driver/ddr/fsl: Add DDR4 support to Freescale DDR driver

Mostly reusing DDR3 driver, this patch adds DDR4 SPD handling, register
calculation and programming.
Signed-off-by: York Sun <yorksun@freescale.com>

README

@@ -458,6 +458,9 @@ The following options need to be configured:
 		CONFIG_SYS_FSL_DDRC_GEN3
 		Freescale DDR3 controller.
 
+		CONFIG_SYS_FSL_DDRC_GEN4
+		Freescale DDR4 controller.
+
 		CONFIG_SYS_FSL_DDRC_ARM_GEN3
 		Freescale DDR3 controller for ARM-based SoCs.
 
@@ -473,7 +476,15 @@ The following options need to be configured:
 
 		CONFIG_SYS_FSL_DDR3
 		Board config to use DDR3. It can be enabled for SoCs with
-		Freescale DDR3 controllers.
+		Freescale DDR3 or DDR3L controllers.
+
+		CONFIG_SYS_FSL_DDR3L
+		Board config to use DDR3L. It can be enabled for SoCs with
+		DDR3L controllers.
+
+		CONFIG_SYS_FSL_DDR4
+		Board config to use DDR4. It can be enabled for SoCs with
+		DDR4 controllers.
 
 		CONFIG_SYS_FSL_IFC_BE
 		Defines the IFC controller register space as Big Endian

arch/powerpc/include/asm/config_mpc85xx.h

@@ -19,8 +19,8 @@
  */
 #define CONFIG_PPC_SPINTABLE_COMPATIBLE
 
-#define FSL_DDR_VER_4_7	47
-#define FSL_DDR_VER_5_0	50
+#include <fsl_ddrc_version.h>
+#define CONFIG_SYS_FSL_DDR_BE
 
 /* IP endianness */
 #define CONFIG_SYS_FSL_IFC_BE
@@ -401,6 +401,7 @@
 #define CONFIG_SYS_NUM_FM1_DTSEC	5
 #define CONFIG_SYS_NUM_FM1_10GEC	1
 #define CONFIG_NUM_DDR_CONTROLLERS	1
+#define CONFIG_SYS_FSL_DDR_VER		FSL_DDR_VER_4_5
 #define CONFIG_SYS_FM_MURAM_SIZE	0x28000
 #define CONFIG_SYS_FSL_TBCLK_DIV	32
 #define CONFIG_SYS_FSL_PCIE_COMPAT	"fsl,qoriq-pcie-v2.2"
@@ -442,6 +443,7 @@
 #define CONFIG_SYS_NUM_FM1_10GEC	1
 #define CONFIG_SYS_NUM_FM2_10GEC	1
 #define CONFIG_NUM_DDR_CONTROLLERS	2
+#define CONFIG_SYS_FSL_DDR_VER		FSL_DDR_VER_4_4
 #define CONFIG_USB_MAX_CONTROLLER_COUNT	2
 #define CONFIG_SYS_FM_MURAM_SIZE	0x28000
 #define CONFIG_SYS_FSL_TBCLK_DIV	16
@@ -490,6 +492,7 @@
 #define CONFIG_SYS_NUM_FM1_DTSEC	5
 #define CONFIG_SYS_NUM_FM1_10GEC	1
 #define CONFIG_NUM_DDR_CONTROLLERS	2
+#define CONFIG_SYS_FSL_DDR_VER		FSL_DDR_VER_4_4
 #define CONFIG_USB_MAX_CONTROLLER_COUNT	2
 #define CONFIG_SYS_FM_MURAM_SIZE	0x28000
 #define CONFIG_SYS_FSL_TBCLK_DIV	32
@@ -527,6 +530,7 @@
 #define CONFIG_SYS_NUM_FM2_DTSEC	5
 #define CONFIG_SYS_NUM_FM2_10GEC	1
 #define CONFIG_NUM_DDR_CONTROLLERS	2
+#define CONFIG_SYS_FSL_DDR_VER		FSL_DDR_VER_4_4
 #define CONFIG_USB_MAX_CONTROLLER_COUNT	2
 #define CONFIG_SYS_FM_MURAM_SIZE	0x28000
 #define CONFIG_SYS_FSL_TBCLK_DIV	16
@@ -553,6 +557,7 @@
 #define CONFIG_TSECV2
 #define CONFIG_SYS_FSL_SEC_COMPAT	4
 #define CONFIG_NUM_DDR_CONTROLLERS	1
+#define CONFIG_SYS_FSL_DDR_VER		FSL_DDR_VER_4_4
 #define CONFIG_USB_MAX_CONTROLLER_COUNT	1
 #define CONFIG_SYS_FSL_DSP_M2_RAM_ADDR	0xb0000000
 #define CONFIG_SYS_FSL_DSP_CCSRBAR_DEFAULT	0xff600000
@@ -571,6 +576,7 @@
 #define CONFIG_TSECV2
 #define CONFIG_SYS_FSL_SEC_COMPAT	4
 #define CONFIG_NUM_DDR_CONTROLLERS	2
+#define CONFIG_SYS_FSL_DDR_VER		FSL_DDR_VER_4_6
 #define CONFIG_USB_MAX_CONTROLLER_COUNT	1
 #define CONFIG_SYS_FSL_DSP_DDR_ADDR	0x40000000
 #define CONFIG_SYS_FSL_DSP_M2_RAM_ADDR	0xb0000000
@@ -704,6 +710,9 @@ defined(CONFIG_PPC_T1020) || defined(CONFIG_PPC_T1022)
 #define CONFIG_SYS_FSL_QORIQ_CHASSIS2	/* Freescale Chassis generation 2 */
 #define CONFIG_SYS_FSL_CORES_PER_CLUSTER 1
 #define CONFIG_SYS_FSL_QMAN_V3		/* QMAN version 3 */
+#ifdef CONFIG_SYS_FSL_DDR4
+#define CONFIG_SYS_FSL_DDRC_GEN4
+#endif
 #if defined(CONFIG_PPC_T1040) || defined(CONFIG_PPC_T1042)
 #define CONFIG_MAX_CPUS			4
 #elif defined(CONFIG_PPC_T1020) || defined(CONFIG_PPC_T1022)
@@ -796,6 +805,7 @@ defined(CONFIG_PPC_T1020) || defined(CONFIG_PPC_T1022)
 #define CONFIG_SYS_FSL_SEC_COMPAT	6
 #define CONFIG_SYS_FSL_ERRATUM_ESDHC111
 #define CONFIG_NUM_DDR_CONTROLLERS	1
+#define CONFIG_SYS_FSL_DDR_VER		FSL_DDR_VER_4_6
 #define CONFIG_SYS_FSL_IFC_BANK_COUNT	8
 #define CONFIG_SYS_CCSRBAR_DEFAULT	0xff700000
 #define CONFIG_SYS_FSL_ERRATUM_A005125
@@ -820,7 +830,8 @@ defined(CONFIG_PPC_T1020) || defined(CONFIG_PPC_T1022)
 
 #if !defined(CONFIG_SYS_FSL_DDRC_GEN1) && \
 	!defined(CONFIG_SYS_FSL_DDRC_GEN2) && \
-	!defined(CONFIG_SYS_FSL_DDRC_GEN3)
+	!defined(CONFIG_SYS_FSL_DDRC_GEN3) && \
+	!defined(CONFIG_SYS_FSL_DDRC_GEN4)
 #define CONFIG_SYS_FSL_DDRC_GEN3
 #endif
 

common/ddr_spd.c

 /*
- * Copyright 2008 Freescale Semiconductor, Inc.
+ * Copyright 2008-2014 Freescale Semiconductor, Inc.
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
@@ -116,3 +116,46 @@ ddr3_spd_check(const ddr3_spd_eeprom_t *spd)
 		return 1;
 	}
 }
+
+unsigned int ddr4_spd_check(const struct ddr4_spd_eeprom_s *spd)
+{
+	char *p = (char *)spd;
+	int csum16;
+	int len;
+	char crc_lsb;	/* byte 126 */
+	char crc_msb;	/* byte 127 */
+
+	len = 126;
+	csum16 = crc16(p, len);
+
+	crc_lsb = (char) (csum16 & 0xff);
+	crc_msb = (char) (csum16 >> 8);
+
+	if (spd->crc[0] != crc_lsb || spd->crc[1] != crc_msb) {
+		printf("SPD checksum unexpected.\n"
+			"Checksum lsb in SPD = %02X, computed SPD = %02X\n"
+			"Checksum msb in SPD = %02X, computed SPD = %02X\n",
+			spd->crc[0], crc_lsb, spd->crc[1], crc_msb);
+		return 1;
+	}
+
+	p = (char *)((ulong)spd + 128);
+	len = 126;
+	csum16 = crc16(p, len);
+
+	crc_lsb = (char) (csum16 & 0xff);
+	crc_msb = (char) (csum16 >> 8);
+
+	if (spd->mod_section.uc[126] != crc_lsb ||
+	    spd->mod_section.uc[127] != crc_msb) {
+		printf("SPD checksum unexpected.\n"
+			"Checksum lsb in SPD = %02X, computed SPD = %02X\n"
+			"Checksum msb in SPD = %02X, computed SPD = %02X\n",
+			spd->mod_section.uc[126],
+			crc_lsb, spd->mod_section.uc[127],
+			crc_msb);
+		return 1;
+	}
+
+	return 0;
+}

drivers/ddr/fsl/Makefile

 #
-# Copyright 2008-2011 Freescale Semiconductor, Inc.
+# Copyright 2008-2014 Freescale Semiconductor, Inc.
 #
 # This program is free software; you can redistribute it and/or
 # modify it under the terms of the GNU General Public License
 # Version 2 as published by the Free Software Foundation.
 #
 
-obj-$(CONFIG_SYS_FSL_DDR1)	+= main.o util.o ctrl_regs.o options.o \
-				   lc_common_dimm_params.o
+obj-$(CONFIG_SYS_FSL_DDR1) += main.o util.o ctrl_regs.o options.o \
+				lc_common_dimm_params.o
+obj-$(CONFIG_SYS_FSL_DDR2) += main.o util.o ctrl_regs.o options.o \
+				lc_common_dimm_params.o
+obj-$(CONFIG_SYS_FSL_DDR3) += main.o util.o ctrl_regs.o options.o \
+				lc_common_dimm_params.o
+obj-$(CONFIG_SYS_FSL_DDR4) += main.o util.o ctrl_regs.o options.o \
+				lc_common_dimm_params.o
 
-obj-$(CONFIG_SYS_FSL_DDR2)	+= main.o util.o ctrl_regs.o options.o \
-				   lc_common_dimm_params.o
-
-obj-$(CONFIG_SYS_FSL_DDR3)	+= main.o util.o ctrl_regs.o options.o \
-				   lc_common_dimm_params.o
 ifdef CONFIG_DDR_SPD
 SPD := y
 endif
@@ -24,6 +25,7 @@ ifdef SPD
 obj-$(CONFIG_SYS_FSL_DDR1)	+= ddr1_dimm_params.o
 obj-$(CONFIG_SYS_FSL_DDR2)	+= ddr2_dimm_params.o
 obj-$(CONFIG_SYS_FSL_DDR3)	+= ddr3_dimm_params.o
+obj-$(CONFIG_SYS_FSL_DDR4)	+= ddr4_dimm_params.o
 endif
 
 obj-$(CONFIG_FSL_DDR_INTERACTIVE)	+= interactive.o
@@ -32,3 +34,4 @@ obj-$(CONFIG_SYS_FSL_DDRC_GEN2)	+= mpc85xx_ddr_gen2.o
 obj-$(CONFIG_SYS_FSL_DDRC_GEN3)	+= mpc85xx_ddr_gen3.o
 obj-$(CONFIG_SYS_FSL_DDR_86XX)		+= mpc86xx_ddr.o
 obj-$(CONFIG_SYS_FSL_DDRC_ARM_GEN3)	+= arm_ddr_gen3.o
+obj-$(CONFIG_SYS_FSL_DDRC_GEN4) += fsl_ddr_gen4.o

drivers/ddr/fsl/ddr4_dimm_params.c

+/*
+ * Copyright 2014 Freescale Semiconductor, Inc.
+ *
+ * calculate the organization and timing parameter
+ * from ddr3 spd, please refer to the spec
+ * JEDEC standard No.21-C 4_01_02_12R23A.pdf
+ *
+ *
+ */
+
+#include <common.h>
+#include <fsl_ddr_sdram.h>
+
+#include <fsl_ddr.h>
+
+/*
+ * Calculate the Density of each Physical Rank.
+ * Returned size is in bytes.
+ *
+ * Total DIMM size =
+ * sdram capacity(bit) / 8 * primary bus width / sdram width
+ *                     * Logical Ranks per DIMM
+ *
+ * where: sdram capacity  = spd byte4[3:0]
+ *        primary bus width = spd byte13[2:0]
+ *        sdram width = spd byte12[2:0]
+ *        Logical Ranks per DIMM = spd byte12[5:3] for SDP, DDP, QDP
+ *                                 spd byte12{5:3] * spd byte6[6:4] for 3DS
+ *
+ * To simplify each rank size = total DIMM size / Number of Package Ranks
+ * where Number of Package Ranks = spd byte12[5:3]
+ *
+ * SPD byte4 - sdram density and banks
+ *	bit[3:0]	size(bit)	size(byte)
+ *	0000		256Mb		32MB
+ *	0001		512Mb		64MB
+ *	0010		1Gb		128MB
+ *	0011		2Gb		256MB
+ *	0100		4Gb		512MB
+ *	0101		8Gb		1GB
+ *	0110		16Gb		2GB
+ *      0111		32Gb		4GB
+ *
+ * SPD byte13 - module memory bus width
+ *	bit[2:0]	primary bus width
+ *	000		8bits
+ *	001		16bits
+ *	010		32bits
+ *	011		64bits
+ *
+ * SPD byte12 - module organization
+ *	bit[2:0]	sdram device width
+ *	000		4bits
+ *	001		8bits
+ *	010		16bits
+ *	011		32bits
+ *
+ * SPD byte12 - module organization
+ *	bit[5:3]	number of package ranks per DIMM
+ *	000		1
+ *	001		2
+ *	010		3
+ *	011		4
+ *
+ * SPD byte6 - SDRAM package type
+ *	bit[6:4]	Die count
+ *	000		1
+ *	001		2
+ *	010		3
+ *	011		4
+ *	100		5
+ *	101		6
+ *	110		7
+ *	111		8
+ *
+ * SPD byte6 - SRAM package type
+ *	bit[1:0]	Signal loading
+ *	00		Not specified
+ *	01		Multi load stack
+ *	10		Sigle load stack (3DS)
+ *	11		Reserved
+ */
+static unsigned long long
+compute_ranksize(const struct ddr4_spd_eeprom_s *spd)
+{
+	unsigned long long bsize;
+
+	int nbit_sdram_cap_bsize = 0;
+	int nbit_primary_bus_width = 0;
+	int nbit_sdram_width = 0;
+	int die_count = 0;
+	bool package_3ds;
+
+	if ((spd->density_banks & 0xf) <= 7)
+		nbit_sdram_cap_bsize = (spd->density_banks & 0xf) + 28;
+	if ((spd->bus_width & 0x7) < 4)
+		nbit_primary_bus_width = (spd->bus_width & 0x7) + 3;
+	if ((spd->organization & 0x7) < 4)
+		nbit_sdram_width = (spd->organization & 0x7) + 2;
+	package_3ds = (spd->package_type & 0x3) == 0x2;
+	if (package_3ds)
+		die_count = (spd->package_type >> 4) & 0x7;
+
+	bsize = 1ULL << (nbit_sdram_cap_bsize - 3 +
+			 nbit_primary_bus_width - nbit_sdram_width +
+			 die_count);
+
+	debug("DDR: DDR III rank density = 0x%16llx\n", bsize);
+
+	return bsize;
+}
+
+#define spd_to_ps(mtb, ftb)	\
+	(mtb * pdimm->mtb_ps + (ftb * pdimm->ftb_10th_ps) / 10)
+/*
+ * ddr_compute_dimm_parameters for DDR3 SPD
+ *
+ * Compute DIMM parameters based upon the SPD information in spd.
+ * Writes the results to the dimm_params_t structure pointed by pdimm.
+ *
+ */
+unsigned int
+ddr_compute_dimm_parameters(const generic_spd_eeprom_t *spd,
+			    dimm_params_t *pdimm,
+			    unsigned int dimm_number)
+{
+	unsigned int retval;
+	int i;
+
+	if (spd->mem_type) {
+		if (spd->mem_type != SPD_MEMTYPE_DDR4) {
+			printf("DIMM %u: is not a DDR4 SPD.\n", dimm_number);
+			return 1;
+		}
+	} else {
+		memset(pdimm, 0, sizeof(dimm_params_t));
+		return 1;
+	}
+
+	retval = ddr4_spd_check(spd);
+	if (retval) {
+		printf("DIMM %u: failed checksum\n", dimm_number);
+		return 2;
+	}
+
+	/*
+	 * The part name in ASCII in the SPD EEPROM is not null terminated.
+	 * Guarantee null termination here by presetting all bytes to 0
+	 * and copying the part name in ASCII from the SPD onto it
+	 */
+	memset(pdimm->mpart, 0, sizeof(pdimm->mpart));
+	if ((spd->info_size_crc & 0xF) > 2)
+		memcpy(pdimm->mpart, spd->mpart, sizeof(pdimm->mpart) - 1);
+
+	/* DIMM organization parameters */
+	pdimm->n_ranks = ((spd->organization >> 3) & 0x7) + 1;
+	pdimm->rank_density = compute_ranksize(spd);
+	pdimm->capacity = pdimm->n_ranks * pdimm->rank_density;
+	pdimm->primary_sdram_width = 1 << (3 + (spd->bus_width & 0x7));
+	if ((spd->bus_width >> 3) & 0x3)
+		pdimm->ec_sdram_width = 8;
+	else
+		pdimm->ec_sdram_width = 0;
+	pdimm->data_width = pdimm->primary_sdram_width
+			  + pdimm->ec_sdram_width;
+	pdimm->device_width = 1 << ((spd->organization & 0x7) + 2);
+
+	/* These are the types defined by the JEDEC DDR3 SPD spec */
+	pdimm->mirrored_dimm = 0;
+	pdimm->registered_dimm = 0;
+	switch (spd->module_type & DDR3_SPD_MODULETYPE_MASK) {
+	case DDR3_SPD_MODULETYPE_RDIMM:
+		/* Registered/buffered DIMMs */
+		pdimm->registered_dimm = 1;
+		break;
+
+	case DDR3_SPD_MODULETYPE_UDIMM:
+	case DDR3_SPD_MODULETYPE_SO_DIMM:
+		/* Unbuffered DIMMs */
+		if (spd->mod_section.unbuffered.addr_mapping & 0x1)
+			pdimm->mirrored_dimm = 1;
+		break;
+
+	default:
+		printf("unknown module_type 0x%02X\n", spd->module_type);
+		return 1;
+	}
+
+	/* SDRAM device parameters */
+	pdimm->n_row_addr = ((spd->addressing >> 3) & 0x7) + 12;
+	pdimm->n_col_addr = (spd->addressing & 0x7) + 9;
+	pdimm->bank_addr_bits = (spd->density_banks >> 4) & 0x3;
+	pdimm->bank_group_bits = (spd->density_banks >> 6) & 0x3;
+
+	/*
+	 * The SPD spec has not the ECC bit,
+	 * We consider the DIMM as ECC capability
+	 * when the extension bus exist
+	 */
+	if (pdimm->ec_sdram_width)
+		pdimm->edc_config = 0x02;
+	else
+		pdimm->edc_config = 0x00;
+
+	/*
+	 * The SPD spec has not the burst length byte
+	 * but DDR4 spec has nature BL8 and BC4,
+	 * BL8 -bit3, BC4 -bit2
+	 */
+	pdimm->burst_lengths_bitmask = 0x0c;
+	pdimm->row_density = __ilog2(pdimm->rank_density);
+
+	/* MTB - medium timebase
+	 * The MTB in the SPD spec is 125ps,
+	 *
+	 * FTB - fine timebase
+	 * use 1/10th of ps as our unit to avoid floating point
+	 * eg, 10 for 1ps, 25 for 2.5ps, 50 for 5ps
+	 */
+	if ((spd->timebases & 0xf) == 0x0) {
+		pdimm->mtb_ps = 125;
+		pdimm->ftb_10th_ps = 10;
+
+	} else {
+		printf("Unknown Timebases\n");
+	}
+
+	/* sdram minimum cycle time */
+	pdimm->tckmin_x_ps = spd_to_ps(spd->tck_min, spd->fine_tck_min);
+
+	/* sdram max cycle time */
+	pdimm->tckmax_ps = spd_to_ps(spd->tck_max, spd->fine_tck_max);
+
+	/*
+	 * CAS latency supported
+	 * bit0 - CL7
+	 * bit4 - CL11
+	 * bit8 - CL15
+	 * bit12- CL19
+	 * bit16- CL23
+	 */
+	pdimm->caslat_x  = (spd->caslat_b1 << 7)	|
+			   (spd->caslat_b2 << 15)	|
+			   (spd->caslat_b3 << 23);
+
+	BUG_ON(spd->caslat_b4 != 0);
+
+	/*
+	 * min CAS latency time
+	 */
+	pdimm->taa_ps = spd_to_ps(spd->taa_min, spd->fine_taa_min);
+
+	/*
+	 * min RAS to CAS delay time
+	 */
+	pdimm->trcd_ps = spd_to_ps(spd->trcd_min, spd->fine_trcd_min);
+
+	/*
+	 * Min Row Precharge Delay Time
+	 */
+	pdimm->trp_ps = spd_to_ps(spd->trp_min, spd->fine_trp_min);
+
+	/* min active to precharge delay time */
+	pdimm->tras_ps = (((spd->tras_trc_ext & 0xf) << 8) +
+			  spd->tras_min_lsb) * pdimm->mtb_ps;
+
+	/* min active to actice/refresh delay time */
+	pdimm->trc_ps = spd_to_ps((((spd->tras_trc_ext & 0xf0) << 4) +
+				   spd->trc_min_lsb), spd->fine_trc_min);
+	/* Min Refresh Recovery Delay Time */
+	pdimm->trfc1_ps = ((spd->trfc1_min_msb << 8) | (spd->trfc1_min_lsb)) *
+		       pdimm->mtb_ps;
+	pdimm->trfc2_ps = ((spd->trfc2_min_msb << 8) | (spd->trfc2_min_lsb)) *
+		       pdimm->mtb_ps;
+	pdimm->trfc4_ps = ((spd->trfc4_min_msb << 8) | (spd->trfc4_min_lsb)) *
+			pdimm->mtb_ps;
+	/* min four active window delay time */
+	pdimm->tfaw_ps = (((spd->tfaw_msb & 0xf) << 8) | spd->tfaw_min) *
+			pdimm->mtb_ps;
+
+	/* min row active to row active delay time, different bank group */
+	pdimm->trrds_ps = spd_to_ps(spd->trrds_min, spd->fine_trrds_min);
+	/* min row active to row active delay time, same bank group */
+	pdimm->trrdl_ps = spd_to_ps(spd->trrdl_min, spd->fine_trrdl_min);
+	/* min CAS to CAS Delay Time (tCCD_Lmin), same bank group */
+	pdimm->tccdl_ps = spd_to_ps(spd->tccdl_min, spd->fine_tccdl_min);
+
+	/*
+	 * Average periodic refresh interval
+	 * tREFI = 7.8 us at normal temperature range
+	 */
+	pdimm->refresh_rate_ps = 7800000;
+
+	for (i = 0; i < 18; i++)
+		pdimm->dq_mapping[i] = spd->mapping[i];
+
+	pdimm->dq_mapping_ors = ((spd->mapping[0] >> 6) & 0x3) == 0 ? 1 : 0;
+
+	return 0;
+}

drivers/ddr/fsl/fsl_ddr_gen4.c

+/*
+ * Copyright 2014 Freescale Semiconductor, Inc.
+ *
+ * SPDX-License-Identifier:	GPL-2.0+
+ */
+
+#include <common.h>
+#include <asm/io.h>
+#include <fsl_ddr_sdram.h>
+#include <asm/processor.h>
+#include <fsl_ddr.h>
+
+#if (CONFIG_CHIP_SELECTS_PER_CTRL > 4)
+#error Invalid setting for CONFIG_CHIP_SELECTS_PER_CTRL
+#endif
+
+/*
+ * regs has the to-be-set values for DDR controller registers
+ * ctrl_num is the DDR controller number
+ * step: 0 goes through the initialization in one pass
+ *       1 sets registers and returns before enabling controller
+ *       2 resumes from step 1 and continues to initialize
+ * Dividing the initialization to two steps to deassert DDR reset signal
+ * to comply with JEDEC specs for RDIMMs.
+ */
+void fsl_ddr_set_memctl_regs(const fsl_ddr_cfg_regs_t *regs,
+			     unsigned int ctrl_num, int step)
+{
+	unsigned int i, bus_width;
+	struct ccsr_ddr __iomem *ddr;
+	u32 temp_sdram_cfg;
+	u32 total_gb_size_per_controller;
+	int timeout;
+
+	switch (ctrl_num) {
+	case 0:
+		ddr = (void *)CONFIG_SYS_FSL_DDR_ADDR;
+		break;
+#if defined(CONFIG_SYS_FSL_DDR2_ADDR) && (CONFIG_NUM_DDR_CONTROLLERS > 1)
+	case 1:
+		ddr = (void *)CONFIG_SYS_FSL_DDR2_ADDR;
+		break;
+#endif
+#if defined(CONFIG_SYS_FSL_DDR3_ADDR) && (CONFIG_NUM_DDR_CONTROLLERS > 2)
+	case 2:
+		ddr = (void *)CONFIG_SYS_FSL_DDR3_ADDR;
+		break;
+#endif
+#if defined(CONFIG_SYS_FSL_DDR4_ADDR) && (CONFIG_NUM_DDR_CONTROLLERS > 3)
+	case 3:
+		ddr = (void *)CONFIG_SYS_FSL_DDR4_ADDR;
+		break;
+#endif
+	default:
+		printf("%s unexpected ctrl_num = %u\n", __func__, ctrl_num);
+		return;
+	}
+
+	if (step == 2)
+		goto step2;
+
+	if (regs->ddr_eor)
+		ddr_out32(&ddr->eor, regs->ddr_eor);
+
+	ddr_out32(&ddr->sdram_clk_cntl, regs->ddr_sdram_clk_cntl);
+
+	for (i = 0; i < CONFIG_CHIP_SELECTS_PER_CTRL; i++) {
+		if (i == 0) {
+			ddr_out32(&ddr->cs0_bnds, regs->cs[i].bnds);
+			ddr_out32(&ddr->cs0_config, regs->cs[i].config);
+			ddr_out32(&ddr->cs0_config_2, regs->cs[i].config_2);
+
+		} else if (i == 1) {
+			ddr_out32(&ddr->cs1_bnds, regs->cs[i].bnds);
+			ddr_out32(&ddr->cs1_config, regs->cs[i].config);
+			ddr_out32(&ddr->cs1_config_2, regs->cs[i].config_2);
+
+		} else if (i == 2) {
+			ddr_out32(&ddr->cs2_bnds, regs->cs[i].bnds);
+			ddr_out32(&ddr->cs2_config, regs->cs[i].config);
+			ddr_out32(&ddr->cs2_config_2, regs->cs[i].config_2);
+
+		} else if (i == 3) {
+			ddr_out32(&ddr->cs3_bnds, regs->cs[i].bnds);
+			ddr_out32(&ddr->cs3_config, regs->cs[i].config);
+			ddr_out32(&ddr->cs3_config_2, regs->cs[i].config_2);
+		}
+	}
+
+	ddr_out32(&ddr->timing_cfg_3, regs->timing_cfg_3);
+	ddr_out32(&ddr->timing_cfg_0, regs->timing_cfg_0);
+	ddr_out32(&ddr->timing_cfg_1, regs->timing_cfg_1);
+	ddr_out32(&ddr->timing_cfg_2, regs->timing_cfg_2);
+	ddr_out32(&ddr->timing_cfg_4, regs->timing_cfg_4);
+	ddr_out32(&ddr->timing_cfg_5, regs->timing_cfg_5);
+	ddr_out32(&ddr->timing_cfg_6, regs->timing_cfg_6);
+	ddr_out32(&ddr->timing_cfg_7, regs->timing_cfg_7);
+	ddr_out32(&ddr->timing_cfg_8, regs->timing_cfg_8);
+	ddr_out32(&ddr->timing_cfg_9, regs->timing_cfg_9);
+	ddr_out32(&ddr->ddr_zq_cntl, regs->ddr_zq_cntl);
+	ddr_out32(&ddr->dq_map_0, regs->dq_map_0);
+	ddr_out32(&ddr->dq_map_1, regs->dq_map_1);
+	ddr_out32(&ddr->dq_map_2, regs->dq_map_2);
+	ddr_out32(&ddr->dq_map_3, regs->dq_map_3);
+	ddr_out32(&ddr->sdram_cfg_2, regs->ddr_sdram_cfg_2);
+	ddr_out32(&ddr->sdram_cfg_3, regs->ddr_sdram_cfg_3);
+	ddr_out32(&ddr->sdram_mode, regs->ddr_sdram_mode);
+	ddr_out32(&ddr->sdram_mode_2, regs->ddr_sdram_mode_2);
+	ddr_out32(&ddr->sdram_mode_3, regs->ddr_sdram_mode_3);
+	ddr_out32(&ddr->sdram_mode_4, regs->ddr_sdram_mode_4);
+	ddr_out32(&ddr->sdram_mode_5, regs->ddr_sdram_mode_5);
+	ddr_out32(&ddr->sdram_mode_6, regs->ddr_sdram_mode_6);
+	ddr_out32(&ddr->sdram_mode_7, regs->ddr_sdram_mode_7);
+	ddr_out32(&ddr->sdram_mode_8, regs->ddr_sdram_mode_8);
+	ddr_out32(&ddr->sdram_mode_9, regs->ddr_sdram_mode_9);
+	ddr_out32(&ddr->sdram_mode_10, regs->ddr_sdram_mode_10);
+	ddr_out32(&ddr->sdram_mode_11, regs->ddr_sdram_mode_11);
+	ddr_out32(&ddr->sdram_mode_12, regs->ddr_sdram_mode_12);
+	ddr_out32(&ddr->sdram_mode_13, regs->ddr_sdram_mode_13);
+	ddr_out32(&ddr->sdram_mode_14, regs->ddr_sdram_mode_14);
+	ddr_out32(&ddr->sdram_mode_15, regs->ddr_sdram_mode_15);
+	ddr_out32(&ddr->sdram_mode_16, regs->ddr_sdram_mode_16);
+	ddr_out32(&ddr->sdram_md_cntl, regs->ddr_sdram_md_cntl);
+	ddr_out32(&ddr->sdram_interval, regs->ddr_sdram_interval);
+	ddr_out32(&ddr->sdram_data_init, regs->ddr_data_init);
+	ddr_out32(&ddr->init_addr, regs->ddr_init_addr);
+	ddr_out32(&ddr->init_ext_addr, regs->ddr_init_ext_addr);
+	ddr_out32(&ddr->ddr_wrlvl_cntl, regs->ddr_wrlvl_cntl);
+#ifndef CONFIG_SYS_FSL_DDR_EMU
+	/*
+	 * Skip these two registers if running on emulator
+	 * because emulator doesn't have skew between bytes.
+	 */
+
+	if (regs->ddr_wrlvl_cntl_2)
+		ddr_out32(&ddr->ddr_wrlvl_cntl_2, regs->ddr_wrlvl_cntl_2);
+	if (regs->ddr_wrlvl_cntl_3)
+		ddr_out32(&ddr->ddr_wrlvl_cntl_3, regs->ddr_wrlvl_cntl_3);
+#endif
+
+	ddr_out32(&ddr->ddr_sr_cntr, regs->ddr_sr_cntr);
+	ddr_out32(&ddr->ddr_sdram_rcw_1, regs->ddr_sdram_rcw_1);
+	ddr_out32(&ddr->ddr_sdram_rcw_2, regs->ddr_sdram_rcw_2);
+	ddr_out32(&ddr->ddr_sdram_rcw_3, regs->ddr_sdram_rcw_3);
+	ddr_out32(&ddr->ddr_sdram_rcw_4, regs->ddr_sdram_rcw_4);
+	ddr_out32(&ddr->ddr_sdram_rcw_5, regs->ddr_sdram_rcw_5);
+	ddr_out32(&ddr->ddr_sdram_rcw_6, regs->ddr_sdram_rcw_6);
+	ddr_out32(&ddr->ddr_cdr1, regs->ddr_cdr1);
+	ddr_out32(&ddr->ddr_cdr2, regs->ddr_cdr2);
+	ddr_out32(&ddr->err_disable, regs->err_disable);
+	ddr_out32(&ddr->err_int_en, regs->err_int_en);
+	for (i = 0; i < 32; i++) {
+		if (regs->debug[i]) {
+			debug("Write to debug_%d as %08x\n",
+			      i+1, regs->debug[i]);
+			ddr_out32(&ddr->debug[i], regs->debug[i]);
+		}
+	}
+
+	/*
+	 * For RDIMMs, JEDEC spec requires clocks to be stable before reset is
+	 * deasserted. Clocks start when any chip select is enabled and clock
+	 * control register is set. Because all DDR components are connected to
+	 * one reset signal, this needs to be done in two steps. Step 1 is to
+	 * get the clocks started. Step 2 resumes after reset signal is
+	 * deasserted.
+	 */
+	if (step == 1) {
+		udelay(200);
+		return;
+	}
+
+step2:
+	/* Set, but do not enable the memory */
+	temp_sdram_cfg = regs->ddr_sdram_cfg;
+	temp_sdram_cfg &= ~(SDRAM_CFG_MEM_EN);
+	ddr_out32(&ddr->sdram_cfg, temp_sdram_cfg);
+
+	/*
+	 * 500 painful micro-seconds must elapse between
+	 * the DDR clock setup and the DDR config enable.
+	 * DDR2 need 200 us, and DDR3 need 500 us from spec,
+	 * we choose the max, that is 500 us for all of case.
+	 */
+	udelay(500);
+	asm volatile("sync;isync");
+
+	/* Let the controller go */
+	temp_sdram_cfg = ddr_in32(&ddr->sdram_cfg) & ~SDRAM_CFG_BI;
+	ddr_out32(&ddr->sdram_cfg, temp_sdram_cfg | SDRAM_CFG_MEM_EN);
+	asm volatile("sync;isync");
+
+	total_gb_size_per_controller = 0;
+	for (i = 0; i < CONFIG_CHIP_SELECTS_PER_CTRL; i++) {
+		if (!(regs->cs[i].config & 0x80000000))
+			continue;
+		total_gb_size_per_controller += 1 << (
+			((regs->cs[i].config >> 14) & 0x3) + 2 +
+			((regs->cs[i].config >> 8) & 0x7) + 12 +
+			((regs->cs[i].config >> 4) & 0x3) + 0 +
+			((regs->cs[i].config >> 0) & 0x7) + 8 +
+			3 - ((regs->ddr_sdram_cfg >> 19) & 0x3) -
+			26);			/* minus 26 (count of 64M) */
+	}
+	if (fsl_ddr_get_intl3r() & 0x80000000)	/* 3-way interleaving */
+		total_gb_size_per_controller *= 3;
+	else if (regs->cs[0].config & 0x20000000) /* 2-way interleaving */