STOP: Fix Infinite Stacking up Stop Processing led by Entry Abort

Key_Cronus_Test=PM_REGRESS

Change-Id: Ib55a367b7e4e1c8b66e7473c696dceaa373b9a84
CQ: SW420964
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/55802
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: Gregory S. Still <stillgs@us.ibm.com>
Reviewed-by: RAHUL BATRA <rbatra@us.ibm.com>
Reviewed-by: Jennifer A. Stofer <stofer@us.ibm.com>

8 files changed, 72 insertions, 18 deletions

diff --git a/import/chips/p9/procedures/ppe_closed/cme/cme_panic_codes.h b/import/chips/p9/procedures/ppe_closed/cme/cme_panic_codes.h
index 665d0c40..8c664dd8 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/cme_panic_codes.h
+++ b/import/chips/p9/procedures/ppe_closed/cme/cme_panic_codes.h
@@ -70,7 +70,7 @@ CME_STOP_SPWU_PROTOCOL_ERROR        = 0x1d06,
 CME_STOP_ENTRY_STOPCLK_FAILED       = 0x1d08,
 CME_STOP_ENTRY_XSTOP_ERROR          = 0x1d09, // NDD1
 CME_STOP_ENTRY_TRAP_INJECT          = 0x1d0a,
-//_UNUSED_1d0d                      = 0x1d0d,
+CME_STOP_ENTRY_HANDOFF_LESSTHAN5    = 0x1d0d,
 //_UNUSED_1d1c                      = 0x1d1c,
 //_UNUSED_1d1d                      = 0x1d1d,
 //_UNUSED_1d1e                      = 0x1d1e,
diff --git a/import/chips/p9/procedures/ppe_closed/cme/iota_lnk_cfg.h b/import/chips/p9/procedures/ppe_closed/cme/iota_lnk_cfg.h
index 430a935c..2bbf930c 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/iota_lnk_cfg.h
+++ b/import/chips/p9/procedures/ppe_closed/cme/iota_lnk_cfg.h
@@ -38,7 +38,7 @@
 #define PPE_DEBUG_PTRS_SIZE      CME_DEBUG_PTRS_SIZE
 
 #define PPE_DUMP_PTR_PSTATE_SIZE 0x4C
-#define PPE_DUMP_PTR_STOP_SIZE   0x28
+#define PPE_DUMP_PTR_STOP_SIZE   0x34
 #define PPE_DUMP_PTR_COMMON_SIZE 0x8
 
 #endif
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 55a22848..00cc3753 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
@@ -39,7 +39,7 @@ CmePstateRecord G_cme_pstate_record __attribute__((section (".dump_ptr_pstate"))
 
 // CME Stop Header and Structure
 #include "p9_cme_stop.h"
-CmeStopRecord G_cme_stop_record __attribute__((section (".dump_ptr_stop"))) = {{0}, {0}, 0, 0, 0, 0, 0, 0, 0, 0, 0, {0}};
+CmeStopRecord G_cme_stop_record __attribute__((section (".dump_ptr_stop"))) = {{0}, {0}, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, {0}};
 
 #if !DISABLE_PERIODIC_CORE_QUIESCE && (NIMBUS_DD_LEVEL == 20 || NIMBUS_DD_LEVEL == 21 || CUMULUS_DD_LEVEL == 10)
 CmeFitRecord G_cme_fit_record = {0, 0, 0, 0, 0xFFFFFFFF, 0};
diff --git a/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_pstate.c b/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_pstate.c
index 3e4275d7..6f1a9b1a 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_pstate.c
+++ b/import/chips/p9/procedures/ppe_closed/cme/pstate_cme/p9_cme_pstate.c
@@ -85,10 +85,12 @@ int send_pig_packet(uint64_t data, uint32_t coreMask)
         // Read PPMPIG status
         CME_GETSCOM(PPM_PIG, coreMask, data_tmp);
     }
-    while (((ppm_pig_t)data_tmp).fields.intr_granted);
+    while ((((ppm_pig_t)data_tmp).fields.pending_source & 0x1));
 
     // Send PIG packet
     CME_PUTSCOM(PPM_PIG, coreMask, data);
+    PK_TRACE_DBG("CME: Sending PIG[%x] at core[%x]",
+                 (uint32_t)(data >> 32), coreMask);
 
     pk_critical_section_exit(&ctx);
 
diff --git a/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop.h b/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop.h
index 3483fced..4179877a 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop.h
+++ b/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop.h
@@ -272,6 +272,8 @@ typedef struct
     uint8_t       act_level[MAX_CORES_PER_CME];
     // uint8_t above is processed by stb/lbz in asm, no additional shifting
 
+    // mark the start and the end of entry process
+    uint32_t      entry_pending;
     // whether core is in running state,
     // used for aborted entry detection or filter wakeup core select in scom address
     uint32_t      core_running;
diff --git a/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_entry.c b/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_entry.c
index 022eb4bc..795944e1 100755
--- a/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_entry.c
+++ b/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_entry.c
@@ -1244,8 +1244,9 @@ p9_cme_stop_entry()
                   (core_wakeup << SHIFT32(17)));
             sync();
             wrteei(0);
-            out32(CME_LCL_EIMR_OR,  (BITS32(12, 6) | BITS32(20, 2)));
+            out32(CME_LCL_EIMR_OR, BITS32(10, 12));
             wrteei(1);
+
 #endif
 
             //===================
@@ -1419,7 +1420,7 @@ p9_cme_stop_entry()
                   (core_wakeup << SHIFT32(17)));
             sync();
             wrteei(0);
-            out32(CME_LCL_EIMR_OR,  (BITS32(12, 6) | BITS32(20, 2)));
+            out32(CME_LCL_EIMR_OR, BITS32(10, 12));
             wrteei(1);
 #endif
 
@@ -1564,23 +1565,45 @@ p9_cme_stop_entry()
                 {
                     core_index = core_mask & 1;
 
+#if DISABLE_STOP8
+
+                    if (G_cme_stop_record.req_level[core_index] >= STOP_LEVEL_11)
+
+#else
+
                     if (G_cme_stop_record.req_level[core_index] >= STOP_LEVEL_8)
+
+#endif
+
                     {
                         CME_PUTSCOM(CPPM_CPMMR_OR, core_mask, BIT64(10));
                         pig.fields.req_intr_type = PIG_TYPE3;
                         G_cme_stop_record.core_blockpc |= core_mask;
                     }
-                    else
+                    else if (G_cme_stop_record.req_level[core_index] >= STOP_LEVEL_5)
                     {
                         CME_PUTSCOM(CPPM_CPMMR_CLR, core_mask, BIT64(10));
                         pig.fields.req_intr_type = PIG_TYPE2;
                         G_cme_stop_record.core_blockpc &= ~core_mask;
                     }
+                    else
+                    {
+                        PK_TRACE_ERR("ERROR: Core[%d] Handoff to SGPE with Requested Stop Level[%d]",
+                                     core_mask, G_cme_stop_record.req_level[core_index]);
+                        PK_PANIC(CME_STOP_ENTRY_HANDOFF_LESSTHAN5);
+                    }
 
                     pig.fields.req_intr_payload = G_cme_stop_record.req_level[core_index];
 
                     // put PIG and Wakeup_Notify_Select back to back as possible
                     send_pig_packet(pig.value, core_mask);
+
+                    do
+                    {
+                        CME_GETSCOM(PPM_PIG, core_mask, scom_data.value);
+                    }
+                    while (scom_data.words.lower & BIT64SH(39));
+
                     CME_PUTSCOM(CPPM_CPMMR_OR, core_mask, BIT64(13));
                     PK_TRACE_DBG("Switch Core[%d] PPM wakeup to STOP-GPE via CPMMR[13]", core_mask);
 
diff --git a/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_irq_handlers.c b/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_irq_handlers.c
index 9c327afe..e64e6950 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_irq_handlers.c
+++ b/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_irq_handlers.c
@@ -83,13 +83,20 @@ p9_cme_stop_pcwu_handler(void)
     {
         PK_TRACE_INF("PCWU Launching exit thread");
 
-        out32(CME_LCL_EIMR_OR, BITS32(10, 12));
+        out32(CME_LCL_EIMR_OR, BITS32(12, 10));
+        g_eimr_override |= BITS64(12, 10);
         wrteei(1);
+
+        // The actual exit sequence
         p9_cme_stop_exit();
     }
 
-    // re-evaluate stop entry & exit enables
-    p9_cme_stop_eval_eimr_override();
+    // in case abort, complete pending entry first
+    if (!G_cme_stop_record.entry_pending)
+    {
+        // re-evaluate stop entry & exit enables
+        p9_cme_stop_eval_eimr_override();
+    }
 }
 
 
@@ -181,13 +188,20 @@ p9_cme_stop_spwu_handler(void)
     {
         PK_TRACE_INF("SPWU Launching exit thread");
 
-        out32(CME_LCL_EIMR_OR, BITS32(10, 12));
+        out32(CME_LCL_EIMR_OR, BITS32(12, 10));
+        g_eimr_override |= BITS64(12, 10);
         wrteei(1);
+
+        // The actual exit sequence
         p9_cme_stop_exit();
     }
 
-    // re-evaluate stop entry & exit enables
-    p9_cme_stop_eval_eimr_override();
+    // in case abort, complete pending entry first
+    if (!G_cme_stop_record.entry_pending)
+    {
+        // re-evaluate stop entry & exit enables
+        p9_cme_stop_eval_eimr_override();
+    }
 }
 
 
@@ -198,12 +212,19 @@ p9_cme_stop_rgwu_handler(void)
     MARK_TRAP(STOP_RGWU_HANDLER)
     PK_TRACE_INF("RGWU Handler Trigger");
 
-    out32(CME_LCL_EIMR_OR, BITS32(10, 12));
+    out32(CME_LCL_EIMR_OR, BITS32(12, 10));
+    g_eimr_override |= BITS64(12, 10);
     wrteei(1);
+
+    // The actual exit sequence
     p9_cme_stop_exit();
 
-    // re-evaluate stop entry & exit enables
-    p9_cme_stop_eval_eimr_override();
+    // in case abort, complete pending entry first
+    if (!G_cme_stop_record.entry_pending)
+    {
+        // re-evaluate stop entry & exit enables
+        p9_cme_stop_eval_eimr_override();
+    }
 }
 
 
@@ -214,12 +235,18 @@ p9_cme_stop_enter_handler(void)
     MARK_TRAP(STOP_ENTER_HANDLER)
     PK_TRACE_INF("PM_ACTIVE Handler Trigger");
 
-    out32(CME_LCL_EIMR_OR, BITS32(10, 12));
+    // Abort Protection
+    out32(CME_LCL_EIMR_OR, BITS32(12, 10));
+    g_eimr_override |= BITS64(12, 10);
+    G_cme_stop_record.entry_pending = 1;
     wrteei(1);
 
     // The actual entry sequence
     p9_cme_stop_entry();
 
+    // Restore Abort Protection
+    G_cme_stop_record.entry_pending = 0;
+
     // re-evaluate stop entry & exit enables
     p9_cme_stop_eval_eimr_override();
 }
diff --git a/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_threads.c b/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_threads.c
index 58310c4c..119791ae 100644
--- a/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_threads.c
+++ b/import/chips/p9/procedures/ppe_closed/cme/stop_cme/p9_cme_stop_threads.c
@@ -60,7 +60,7 @@ p9_cme_stop_core_error_handler(uint32_t core, uint32_t core_error, uint32_t pani
 void
 p9_cme_stop_eval_eimr_override()
 {
-    g_eimr_override      &= ~(BITS64(12, 6) | BITS64(20, 2));
+    g_eimr_override      &= ~BITS64(12, 10);
     data64_t mask_irqs    = {0};
     mask_irqs.words.lower = 0;