1 /*
2  * (C) Copyright 2017 Theobroma Systems Design und Consulting GmbH
3  *
4  * SPDX-License-Identifier:	GPL-2.0
5  */
6 
7 #include <common.h>
8 #include <clk.h>
9 #include <dm.h>
10 #include <dt-bindings/memory/rk3368-dmc.h>
11 #include <dt-structs.h>
12 #include <ram.h>
13 #include <regmap.h>
14 #include <syscon.h>
15 #include <asm/io.h>
16 #include <asm/arch/clock.h>
17 #include <asm/arch/cru_rk3368.h>
18 #include <asm/arch/grf_rk3368.h>
19 #include <asm/arch/ddr_rk3368.h>
20 #include <asm/arch/sdram.h>
21 #include <asm/arch/sdram_common.h>
22 
23 DECLARE_GLOBAL_DATA_PTR;
24 
25 struct dram_info {
26 	struct ram_info info;
27 	struct clk ddr_clk;
28 	struct rk3368_cru *cru;
29 	struct rk3368_grf *grf;
30 	struct rk3368_ddr_pctl *pctl;
31 	struct rk3368_ddrphy *phy;
32 	struct rk3368_pmu_grf *pmugrf;
33 	struct rk3368_msch *msch;
34 };
35 
36 struct rk3368_sdram_params {
37 #if CONFIG_IS_ENABLED(OF_PLATDATA)
38 	struct dtd_rockchip_rk3368_dmc of_plat;
39 #endif
40 	struct rk3288_sdram_pctl_timing pctl_timing;
41 	u32 trefi_mem_ddr3;
42 	struct rk3288_sdram_channel chan;
43 	struct regmap *map;
44 	u32 ddr_freq;
45 	u32 memory_schedule;
46 	u32 ddr_speed_bin;
47 	u32 tfaw_mult;
48 };
49 
50 /* PTCL bits */
51 enum {
52 	/* PCTL_DFISTCFG0 */
53 	DFI_INIT_START = BIT(0),
54 	DFI_DATA_BYTE_DISABLE_EN = BIT(2),
55 
56 	/* PCTL_DFISTCFG1 */
57 	DFI_DRAM_CLK_SR_EN = BIT(0),
58 	DFI_DRAM_CLK_DPD_EN = BIT(1),
59 	ODT_LEN_BL8_W_SHIFT = 16,
60 
61 	/* PCTL_DFISTCFG2 */
62 	DFI_PARITY_INTR_EN = BIT(0),
63 	DFI_PARITY_EN = BIT(1),
64 
65 	/* PCTL_DFILPCFG0 */
66 	TLP_RESP_TIME_SHIFT = 16,
67 	LP_SR_EN = BIT(8),
68 	LP_PD_EN = BIT(0),
69 
70 	/* PCTL_DFIODTCFG */
71 	RANK0_ODT_WRITE_SEL = BIT(3),
72 	RANK1_ODT_WRITE_SEL = BIT(11),
73 
74 	/* PCTL_SCFG */
75 	HW_LOW_POWER_EN = BIT(0),
76 
77 	/* PCTL_MCMD */
78 	START_CMD = BIT(31),
79 	MCMD_RANK0 = BIT(20),
80 	MCMD_RANK1 = BIT(21),
81 	DESELECT_CMD = 0,
82 	PREA_CMD,
83 	REF_CMD,
84 	MRS_CMD,
85 	ZQCS_CMD,
86 	ZQCL_CMD,
87 	RSTL_CMD,
88 	MRR_CMD	= 8,
89 	DPDE_CMD,
90 
91 	/* PCTL_POWCTL */
92 	POWER_UP_START = BIT(0),
93 
94 	/* PCTL_POWSTAT */
95 	POWER_UP_DONE = BIT(0),
96 
97 	/* PCTL_SCTL */
98 	INIT_STATE = 0,
99 	CFG_STATE,
100 	GO_STATE,
101 	SLEEP_STATE,
102 	WAKEUP_STATE,
103 
104 	/* PCTL_STAT */
105 	LP_TRIG_SHIFT = 4,
106 	LP_TRIG_MASK = 7,
107 	PCTL_STAT_MSK = 7,
108 	INIT_MEM = 0,
109 	CONFIG,
110 	CONFIG_REQ,
111 	ACCESS,
112 	ACCESS_REQ,
113 	LOW_POWER,
114 	LOW_POWER_ENTRY_REQ,
115 	LOW_POWER_EXIT_REQ,
116 
117 	/* PCTL_MCFG */
118 	DDR2_DDR3_BL_8 = BIT(0),
119 	DDR3_EN = BIT(5),
120 	TFAW_TRRD_MULT4 = (0 << 18),
121 	TFAW_TRRD_MULT5 = (1 << 18),
122 	TFAW_TRRD_MULT6 = (2 << 18),
123 };
124 
125 #define DDR3_MR0_WR(n) \
126 	((n <= 8) ? ((n - 4) << 9) : (((n >> 1) & 0x7) << 9))
127 #define DDR3_MR0_CL(n) \
128 	((((n - 4) & 0x7) << 4) | (((n - 4) & 0x8) >> 2))
129 #define DDR3_MR0_BL8 \
130 	(0 << 0)
131 #define DDR3_MR0_DLL_RESET \
132 	(1 << 8)
133 #define DDR3_MR1_RTT120OHM \
134 	((0 << 9) | (1 << 6) | (0 << 2))
135 #define DDR3_MR2_TWL(n) \
136 	(((n - 5) & 0x7) << 3)
137 
138 
139 #ifdef CONFIG_TPL_BUILD
140 
141 static void ddr_set_noc_spr_err_stall(struct rk3368_grf *grf, bool enable)
142 {
143 	if (enable)
144 		rk_setreg(&grf->ddrc0_con0, NOC_RSP_ERR_STALL);
145 	else
146 		rk_clrreg(&grf->ddrc0_con0, NOC_RSP_ERR_STALL);
147 }
148 
149 static void ddr_set_ddr3_mode(struct rk3368_grf *grf, bool ddr3_mode)
150 {
151 	if (ddr3_mode)
152 		rk_setreg(&grf->ddrc0_con0, MSCH0_MAINDDR3_DDR3);
153 	else
154 		rk_clrreg(&grf->ddrc0_con0, MSCH0_MAINDDR3_DDR3);
155 }
156 
157 static void ddrphy_config(struct rk3368_ddrphy *phy,
158 			  u32 tcl, u32 tal, u32 tcwl)
159 {
160 	int i;
161 
162 	/* Set to DDR3 mode */
163 	clrsetbits_le32(&phy->reg[1], 0x3, 0x0);
164 
165 	/* DDRPHY_REGB: CL, AL */
166 	clrsetbits_le32(&phy->reg[0xb], 0xff, tcl << 4 | tal);
167 	/* DDRPHY_REGC: CWL */
168 	clrsetbits_le32(&phy->reg[0xc], 0x0f, tcwl);
169 
170 	/* Update drive-strength */
171 	writel(0xcc, &phy->reg[0x11]);
172 	writel(0xaa, &phy->reg[0x16]);
173 	/*
174 	 * Update NRCOMP/PRCOMP for all 4 channels (for details of all
175 	 * affected registers refer to the documentation of DDRPHY_REG20
176 	 * and DDRPHY_REG21 in the RK3368 TRM.
177 	 */
178 	for (i = 0; i < 4; ++i) {
179 		writel(0xcc, &phy->reg[0x20 + i * 0x10]);
180 		writel(0x44, &phy->reg[0x21 + i * 0x10]);
181 	}
182 
183 	/* Enable write-leveling calibration bypass */
184 	setbits_le32(&phy->reg[2], BIT(3));
185 }
186 
187 static void copy_to_reg(u32 *dest, const u32 *src, u32 n)
188 {
189 	int i;
190 
191 	for (i = 0; i < n / sizeof(u32); i++)
192 		writel(*src++, dest++);
193 }
194 
195 static void send_command(struct rk3368_ddr_pctl *pctl, u32 rank, u32 cmd)
196 {
197 	u32 mcmd = START_CMD | cmd | rank;
198 
199 	debug("%s: writing %x to MCMD\n", __func__, mcmd);
200 	writel(mcmd, &pctl->mcmd);
201 	while (readl(&pctl->mcmd) & START_CMD)
202 		/* spin */;
203 }
204 
205 static void send_mrs(struct rk3368_ddr_pctl *pctl,
206 			    u32 rank, u32 mr_num, u32 mr_data)
207 {
208 	u32 mcmd = START_CMD | MRS_CMD | rank | (mr_num << 17) | (mr_data << 4);
209 
210 	debug("%s: writing %x to MCMD\n", __func__, mcmd);
211 	writel(mcmd, &pctl->mcmd);
212 	while (readl(&pctl->mcmd) & START_CMD)
213 		/* spin */;
214 }
215 
216 static int memory_init(struct rk3368_ddr_pctl *pctl,
217 		       struct rk3368_sdram_params *params)
218 {
219 	u32 mr[4];
220 	const ulong timeout_ms = 500;
221 	ulong tmp;
222 
223 	/*
224 	 * Power up DRAM by DDR_PCTL_POWCTL[0] register of PCTL and
225 	 * wait power up DRAM finish with DDR_PCTL_POWSTAT[0] register
226 	 * of PCTL.
227 	 */
228 	writel(POWER_UP_START, &pctl->powctl);
229 
230 	tmp = get_timer(0);
231 	do {
232 		if (get_timer(tmp) > timeout_ms) {
233 			pr_err("%s: POWER_UP_START did not complete in %ld ms\n",
234 			      __func__, timeout_ms);
235 			return -ETIME;
236 		}
237 	} while (!(readl(&pctl->powstat) & POWER_UP_DONE));
238 
239 	/* Configure MR0 through MR3 */
240 	mr[0] = DDR3_MR0_WR(params->pctl_timing.twr) |
241 		DDR3_MR0_CL(params->pctl_timing.tcl) |
242 		DDR3_MR0_DLL_RESET;
243 	mr[1] = DDR3_MR1_RTT120OHM;
244 	mr[2] = DDR3_MR2_TWL(params->pctl_timing.tcwl);
245 	mr[3] = 0;
246 
247 	/*
248 	 * Also see RK3368 Technical Reference Manual:
249 	 *   "16.6.2 Initialization (DDR3 Initialization Sequence)"
250 	 */
251 	send_command(pctl, MCMD_RANK0 | MCMD_RANK1, DESELECT_CMD);
252 	udelay(1);
253 	send_command(pctl, MCMD_RANK0 | MCMD_RANK1, PREA_CMD);
254 	send_mrs(pctl, MCMD_RANK0 | MCMD_RANK1, 2, mr[2]);
255 	send_mrs(pctl, MCMD_RANK0 | MCMD_RANK1, 3, mr[3]);
256 	send_mrs(pctl, MCMD_RANK0 | MCMD_RANK1, 1, mr[1]);
257 	send_mrs(pctl, MCMD_RANK0 | MCMD_RANK1, 0, mr[0]);
258 	send_command(pctl, MCMD_RANK0 | MCMD_RANK1, ZQCL_CMD);
259 
260 	return 0;
261 }
262 
263 static void move_to_config_state(struct rk3368_ddr_pctl *pctl)
264 {
265 	/*
266 	 * Also see RK3368 Technical Reference Manual:
267 	 *   "16.6.1 State transition of PCTL (Moving to Config State)"
268 	 */
269 	u32 state = readl(&pctl->stat) & PCTL_STAT_MSK;
270 
271 	switch (state) {
272 	case LOW_POWER:
273 		writel(WAKEUP_STATE, &pctl->sctl);
274 		while ((readl(&pctl->stat) & PCTL_STAT_MSK) != ACCESS)
275 			/* spin */;
276 
277 		/* fall-through */
278 	case ACCESS:
279 	case INIT_MEM:
280 		writel(CFG_STATE, &pctl->sctl);
281 		while ((readl(&pctl->stat) & PCTL_STAT_MSK) != CONFIG)
282 			/* spin */;
283 		break;
284 
285 	case CONFIG:
286 		return;
287 
288 	default:
289 		break;
290 	}
291 }
292 
293 static void move_to_access_state(struct rk3368_ddr_pctl *pctl)
294 {
295 	/*
296 	 * Also see RK3368 Technical Reference Manual:
297 	 *   "16.6.1 State transition of PCTL (Moving to Access State)"
298 	 */
299 	u32 state = readl(&pctl->stat) & PCTL_STAT_MSK;
300 
301 	switch (state) {
302 	case LOW_POWER:
303 		if (((readl(&pctl->stat) >> LP_TRIG_SHIFT) &
304 		     LP_TRIG_MASK) == 1)
305 			return;
306 
307 		writel(WAKEUP_STATE, &pctl->sctl);
308 		while ((readl(&pctl->stat) & PCTL_STAT_MSK) != ACCESS)
309 			/* spin */;
310 
311 		/* fall-through */
312 	case INIT_MEM:
313 		writel(CFG_STATE, &pctl->sctl);
314 		while ((readl(&pctl->stat) & PCTL_STAT_MSK) != CONFIG)
315 			/* spin */;
316 
317 		/* fall-through */
318 	case CONFIG:
319 		writel(GO_STATE, &pctl->sctl);
320 		while ((readl(&pctl->stat) & PCTL_STAT_MSK) == CONFIG)
321 			/* spin */;
322 		break;
323 
324 	case ACCESS:
325 		return;
326 
327 	default:
328 		break;
329 	}
330 }
331 
332 static void ddrctl_reset(struct rk3368_cru *cru)
333 {
334 	const u32 ctl_reset = BIT(3) | BIT(2);
335 	const u32 phy_reset = BIT(1) | BIT(0);
336 
337 	/*
338 	 * The PHY reset should be released before the PCTL reset.
339 	 *
340 	 * Note that the following sequence (including the number of
341 	 * us to delay between releasing the PHY and PCTL reset) has
342 	 * been adapted per feedback received from Rockchips, so do
343 	 * not try to optimise.
344 	 */
345 	rk_setreg(&cru->softrst_con[10], ctl_reset | phy_reset);
346 	udelay(1);
347 	rk_clrreg(&cru->softrst_con[10], phy_reset);
348 	udelay(5);
349 	rk_clrreg(&cru->softrst_con[10], ctl_reset);
350 }
351 
352 static void ddrphy_reset(struct rk3368_ddrphy *ddrphy)
353 {
354 	/*
355 	 * The analog part of the PHY should be release at least 1000
356 	 * DRAM cycles before the digital part of the PHY (waiting for
357 	 * 5us will ensure this for a DRAM clock as low as 200MHz).
358 	 */
359 	clrbits_le32(&ddrphy->reg[0], BIT(3) | BIT(2));
360 	udelay(1);
361 	setbits_le32(&ddrphy->reg[0], BIT(2));
362 	udelay(5);
363 	setbits_le32(&ddrphy->reg[0], BIT(3));
364 }
365 
366 static void ddrphy_config_delays(struct rk3368_ddrphy *ddrphy, u32 freq)
367 {
368 	u32 dqs_dll_delay;
369 
370 	setbits_le32(&ddrphy->reg[0x13], BIT(4));
371 	clrbits_le32(&ddrphy->reg[0x14], BIT(3));
372 
373 	setbits_le32(&ddrphy->reg[0x26], BIT(4));
374 	clrbits_le32(&ddrphy->reg[0x27], BIT(3));
375 
376 	setbits_le32(&ddrphy->reg[0x36], BIT(4));
377 	clrbits_le32(&ddrphy->reg[0x37], BIT(3));
378 
379 	setbits_le32(&ddrphy->reg[0x46], BIT(4));
380 	clrbits_le32(&ddrphy->reg[0x47], BIT(3));
381 
382 	setbits_le32(&ddrphy->reg[0x56], BIT(4));
383 	clrbits_le32(&ddrphy->reg[0x57], BIT(3));
384 
385 	if (freq <= 400000000)
386 		setbits_le32(&ddrphy->reg[0xa4], 0x1f);
387 	else
388 		clrbits_le32(&ddrphy->reg[0xa4], 0x1f);
389 
390 	if (freq < 681000000)
391 		dqs_dll_delay = 3; /* 67.5 degree delay */
392 	else
393 		dqs_dll_delay = 2; /* 45 degree delay */
394 
395 	writel(dqs_dll_delay, &ddrphy->reg[0x28]);
396 	writel(dqs_dll_delay, &ddrphy->reg[0x38]);
397 	writel(dqs_dll_delay, &ddrphy->reg[0x48]);
398 	writel(dqs_dll_delay, &ddrphy->reg[0x58]);
399 }
400 
401 static int dfi_cfg(struct rk3368_ddr_pctl *pctl)
402 {
403 	const ulong timeout_ms = 200;
404 	ulong tmp;
405 
406 	writel(DFI_DATA_BYTE_DISABLE_EN, &pctl->dfistcfg0);
407 
408 	writel(DFI_DRAM_CLK_SR_EN | DFI_DRAM_CLK_DPD_EN,
409 	       &pctl->dfistcfg1);
410 	writel(DFI_PARITY_INTR_EN | DFI_PARITY_EN, &pctl->dfistcfg2);
411 	writel(7 << TLP_RESP_TIME_SHIFT | LP_SR_EN | LP_PD_EN,
412 	       &pctl->dfilpcfg0);
413 
414 	writel(1, &pctl->dfitphyupdtype0);
415 
416 	writel(0x1f, &pctl->dfitphyrdlat);
417 	writel(0, &pctl->dfitphywrdata);
418 	writel(0, &pctl->dfiupdcfg);  /* phyupd and ctrlupd disabled */
419 
420 	setbits_le32(&pctl->dfistcfg0, DFI_INIT_START);
421 
422 	tmp = get_timer(0);
423 	do {
424 		if (get_timer(tmp) > timeout_ms) {
425 			pr_err("%s: DFI init did not complete within %ld ms\n",
426 			      __func__, timeout_ms);
427 			return -ETIME;
428 		}
429 	} while ((readl(&pctl->dfiststat0) & 1) == 0);
430 
431 	return 0;
432 }
433 
434 static inline u32 ps_to_tCK(const u32 ps, const ulong freq)
435 {
436 	const ulong MHz = 1000000;
437 	return DIV_ROUND_UP(ps * freq, 1000000 * MHz);
438 }
439 
440 static inline u32 ns_to_tCK(const u32 ns, const ulong freq)
441 {
442 	return ps_to_tCK(ns * 1000, freq);
443 }
444 
445 static inline u32 tCK_to_ps(const ulong tCK, const ulong freq)
446 {
447 	const ulong MHz = 1000000;
448 	return DIV_ROUND_UP(tCK * 1000000 * MHz, freq);
449 }
450 
451 static int pctl_calc_timings(struct rk3368_sdram_params *params,
452 			      ulong freq)
453 {
454 	struct rk3288_sdram_pctl_timing *pctl_timing = &params->pctl_timing;
455 	const ulong MHz = 1000000;
456 	u32 tccd;
457 	u32 tfaw_as_ps;
458 
459 	if (params->ddr_speed_bin != DDR3_1600K) {
460 		pr_err("%s: unimplemented DDR3 speed bin %d\n",
461 		      __func__, params->ddr_speed_bin);
462 		return -1;
463 	}
464 
465 	/* PCTL is clocked at 1/2 the DRAM clock; err on the side of caution */
466 	pctl_timing->togcnt1u = DIV_ROUND_UP(freq, 2 * MHz);
467 	pctl_timing->togcnt100n = DIV_ROUND_UP(freq / 10, 2 * MHz);
468 
469 	pctl_timing->tinit = 200;                 /* 200 usec                */
470 	pctl_timing->trsth = 500;                 /* 500 usec                */
471 	pctl_timing->trefi = 78;                  /* 7.8usec = 78 * 100ns    */
472 	params->trefi_mem_ddr3 = ns_to_tCK(pctl_timing->trefi * 100, freq);
473 
474 	if (freq <= (400 * MHz)) {
475 		pctl_timing->tcl = 6;
476 		pctl_timing->tcwl = 10;
477 	} else if (freq <= (533 * MHz)) {
478 		pctl_timing->tcl = 8;
479 		pctl_timing->tcwl = 6;
480 	} else if (freq <= (666 * MHz)) {
481 		pctl_timing->tcl = 10;
482 		pctl_timing->tcwl = 7;
483 	} else {
484 		pctl_timing->tcl = 11;
485 		pctl_timing->tcwl = 8;
486 	}
487 
488 	pctl_timing->tmrd = 4;                    /* 4 tCK (all speed bins)  */
489 	pctl_timing->trfc = ns_to_tCK(350, freq); /* tRFC: 350 (max) @ 8GBit */
490 	pctl_timing->trp = max(4u, ps_to_tCK(13750, freq));
491 	/*
492 	 * JESD-79:
493 	 *   READ to WRITE Command Delay = RL + tCCD / 2 + 2tCK - WL
494 	 */
495 	tccd = 4;
496 	pctl_timing->trtw = pctl_timing->tcl + tccd/2 + 2 - pctl_timing->tcwl;
497 	pctl_timing->tal = 0;
498 	pctl_timing->tras = ps_to_tCK(35000, freq);
499 	pctl_timing->trc = ps_to_tCK(48750, freq);
500 	pctl_timing->trcd = ps_to_tCK(13750, freq);
501 	pctl_timing->trrd = max(4u, ps_to_tCK(7500, freq));
502 	pctl_timing->trtp = max(4u, ps_to_tCK(7500, freq));
503 	pctl_timing->twr = ps_to_tCK(15000, freq);
504 	/* The DDR3 mode-register does only support even values for tWR > 8. */
505 	if (pctl_timing->twr > 8)
506 		pctl_timing->twr = (pctl_timing->twr + 1) & ~1;
507 	pctl_timing->twtr = max(4u, ps_to_tCK(7500, freq));
508 	pctl_timing->texsr = 512;                 /* tEXSR(max) is tDLLLK    */
509 	pctl_timing->txp = max(3u, ps_to_tCK(6000, freq));
510 	pctl_timing->txpdll = max(10u, ps_to_tCK(24000, freq));
511 	pctl_timing->tzqcs = max(64u, ps_to_tCK(80000, freq));
512 	pctl_timing->tzqcsi = 10000;               /* as used by Rockchip    */
513 	pctl_timing->tdqs = 1;                     /* fixed for DDR3         */
514 	pctl_timing->tcksre = max(5u, ps_to_tCK(10000, freq));
515 	pctl_timing->tcksrx = max(5u, ps_to_tCK(10000, freq));
516 	pctl_timing->tcke = max(3u, ps_to_tCK(5000, freq));
517 	pctl_timing->tmod = max(12u, ps_to_tCK(15000, freq));
518 	pctl_timing->trstl = ns_to_tCK(100, freq);
519 	pctl_timing->tzqcl = max(256u, ps_to_tCK(320000, freq));   /* tZQoper */
520 	pctl_timing->tmrr = 0;
521 	pctl_timing->tckesr = pctl_timing->tcke + 1;  /* JESD-79: tCKE + 1tCK */
522 	pctl_timing->tdpd = 0;    /* RK3368 TRM: "allowed values for DDR3: 0" */
523 
524 
525 	/*
526 	 * The controller can represent tFAW as 4x, 5x or 6x tRRD only.
527 	 * We want to use the smallest multiplier that satisfies the tFAW
528 	 * requirements of the given speed-bin.  If necessary, we stretch out
529 	 * tRRD to allow us to operate on a 6x multiplier for tFAW.
530 	 */
531 	tfaw_as_ps = 40000;      /* 40ns: tFAW for DDR3-1600K, 2KB page-size */
532 	if (tCK_to_ps(pctl_timing->trrd * 6, freq) < tfaw_as_ps) {
533 		/* If tFAW is > 6 x tRRD, we need to stretch tRRD */
534 		pctl_timing->trrd = ps_to_tCK(DIV_ROUND_UP(40000, 6), freq);
535 		params->tfaw_mult = TFAW_TRRD_MULT6;
536 	} else if (tCK_to_ps(pctl_timing->trrd * 5, freq) < tfaw_as_ps) {
537 		params->tfaw_mult = TFAW_TRRD_MULT6;
538 	} else if (tCK_to_ps(pctl_timing->trrd * 4, freq) < tfaw_as_ps) {
539 		params->tfaw_mult = TFAW_TRRD_MULT5;
540 	} else {
541 		params->tfaw_mult = TFAW_TRRD_MULT4;
542 	}
543 
544 	return 0;
545 }
546 
547 static void pctl_cfg(struct rk3368_ddr_pctl *pctl,
548 		     struct rk3368_sdram_params *params,
549 		     struct rk3368_grf *grf)
550 {
551 	/* Configure PCTL timing registers */
552 	params->pctl_timing.trefi |= BIT(31);   /* see PCTL_TREFI */
553 	copy_to_reg(&pctl->togcnt1u, &params->pctl_timing.togcnt1u,
554 		    sizeof(params->pctl_timing));
555 	writel(params->trefi_mem_ddr3, &pctl->trefi_mem_ddr3);
556 
557 	/* Set up ODT write selector and ODT write length */
558 	writel((RANK0_ODT_WRITE_SEL | RANK1_ODT_WRITE_SEL), &pctl->dfiodtcfg);
559 	writel(7 << ODT_LEN_BL8_W_SHIFT, &pctl->dfiodtcfg1);
560 
561 	/* Set up the CL/CWL-dependent timings of DFI */
562 	writel((params->pctl_timing.tcl - 1) / 2 - 1, &pctl->dfitrddataen);
563 	writel((params->pctl_timing.tcwl - 1) / 2 - 1, &pctl->dfitphywrlat);
564 
565 	/* DDR3 */
566 	writel(params->tfaw_mult | DDR3_EN | DDR2_DDR3_BL_8, &pctl->mcfg);
567 	writel(0x001c0004, &grf->ddrc0_con0);
568 
569 	setbits_le32(&pctl->scfg, HW_LOW_POWER_EN);
570 }
571 
572 static int ddrphy_data_training(struct rk3368_ddr_pctl *pctl,
573 				struct rk3368_ddrphy *ddrphy)
574 {
575 	const u32 trefi = readl(&pctl->trefi);
576 	const ulong timeout_ms = 500;
577 	ulong tmp;
578 
579 	/* disable auto-refresh */
580 	writel(0 | BIT(31), &pctl->trefi);
581 
582 	clrsetbits_le32(&ddrphy->reg[2], 0x33, 0x20);
583 	clrsetbits_le32(&ddrphy->reg[2], 0x33, 0x21);
584 
585 	tmp = get_timer(0);
586 	do {
587 		if (get_timer(tmp) > timeout_ms) {
588 			pr_err("%s: did not complete within %ld ms\n",
589 			      __func__, timeout_ms);
590 			return -ETIME;
591 		}
592 	} while ((readl(&ddrphy->reg[0xff]) & 0xf) != 0xf);
593 
594 	send_command(pctl, MCMD_RANK0 | MCMD_RANK1, PREA_CMD);
595 	clrsetbits_le32(&ddrphy->reg[2], 0x33, 0x20);
596 	/* resume auto-refresh */
597 	writel(trefi | BIT(31), &pctl->trefi);
598 
599 	return 0;
600 }
601 
602 static int sdram_col_row_detect(struct udevice *dev)
603 {
604 	struct dram_info *priv = dev_get_priv(dev);
605 	struct rk3368_sdram_params *params = dev_get_platdata(dev);
606 	struct rk3368_ddr_pctl *pctl = priv->pctl;
607 	struct rk3368_msch *msch = priv->msch;
608 	const u32 test_pattern = 0x5aa5f00f;
609 	int row, col;
610 	uintptr_t addr;
611 
612 	move_to_config_state(pctl);
613 	writel(6, &msch->ddrconf);
614 	move_to_access_state(pctl);
615 
616 	/* Detect col */
617 	for (col = 11; col >= 9; col--) {
618 		writel(0, CONFIG_SYS_SDRAM_BASE);
619 		addr = CONFIG_SYS_SDRAM_BASE +
620 			(1 << (col + params->chan.bw - 1));
621 		writel(test_pattern, addr);
622 		if ((readl(addr) == test_pattern) &&
623 		    (readl(CONFIG_SYS_SDRAM_BASE) == 0))
624 			break;
625 	}
626 
627 	if (col == 8) {
628 		pr_err("%s: col detect error\n", __func__);
629 		return -EINVAL;
630 	}
631 
632 	move_to_config_state(pctl);
633 	writel(15, &msch->ddrconf);
634 	move_to_access_state(pctl);
635 
636 	/* Detect row*/
637 	for (row = 16; row >= 12; row--) {
638 		writel(0, CONFIG_SYS_SDRAM_BASE);
639 		addr = CONFIG_SYS_SDRAM_BASE + (1 << (row + 15 - 1));
640 		writel(test_pattern, addr);
641 		if ((readl(addr) == test_pattern) &&
642 		    (readl(CONFIG_SYS_SDRAM_BASE) == 0))
643 			break;
644 	}
645 
646 	if (row == 11) {
647 		pr_err("%s: row detect error\n", __func__);
648 		return -EINVAL;
649 	}
650 
651 	/* Record results */
652 	debug("%s: col %d, row %d\n", __func__, col, row);
653 	params->chan.col = col;
654 	params->chan.cs0_row = row;
655 	params->chan.cs1_row = row;
656 	params->chan.row_3_4 = 0;
657 
658 	return 0;
659 }
660 
661 static int msch_niu_config(struct rk3368_msch *msch,
662 			   struct rk3368_sdram_params *params)
663 {
664 	int i;
665 	const u8 cols =	params->chan.col - ((params->chan.bw == 2) ? 0 : 1);
666 	const u8 rows = params->chan.cs0_row;
667 
668 	/*
669 	 * The DDR address-translation table always assumes a 32bit
670 	 * bus and the comparison below takes care of adjusting for
671 	 * a 16bit bus (i.e. one column-address is consumed).
672 	 */
673 	const struct {
674 		u8 rows;
675 		u8 columns;
676 		u8 type;
677 	} ddrconf_table[] = {
678 		/*
679 		 * C-B-R-D patterns are first. For these we require an
680 		 * exact match for the columns and rows (as there's
681 		 * one entry per possible configuration).
682 		 */
683 		[0] =  { .rows = 13, .columns = 10, .type = DMC_MSCH_CBRD },
684 		[1] =  { .rows = 14, .columns = 10, .type = DMC_MSCH_CBRD },
685 		[2] =  { .rows = 15, .columns = 10, .type = DMC_MSCH_CBRD },
686 		[3] =  { .rows = 16, .columns = 10, .type = DMC_MSCH_CBRD },
687 		[4] =  { .rows = 14, .columns = 11, .type = DMC_MSCH_CBRD },
688 		[5] =  { .rows = 15, .columns = 11, .type = DMC_MSCH_CBRD },
689 		[6] =  { .rows = 16, .columns = 11, .type = DMC_MSCH_CBRD },
690 		[7] =  { .rows = 13, .columns = 9, .type = DMC_MSCH_CBRD },
691 		[8] =  { .rows = 14, .columns = 9, .type = DMC_MSCH_CBRD },
692 		[9] =  { .rows = 15, .columns = 9, .type = DMC_MSCH_CBRD },
693 		[10] = { .rows = 16, .columns = 9, .type = DMC_MSCH_CBRD },
694 		/*
695 		 * 11 through 13 are C-R-B-D patterns. These are
696 		 * matched for an exact number of columns and to
697 		 * ensure that the hardware uses at least as many rows
698 		 * as the pattern requires (i.e. we make sure that
699 		 * there's no gaps up until we hit the device/chip-select;
700 		 * however, these patterns can accept up to 16 rows,
701 		 * as the row-address continues right after the CS
702 		 * switching)
703 		 */
704 		[11] = { .rows = 15, .columns = 10, .type = DMC_MSCH_CRBD },
705 		[12] = { .rows = 14, .columns = 11, .type = DMC_MSCH_CRBD },
706 		[13] = { .rows = 13, .columns = 10, .type = DMC_MSCH_CRBD },
707 		/*
708 		 * 14 and 15 are catch-all variants using a C-B-D-R
709 		 * scheme (i.e. alternating the chip-select every time
710 		 * C-B overflows) and stuffing the remaining C-bits
711 		 * into the top. Matching needs to make sure that the
712 		 * number of columns is either an exact match (i.e. we
713 		 * can use less the the maximum number of rows) -or-
714 		 * that the columns exceed what is given in this table
715 		 * and the rows are an exact match (in which case the
716 		 * remaining C-bits will be stuffed onto the top after
717 		 * the device/chip-select switches).
718 		 */
719 		[14] = { .rows = 16, .columns = 10, .type = DMC_MSCH_CBDR },
720 		[15] = { .rows = 16, .columns = 9, .type = DMC_MSCH_CBDR },
721 	};
722 
723 	/*
724 	 * For C-B-R-D, we need an exact match (i.e. both for the number of
725 	 * columns and rows), while for C-B-D-R, only the the number of
726 	 * columns needs to match.
727 	 */
728 	for (i = 0; i < ARRAY_SIZE(ddrconf_table); i++) {
729 		bool match = false;
730 
731 		/* If this entry if for a different matcher, then skip it */
732 		if (ddrconf_table[i].type != params->memory_schedule)
733 			continue;
734 
735 		/*
736 		 * Match according to the rules (exact/inexact/at-least)
737 		 * documented in the ddrconf_table above.
738 		 */
739 		switch (params->memory_schedule) {
740 		case DMC_MSCH_CBRD:
741 			match = (ddrconf_table[i].columns == cols) &&
742 				(ddrconf_table[i].rows == rows);
743 			break;
744 
745 		case DMC_MSCH_CRBD:
746 			match = (ddrconf_table[i].columns == cols) &&
747 				(ddrconf_table[i].rows <= rows);
748 			break;
749 
750 		case DMC_MSCH_CBDR:
751 			match = (ddrconf_table[i].columns == cols) ||
752 				((ddrconf_table[i].columns <= cols) &&
753 				 (ddrconf_table[i].rows == rows));
754 			break;
755 
756 		default:
757 			break;
758 		}
759 
760 		if (match) {
761 			debug("%s: setting ddrconf 0x%x\n", __func__, i);
762 			writel(i, &msch->ddrconf);
763 			return 0;
764 		}
765 	}
766 
767 	pr_err("%s: ddrconf (NIU config) not found\n", __func__);
768 	return -EINVAL;
769 }
770 
771 static void dram_all_config(struct udevice *dev)
772 {
773 	struct dram_info *priv = dev_get_priv(dev);
774 	struct rk3368_pmu_grf *pmugrf = priv->pmugrf;
775 	struct rk3368_sdram_params *params = dev_get_platdata(dev);
776 	const struct rk3288_sdram_channel *info = &params->chan;
777 	u32 sys_reg = 0;
778 	const int chan = 0;
779 
780 	sys_reg |= DDR3 << SYS_REG_DDRTYPE_SHIFT;
781 	sys_reg |= 0 << SYS_REG_NUM_CH_SHIFT;
782 
783 	sys_reg |= info->row_3_4 << SYS_REG_ROW_3_4_SHIFT(chan);
784 	sys_reg |= 1 << SYS_REG_CHINFO_SHIFT(chan);
785 	sys_reg |= (info->rank - 1) << SYS_REG_RANK_SHIFT(chan);
786 	sys_reg |= (info->col - 9) << SYS_REG_COL_SHIFT(chan);
787 	sys_reg |= info->bk == 3 ? 0 : 1 << SYS_REG_BK_SHIFT(chan);
788 	sys_reg |= (info->cs0_row - 13) << SYS_REG_CS0_ROW_SHIFT(chan);
789 	sys_reg |= (info->cs1_row - 13) << SYS_REG_CS1_ROW_SHIFT(chan);
790 	sys_reg |= (2 >> info->bw) << SYS_REG_BW_SHIFT(chan);
791 	sys_reg |= (2 >> info->dbw) << SYS_REG_DBW_SHIFT(chan);
792 
793 	writel(sys_reg, &pmugrf->os_reg[2]);
794 }
795 
796 static int setup_sdram(struct udevice *dev)
797 {
798 	struct dram_info *priv = dev_get_priv(dev);
799 	struct rk3368_sdram_params *params = dev_get_platdata(dev);
800 
801 	struct rk3368_ddr_pctl *pctl = priv->pctl;
802 	struct rk3368_ddrphy *ddrphy = priv->phy;
803 	struct rk3368_cru *cru = priv->cru;
804 	struct rk3368_grf *grf = priv->grf;
805 	struct rk3368_msch *msch = priv->msch;
806 
807 	int ret;
808 
809 	/* The input clock (i.e. DPLL) needs to be 2x the DRAM frequency */
810 	ret = clk_set_rate(&priv->ddr_clk, 2 * params->ddr_freq);
811 	if (ret < 0) {
812 		debug("%s: could not set DDR clock: %d\n", __func__, ret);
813 		return ret;
814 	}
815 
816 	/* Update the read-latency for the RK3368 */
817 	writel(0x32, &msch->readlatency);
818 
819 	/* Initialise the DDR PCTL and DDR PHY */
820 	ddrctl_reset(cru);
821 	ddrphy_reset(ddrphy);
822 	ddrphy_config_delays(ddrphy, params->ddr_freq);
823 	dfi_cfg(pctl);
824 	/* Configure relative system information of grf_ddrc0_con0 register */
825 	ddr_set_ddr3_mode(grf, true);
826 	ddr_set_noc_spr_err_stall(grf, true);
827 	/* Calculate timings */
828 	pctl_calc_timings(params, params->ddr_freq);
829 	/* Initialise the device timings in protocol controller */
830 	pctl_cfg(pctl, params, grf);
831 	/* Configure AL, CL ... information of PHY registers */
832 	ddrphy_config(ddrphy,
833 		      params->pctl_timing.tcl,
834 		      params->pctl_timing.tal,
835 		      params->pctl_timing.tcwl);
836 
837 	/* Initialize DRAM and configure with mode-register values */
838 	ret = memory_init(pctl, params);
839 	if (ret)
840 		goto error;
841 
842 	move_to_config_state(pctl);
843 	/* Perform data-training */
844 	ddrphy_data_training(pctl, ddrphy);
845 	move_to_access_state(pctl);
846 
847 	/* TODO(prt): could detect rank in training... */
848 	params->chan.rank = 2;
849 	/* TODO(prt): bus width is not auto-detected (yet)... */
850 	params->chan.bw = 2;  /* 32bit wide bus */
851 	params->chan.dbw = params->chan.dbw;  /* 32bit wide bus */
852 
853 	/* DDR3 is always 8 bank */
854 	params->chan.bk = 3;
855 	/* Detect col and row number */
856 	ret = sdram_col_row_detect(dev);
857 	if (ret)
858 		goto error;
859 
860 	/* Configure NIU DDR configuration */
861 	ret = msch_niu_config(msch, params);
862 	if (ret)
863 		goto error;
864 
865 	/* set up OS_REG to communicate w/ next stage and OS */
866 	dram_all_config(dev);
867 
868 	return 0;
869 
870 error:
871 	printf("DRAM init failed!\n");
872 	hang();
873 }
874 #endif
875 
876 static int rk3368_dmc_ofdata_to_platdata(struct udevice *dev)
877 {
878 	int ret = 0;
879 
880 #if !CONFIG_IS_ENABLED(OF_PLATDATA)
881 	struct rk3368_sdram_params *plat = dev_get_platdata(dev);
882 
883 	ret = regmap_init_mem(dev, &plat->map);
884 	if (ret)
885 		return ret;
886 #endif
887 
888 	return ret;
889 }
890 
891 #if CONFIG_IS_ENABLED(OF_PLATDATA)
892 static int conv_of_platdata(struct udevice *dev)
893 {
894 	struct rk3368_sdram_params *plat = dev_get_platdata(dev);
895 	struct dtd_rockchip_rk3368_dmc *of_plat = &plat->of_plat;
896 
897 	plat->ddr_freq = of_plat->rockchip_ddr_frequency;
898 	plat->ddr_speed_bin = of_plat->rockchip_ddr_speed_bin;
899 	plat->memory_schedule = of_plat->rockchip_memory_schedule;
900 
901 	return 0;
902 }
903 #endif
904 
905 static int rk3368_dmc_probe(struct udevice *dev)
906 {
907 #ifdef CONFIG_TPL_BUILD
908 	struct rk3368_sdram_params *plat = dev_get_platdata(dev);
909 	struct rk3368_ddr_pctl *pctl;
910 	struct rk3368_ddrphy *ddrphy;
911 	struct rk3368_cru *cru;
912 	struct rk3368_grf *grf;
913 	struct rk3368_msch *msch;
914 	int ret;
915 	struct udevice *dev_clk;
916 #endif
917 	struct dram_info *priv = dev_get_priv(dev);
918 
919 #if CONFIG_IS_ENABLED(OF_PLATDATA)
920 	ret = conv_of_platdata(dev);
921 	if (ret)
922 		return ret;
923 #endif
924 
925 	priv->pmugrf = syscon_get_first_range(ROCKCHIP_SYSCON_PMUGRF);
926 	debug("%s: pmugrf=%p\n", __func__, priv->pmugrf);
927 
928 #ifdef CONFIG_TPL_BUILD
929 	pctl = (struct rk3368_ddr_pctl *)plat->of_plat.reg[0];
930 	ddrphy = (struct rk3368_ddrphy *)plat->of_plat.reg[2];
931 	msch = syscon_get_first_range(ROCKCHIP_SYSCON_MSCH);
932 	grf = syscon_get_first_range(ROCKCHIP_SYSCON_GRF);
933 
934 	priv->pctl = pctl;
935 	priv->phy = ddrphy;
936 	priv->msch = msch;
937 	priv->grf = grf;
938 
939 	ret = rockchip_get_clk(&dev_clk);
940 	if (ret)
941 		return ret;
942 	priv->ddr_clk.id = CLK_DDR;
943 	ret = clk_request(dev_clk, &priv->ddr_clk);
944 	if (ret)
945 		return ret;
946 
947 	cru = rockchip_get_cru();
948 	priv->cru = cru;
949 	if (IS_ERR(priv->cru))
950 		return PTR_ERR(priv->cru);
951 
952 	ret = setup_sdram(dev);
953 	if (ret)
954 		return ret;
955 #endif
956 
957 	priv->info.base = 0;
958 	priv->info.size =
959 		rockchip_sdram_size((phys_addr_t)&priv->pmugrf->os_reg[2]);
960 
961 	/*
962 	* we use the 0x00000000~0xfdffffff space since 0xff000000~0xffffffff
963 	* is SoC register space (i.e. reserved), and 0xfe000000~0xfeffffff is
964 	* inaccessible for some IP controller.
965 	*/
966 	priv->info.size = min(priv->info.size, (size_t)0xfe000000);
967 
968 	return 0;
969 }
970 
971 static int rk3368_dmc_get_info(struct udevice *dev, struct ram_info *info)
972 {
973 	struct dram_info *priv = dev_get_priv(dev);
974 
975 	*info = priv->info;
976 	return 0;
977 }
978 
979 static struct ram_ops rk3368_dmc_ops = {
980 	.get_info = rk3368_dmc_get_info,
981 };
982 
983 
984 static const struct udevice_id rk3368_dmc_ids[] = {
985 	{ .compatible = "rockchip,rk3368-dmc" },
986 	{ }
987 };
988 
989 U_BOOT_DRIVER(dmc_rk3368) = {
990 	.name = "rockchip_rk3368_dmc",
991 	.id = UCLASS_RAM,
992 	.of_match = rk3368_dmc_ids,
993 	.ops = &rk3368_dmc_ops,
994 	.probe = rk3368_dmc_probe,
995 	.priv_auto_alloc_size = sizeof(struct dram_info),
996 	.ofdata_to_platdata = rk3368_dmc_ofdata_to_platdata,
997 	.probe = rk3368_dmc_probe,
998 	.priv_auto_alloc_size = sizeof(struct dram_info),
999 	.platdata_auto_alloc_size = sizeof(struct rk3368_sdram_params),
1000 };
1001