northbridge/amd/amdmct: Clear memory before enabling ECC

The existing code enabled ECC before clearing memory.  As the
AMD CPUs will generate MCEs on any invalid check bits, this
resulted in random lockups during memory training due to the
uniniailized check bits.

Initialize ECC check bits before enabling ECC hardware.

Change-Id: I992e7040520570893ba6a213138dd57bfa14733b
Signed-off-by: Timothy Pearson <tpearson@raptorengineeringinc.com>
Reviewed-on: http://review.coreboot.org/11996
Tested-by: build bot (Jenkins)
Reviewed-by: Martin Roth <martinroth@google.com>

src/northbridge/amd/amdmct/mct_ddr3/mct_d.c

@@ -47,8 +47,6 @@ static void DCTMemClr_Init_D(struct MCTStatStruc *pMCTstat,
 				struct DCTStatStruc *pDCTstat);
 static void DCTMemClr_Sync_D(struct MCTStatStruc *pMCTstat,
 				struct DCTStatStruc *pDCTstat);
-static void MCTMemClrSync_D(struct MCTStatStruc *pMCTstat,
-				struct DCTStatStruc *pDCTstatA);
 static u8 NodePresent_D(u8 Node);
 static void SyncDCTsReady_D(struct MCTStatStruc *pMCTstat,
 				struct DCTStatStruc *pDCTstatA);
@@ -1507,10 +1505,11 @@ restartinit:
 		InterleaveChannels_D(pMCTstat, pDCTstatA);
 
 		printk(BIOS_DEBUG, "mctAutoInitMCT_D: ECCInit_D\n");
-		if (ECCInit_D(pMCTstat, pDCTstatA)) {		/* Setup ECC control and ECC check-bits*/
+		ECCInit_D(pMCTstat, pDCTstatA);			/* Setup ECC control and ECC check-bits*/
-			printk(BIOS_DEBUG, "mctAutoInitMCT_D: MCTMemClr_D\n");
+
-			MCTMemClr_D(pMCTstat,pDCTstatA);
+		/* mctDoWarmResetMemClr_D(); */
-		}
+		printk(BIOS_DEBUG, "mctAutoInitMCT_D: MCTMemClr_D\n");
+		MCTMemClr_D(pMCTstat,pDCTstatA);
 
 		printk(BIOS_DEBUG, "mctAutoInitMCT_D: mct_ForceNBPState0_Dis_Fam15\n");
 		for (Node = 0; Node < MAX_NODES_SUPPORTED; Node++) {
@@ -2102,9 +2101,6 @@ static void DQSTiming_D(struct MCTStatStruc *pMCTstat,
 
 	/* FIXME - currently uses calculated value	TrainMaxReadLatency_D(pMCTstat, pDCTstatA); */
 	mctHookAfterAnyTraining();
-
-	/* mctDoWarmResetMemClr_D(); */
-	MCTMemClr_D(pMCTstat, pDCTstatA);
 }
 
 static void LoadDQSSigTmgRegs_D(struct MCTStatStruc *pMCTstat,
@@ -2392,26 +2388,6 @@ static void DCTMemClr_Init_D(struct MCTStatStruc *pMCTstat,
 	}
 }
 
-static void MCTMemClrSync_D(struct MCTStatStruc *pMCTstat,
-				struct DCTStatStruc *pDCTstatA)
-{
-	/* Ensures that memory clear has completed on all node.*/
-	u8 Node;
-	struct DCTStatStruc *pDCTstat;
-
-	if (!mctGet_NVbits(NV_DQSTrainCTL)){
-		/* callback to wrapper: mctDoWarmResetMemClr_D */
-	} else {	/* NV_DQSTrainCTL == 1 */
-		for (Node = 0; Node < MAX_NODES_SUPPORTED; Node++) {
-			pDCTstat = pDCTstatA + Node;
-
-			if (pDCTstat->NodePresent) {
-				DCTMemClr_Sync_D(pMCTstat, pDCTstat);
-			}
-		}
-	}
-}
-
 static void DCTMemClr_Sync_D(struct MCTStatStruc *pMCTstat,
 				struct DCTStatStruc *pDCTstat)
 {
@@ -3097,6 +3073,8 @@ static void GetPresetmaxF_D(struct MCTStatStruc *pMCTstat,
 	u16 proposedFreq;
 	u16 word;
 
+	printk(BIOS_DEBUG, "%s: Start\n", __func__);
+
 	/* Get CPU Si Revision defined limit (NPT) */
 	if (is_fam15h())
 		proposedFreq = 933;
@@ -3121,6 +3099,8 @@ static void GetPresetmaxF_D(struct MCTStatStruc *pMCTstat,
 		pDCTstat->PresetmaxFreq = word;
 	}
 	/* Check F3xE8[DdrMaxRate] for maximum DRAM data rate support */
+
+	printk(BIOS_DEBUG, "%s: Done\n", __func__);
 }
 
 static void SPDGetTCL_D(struct MCTStatStruc *pMCTstat,

src/northbridge/amd/amdmct/mct_ddr3/mctecc_d.c

@@ -84,6 +84,10 @@ u8 ECCInit_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA)
 	u32 val;
 	u16 nvbits;
 
+	uint32_t dword;
+	uint8_t sync_flood_on_dram_err[MAX_NODES_SUPPORTED];
+	uint8_t sync_flood_on_any_uc_err[MAX_NODES_SUPPORTED];
+
 	mctHookBeforeECC();
 
 	/* Construct these booleans, based on setup options, for easy handling
@@ -107,6 +111,25 @@ u8 ECCInit_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA)
 	nvbits = mctGet_NVbits(NV_DramBKScrub);
 	OF_ScrubCTL |= nvbits;
 
+	/* Prevent lockups on DRAM errors during ECC init */
+	for (Node = 0; Node < MAX_NODES_SUPPORTED; Node++) {
+		struct DCTStatStruc *pDCTstat;
+		pDCTstat = pDCTstatA + Node;
+
+		if (NodePresent_D(Node)) {
+			dword = Get_NB32(pDCTstat->dev_nbmisc, 0x44);
+			sync_flood_on_dram_err[Node] = (dword >> 30) & 0x1;
+			sync_flood_on_any_uc_err[Node] = (dword >> 21) & 0x1;
+			dword &= ~(0x1 << 30);
+			dword &= ~(0x1 << 21);
+			Set_NB32(pDCTstat->dev_nbmisc, 0x44, dword);
+
+			/* Clear the RAM before enabling ECC to prevent MCE-related lockups */
+			DCTMemClr_Init_D(pMCTstat, pDCTstat);
+			DCTMemClr_Sync_D(pMCTstat, pDCTstat);
+		}
+	}
+
 	AllECC = 1;
 	MemClrECC = 0;
 	for (Node = 0; Node < MAX_NODES_SUPPORTED; Node++) {
@@ -153,7 +176,7 @@ u8 ECCInit_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA)
 			}	/* Node has Dram */
 
 			if (MemClrECC) {
-				MCTMemClrSync_D(pMCTstat, pDCTstatA);
+				DCTMemClr_Sync_D(pMCTstat, pDCTstat);
 			}
 
 			if (pDCTstat->LogicalCPUID & (AMD_DR_GT_D0 | AMD_FAM15_ALL)) {
@@ -166,6 +189,19 @@ u8 ECCInit_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA)
 		}	/* if Node present */
 	}
 
+	/* Restore previous MCA error handling settings */
+	for (Node = 0; Node < MAX_NODES_SUPPORTED; Node++) {
+		struct DCTStatStruc *pDCTstat;
+		pDCTstat = pDCTstatA + Node;
+
+		if (NodePresent_D(Node)) {
+			dword = Get_NB32(pDCTstat->dev_nbmisc, 0x44);
+			dword |= (sync_flood_on_dram_err[Node] & 0x1) << 30;
+			dword |= (sync_flood_on_any_uc_err[Node] & 0x1) << 21;
+			Set_NB32(pDCTstat->dev_nbmisc, 0x44, dword);
+		}
+	}
+
 	if(AllECC)
 		pMCTstat->GStatus |= 1<<GSB_ECCDIMMs;
 	else