PM: Fix DB0 Hang

        Key_Cronus_Test=PM_REGRESS

Change-Id: I706ec7b87e777b736153d5765ced0a3f6cea5d96
CQ: SW470688
Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/81266
Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com>
Tested-by: PPE CI <ppe-ci+hostboot@us.ibm.com>
Tested-by: Cronus HW CI <cronushw-ci+hostboot@us.ibm.com>
Tested-by: Hostboot CI <hostboot-ci+hostboot@us.ibm.com>
Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com>
Tested-by: HWSV CI <hwsv-ci+hostboot@us.ibm.com>
Reviewed-by: YUE DU <daviddu@us.ibm.com>
Reviewed-by: RANGANATHPRASAD G. BRAHMASAMUDRA <prasadbgr@in.ibm.com>
Reviewed-by: Jennifer A Stofer <stofer@us.ibm.com>

7 files changed, 175 insertions, 13 deletions

diff --git a/import/chips/p9/procedures/hwp/lib/p9_pm_hcd_flags.h b/import/chips/p9/procedures/hwp/lib/p9_pm_hcd_flags.h
index fcc74dbd..1b940919 100644
--- a/import/chips/p9/procedures/hwp/lib/p9_pm_hcd_flags.h
+++ b/import/chips/p9/procedures/hwp/lib/p9_pm_hcd_flags.h
@@ -129,6 +129,7 @@ enum PM_CME_FLAGS_DEFS
     CME_FLAGS_DROOP_SUSPEND_ENTRY                       = 14,
     CME_FLAGS_SAFE_MODE                                 = 16,
     CME_FLAGS_PSTATES_SUSPENDED                         = 17,
+    CME_FLAGS_DB0_COMM_RECV_STARVATION_CNT_ENABLED      = 18,
     CME_FLAGS_SPWU_CHECK_ENABLE                         = 22,
     CME_FLAGS_BLOCK_ENTRY_STOP11                        = 23,
     CME_FLAGS_PSTATES_ENABLED                           = 24,
diff --git a/import/chips/p9/procedures/ppe/iota/iota_uih.c b/import/chips/p9/procedures/ppe/iota/iota_uih.c
index 546def67..28272661 100644
--- a/import/chips/p9/procedures/ppe/iota/iota_uih.c
+++ b/import/chips/p9/procedures/ppe/iota/iota_uih.c
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HCODE Project                                                */
 /*                                                                        */
-/* COPYRIGHT 2017                                                         */
+/* COPYRIGHT 2017,2019                                                    */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -32,6 +32,9 @@ int      g_eimr_stack_ctr = -1;
 uint64_t g_eimr_override_stack[IOTA_NUM_EXT_IRQ_PRIORITIES];
 uint64_t g_eimr_override = 0x0000000000000000;
 uint64_t g_ext_irq_vector = 0;
+uint32_t g_db0_pending_fit_tick_count = 0;
+uint32_t g_comm_recv_pending_fit_tick_count = 0;
+uint32_t g_intercme_in0_pending_tick_count  = 0;
 
 // Unified IRQ priority and masking handler.
 // - Locates the highest priority IRQ task vector that has at least one of its
@@ -51,13 +54,50 @@ uint32_t iota_uih(void)
 
     do
     {
-        if(ext_irq_vectors_cme[iPrtyLvl][IDX_PRTY_VEC] & g_ext_irq_vector)
+        //Note: Special handling of DB0/COMM_RECV to handle the db0/comm_recv
+        //starvation case.
+        //
+        //Reason: DB0(Quad Manager CME) and COMM_RECV(Sibling CME) are lower priority
+        //than the STOP related interrupts,
+        //and can stay pending for very long time(~ms scale) on systems with
+        //high frequency of STOP requests. This can then prevent PGPE from
+        //completing OCC directed IPC operations within the expected
+        //time bounds(< 8ms)
+        //
+        //Mechanism:
+        //1)In FIT: Every FIT tick, we check if DB0(on Quad manager)/COMM_RECV(on Sibling CME)
+        //is pending. If DB0(on Quad manager)/COMM_RECV(on Sibling CME) is seen pending for
+        //more than DB0_FIT_TICK_THRESHOLD/COMM_RECV_FIT_TICK_THRESHOLD FIT ticks,
+        //then we take action in UIH
+        //
+        //2)In UIH: We set priority level to IDX_PRTY_LVL_DB0/IDX_PRTY_LVL_COMM_RECVD, and  mask
+        //everything except Priority 0(xstop, exceptions, etc). This then allows a
+        //pending DB0 to complete
+        if(g_db0_pending_fit_tick_count > DB0_FIT_TICK_THRESHOLD)
+        {
+            bFound = 1;
+            iPrtyLvl = IDX_PRTY_LVL_DB0;
+            break;
+        }
+        else if(g_comm_recv_pending_fit_tick_count > COMM_RECV_FIT_TICK_THRESHOLD)
+        {
+            bFound = 1;
+            iPrtyLvl = IDX_PRTY_LVL_COMM_RECVD;
+            break;
+        }
+        else if(g_intercme_in0_pending_tick_count > INTERCME_IN0_FIT_TICK_THRESHOLD)
+        {
+            bFound = 1;
+            iPrtyLvl = IDX_PRTY_LVL_INTERCME_IN0;
+            break;
+        }
+        else if(ext_irq_vectors_cme[iPrtyLvl][IDX_PRTY_VEC] & g_ext_irq_vector)
         {
             bFound = 1;
             break;
         }
     }
-    while(++iPrtyLvl < (IOTA_NUM_EXT_IRQ_PRIORITIES - 1)); //No need to check DISABLED.
+    while(++iPrtyLvl < (IOTA_NUM_EXT_IRQ_PRIORITIES - 1));   //No need to check DISABLED.
 
     // Only manipulate EIMR masks for task level prty levels.
     // Let shared non-task IRQs (iPrtyLvl=0) be processed by
@@ -82,9 +122,38 @@ uint32_t iota_uih(void)
         }
 
         // 3. Write the new mask for this priority level.
-        out64(CME_LCL_EIMR, ext_irq_vectors_cme[iPrtyLvl][IDX_MASK_VEC] |
-              g_eimr_override);
-
+        //Note: Special handling of DB0/COMM_RECV to handle the db0/comm_recv
+        //starvation case.
+        //
+        //Reason: DB0(Quad Manager CME) and COMM_RECV(Sibling CME) are lower priority
+        //than the STOP related interrupts,
+        //and can stay pending for very long time(~ms scale) on systems with
+        //high frequency of STOP requests. This can then prevent PGPE from
+        //completing OCC directed IPC operations within the expected
+        //time bounds(< 8ms)
+        //
+        //Mechanism:
+        //1)In FIT: Every FIT tick, we check if DB0(on Quad manager)/COMM_RECV(on Sibling CME)
+        //is pending. If DB0(on Quad manager)/COMM_RECV(on Sibling CME) is seen pending for
+        //more than DB0_FIT_TICK_THRESHOLD/COMM_RECV_FIT_TICK_THRESHOLD FIT ticks,
+        //then we take action in UIH
+        //
+        //2)In UIH: We set priority level to IDX_PRTY_LVL_DB0/IDX_PRTY_LVL_COMM_RECVD, and  mask
+        //everything except Priority 0(xstop, exceptions, etc). This then allows a
+        //pending DB0 to complete
+        if ((g_db0_pending_fit_tick_count > DB0_FIT_TICK_THRESHOLD) ||
+            (g_comm_recv_pending_fit_tick_count > COMM_RECV_FIT_TICK_THRESHOLD) ||
+            (g_intercme_in0_pending_tick_count > INTERCME_IN0_FIT_TICK_THRESHOLD))
+        {
+            PK_TRACE_INF("UIH: Starvation Detected. Overriding Mask!");
+            out64(CME_LCL_EIMR, (ext_irq_vectors_cme[0][IDX_MASK_VEC] |
+                                 g_eimr_override));
+        }
+        else
+        {
+            out64(CME_LCL_EIMR, ext_irq_vectors_cme[iPrtyLvl][IDX_MASK_VEC] |
+                  g_eimr_override);
+        }
     }
     else
     {
diff --git a/import/chips/p9/procedures/ppe_closed/cme/p9_cme_iota_main.c b/import/chips/p9/procedures/ppe_closed/cme/p9_cme_iota_main.c
index f9632f45..0818517e 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/p9_cme_iota_main.c
+++ b/import/chips/p9/procedures/ppe_closed/cme/p9_cme_iota_main.c
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HCODE Project                                                */
 /*                                                                        */
-/* COPYRIGHT 2017,2018                                                    */
+/* COPYRIGHT 2017,2019                                                    */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -46,6 +46,8 @@ CmeFitRecord G_cme_fit_record = {0, 0, 0, 0, 0xFFFFFFFF, 0};
 #endif
 
 
+void p9_cme_pstate_db0_comm_recv_intercme_in0_pending_counter();
+
 uint32_t G_CME_LCL_EINR      = CME_LCL_EINR;
 uint32_t G_CME_LCL_EISR      = CME_LCL_EISR;
 uint32_t G_CME_LCL_EISR_CLR  = CME_LCL_EISR_CLR;
@@ -102,6 +104,8 @@ void fit_handler()
     p9_cme_core_livelock_buster();
 #endif
 
+    //Handle DB0/Comm_Recv starvation case
+    p9_cme_pstate_db0_comm_recv_intercme_in0_pending_counter();
 }
 #endif //fit handler
 
diff --git a/import/chips/p9/procedures/ppe_closed/cme/p9_cme_irq.h b/import/chips/p9/procedures/ppe_closed/cme/p9_cme_irq.h
index c5a29801..170eb88f 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/p9_cme_irq.h
+++ b/import/chips/p9/procedures/ppe_closed/cme/p9_cme_irq.h
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HCODE Project                                                */
 /*                                                                        */
-/* COPYRIGHT 2015,2018                                                    */
+/* COPYRIGHT 2015,2019                                                    */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -48,6 +48,19 @@
 
 #include <stdint.h>
 
+//CME_TSEL is set to 8 which means FIT has period of 1.04ms when
+//Nest Freq is 2000Mhz. Ideally, should calculate period of FIT based
+//on nest frequency, but nest frequency is NOT plumbed to CME and we
+//don't need to be highly accurate here.
+//Note, from PGPE perspective, the latency of the DB0 operation depends
+//on the amount of time DB0 is pending on Quad Manager plus COMM_RECV is pending
+//on sibling. This is because COMM_RECV interrupt is triggered by the DB0
+//handler on the quad manager. Therefore, we must set the COMM_RECV_TICK_THRESHOLD
+//to be smaller.
+#define DB0_FIT_TICK_THRESHOLD          1 //Threshold for DB0 pending count(2ms)
+#define COMM_RECV_FIT_TICK_THRESHOLD    1 //Threshold for COMM_RECV pending countr(2ms)
+#define INTERCME_IN0_FIT_TICK_THRESHOLD 1 //Threshold for COMM_RECV pending countr(2ms)
+
 // Priority Levels
 #define IDX_PRTY_LVL_HIPRTY         0
 #define IDX_PRTY_LVL_DB3            1
diff --git a/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_intercme.c b/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_intercme.c
index e802448e..9497442c 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_intercme.c
+++ b/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_intercme.c
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HCODE Project                                                */
 /*                                                                        */
-/* COPYRIGHT 2016,2018                                                    */
+/* COPYRIGHT 2016,2019                                                    */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -43,17 +43,21 @@
 //
 extern CmePstateRecord G_cme_pstate_record;
 extern CmeRecord G_cme_record;
+extern uint32_t g_comm_recv_pending_fit_tick_count;
+extern uint32_t g_intercme_in0_pending_tick_count;
 
 //
 //InterCME_IN0 handler
 //
 void p9_cme_pstate_intercme_in0_irq_handler(void)
 {
+    g_intercme_in0_pending_tick_count = 0;
     p9_cme_pstate_process_db0_sibling();
 }
 
 void p9_cme_pstate_intercme_msg_handler(void)
 {
+    g_comm_recv_pending_fit_tick_count = 0;
     p9_cme_pstate_sibling_lock_and_intercme_protocol(INTERCME_MSG_LOCK_WAIT_ON_RECV);
 }
 
@@ -119,6 +123,8 @@ void p9_cme_pstate_process_db0_sibling()
         //Unmask EIMR[OCC_HEARTBEAT_LOST/4]
         g_eimr_override &= ~BIT64(4);
 
+        out32(G_CME_LCL_FLAGS_OR, BIT32(CME_FLAGS_DB0_COMM_RECV_STARVATION_CNT_ENABLED));//Set Starvation Count enabled
+
         //Clear Core GPMMR RESET_STATE_INDICATOR bit to show pstates have started
         CME_PUTSCOM(PPM_GPMMR_CLR, G_cme_record.core_enabled, BIT64(15));
     }
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 0c3f8cde..cba9f9e8 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
@@ -56,6 +56,9 @@ extern CmePstateRecord G_cme_pstate_record;
 extern cmeHeader_t* G_cmeHeader;
 extern LocalPstateParmBlock* G_lppb;
 extern uint8_t G_vdm_threshold_table[];
+extern uint32_t g_db0_pending_fit_tick_count;
+extern uint32_t g_comm_recv_pending_fit_tick_count;
+extern uint32_t g_intercme_in0_pending_tick_count;
 cppm_cmedb0_t G_dbData;
 
 
@@ -75,6 +78,7 @@ void p9_cme_pstate_db0_start();
 void p9_cme_pstate_db0_glb_bcast();
 void p9_cme_pstate_db0_clip_bcast();
 void p9_cme_pstate_update();
+void p9_cme_pstate_db0_comm_recv_intercme_in0_pending_counter();
 
 //
 //Doorbell0 interrupt handler
@@ -167,6 +171,67 @@ void p9_cme_pstate_db0_handler(void)
 }
 
 //
+//Doorbell0/Comm Recv pending counter(called every FIT tick)
+//
+void p9_cme_pstate_db0_comm_recv_intercme_in0_pending_counter()
+{
+    //Note: Special handling of DB0/COMM_RECV to handle the db0/comm_recv
+    //starvation case.
+    //
+    //Reason: DB0(Quad Manager CME) and COMM_RECV(Sibling CME) are lower priority
+    //than the STOP related interrupts,
+    //and can stay pending for very long time(~ms scale) on systems with
+    //high frequency of STOP requests. This can then prevent PGPE from
+    //completing OCC directed IPC operations within the expected
+    //time bounds(< 8ms)
+    //
+    //Mechanism:
+    //1)In FIT: Every FIT tick, we check if DB0(on Quad manager)/COMM_RECV(on Sibling CME)
+    //is pending. If DB0(on Quad manager)/COMM_RECV(on Sibling CME) is seen pending for
+    //more than DB0_FIT_TICK_THRESHOLD/COMM_RECV_FIT_TICK_THRESHOLD FIT ticks,
+    //then we take action in UIH
+    //
+    //2)In UIH: We set priority level to IDX_PRTY_LVL_DB0/IDX_PRTY_LVL_COMM_RECVD, and  mask
+    //everything except Priority 0(xstop, exceptions, etc). This then allows a
+    //pending DB0 to complete
+    uint32_t cme_flags = in32(G_CME_LCL_FLAGS);
+
+    if (cme_flags & BIT32(CME_FLAGS_DB0_COMM_RECV_STARVATION_CNT_ENABLED))
+    {
+        if(G_cme_pstate_record.qmFlag)
+        {
+
+            if (cme_flags & BIT32(CME_FLAGS_CORE0_GOOD))
+            {
+                if (in32_sh(CME_LCL_EISR) & BIT64SH(36))
+                {
+                    g_db0_pending_fit_tick_count++;
+                }
+            }
+            else
+            {
+                if (in32_sh(CME_LCL_EISR) & BIT64SH(37))
+                {
+                    g_db0_pending_fit_tick_count++;
+                }
+            }
+        }
+        else
+        {
+            if (in32(CME_LCL_EISR) & BIT32(29))
+            {
+                g_comm_recv_pending_fit_tick_count++;
+            }
+
+            if(in32(CME_LCL_EISR) & BIT32(7))
+            {
+                g_intercme_in0_pending_tick_count++;
+            }
+        }
+    }
+}
+
+//
 //Doorbell3 interrupt handler
 //
 //Note: This enabled on both QuadManagerCME and SiblingCME
@@ -660,6 +725,9 @@ void p9_cme_pstate_process_db0()
     G_cme_pstate_record.updateAnalogError = 0;
     uint64_t scom_data;
 
+    //Clear out db0_pending_tick_count
+    g_db0_pending_fit_tick_count = 0;
+
     PK_TRACE_INF("PSTATE: Process DB0 Enter");
 
     //Clear EISR and read DB0 register
@@ -856,7 +924,7 @@ inline void p9_cme_pstate_register()
                     }
                 }
 
-                PK_TRACE_INF("PSTATE: Sib Register MsgCnt=%d", msgCnt);
+                PK_TRACE_DBG("PSTATE: Sib Register MsgCnt=%d", msgCnt);
             }
         }
     }
@@ -894,6 +962,7 @@ void p9_cme_pstate_db0_start()
         ack = MSGID_PCB_TYPE4_ACK_PSTATE_PROTO_ACK;
 
         out32(G_CME_LCL_FLAGS_OR, BIT32(24));//Set Pstates Enabled
+        out32(G_CME_LCL_FLAGS_OR, BIT32(CME_FLAGS_DB0_COMM_RECV_STARVATION_CNT_ENABLED));//Set Starvation Count enabled
 
         //Enable PMCR Interrupts (for good cores) when this task is done
         g_eimr_override &= ~(uint64_t)(G_cme_record.core_enabled << SHIFT64(35));
@@ -1035,7 +1104,7 @@ inline void p9_cme_pstate_db0_pmsr_updt()
     //Set Core GPMMR RESET_STATE_INDICATOR bit to show pstates have stopped
     CME_PUTSCOM(PPM_GPMMR_OR, G_cme_record.core_enabled, BIT64(15));
 
-    PK_TRACE_INF("PSTATE: DB0 Safe Mode Exit");
+    PK_TRACE_INF("PSTATE: DB0 PMSR Updt Exit");
 }
 
 void p9_cme_pstate_notify_sib(INTERCME_DIRECT_INTF intf)
@@ -1058,7 +1127,7 @@ inline void p9_cme_pstate_freq_update(uint32_t cme_flags)
     else
     {
         PK_TRACE_INF("PSTATE: Freq Updt Enter");
-        PK_TRACE_INF("PSTATE: Dpll0=0x%x", G_lppb->dpll_pstate0_value);
+        PK_TRACE_DBG("PSTATE: Dpll0=0x%x", G_lppb->dpll_pstate0_value);
 
         //Adjust DPLL
         qppm_dpll_freq_t dpllFreq;
diff --git a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_pstate.c b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_pstate.c
index 2ed5aa58..4d6348d9 100644
--- a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_pstate.c
+++ b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_pstate.c
@@ -2542,7 +2542,7 @@ void p9_pgpe_pstate_wov_init()
     G_pgpe_pstate_record.wov.avg_freq_gt_target_freq = 0;
     G_pgpe_pstate_record.wov.freq_loss_tenths_gt_max_droop_tenths = 0;
     G_pgpe_pstate_record.wov.status = WOV_DISABLED;
-    G_pgpe_pstate_record.wov.info = 0xdeadbeef;
+    G_pgpe_pstate_record.wov.info = 0xdeadde04;
 }
 
 //