northbridge/amd/amdmct/mct_ddr3: Fix lockups and wasted time during ECC init

The existing ECC initialization algorithm contained several bugs on both
Family 10h and Family 15h processors, including activation of ECC scrub
before DRAM setup was completed, in violation of both BKDG and errata
recommendations.

Change-Id: I09a8ea83024186b7ece7d78a4bef1201ab34ff8a
Signed-off-by: Timothy Pearson <tpearson@raptorengineeringinc.com>
Reviewed-on: http://review.coreboot.org/12002
Tested-by: build bot (Jenkins)
Reviewed-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>

4 files changed, 147 insertions, 61 deletions

diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c b/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c
index 1312b34bce..24d6c6c8a2 100644
--- a/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c
+++ b/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c
@@ -1517,11 +1517,12 @@ restartinit:
 		InterleaveChannels_D(pMCTstat, pDCTstatA);
 
 		printk(BIOS_DEBUG, "mctAutoInitMCT_D: ECCInit_D\n");
-		ECCInit_D(pMCTstat, pDCTstatA);			/* Setup ECC control and ECC check-bits*/
-
-		/* mctDoWarmResetMemClr_D(); */
-		printk(BIOS_DEBUG, "mctAutoInitMCT_D: MCTMemClr_D\n");
-		MCTMemClr_D(pMCTstat,pDCTstatA);
+		if (!ECCInit_D(pMCTstat, pDCTstatA)) {			/* Setup ECC control and ECC check-bits*/
+			/* Memory was not cleared during ECC setup */
+			/* 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++) {
@@ -1714,7 +1715,6 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat,
 		uint8_t x8_present = 0;
 		uint8_t memclk_index;
 		uint8_t interleave_channels = 0;
-		uint8_t redirect_ecc_scrub = 0;
 		uint16_t trdrdsddc;
 		uint16_t trdrddd;
 		uint16_t cdd_trdrddd;
@@ -1752,9 +1752,6 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat,
 		if (pDCTstat->DIMMValidDCT[0] && pDCTstat->DIMMValidDCT[1] && mctGet_NVbits(NV_Unganged))
 			interleave_channels = 1;
 
-		if ((pMCTstat->GStatus & 1 << GSB_ECCDIMMs) && mctGet_NVbits(NV_ECCRedir))
-			redirect_ecc_scrub = 1;
-
 		dword = (Get_NB32_DCT(dev, dct, 0x240) >> 4) & 0xf;
 		if (dword > 6)
 			read_odt_delay = dword - 6;
@@ -1946,21 +1943,10 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat,
 		dword |= (interleave_channels & 0x1) << 2;
 		Set_NB32_DCT(dev, dct, 0x110, dword);			/* DRAM Controller Select Low */
 
-		dword = Get_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x58);	/* Scrub Rate Control */
-		dword &= ~(0x1f << 24);					/* L3Scrub = NV_L3BKScrub */
-		dword |= (mctGet_NVbits(NV_L3BKScrub) & 0x1f) << 24;
-		dword &= ~(0x1f);					/* DramScrub = NV_DramBKScrub */
-		dword |= mctGet_NVbits(NV_DramBKScrub) & 0x1f;
-		Set_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x58, dword);	/* Scrub Rate Control */
-
-		dword = Get_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x5c);	/* DRAM Scrub Address Low */
-		dword &= ~(0x1);					/* ScrubReDirEn = redirect_ecc_scrub */
-		dword |= redirect_ecc_scrub & 0x1;
-		Set_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x5c, dword);	/* DRAM Scrub Address Low */
-
-		dword = Get_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x1b8);	/* L3 Control 1 */
-		dword &= ~(0x1 << 4);					/* L3ScrbRedirDis = 0 */
-		Set_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x1b8, dword);	/* L3 Control 1 */
+		/* NOTE
+		 * ECC-related setup is performed as part of ECCInit_D and must not be located here,
+		 * otherwise semi-random lockups will occur due to misconfigured scrubbing hardware!
+		 */
 
 		/* FIXME
 		 * The BKDG-recommended settings cause memory corruption on the ASUS KGPE-D16.
@@ -2002,11 +1988,17 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat,
 		dword |= ((((dword >> 8) & 0x1f) + 1) << 16);
 		Set_NB32_DCT(dev, dct, 0x21c, dword);			/* DRAM Timing 6 */
 
+		/* Configure partial power down delay */
+		dword = Get_NB32(dev, 0x244);				/* DRAM Controller Miscellaneous 3 */
+		dword &= ~0xf;						/* PrtlChPDDynDly = 0x2 */
+		dword |= 0x2;
+		Set_NB32(dev, 0x244, dword);				/* DRAM Controller Miscellaneous 3 */
+
 		/* Enable prefetchers */
-		dword = Get_NB32_DCT(dev, dct, 0x110);			/* Memory Controller Configuration High */
+		dword = Get_NB32(dev, 0x11c);				/* Memory Controller Configuration High */
 		dword &= ~(0x1 << 13);					/* PrefIoDis = 0 */
 		dword &= ~(0x1 << 12);					/* PrefCpuDis = 0 */
-		Set_NB32_DCT(dev, dct, 0x110, dword);			/* Memory Controller Configuration High */
+		Set_NB32(dev, 0x11c, dword);				/* Memory Controller Configuration High */
 	}
 }
 
@@ -2111,6 +2103,19 @@ static void DQSTiming_D(struct MCTStatStruc *pMCTstat,
 		pMCTstat->GStatus |= 1 << GSB_ConfigRestored;
 	}
 
+	if (is_fam15h()) {
+		uint8_t Node;
+		struct DCTStatStruc *pDCTstat;
+
+		/* Switch DCT control register to DCT 0 per Erratum 505 */
+		for (Node = 0; Node < MAX_NODES_SUPPORTED; Node++) {
+			pDCTstat = pDCTstatA + Node;
+			if (pDCTstat->NodePresent) {
+				fam15h_switch_dct(pDCTstat->dev_map, 0);
+			}
+		}
+	}
+
 	/* FIXME - currently uses calculated value	TrainMaxReadLatency_D(pMCTstat, pDCTstatA); */
 	mctHookAfterAnyTraining();
 }
@@ -2357,6 +2362,7 @@ static void MCTMemClr_D(struct MCTStatStruc *pMCTstat,
 	 * status are checked to ensure that memclr has completed.
 	 */
 	u8 Node;
+	uint32_t dword;
 	struct DCTStatStruc *pDCTstat;
 
 	if (!mctGet_NVbits(NV_DQSTrainCTL)){
@@ -2377,6 +2383,18 @@ static void MCTMemClr_D(struct MCTStatStruc *pMCTstat,
 			}
 		}
 	}
+
+	for (Node = 0; Node < MAX_NODES_SUPPORTED; Node++) {
+		pDCTstat = pDCTstatA + Node;
+
+		/* Configure and enable prefetchers */
+		if (is_fam15h())
+			dword = 0x0ce00f41;	/* BKDG recommended */
+		else
+			dword = 0x0fe40fc0;	/* BKDG recommended */
+		dword |= MCCH_FlushWrOnStpGnt;	/* Set for S3 */
+		Set_NB32(pDCTstat->dev_dct, 0x11c, dword);
+	}
 }
 
 static void DCTMemClr_Init_D(struct MCTStatStruc *pMCTstat,
@@ -2384,48 +2402,59 @@ static void DCTMemClr_Init_D(struct MCTStatStruc *pMCTstat,
 {
 	u32 val;
 	u32 dev;
-	u32 reg;
+	uint32_t dword;
 
 	/* Initiates a memory clear operation on one node */
 	if (pDCTstat->DCTSysLimit) {
 		dev = pDCTstat->dev_dct;
-		reg = 0x110;
+
+		/* Disable prefetchers */
+		dword = Get_NB32(dev, 0x11c);		/* Memory Controller Configuration High */
+		dword |= 0x1 << 13;			/* PrefIoDis = 1 */
+		dword |= 0x1 << 12;			/* PrefCpuDis = 1 */
+		Set_NB32(dev, 0x11c, dword);		/* Memory Controller Configuration High */
 
 		do {
-			val = Get_NB32(dev, reg);
+			val = Get_NB32(dev, 0x110);
 		} while (val & (1 << MemClrBusy));
 
 		val |= (1 << MemClrInit);
-		Set_NB32(dev, reg, val);
+		Set_NB32(dev, 0x110, val);
 	}
 }
 
 static void DCTMemClr_Sync_D(struct MCTStatStruc *pMCTstat,
 				struct DCTStatStruc *pDCTstat)
 {
-	u32 val;
-	u32 dev = pDCTstat->dev_dct;
-	u32 reg;
+	uint32_t dword;
+	uint32_t dev = pDCTstat->dev_dct;
+
+	printk(BIOS_DEBUG, "%s: Start\n", __func__);
 
 	/* Ensure that a memory clear operation has completed on one node */
 	if (pDCTstat->DCTSysLimit){
-		reg = 0x110;
-
+		printk(BIOS_DEBUG, "%s: Waiting for memory clear to complete", __func__);
 		do {
-			val = Get_NB32(dev, reg);
-		} while (val & (1 << MemClrBusy));
+			dword = Get_NB32(dev, 0x110);
+
+			printk(BIOS_DEBUG, ".");
+		} while (dword & (1 << MemClrBusy));
 
+		printk(BIOS_DEBUG, "\n");
 		do {
-			val = Get_NB32(dev, reg);
-		} while (!(val & (1 << Dr_MemClrStatus)));
+			printk(BIOS_DEBUG, ".");
+			dword = Get_NB32(dev, 0x110);
+		} while (!(dword & (1 << Dr_MemClrStatus)));
+		printk(BIOS_DEBUG, "\n");
 	}
 
-	if (is_fam15h())
-		val = 0x0ce00f41;	/* BKDG recommended */
-	else
-		val = 0x0fe40fc0;	/* BKDG recommended */
-	val |= MCCH_FlushWrOnStpGnt;	/* Set for S3 */
-	Set_NB32(dev, 0x11c, val);
+	/* Enable prefetchers */
+	dword = Get_NB32(dev, 0x11c);		/* Memory Controller Configuration High */
+	dword &= ~(0x1 << 13);			/* PrefIoDis = 0 */
+	dword &= ~(0x1 << 12);			/* PrefCpuDis = 0 */
+	Set_NB32(dev, 0x11c, dword);		/* Memory Controller Configuration High */
+
+	printk(BIOS_DEBUG, "%s: Done\n", __func__);
 }
 
 static u8 NodePresent_D(u8 Node)
@@ -3366,8 +3395,8 @@ static u8 AutoConfig_D(struct MCTStatStruc *pMCTstat,
 	dev = pDCTstat->dev_dct;
 
 	/* Build Dram Control Register Value */
-	DramConfigMisc2 = Get_NB32_DCT(dev, dct, 0xA8);		/* Dram Control*/
-	DramControl = Get_NB32_DCT(dev, dct, 0x78);		/* Dram Control*/
+	DramConfigMisc2 = Get_NB32_DCT(dev, dct, 0xa8);		/* Dram Miscellaneous 2 */
+	DramControl = Get_NB32_DCT(dev, dct, 0x78);		/* Dram Control */
 
 	/* FIXME: Skip mct_checkForDxSupport */
 	/* REV_CALL mct_DoRdPtrInit if not Dx */
@@ -3422,9 +3451,15 @@ static u8 AutoConfig_D(struct MCTStatStruc *pMCTstat,
 			/* set only if x8 Registered DIMMs in System*/
 			DramConfigHi |= 1 << RDqsEn;
 
-	if (mctGet_NVbits(NV_CKE_CTL))
-		/*Chip Select control of CKE*/
-		DramConfigHi |= 1 << 16;
+	if (pDCTstat->LogicalCPUID & AMD_FAM15_ALL) {
+		DramConfigLo |= 1 << 25;	/* PendRefPaybackS3En = 1 */
+		DramConfigLo |= 1 << 24;	/* StagRefEn = 1 */
+		DramConfigHi |= 1 << 16;	/* PowerDownMode = 1 */
+	} else {
+		if (mctGet_NVbits(NV_CKE_CTL))
+			/*Chip Select control of CKE*/
+			DramConfigHi |= 1 << 16;
+	}
 
 	/* Control Bank Swizzle */
 	if (0) /* call back not needed mctBankSwizzleControl_D()) */
@@ -4132,8 +4167,7 @@ static void mct_preInitDCT(struct MCTStatStruc *pMCTstat,
 
 	if (load_spd_hashes_from_nvram(pMCTstat, pDCTstat) < 0) {
 		pDCTstat->spd_data.nvram_spd_match = 0;
-	}
-	else {
+	} else {
 		compare_nvram_spd_hashes(pMCTstat, pDCTstat);
 	}
 #else
@@ -4331,8 +4365,8 @@ static u8 mct_PlatformSpec(struct MCTStatStruc *pMCTstat,
 	}
 	for (i=i_start; i<i_end; i++) {
 		index_reg = 0x98;
-		Set_NB32_index_wait_DCT(dev, i, index_reg, 0x00, pDCTstat->CH_ODC_CTL[i]); /* Channel A Output Driver Compensation Control */
-		Set_NB32_index_wait_DCT(dev, i, index_reg, 0x04, pDCTstat->CH_ADDR_TMG[i]); /* Channel A Output Driver Compensation Control */
+		Set_NB32_index_wait_DCT(dev, i, index_reg, 0x00, pDCTstat->CH_ODC_CTL[i]); /* Channel A/B Output Driver Compensation Control */
+		Set_NB32_index_wait_DCT(dev, i, index_reg, 0x04, pDCTstat->CH_ADDR_TMG[i]); /* Channel A/B Output Driver Compensation Control */
 	}
 
 	return pDCTstat->ErrCode;
@@ -6130,11 +6164,11 @@ void ProgDramMRSReg_D(struct MCTStatStruc *pMCTstat,
 		DramMRS |= 1 << 1;
 
 	dword = Get_NB32_DCT(pDCTstat->dev_dct, dct, 0x84);
+	dword |= DramMRS;
 	if (is_fam15h())
 		dword &= ~0x00800003;
 	else
 		dword &= ~0x00fc2f8f;
-	dword |= DramMRS;
 	Set_NB32_DCT(pDCTstat->dev_dct, dct, 0x84, dword);
 }
 
diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mctecc_d.c b/src/northbridge/amd/amdmct/mct_ddr3/mctecc_d.c
index 8701baebf7..1be46b1351 100644
--- a/src/northbridge/amd/amdmct/mct_ddr3/mctecc_d.c
+++ b/src/northbridge/amd/amdmct/mct_ddr3/mctecc_d.c
@@ -88,8 +88,13 @@ u8 ECCInit_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA)
 	uint8_t sync_flood_on_dram_err[MAX_NODES_SUPPORTED];
 	uint8_t sync_flood_on_any_uc_err[MAX_NODES_SUPPORTED];
 
+	uint8_t redirect_ecc_scrub = 0;
+
 	mctHookBeforeECC();
 
+	if ((pMCTstat->GStatus & 1 << GSB_ECCDIMMs) && mctGet_NVbits(NV_ECCRedir))
+		redirect_ecc_scrub = 1;
+
 	/* Construct these booleans, based on setup options, for easy handling
 	later in this procedure */
 	OB_NBECC = mctGet_NVbits(NV_NBECC);			/* MCA ECC (MCE) enable bit */
@@ -226,12 +231,12 @@ u8 ECCInit_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA)
 					}
 					dev = pDCTstat->dev_nbmisc;
 					val = curBase << 8;
-					if(OB_ECCRedir) {
-						val |= (1<<0); /* enable redirection */
+					if (OB_ECCRedir) {
+						val |= (1<<0);			/* enable redirection */
 					}
-					Set_NB32(dev, 0x5C, val); /* Dram Scrub Addr Low */
+					Set_NB32(dev, 0x5c, val);		/* Dram Scrub Addr Low */
 					val = curBase>>24;
-					Set_NB32(dev, 0x60, val); /* Dram Scrub Addr High */
+					Set_NB32(dev, 0x60, val);		/* Dram Scrub Addr High */
 					Set_NB32(dev, 0x58, OF_ScrubCTL);	/*Scrub Control */
 
 					if (!is_fam15h()) {
@@ -248,6 +253,32 @@ u8 ECCInit_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA)
 							}
 						}
 					}
+
+					if (is_fam15h()) {
+						uint8_t dct;
+
+						/* Disable training mode
+						 * See fam15EnableTrainingMode for the non-ECC training mode tear-down code
+						 */
+						for (dct = 0; dct < 2; dct++) {
+							/* NOTE: Reads use DCT 0 and writes use the current DCT per Erratum 505 */
+							dword = Get_NB32_DCT(pDCTstat->dev_nbmisc, 0, 0x58);	/* Scrub Rate Control */
+							dword &= ~(0x1f << 24);					/* L3Scrub = NV_L3BKScrub */
+							dword |= (mctGet_NVbits(NV_L3BKScrub) & 0x1f) << 24;
+							dword &= ~(0x1f);					/* DramScrub = NV_DramBKScrub */
+							dword |= mctGet_NVbits(NV_DramBKScrub) & 0x1f;
+							Set_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x58, dword);	/* Scrub Rate Control */
+
+							dword = Get_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x5c);	/* DRAM Scrub Address Low */
+							dword &= ~(0x1);					/* ScrubReDirEn = redirect_ecc_scrub */
+							dword |= redirect_ecc_scrub & 0x1;
+							Set_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x5c, dword);	/* DRAM Scrub Address Low */
+
+							dword = Get_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x1b8);	/* L3 Control 1 */
+							dword &= ~(0x1 << 4);					/* L3ScrbRedirDis = 0 */
+							Set_NB32_DCT(pDCTstat->dev_nbmisc, dct, 0x1b8, dword);	/* L3 Control 1 */
+						}
+					}
 				}	/* this node has ECC enabled dram */
 			}	/*Node has Dram */
 		}	/*if Node present */
diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mctproc.c b/src/northbridge/amd/amdmct/mct_ddr3/mctproc.c
index a86c319a88..77acaec3a4 100644
--- a/src/northbridge/amd/amdmct/mct_ddr3/mctproc.c
+++ b/src/northbridge/amd/amdmct/mct_ddr3/mctproc.c
@@ -19,7 +19,27 @@ u32 mct_SetDramConfigMisc2(struct DCTStatStruc *pDCTstat, u8 dct, u32 misc2)
 {
 	u32 val;
 
-	if (pDCTstat->LogicalCPUID & (AMD_DR_Dx | AMD_DR_Cx)) {
+	/* FIXME
+	 * Mainboards need to be able to specify the maximum number of DIMMs installable per channel
+	 * For now assume a maximum of 2 DIMMs per channel can be installed
+	 */
+	uint8_t MaxDimmsInstallable = 2;
+
+	if (pDCTstat->LogicalCPUID & AMD_FAM15_ALL) {
+		uint8_t cs_mux_45;
+		uint8_t cs_mux_67;
+
+		/* BKDG v3.14 Table 200 / Table 201 */
+		if (MaxDimmsInstallable < 3) {
+			cs_mux_45 = 1;
+			cs_mux_67 = 1;
+		} else {
+			cs_mux_45 = 0;
+			cs_mux_67 = 0;
+		}
+		misc2 |= (cs_mux_45 & 0x1) << 26;
+		misc2 |= (cs_mux_67 & 0x1) << 27;
+	} else if (pDCTstat->LogicalCPUID & (AMD_DR_Dx | AMD_DR_Cx)) {
 		if (pDCTstat->Status & (1 << SB_Registered)) {
 			misc2 |= 1 << SubMemclkRegDly;
 			if (mctGet_NVbits(NV_MAX_DIMMS) == 8)
diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mhwlc_d.c b/src/northbridge/amd/amdmct/mct_ddr3/mhwlc_d.c
index f3915a2438..48b72caf36 100644
--- a/src/northbridge/amd/amdmct/mct_ddr3/mhwlc_d.c
+++ b/src/northbridge/amd/amdmct/mct_ddr3/mhwlc_d.c
@@ -205,7 +205,8 @@ void AgesaHwWlPhase2(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTsta
 			uint16_t total_delay_seed = ((pDCTData->WLSeedGrossDelay[index+ByteLane] & 0x1f) << 5) | (pDCTData->WLSeedFineDelay[index+ByteLane] & 0x1f);
 			uint16_t total_delay_phy = ((pDCTData->WLGrossDelay[index+ByteLane] & 0x1f) << 5) | (pDCTData->WLFineDelay[index+ByteLane] & 0x1f);
 			if (abs(total_delay_phy - total_delay_seed) > 0x20) {
-				printk(BIOS_DEBUG, "%s: overriding faulty phy value\n", __func__);
+				printk(BIOS_DEBUG, "%s: overriding faulty phy value (seed: %04x phy: %04x step: %04x)\n", __func__,
+					total_delay_seed, total_delay_phy, abs(total_delay_phy - total_delay_seed));
 				pDCTData->WLGrossDelay[index+ByteLane] = pDCTData->WLSeedGrossDelay[index+ByteLane];
 				pDCTData->WLFineDelay[index+ByteLane] = pDCTData->WLSeedFineDelay[index+ByteLane];
 			}