xref: /rk3399_ARM-atf/plat/rockchip/rk3399/drivers/dram/suspend.c (revision 82cb2c1ad9897473743f08437d0a3995bed561b9)
1 /*
2  * Copyright (c) 2016, ARM Limited and Contributors. All rights reserved.
3  *
4  * SPDX-License-Identifier: BSD-3-Clause
5  */
6 #include <debug.h>
7 #include <arch_helpers.h>
8 #include <platform_def.h>
9 #include <plat_private.h>
10 #include <dram.h>
11 #include <pmu_regs.h>
12 #include <rk3399_def.h>
13 #include <secure.h>
14 #include <soc.h>
15 #include <suspend.h>
16 
17 #define PMUGRF_OS_REG0			0x300
18 #define PMUGRF_OS_REG1			0x304
19 #define PMUGRF_OS_REG2			0x308
20 #define PMUGRF_OS_REG3			0x30c
21 
22 #define CRU_SFTRST_DDR_CTRL(ch, n)	((0x1 << (8 + 16 + (ch) * 4)) | \
23 					 ((n) << (8 + (ch) * 4)))
24 #define CRU_SFTRST_DDR_PHY(ch, n)	((0x1 << (9 + 16 + (ch) * 4)) | \
25 					 ((n) << (9 + (ch) * 4)))
26 
27 #define FBDIV_ENC(n)			((n) << 16)
28 #define FBDIV_DEC(n)			(((n) >> 16) & 0xfff)
29 #define POSTDIV2_ENC(n)			((n) << 12)
30 #define POSTDIV2_DEC(n)			(((n) >> 12) & 0x7)
31 #define POSTDIV1_ENC(n)			((n) << 8)
32 #define POSTDIV1_DEC(n)			(((n) >> 8) & 0x7)
33 #define REFDIV_ENC(n)			(n)
34 #define REFDIV_DEC(n)			((n) & 0x3f)
35 
36 /* PMU CRU */
37 #define PMUCRU_RSTNHOLD_CON0		0x120
38 #define PMUCRU_RSTNHOLD_CON1		0x124
39 
40 #define PRESET_GPIO0_HOLD(n)		(((n) << 7) | WMSK_BIT(7))
41 #define PRESET_GPIO1_HOLD(n)		(((n) << 8) | WMSK_BIT(8))
42 
43 #define SYS_COUNTER_FREQ_IN_MHZ		(SYS_COUNTER_FREQ_IN_TICKS / 1000000)
44 
45 /*
46  * Copy @num registers from @src to @dst
47  */
48 __sramfunc void sram_regcpy(uintptr_t dst, uintptr_t src, uint32_t num)
49 {
50 	while (num--) {
51 		mmio_write_32(dst, mmio_read_32(src));
52 		dst += sizeof(uint32_t);
53 		src += sizeof(uint32_t);
54 	}
55 }
56 
57 static __sramfunc uint32_t sram_get_timer_value(void)
58 {
59 	/*
60 	 * Generic delay timer implementation expects the timer to be a down
61 	 * counter. We apply bitwise NOT operator to the tick values returned
62 	 * by read_cntpct_el0() to simulate the down counter.
63 	 */
64 	return (uint32_t)(~read_cntpct_el0());
65 }
66 
67 static __sramfunc void sram_udelay(uint32_t usec)
68 {
69 	uint32_t start, cnt, delta, delta_us;
70 
71 	/* counter is decreasing */
72 	start = sram_get_timer_value();
73 	do {
74 		cnt = sram_get_timer_value();
75 		if (cnt > start) {
76 			delta = UINT32_MAX - cnt;
77 			delta += start;
78 		} else
79 			delta = start - cnt;
80 		delta_us = (delta * SYS_COUNTER_FREQ_IN_MHZ);
81 	} while (delta_us < usec);
82 }
83 
84 static __sramfunc void configure_sgrf(void)
85 {
86 	/*
87 	 * SGRF_DDR_RGN_DPLL_CLK and SGRF_DDR_RGN_RTC_CLK:
88 	 * IC ECO bug, need to set this register.
89 	 *
90 	 * SGRF_DDR_RGN_BYPS:
91 	 * After the PD_CENTER suspend/resume, the DDR region
92 	 * related registers in the SGRF will be reset, we
93 	 * need to re-initialize them.
94 	 */
95 	mmio_write_32(SGRF_BASE + SGRF_DDRRGN_CON0_16(16),
96 		      SGRF_DDR_RGN_DPLL_CLK |
97 		      SGRF_DDR_RGN_RTC_CLK |
98 		      SGRF_DDR_RGN_BYPS);
99 }
100 
101 static __sramfunc void rkclk_ddr_reset(uint32_t channel, uint32_t ctl,
102 		uint32_t phy)
103 {
104 	channel &= 0x1;
105 	ctl &= 0x1;
106 	phy &= 0x1;
107 	mmio_write_32(CRU_BASE + CRU_SOFTRST_CON(4),
108 		      CRU_SFTRST_DDR_CTRL(channel, ctl) |
109 		      CRU_SFTRST_DDR_PHY(channel, phy));
110 }
111 
112 static __sramfunc void phy_pctrl_reset(uint32_t ch)
113 {
114 	rkclk_ddr_reset(ch, 1, 1);
115 	sram_udelay(10);
116 	rkclk_ddr_reset(ch, 1, 0);
117 	sram_udelay(10);
118 	rkclk_ddr_reset(ch, 0, 0);
119 	sram_udelay(10);
120 }
121 
122 static __sramfunc void phy_dll_bypass_set(uint32_t ch, uint32_t hz)
123 {
124 	if (hz <= 125 * MHz) {
125 		/* phy_sw_master_mode_X PHY_86/214/342/470 4bits offset_8 */
126 		mmio_setbits_32(PHY_REG(ch, 86), (0x3 << 2) << 8);
127 		mmio_setbits_32(PHY_REG(ch, 214), (0x3 << 2) << 8);
128 		mmio_setbits_32(PHY_REG(ch, 342), (0x3 << 2) << 8);
129 		mmio_setbits_32(PHY_REG(ch, 470), (0x3 << 2) << 8);
130 		/* phy_adrctl_sw_master_mode PHY_547/675/803 4bits offset_16 */
131 		mmio_setbits_32(PHY_REG(ch, 547), (0x3 << 2) << 16);
132 		mmio_setbits_32(PHY_REG(ch, 675), (0x3 << 2) << 16);
133 		mmio_setbits_32(PHY_REG(ch, 803), (0x3 << 2) << 16);
134 	} else {
135 		/* phy_sw_master_mode_X PHY_86/214/342/470 4bits offset_8 */
136 		mmio_clrbits_32(PHY_REG(ch, 86), (0x3 << 2) << 8);
137 		mmio_clrbits_32(PHY_REG(ch, 214), (0x3 << 2) << 8);
138 		mmio_clrbits_32(PHY_REG(ch, 342), (0x3 << 2) << 8);
139 		mmio_clrbits_32(PHY_REG(ch, 470), (0x3 << 2) << 8);
140 		/* phy_adrctl_sw_master_mode PHY_547/675/803 4bits offset_16 */
141 		mmio_clrbits_32(PHY_REG(ch, 547), (0x3 << 2) << 16);
142 		mmio_clrbits_32(PHY_REG(ch, 675), (0x3 << 2) << 16);
143 		mmio_clrbits_32(PHY_REG(ch, 803), (0x3 << 2) << 16);
144 	}
145 }
146 
147 static __sramfunc void set_cs_training_index(uint32_t ch, uint32_t rank)
148 {
149 	/* PHY_8/136/264/392 phy_per_cs_training_index_X 1bit offset_24 */
150 	mmio_clrsetbits_32(PHY_REG(ch, 8), 0x1 << 24, rank << 24);
151 	mmio_clrsetbits_32(PHY_REG(ch, 136), 0x1 << 24, rank << 24);
152 	mmio_clrsetbits_32(PHY_REG(ch, 264), 0x1 << 24, rank << 24);
153 	mmio_clrsetbits_32(PHY_REG(ch, 392), 0x1 << 24, rank << 24);
154 }
155 
156 static __sramfunc void select_per_cs_training_index(uint32_t ch, uint32_t rank)
157 {
158 	/* PHY_84 PHY_PER_CS_TRAINING_EN_0 1bit offset_16 */
159 	if ((mmio_read_32(PHY_REG(ch, 84)) >> 16) & 1)
160 		set_cs_training_index(ch, rank);
161 }
162 
163 static void override_write_leveling_value(uint32_t ch)
164 {
165 	uint32_t byte;
166 
167 	/* PHY_896 PHY_FREQ_SEL_MULTICAST_EN 1bit offset_0 */
168 	mmio_setbits_32(PHY_REG(ch, 896), 1);
169 
170 	/*
171 	 * PHY_8/136/264/392
172 	 * phy_per_cs_training_multicast_en_X 1bit offset_16
173 	 */
174 	mmio_clrsetbits_32(PHY_REG(ch, 8), 0x1 << 16, 1 << 16);
175 	mmio_clrsetbits_32(PHY_REG(ch, 136), 0x1 << 16, 1 << 16);
176 	mmio_clrsetbits_32(PHY_REG(ch, 264), 0x1 << 16, 1 << 16);
177 	mmio_clrsetbits_32(PHY_REG(ch, 392), 0x1 << 16, 1 << 16);
178 
179 	for (byte = 0; byte < 4; byte++)
180 		mmio_clrsetbits_32(PHY_REG(ch, 63 + (128 * byte)),
181 				   0xffff << 16,
182 				   0x200 << 16);
183 
184 	/* PHY_896 PHY_FREQ_SEL_MULTICAST_EN 1bit offset_0 */
185 	mmio_clrbits_32(PHY_REG(ch, 896), 1);
186 
187 	/* CTL_200 ctrlupd_req 1bit offset_8 */
188 	mmio_clrsetbits_32(CTL_REG(ch, 200), 0x1 << 8, 0x1 << 8);
189 }
190 
191 static __sramfunc int data_training(uint32_t ch,
192 		struct rk3399_sdram_params *sdram_params,
193 		uint32_t training_flag)
194 {
195 	uint32_t obs_0, obs_1, obs_2, obs_3, obs_err = 0;
196 	uint32_t rank = sdram_params->ch[ch].rank;
197 	uint32_t rank_mask;
198 	uint32_t i, tmp;
199 
200 	if (sdram_params->dramtype == LPDDR4)
201 		rank_mask = (rank == 1) ? 0x5 : 0xf;
202 	else
203 		rank_mask = (rank == 1) ? 0x1 : 0x3;
204 
205 	/* PHY_927 PHY_PAD_DQS_DRIVE  RPULL offset_22 */
206 	mmio_setbits_32(PHY_REG(ch, 927), (1 << 22));
207 
208 	if (training_flag == PI_FULL_TRAINING) {
209 		if (sdram_params->dramtype == LPDDR4) {
210 			training_flag = PI_WRITE_LEVELING |
211 					PI_READ_GATE_TRAINING |
212 					PI_READ_LEVELING |
213 					PI_WDQ_LEVELING;
214 		} else if (sdram_params->dramtype == LPDDR3) {
215 			training_flag = PI_CA_TRAINING | PI_WRITE_LEVELING |
216 					PI_READ_GATE_TRAINING;
217 		} else if (sdram_params->dramtype == DDR3) {
218 			training_flag = PI_WRITE_LEVELING |
219 					PI_READ_GATE_TRAINING |
220 					PI_READ_LEVELING;
221 		}
222 	}
223 
224 	/* ca training(LPDDR4,LPDDR3 support) */
225 	if ((training_flag & PI_CA_TRAINING) == PI_CA_TRAINING) {
226 		for (i = 0; i < 4; i++) {
227 			if (!(rank_mask & (1 << i)))
228 				continue;
229 
230 			select_per_cs_training_index(ch, i);
231 			/* PI_100 PI_CALVL_EN:RW:8:2 */
232 			mmio_clrsetbits_32(PI_REG(ch, 100), 0x3 << 8, 0x2 << 8);
233 
234 			/* PI_92 PI_CALVL_REQ:WR:16:1,PI_CALVL_CS:RW:24:2 */
235 			mmio_clrsetbits_32(PI_REG(ch, 92),
236 					   (0x1 << 16) | (0x3 << 24),
237 					   (0x1 << 16) | (i << 24));
238 			while (1) {
239 				/* PI_174 PI_INT_STATUS:RD:8:18 */
240 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
241 
242 				/*
243 				 * check status obs
244 				 * PHY_532/660/788 phy_adr_calvl_obs1_:0:32
245 				 */
246 				obs_0 = mmio_read_32(PHY_REG(ch, 532));
247 				obs_1 = mmio_read_32(PHY_REG(ch, 660));
248 				obs_2 = mmio_read_32(PHY_REG(ch, 788));
249 				if (((obs_0 >> 30) & 0x3) ||
250 				    ((obs_1 >> 30) & 0x3) ||
251 				    ((obs_2 >> 30) & 0x3))
252 					obs_err = 1;
253 				if ((((tmp >> 11) & 0x1) == 0x1) &&
254 				    (((tmp >> 13) & 0x1) == 0x1) &&
255 				    (((tmp >> 5) & 0x1) == 0x0) &&
256 				    (obs_err == 0))
257 					break;
258 				else if ((((tmp >> 5) & 0x1) == 0x1) ||
259 					 (obs_err == 1))
260 					return -1;
261 			}
262 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
263 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
264 		}
265 		mmio_clrbits_32(PI_REG(ch, 100), 0x3 << 8);
266 	}
267 
268 	/* write leveling(LPDDR4,LPDDR3,DDR3 support) */
269 	if ((training_flag & PI_WRITE_LEVELING) == PI_WRITE_LEVELING) {
270 		for (i = 0; i < rank; i++) {
271 			select_per_cs_training_index(ch, i);
272 			/* PI_60 PI_WRLVL_EN:RW:8:2 */
273 			mmio_clrsetbits_32(PI_REG(ch, 60), 0x3 << 8, 0x2 << 8);
274 			/* PI_59 PI_WRLVL_REQ:WR:8:1,PI_WRLVL_CS:RW:16:2 */
275 			mmio_clrsetbits_32(PI_REG(ch, 59),
276 					   (0x1 << 8) | (0x3 << 16),
277 					   (0x1 << 8) | (i << 16));
278 
279 			while (1) {
280 				/* PI_174 PI_INT_STATUS:RD:8:18 */
281 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
282 
283 				/*
284 				 * check status obs, if error maybe can not
285 				 * get leveling done PHY_40/168/296/424
286 				 * phy_wrlvl_status_obs_X:0:13
287 				 */
288 				obs_0 = mmio_read_32(PHY_REG(ch, 40));
289 				obs_1 = mmio_read_32(PHY_REG(ch, 168));
290 				obs_2 = mmio_read_32(PHY_REG(ch, 296));
291 				obs_3 = mmio_read_32(PHY_REG(ch, 424));
292 				if (((obs_0 >> 12) & 0x1) ||
293 				    ((obs_1 >> 12) & 0x1) ||
294 				    ((obs_2 >> 12) & 0x1) ||
295 				    ((obs_3 >> 12) & 0x1))
296 					obs_err = 1;
297 				if ((((tmp >> 10) & 0x1) == 0x1) &&
298 				    (((tmp >> 13) & 0x1) == 0x1) &&
299 				    (((tmp >> 4) & 0x1) == 0x0) &&
300 				    (obs_err == 0))
301 					break;
302 				else if ((((tmp >> 4) & 0x1) == 0x1) ||
303 					 (obs_err == 1))
304 					return -1;
305 			}
306 
307 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
308 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
309 		}
310 		override_write_leveling_value(ch);
311 		mmio_clrbits_32(PI_REG(ch, 60), 0x3 << 8);
312 	}
313 
314 	/* read gate training(LPDDR4,LPDDR3,DDR3 support) */
315 	if ((training_flag & PI_READ_GATE_TRAINING) == PI_READ_GATE_TRAINING) {
316 		for (i = 0; i < rank; i++) {
317 			select_per_cs_training_index(ch, i);
318 			/* PI_80 PI_RDLVL_GATE_EN:RW:24:2 */
319 			mmio_clrsetbits_32(PI_REG(ch, 80), 0x3 << 24,
320 					   0x2 << 24);
321 			/*
322 			 * PI_74 PI_RDLVL_GATE_REQ:WR:16:1
323 			 * PI_RDLVL_CS:RW:24:2
324 			 */
325 			mmio_clrsetbits_32(PI_REG(ch, 74),
326 					   (0x1 << 16) | (0x3 << 24),
327 					   (0x1 << 16) | (i << 24));
328 
329 			while (1) {
330 				/* PI_174 PI_INT_STATUS:RD:8:18 */
331 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
332 
333 				/*
334 				 * check status obs
335 				 * PHY_43/171/299/427
336 				 *     PHY_GTLVL_STATUS_OBS_x:16:8
337 				 */
338 				obs_0 = mmio_read_32(PHY_REG(ch, 43));
339 				obs_1 = mmio_read_32(PHY_REG(ch, 171));
340 				obs_2 = mmio_read_32(PHY_REG(ch, 299));
341 				obs_3 = mmio_read_32(PHY_REG(ch, 427));
342 				if (((obs_0 >> (16 + 6)) & 0x3) ||
343 				    ((obs_1 >> (16 + 6)) & 0x3) ||
344 				    ((obs_2 >> (16 + 6)) & 0x3) ||
345 				    ((obs_3 >> (16 + 6)) & 0x3))
346 					obs_err = 1;
347 				if ((((tmp >> 9) & 0x1) == 0x1) &&
348 				    (((tmp >> 13) & 0x1) == 0x1) &&
349 				    (((tmp >> 3) & 0x1) == 0x0) &&
350 				    (obs_err == 0))
351 					break;
352 				else if ((((tmp >> 3) & 0x1) == 0x1) ||
353 					 (obs_err == 1))
354 					return -1;
355 			}
356 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
357 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
358 		}
359 		mmio_clrbits_32(PI_REG(ch, 80), 0x3 << 24);
360 	}
361 
362 	/* read leveling(LPDDR4,LPDDR3,DDR3 support) */
363 	if ((training_flag & PI_READ_LEVELING) == PI_READ_LEVELING) {
364 		for (i = 0; i < rank; i++) {
365 			select_per_cs_training_index(ch, i);
366 			/* PI_80 PI_RDLVL_EN:RW:16:2 */
367 			mmio_clrsetbits_32(PI_REG(ch, 80), 0x3 << 16,
368 					   0x2 << 16);
369 			/* PI_74 PI_RDLVL_REQ:WR:8:1,PI_RDLVL_CS:RW:24:2 */
370 			mmio_clrsetbits_32(PI_REG(ch, 74),
371 					   (0x1 << 8) | (0x3 << 24),
372 					   (0x1 << 8) | (i << 24));
373 			while (1) {
374 				/* PI_174 PI_INT_STATUS:RD:8:18 */
375 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
376 
377 				/*
378 				 * make sure status obs not report error bit
379 				 * PHY_46/174/302/430
380 				 *     phy_rdlvl_status_obs_X:16:8
381 				 */
382 				if ((((tmp >> 8) & 0x1) == 0x1) &&
383 				    (((tmp >> 13) & 0x1) == 0x1) &&
384 				    (((tmp >> 2) & 0x1) == 0x0))
385 					break;
386 				else if (((tmp >> 2) & 0x1) == 0x1)
387 					return -1;
388 			}
389 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
390 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
391 		}
392 		mmio_clrbits_32(PI_REG(ch, 80), 0x3 << 16);
393 	}
394 
395 	/* wdq leveling(LPDDR4 support) */
396 	if ((training_flag & PI_WDQ_LEVELING) == PI_WDQ_LEVELING) {
397 		for (i = 0; i < 4; i++) {
398 			if (!(rank_mask & (1 << i)))
399 				continue;
400 
401 			select_per_cs_training_index(ch, i);
402 			/*
403 			 * disable PI_WDQLVL_VREF_EN before wdq leveling?
404 			 * PI_181 PI_WDQLVL_VREF_EN:RW:8:1
405 			 */
406 			mmio_clrbits_32(PI_REG(ch, 181), 0x1 << 8);
407 			/* PI_124 PI_WDQLVL_EN:RW:16:2 */
408 			mmio_clrsetbits_32(PI_REG(ch, 124), 0x3 << 16,
409 					   0x2 << 16);
410 			/* PI_121 PI_WDQLVL_REQ:WR:8:1,PI_WDQLVL_CS:RW:16:2 */
411 			mmio_clrsetbits_32(PI_REG(ch, 121),
412 					   (0x1 << 8) | (0x3 << 16),
413 					   (0x1 << 8) | (i << 16));
414 			while (1) {
415 				/* PI_174 PI_INT_STATUS:RD:8:18 */
416 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
417 				if ((((tmp >> 12) & 0x1) == 0x1) &&
418 				    (((tmp >> 13) & 0x1) == 0x1) &&
419 				    (((tmp >> 6) & 0x1) == 0x0))
420 					break;
421 				else if (((tmp >> 6) & 0x1) == 0x1)
422 					return -1;
423 			}
424 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
425 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
426 		}
427 		mmio_clrbits_32(PI_REG(ch, 124), 0x3 << 16);
428 	}
429 
430 	/* PHY_927 PHY_PAD_DQS_DRIVE  RPULL offset_22 */
431 	mmio_clrbits_32(PHY_REG(ch, 927), (1 << 22));
432 
433 	return 0;
434 }
435 
436 static __sramfunc void set_ddrconfig(struct rk3399_sdram_params *sdram_params,
437 		unsigned char channel, uint32_t ddrconfig)
438 {
439 	/* only need to set ddrconfig */
440 	struct rk3399_sdram_channel *ch = &sdram_params->ch[channel];
441 	unsigned int cs0_cap = 0;
442 	unsigned int cs1_cap = 0;
443 
444 	cs0_cap = (1 << (ch->cs0_row + ch->col + ch->bk + ch->bw - 20));
445 	if (ch->rank > 1)
446 		cs1_cap = cs0_cap >> (ch->cs0_row - ch->cs1_row);
447 	if (ch->row_3_4) {
448 		cs0_cap = cs0_cap * 3 / 4;
449 		cs1_cap = cs1_cap * 3 / 4;
450 	}
451 
452 	mmio_write_32(MSCH_BASE(channel) + MSCH_DEVICECONF,
453 		      ddrconfig | (ddrconfig << 6));
454 	mmio_write_32(MSCH_BASE(channel) + MSCH_DEVICESIZE,
455 		      ((cs0_cap / 32) & 0xff) | (((cs1_cap / 32) & 0xff) << 8));
456 }
457 
458 static __sramfunc void dram_all_config(struct rk3399_sdram_params *sdram_params)
459 {
460 	unsigned int i;
461 
462 	for (i = 0; i < 2; i++) {
463 		struct rk3399_sdram_channel *info = &sdram_params->ch[i];
464 		struct rk3399_msch_timings *noc = &info->noc_timings;
465 
466 		if (sdram_params->ch[i].col == 0)
467 			continue;
468 
469 		mmio_write_32(MSCH_BASE(i) + MSCH_DDRTIMINGA0,
470 			      noc->ddrtiminga0.d32);
471 		mmio_write_32(MSCH_BASE(i) + MSCH_DDRTIMINGB0,
472 			      noc->ddrtimingb0.d32);
473 		mmio_write_32(MSCH_BASE(i) + MSCH_DDRTIMINGC0,
474 			      noc->ddrtimingc0.d32);
475 		mmio_write_32(MSCH_BASE(i) + MSCH_DEVTODEV0,
476 			      noc->devtodev0.d32);
477 		mmio_write_32(MSCH_BASE(i) + MSCH_DDRMODE, noc->ddrmode.d32);
478 
479 		/* rank 1 memory clock disable (dfi_dram_clk_disable = 1) */
480 		if (sdram_params->ch[i].rank == 1)
481 			mmio_setbits_32(CTL_REG(i, 276), 1 << 17);
482 	}
483 
484 	DDR_STRIDE(sdram_params->stride);
485 
486 	/* reboot hold register set */
487 	mmio_write_32(PMUCRU_BASE + CRU_PMU_RSTHOLD_CON(1),
488 		      CRU_PMU_SGRF_RST_RLS |
489 		      PRESET_GPIO0_HOLD(1) |
490 		      PRESET_GPIO1_HOLD(1));
491 	mmio_clrsetbits_32(CRU_BASE + CRU_GLB_RST_CON, 0x3, 0x3);
492 }
493 
494 static __sramfunc void pctl_cfg(uint32_t ch,
495 		struct rk3399_sdram_params *sdram_params)
496 {
497 	const uint32_t *params_ctl = sdram_params->pctl_regs.denali_ctl;
498 	const uint32_t *params_phy = sdram_params->phy_regs.denali_phy;
499 	const uint32_t *params_pi = sdram_params->pi_regs.denali_pi;
500 	uint32_t tmp, tmp1, tmp2;
501 
502 	/*
503 	 * Workaround controller bug:
504 	 * Do not program DRAM_CLASS until NO_PHY_IND_TRAIN_INT is programmed
505 	 */
506 	sram_regcpy(CTL_REG(ch, 1), (uintptr_t)&params_ctl[1],
507 		    CTL_REG_NUM - 1);
508 	mmio_write_32(CTL_REG(ch, 0), params_ctl[0]);
509 	sram_regcpy(PI_REG(ch, 0), (uintptr_t)&params_pi[0],
510 		    PI_REG_NUM);
511 
512 	mmio_write_32(PHY_REG(ch, 910), params_phy[910]);
513 	mmio_write_32(PHY_REG(ch, 911), params_phy[911]);
514 	mmio_write_32(PHY_REG(ch, 912), params_phy[912]);
515 
516 	mmio_clrsetbits_32(CTL_REG(ch, 68), PWRUP_SREFRESH_EXIT,
517 				PWRUP_SREFRESH_EXIT);
518 
519 	/* PHY_DLL_RST_EN */
520 	mmio_clrsetbits_32(PHY_REG(ch, 957), 0x3 << 24, 1 << 24);
521 	dmbst();
522 
523 	mmio_setbits_32(PI_REG(ch, 0), START);
524 	mmio_setbits_32(CTL_REG(ch, 0), START);
525 
526 	/* wait lock */
527 	while (1) {
528 		tmp = mmio_read_32(PHY_REG(ch, 920));
529 		tmp1 = mmio_read_32(PHY_REG(ch, 921));
530 		tmp2 = mmio_read_32(PHY_REG(ch, 922));
531 		if ((((tmp >> 16) & 0x1) == 0x1) &&
532 		     (((tmp1 >> 16) & 0x1) == 0x1) &&
533 		     (((tmp1 >> 0) & 0x1) == 0x1) &&
534 		     (((tmp2 >> 0) & 0x1) == 0x1))
535 			break;
536 		/* if PLL bypass,don't need wait lock */
537 		if (mmio_read_32(PHY_REG(ch, 911)) & 0x1)
538 			break;
539 	}
540 
541 	sram_regcpy(PHY_REG(ch, 896), (uintptr_t)&params_phy[896], 63);
542 	sram_regcpy(PHY_REG(ch, 0), (uintptr_t)&params_phy[0], 91);
543 	sram_regcpy(PHY_REG(ch, 128), (uintptr_t)&params_phy[128], 91);
544 	sram_regcpy(PHY_REG(ch, 256), (uintptr_t)&params_phy[256], 91);
545 	sram_regcpy(PHY_REG(ch, 384), (uintptr_t)&params_phy[384], 91);
546 	sram_regcpy(PHY_REG(ch, 512), (uintptr_t)&params_phy[512], 38);
547 	sram_regcpy(PHY_REG(ch, 640), (uintptr_t)&params_phy[640], 38);
548 	sram_regcpy(PHY_REG(ch, 768), (uintptr_t)&params_phy[768], 38);
549 }
550 
551 static __sramfunc int dram_switch_to_next_index(
552 		struct rk3399_sdram_params *sdram_params)
553 {
554 	uint32_t ch, ch_count;
555 	uint32_t fn = ((mmio_read_32(CTL_REG(0, 111)) >> 16) + 1) & 0x1;
556 
557 	mmio_write_32(CIC_BASE + CIC_CTRL0,
558 		      (((0x3 << 4) | (1 << 2) | 1) << 16) |
559 		      (fn << 4) | (1 << 2) | 1);
560 	while (!(mmio_read_32(CIC_BASE + CIC_STATUS0) & (1 << 2)))
561 		;
562 
563 	mmio_write_32(CIC_BASE + CIC_CTRL0, 0x20002);
564 	while (!(mmio_read_32(CIC_BASE + CIC_STATUS0) & (1 << 0)))
565 		;
566 
567 	ch_count = sdram_params->num_channels;
568 
569 	/* LPDDR4 f2 cann't do training, all training will fail */
570 	for (ch = 0; ch < ch_count; ch++) {
571 		mmio_clrsetbits_32(PHY_REG(ch, 896), (0x3 << 8) | 1,
572 				   fn << 8);
573 
574 		/* data_training failed */
575 		if (data_training(ch, sdram_params, PI_FULL_TRAINING))
576 			return -1;
577 	}
578 
579 	return 0;
580 }
581 
582 /*
583  * Needs to be done for both channels at once in case of a shared reset signal
584  * between channels.
585  */
586 static __sramfunc int pctl_start(uint32_t channel_mask,
587 		struct rk3399_sdram_params *sdram_params)
588 {
589 	uint32_t count;
590 	uint32_t byte;
591 
592 	mmio_setbits_32(CTL_REG(0, 68), PWRUP_SREFRESH_EXIT);
593 	mmio_setbits_32(CTL_REG(1, 68), PWRUP_SREFRESH_EXIT);
594 
595 	/* need de-access IO retention before controller START */
596 	if (channel_mask & (1 << 0))
597 		mmio_setbits_32(PMU_BASE + PMU_PWRMODE_CON, (1 << 19));
598 	if (channel_mask & (1 << 1))
599 		mmio_setbits_32(PMU_BASE + PMU_PWRMODE_CON, (1 << 23));
600 
601 	/* PHY_DLL_RST_EN */
602 	if (channel_mask & (1 << 0))
603 		mmio_clrsetbits_32(PHY_REG(0, 957), 0x3 << 24,
604 				   0x2 << 24);
605 	if (channel_mask & (1 << 1))
606 		mmio_clrsetbits_32(PHY_REG(1, 957), 0x3 << 24,
607 				   0x2 << 24);
608 
609 	/* check ERROR bit */
610 	if (channel_mask & (1 << 0)) {
611 		count = 0;
612 		while (!(mmio_read_32(CTL_REG(0, 203)) & (1 << 3))) {
613 			/* CKE is low, loop 10ms */
614 			if (count > 100)
615 				return -1;
616 
617 			sram_udelay(100);
618 			count++;
619 		}
620 
621 		mmio_clrbits_32(CTL_REG(0, 68), PWRUP_SREFRESH_EXIT);
622 
623 		/* Restore the PHY_RX_CAL_DQS value */
624 		for (byte = 0; byte < 4; byte++)
625 			mmio_clrsetbits_32(PHY_REG(0, 57 + 128 * byte),
626 					   0xfff << 16,
627 					   sdram_params->rx_cal_dqs[0][byte]);
628 	}
629 	if (channel_mask & (1 << 1)) {
630 		count = 0;
631 		while (!(mmio_read_32(CTL_REG(1, 203)) & (1 << 3))) {
632 			/* CKE is low, loop 10ms */
633 			if (count > 100)
634 				return -1;
635 
636 			sram_udelay(100);
637 			count++;
638 		}
639 
640 		mmio_clrbits_32(CTL_REG(1, 68), PWRUP_SREFRESH_EXIT);
641 
642 		/* Restore the PHY_RX_CAL_DQS value */
643 		for (byte = 0; byte < 4; byte++)
644 			mmio_clrsetbits_32(PHY_REG(1, 57 + 128 * byte),
645 					   0xfff << 16,
646 					   sdram_params->rx_cal_dqs[1][byte]);
647 	}
648 
649 	return 0;
650 }
651 
652 void dmc_save(void)
653 {
654 	struct rk3399_sdram_params *sdram_params = &sdram_config;
655 	uint32_t *params_ctl;
656 	uint32_t *params_pi;
657 	uint32_t *params_phy;
658 	uint32_t refdiv, postdiv2, postdiv1, fbdiv;
659 	uint32_t tmp, ch, byte;
660 
661 	params_ctl = sdram_params->pctl_regs.denali_ctl;
662 	params_pi = sdram_params->pi_regs.denali_pi;
663 	params_phy = sdram_params->phy_regs.denali_phy;
664 
665 	fbdiv = mmio_read_32(CRU_BASE + CRU_PLL_CON(DPLL_ID, 0)) & 0xfff;
666 	tmp = mmio_read_32(CRU_BASE + CRU_PLL_CON(DPLL_ID, 1));
667 	postdiv2 = POSTDIV2_DEC(tmp);
668 	postdiv1 = POSTDIV1_DEC(tmp);
669 	refdiv = REFDIV_DEC(tmp);
670 
671 	sdram_params->ddr_freq = ((fbdiv * 24) /
672 				(refdiv * postdiv1 * postdiv2)) * MHz;
673 
674 	INFO("sdram_params->ddr_freq = %d\n", sdram_params->ddr_freq);
675 	sdram_params->odt = (((mmio_read_32(PHY_REG(0, 5)) >> 16) &
676 			       0x7) != 0) ? 1 : 0;
677 
678 	/* copy the registers CTL PI and PHY */
679 	sram_regcpy((uintptr_t)&params_ctl[0], CTL_REG(0, 0), CTL_REG_NUM);
680 
681 	/* mask DENALI_CTL_00_DATA.START, only copy here, will trigger later */
682 	params_ctl[0] &= ~(0x1 << 0);
683 
684 	sram_regcpy((uintptr_t)&params_pi[0], PI_REG(0, 0),
685 		    PI_REG_NUM);
686 
687 	/* mask DENALI_PI_00_DATA.START, only copy here, will trigger later*/
688 	params_pi[0] &= ~(0x1 << 0);
689 
690 	sram_regcpy((uintptr_t)&params_phy[0], PHY_REG(0, 0), 91);
691 	sram_regcpy((uintptr_t)&params_phy[128], PHY_REG(0, 128), 91);
692 	sram_regcpy((uintptr_t)&params_phy[256], PHY_REG(0, 256), 91);
693 	sram_regcpy((uintptr_t)&params_phy[384], PHY_REG(0, 384), 91);
694 	sram_regcpy((uintptr_t)&params_phy[512], PHY_REG(0, 512), 38);
695 	sram_regcpy((uintptr_t)&params_phy[640], PHY_REG(0, 640), 38);
696 	sram_regcpy((uintptr_t)&params_phy[768], PHY_REG(0, 768), 38);
697 	sram_regcpy((uintptr_t)&params_phy[896], PHY_REG(0, 896), 63);
698 
699 	for (ch = 0; ch < sdram_params->num_channels; ch++) {
700 		for (byte = 0; byte < 4; byte++)
701 			sdram_params->rx_cal_dqs[ch][byte] = (0xfff << 16) &
702 				mmio_read_32(PHY_REG(ch, 57 + byte * 128));
703 	}
704 
705 	/* set DENALI_PHY_957_DATA.PHY_DLL_RST_EN = 0x1 */
706 	params_phy[957] &= ~(0x3 << 24);
707 	params_phy[957] |= 1 << 24;
708 	params_phy[896] |= 1;
709 	params_phy[896] &= ~(0x3 << 8);
710 }
711 
712 __sramfunc void dmc_restore(void)
713 {
714 	struct rk3399_sdram_params *sdram_params = &sdram_config;
715 	uint32_t channel_mask = 0;
716 	uint32_t channel;
717 
718 	configure_sgrf();
719 
720 retry:
721 	for (channel = 0; channel < sdram_params->num_channels; channel++) {
722 		phy_pctrl_reset(channel);
723 		phy_dll_bypass_set(channel, sdram_params->ddr_freq);
724 		if (channel >= sdram_params->num_channels)
725 			continue;
726 
727 		pctl_cfg(channel, sdram_params);
728 	}
729 
730 	for (channel = 0; channel < 2; channel++) {
731 		if (sdram_params->ch[channel].col)
732 			channel_mask |= 1 << channel;
733 	}
734 
735 	if (pctl_start(channel_mask, sdram_params) < 0)
736 		goto retry;
737 
738 	for (channel = 0; channel < sdram_params->num_channels; channel++) {
739 		/* LPDDR2/LPDDR3 need to wait DAI complete, max 10us */
740 		if (sdram_params->dramtype == LPDDR3)
741 			sram_udelay(10);
742 
743 		/* If traning fail, retry to do it again. */
744 		if (data_training(channel, sdram_params, PI_FULL_TRAINING))
745 			goto retry;
746 
747 		set_ddrconfig(sdram_params, channel,
748 			      sdram_params->ch[channel].ddrconfig);
749 	}
750 
751 	dram_all_config(sdram_params);
752 
753 	/* Switch to index 1 and prepare for DDR frequency switch. */
754 	dram_switch_to_next_index(sdram_params);
755 }
756