ERRL_MAX_SLOTS )
         {
             ssx_critical_section_enter(SSX_NONCRITICAL, &l_ctx);
@@ -118,8 +122,11 @@ uint8_t getErrSlotNumAndErrId(
             *o_timeStamp = ssx_timebase_get();
             // save of counter and then increment it
             // Note: Internal caller so assuming valid pointer
+
+            //Provide next ErrorId; G_occErrIdCounter should never be 0.
             *o_errlId = ((++G_occErrIdCounter) == 0) ? ++G_occErrIdCounter : G_occErrIdCounter;
 
+            //Set slot bit in list of used up slots.
             G_occErrSlotBits |= (ERRL_SLOT_SHIFT >> l_slot);
             ssx_critical_section_exit(&l_ctx);
 
diff --git a/src/occ/thread/chom.c b/src/occ/thread/chom.c
index 9c8f85e..b75bda8 100755
--- a/src/occ/thread/chom.c
+++ b/src/occ/thread/chom.c
@@ -380,7 +380,7 @@ void chom_gen_periodic_log()
             THRD_MID_GEN_CALLHOME_LOG,    // modId
             GEN_CALLHOME_LOG,             // reasoncode
             OCC_NO_EXTENDED_RC,           // Extended reason code
-            ERRL_SEV_INFORMATIONAL,       // Severity
+            ERRL_SEV_CALLHOME_DATA,       // Severity; will be changed to informational in errl.c
             NULL,                         // Trace Buf
             0,                            // Trace Size
             0,                            // userdata1