PM: clear Hcode error injection bits upon injection and malf alert

- Also deals with PM callout order issue with the malf alert assertion

Key_Cronus_Test=PM_REGRESS

Change-Id: I78b41c59d09369c60891beb41ff9debc986e59b4
CQ: SW434547
Reviewed-on: http://ralgit01.raleigh.ibm.com/gerrit1/61020
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: Jennifer A. Stofer <stofer@us.ibm.com>

12 files changed, 89 insertions, 10 deletions

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 15890fed..27936797 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
@@ -587,6 +587,8 @@ void p9_cme_pstate_process_db0()
 
     if( scom_data & BIT64(CPPM_CSAR_PSTATE_HCODE_ERROR_INJECT) )
     {
+        // Clear the injection so things are not permenently stuck
+        CME_PUTSCOM(CPPM_CSAR_CLR, CME_MASK_BC, BIT64(CPPM_CSAR_PSTATE_HCODE_ERROR_INJECT));
         PK_TRACE_ERR("CME PSTATE ERROR INJECT TRAP");
         PK_PANIC(CME_PSTATE_TRAP_INJECT);
     }
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 09df7cb4..13ad0f8d 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
@@ -1368,6 +1368,8 @@ p9_cme_stop_entry()
 
             if( BIT64(CPPM_CSAR_STOP_HCODE_ERROR_INJECT) & scom_data.value )
             {
+                // Clear the injection so things are not permenently stuck
+                CME_PUTSCOM(CPPM_CSAR_CLR, core, BIT64(CPPM_CSAR_STOP_HCODE_ERROR_INJECT));
                 PK_TRACE_DBG("CME STOP ENTRY ERROR INJECT TRAP");
                 PK_PANIC(CME_STOP_ENTRY_TRAP_INJECT);
             }
diff --git a/import/chips/p9/procedures/ppe_closed/lib/p9_stop_recovery_trigger.c b/import/chips/p9/procedures/ppe_closed/lib/p9_stop_recovery_trigger.c
index 93df097e..586d0369 100644
--- a/import/chips/p9/procedures/ppe_closed/lib/p9_stop_recovery_trigger.c
+++ b/import/chips/p9/procedures/ppe_closed/lib/p9_stop_recovery_trigger.c
@@ -28,6 +28,8 @@
 #include "ppe42_scom.h"
 #include "ppehw_common.h"
 #include "gpehw_common.h"
+#include "ocb_firmware_registers.h"
+#include "cppm_register_addresses.h"
 #include "ocb_register_addresses.h"
 #include "p9_pm_hcd_flags.h"
 #include "p9_stop_recovery_trigger.h"
@@ -39,32 +41,91 @@
 void
 p9_stop_recovery_trigger()
 {
+    data64_t    scom_data  = {0};
+    uint32_t    qloop      = 0;
+    uint32_t    cloop      = 0;
+    uint32_t    cindex     = 0;
+    uint32_t    m_1c       = 0;
+    ocb_ccsr_t  ccsr       = {0};
+
+    // Clear ALL the error injections to have any hope that the next
+    // special wake-up will succeed.  These have to be done first to
+    // avoid races with Hypervisors setting special wake-ups
+
+    // SGPE and PGPE.
+    scom_data.words.upper = BIT32(OCCFLG2_SGPE_HCODE_STOP_REQ_ERR_INJ) |
+                            BIT32(OCCFLG2_PGPE_HCODE_PSTATE_REQ_ERR_INJ) |
+                            BIT32(OCCFLG2_PGPE_HCODE_FIT_ERR_INJ);
+    out64(OCB_OCCFLG2_CLR, scom_data.value);
+
+    // CMEs
+    // The in-progress bit are not checked as this is an error case
+    ccsr.value = in32(OCB_CCSR);
+
+    for(qloop = 0;
+        qloop < MAX_QUADS;
+        qloop++)
+    {
+        for(cloop = 0; cloop < CORES_PER_QUAD; cloop++)
+        {
+            scom_data.value = 0;
+            cindex = (qloop << 2) + cloop;
+            m_1c = BIT32(cindex);
+
+            if(ccsr.value & m_1c)
+            {
+                PK_TRACE_DBG("core[%d][%d] error inject being cleared", qloop, cindex);
+                scom_data.words.upper = BIT32(CPPM_CSAR_FIT_HCODE_ERROR_INJECT) |
+                                        BIT32(CPPM_CSAR_PSTATE_HCODE_ERROR_INJECT) |
+                                        BIT32(CPPM_CSAR_STOP_HCODE_ERROR_INJECT);
+                GPE_PUTSCOM(GPE_SCOM_ADDR_CORE(CPPM_CSAR_CLR, cindex), scom_data.value);
+            }
+        }
+    }
+
     if (in32(OCB_OCCFLG2) & BIT32(STOP_RECOVERY_TRIGGER_ENABLE))
     {
-        uint64_t scom_data = 0;
         PK_TRACE_INF("WARNING: STOP RECOVERY TRIGGER!");
 
-        PK_TRACE("1. Set ADU lock for exclusive use for a timeout of 500ms.");
+        // Has to be set first to avoid races with Hypervisors reading
+        // this bit
+        PK_TRACE("1. Set OCCFLG2[28] PM Callout Active");
+        out32(OCB_OCCFLG2_OR, BIT32(PM_CALLOUT_ACTIVE));
+
+        PK_TRACE("2. Set ADU lock for exclusive use for a timeout of 500ms.");
 
         do
         {
-
             GPE_PUTSCOM(0x90001, 0x0010000000000000ull);
-            GPE_GETSCOM(0x90001, scom_data);
-
+            GPE_GETSCOM(0x90001, scom_data.value);
         }
-        while (!(scom_data  &  0x0010000000000000ull));
+        while (!(scom_data.value  &  0x0010000000000000ull));
 
-        PK_TRACE("2. Cleanup/reset ADU");
+        PK_TRACE("3. Cleanup/reset ADU");
+        // 3      Clear Status
+        // 4      Reset FSM
+        // 11     Locked
         GPE_PUTSCOM(0x90001, 0x1810000000000000ull);
 
-        PK_TRACE("3. Setup PowerBus 'address' field for malf alert");
+        PK_TRACE("4. Setup PowerBus 'address' field for malf alert");
+        // 10     Set Malfunction Alert Error
         GPE_PUTSCOM(0x90000, 0x0000100000000000ull);
 
         PK_TRACE("4. Setup PowerBus command type and launch malfunction");
+        // 2      Start Op
+        // 6      AX Type
+        // 11     Locked
+        // 16:18  Scope 16:18 = VG = 101.
+        // 25:31  TType = 011 0100
+        // 32:39  TSize = 0000 0100 = 2B
         GPE_PUTSCOM(0x90001, 0x2210A03104000000ull);
 
-        PK_TRACE("5. Set OCCFLG2[28] PM Callout Active");
-        out32(OCB_OCCFLG2_OR, BIT32(PM_CALLOUT_ACTIVE));
+        PK_TRACE("5. Cleanup/reset ADU");
+        // See above
+        GPE_PUTSCOM(0x90001, 0x1810000000000000ull);
+
+        PK_TRACE("6: Unlock ADU");
+        // 11      Clear to remove lock
+        GPE_PUTSCOM(0x90001, 0x0000000000000000ull);
     }
 }
diff --git a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe.h b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe.h
index 38fbbed2..2c733d7c 100644
--- a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe.h
+++ b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe.h
@@ -129,6 +129,7 @@ extern uint32_t G_OCB_OCCFLG;
 extern uint32_t G_OCB_OCCFLG_OR;
 extern uint32_t G_OCB_OCCFLG_CLR;
 extern uint32_t G_OCB_OCCFLG2;
+extern uint32_t G_OCB_OCCFLG2_CLR;
 extern uint32_t G_OCB_OISR0_CLR;
 extern uint32_t G_OCB_OIMR1_OR;
 extern uint32_t G_OCB_OIMR1_CLR;
diff --git a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_fit.c b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_fit.c
index 93ded7c6..849e330c 100644
--- a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_fit.c
+++ b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_fit.c
@@ -244,6 +244,8 @@ __attribute__((always_inline)) inline void handle_occflg_requests()
 
     if(in32(G_OCB_OCCFLG2) & BIT32(OCCFLG2_PGPE_HCODE_FIT_ERR_INJ))
     {
+        // Clear the injection so things are not permenently stuck
+        out32(G_OCB_OCCFLG2_CLR, BIT32(OCCFLG2_PGPE_HCODE_FIT_ERR_INJ));
         PK_TRACE_ERR("FIT_IPC_ERROR_INJECT TRAP");
         PK_PANIC(PGPE_SET_PMCR_TRAP_INJECT);
     }
diff --git a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_ipc_handlers.c b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_ipc_handlers.c
index 740bb85e..14cdd90e 100644
--- a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_ipc_handlers.c
+++ b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_ipc_handlers.c
@@ -175,6 +175,8 @@ void p9_pgpe_ipc_405_set_pmcr(ipc_msg_t* cmd, void* arg)
 
     if(in32(G_OCB_OCCFLG2) & BIT32(OCCFLG2_PGPE_HCODE_PSTATE_REQ_ERR_INJ))
     {
+        // Clear the injection so things are not permenently stuck
+        out32(G_OCB_OCCFLG2_CLR, BIT32(OCCFLG2_PGPE_HCODE_PSTATE_REQ_ERR_INJ));
         PK_TRACE_ERR("SET PMCR IPC ERROR INJECT TRAP");
         PK_PANIC(PGPE_SET_PMCR_TRAP_INJECT);
     }
diff --git a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_irq_handlers.c b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_irq_handlers.c
index 3234c59a..17d66197 100644
--- a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_irq_handlers.c
+++ b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_irq_handlers.c
@@ -306,6 +306,8 @@ void p9_pgpe_irq_handler_pcb_type1(void* arg, PkIrqId irq)
     //If error injection bit is set in OCC Scratch 2, then halt PGPE immediately
     if(in32(G_OCB_OCCFLG2) & BIT32(OCCFLG2_PGPE_HCODE_PSTATE_REQ_ERR_INJ))
     {
+        // Clear the injection so things are not permenently stuck
+        out32(G_OCB_OCCFLG2_CLR, BIT32(OCCFLG2_PGPE_HCODE_PSTATE_REQ_ERR_INJ));
         PK_TRACE_ERR("PCB TYPE1 ERROR INJECT TRAP");
         PK_PANIC(PGPE_SET_PMCR_TRAP_INJECT);
     }
diff --git a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_main.c b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_main.c
index 76450317..6c4c0b26 100644
--- a/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_main.c
+++ b/import/chips/p9/procedures/ppe_closed/pgpe/pstate_gpe/p9_pgpe_main.c
@@ -91,6 +91,7 @@ uint32_t G_OCB_OCCFLG = OCB_OCCFLG;
 uint32_t G_OCB_OCCFLG_OR = OCB_OCCFLG_OR;
 uint32_t G_OCB_OCCFLG_CLR = OCB_OCCFLG_CLR;
 uint32_t G_OCB_OCCFLG2 = OCB_OCCFLG2;
+uint32_t G_OCB_OCCFLG2_CLR = OCB_OCCFLG2_CLR;
 uint32_t G_OCB_OISR0_CLR = OCB_OISR0_CLR;
 uint32_t G_OCB_OIMR0_OR = OCB_OIMR0_OR;
 uint32_t G_OCB_OIMR1_OR = OCB_OIMR1_OR;
diff --git a/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_main.C b/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_main.C
index 825ae9c0..792ffbc9 100644
--- a/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_main.C
+++ b/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_main.C
@@ -44,6 +44,7 @@ uint32_t G_OCB_OCCFLG       = OCB_OCCFLG;
 uint32_t G_OCB_OCCFLG_CLR   = OCB_OCCFLG_CLR;
 uint32_t G_OCB_OCCFLG_OR    = OCB_OCCFLG_OR;
 uint32_t G_OCB_OCCFLG2      = OCB_OCCFLG2;
+uint32_t G_OCB_OCCFLG2_CLR  = OCB_OCCFLG2_CLR;
 uint32_t G_OCB_OCCS2        = OCB_OCCS2;
 uint32_t G_OCB_OISR0_CLR    = OCB_OISR0_CLR;
 uint32_t G_OCB_OISR1        = OCB_OISR1;
diff --git a/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop.h b/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop.h
index d4ef5dff..6365b932 100644
--- a/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop.h
+++ b/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop.h
@@ -80,6 +80,7 @@ extern uint32_t G_OCB_OCCFLG;
 extern uint32_t G_OCB_OCCFLG_CLR;
 extern uint32_t G_OCB_OCCFLG_OR;
 extern uint32_t G_OCB_OCCFLG2;
+extern uint32_t G_OCB_OCCFLG2_CLR;
 extern uint32_t G_OCB_OCCS2;
 extern uint32_t G_OCB_OISR0_CLR;
 extern uint32_t G_OCB_OISR1;
diff --git a/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop_entry.c b/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop_entry.c
index 2c92bdb4..10ddfb34 100644
--- a/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop_entry.c
+++ b/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop_entry.c
@@ -71,6 +71,8 @@ p9_sgpe_stop_entry()
 
     if( in32(G_OCB_OCCFLG2) & BIT32(OCCFLG2_SGPE_HCODE_STOP_REQ_ERR_INJ))
     {
+        // Clear the injection so things are not permenently stuck
+        out32(G_OCB_OCCFLG2_CLR, BIT32(OCCFLG2_SGPE_HCODE_STOP_REQ_ERR_INJ));
         PK_TRACE_ERR("SGPE STOP ENTRY ERROR INJECT TRAP");
         PK_PANIC(SGPE_STOP_ENTRY_TRAP_INJECT);
     }
diff --git a/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop_exit.c b/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop_exit.c
index bca3a1de..c09baae9 100644
--- a/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop_exit.c
+++ b/import/chips/p9/procedures/ppe_closed/sgpe/stop_gpe/p9_sgpe_stop_exit.c
@@ -362,6 +362,8 @@ p9_sgpe_stop_exit()
 
     if(in32(G_OCB_OCCFLG2) & BIT32(OCCFLG2_SGPE_HCODE_STOP_REQ_ERR_INJ))
     {
+        // Clear the injection so things are not permenently stuck
+        out32(G_OCB_OCCFLG2_CLR, BIT32(OCCFLG2_SGPE_HCODE_STOP_REQ_ERR_INJ));
         PK_TRACE_ERR("SGPE STOP EXIT ERROR INJECT TRAP");
         PK_PANIC(SGPE_STOP_EXIT_TRAP_INJECT);
     }