From 1ea9b05694da7ee61d49d9cd2b7e533a98e42321 Mon Sep 17 00:00:00 2001
From: Angel Pons <th3fanbus@gmail.com>
Date: Sat, 7 May 2022 21:49:40 +0200
Subject: [PATCH 06/20] haswell NRI: Add DDR3 JEDEC reset and init

Implement JEDEC reset and init sequence for DDR3. The MRS commands are
issued through the REUT (Robust Electrical Unified Testing) hardware.

Change-Id: I2a0c066537021b587599228086727cb1e041bff5
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
---
 .../intel/haswell/native_raminit/Makefile.mk  |   3 +
 .../intel/haswell/native_raminit/ddr3.c       | 217 ++++++++++++++++++
 .../haswell/native_raminit/io_comp_control.c  |  19 ++
 .../haswell/native_raminit/jedec_reset.c      | 120 ++++++++++
 .../haswell/native_raminit/raminit_main.c     |   2 +
 .../haswell/native_raminit/raminit_native.h   |  99 ++++++++
 .../haswell/native_raminit/reg_structs.h      | 154 +++++++++++++
 .../intel/haswell/native_raminit/reut.c       | 196 ++++++++++++++++
 .../intel/haswell/registers/mchbar.h          |  21 ++
 src/southbridge/intel/lynxpoint/pch.h         |   2 +
 10 files changed, 833 insertions(+)
 create mode 100644 src/northbridge/intel/haswell/native_raminit/ddr3.c
 create mode 100644 src/northbridge/intel/haswell/native_raminit/jedec_reset.c
 create mode 100644 src/northbridge/intel/haswell/native_raminit/reut.c

diff --git a/src/northbridge/intel/haswell/native_raminit/Makefile.mk b/src/northbridge/intel/haswell/native_raminit/Makefile.mk
index 37d527e972..e9212df9e6 100644
--- a/src/northbridge/intel/haswell/native_raminit/Makefile.mk
+++ b/src/northbridge/intel/haswell/native_raminit/Makefile.mk
@@ -1,11 +1,14 @@
 ## SPDX-License-Identifier: GPL-2.0-or-later
 
 romstage-y += configure_mc.c
+romstage-y += ddr3.c
+romstage-y += jedec_reset.c
 romstage-y += lookup_timings.c
 romstage-y += init_mpll.c
 romstage-y += io_comp_control.c
 romstage-y += memory_map.c
 romstage-y += raminit_main.c
 romstage-y += raminit_native.c
+romstage-y += reut.c
 romstage-y += spd_bitmunching.c
 romstage-y += timings_refresh.c
diff --git a/src/northbridge/intel/haswell/native_raminit/ddr3.c b/src/northbridge/intel/haswell/native_raminit/ddr3.c
new file mode 100644
index 0000000000..6ddb11488b
--- /dev/null
+++ b/src/northbridge/intel/haswell/native_raminit/ddr3.c
@@ -0,0 +1,217 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#include <assert.h>
+#include <console/console.h>
+#include <northbridge/intel/haswell/haswell.h>
+#include <types.h>
+
+#include "raminit_native.h"
+
+#define DDR3_RTTNOM(a, b, c) (((a) << 9) | ((b) << 6) | ((c) << 2))
+
+uint16_t encode_ddr3_rttnom(const uint32_t rttnom)
+{
+	switch (rttnom) {
+	case 0:		return DDR3_RTTNOM(0, 0, 0);	/* RttNom is disabled */
+	case 20:	return DDR3_RTTNOM(1, 0, 0);	/* RZQ/12 */
+	case 30:	return DDR3_RTTNOM(1, 0, 1);	/* RZQ/8 */
+	case 40:	return DDR3_RTTNOM(0, 1, 1);	/* RZQ/6 */
+	case 60:	return DDR3_RTTNOM(0, 0, 1);	/* RZQ/4 */
+	case 120:	return DDR3_RTTNOM(0, 1, 0);	/* RZQ/2 */
+	}
+	printk(BIOS_ERR, "%s: Invalid rtt_nom value %u\n", __func__, rttnom);
+	return 0;
+}
+
+static const uint8_t jedec_wr_t[12] = { 1, 2, 3, 4, 5, 5, 6, 6, 7, 7, 0, 0 };
+
+static void ddr3_program_mr0(struct sysinfo *ctrl, const uint8_t dll_reset)
+{
+	assert(ctrl->tWR >= 5 && ctrl->tWR <= 16);
+	assert(ctrl->tAA >= 4);
+	const uint8_t jedec_cas = ctrl->tAA - 4;
+	const union {
+		struct __packed {
+			uint16_t burst_length     : 2; // Bits  1:0
+			uint16_t cas_latency_msb  : 1; // Bits  2:2
+			uint16_t read_burst_type  : 1; // Bits  3:3
+			uint16_t cas_latency_low  : 3; // Bits  6:4
+			uint16_t test_mode        : 1; // Bits  7:7
+			uint16_t dll_reset        : 1; // Bits  8:8
+			uint16_t write_recovery   : 3; // Bits 11:9
+			uint16_t precharge_pd_dll : 1; // Bits 12:12
+			uint16_t                  : 3; // Bits 15:13
+		};
+		uint16_t raw;
+	} mr0reg = {
+		.burst_length     = 0,
+		.cas_latency_msb  = !!(jedec_cas & BIT(3)),
+		.read_burst_type  = 0,
+		.cas_latency_low  = jedec_cas & 0x7,
+		.dll_reset        = 1,
+		.write_recovery   = jedec_wr_t[ctrl->tWR - 5],
+		.precharge_pd_dll = 0,
+	};
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		if (!does_ch_exist(ctrl, channel))
+			continue;
+
+		for (uint8_t slot = 0; slot < NUM_SLOTS; slot++) {
+			if (!rank_in_ch(ctrl, slot + slot, channel))
+				continue;
+
+			if (!ctrl->restore_mrs)
+				ctrl->mr0[channel][slot] = mr0reg.raw;
+		}
+		reut_issue_mrs_all(ctrl, channel, 0, ctrl->mr0[channel]);
+	}
+}
+
+void ddr3_program_mr1(struct sysinfo *ctrl, const uint8_t wl_mode, const uint8_t q_off)
+{
+	/*
+	 * JESD79-3F (JEDEC DDR3 spec) refers to bit 0 of MR1 as 'DLL Enable'.
+	 * However, its encoding is weird, and 'DLL Disable' makes more sense.
+	 *
+	 * Moreover, bit 5 is part of ODIC (Output Driver Impedance Control),
+	 * but all encodings where MR1 bit 5 is 1 are reserved. Thus, omit it.
+	 */
+	union {
+		struct __packed {
+			uint16_t dll_disable      : 1; // Bits  0:0
+			uint16_t od_impedance_ctl : 1; // Bits  1:1
+			uint16_t odt_rtt_nom_low  : 1; // Bits  2:2
+			uint16_t additive_latency : 2; // Bits  4:3
+			uint16_t                  : 1; // Bits  5:5
+			uint16_t odt_rtt_nom_mid  : 1; // Bits  6:6
+			uint16_t write_level_mode : 1; // Bits  7:7
+			uint16_t                  : 1; // Bits  8:8
+			uint16_t odt_rtt_nom_high : 1; // Bits  9:9
+			uint16_t                  : 1; // Bits 10:10
+			uint16_t t_dqs            : 1; // Bits 11:11
+			uint16_t q_off            : 1; // Bits 12:12
+			uint16_t                  : 3; // Bits 15:13
+		};
+		uint16_t raw;
+	} mr1reg = {
+		.dll_disable      = 0,
+		.od_impedance_ctl = 1,	/* RZQ/7 */
+		.additive_latency = 0,
+		.write_level_mode = wl_mode,
+		.t_dqs            = 0,
+		.q_off            = q_off,
+	};
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		if (!does_ch_exist(ctrl, channel))
+			continue;
+
+		mr1reg.raw &= ~RTTNOM_MASK;
+		mr1reg.raw |= encode_ddr3_rttnom(ctrl->dpc[channel] == 2 ? 60 : 0);
+		for (uint8_t slot = 0; slot < NUM_SLOTS; slot++) {
+			if (!rank_in_ch(ctrl, slot + slot, channel))
+				continue;
+
+			if (!ctrl->restore_mrs)
+				ctrl->mr1[channel][slot] = mr1reg.raw;
+		}
+		reut_issue_mrs_all(ctrl, channel, 1, ctrl->mr1[channel]);
+	}
+}
+
+enum {
+	RTT_WR_OFF = 0,
+	RTT_WR_60  = 1,
+	RTT_WR_120 = 2,
+};
+
+static void ddr3_program_mr2(struct sysinfo *ctrl)
+{
+	assert(ctrl->tCWL >= 5);
+	const bool dimm_srt = ctrl->flags.ext_temp_refresh && !ctrl->flags.asr;
+
+	const union {
+		struct __packed {
+			uint16_t partial_array_sr  : 3; // Bits  0:2
+			uint16_t cas_write_latency : 3; // Bits  5:3
+			uint16_t auto_self_refresh : 1; // Bits  6:6
+			uint16_t self_refresh_temp : 1; // Bits  7:7
+			uint16_t                   : 1; // Bits  8:8
+			uint16_t odt_rtt_wr        : 2; // Bits 10:9
+			uint16_t                   : 5; // Bits 15:11
+		};
+		uint16_t raw;
+	} mr2reg = {
+		.partial_array_sr  = 0,
+		.cas_write_latency = ctrl->tCWL - 5,
+		.auto_self_refresh = ctrl->flags.asr,
+		.self_refresh_temp = dimm_srt,
+		.odt_rtt_wr        = is_hsw_ult() ? RTT_WR_120 : RTT_WR_60,
+	};
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		if (!does_ch_exist(ctrl, channel))
+			continue;
+
+		for (uint8_t slot = 0; slot < NUM_SLOTS; slot++) {
+			if (!rank_in_ch(ctrl, slot + slot, channel))
+				continue;
+
+			if (!ctrl->restore_mrs)
+				ctrl->mr2[channel][slot] = mr2reg.raw;
+		}
+		/* MR2 shadow register is similar but not identical to MR2 */
+		if (!ctrl->restore_mrs) {
+			union tc_mr2_shadow_reg tc_mr2_shadow = {
+				.raw = mr2reg.raw & 0x073f,
+			};
+			for (uint8_t slot = 0; slot < NUM_SLOTS; slot++) {
+				if (!rank_in_ch(ctrl, slot + slot, channel))
+					continue;
+
+				if (dimm_srt)
+					tc_mr2_shadow.srt_available |= BIT(slot);
+
+				if (ctrl->rank_mirrored[channel] & BIT(slot + slot + 1))
+					tc_mr2_shadow.addr_bit_swizzle |= BIT(slot);
+			}
+			mchbar_write32(TC_MR2_SHADOW_ch(channel), tc_mr2_shadow.raw);
+		}
+		reut_issue_mrs_all(ctrl, channel, 2, ctrl->mr2[channel]);
+	}
+}
+
+static void ddr3_program_mr3(struct sysinfo *ctrl, const uint8_t mpr_mode)
+{
+	const union {
+		struct __packed {
+			uint16_t mpr_loc  :  2; // Bits  1:0
+			uint16_t mpr_mode :  1; // Bits  2:2
+			uint16_t          : 13; // Bits 15:3
+		};
+		uint16_t raw;
+	} mr3reg = {
+		.mpr_loc  = 0,
+		.mpr_mode = mpr_mode,
+	};
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		if (!does_ch_exist(ctrl, channel))
+			continue;
+
+		for (uint8_t slot = 0; slot < NUM_SLOTS; slot++) {
+			if (!rank_in_ch(ctrl, slot + slot, channel))
+				continue;
+
+			if (!ctrl->restore_mrs)
+				ctrl->mr3[channel][slot] = mr3reg.raw;
+		}
+		reut_issue_mrs_all(ctrl, channel, 3, ctrl->mr3[channel]);
+	}
+}
+
+enum raminit_status ddr3_jedec_init(struct sysinfo *ctrl)
+{
+	ddr3_program_mr2(ctrl);
+	ddr3_program_mr3(ctrl, 0);
+	ddr3_program_mr1(ctrl, 0, 0);
+	ddr3_program_mr0(ctrl, 1);
+	return reut_issue_zq(ctrl, ctrl->chanmap, ZQ_INIT);
+}
diff --git a/src/northbridge/intel/haswell/native_raminit/io_comp_control.c b/src/northbridge/intel/haswell/native_raminit/io_comp_control.c
index d45b608dd3..8a55fd81b2 100644
--- a/src/northbridge/intel/haswell/native_raminit/io_comp_control.c
+++ b/src/northbridge/intel/haswell/native_raminit/io_comp_control.c
@@ -8,6 +8,25 @@
 
 #include "raminit_native.h"
 
+enum raminit_status io_reset(void)
+{
+	union mc_init_state_g_reg mc_init_state_g = {
+		.raw = mchbar_read32(MC_INIT_STATE_G),
+	};
+	mc_init_state_g.reset_io = 1;
+	mchbar_write32(MC_INIT_STATE_G, mc_init_state_g.raw);
+	struct stopwatch timer;
+	stopwatch_init_msecs_expire(&timer, 2000);
+	do {
+		mc_init_state_g.raw = mchbar_read32(MC_INIT_STATE_G);
+		if (mc_init_state_g.reset_io == 0)
+			return RAMINIT_STATUS_SUCCESS;
+
+	} while (!stopwatch_expired(&timer));
+	printk(BIOS_ERR, "Timed out waiting for DDR I/O reset to complete\n");
+	return RAMINIT_STATUS_POLL_TIMEOUT;
+}
+
 enum raminit_status wait_for_first_rcomp(void)
 {
 	struct stopwatch timer;
diff --git a/src/northbridge/intel/haswell/native_raminit/jedec_reset.c b/src/northbridge/intel/haswell/native_raminit/jedec_reset.c
new file mode 100644
index 0000000000..de0f676758
--- /dev/null
+++ b/src/northbridge/intel/haswell/native_raminit/jedec_reset.c
@@ -0,0 +1,120 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#include <console/console.h>
+#include <delay.h>
+#include <northbridge/intel/haswell/haswell.h>
+#include <southbridge/intel/lynxpoint/pch.h>
+#include <types.h>
+#include <timer.h>
+
+#include "raminit_native.h"
+
+static void assert_reset(const bool do_reset)
+{
+	if (is_hsw_ult()) {
+		uint32_t pm_cfg2 = RCBA32(PM_CFG2);
+		if (do_reset)
+			pm_cfg2 &= ~PM_CFG2_DRAM_RESET_CTL;
+		else
+			pm_cfg2 |= PM_CFG2_DRAM_RESET_CTL;
+		RCBA32(PM_CFG2) = pm_cfg2;
+	} else {
+		union mc_init_state_g_reg mc_init_state_g = {
+			.raw = mchbar_read32(MC_INIT_STATE_G),
+		};
+		mc_init_state_g.ddr_not_reset = !do_reset;
+		mchbar_write32(MC_INIT_STATE_G, mc_init_state_g.raw);
+	}
+}
+
+/*
+ * Perform JEDEC reset.
+ *
+ * If RTT_NOM is to be enabled in MR1, the ODT input signal must be
+ * statically held low in our system since RTT_NOM is always enabled.
+ */
+static void jedec_reset(struct sysinfo *ctrl)
+{
+	if (is_hsw_ult())
+		assert_reset(false);
+
+	union mc_init_state_g_reg mc_init_state_g = {
+		.ddr_not_reset     = 1,
+		.safe_self_refresh = 1,
+	};
+	mchbar_write32(MC_INIT_STATE_G, mc_init_state_g.raw);
+
+	union reut_misc_cke_ctrl_reg reut_misc_cke_ctrl = {
+		.cke_override = 0xf,
+		.cke_on       = 0,
+	};
+	mchbar_write32(REUT_MISC_CKE_CTRL, reut_misc_cke_ctrl.raw);
+
+	assert_reset(true);
+
+	/** TODO: check and switch DDR3 voltage here (mainboard-specific) **/
+
+	udelay(200);
+
+	assert_reset(false);
+
+	udelay(500);
+
+	mc_init_state_g.dclk_enable = 1;
+	mchbar_write32(MC_INIT_STATE_G, mc_init_state_g.raw);
+
+	/* Delay at least 20 nanoseconds for tCKSRX */
+	tick_delay(1);
+
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		reut_misc_cke_ctrl.cke_on = ctrl->rankmap[channel];
+		mchbar_write32(REUT_ch_MISC_CKE_CTRL(channel), reut_misc_cke_ctrl.raw);
+	}
+
+	/*
+	 * Wait minimum of reset CKE exit time, tXPR.
+	 * Spec says MAX(tXS, 5 tCK). 5 tCK is 10 ns.
+	 */
+	tick_delay(1);
+}
+
+enum raminit_status do_jedec_init(struct sysinfo *ctrl)
+{
+	/* Never do a JEDEC reset in S3 resume */
+	if (ctrl->bootmode == BOOTMODE_S3)
+		return RAMINIT_STATUS_SUCCESS;
+
+	enum raminit_status status = io_reset();
+	if (status)
+		return status;
+
+	status = wait_for_first_rcomp();
+	if (status)
+		return status;
+
+	/* Force ODT low (JEDEC spec) */
+	const union reut_misc_odt_ctrl_reg reut_misc_odt_ctrl = {
+		.odt_override = 0xf,
+		.odt_on       = 0,
+	};
+	mchbar_write32(REUT_MISC_ODT_CTRL, reut_misc_odt_ctrl.raw);
+
+	/*
+	 * Note: Haswell MRC does not clear ODT override for LPDDR3. However,
+	 * Broadwell MRC does. Hell suspects this difference is important, as
+	 * there is an erratum in the specification update for Broadwell:
+	 *
+	 * Erratum BDM74: LPDDR3 Memory Training May Cause Platform Boot Failure
+	 */
+	if (ctrl->lpddr)
+		die("%s: LPDDR-specific JEDEC init not implemented\n", __func__);
+
+	jedec_reset(ctrl);
+	status = ddr3_jedec_init(ctrl);
+	if (!status)
+		ctrl->restore_mrs = true;
+
+	/* Release ODT override */
+	mchbar_write32(REUT_MISC_ODT_CTRL, 0);
+	return status;
+}
diff --git a/src/northbridge/intel/haswell/native_raminit/raminit_main.c b/src/northbridge/intel/haswell/native_raminit/raminit_main.c
index 559dfc3a4e..94b268468c 100644
--- a/src/northbridge/intel/haswell/native_raminit/raminit_main.c
+++ b/src/northbridge/intel/haswell/native_raminit/raminit_main.c
@@ -24,6 +24,7 @@ static const struct task_entry cold_boot[] = {
 	{ convert_timings,                                        true, "CONVTIM",    },
 	{ configure_mc,                                           true, "CONFMC",     },
 	{ configure_memory_map,                                   true, "MEMMAP",     },
+	{ do_jedec_init,                                          true, "JEDECINIT",  },
 };
 
 /* Return a generic stepping value to make stepping checks simpler */
@@ -57,6 +58,7 @@ static void initialize_ctrl(struct sysinfo *ctrl)
 	ctrl->stepping = get_stepping(ctrl->cpu);
 	ctrl->vdd_mv = is_hsw_ult() ? 1350 : 1500; /** FIXME: Hardcoded, does it matter? **/
 	ctrl->dq_pins_interleaved = cfg->dq_pins_interleaved;
+	ctrl->restore_mrs = false;
 	ctrl->bootmode = bootmode;
 }
 
diff --git a/src/northbridge/intel/haswell/native_raminit/raminit_native.h b/src/northbridge/intel/haswell/native_raminit/raminit_native.h
index 4763b25e8d..4bc2a4955f 100644
--- a/src/northbridge/intel/haswell/native_raminit/raminit_native.h
+++ b/src/northbridge/intel/haswell/native_raminit/raminit_native.h
@@ -27,6 +27,30 @@
 /* Always use 12 legs for emphasis (not trained) */
 #define TXEQFULLDRV		(3 << 4)
 
+/* DDR3 mode register bits */
+#define MR0_DLL_RESET		BIT(8)
+
+#define MR1_WL_ENABLE		BIT(7)
+#define MR1_QOFF_ENABLE		BIT(12) /* If set, output buffers disabled */
+
+#define RTTNOM_MASK		(BIT(9) | BIT(6) | BIT(2))
+
+/* ZQ calibration types */
+enum {
+	ZQ_INIT,	/* DDR3: ZQCL with tZQinit, LPDDR3: ZQ Init  with tZQinit  */
+	ZQ_LONG,	/* DDR3: ZQCL with tZQoper, LPDDR3: ZQ Long  with tZQCL    */
+	ZQ_SHORT,	/* DDR3: ZQCS with tZQCS,   LPDDR3: ZQ Short with tZQCS    */
+	ZQ_RESET,	/* DDR3: not used,          LPDDR3: ZQ Reset with tZQreset */
+};
+
+/* REUT initialisation modes */
+enum {
+	REUT_MODE_IDLE = 0,
+	REUT_MODE_TEST = 1,
+	REUT_MODE_MRS  = 2,
+	REUT_MODE_NOP  = 3, /* Normal operation mode */
+};
+
 enum command_training_iteration {
 	CT_ITERATION_CLOCK = 0,
 	CT_ITERATION_CMD_NORTH,
@@ -50,6 +74,7 @@ enum raminit_status {
 	RAMINIT_STATUS_UNSUPPORTED_MEMORY,
 	RAMINIT_STATUS_MPLL_INIT_FAILURE,
 	RAMINIT_STATUS_POLL_TIMEOUT,
+	RAMINIT_STATUS_REUT_ERROR,
 	RAMINIT_STATUS_UNSPECIFIED_ERROR, /** TODO: Deprecated in favor of specific values **/
 };
 
@@ -72,6 +97,7 @@ struct sysinfo {
 	uint32_t cpu;		/* CPUID value */
 
 	bool dq_pins_interleaved;
+	bool restore_mrs;
 
 	/** TODO: ECC support untested **/
 	bool is_ecc;
@@ -161,6 +187,11 @@ struct sysinfo {
 	union tc_bank_rank_b_reg tc_bankrank_b[NUM_CHANNELS];
 	union tc_bank_rank_c_reg tc_bankrank_c[NUM_CHANNELS];
 	union tc_bank_rank_d_reg tc_bankrank_d[NUM_CHANNELS];
+
+	uint16_t mr0[NUM_CHANNELS][NUM_SLOTS];
+	uint16_t mr1[NUM_CHANNELS][NUM_SLOTS];
+	uint16_t mr2[NUM_CHANNELS][NUM_SLOTS];
+	uint16_t mr3[NUM_CHANNELS][NUM_SLOTS];
 };
 
 static inline bool is_hsw_ult(void)
@@ -196,6 +227,53 @@ static inline void clear_data_offset_train_all(struct sysinfo *ctrl)
 	memset(ctrl->data_offset_train, 0, sizeof(ctrl->data_offset_train));
 }
 
+/* Number of ticks to wait in units of 69.841279 ns (citation needed) */
+static inline void tick_delay(const uint32_t delay)
+{
+	/* Just perform reads to a random register */
+	for (uint32_t start = 0; start <= delay; start++)
+		mchbar_read32(REUT_ERR_DATA_STATUS);
+}
+
+/*
+ * 64-bit MCHBAR registers need to be accessed atomically. If one uses
+ * two 32-bit ops instead, there will be problems with the REUT's CADB
+ * (Command Address Data Buffer): hardware automatically advances the
+ * pointer into the register file after a write to the input register.
+ */
+static inline uint64_t mchbar_read64(const uintptr_t x)
+{
+	const uint64_t *offset = (uint64_t *)(CONFIG_FIXED_MCHBAR_MMIO_BASE + x);
+	uint64_t mmxsave, v;
+	asm volatile (
+		"\n\t movq %%mm0, %0"
+		"\n\t movq %2, %%mm0"
+		"\n\t movq %%mm0, %1"
+		"\n\t movq %3, %%mm0"
+		"\n\t emms"
+		: "=m"(mmxsave),
+		  "=m"(v)
+		: "m"(offset[0]),
+		  "m"(mmxsave));
+	return v;
+}
+
+static inline void mchbar_write64(const uintptr_t x, const uint64_t v)
+{
+	const uint64_t *offset = (uint64_t *)(CONFIG_FIXED_MCHBAR_MMIO_BASE + x);
+	uint64_t mmxsave;
+	asm volatile (
+		"\n\t movq %%mm0, %0"
+		"\n\t movq %2, %%mm0"
+		"\n\t movq %%mm0, %1"
+		"\n\t movq %3, %%mm0"
+		"\n\t emms"
+		: "=m"(mmxsave)
+		: "m"(offset[0]),
+		  "m"(v),
+		  "m"(mmxsave));
+}
+
 void raminit_main(enum raminit_boot_mode bootmode);
 
 enum raminit_status collect_spd_info(struct sysinfo *ctrl);
@@ -203,6 +281,7 @@ enum raminit_status initialise_mpll(struct sysinfo *ctrl);
 enum raminit_status convert_timings(struct sysinfo *ctrl);
 enum raminit_status configure_mc(struct sysinfo *ctrl);
 enum raminit_status configure_memory_map(struct sysinfo *ctrl);
+enum raminit_status do_jedec_init(struct sysinfo *ctrl);
 
 void configure_timings(struct sysinfo *ctrl);
 void configure_refresh(struct sysinfo *ctrl);
@@ -215,8 +294,28 @@ uint32_t get_tXS_offset(uint32_t mem_clock_mhz);
 uint32_t get_tZQOPER(uint32_t mem_clock_mhz, bool lpddr);
 uint32_t get_tZQCS(uint32_t mem_clock_mhz, bool lpddr);
 
+enum raminit_status io_reset(void);
 enum raminit_status wait_for_first_rcomp(void);
 
+uint16_t encode_ddr3_rttnom(uint32_t rttnom);
+void ddr3_program_mr1(struct sysinfo *ctrl, uint8_t wl_mode, uint8_t q_off);
+enum raminit_status ddr3_jedec_init(struct sysinfo *ctrl);
+
+void reut_issue_mrs(
+	struct sysinfo *ctrl,
+	uint8_t channel,
+	uint8_t rankmask,
+	uint8_t mr,
+	uint16_t val);
+
+void reut_issue_mrs_all(
+	struct sysinfo *ctrl,
+	uint8_t channel,
+	uint8_t mr,
+	const uint16_t val[NUM_SLOTS]);
+
+enum raminit_status reut_issue_zq(struct sysinfo *ctrl, uint8_t chanmask, uint8_t zq_type);
+
 uint8_t get_rx_bias(const struct sysinfo *ctrl);
 
 uint8_t get_tCWL(uint32_t mem_clock_mhz);
diff --git a/src/northbridge/intel/haswell/native_raminit/reg_structs.h b/src/northbridge/intel/haswell/native_raminit/reg_structs.h
index 70487e1640..9929f617fe 100644
--- a/src/northbridge/intel/haswell/native_raminit/reg_structs.h
+++ b/src/northbridge/intel/haswell/native_raminit/reg_structs.h
@@ -335,6 +335,127 @@ union mcscheds_cbit_reg {
 	uint32_t raw;
 };
 
+union reut_pat_cadb_prog_reg {
+	struct __packed {
+		uint32_t addr : 16; // Bits 15:0
+		uint32_t      :  8; // Bits 23:16
+		uint32_t bank :  3; // Bits 26:24
+		uint32_t      :  5; // Bits 31:27
+		uint32_t cs   :  4; // Bits 35:32
+		uint32_t      :  4; // Bits 39:36
+		uint32_t cmd  :  3; // Bits 42:40
+		uint32_t      :  5; // Bits 47:43
+		uint32_t odt  :  4; // Bits 51:48
+		uint32_t      :  4; // Bits 55:52
+		uint32_t cke  :  4; // Bits 59:56
+		uint32_t      :  4; // Bits 63:60
+	};
+	uint64_t raw;
+	uint32_t raw32[2];
+};
+
+union reut_pat_cadb_mrs_reg {
+	struct __packed {
+		uint32_t delay_gap : 3; // Bits  2:0
+		uint32_t           : 5; // Bits  7:3
+		uint32_t start_ptr : 3; // Bits 10:8
+		uint32_t           : 5; // Bits 15:11
+		uint32_t end_ptr   : 3; // Bits 18:16
+		uint32_t           : 5; // Bits 23:19
+		uint32_t curr_ptr  : 3; // Bits 26:24
+		uint32_t           : 5; // Bits 31:27
+	};
+	uint32_t raw;
+};
+
+union reut_seq_cfg_reg {
+	struct __packed {
+		uint32_t                               :  3; // Bits  2:0
+		uint32_t stop_base_seq_on_wrap_trigger :  1; // Bits  3:3
+		uint32_t                               :  1; // Bits  4:4
+		uint32_t address_update_rate_mode      :  1; // Bits  5:5
+		uint32_t                               :  1; // Bits  6:6
+		uint32_t enable_dummy_reads            :  1; // Bits  7:7
+		uint32_t                               :  2; // Bits  9:8
+		uint32_t enable_constant_write_strobe  :  1; // Bits 10:10
+		uint32_t global_control                :  1; // Bits 11:11
+		uint32_t initialization_mode           :  2; // Bits 13:12
+		uint32_t                               :  2; // Bits 15:14
+		uint32_t early_steppings_loop_count    :  5; // Bits 20:16   *** Not on C0 ***
+		uint32_t                               :  3; // Bits 23:21
+		uint32_t subsequence_start_pointer     :  3; // Bits 26:24
+		uint32_t                               :  1; // Bits 27:27
+		uint32_t subsequence_end_pointer       :  3; // Bits 30:28
+		uint32_t                               :  1; // Bits 31:31
+		uint32_t start_test_delay              : 10; // Bits 41:32
+		uint32_t                               : 22; // Bits 63:42
+	};
+	uint64_t raw;
+	uint32_t raw32[2];
+};
+
+union reut_seq_ctl_reg {
+	struct __packed {
+		uint32_t start_test    :  1; // Bits  0:0
+		uint32_t stop_test     :  1; // Bits  1:1
+		uint32_t clear_errors  :  1; // Bits  2:2
+		uint32_t               :  1; // Bits  3:3
+		uint32_t stop_on_error :  1; // Bits  4:4
+		uint32_t               : 27; // Bits 31:5
+	};
+	uint32_t raw;
+};
+
+union reut_global_err_reg {
+	struct __packed {
+		uint32_t ch_error     :  2; // Bits  1:0
+		uint32_t              : 14; // Bits 15:2
+		uint32_t ch_test_done :  2; // Bits 17:16
+		uint32_t              : 14; // Bits 31:18
+	};
+	uint32_t raw;
+};
+
+union reut_misc_cke_ctrl_reg {
+	struct __packed {
+		uint32_t cke_override           :  4; // Bits  3:0
+		uint32_t                        :  4; // Bits  7:4
+		uint32_t cke_en_start_test_sync :  1; // Bits  8:8
+		uint32_t                        :  7; // Bits 15:9
+		uint32_t cke_on                 :  4; // Bits 19:16
+		uint32_t                        : 12; // Bits 31:20
+	};
+	uint32_t raw;
+};
+
+union reut_misc_odt_ctrl_reg {
+	struct __packed {
+		uint32_t odt_override     :  4; // Bits  3:0
+		uint32_t                  : 12; // Bits 15:4
+		uint32_t odt_on           :  4; // Bits 19:16
+		uint32_t                  : 11; // Bits 30:20
+		uint32_t mpr_train_ddr_on :  1; // Bits 31:31
+	};
+	uint32_t raw;
+};
+
+union mcscheds_dft_misc_reg {
+	struct __packed {
+		uint32_t wdar                 :  1; // Bits  0:0
+		uint32_t safe_mask_sel        :  3; // Bits  3:1
+		uint32_t force_rcv_en         :  1; // Bits  4:4
+		uint32_t                      :  3; // Bits  7:5
+		uint32_t ddr_qualifier        :  2; // Bits  9:8
+		uint32_t qualifier_length     :  2; // Bits 11:10
+		uint32_t wdb_block_en         :  1; // Bits 12:12
+		uint32_t rt_dft_read_ptr      :  4; // Bits 16:13
+		uint32_t rt_dft_read_enable   :  1; // Bits 17:17
+		uint32_t rt_dft_read_sel_addr :  1; // Bits 18:18
+		uint32_t                      : 13; // Bits 31:19
+	};
+	uint32_t raw;
+};
+
 union tc_bank_reg {
 	struct __packed {
 		uint32_t tRCD      : 5; // Bits  4:0
@@ -428,6 +549,18 @@ union tc_srftp_reg {
 	uint32_t raw;
 };
 
+union tc_mr2_shadow_reg {
+	struct __packed {
+		uint32_t mr2_shadow_low   :  6; // Bits  5:0
+		uint32_t srt_available    :  2; // Bits  7:6
+		uint32_t mr2_shadow_high  :  3; // Bits 10:8
+		uint32_t                  :  3; // Bits 13:11
+		uint32_t addr_bit_swizzle :  2; // Bits 15:14
+		uint32_t                  : 16; // Bits 31:16
+	};
+	uint32_t raw;
+};
+
 union mcmain_command_rate_limit_reg {
 	struct __packed {
 		uint32_t enable_cmd_limit :  1; // Bits  0:0
@@ -483,6 +616,27 @@ union mad_zr_reg {
 	uint32_t raw;
 };
 
+union mc_init_state_g_reg {
+	struct __packed {
+		uint32_t pu_mrc_done        :  1; // Bits  0:0
+		uint32_t ddr_not_reset      :  1; // Bits  1:1
+		uint32_t                    :  1; // Bits  2:2
+		uint32_t refresh_enable     :  1; // Bits  3:3
+		uint32_t                    :  1; // Bits  4:4
+		uint32_t mc_init_done_ack   :  1; // Bits  5:5
+		uint32_t                    :  1; // Bits  6:6
+		uint32_t mrc_done           :  1; // Bits  7:7
+		uint32_t safe_self_refresh  :  1; // Bits  8:8
+		uint32_t                    :  1; // Bits  9:9
+		uint32_t hvm_gate_ddr_reset :  1; // Bits 10:10
+		uint32_t                    : 11; // Bits 21:11
+		uint32_t dclk_enable        :  1; // Bits 22:22
+		uint32_t reset_io           :  1; // Bits 23:23
+		uint32_t                    :  8; // Bits 31:24
+	};
+	uint32_t raw;
+};
+
 /* Same definition for P_COMP, M_COMP, D_COMP */
 union pcu_comp_reg {
 	struct __packed {
diff --git a/src/northbridge/intel/haswell/native_raminit/reut.c b/src/northbridge/intel/haswell/native_raminit/reut.c
new file mode 100644
index 0000000000..31019f74a1
--- /dev/null
+++ b/src/northbridge/intel/haswell/native_raminit/reut.c
@@ -0,0 +1,196 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#include <console/console.h>
+#include <delay.h>
+#include <northbridge/intel/haswell/haswell.h>
+#include <timer.h>
+#include <types.h>
+
+#include "raminit_native.h"
+
+enum {
+	CADB_CMD_MRS = 0,
+	CADB_CMD_REF = 1,
+	CADB_CMD_PRE = 2,
+	CADB_CMD_ACT = 3,
+	CADB_CMD_WR  = 4,
+	CADB_CMD_RD  = 5,
+	CADB_CMD_ZQ  = 6,
+	CADB_CMD_NOP = 7,
+};
+
+/*
+ * DDR3 rank mirror swaps the following pins: A3<->A4, A5<->A6, A7<->A8, BA0<->BA1
+ *
+ * Note that the swapped bits are contiguous. We can use some XOR magic to swap the bits.
+ * Address lanes are at bits 0..15 and bank selects are at bits 24..26 on the REUT register.
+ */
+#define MIRROR_BITS	(BIT(24) | BIT(7) | BIT(5) | BIT(3))
+static uint64_t cadb_prog_rank_mirror(const uint64_t cadb_prog)
+{
+	/* First XOR: find which pairs of bits are different (need swapping) */
+	const uint64_t tmp64 = (cadb_prog ^ (cadb_prog >> 1)) & MIRROR_BITS;
+
+	/* Second XOR: invert the pairs of bits that have different values */
+	return cadb_prog ^ (tmp64 | tmp64 << 1);
+}
+
+static enum raminit_status reut_write_cadb_cmd(
+	struct sysinfo *ctrl,
+	const uint8_t channel,
+	const uint8_t rankmask,
+	const uint8_t cmd,
+	const uint8_t bank,
+	const uint16_t valarr[NUM_SLOTRANKS],
+	const uint8_t delay)
+{
+	union mcscheds_dft_misc_reg dft_misc = {
+		.raw = mchbar_read32(MCSCHEDS_DFT_MISC),
+	};
+	dft_misc.ddr_qualifier = 0;
+	mchbar_write32(MCSCHEDS_DFT_MISC, dft_misc.raw);
+
+	/* Pointer will be dynamically incremented after a write to CADB_PROG register */
+	mchbar_write8(REUT_ch_PAT_CADB_WRITE_PTR(channel), 0);
+
+	uint8_t count = 0;
+	for (uint8_t rank = 0; rank < NUM_SLOTRANKS; rank++) {
+		if (!(ctrl->rankmap[channel] & BIT(rank) & rankmask))
+			continue;
+
+		union reut_pat_cadb_prog_reg reut_cadb_prog = {
+			.addr = valarr[rank],
+			.bank = bank,
+			.cs   = ~BIT(rank), /* CS is active low */
+			.cmd  = cmd,
+			.cke  = 0xf,
+		};
+		if (ctrl->rank_mirrored[channel] & BIT(rank))
+			reut_cadb_prog.raw = cadb_prog_rank_mirror(reut_cadb_prog.raw);
+
+		mchbar_write64(REUT_ch_PAT_CADB_PROG(channel), reut_cadb_prog.raw);
+		count++;
+	}
+	if (!count) {
+		printk(BIOS_ERR, "%s: rankmask is invalid\n", __func__);
+		return RAMINIT_STATUS_UNSPECIFIED_ERROR;	/** FIXME: Is this needed? **/
+	}
+	const union reut_pat_cadb_mrs_reg reut_cadb_mrs = {
+		.delay_gap = delay ? delay : 3,
+		.end_ptr   = count - 1,
+	};
+	mchbar_write32(REUT_ch_PAT_CADB_MRS(channel), reut_cadb_mrs.raw);
+
+	const uint32_t reut_seq_cfg_save = mchbar_read32(REUT_ch_SEQ_CFG(channel));
+	union reut_seq_cfg_reg reut_seq_cfg = {
+		.raw = reut_seq_cfg_save,
+	};
+	reut_seq_cfg.global_control = 0;
+	reut_seq_cfg.initialization_mode = REUT_MODE_MRS;
+	mchbar_write32(REUT_ch_SEQ_CFG(channel), reut_seq_cfg.raw);
+	mchbar_write32(REUT_ch_SEQ_CTL(channel), (union reut_seq_ctl_reg) {
+		.start_test   = 1,
+		.clear_errors = 1,
+	}.raw);
+	enum raminit_status status = RAMINIT_STATUS_SUCCESS;
+	union reut_global_err_reg reut_global_err;
+	struct stopwatch timer;
+	stopwatch_init_msecs_expire(&timer, 100);
+	do {
+		reut_global_err.raw = mchbar_read32(REUT_GLOBAL_ERR);
+		if (reut_global_err.ch_error & BIT(channel)) {
+			printk(BIOS_ERR, "Unexpected REUT error for channel %u\n", channel);
+			status = RAMINIT_STATUS_REUT_ERROR;
+			break;
+		}
+		if (stopwatch_expired(&timer)) {
+			printk(BIOS_ERR, "%s: REUT timed out!\n", __func__);
+			status = RAMINIT_STATUS_POLL_TIMEOUT;
+			break;
+		}
+	} while (!(reut_global_err.ch_test_done & BIT(channel)));
+	mchbar_write32(REUT_ch_SEQ_CTL(channel), (union reut_seq_ctl_reg) {
+		.clear_errors = 1,
+	}.raw);
+	mchbar_write32(REUT_ch_SEQ_CFG(channel), reut_seq_cfg_save);
+	return status;
+}
+
+static enum raminit_status reut_write_cadb_cmd_all(
+	struct sysinfo *ctrl,
+	const uint8_t channel,
+	const uint8_t rankmask,
+	const uint8_t cmd,
+	const uint8_t bank,
+	const uint16_t val,
+	const uint8_t delay)
+{
+	const uint16_t valarr[NUM_SLOTRANKS] = { val, val, val, val };
+	return reut_write_cadb_cmd(ctrl, channel, rankmask, cmd, bank, valarr, delay);
+}
+
+void reut_issue_mrs(
+	struct sysinfo *ctrl,
+	const uint8_t channel,
+	const uint8_t rankmask,
+	const uint8_t mr,
+	const uint16_t val)
+{
+	reut_write_cadb_cmd_all(ctrl, channel, rankmask, CADB_CMD_MRS, mr, val, 0);
+}
+
+void reut_issue_mrs_all(
+	struct sysinfo *ctrl,
+	const uint8_t channel,
+	const uint8_t mr,
+	const uint16_t val[NUM_SLOTS])
+{
+	const uint16_t valarr[NUM_SLOTRANKS] = { val[0], val[0], val[1], val[1] };
+	reut_write_cadb_cmd(ctrl, channel, 0xf, CADB_CMD_MRS, mr, valarr, 0);
+}
+
+enum raminit_status reut_issue_zq(struct sysinfo *ctrl, uint8_t chanmask, uint8_t zq_type)
+{
+	/** TODO: Issuing ZQ commands differs for LPDDR **/
+	if (ctrl->lpddr)
+		die("%s: LPDDR not yet supported in ZQ calibration\n", __func__);
+
+	__maybe_unused uint8_t opcode; /* NOTE: Only used for LPDDR */
+	uint16_t zq = 0;
+	switch (zq_type) {
+	case ZQ_INIT:
+		zq = BIT(10);
+		opcode = 0xff;
+		break;
+	case ZQ_LONG:
+		zq = BIT(10);
+		opcode = 0xab;
+		break;
+	case ZQ_SHORT:
+		opcode = 0x56;
+		break;
+	case ZQ_RESET:
+		opcode = 0xc3;
+		break;
+	default:
+		die("%s: ZQ type %u is invalid\n", __func__, zq_type);
+	}
+
+	/* ZQCS on single-channel needs a longer delay */
+	const uint8_t delay = zq_type == ZQ_SHORT && (!ctrl->dpc[0] || !ctrl->dpc[1]) ? 7 : 1;
+	enum raminit_status status = RAMINIT_STATUS_SUCCESS;
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		if (!(BIT(channel) & chanmask) || !does_ch_exist(ctrl, channel))
+			continue;
+
+		status = reut_write_cadb_cmd_all(ctrl, channel, 0xf, CADB_CMD_ZQ, 0, zq, delay);
+		if (status)
+			break;
+	}
+
+	/* Wait a bit after ZQ INIT and ZQCL commands */
+	if (zq)
+		udelay(1);
+
+	return status;
+}
diff --git a/src/northbridge/intel/haswell/registers/mchbar.h b/src/northbridge/intel/haswell/registers/mchbar.h
index 2acc5cbbc8..4fc78a7f43 100644
--- a/src/northbridge/intel/haswell/registers/mchbar.h
+++ b/src/northbridge/intel/haswell/registers/mchbar.h
@@ -96,15 +96,36 @@
 
 #define SC_WR_ADD_DELAY_ch(ch)			_MCMAIN_C(0x40d0, ch)
 
+#define REUT_ch_MISC_CKE_CTRL(ch)		_MCMAIN_C(0x4190, ch)
+
+#define REUT_ch_PAT_CADB_MRS(ch)		_MCMAIN_C(0x419c, ch)
+
+#define REUT_ch_PAT_CADB_WRITE_PTR(ch)		_MCMAIN_C(0x41bc, ch)
+#define REUT_ch_PAT_CADB_PROG(ch)		_MCMAIN_C(0x41c0, ch)
+
 #define TC_ZQCAL_ch(ch)				_MCMAIN_C(0x4290, ch)
 #define TC_RFP_ch(ch)				_MCMAIN_C(0x4294, ch)
 #define TC_RFTP_ch(ch)				_MCMAIN_C(0x4298, ch)
+#define TC_MR2_SHADOW_ch(ch)			_MCMAIN_C(0x429c, ch)
 #define MC_INIT_STATE_ch(ch)			_MCMAIN_C(0x42a0, ch)
 #define TC_SRFTP_ch(ch)				_MCMAIN_C(0x42a4, ch)
 
+#define REUT_GLOBAL_ERR				0x4804
+
+#define REUT_ch_SEQ_CFG(ch)			(0x48a8 + 8 * (ch))
+
+#define REUT_ch_SEQ_CTL(ch)			(0x48b8 + 4 * (ch))
+
 /* MCMAIN broadcast */
 #define MCSCHEDS_CBIT		0x4c20
 
+#define MCSCHEDS_DFT_MISC	0x4c30
+
+#define REUT_ERR_DATA_STATUS	0x4ce0
+
+#define REUT_MISC_CKE_CTRL	0x4d90
+#define REUT_MISC_ODT_CTRL	0x4d94
+
 #define MCMNTS_SC_WDBWM		0x4f8c
 
 /* MCDECS */
diff --git a/src/southbridge/intel/lynxpoint/pch.h b/src/southbridge/intel/lynxpoint/pch.h
index 07f4b9dc16..5b3696347c 100644
--- a/src/southbridge/intel/lynxpoint/pch.h
+++ b/src/southbridge/intel/lynxpoint/pch.h
@@ -586,6 +586,8 @@ void mainboard_config_rcba(void);
 #define ACPIIRQEN	0x31e0	/* 32bit */
 #define OIC		0x31fe	/* 16bit */
 #define PRSTS		0x3310	/* 32bit */
+#define PM_CFG2		0x333c	/* 32bit */
+#define  PM_CFG2_DRAM_RESET_CTL	(1 << 26)	/* ULT only */
 #define PMSYNC_CONFIG	0x33c4	/* 32bit */
 #define PMSYNC_CONFIG2	0x33cc	/* 32bit */
 #define SOFT_RESET_CTRL 0x38f4
-- 
2.39.2