xref: /rk3399_ARM-atf/plat/rockchip/rk3399/drivers/dram/suspend.c (revision c82eef6ce57a5d1f8408778f191928e958c39878)
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 set_cs_training_index(uint32_t ch, uint32_t rank)
123 {
124 	uint32_t byte;
125 
126 	/* PHY_8/136/264/392 phy_per_cs_training_index_X 1bit offset_24 */
127 	for (byte = 0; byte < 4; byte++)
128 		mmio_clrsetbits_32(PHY_REG(ch, 8 + (128 * byte)), 0x1 << 24,
129 				   rank << 24);
130 }
131 
132 static __sramfunc void select_per_cs_training_index(uint32_t ch, uint32_t rank)
133 {
134 	/* PHY_84 PHY_PER_CS_TRAINING_EN_0 1bit offset_16 */
135 	if ((mmio_read_32(PHY_REG(ch, 84)) >> 16) & 1)
136 		set_cs_training_index(ch, rank);
137 }
138 
139 static void override_write_leveling_value(uint32_t ch)
140 {
141 	uint32_t byte;
142 
143 	for (byte = 0; byte < 4; byte++) {
144 		/*
145 		 * PHY_8/136/264/392
146 		 * phy_per_cs_training_multicast_en_X 1bit offset_16
147 		 */
148 		mmio_clrsetbits_32(PHY_REG(ch, 8 + (128 * byte)), 0x1 << 16,
149 				   1 << 16);
150 		mmio_clrsetbits_32(PHY_REG(ch, 63 + (128 * byte)),
151 				   0xffff << 16,
152 				   0x200 << 16);
153 	}
154 
155 	/* CTL_200 ctrlupd_req 1bit offset_8 */
156 	mmio_clrsetbits_32(CTL_REG(ch, 200), 0x1 << 8, 0x1 << 8);
157 }
158 
159 static __sramfunc int data_training(uint32_t ch,
160 		struct rk3399_sdram_params *sdram_params,
161 		uint32_t training_flag)
162 {
163 	uint32_t obs_0, obs_1, obs_2, obs_3, obs_err = 0;
164 	uint32_t rank = sdram_params->ch[ch].rank;
165 	uint32_t rank_mask;
166 	uint32_t i, tmp;
167 
168 	if (sdram_params->dramtype == LPDDR4)
169 		rank_mask = (rank == 1) ? 0x5 : 0xf;
170 	else
171 		rank_mask = (rank == 1) ? 0x1 : 0x3;
172 
173 	/* PHY_927 PHY_PAD_DQS_DRIVE  RPULL offset_22 */
174 	mmio_setbits_32(PHY_REG(ch, 927), (1 << 22));
175 
176 	if (training_flag == PI_FULL_TRAINING) {
177 		if (sdram_params->dramtype == LPDDR4) {
178 			training_flag = PI_WRITE_LEVELING |
179 					PI_READ_GATE_TRAINING |
180 					PI_READ_LEVELING |
181 					PI_WDQ_LEVELING;
182 		} else if (sdram_params->dramtype == LPDDR3) {
183 			training_flag = PI_CA_TRAINING | PI_WRITE_LEVELING |
184 					PI_READ_GATE_TRAINING;
185 		} else if (sdram_params->dramtype == DDR3) {
186 			training_flag = PI_WRITE_LEVELING |
187 					PI_READ_GATE_TRAINING |
188 					PI_READ_LEVELING;
189 		}
190 	}
191 
192 	/* ca training(LPDDR4,LPDDR3 support) */
193 	if ((training_flag & PI_CA_TRAINING) == PI_CA_TRAINING) {
194 		for (i = 0; i < 4; i++) {
195 			if (!(rank_mask & (1 << i)))
196 				continue;
197 
198 			select_per_cs_training_index(ch, i);
199 			/* PI_100 PI_CALVL_EN:RW:8:2 */
200 			mmio_clrsetbits_32(PI_REG(ch, 100), 0x3 << 8, 0x2 << 8);
201 
202 			/* PI_92 PI_CALVL_REQ:WR:16:1,PI_CALVL_CS:RW:24:2 */
203 			mmio_clrsetbits_32(PI_REG(ch, 92),
204 					   (0x1 << 16) | (0x3 << 24),
205 					   (0x1 << 16) | (i << 24));
206 			while (1) {
207 				/* PI_174 PI_INT_STATUS:RD:8:18 */
208 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
209 
210 				/*
211 				 * check status obs
212 				 * PHY_532/660/788 phy_adr_calvl_obs1_:0:32
213 				 */
214 				obs_0 = mmio_read_32(PHY_REG(ch, 532));
215 				obs_1 = mmio_read_32(PHY_REG(ch, 660));
216 				obs_2 = mmio_read_32(PHY_REG(ch, 788));
217 				if (((obs_0 >> 30) & 0x3) ||
218 				    ((obs_1 >> 30) & 0x3) ||
219 				    ((obs_2 >> 30) & 0x3))
220 					obs_err = 1;
221 				if ((((tmp >> 11) & 0x1) == 0x1) &&
222 				    (((tmp >> 13) & 0x1) == 0x1) &&
223 				    (((tmp >> 5) & 0x1) == 0x0) &&
224 				    (obs_err == 0))
225 					break;
226 				else if ((((tmp >> 5) & 0x1) == 0x1) ||
227 					 (obs_err == 1))
228 					return -1;
229 			}
230 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
231 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
232 		}
233 		mmio_clrbits_32(PI_REG(ch, 100), 0x3 << 8);
234 	}
235 
236 	/* write leveling(LPDDR4,LPDDR3,DDR3 support) */
237 	if ((training_flag & PI_WRITE_LEVELING) == PI_WRITE_LEVELING) {
238 		for (i = 0; i < rank; i++) {
239 			select_per_cs_training_index(ch, i);
240 			/* PI_60 PI_WRLVL_EN:RW:8:2 */
241 			mmio_clrsetbits_32(PI_REG(ch, 60), 0x3 << 8, 0x2 << 8);
242 			/* PI_59 PI_WRLVL_REQ:WR:8:1,PI_WRLVL_CS:RW:16:2 */
243 			mmio_clrsetbits_32(PI_REG(ch, 59),
244 					   (0x1 << 8) | (0x3 << 16),
245 					   (0x1 << 8) | (i << 16));
246 
247 			while (1) {
248 				/* PI_174 PI_INT_STATUS:RD:8:18 */
249 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
250 
251 				/*
252 				 * check status obs, if error maybe can not
253 				 * get leveling done PHY_40/168/296/424
254 				 * phy_wrlvl_status_obs_X:0:13
255 				 */
256 				obs_0 = mmio_read_32(PHY_REG(ch, 40));
257 				obs_1 = mmio_read_32(PHY_REG(ch, 168));
258 				obs_2 = mmio_read_32(PHY_REG(ch, 296));
259 				obs_3 = mmio_read_32(PHY_REG(ch, 424));
260 				if (((obs_0 >> 12) & 0x1) ||
261 				    ((obs_1 >> 12) & 0x1) ||
262 				    ((obs_2 >> 12) & 0x1) ||
263 				    ((obs_3 >> 12) & 0x1))
264 					obs_err = 1;
265 				if ((((tmp >> 10) & 0x1) == 0x1) &&
266 				    (((tmp >> 13) & 0x1) == 0x1) &&
267 				    (((tmp >> 4) & 0x1) == 0x0) &&
268 				    (obs_err == 0))
269 					break;
270 				else if ((((tmp >> 4) & 0x1) == 0x1) ||
271 					 (obs_err == 1))
272 					return -1;
273 			}
274 
275 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
276 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
277 		}
278 		override_write_leveling_value(ch);
279 		mmio_clrbits_32(PI_REG(ch, 60), 0x3 << 8);
280 	}
281 
282 	/* read gate training(LPDDR4,LPDDR3,DDR3 support) */
283 	if ((training_flag & PI_READ_GATE_TRAINING) == PI_READ_GATE_TRAINING) {
284 		for (i = 0; i < rank; i++) {
285 			select_per_cs_training_index(ch, i);
286 			/* PI_80 PI_RDLVL_GATE_EN:RW:24:2 */
287 			mmio_clrsetbits_32(PI_REG(ch, 80), 0x3 << 24,
288 					   0x2 << 24);
289 			/*
290 			 * PI_74 PI_RDLVL_GATE_REQ:WR:16:1
291 			 * PI_RDLVL_CS:RW:24:2
292 			 */
293 			mmio_clrsetbits_32(PI_REG(ch, 74),
294 					   (0x1 << 16) | (0x3 << 24),
295 					   (0x1 << 16) | (i << 24));
296 
297 			while (1) {
298 				/* PI_174 PI_INT_STATUS:RD:8:18 */
299 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
300 
301 				/*
302 				 * check status obs
303 				 * PHY_43/171/299/427
304 				 *     PHY_GTLVL_STATUS_OBS_x:16:8
305 				 */
306 				obs_0 = mmio_read_32(PHY_REG(ch, 43));
307 				obs_1 = mmio_read_32(PHY_REG(ch, 171));
308 				obs_2 = mmio_read_32(PHY_REG(ch, 299));
309 				obs_3 = mmio_read_32(PHY_REG(ch, 427));
310 				if (((obs_0 >> (16 + 6)) & 0x3) ||
311 				    ((obs_1 >> (16 + 6)) & 0x3) ||
312 				    ((obs_2 >> (16 + 6)) & 0x3) ||
313 				    ((obs_3 >> (16 + 6)) & 0x3))
314 					obs_err = 1;
315 				if ((((tmp >> 9) & 0x1) == 0x1) &&
316 				    (((tmp >> 13) & 0x1) == 0x1) &&
317 				    (((tmp >> 3) & 0x1) == 0x0) &&
318 				    (obs_err == 0))
319 					break;
320 				else if ((((tmp >> 3) & 0x1) == 0x1) ||
321 					 (obs_err == 1))
322 					return -1;
323 			}
324 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
325 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
326 		}
327 		mmio_clrbits_32(PI_REG(ch, 80), 0x3 << 24);
328 	}
329 
330 	/* read leveling(LPDDR4,LPDDR3,DDR3 support) */
331 	if ((training_flag & PI_READ_LEVELING) == PI_READ_LEVELING) {
332 		for (i = 0; i < rank; i++) {
333 			select_per_cs_training_index(ch, i);
334 			/* PI_80 PI_RDLVL_EN:RW:16:2 */
335 			mmio_clrsetbits_32(PI_REG(ch, 80), 0x3 << 16,
336 					   0x2 << 16);
337 			/* PI_74 PI_RDLVL_REQ:WR:8:1,PI_RDLVL_CS:RW:24:2 */
338 			mmio_clrsetbits_32(PI_REG(ch, 74),
339 					   (0x1 << 8) | (0x3 << 24),
340 					   (0x1 << 8) | (i << 24));
341 			while (1) {
342 				/* PI_174 PI_INT_STATUS:RD:8:18 */
343 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
344 
345 				/*
346 				 * make sure status obs not report error bit
347 				 * PHY_46/174/302/430
348 				 *     phy_rdlvl_status_obs_X:16:8
349 				 */
350 				if ((((tmp >> 8) & 0x1) == 0x1) &&
351 				    (((tmp >> 13) & 0x1) == 0x1) &&
352 				    (((tmp >> 2) & 0x1) == 0x0))
353 					break;
354 				else if (((tmp >> 2) & 0x1) == 0x1)
355 					return -1;
356 			}
357 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
358 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
359 		}
360 		mmio_clrbits_32(PI_REG(ch, 80), 0x3 << 16);
361 	}
362 
363 	/* wdq leveling(LPDDR4 support) */
364 	if ((training_flag & PI_WDQ_LEVELING) == PI_WDQ_LEVELING) {
365 		for (i = 0; i < 4; i++) {
366 			if (!(rank_mask & (1 << i)))
367 				continue;
368 
369 			select_per_cs_training_index(ch, i);
370 			/*
371 			 * disable PI_WDQLVL_VREF_EN before wdq leveling?
372 			 * PI_181 PI_WDQLVL_VREF_EN:RW:8:1
373 			 */
374 			mmio_clrbits_32(PI_REG(ch, 181), 0x1 << 8);
375 			/* PI_124 PI_WDQLVL_EN:RW:16:2 */
376 			mmio_clrsetbits_32(PI_REG(ch, 124), 0x3 << 16,
377 					   0x2 << 16);
378 			/* PI_121 PI_WDQLVL_REQ:WR:8:1,PI_WDQLVL_CS:RW:16:2 */
379 			mmio_clrsetbits_32(PI_REG(ch, 121),
380 					   (0x1 << 8) | (0x3 << 16),
381 					   (0x1 << 8) | (i << 16));
382 			while (1) {
383 				/* PI_174 PI_INT_STATUS:RD:8:18 */
384 				tmp = mmio_read_32(PI_REG(ch, 174)) >> 8;
385 				if ((((tmp >> 12) & 0x1) == 0x1) &&
386 				    (((tmp >> 13) & 0x1) == 0x1) &&
387 				    (((tmp >> 6) & 0x1) == 0x0))
388 					break;
389 				else if (((tmp >> 6) & 0x1) == 0x1)
390 					return -1;
391 			}
392 			/* clear interrupt,PI_175 PI_INT_ACK:WR:0:17 */
393 			mmio_write_32(PI_REG(ch, 175), 0x00003f7c);
394 		}
395 		mmio_clrbits_32(PI_REG(ch, 124), 0x3 << 16);
396 	}
397 
398 	/* PHY_927 PHY_PAD_DQS_DRIVE  RPULL offset_22 */
399 	mmio_clrbits_32(PHY_REG(ch, 927), (1 << 22));
400 
401 	return 0;
402 }
403 
404 static __sramfunc void set_ddrconfig(struct rk3399_sdram_params *sdram_params,
405 		unsigned char channel, uint32_t ddrconfig)
406 {
407 	/* only need to set ddrconfig */
408 	struct rk3399_sdram_channel *ch = &sdram_params->ch[channel];
409 	unsigned int cs0_cap = 0;
410 	unsigned int cs1_cap = 0;
411 
412 	cs0_cap = (1 << (ch->cs0_row + ch->col + ch->bk + ch->bw - 20));
413 	if (ch->rank > 1)
414 		cs1_cap = cs0_cap >> (ch->cs0_row - ch->cs1_row);
415 	if (ch->row_3_4) {
416 		cs0_cap = cs0_cap * 3 / 4;
417 		cs1_cap = cs1_cap * 3 / 4;
418 	}
419 
420 	mmio_write_32(MSCH_BASE(channel) + MSCH_DEVICECONF,
421 		      ddrconfig | (ddrconfig << 6));
422 	mmio_write_32(MSCH_BASE(channel) + MSCH_DEVICESIZE,
423 		      ((cs0_cap / 32) & 0xff) | (((cs1_cap / 32) & 0xff) << 8));
424 }
425 
426 static __sramfunc void dram_all_config(struct rk3399_sdram_params *sdram_params)
427 {
428 	unsigned int i;
429 
430 	for (i = 0; i < 2; i++) {
431 		struct rk3399_sdram_channel *info = &sdram_params->ch[i];
432 		struct rk3399_msch_timings *noc = &info->noc_timings;
433 
434 		if (sdram_params->ch[i].col == 0)
435 			continue;
436 
437 		mmio_write_32(MSCH_BASE(i) + MSCH_DDRTIMINGA0,
438 			      noc->ddrtiminga0.d32);
439 		mmio_write_32(MSCH_BASE(i) + MSCH_DDRTIMINGB0,
440 			      noc->ddrtimingb0.d32);
441 		mmio_write_32(MSCH_BASE(i) + MSCH_DDRTIMINGC0,
442 			      noc->ddrtimingc0.d32);
443 		mmio_write_32(MSCH_BASE(i) + MSCH_DEVTODEV0,
444 			      noc->devtodev0.d32);
445 		mmio_write_32(MSCH_BASE(i) + MSCH_DDRMODE, noc->ddrmode.d32);
446 
447 		/* rank 1 memory clock disable (dfi_dram_clk_disable = 1) */
448 		if (sdram_params->ch[i].rank == 1)
449 			mmio_setbits_32(CTL_REG(i, 276), 1 << 17);
450 	}
451 
452 	DDR_STRIDE(sdram_params->stride);
453 
454 	/* reboot hold register set */
455 	mmio_write_32(PMUCRU_BASE + CRU_PMU_RSTHOLD_CON(1),
456 		      CRU_PMU_SGRF_RST_RLS |
457 		      PRESET_GPIO0_HOLD(1) |
458 		      PRESET_GPIO1_HOLD(1));
459 	mmio_clrsetbits_32(CRU_BASE + CRU_GLB_RST_CON, 0x3, 0x3);
460 }
461 
462 static __sramfunc void pctl_cfg(uint32_t ch,
463 		struct rk3399_sdram_params *sdram_params)
464 {
465 	const uint32_t *params_ctl = sdram_params->pctl_regs.denali_ctl;
466 	const uint32_t *params_pi = sdram_params->pi_regs.denali_pi;
467 	const struct rk3399_ddr_publ_regs *phy_regs = &sdram_params->phy_regs;
468 	uint32_t tmp, tmp1, tmp2, i;
469 
470 	/*
471 	 * Workaround controller bug:
472 	 * Do not program DRAM_CLASS until NO_PHY_IND_TRAIN_INT is programmed
473 	 */
474 	sram_regcpy(CTL_REG(ch, 1), (uintptr_t)&params_ctl[1],
475 		    CTL_REG_NUM - 1);
476 	mmio_write_32(CTL_REG(ch, 0), params_ctl[0]);
477 	sram_regcpy(PI_REG(ch, 0), (uintptr_t)&params_pi[0],
478 		    PI_REG_NUM);
479 
480 	sram_regcpy(PHY_REG(ch, 910), (uintptr_t)&phy_regs->phy896[910 - 896],
481 		    3);
482 
483 	mmio_clrsetbits_32(CTL_REG(ch, 68), PWRUP_SREFRESH_EXIT,
484 				PWRUP_SREFRESH_EXIT);
485 
486 	/* PHY_DLL_RST_EN */
487 	mmio_clrsetbits_32(PHY_REG(ch, 957), 0x3 << 24, 1 << 24);
488 	dmbst();
489 
490 	mmio_setbits_32(PI_REG(ch, 0), START);
491 	mmio_setbits_32(CTL_REG(ch, 0), START);
492 
493 	/* wait lock */
494 	while (1) {
495 		tmp = mmio_read_32(PHY_REG(ch, 920));
496 		tmp1 = mmio_read_32(PHY_REG(ch, 921));
497 		tmp2 = mmio_read_32(PHY_REG(ch, 922));
498 		if ((((tmp >> 16) & 0x1) == 0x1) &&
499 		     (((tmp1 >> 16) & 0x1) == 0x1) &&
500 		     (((tmp1 >> 0) & 0x1) == 0x1) &&
501 		     (((tmp2 >> 0) & 0x1) == 0x1))
502 			break;
503 		/* if PLL bypass,don't need wait lock */
504 		if (mmio_read_32(PHY_REG(ch, 911)) & 0x1)
505 			break;
506 	}
507 
508 	sram_regcpy(PHY_REG(ch, 896), (uintptr_t)&phy_regs->phy896[0], 63);
509 
510 	for (i = 0; i < 4; i++)
511 		sram_regcpy(PHY_REG(ch, 128 * i),
512 			    (uintptr_t)&phy_regs->phy0[i][0], 91);
513 
514 	for (i = 0; i < 3; i++)
515 		sram_regcpy(PHY_REG(ch, 512 + 128 * i),
516 				(uintptr_t)&phy_regs->phy512[i][0], 38);
517 }
518 
519 static __sramfunc int dram_switch_to_next_index(
520 		struct rk3399_sdram_params *sdram_params)
521 {
522 	uint32_t ch, ch_count;
523 	uint32_t fn = ((mmio_read_32(CTL_REG(0, 111)) >> 16) + 1) & 0x1;
524 
525 	mmio_write_32(CIC_BASE + CIC_CTRL0,
526 		      (((0x3 << 4) | (1 << 2) | 1) << 16) |
527 		      (fn << 4) | (1 << 2) | 1);
528 	while (!(mmio_read_32(CIC_BASE + CIC_STATUS0) & (1 << 2)))
529 		;
530 
531 	mmio_write_32(CIC_BASE + CIC_CTRL0, 0x20002);
532 	while (!(mmio_read_32(CIC_BASE + CIC_STATUS0) & (1 << 0)))
533 		;
534 
535 	ch_count = sdram_params->num_channels;
536 
537 	/* LPDDR4 f2 cann't do training, all training will fail */
538 	for (ch = 0; ch < ch_count; ch++) {
539 		mmio_clrsetbits_32(PHY_REG(ch, 896), (0x3 << 8) | 1,
540 				   fn << 8);
541 
542 		/* data_training failed */
543 		if (data_training(ch, sdram_params, PI_FULL_TRAINING))
544 			return -1;
545 	}
546 
547 	return 0;
548 }
549 
550 /*
551  * Needs to be done for both channels at once in case of a shared reset signal
552  * between channels.
553  */
554 static __sramfunc int pctl_start(uint32_t channel_mask,
555 		struct rk3399_sdram_params *sdram_params)
556 {
557 	uint32_t count;
558 	uint32_t byte;
559 
560 	mmio_setbits_32(CTL_REG(0, 68), PWRUP_SREFRESH_EXIT);
561 	mmio_setbits_32(CTL_REG(1, 68), PWRUP_SREFRESH_EXIT);
562 
563 	/* need de-access IO retention before controller START */
564 	if (channel_mask & (1 << 0))
565 		mmio_setbits_32(PMU_BASE + PMU_PWRMODE_CON, (1 << 19));
566 	if (channel_mask & (1 << 1))
567 		mmio_setbits_32(PMU_BASE + PMU_PWRMODE_CON, (1 << 23));
568 
569 	/* PHY_DLL_RST_EN */
570 	if (channel_mask & (1 << 0))
571 		mmio_clrsetbits_32(PHY_REG(0, 957), 0x3 << 24,
572 				   0x2 << 24);
573 	if (channel_mask & (1 << 1))
574 		mmio_clrsetbits_32(PHY_REG(1, 957), 0x3 << 24,
575 				   0x2 << 24);
576 
577 	/* check ERROR bit */
578 	if (channel_mask & (1 << 0)) {
579 		count = 0;
580 		while (!(mmio_read_32(CTL_REG(0, 203)) & (1 << 3))) {
581 			/* CKE is low, loop 10ms */
582 			if (count > 100)
583 				return -1;
584 
585 			sram_udelay(100);
586 			count++;
587 		}
588 
589 		mmio_clrbits_32(CTL_REG(0, 68), PWRUP_SREFRESH_EXIT);
590 
591 		/* Restore the PHY_RX_CAL_DQS value */
592 		for (byte = 0; byte < 4; byte++)
593 			mmio_clrsetbits_32(PHY_REG(0, 57 + 128 * byte),
594 					   0xfff << 16,
595 					   sdram_params->rx_cal_dqs[0][byte]);
596 	}
597 	if (channel_mask & (1 << 1)) {
598 		count = 0;
599 		while (!(mmio_read_32(CTL_REG(1, 203)) & (1 << 3))) {
600 			/* CKE is low, loop 10ms */
601 			if (count > 100)
602 				return -1;
603 
604 			sram_udelay(100);
605 			count++;
606 		}
607 
608 		mmio_clrbits_32(CTL_REG(1, 68), PWRUP_SREFRESH_EXIT);
609 
610 		/* Restore the PHY_RX_CAL_DQS value */
611 		for (byte = 0; byte < 4; byte++)
612 			mmio_clrsetbits_32(PHY_REG(1, 57 + 128 * byte),
613 					   0xfff << 16,
614 					   sdram_params->rx_cal_dqs[1][byte]);
615 	}
616 
617 	return 0;
618 }
619 
620 void dmc_save(void)
621 {
622 	struct rk3399_sdram_params *sdram_params = &sdram_config;
623 	struct rk3399_ddr_publ_regs *phy_regs;
624 	uint32_t *params_ctl;
625 	uint32_t *params_pi;
626 	uint32_t refdiv, postdiv2, postdiv1, fbdiv;
627 	uint32_t tmp, ch, byte, i;
628 
629 	phy_regs = &sdram_params->phy_regs;
630 	params_ctl = sdram_params->pctl_regs.denali_ctl;
631 	params_pi = sdram_params->pi_regs.denali_pi;
632 
633 	fbdiv = mmio_read_32(CRU_BASE + CRU_PLL_CON(DPLL_ID, 0)) & 0xfff;
634 	tmp = mmio_read_32(CRU_BASE + CRU_PLL_CON(DPLL_ID, 1));
635 	postdiv2 = POSTDIV2_DEC(tmp);
636 	postdiv1 = POSTDIV1_DEC(tmp);
637 	refdiv = REFDIV_DEC(tmp);
638 
639 	sdram_params->ddr_freq = ((fbdiv * 24) /
640 				(refdiv * postdiv1 * postdiv2)) * MHz;
641 
642 	INFO("sdram_params->ddr_freq = %d\n", sdram_params->ddr_freq);
643 	sdram_params->odt = (((mmio_read_32(PHY_REG(0, 5)) >> 16) &
644 			       0x7) != 0) ? 1 : 0;
645 
646 	/* copy the registers CTL PI and PHY */
647 	sram_regcpy((uintptr_t)&params_ctl[0], CTL_REG(0, 0), CTL_REG_NUM);
648 
649 	/* mask DENALI_CTL_00_DATA.START, only copy here, will trigger later */
650 	params_ctl[0] &= ~(0x1 << 0);
651 
652 	sram_regcpy((uintptr_t)&params_pi[0], PI_REG(0, 0),
653 		    PI_REG_NUM);
654 
655 	/* mask DENALI_PI_00_DATA.START, only copy here, will trigger later*/
656 	params_pi[0] &= ~(0x1 << 0);
657 
658 	for (i = 0; i < 4; i++)
659 		sram_regcpy((uintptr_t)&phy_regs->phy0[i][0],
660 			    PHY_REG(0, 128 * i), 91);
661 
662 	for (i = 0; i < 3; i++)
663 		sram_regcpy((uintptr_t)&phy_regs->phy512[i][0],
664 			    PHY_REG(0, 512 + 128 * i), 38);
665 
666 	sram_regcpy((uintptr_t)&phy_regs->phy896[0], PHY_REG(0, 896), 63);
667 
668 	for (ch = 0; ch < sdram_params->num_channels; ch++) {
669 		for (byte = 0; byte < 4; byte++)
670 			sdram_params->rx_cal_dqs[ch][byte] = (0xfff << 16) &
671 				mmio_read_32(PHY_REG(ch, 57 + byte * 128));
672 	}
673 
674 	/* set DENALI_PHY_957_DATA.PHY_DLL_RST_EN = 0x1 */
675 	phy_regs->phy896[957 - 896] &= ~(0x3 << 24);
676 	phy_regs->phy896[957 - 896] |= 1 << 24;
677 	phy_regs->phy896[0] |= 1;
678 	phy_regs->phy896[0] &= ~(0x3 << 8);
679 }
680 
681 __sramfunc void dmc_restore(void)
682 {
683 	struct rk3399_sdram_params *sdram_params = &sdram_config;
684 	uint32_t channel_mask = 0;
685 	uint32_t channel;
686 
687 	configure_sgrf();
688 
689 retry:
690 	for (channel = 0; channel < sdram_params->num_channels; channel++) {
691 		phy_pctrl_reset(channel);
692 		pctl_cfg(channel, sdram_params);
693 	}
694 
695 	for (channel = 0; channel < 2; channel++) {
696 		if (sdram_params->ch[channel].col)
697 			channel_mask |= 1 << channel;
698 	}
699 
700 	if (pctl_start(channel_mask, sdram_params) < 0)
701 		goto retry;
702 
703 	for (channel = 0; channel < sdram_params->num_channels; channel++) {
704 		/* LPDDR2/LPDDR3 need to wait DAI complete, max 10us */
705 		if (sdram_params->dramtype == LPDDR3)
706 			sram_udelay(10);
707 
708 		/* If traning fail, retry to do it again. */
709 		if (data_training(channel, sdram_params, PI_FULL_TRAINING))
710 			goto retry;
711 
712 		set_ddrconfig(sdram_params, channel,
713 			      sdram_params->ch[channel].ddrconfig);
714 	}
715 
716 	dram_all_config(sdram_params);
717 
718 	/* Switch to index 1 and prepare for DDR frequency switch. */
719 	dram_switch_to_next_index(sdram_params);
720 }
721