rockchip: add support save/restore configuration for DDR during enter S3

This patch intend to support save the registers of the DDR controller
and PHY before suspend, and restore them after resume.

Change-Id: Ia10b476c0b837628ac0f365416a7118292753e96
Signed-off-by: Xing Zheng <zhengxing@rock-chips.com>
Signed-off-by: Derek Basehore <dbasehore@chromium.org>
Signed-off-by: Caesar Wang <wxt@rock-chips.com>

plat/rockchip/rk3399/drivers/dram/suspend.c

+/*
+ * Copyright (c) 2016, ARM Limited and Contributors. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * Redistributions of source code must retain the above copyright notice, this
+ * list of conditions and the following disclaimer.
+ *
+ * Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * Neither the name of ARM nor the names of its contributors may be used
+ * to endorse or promote products derived from this software without specific
+ * prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+#include <debug.h>
+#include <arch_helpers.h>
+#include <platform_def.h>
+#include <plat_private.h>
+#include <dram.h>
+#include <pmu_regs.h>
+#include <rk3399_def.h>
+#include <soc.h>
+#include <suspend.h>
+
+#define PMUGRF_OS_REG0			0x300
+#define PMUGRF_OS_REG1			0x304
+#define PMUGRF_OS_REG2			0x308
+#define PMUGRF_OS_REG3			0x30c
+
+#define CRU_SFTRST_DDR_CTRL(ch, n)	((0x1 << (8 + 16 + (ch) * 4)) | \
+					 ((n) << (8 + (ch) * 4)))
+#define CRU_SFTRST_DDR_PHY(ch, n)	((0x1 << (9 + 16 + (ch) * 4)) | \
+					 ((n) << (9 + (ch) * 4)))
+
+#define FBDIV_ENC(n)			((n) << 16)
+#define FBDIV_DEC(n)			(((n) >> 16) & 0xfff)
+#define POSTDIV2_ENC(n)			((n) << 12)
+#define POSTDIV2_DEC(n)			(((n) >> 12) & 0x7)
+#define POSTDIV1_ENC(n)			((n) << 8)
+#define POSTDIV1_DEC(n)			(((n) >> 8) & 0x7)
+#define REFDIV_ENC(n)			(n)
+#define REFDIV_DEC(n)			((n) & 0x3f)
+
+/* PMU CRU */
+#define PMUCRU_RSTNHOLD_CON0		0x120
+#define PMUCRU_RSTNHOLD_CON1		0x124
+
+#define PRESET_GPIO0_HOLD(n)		(((n) << 7) | WMSK_BIT(7))
+#define PRESET_GPIO1_HOLD(n)		(((n) << 8) | WMSK_BIT(8))
+
+#define SYS_COUNTER_FREQ_IN_MHZ		(SYS_COUNTER_FREQ_IN_TICKS / 1000000)
+
+/*
+ * Copy @num registers from @src to @dst
+ */
+__sramfunc void sram_regcpy(uintptr_t dst, uintptr_t src, uint32_t num)
+{
+	while (num--) {
+		mmio_write_32(dst, mmio_read_32(src));
+		dst += sizeof(uint32_t);
+		src += sizeof(uint32_t);
+	}
+}
+
+static __sramfunc uint32_t sram_get_timer_value(void)
+{
+	/*
+	 * Generic delay timer implementation expects the timer to be a down
+	 * counter. We apply bitwise NOT operator to the tick values returned
+	 * by read_cntpct_el0() to simulate the down counter.
+	 */
+	return (uint32_t)(~read_cntpct_el0());
+}
+
+static __sramfunc void sram_udelay(uint32_t usec)
+{
+	uint32_t start, cnt, delta, delta_us;
+
+	/* counter is decreasing */
+	start = sram_get_timer_value();
+	do {
+		cnt = sram_get_timer_value();
+		if (cnt > start) {
+			delta = UINT32_MAX - cnt;
+			delta += start;
+		} else
+			delta = start - cnt;
+		delta_us = (delta * SYS_COUNTER_FREQ_IN_MHZ);
+	} while (delta_us < usec);
+}
+
+static __sramfunc void configure_sgrf(void)
+{
+	/*
+	 * SGRF_DDR_RGN_DPLL_CLK and SGRF_DDR_RGN_RTC_CLK:
+	 * IC ECO bug, need to set this register.
+	 *
+	 * SGRF_DDR_RGN_BYPS:
+	 * After the PD_CENTER suspend/resume, the DDR region
+	 * related registers in the SGRF will be reset, we
+	 * need to re-initialize them.
+	 */
+	mmio_write_32(SGRF_BASE + SGRF_DDRRGN_CON0_16(16),
+		      SGRF_DDR_RGN_DPLL_CLK |
+		      SGRF_DDR_RGN_RTC_CLK |
+		      SGRF_DDR_RGN_BYPS);
+}
+
+static __sramfunc void rkclk_ddr_reset(uint32_t channel, uint32_t ctl,
+		uint32_t phy)
+{
+	channel &= 0x1;
+	ctl &= 0x1;
+	phy &= 0x1;
+	mmio_write_32(CRU_BASE + CRU_SOFTRST_CON(4),
+		      CRU_SFTRST_DDR_CTRL(channel, ctl) |
+		      CRU_SFTRST_DDR_PHY(channel, phy));
+}
+
+static __sramfunc void phy_pctrl_reset(uint32_t ch)
+{
+	rkclk_ddr_reset(ch, 1, 1);
+	sram_udelay(10);
+	rkclk_ddr_reset(ch, 1, 0);
+	sram_udelay(10);
+	rkclk_ddr_reset(ch, 0, 0);
+	sram_udelay(10);
+}
+
+static __sramfunc void phy_dll_bypass_set(uint32_t ch, uint32_t hz)
+{
+	if (hz <= 125 * MHz) {
+		/* phy_sw_master_mode_X PHY_86/214/342/470 4bits offset_8 */
+		mmio_setbits_32(PHY_REG(ch, 86), (0x3 << 2) << 8);
+		mmio_setbits_32(PHY_REG(ch, 214), (0x3 << 2) << 8);
+		mmio_setbits_32(PHY_REG(ch, 342), (0x3 << 2) << 8);
+		mmio_setbits_32(PHY_REG(ch, 470), (0x3 << 2) << 8);
+		/* phy_adrctl_sw_master_mode PHY_547/675/803 4bits offset_16 */
+		mmio_setbits_32(PHY_REG(ch, 547), (0x3 << 2) << 16);
+		mmio_setbits_32(PHY_REG(ch, 675), (0x3 << 2) << 16);
+		mmio_setbits_32(PHY_REG(ch, 803), (0x3 << 2) << 16);
+	} else {
+		/* phy_sw_master_mode_X PHY_86/214/342/470 4bits offset_8 */
+		mmio_clrbits_32(PHY_REG(ch, 86), (0x3 << 2) << 8);
+		mmio_clrbits_32(PHY_REG(ch, 214), (0x3 << 2) << 8);
+		mmio_clrbits_32(PHY_REG(ch, 342), (0x3 << 2) << 8);
+		mmio_clrbits_32(PHY_REG(ch, 470), (0x3 << 2) << 8);
+		/* phy_adrctl_sw_master_mode PHY_547/675/803 4bits offset_16 */
+		mmio_clrbits_32(PHY_REG(ch, 547), (0x3 << 2) << 16);
+		mmio_clrbits_32(PHY_REG(ch, 675), (0x3 << 2) << 16);
+		mmio_clrbits_32(PHY_REG(ch, 803), (0x3 << 2) << 16);
+	}
+}
+
+static __sramfunc void set_cs_training_index(uint32_t ch, uint32_t rank)
+{
+	/* PHY_8/136/264/392 phy_per_cs_training_index_X 1bit offset_24 */
+	mmio_clrsetbits_32(PHY_REG(ch, 8), 0x1 << 24, rank << 24);
+	mmio_clrsetbits_32(PHY_REG(ch, 136), 0x1 << 24, rank << 24);
+	mmio_clrsetbits_32(PHY_REG(ch, 264), 0x1 << 24, rank << 24);
+	mmio_clrsetbits_32(PHY_REG(ch, 392), 0x1 << 24, rank << 24);
+}
+
+static __sramfunc void select_per_cs_training_index(uint32_t ch, uint32_t rank)
+{
+	/* PHY_84 PHY_PER_CS_TRAINING_EN_0 1bit offset_16 */
+	if ((mmio_read_32(PHY_REG(ch, 84)) >> 16) & 1)
+		set_cs_training_index(ch, rank);
+}
+
+static void override_write_leveling_value(uint32_t ch)
+{
+	uint32_t byte;
+
+	/* PHY_896 PHY_FREQ_SEL_MULTICAST_EN 1bit offset_0 */
+	mmio_setbits_32(PHY_REG(ch, 896), 1);
+
+	/*
+	 * PHY_8/136/264/392
+	 * phy_per_cs_training_multicast_en_X 1bit offset_16
+	 */
+	mmio_clrsetbits_32(PHY_REG(ch, 8), 0x1 << 16, 1 << 16);
+	mmio_clrsetbits_32(PHY_REG(ch, 136), 0x1 << 16, 1 << 16);
+	mmio_clrsetbits_32(PHY_REG(ch, 264), 0x1 << 16, 1 << 16);
+	mmio_clrsetbits_32(PHY_REG(ch, 392), 0x1 << 16, 1 << 16);
+
+	for (byte = 0; byte < 4; byte++)
+		mmio_clrsetbits_32(PHY_REG(ch, 63 + (128 * byte)),
+				   0xffff << 16,
+				   0x200 << 16);
+
+	/* PHY_896 PHY_FREQ_SEL_MULTICAST_EN 1bit offset_0 */
+	mmio_clrbits_32(PHY_REG(ch, 896), 1);
+
+	/* CTL_200 ctrlupd_req 1bit offset_8 */
+	mmio_clrsetbits_32(CTL_REG(ch, 200), 0x1 << 8, 0x1 << 8);
+}
+
+static __sramfunc int data_training(uint32_t ch,
+		struct rk3399_sdram_params *sdram_params,
+		uint32_t training_flag)
+{
+	uint32_t obs_0, obs_1, obs_2, obs_3, obs_err = 0;
+	uint32_t rank = sdram_params->ch[ch].rank;
+	uint32_t rank_mask;
+	uint32_t i, tmp;
+
+	if (sdram_params->dramtype == LPDDR4)
+		rank_mask = (rank == 1) ? 0x5 : 0xf;
+	else
+		rank_mask = (rank == 1) ? 0x1 : 0x3;
+
+	/* PHY_927 PHY_PAD_DQS_DRIVE  RPULL offset_22 */
+	mmio_setbits_32(PHY_REG(ch, 927), (1 << 22));
+
+	if (training_flag == PI_FULL_TRAINING) {
+		if (sdram_params->dramtype == LPDDR4) {
+			training_flag = PI_WRITE_LEVELING |
+					PI_READ_GATE_TRAINING |
+					PI_READ_LEVELING |
+					PI_WDQ_LEVELING;
+		} else if (sdram_params->dramtype == LPDDR3) {
+			training_flag = PI_CA_TRAINING | PI_WRITE_LEVELING |
+					PI_READ_GATE_TRAINING;
+		} else if (sdram_params->dramtype == DDR3) {
+			training_flag = PI_WRITE_LEVELING |
+					PI_READ_GATE_TRAINING |
+					PI_READ_LEVELING;
+		}
+	}
+
+	/* ca training(LPDDR4,LPDDR3 support) */
+	if ((training_flag & PI_CA_TRAINING) == PI_CA_TRAINING) {
+		for (i = 0; i < 4; i++) {
+			if (!(rank_mask & (1 << i)))
+				continue;
+
+			select_per_cs_training_index(ch, i);
+			/* PI_100 PI_CALVL_EN:RW:8:2 */
+			mmio_clrsetbits_32(PI_REG(ch, 100), 0x3 << 8, 0x2 << 8);
+
+			/* PI_92 PI_CALVL_REQ:WR:16:1,PI_CALVL_CS:RW:24:2 */
+			mmio_clrsetbits_32(PI_REG(ch, 92),
+					   (0x1 << 16) | (0x3 << 24),
+					   (0x1 << 16) | (i << 24));
+			while (1) {
+				/* PI_174 PI_INT_STATUS:RD:8:18 */
+				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
+
+				/*
+				 * check status obs
+				 * PHY_532/660/788 phy_adr_calvl_obs1_:0:32
+				 */
+				obs_0 = mmio_read_32(PHY_REG(ch, 532));
+				obs_1 = mmio_read_32(PHY_REG(ch, 660));
+				obs_2 = mmio_read_32(PHY_REG(ch, 788));
+				if (((obs_0 >> 30) & 0x3) ||
+				    ((obs_1 >> 30) & 0x3) ||
+				    ((obs_2 >> 30) & 0x3))
+					obs_err = 1;
+				if ((((tmp >> 11) & 0x1) == 0x1) &&
+				    (((tmp >> 13) & 0x1) == 0x1) &&
+				    (((tmp >> 5) & 0x1) == 0x0) &&
+				    (obs_err == 0))
+					break;
+				else if ((((tmp >> 5) & 0x1) == 0x1) ||
+					 (obs_err == 1))
+					return -1;
+			}
+			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
+			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
+		}
+		mmio_clrbits_32(PI_REG(ch, 100), 0x3 << 8);
+	}
+
+	/* write leveling(LPDDR4,LPDDR3,DDR3 support) */
+	if ((training_flag & PI_WRITE_LEVELING) == PI_WRITE_LEVELING) {
+		for (i = 0; i < rank; i++) {
+			select_per_cs_training_index(ch, i);
+			/* PI_60 PI_WRLVL_EN:RW:8:2 */
+			mmio_clrsetbits_32(PI_REG(ch, 60), 0x3 << 8, 0x2 << 8);
+			/* PI_59 PI_WRLVL_REQ:WR:8:1,PI_WRLVL_CS:RW:16:2 */
+			mmio_clrsetbits_32(PI_REG(ch, 59),
+					   (0x1 << 8) | (0x3 << 16),
+					   (0x1 << 8) | (i << 16));
+
+			while (1) {
+				/* PI_174 PI_INT_STATUS:RD:8:18 */
+				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
+
+				/*
+				 * check status obs, if error maybe can not
+				 * get leveling done PHY_40/168/296/424
+				 * phy_wrlvl_status_obs_X:0:13
+				 */
+				obs_0 = mmio_read_32(PHY_REG(ch, 40));
+				obs_1 = mmio_read_32(PHY_REG(ch, 168));
+				obs_2 = mmio_read_32(PHY_REG(ch, 296));
+				obs_3 = mmio_read_32(PHY_REG(ch, 424));
+				if (((obs_0 >> 12) & 0x1) ||
+				    ((obs_1 >> 12) & 0x1) ||
+				    ((obs_2 >> 12) & 0x1) ||
+				    ((obs_3 >> 12) & 0x1))
+					obs_err = 1;
+				if ((((tmp >> 10) & 0x1) == 0x1) &&
+				    (((tmp >> 13) & 0x1) == 0x1) &&
+				    (((tmp >> 4) & 0x1) == 0x0) &&
+				    (obs_err == 0))
+					break;
+				else if ((((tmp >> 4) & 0x1) == 0x1) ||
+					 (obs_err == 1))
+					return -1;
+			}
+
+			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
+			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
+		}
+		override_write_leveling_value(ch);
+		mmio_clrbits_32(PI_REG(ch, 60), 0x3 << 8);
+	}
+
+	/* read gate training(LPDDR4,LPDDR3,DDR3 support) */
+	if ((training_flag & PI_READ_GATE_TRAINING) == PI_READ_GATE_TRAINING) {
+		for (i = 0; i < rank; i++) {
+			select_per_cs_training_index(ch, i);
+			/* PI_80 PI_RDLVL_GATE_EN:RW:24:2 */
+			mmio_clrsetbits_32(PI_REG(ch, 80), 0x3 << 24,
+					   0x2 << 24);
+			/*
+			 * PI_74 PI_RDLVL_GATE_REQ:WR:16:1
+			 * PI_RDLVL_CS:RW:24:2
+			 */
+			mmio_clrsetbits_32(PI_REG(ch, 74),
+					   (0x1 << 16) | (0x3 << 24),
+					   (0x1 << 16) | (i << 24));
+
+			while (1) {
+				/* PI_174 PI_INT_STATUS:RD:8:18 */
+				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
+
+				/*
+				 * check status obs
+				 * PHY_43/171/299/427
+				 *     PHY_GTLVL_STATUS_OBS_x:16:8
+				 */
+				obs_0 = mmio_read_32(PHY_REG(ch, 43));
+				obs_1 = mmio_read_32(PHY_REG(ch, 171));
+				obs_2 = mmio_read_32(PHY_REG(ch, 299));
+				obs_3 = mmio_read_32(PHY_REG(ch, 427));
+				if (((obs_0 >> (16 + 6)) & 0x3) ||
+				    ((obs_1 >> (16 + 6)) & 0x3) ||
+				    ((obs_2 >> (16 + 6)) & 0x3) ||
+				    ((obs_3 >> (16 + 6)) & 0x3))
+					obs_err = 1;
+				if ((((tmp >> 9) & 0x1) == 0x1) &&
+				    (((tmp >> 13) & 0x1) == 0x1) &&
+				    (((tmp >> 3) & 0x1) == 0x0) &&
+				    (obs_err == 0))
+					break;
+				else if ((((tmp >> 3) & 0x1) == 0x1) ||
+					 (obs_err == 1))
+					return -1;
+			}
+			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
+			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
+		}
+		mmio_clrbits_32(PI_REG(ch, 80), 0x3 << 24);
+	}
+
+	/* read leveling(LPDDR4,LPDDR3,DDR3 support) */
+	if ((training_flag & PI_READ_LEVELING) == PI_READ_LEVELING) {
+		for (i = 0; i < rank; i++) {
+			select_per_cs_training_index(ch, i);
+			/* PI_80 PI_RDLVL_EN:RW:16:2 */
+			mmio_clrsetbits_32(PI_REG(ch, 80), 0x3 << 16,
+					   0x2 << 16);
+			/* PI_74 PI_RDLVL_REQ:WR:8:1,PI_RDLVL_CS:RW:24:2 */
+			mmio_clrsetbits_32(PI_REG(ch, 74),
+					   (0x1 << 8) | (0x3 << 24),
+					   (0x1 << 8) | (i << 24));
+			while (1) {
+				/* PI_174 PI_INT_STATUS:RD:8:18 */
+				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
+
+				/*
+				 * make sure status obs not report error bit
+				 * PHY_46/174/302/430
+				 *     phy_rdlvl_status_obs_X:16:8
+				 */
+				if ((((tmp >> 8) & 0x1) == 0x1) &&
+				    (((tmp >> 13) & 0x1) == 0x1) &&
+				    (((tmp >> 2) & 0x1) == 0x0))
+					break;
+				else if (((tmp >> 2) & 0x1) == 0x1)
+					return -1;
+			}
+			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
+			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
+		}
+		mmio_clrbits_32(PI_REG(ch, 80), 0x3 << 16);
+	}
+
+	/* wdq leveling(LPDDR4 support) */
+	if ((training_flag & PI_WDQ_LEVELING) == PI_WDQ_LEVELING) {
+		for (i = 0; i < 4; i++) {
+			if (!(rank_mask & (1 << i)))
+				continue;
+
+			select_per_cs_training_index(ch, i);
+			/*
+			 * disable PI_WDQLVL_VREF_EN before wdq leveling?
+			 * PI_181 PI_WDQLVL_VREF_EN:RW:8:1
+			 */
+			mmio_clrbits_32(PI_REG(ch, 181), 0x1 << 8);
+			/* PI_124 PI_WDQLVL_EN:RW:16:2 */
+			mmio_clrsetbits_32(PI_REG(ch, 124), 0x3 << 16,
+					   0x2 << 16);
+			/* PI_121 PI_WDQLVL_REQ:WR:8:1,PI_WDQLVL_CS:RW:16:2 */
+			mmio_clrsetbits_32(PI_REG(ch, 121),
+					   (0x1 << 8) | (0x3 << 16),
+					   (0x1 << 8) | (i << 16));
+			while (1) {
+				/* PI_174 PI_INT_STATUS:RD:8:18 */
+				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
+				if ((((tmp >> 12) & 0x1) == 0x1) &&
+				    (((tmp >> 13) & 0x1) == 0x1) &&
+				    (((tmp >> 6) & 0x1) == 0x0))
+					break;
+				else if (((tmp >> 6) & 0x1) == 0x1)
+					return -1;
+			}
+			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
+			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
+		}
+		mmio_clrbits_32(PI_REG(ch, 124), 0x3 << 16);
+	}
+
+	/* PHY_927 PHY_PAD_DQS_DRIVE  RPULL offset_22 */
+	mmio_clrbits_32(PHY_REG(ch, 927), (1 << 22));
+
+	return 0;
+}
+
+static __sramfunc void set_ddrconfig(struct rk3399_sdram_params *sdram_params,
+		unsigned char channel, uint32_t ddrconfig)
+{
+	/* only need to set ddrconfig */
+	struct rk3399_sdram_channel *ch = &sdram_params->ch[channel];
+	unsigned int cs0_cap = 0;
+	unsigned int cs1_cap = 0;
+
+	cs0_cap = (1 << (ch->cs0_row + ch->col + ch->bk + ch->bw - 20));
+	if (ch->rank > 1)
+		cs1_cap = cs0_cap >> (ch->cs0_row - ch->cs1_row);
+	if (ch->row_3_4) {
+		cs0_cap = cs0_cap * 3 / 4;
+		cs1_cap = cs1_cap * 3 / 4;
+	}
+
+	mmio_write_32(MSCH_BASE(channel) + MSCH_DEVICECONF,
+		      ddrconfig | (ddrconfig << 6));
+	mmio_write_32(MSCH_BASE(channel) + MSCH_DEVICESIZE,
+		      ((cs0_cap / 32) & 0xff) | (((cs1_cap / 32) & 0xff) << 8));
+}
+
+static __sramfunc void dram_all_config(struct rk3399_sdram_params *sdram_params)
+{
+	unsigned int i;
+
+	for (i = 0; i < 2; i++) {
+		struct rk3399_sdram_channel *info = &sdram_params->ch[i];
+		struct rk3399_msch_timings *noc = &info->noc_timings;
+
+		if (sdram_params->ch[i].col == 0)
+			continue;
+
+		mmio_write_32(MSCH_BASE(i) + MSCH_DDRTIMINGA0,
+			      noc->ddrtiminga0.d32);
+		mmio_write_32(MSCH_BASE(i) + MSCH_DDRTIMINGB0,
+			      noc->ddrtimingb0.d32);
+		mmio_write_32(MSCH_BASE(i) + MSCH_DDRTIMINGC0,
+			      noc->ddrtimingc0.d32);
+		mmio_write_32(MSCH_BASE(i) + MSCH_DEVTODEV0,
+			      noc->devtodev0.d32);
+		mmio_write_32(MSCH_BASE(i) + MSCH_DDRMODE, noc->ddrmode.d32);
+
+		/* rank 1 memory clock disable (dfi_dram_clk_disable = 1) */
+		if (sdram_params->ch[i].rank == 1)
+			mmio_setbits_32(CTL_REG(i, 276), 1 << 17);
+	}
+
+	DDR_STRIDE(sdram_params->stride);
+
+	/* reboot hold register set */
+	mmio_write_32(PMUCRU_BASE + CRU_PMU_RSTHOLD_CON(1),
+		      CRU_PMU_SGRF_RST_RLS |
+		      PRESET_GPIO0_HOLD(1) |
+		      PRESET_GPIO1_HOLD(1));
+	mmio_clrsetbits_32(CRU_BASE + CRU_GLB_RST_CON, 0x3, 0x3);
+}
+
+static __sramfunc void pctl_cfg(uint32_t ch,
+		struct rk3399_sdram_params *sdram_params)
+{
+	const uint32_t *params_ctl = sdram_params->pctl_regs.denali_ctl;
+	const uint32_t *params_phy = sdram_params->phy_regs.denali_phy;
+	const uint32_t *params_pi = sdram_params->pi_regs.denali_pi;
+	uint32_t tmp, tmp1, tmp2;
+
+	/*
+	 * Workaround controller bug:
+	 * Do not program DRAM_CLASS until NO_PHY_IND_TRAIN_INT is programmed
+	 */
+	sram_regcpy(CTL_REG(ch, 1), (uintptr_t)&params_ctl[1],
+		    CTL_REG_NUM - 1);
+	mmio_write_32(CTL_REG(ch, 0), params_ctl[0]);
+	sram_regcpy(PI_REG(ch, 0), (uintptr_t)&params_pi[0],
+		    PI_REG_NUM);
+
+	mmio_write_32(PHY_REG(ch, 910), params_phy[910]);
+	mmio_write_32(PHY_REG(ch, 911), params_phy[911]);
+	mmio_write_32(PHY_REG(ch, 912), params_phy[912]);
+
+	mmio_clrsetbits_32(CTL_REG(ch, 68), PWRUP_SREFRESH_EXIT,
+				PWRUP_SREFRESH_EXIT);
+
+	/* PHY_DLL_RST_EN */
+	mmio_clrsetbits_32(PHY_REG(ch, 957), 0x3 << 24, 1 << 24);
+	dmbst();
+
+	mmio_setbits_32(PI_REG(ch, 0), START);
+	mmio_setbits_32(CTL_REG(ch, 0), START);
+
+	/* wait lock */
+	while (1) {
+		tmp = mmio_read_32(PHY_REG(ch, 920));
+		tmp1 = mmio_read_32(PHY_REG(ch, 921));
+		tmp2 = mmio_read_32(PHY_REG(ch, 922));
+		if ((((tmp >> 16) & 0x1) == 0x1) &&
+		     (((tmp1 >> 16) & 0x1) == 0x1) &&
+		     (((tmp1 >> 0) & 0x1) == 0x1) &&
+		     (((tmp2 >> 0) & 0x1) == 0x1))
+			break;
+		/* if PLL bypass,don't need wait lock */
+		if (mmio_read_32(PHY_REG(ch, 911)) & 0x1)
+			break;
+	}
+
+	sram_regcpy(PHY_REG(ch, 896), (uintptr_t)&params_phy[896], 63);
+	sram_regcpy(PHY_REG(ch, 0), (uintptr_t)&params_phy[0], 91);
+	sram_regcpy(PHY_REG(ch, 128), (uintptr_t)&params_phy[128], 91);
+	sram_regcpy(PHY_REG(ch, 256), (uintptr_t)&params_phy[256], 91);
+	sram_regcpy(PHY_REG(ch, 384), (uintptr_t)&params_phy[384], 91);
+	sram_regcpy(PHY_REG(ch, 512), (uintptr_t)&params_phy[512], 38);
+	sram_regcpy(PHY_REG(ch, 640), (uintptr_t)&params_phy[640], 38);
+	sram_regcpy(PHY_REG(ch, 768), (uintptr_t)&params_phy[768], 38);
+}
+
+static __sramfunc int dram_switch_to_phy_index1(
+		struct rk3399_sdram_params *sdram_params)
+{
+	uint32_t ch, ch_count;
+
+	mmio_write_32(CIC_BASE + CIC_CTRL0,
+		      (((0x3 << 4) | (1 << 2) | 1) << 16) |
+		      (1 << 4) | (1 << 2) | 1);
+	while (!(mmio_read_32(CIC_BASE + CIC_STATUS0) & (1 << 2)))
+		;
+
+	mmio_write_32(CIC_BASE + CIC_CTRL0, 0x20002);
+	while (!(mmio_read_32(CIC_BASE + CIC_STATUS0) & (1 << 0)))
+		;
+
+	ch_count = sdram_params->num_channels;
+
+	/* LPDDR4 f2 cann't do training, all training will fail */
+	for (ch = 0; ch < ch_count; ch++) {
+		mmio_clrsetbits_32(PHY_REG(ch, 896), (0x3 << 8) | 1,
+				   1 << 8);
+
+		/* data_training failed */
+		if (data_training(ch, sdram_params, PI_FULL_TRAINING))
+			return -1;
+	}
+
+	return 0;
+}
+
+/*
+ * Needs to be done for both channels at once in case of a shared reset signal
+ * between channels.
+ */
+static __sramfunc int pctl_start(uint32_t channel_mask,
+		struct rk3399_sdram_params *sdram_params)
+{
+	uint32_t count;
+
+	mmio_setbits_32(CTL_REG(0, 68), PWRUP_SREFRESH_EXIT);
+	mmio_setbits_32(CTL_REG(1, 68), PWRUP_SREFRESH_EXIT);
+
+	/* need de-access IO retention before controller START */
+	if (channel_mask & (1 << 0))
+		mmio_setbits_32(PMU_BASE + PMU_PWRMODE_CON, (1 << 19));
+	if (channel_mask & (1 << 1))
+		mmio_setbits_32(PMU_BASE + PMU_PWRMODE_CON, (1 << 23));
+
+	/* PHY_DLL_RST_EN */
+	if (channel_mask & (1 << 0))
+		mmio_clrsetbits_32(PHY_REG(0, 957), 0x3 << 24,
+				   0x2 << 24);
+	if (channel_mask & (1 << 1))
+		mmio_clrsetbits_32(PHY_REG(1, 957), 0x3 << 24,
+				   0x2 << 24);
+
+	/* check ERROR bit */
+	if (channel_mask & (1 << 0)) {
+		count = 0;
+		while (!(mmio_read_32(CTL_REG(0, 203)) & (1 << 3))) {
+			/* CKE is low, loop 10ms */
+			if (count > 100)
+				return -1;
+
+			sram_udelay(100);
+			count++;
+		}
+
+		mmio_clrbits_32(CTL_REG(0, 68), PWRUP_SREFRESH_EXIT);
+	}
+	if (channel_mask & (1 << 1)) {
+		count = 0;
+		while (!(mmio_read_32(CTL_REG(1, 203)) & (1 << 3))) {
+			/* CKE is low, loop 10ms */
+			if (count > 100)
+				return -1;
+
+			sram_udelay(100);
+			count++;
+		}
+
+		mmio_clrbits_32(CTL_REG(1, 68), PWRUP_SREFRESH_EXIT);
+	}
+
+	return 0;
+}
+
+void dmc_save(void)
+{
+	struct rk3399_sdram_params *sdram_params = &sdram_config;
+	uint32_t *params_ctl;
+	uint32_t *params_pi;
+	uint32_t *params_phy;
+	uint32_t refdiv, postdiv2, postdiv1, fbdiv;
+	uint32_t tmp;
+
+	params_ctl = sdram_params->pctl_regs.denali_ctl;
+	params_pi = sdram_params->pi_regs.denali_pi;
+	params_phy = sdram_params->phy_regs.denali_phy;
+
+	fbdiv = mmio_read_32(CRU_BASE + CRU_PLL_CON(DPLL_ID, 0)) & 0xfff;
+	tmp = mmio_read_32(CRU_BASE + CRU_PLL_CON(DPLL_ID, 1));
+	postdiv2 = POSTDIV2_DEC(tmp);
+	postdiv1 = POSTDIV1_DEC(tmp);
+	refdiv = REFDIV_DEC(tmp);
+
+	sdram_params->ddr_freq = ((fbdiv * 24) /
+				(refdiv * postdiv1 * postdiv2)) * MHz;
+
+	INFO("sdram_params->ddr_freq = %d\n", sdram_params->ddr_freq);
+	sdram_params->odt = (((mmio_read_32(PHY_REG(0, 5)) >> 16) &
+			       0x7) != 0) ? 1 : 0;
+
+	/* copy the registers CTL PI and PHY */
+	sram_regcpy((uintptr_t)&params_ctl[0], CTL_REG(0, 0), CTL_REG_NUM);
+
+	/* mask DENALI_CTL_00_DATA.START, only copy here, will trigger later */
+	params_ctl[0] &= ~(0x1 << 0);
+
+	sram_regcpy((uintptr_t)&params_pi[0], PI_REG(0, 0),
+		    PI_REG_NUM);
+
+	/* mask DENALI_PI_00_DATA.START, only copy here, will trigger later*/
+	params_pi[0] &= ~(0x1 << 0);
+
+	sram_regcpy((uintptr_t)&params_phy[0], PHY_REG(0, 0), 91);
+	sram_regcpy((uintptr_t)&params_phy[128], PHY_REG(0, 128), 91);
+	sram_regcpy((uintptr_t)&params_phy[256], PHY_REG(0, 256), 91);
+	sram_regcpy((uintptr_t)&params_phy[384], PHY_REG(0, 384), 91);
+	sram_regcpy((uintptr_t)&params_phy[512], PHY_REG(0, 512), 38);
+	sram_regcpy((uintptr_t)&params_phy[640], PHY_REG(0, 640), 38);
+	sram_regcpy((uintptr_t)&params_phy[768], PHY_REG(0, 768), 38);
+	sram_regcpy((uintptr_t)&params_phy[896], PHY_REG(0, 896), 63);
+
+	/* set DENALI_PHY_957_DATA.PHY_DLL_RST_EN = 0x1 */
+	params_phy[957] &= ~(0x3 << 24);
+	params_phy[957] |= 1 << 24;
+	params_phy[896] |= 1;
+	params_phy[896] &= ~(0x3 << 8);
+}
+
+__sramfunc void dmc_restore(void)
+{
+	struct rk3399_sdram_params *sdram_params = &sdram_config;
+	uint32_t channel_mask = 0;
+	uint32_t channel;
+
+	configure_sgrf();
+
+retry:
+	for (channel = 0; channel < sdram_params->num_channels; channel++) {
+		phy_pctrl_reset(channel);
+		phy_dll_bypass_set(channel, sdram_params->ddr_freq);
+		if (channel >= sdram_params->num_channels)
+			continue;
+
+		pctl_cfg(channel, sdram_params);
+	}
+
+	for (channel = 0; channel < 2; channel++) {
+		if (sdram_params->ch[channel].col)
+			channel_mask |= 1 << channel;
+	}
+
+	if (pctl_start(channel_mask, sdram_params) < 0)
+		goto retry;
+
+	for (channel = 0; channel < sdram_params->num_channels; channel++) {
+		/* LPDDR2/LPDDR3 need to wait DAI complete, max 10us */
+		if (sdram_params->dramtype == LPDDR3)
+			sram_udelay(10);
+
+		/* If traning fail, retry to do it again. */
+		if (data_training(channel, sdram_params, PI_FULL_TRAINING))
+			goto retry;
+
+		set_ddrconfig(sdram_params, channel,
+			      sdram_params->ch[channel].ddrconfig);
+	}
+
+	dram_all_config(sdram_params);
+
+	/* Switch to index 1 and prepare for DDR frequency switch. */
+	dram_switch_to_phy_index1(sdram_params);
+}

plat/rockchip/rk3399/drivers/dram/suspend.h

+/*
+ * Copyright (c) 2016, ARM Limited and Contributors. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * Redistributions of source code must retain the above copyright notice, this
+ * list of conditions and the following disclaimer.
+ *
+ * Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * Neither the name of ARM nor the names of its contributors may be used
+ * to endorse or promote products derived from this software without specific
+ * prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __SOC_ROCKCHIP_RK3399_SUSPEND_H__
+#define __SOC_ROCKCHIP_RK3399_SUSPEND_H__
+#include <dram.h>
+
+#define KHz (1000)
+#define MHz (1000 * KHz)
+#define GHz (1000 * MHz)
+
+#define PI_CA_TRAINING		(1 << 0)
+#define PI_WRITE_LEVELING	(1 << 1)
+#define PI_READ_GATE_TRAINING	(1 << 2)
+#define PI_READ_LEVELING	(1 << 3)
+#define PI_WDQ_LEVELING		(1 << 4)
+#define PI_FULL_TRAINING	(0xff)
+
+void dmc_save(void);
+__sramfunc void dmc_restore(void);
+__sramfunc void sram_regcpy(uintptr_t dst, uintptr_t src, uint32_t num);
+
+#endif /* __DRAM_H__ */

plat/rockchip/rk3399/drivers/soc/soc.h

@@ -245,6 +245,8 @@ struct deepsleep_data_s {
 #define SGRF_PMU_SLV_CON1_CFG		(SGRF_SLV_S_WMSK | \
 					SGRF_PMUSRAM_S)
 /* ddr region */
+#define SGRF_DDR_RGN_DPLL_CLK	BIT_WITH_WMSK(15) /* DDR PLL output clock */
+#define SGRF_DDR_RGN_RTC_CLK	BIT_WITH_WMSK(14) /* 32K clock for DDR PLL */
 #define SGRF_DDR_RGN_BYPS	BIT_WITH_WMSK(9) /* All of ddr rgn  is ns */
 
 /* The MST access the ddr rgn n with secure attribution */

plat/rockchip/rk3399/platform.mk

@@ -80,6 +80,7 @@ BL31_SOURCES            +=      ${RK_GIC_SOURCES}
 				${RK_PLAT_SOC}/drivers/pwm/pwm.c	\
                                 ${RK_PLAT_SOC}/drivers/soc/soc.c		\
 				${RK_PLAT_SOC}/drivers/dram/dfs.c		\
+                                ${RK_PLAT_SOC}/drivers/dram/suspend.c           \
 				${RK_PLAT_SOC}/drivers/dram/dram.c		\
 				${RK_PLAT_SOC}/drivers/dram/dram_spec_timing.c