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