PM: Fixes CME Registration Protocol Hang during PM_SUSPEND

    Key_Cronus_Test=PM_REGRESS

Change-Id: I64acf1e08f3e8544c020d50d5677fe7ef036ab46
CQ: SW419088
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/54934
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Tested-by: Cronus HW CI <cronushw-ci+hostboot@us.ibm.com>
Reviewed-by: RANGANATHPRASAD G. BRAHMASAMUDRA <prasadbgr@in.ibm.com>
Reviewed-by: Gregory S. Still <stillgs@us.ibm.com>
Reviewed-by: Brian T. Vanderpool <vanderp@us.ibm.com>
Reviewed-by: Jennifer A. Stofer <stofer@us.ibm.com>

3 files changed, 147 insertions, 70 deletions

diff --git a/import/chips/p9/common/pmlib/include/pstate_pgpe_cme_api.h b/import/chips/p9/common/pmlib/include/pstate_pgpe_cme_api.h
index 7e1c0488..7a51f0a5 100644
--- a/import/chips/p9/common/pmlib/include/pstate_pgpe_cme_api.h
+++ b/import/chips/p9/common/pmlib/include/pstate_pgpe_cme_api.h
@@ -98,7 +98,8 @@ enum MESSAGE_ID_DB0
     MSGID_DB0_STOP_PSTATE_BROADCAST         = 4,
     MSGID_DB0_CLIP_BROADCAST                = 5,
     MSGID_DB0_PMSR_UPDT                     = 6,
-    MSGID_DB0_VALID_END                     = 6 //This is for error checking
+    MSGID_DB0_REGISTER_DONE                 = 7,
+    MSGID_DB0_VALID_END                     = 7 //This is for error checking
 };
 
 enum MESSAGE_ID_DB3
@@ -262,5 +263,23 @@ typedef union pgpe_db0_pmsr_updt
     } fields;
 } pgpe_db0_pmsr_updt_t;
 
+//
+//PGPE-CME Doorbell0(Pstate Abort)
+//
+typedef union pgpe_db0_pstate_start_abort
+{
+    uint64_t value;
+    struct
+    {
+#ifdef _BIG_ENDIAN
+        uint64_t msg_id : 8;
+        uint64_t reserved: 56;
+#else
+        uint64_t reserved: 56;
+        uint64_t msg_id : 8;
+#endif
+    } fields;
+} pgpe_db0_pstate_start_abort_t;
+
 
 #endif //__PSTATE_PGPE_CME_API_H__
diff --git a/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_thread_db.c b/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_thread_db.c
index 955bd058..caa9e666 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_thread_db.c
+++ b/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_thread_db.c
@@ -436,71 +436,71 @@ void p9_cme_pstate_process_db0()
 
     uint32_t cme_flags = in32(CME_LCL_FLAGS);
 
-    //Ignore if Doorbell0 if DB0_PROCESSING_ENABLE=0. This bit is zero
-    //upon CME boot. PGPE will set it right before sending Pstate Starts.
-    //PGPE clears it before ACKing Stop Entry Notify(this CME is about to powered off)
-    //back to SGPE
-    if (in32(CME_LCL_SRTCH0) & BIT32(CME_SCRATCH_DB0_PROCESSING_ENABLE))
-    {
-        //Process DB0
-        //Start Pstates and Pstates NOT enabled
-        if((G_dbData.fields.cme_message_number0 == MSGID_DB0_START_PSTATE_BROADCAST) &&
-           (!(cme_flags & BIT32(CME_FLAGS_PSTATES_ENABLED))))
-        {
-            //This will Clear EISR[DB0_C0/1] to prevent another DB0 interrupt
-            //PGPE multicast Global Bcast/Clip Bcast DB0, and it's possible that
-            //the DB0 interrupt was taken as a result of multicast operation, but
-            //the value of DB0 read corresponds to Pstate Start. In such a case,
-            //another DB0 interrupt can happen, and it appears as PGPE has sent
-            //a second Pstate Start causing CME to ACK back with error and Halt
-            p9_cme_pstate_db0_start();
-        }
-        //Global Actual Broadcast and Pstates enabled
-        else if((G_dbData.fields.cme_message_number0 == MSGID_DB0_GLOBAL_ACTUAL_BROADCAST) ||
-                (G_dbData.fields.cme_message_number0 == MSGID_DB0_DB3_GLOBAL_ACTUAL_BROADCAST))
-        {
-            //Process Global Bcast only if Pstates are enabled.
-            //Otherwise, ignore. The reason is PGPE multicasts Global Bcast, and doorbell0
-            //can be written while this CME is powered-off or or about to be powered-off.
-            //For Pstate Start and Stop PGPE only unicasts.
-            if (cme_flags & BIT32(CME_FLAGS_PSTATES_ENABLED))
-            {
-                p9_cme_pstate_db0_glb_bcast();
-            }
-        }
-        //Stop Pstates and Pstates enabled
-        else if((G_dbData.fields.cme_message_number0 == MSGID_DB0_STOP_PSTATE_BROADCAST) &&
-                (cme_flags & BIT32(CME_FLAGS_PSTATES_ENABLED)))
-        {
-            p9_cme_pstate_db0_stop();
-        }
-        //Pmin or Pmax Update
-        else if(G_dbData.fields.cme_message_number0 == MSGID_DB0_CLIP_BROADCAST)
+    //Process DB0
+    //Start Pstates and Pstates NOT enabled
+    if((G_dbData.fields.cme_message_number0 == MSGID_DB0_START_PSTATE_BROADCAST) &&
+       (!(cme_flags & BIT32(CME_FLAGS_PSTATES_ENABLED))))
+    {
+        //This will Clear EISR[DB0_C0/1] to prevent another DB0 interrupt
+        //PGPE multicast Global Bcast/Clip Bcast DB0, and it's possible that
+        //the DB0 interrupt was taken as a result of multicast operation, but
+        //the value of DB0 read corresponds to Pstate Start. In such a case,
+        //another DB0 interrupt can happen, and it appears as PGPE has sent
+        //a second Pstate Start causing CME to ACK back with error and Halt
+        p9_cme_pstate_db0_start();
+    }
+    //Global Actual Broadcast and Pstates enabled
+    else if((G_dbData.fields.cme_message_number0 == MSGID_DB0_GLOBAL_ACTUAL_BROADCAST) ||
+            (G_dbData.fields.cme_message_number0 == MSGID_DB0_DB3_GLOBAL_ACTUAL_BROADCAST))
+    {
+        //Process Global Bcast only if Pstates are enabled.
+        //Otherwise, ignore. The reason is PGPE multicasts Global Bcast, and doorbell0
+        //can be written while this CME is powered-off or or about to be powered-off.
+        //For Pstate Start and Stop PGPE only unicasts.
+        if (cme_flags & BIT32(CME_FLAGS_PSTATES_ENABLED))
         {
-            p9_cme_pstate_db0_clip_bcast();
+            p9_cme_pstate_db0_glb_bcast();
         }
-        else if(G_dbData.fields.cme_message_number0 == MSGID_DB0_PMSR_UPDT)
+    }
+    //Stop Pstates and Pstates enabled
+    else if((G_dbData.fields.cme_message_number0 == MSGID_DB0_STOP_PSTATE_BROADCAST) &&
+            (cme_flags & BIT32(CME_FLAGS_PSTATES_ENABLED)))
+    {
+        p9_cme_pstate_db0_stop();
+    }
+    //Pmin or Pmax Update
+    else if(G_dbData.fields.cme_message_number0 == MSGID_DB0_CLIP_BROADCAST)
+    {
+        p9_cme_pstate_db0_clip_bcast();
+    }
+    else if(G_dbData.fields.cme_message_number0 == MSGID_DB0_PMSR_UPDT)
+    {
+        p9_cme_pstate_db0_pmsr_updt();
+    }
+    else if(G_dbData.fields.cme_message_number0 == MSGID_DB0_REGISTER_DONE)
+    {
+        ppmPigData.value = 0;
+        ppmPigData.fields.req_intr_type = 4;
+        ppmPigData.fields.req_intr_payload = MSGID_PCB_TYPE4_ACK_PSTATE_PROTO_ACK;
+        send_pig_packet(ppmPigData.value, G_cme_pstate_record.firstGoodCoreMask);
+    }
+    //Otherwise, send an ERR ACK to PGPE and Halt
+    else
+    {
+        ppmPigData.value = 0;
+        ppmPigData.fields.req_intr_type = 4;
+        ppmPigData.fields.req_intr_payload = MSGID_PCB_TYPE4_ACK_ERROR;
+        send_pig_packet(ppmPigData.value, G_cme_pstate_record.firstGoodCoreMask);
+        PK_TRACE_INF("PSTATE: Bad DB0=0x%x", (uint8_t)G_dbData.fields.cme_message_number0);
+
+        if(G_dbData.fields.cme_message_number0 < MSGID_DB0_VALID_START ||
+           G_dbData.fields.cme_message_number0 > MSGID_DB0_VALID_END)
         {
-            p9_cme_pstate_db0_pmsr_updt();
+            PK_PANIC(CME_PSTATE_UNEXPECTED_DB0_MSGID);
         }
-        //Otherwise, send an ERR ACK to PGPE and Halt
         else
         {
-            ppmPigData.value = 0;
-            ppmPigData.fields.req_intr_type = 4;
-            ppmPigData.fields.req_intr_payload = MSGID_PCB_TYPE4_ACK_ERROR;
-            send_pig_packet(ppmPigData.value, G_cme_pstate_record.firstGoodCoreMask);
-            PK_TRACE_INF("PSTATE: Bad DB0=0x%x", (uint8_t)G_dbData.fields.cme_message_number0);
-
-            if(G_dbData.fields.cme_message_number0 < MSGID_DB0_VALID_START ||
-               G_dbData.fields.cme_message_number0 > MSGID_DB0_VALID_END)
-            {
-                PK_PANIC(CME_PSTATE_UNEXPECTED_DB0_MSGID);
-            }
-            else
-            {
-                PK_PANIC(CME_PSTATE_INVALID_DB0_MSGID);
-            }
+            PK_PANIC(CME_PSTATE_INVALID_DB0_MSGID);
         }
     }
 
@@ -541,8 +541,11 @@ inline void p9_cme_pstate_register()
 
             PK_TRACE_INF("PSTATE: DB0 Processing is Enabled");
 
-            //PGPE will send Pmin, Pmax and Pstate Start Doorbells
-            //Pstate Start is the last DB0
+            //PGPE sends MSGID_DB0_REGISTER_DONE, if Pstates aren't active anymore.
+            //Otherwise, PGPE sends DB0 in the following order
+            //1. MSGID_DB0_CLIP_BROADCAST
+            //2. MSGID_DB0_CLIP_BROADCAST
+            //3. MSGID_DB0_START_PSTATE_BROADCAST
             while(!done)
             {
                 uint32_t db0_check = G_cme_pstate_record.firstGoodCoreMask << SHIFT64SH(37);
@@ -565,6 +568,15 @@ inline void p9_cme_pstate_register()
                 {
                     p9_cme_pstate_db0_clip_bcast();
                 }
+                //PGPE sends this if Pstates aren't active anymore
+                else if ((G_dbData.fields.cme_message_number0 ==  MSGID_DB0_REGISTER_DONE))
+                {
+                    done = 1;
+                    ppmPigData.value = 0;
+                    ppmPigData.fields.req_intr_type = 4;
+                    ppmPigData.fields.req_intr_payload = MSGID_PCB_TYPE4_ACK_PSTATE_PROTO_ACK;
+                    send_pig_packet(ppmPigData.value, G_cme_pstate_record.firstGoodCoreMask);
+                }
             }
         }
 
@@ -576,15 +588,32 @@ inline void p9_cme_pstate_register()
 
         if (cme_flags & BIT32(CME_FLAGS_WAIT_ON_PSTATE_START))
         {
-            //PGPE sends Pmin, Pmax, and Pstate start DB0, so
-            //total of three
             PK_TRACE_INF("PSTATE: Wait on Pstate Start");
 
-            while(msgCnt != 3)
+            //PGPE sends MSGID_DB0_REGISTER_DONE, if Pstates aren't active anymore.
+            //Otherwise, PGPE sends DB0 in the following order
+            //1. MSGID_DB0_CLIP_BROADCAST
+            //2. MSGID_DB0_CLIP_BROADCAST
+            //3. MSGID_DB0_START_PSTATE_BROADCAST
+            while(!done)
             {
-                p9_cme_pstate_sibling_lock_and_intercme_protocol(1);
-                msgCnt++;
-                PK_TRACE_INF("PSTATE: Sib Register MsgCnt=%d", msgCnt);
+                if ((G_dbData.fields.cme_message_number0 ==  MSGID_DB0_REGISTER_DONE))
+                {
+                    done = 1;
+                    PK_TRACE_INF("PSTATE: Sib Register Got Pstate Register Done");
+                }
+                else
+                {
+                    p9_cme_pstate_sibling_lock_and_intercme_protocol(1);
+                    msgCnt++;
+
+                    if (msgCnt == 3)
+                    {
+                        done = 1;
+                    }
+
+                    PK_TRACE_INF("PSTATE: Sib Register MsgCnt=%d", msgCnt);
+                }
             }
         }
     }
diff --git a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_thread_process_requests.c b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_thread_process_requests.c
index ea11dd1b..2a13c6de 100644
--- a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_thread_process_requests.c
+++ b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_thread_process_requests.c
@@ -909,8 +909,9 @@ void p9_pgpe_process_registration()
     uint32_t  q, c, oldActiveDB, oldActiveQuads, unicastCoresVector = 0, quadsRegisterProcess;
     uint32_t quadAckExpect = 0;
     uint64_t value;
-    pgpe_db0_start_ps_bcast_t db0_glb_bcast;
-    pgpe_db0_clip_bcast_t     db0_clip_bcast;
+    pgpe_db0_start_ps_bcast_t   db0_glb_bcast;
+    pgpe_db0_clip_bcast_t       db0_clip_bcast;
+    pgpe_db0_pstate_start_abort_t     db0_pstate_start_abort;
 
     //Save it for global bcast sync in case GlobalPSTarget is higher in
     //after doing auctions with just registered quads
@@ -1096,6 +1097,34 @@ void p9_pgpe_process_registration()
             }
         }
     }
+    //Send REGISTER_DONE if Pstates aren't active anymore
+    else
+    {
+        for (q = 0; q < MAX_QUADS; q++)
+        {
+            if(quadsRegisterProcess & QUAD_MASK(q))
+            {
+                //Give Quad Manager CME control of DPLL through inter-ppm
+                //SGPE sets up the DPLL_SEL bits before booting CME
+                GPE_PUTSCOM(GPE_SCOM_ADDR_QUAD(QPPM_QPMMR_OR, q), BIT64(26));
+
+                for (c = FIRST_CORE_FROM_QUAD(q); c < LAST_CORE_FROM_QUAD(q); c++)
+                {
+                    if (G_pgpe_pstate_record.activeDB & CORE_MASK(c))
+                    {
+                        unicastCoresVector |= CORE_MASK(c);
+                    }
+                }
+
+                quadAckExpect |= QUAD_MASK(q);
+            }
+        }
+
+        db0_pstate_start_abort.value = 0;
+        db0_pstate_start_abort.fields.msg_id = MSGID_DB0_REGISTER_DONE;
+        p9_pgpe_send_db0(db0_pstate_start_abort.value, unicastCoresVector, PGPE_DB0_UNICAST, PGPE_DB0_ACK_WAIT_CME,
+                         quadAckExpect);
+    }
 
     PK_TRACE_DBG("PTH: Register Exit");
 }