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 <hang.h>
10#include <log.h>
11#include <dt-bindings/memory/rk3368-dmc.h>
12#include <dt-structs.h>
13#include <ram.h>
14#include <regmap.h>
15#include <syscon.h>
16#include <asm/arch-rockchip/clock.h>
17#include <asm/arch-rockchip/cru_rk3368.h>
18#include <asm/arch-rockchip/grf_rk3368.h>
19#include <asm/arch-rockchip/hardware.h>
20#include <asm/arch-rockchip/ddr_rk3368.h>
21#include <asm/arch-rockchip/sdram.h>
22#include <asm/arch-rockchip/sdram_rk3288.h>
23#include <linux/bitops.h>
24#include <linux/delay.h>
25#include <linux/err.h>
26#include <linux/printk.h>
27
28struct dram_info {
29	struct ram_info info;
30	struct clk ddr_clk;
31	struct rk3368_cru *cru;
32	struct rk3368_grf *grf;
33	struct rk3368_ddr_pctl *pctl;
34	struct rk3368_ddrphy *phy;
35	struct rk3368_pmu_grf *pmugrf;
36	struct rk3368_msch *msch;
37};
38
39struct rk3368_sdram_params {
40#if CONFIG_IS_ENABLED(OF_PLATDATA)
41	struct dtd_rockchip_rk3368_dmc of_plat;
42#endif
43	struct rk3288_sdram_pctl_timing pctl_timing;
44	u32 trefi_mem_ddr3;
45	struct rk3288_sdram_channel chan;
46	struct regmap *map;
47	u32 ddr_freq;
48	u32 memory_schedule;
49	u32 ddr_speed_bin;
50	u32 tfaw_mult;
51};
52
53/* PTCL bits */
54enum {
55	/* PCTL_DFISTCFG0 */
56	DFI_INIT_START = BIT(0),
57	DFI_DATA_BYTE_DISABLE_EN = BIT(2),
58
59	/* PCTL_DFISTCFG1 */
60	DFI_DRAM_CLK_SR_EN = BIT(0),
61	DFI_DRAM_CLK_DPD_EN = BIT(1),
62	ODT_LEN_BL8_W_SHIFT = 16,
63
64	/* PCTL_DFISTCFG2 */
65	DFI_PARITY_INTR_EN = BIT(0),
66	DFI_PARITY_EN = BIT(1),
67
68	/* PCTL_DFILPCFG0 */
69	TLP_RESP_TIME_SHIFT = 16,
70	LP_SR_EN = BIT(8),
71	LP_PD_EN = BIT(0),
72
73	/* PCTL_DFIODTCFG */
74	RANK0_ODT_WRITE_SEL = BIT(3),
75	RANK1_ODT_WRITE_SEL = BIT(11),
76
77	/* PCTL_SCFG */
78	HW_LOW_POWER_EN = BIT(0),
79
80	/* PCTL_MCMD */
81	START_CMD = BIT(31),
82	MCMD_RANK0 = BIT(20),
83	MCMD_RANK1 = BIT(21),
84	DESELECT_CMD = 0,
85	PREA_CMD,
86	REF_CMD,
87	MRS_CMD,
88	ZQCS_CMD,
89	ZQCL_CMD,
90	RSTL_CMD,
91	MRR_CMD	= 8,
92	DPDE_CMD,
93
94	/* PCTL_POWCTL */
95	POWER_UP_START = BIT(0),
96
97	/* PCTL_POWSTAT */
98	POWER_UP_DONE = BIT(0),
99
100	/* PCTL_SCTL */
101	INIT_STATE = 0,
102	CFG_STATE,
103	GO_STATE,
104	SLEEP_STATE,
105	WAKEUP_STATE,
106
107	/* PCTL_STAT */
108	LP_TRIG_SHIFT = 4,
109	LP_TRIG_MASK = 7,
110	PCTL_STAT_MSK = 7,
111	INIT_MEM = 0,
112	CONFIG,
113	CFG_REQ,
114	ACCESS,
115	ACCESS_REQ,
116	LOW_POWER,
117	LOW_POWER_ENTRY_REQ,
118	LOW_POWER_EXIT_REQ,
119
120	/* PCTL_MCFG */
121	DDR2_DDR3_BL_8 = BIT(0),
122	DDR3_EN = BIT(5),
123	TFAW_TRRD_MULT4 = (0 << 18),
124	TFAW_TRRD_MULT5 = (1 << 18),
125	TFAW_TRRD_MULT6 = (2 << 18),
126};
127
128#define DDR3_MR0_WR(n) \
129	((n <= 8) ? ((n - 4) << 9) : (((n >> 1) & 0x7) << 9))
130#define DDR3_MR0_CL(n) \
131	((((n - 4) & 0x7) << 4) | (((n - 4) & 0x8) >> 2))
132#define DDR3_MR0_BL8 \
133	(0 << 0)
134#define DDR3_MR0_DLL_RESET \
135	(1 << 8)
136#define DDR3_MR1_RTT120OHM \
137	((0 << 9) | (1 << 6) | (0 << 2))
138#define DDR3_MR2_TWL(n) \
139	(((n - 5) & 0x7) << 3)
140
141
142#ifdef CONFIG_TPL_BUILD
143
144static void ddr_set_noc_spr_err_stall(struct rk3368_grf *grf, bool enable)
145{
146	if (enable)
147		rk_setreg(&grf->ddrc0_con0, NOC_RSP_ERR_STALL);
148	else
149		rk_clrreg(&grf->ddrc0_con0, NOC_RSP_ERR_STALL);
150}
151
152static void ddr_set_ddr3_mode(struct rk3368_grf *grf, bool ddr3_mode)
153{
154	if (ddr3_mode)
155		rk_setreg(&grf->ddrc0_con0, MSCH0_MAINDDR3_DDR3);
156	else
157		rk_clrreg(&grf->ddrc0_con0, MSCH0_MAINDDR3_DDR3);
158}
159
160static void ddrphy_config(struct rk3368_ddrphy *phy,
161			  u32 tcl, u32 tal, u32 tcwl)
162{
163	int i;
164
165	/* Set to DDR3 mode */
166	clrsetbits_le32(&phy->reg[1], 0x3, 0x0);
167
168	/* DDRPHY_REGB: CL, AL */
169	clrsetbits_le32(&phy->reg[0xb], 0xff, tcl << 4 | tal);
170	/* DDRPHY_REGC: CWL */
171	clrsetbits_le32(&phy->reg[0xc], 0x0f, tcwl);
172
173	/* Update drive-strength */
174	writel(0xcc, &phy->reg[0x11]);
175	writel(0xaa, &phy->reg[0x16]);
176	/*
177	 * Update NRCOMP/PRCOMP for all 4 channels (for details of all
178	 * affected registers refer to the documentation of DDRPHY_REG20
179	 * and DDRPHY_REG21 in the RK3368 TRM.
180	 */
181	for (i = 0; i < 4; ++i) {
182		writel(0xcc, &phy->reg[0x20 + i * 0x10]);
183		writel(0x44, &phy->reg[0x21 + i * 0x10]);
184	}
185
186	/* Enable write-leveling calibration bypass */
187	setbits_le32(&phy->reg[2], BIT(3));
188}
189
190static void copy_to_reg(u32 *dest, const u32 *src, u32 n)
191{
192	int i;
193
194	for (i = 0; i < n / sizeof(u32); i++)
195		writel(*src++, dest++);
196}
197
198static void send_command(struct rk3368_ddr_pctl *pctl, u32 rank, u32 cmd)
199{
200	u32 mcmd = START_CMD | cmd | rank;
201
202	debug("%s: writing %x to MCMD\n", __func__, mcmd);
203	writel(mcmd, &pctl->mcmd);
204	while (readl(&pctl->mcmd) & START_CMD)
205		/* spin */;
206}
207
208static void send_mrs(struct rk3368_ddr_pctl *pctl,
209			    u32 rank, u32 mr_num, u32 mr_data)
210{
211	u32 mcmd = START_CMD | MRS_CMD | rank | (mr_num << 17) | (mr_data << 4);
212
213	debug("%s: writing %x to MCMD\n", __func__, mcmd);
214	writel(mcmd, &pctl->mcmd);
215	while (readl(&pctl->mcmd) & START_CMD)
216		/* spin */;
217}
218
219static int memory_init(struct rk3368_ddr_pctl *pctl,
220		       struct rk3368_sdram_params *params)
221{
222	u32 mr[4];
223	const ulong timeout_ms = 500;
224	ulong tmp;
225
226	/*
227	 * Power up DRAM by DDR_PCTL_POWCTL[0] register of PCTL and
228	 * wait power up DRAM finish with DDR_PCTL_POWSTAT[0] register
229	 * of PCTL.
230	 */
231	writel(POWER_UP_START, &pctl->powctl);
232
233	tmp = get_timer(0);
234	do {
235		if (get_timer(tmp) > timeout_ms) {
236			pr_err("%s: POWER_UP_START did not complete in %ld ms\n",
237			      __func__, timeout_ms);
238			return -ETIME;
239		}
240	} while (!(readl(&pctl->powstat) & POWER_UP_DONE));
241
242	/* Configure MR0 through MR3 */
243	mr[0] = DDR3_MR0_WR(params->pctl_timing.twr) |
244		DDR3_MR0_CL(params->pctl_timing.tcl) |
245		DDR3_MR0_DLL_RESET;
246	mr[1] = DDR3_MR1_RTT120OHM;
247	mr[2] = DDR3_MR2_TWL(params->pctl_timing.tcwl);
248	mr[3] = 0;
249
250	/*
251	 * Also see RK3368 Technical Reference Manual:
252	 *   "16.6.2 Initialization (DDR3 Initialization Sequence)"
253	 */
254	send_command(pctl, MCMD_RANK0 | MCMD_RANK1, DESELECT_CMD);
255	udelay(1);
256	send_command(pctl, MCMD_RANK0 | MCMD_RANK1, PREA_CMD);
257	send_mrs(pctl, MCMD_RANK0 | MCMD_RANK1, 2, mr[2]);
258	send_mrs(pctl, MCMD_RANK0 | MCMD_RANK1, 3, mr[3]);
259	send_mrs(pctl, MCMD_RANK0 | MCMD_RANK1, 1, mr[1]);
260	send_mrs(pctl, MCMD_RANK0 | MCMD_RANK1, 0, mr[0]);
261	send_command(pctl, MCMD_RANK0 | MCMD_RANK1, ZQCL_CMD);
262
263	return 0;
264}
265
266static void move_to_config_state(struct rk3368_ddr_pctl *pctl)
267{
268	/*
269	 * Also see RK3368 Technical Reference Manual:
270	 *   "16.6.1 State transition of PCTL (Moving to Config State)"
271	 */
272	u32 state = readl(&pctl->stat) & PCTL_STAT_MSK;
273
274	switch (state) {
275	case LOW_POWER:
276		writel(WAKEUP_STATE, &pctl->sctl);
277		while ((readl(&pctl->stat) & PCTL_STAT_MSK) != ACCESS)
278			/* spin */;
279
280		/* fall-through */
281	case ACCESS:
282	case INIT_MEM:
283		writel(CFG_STATE, &pctl->sctl);
284		while ((readl(&pctl->stat) & PCTL_STAT_MSK) != CONFIG)
285			/* spin */;
286		break;
287
288	case CONFIG:
289		return;
290
291	default:
292		break;
293	}
294}
295
296static void move_to_access_state(struct rk3368_ddr_pctl *pctl)
297{
298	/*
299	 * Also see RK3368 Technical Reference Manual:
300	 *   "16.6.1 State transition of PCTL (Moving to Access State)"
301	 */
302	u32 state = readl(&pctl->stat) & PCTL_STAT_MSK;
303
304	switch (state) {
305	case LOW_POWER:
306		if (((readl(&pctl->stat) >> LP_TRIG_SHIFT) &
307		     LP_TRIG_MASK) == 1)
308			return;
309
310		writel(WAKEUP_STATE, &pctl->sctl);
311		while ((readl(&pctl->stat) & PCTL_STAT_MSK) != ACCESS)
312			/* spin */;
313
314		/* fall-through */
315	case INIT_MEM:
316		writel(CFG_STATE, &pctl->sctl);
317		while ((readl(&pctl->stat) & PCTL_STAT_MSK) != CONFIG)
318			/* spin */;
319
320		/* fall-through */
321	case CONFIG:
322		writel(GO_STATE, &pctl->sctl);
323		while ((readl(&pctl->stat) & PCTL_STAT_MSK) == CONFIG)
324			/* spin */;
325		break;
326
327	case ACCESS:
328		return;
329
330	default:
331		break;
332	}
333}
334
335static void ddrctl_reset(struct rk3368_cru *cru)
336{
337	const u32 ctl_reset = BIT(3) | BIT(2);
338	const u32 phy_reset = BIT(1) | BIT(0);
339
340	/*
341	 * The PHY reset should be released before the PCTL reset.
342	 *
343	 * Note that the following sequence (including the number of
344	 * us to delay between releasing the PHY and PCTL reset) has
345	 * been adapted per feedback received from Rockchips, so do
346	 * not try to optimise.
347	 */
348	rk_setreg(&cru->softrst_con[10], ctl_reset | phy_reset);
349	udelay(1);
350	rk_clrreg(&cru->softrst_con[10], phy_reset);
351	udelay(5);
352	rk_clrreg(&cru->softrst_con[10], ctl_reset);
353}
354
355static void ddrphy_reset(struct rk3368_ddrphy *ddrphy)
356{
357	/*
358	 * The analog part of the PHY should be release at least 1000
359	 * DRAM cycles before the digital part of the PHY (waiting for
360	 * 5us will ensure this for a DRAM clock as low as 200MHz).
361	 */
362	clrbits_le32(&ddrphy->reg[0], BIT(3) | BIT(2));
363	udelay(1);
364	setbits_le32(&ddrphy->reg[0], BIT(2));
365	udelay(5);
366	setbits_le32(&ddrphy->reg[0], BIT(3));
367}
368
369static void ddrphy_config_delays(struct rk3368_ddrphy *ddrphy, u32 freq)
370{
371	u32 dqs_dll_delay;
372
373	setbits_le32(&ddrphy->reg[0x13], BIT(4));
374	clrbits_le32(&ddrphy->reg[0x14], BIT(3));
375
376	setbits_le32(&ddrphy->reg[0x26], BIT(4));
377	clrbits_le32(&ddrphy->reg[0x27], BIT(3));
378
379	setbits_le32(&ddrphy->reg[0x36], BIT(4));
380	clrbits_le32(&ddrphy->reg[0x37], BIT(3));
381
382	setbits_le32(&ddrphy->reg[0x46], BIT(4));
383	clrbits_le32(&ddrphy->reg[0x47], BIT(3));
384
385	setbits_le32(&ddrphy->reg[0x56], BIT(4));
386	clrbits_le32(&ddrphy->reg[0x57], BIT(3));
387
388	if (freq <= 400000000)
389		setbits_le32(&ddrphy->reg[0xa4], 0x1f);
390	else
391		clrbits_le32(&ddrphy->reg[0xa4], 0x1f);
392
393	if (freq < 681000000)
394		dqs_dll_delay = 3; /* 67.5 degree delay */
395	else
396		dqs_dll_delay = 2; /* 45 degree delay */
397
398	writel(dqs_dll_delay, &ddrphy->reg[0x28]);
399	writel(dqs_dll_delay, &ddrphy->reg[0x38]);
400	writel(dqs_dll_delay, &ddrphy->reg[0x48]);
401	writel(dqs_dll_delay, &ddrphy->reg[0x58]);
402}
403
404static int dfi_cfg(struct rk3368_ddr_pctl *pctl)
405{
406	const ulong timeout_ms = 200;
407	ulong tmp;
408
409	writel(DFI_DATA_BYTE_DISABLE_EN, &pctl->dfistcfg0);
410
411	writel(DFI_DRAM_CLK_SR_EN | DFI_DRAM_CLK_DPD_EN,
412	       &pctl->dfistcfg1);
413	writel(DFI_PARITY_INTR_EN | DFI_PARITY_EN, &pctl->dfistcfg2);
414	writel(7 << TLP_RESP_TIME_SHIFT | LP_SR_EN | LP_PD_EN,
415	       &pctl->dfilpcfg0);
416
417	writel(1, &pctl->dfitphyupdtype0);
418
419	writel(0x1f, &pctl->dfitphyrdlat);
420	writel(0, &pctl->dfitphywrdata);
421	writel(0, &pctl->dfiupdcfg);  /* phyupd and ctrlupd disabled */
422
423	setbits_le32(&pctl->dfistcfg0, DFI_INIT_START);
424
425	tmp = get_timer(0);
426	do {
427		if (get_timer(tmp) > timeout_ms) {
428			pr_err("%s: DFI init did not complete within %ld ms\n",
429			      __func__, timeout_ms);
430			return -ETIME;
431		}
432	} while ((readl(&pctl->dfiststat0) & 1) == 0);
433
434	return 0;
435}
436
437static inline u32 ps_to_tCK(const u32 ps, const ulong freq)
438{
439	const ulong MHz = 1000000;
440	return DIV_ROUND_UP(ps * freq, 1000000 * MHz);
441}
442
443static inline u32 ns_to_tCK(const u32 ns, const ulong freq)
444{
445	return ps_to_tCK(ns * 1000, freq);
446}
447
448static inline u32 tCK_to_ps(const ulong tCK, const ulong freq)
449{
450	const ulong MHz = 1000000;
451	return DIV_ROUND_UP(tCK * 1000000 * MHz, freq);
452}
453
454static int pctl_calc_timings(struct rk3368_sdram_params *params,
455			      ulong freq)
456{
457	struct rk3288_sdram_pctl_timing *pctl_timing = &params->pctl_timing;
458	const ulong MHz = 1000000;
459	u32 tccd;
460	u32 tfaw_as_ps;
461
462	if (params->ddr_speed_bin != DDR3_1600K) {
463		pr_err("%s: unimplemented DDR3 speed bin %d\n",
464		      __func__, params->ddr_speed_bin);
465		return -1;
466	}
467
468	/* PCTL is clocked at 1/2 the DRAM clock; err on the side of caution */
469	pctl_timing->togcnt1u = DIV_ROUND_UP(freq, 2 * MHz);
470	pctl_timing->togcnt100n = DIV_ROUND_UP(freq / 10, 2 * MHz);
471
472	pctl_timing->tinit = 200;                 /* 200 usec                */
473	pctl_timing->trsth = 500;                 /* 500 usec                */
474	pctl_timing->trefi = 78;                  /* 7.8usec = 78 * 100ns    */
475	params->trefi_mem_ddr3 = ns_to_tCK(pctl_timing->trefi * 100, freq);
476
477	if (freq <= (400 * MHz)) {
478		pctl_timing->tcl = 6;
479		pctl_timing->tcwl = 10;
480	} else if (freq <= (533 * MHz)) {
481		pctl_timing->tcl = 8;
482		pctl_timing->tcwl = 6;
483	} else if (freq <= (666 * MHz)) {
484		pctl_timing->tcl = 10;
485		pctl_timing->tcwl = 7;
486	} else {
487		pctl_timing->tcl = 11;
488		pctl_timing->tcwl = 8;
489	}
490
491	pctl_timing->tmrd = 4;                    /* 4 tCK (all speed bins)  */
492	pctl_timing->trfc = ns_to_tCK(350, freq); /* tRFC: 350 (max) @ 8GBit */
493	pctl_timing->trp = max(4u, ps_to_tCK(13750, freq));
494	/*
495	 * JESD-79:
496	 *   READ to WRITE Command Delay = RL + tCCD / 2 + 2tCK - WL
497	 */
498	tccd = 4;
499	pctl_timing->trtw = pctl_timing->tcl + tccd/2 + 2 - pctl_timing->tcwl;
500	pctl_timing->tal = 0;
501	pctl_timing->tras = ps_to_tCK(35000, freq);
502	pctl_timing->trc = ps_to_tCK(48750, freq);
503	pctl_timing->trcd = ps_to_tCK(13750, freq);
504	pctl_timing->trrd = max(4u, ps_to_tCK(7500, freq));
505	pctl_timing->trtp = max(4u, ps_to_tCK(7500, freq));
506	pctl_timing->twr = ps_to_tCK(15000, freq);
507	/* The DDR3 mode-register does only support even values for tWR > 8. */
508	if (pctl_timing->twr > 8)
509		pctl_timing->twr = (pctl_timing->twr + 1) & ~1;
510	pctl_timing->twtr = max(4u, ps_to_tCK(7500, freq));
511	pctl_timing->texsr = 512;                 /* tEXSR(max) is tDLLLK    */
512	pctl_timing->txp = max(3u, ps_to_tCK(6000, freq));
513	pctl_timing->txpdll = max(10u, ps_to_tCK(24000, freq));
514	pctl_timing->tzqcs = max(64u, ps_to_tCK(80000, freq));
515	pctl_timing->tzqcsi = 10000;               /* as used by Rockchip    */
516	pctl_timing->tdqs = 1;                     /* fixed for DDR3         */
517	pctl_timing->tcksre = max(5u, ps_to_tCK(10000, freq));
518	pctl_timing->tcksrx = max(5u, ps_to_tCK(10000, freq));
519	pctl_timing->tcke = max(3u, ps_to_tCK(5000, freq));
520	pctl_timing->tmod = max(12u, ps_to_tCK(15000, freq));
521	pctl_timing->trstl = ns_to_tCK(100, freq);
522	pctl_timing->tzqcl = max(256u, ps_to_tCK(320000, freq));   /* tZQoper */
523	pctl_timing->tmrr = 0;
524	pctl_timing->tckesr = pctl_timing->tcke + 1;  /* JESD-79: tCKE + 1tCK */
525	pctl_timing->tdpd = 0;    /* RK3368 TRM: "allowed values for DDR3: 0" */
526
527
528	/*
529	 * The controller can represent tFAW as 4x, 5x or 6x tRRD only.
530	 * We want to use the smallest multiplier that satisfies the tFAW
531	 * requirements of the given speed-bin.  If necessary, we stretch out
532	 * tRRD to allow us to operate on a 6x multiplier for tFAW.
533	 */
534	tfaw_as_ps = 40000;      /* 40ns: tFAW for DDR3-1600K, 2KB page-size */
535	if (tCK_to_ps(pctl_timing->trrd * 6, freq) < tfaw_as_ps) {
536		/* If tFAW is > 6 x tRRD, we need to stretch tRRD */
537		pctl_timing->trrd = ps_to_tCK(DIV_ROUND_UP(40000, 6), freq);
538		params->tfaw_mult = TFAW_TRRD_MULT6;
539	} else if (tCK_to_ps(pctl_timing->trrd * 5, freq) < tfaw_as_ps) {
540		params->tfaw_mult = TFAW_TRRD_MULT6;
541	} else if (tCK_to_ps(pctl_timing->trrd * 4, freq) < tfaw_as_ps) {
542		params->tfaw_mult = TFAW_TRRD_MULT5;
543	} else {
544		params->tfaw_mult = TFAW_TRRD_MULT4;
545	}
546
547	return 0;
548}
549
550static void pctl_cfg(struct rk3368_ddr_pctl *pctl,
551		     struct rk3368_sdram_params *params,
552		     struct rk3368_grf *grf)
553{
554	/* Configure PCTL timing registers */
555	params->pctl_timing.trefi |= BIT(31);   /* see PCTL_TREFI */
556	copy_to_reg(&pctl->togcnt1u, &params->pctl_timing.togcnt1u,
557		    sizeof(params->pctl_timing));
558	writel(params->trefi_mem_ddr3, &pctl->trefi_mem_ddr3);
559
560	/* Set up ODT write selector and ODT write length */
561	writel((RANK0_ODT_WRITE_SEL | RANK1_ODT_WRITE_SEL), &pctl->dfiodtcfg);
562	writel(7 << ODT_LEN_BL8_W_SHIFT, &pctl->dfiodtcfg1);
563
564	/* Set up the CL/CWL-dependent timings of DFI */
565	writel((params->pctl_timing.tcl - 1) / 2 - 1, &pctl->dfitrddataen);
566	writel((params->pctl_timing.tcwl - 1) / 2 - 1, &pctl->dfitphywrlat);
567
568	/* DDR3 */
569	writel(params->tfaw_mult | DDR3_EN | DDR2_DDR3_BL_8, &pctl->mcfg);
570	writel(0x001c0004, &grf->ddrc0_con0);
571
572	setbits_le32(&pctl->scfg, HW_LOW_POWER_EN);
573}
574
575static int ddrphy_data_training(struct rk3368_ddr_pctl *pctl,
576				struct rk3368_ddrphy *ddrphy)
577{
578	const u32 trefi = readl(&pctl->trefi);
579	const ulong timeout_ms = 500;
580	ulong tmp;
581
582	/* disable auto-refresh */
583	writel(0 | BIT(31), &pctl->trefi);
584
585	clrsetbits_le32(&ddrphy->reg[2], 0x33, 0x20);
586	clrsetbits_le32(&ddrphy->reg[2], 0x33, 0x21);
587
588	tmp = get_timer(0);
589	do {
590		if (get_timer(tmp) > timeout_ms) {
591			pr_err("%s: did not complete within %ld ms\n",
592			      __func__, timeout_ms);
593			return -ETIME;
594		}
595	} while ((readl(&ddrphy->reg[0xff]) & 0xf) != 0xf);
596
597	send_command(pctl, MCMD_RANK0 | MCMD_RANK1, PREA_CMD);
598	clrsetbits_le32(&ddrphy->reg[2], 0x33, 0x20);
599	/* resume auto-refresh */
600	writel(trefi | BIT(31), &pctl->trefi);
601
602	return 0;
603}
604
605static int sdram_col_row_detect(struct udevice *dev)
606{
607	struct dram_info *priv = dev_get_priv(dev);
608	struct rk3368_sdram_params *params = dev_get_plat(dev);
609	struct rk3368_ddr_pctl *pctl = priv->pctl;
610	struct rk3368_msch *msch = priv->msch;
611	const u32 test_pattern = 0x5aa5f00f;
612	int row, col;
613	uintptr_t addr;
614
615	move_to_config_state(pctl);
616	writel(6, &msch->ddrconf);
617	move_to_access_state(pctl);
618
619	/* Detect col */
620	for (col = 11; col >= 9; col--) {
621		writel(0, CFG_SYS_SDRAM_BASE);
622		addr = CFG_SYS_SDRAM_BASE +
623			(1 << (col + params->chan.bw - 1));
624		writel(test_pattern, addr);
625		if ((readl(addr) == test_pattern) &&
626		    (readl(CFG_SYS_SDRAM_BASE) == 0))
627			break;
628	}
629
630	if (col == 8) {
631		pr_err("%s: col detect error\n", __func__);
632		return -EINVAL;
633	}
634
635	move_to_config_state(pctl);
636	writel(15, &msch->ddrconf);
637	move_to_access_state(pctl);
638
639	/* Detect row*/
640	for (row = 16; row >= 12; row--) {
641		writel(0, CFG_SYS_SDRAM_BASE);
642		addr = CFG_SYS_SDRAM_BASE + (1 << (row + 15 - 1));
643		writel(test_pattern, addr);
644		if ((readl(addr) == test_pattern) &&
645		    (readl(CFG_SYS_SDRAM_BASE) == 0))
646			break;
647	}
648
649	if (row == 11) {
650		pr_err("%s: row detect error\n", __func__);
651		return -EINVAL;
652	}
653
654	/* Record results */
655	debug("%s: col %d, row %d\n", __func__, col, row);
656	params->chan.col = col;
657	params->chan.cs0_row = row;
658	params->chan.cs1_row = row;
659	params->chan.row_3_4 = 0;
660
661	return 0;
662}
663
664static int msch_niu_config(struct rk3368_msch *msch,
665			   struct rk3368_sdram_params *params)
666{
667	int i;
668	const u8 cols =	params->chan.col - ((params->chan.bw == 2) ? 0 : 1);
669	const u8 rows = params->chan.cs0_row;
670
671	/*
672	 * The DDR address-translation table always assumes a 32bit
673	 * bus and the comparison below takes care of adjusting for
674	 * a 16bit bus (i.e. one column-address is consumed).
675	 */
676	const struct {
677		u8 rows;
678		u8 columns;
679		u8 type;
680	} ddrconf_table[] = {
681		/*
682		 * C-B-R-D patterns are first. For these we require an
683		 * exact match for the columns and rows (as there's
684		 * one entry per possible configuration).
685		 */
686		[0] =  { .rows = 13, .columns = 10, .type = DMC_MSCH_CBRD },
687		[1] =  { .rows = 14, .columns = 10, .type = DMC_MSCH_CBRD },
688		[2] =  { .rows = 15, .columns = 10, .type = DMC_MSCH_CBRD },
689		[3] =  { .rows = 16, .columns = 10, .type = DMC_MSCH_CBRD },
690		[4] =  { .rows = 14, .columns = 11, .type = DMC_MSCH_CBRD },
691		[5] =  { .rows = 15, .columns = 11, .type = DMC_MSCH_CBRD },
692		[6] =  { .rows = 16, .columns = 11, .type = DMC_MSCH_CBRD },
693		[7] =  { .rows = 13, .columns = 9, .type = DMC_MSCH_CBRD },
694		[8] =  { .rows = 14, .columns = 9, .type = DMC_MSCH_CBRD },
695		[9] =  { .rows = 15, .columns = 9, .type = DMC_MSCH_CBRD },
696		[10] = { .rows = 16, .columns = 9, .type = DMC_MSCH_CBRD },
697		/*
698		 * 11 through 13 are C-R-B-D patterns. These are
699		 * matched for an exact number of columns and to
700		 * ensure that the hardware uses at least as many rows
701		 * as the pattern requires (i.e. we make sure that
702		 * there's no gaps up until we hit the device/chip-select;
703		 * however, these patterns can accept up to 16 rows,
704		 * as the row-address continues right after the CS
705		 * switching)
706		 */
707		[11] = { .rows = 15, .columns = 10, .type = DMC_MSCH_CRBD },
708		[12] = { .rows = 14, .columns = 11, .type = DMC_MSCH_CRBD },
709		[13] = { .rows = 13, .columns = 10, .type = DMC_MSCH_CRBD },
710		/*
711		 * 14 and 15 are catch-all variants using a C-B-D-R
712		 * scheme (i.e. alternating the chip-select every time
713		 * C-B overflows) and stuffing the remaining C-bits
714		 * into the top. Matching needs to make sure that the
715		 * number of columns is either an exact match (i.e. we
716		 * can use less the the maximum number of rows) -or-
717		 * that the columns exceed what is given in this table
718		 * and the rows are an exact match (in which case the
719		 * remaining C-bits will be stuffed onto the top after
720		 * the device/chip-select switches).
721		 */
722		[14] = { .rows = 16, .columns = 10, .type = DMC_MSCH_CBDR },
723		[15] = { .rows = 16, .columns = 9, .type = DMC_MSCH_CBDR },
724	};
725
726	/*
727	 * For C-B-R-D, we need an exact match (i.e. both for the number of
728	 * columns and rows), while for C-B-D-R, only the the number of
729	 * columns needs to match.
730	 */
731	for (i = 0; i < ARRAY_SIZE(ddrconf_table); i++) {
732		bool match = false;
733
734		/* If this entry if for a different matcher, then skip it */
735		if (ddrconf_table[i].type != params->memory_schedule)
736			continue;
737
738		/*
739		 * Match according to the rules (exact/inexact/at-least)
740		 * documented in the ddrconf_table above.
741		 */
742		switch (params->memory_schedule) {
743		case DMC_MSCH_CBRD:
744			match = (ddrconf_table[i].columns == cols) &&
745				(ddrconf_table[i].rows == rows);
746			break;
747
748		case DMC_MSCH_CRBD:
749			match = (ddrconf_table[i].columns == cols) &&
750				(ddrconf_table[i].rows <= rows);
751			break;
752
753		case DMC_MSCH_CBDR:
754			match = (ddrconf_table[i].columns == cols) ||
755				((ddrconf_table[i].columns <= cols) &&
756				 (ddrconf_table[i].rows == rows));
757			break;
758
759		default:
760			break;
761		}
762
763		if (match) {
764			debug("%s: setting ddrconf 0x%x\n", __func__, i);
765			writel(i, &msch->ddrconf);
766			return 0;
767		}
768	}
769
770	pr_err("%s: ddrconf (NIU config) not found\n", __func__);
771	return -EINVAL;
772}
773
774static void dram_all_config(struct udevice *dev)
775{
776	struct dram_info *priv = dev_get_priv(dev);
777	struct rk3368_pmu_grf *pmugrf = priv->pmugrf;
778	struct rk3368_sdram_params *params = dev_get_plat(dev);
779	const struct rk3288_sdram_channel *info = &params->chan;
780	u32 sys_reg = 0;
781	const int chan = 0;
782
783	sys_reg |= DDR3 << SYS_REG_DDRTYPE_SHIFT;
784	sys_reg |= 0 << SYS_REG_NUM_CH_SHIFT;
785
786	sys_reg |= info->row_3_4 << SYS_REG_ROW_3_4_SHIFT(chan);
787	sys_reg |= 1 << SYS_REG_CHINFO_SHIFT(chan);
788	sys_reg |= (info->rank - 1) << SYS_REG_RANK_SHIFT(chan);
789	sys_reg |= (info->col - 9) << SYS_REG_COL_SHIFT(chan);
790	sys_reg |= info->bk == 3 ? 0 : 1 << SYS_REG_BK_SHIFT(chan);
791	sys_reg |= (info->cs0_row - 13) << SYS_REG_CS0_ROW_SHIFT(chan);
792	sys_reg |= (info->cs1_row - 13) << SYS_REG_CS1_ROW_SHIFT(chan);
793	sys_reg |= (2 >> info->bw) << SYS_REG_BW_SHIFT(chan);
794	sys_reg |= (2 >> info->dbw) << SYS_REG_DBW_SHIFT(chan);
795
796	writel(sys_reg, &pmugrf->os_reg[2]);
797}
798
799static int setup_sdram(struct udevice *dev)
800{
801	struct dram_info *priv = dev_get_priv(dev);
802	struct rk3368_sdram_params *params = dev_get_plat(dev);
803
804	struct rk3368_ddr_pctl *pctl = priv->pctl;
805	struct rk3368_ddrphy *ddrphy = priv->phy;
806	struct rk3368_cru *cru = priv->cru;
807	struct rk3368_grf *grf = priv->grf;
808	struct rk3368_msch *msch = priv->msch;
809
810	int ret;
811
812	/* The input clock (i.e. DPLL) needs to be 2x the DRAM frequency */
813	ret = clk_set_rate(&priv->ddr_clk, 2 * params->ddr_freq);
814	if (ret < 0) {
815		debug("%s: could not set DDR clock: %d\n", __func__, ret);
816		return ret;
817	}
818
819	/* Update the read-latency for the RK3368 */
820	writel(0x32, &msch->readlatency);
821
822	/* Initialise the DDR PCTL and DDR PHY */
823	ddrctl_reset(cru);
824	ddrphy_reset(ddrphy);
825	ddrphy_config_delays(ddrphy, params->ddr_freq);
826	dfi_cfg(pctl);
827	/* Configure relative system information of grf_ddrc0_con0 register */
828	ddr_set_ddr3_mode(grf, true);
829	ddr_set_noc_spr_err_stall(grf, true);
830	/* Calculate timings */
831	pctl_calc_timings(params, params->ddr_freq);
832	/* Initialise the device timings in protocol controller */
833	pctl_cfg(pctl, params, grf);
834	/* Configure AL, CL ... information of PHY registers */
835	ddrphy_config(ddrphy,
836		      params->pctl_timing.tcl,
837		      params->pctl_timing.tal,
838		      params->pctl_timing.tcwl);
839
840	/* Initialize DRAM and configure with mode-register values */
841	ret = memory_init(pctl, params);
842	if (ret)
843		goto error;
844
845	move_to_config_state(pctl);
846	/* Perform data-training */
847	ddrphy_data_training(pctl, ddrphy);
848	move_to_access_state(pctl);
849
850	/* TODO(prt): could detect rank in training... */
851#ifdef CONFIG_TARGET_EVB_PX5
852	params->chan.rank = 1;
853#else
854	params->chan.rank = 2;
855#endif
856	/* TODO(prt): bus width is not auto-detected (yet)... */
857	params->chan.bw = 2;  /* 32bit wide bus */
858	params->chan.dbw = params->chan.dbw;  /* 32bit wide bus */
859
860	/* DDR3 is always 8 bank */
861	params->chan.bk = 3;
862	/* Detect col and row number */
863	ret = sdram_col_row_detect(dev);
864	if (ret)
865		goto error;
866
867	/* Configure NIU DDR configuration */
868	ret = msch_niu_config(msch, params);
869	if (ret)
870		goto error;
871
872	/* set up OS_REG to communicate w/ next stage and OS */
873	dram_all_config(dev);
874
875	return 0;
876
877error:
878	printf("DRAM init failed!\n");
879	hang();
880}
881#endif
882
883static int rk3368_dmc_of_to_plat(struct udevice *dev)
884{
885	int ret = 0;
886
887	if (CONFIG_IS_ENABLED(OF_REAL)) {
888		struct rk3368_sdram_params *plat = dev_get_plat(dev);
889
890		ret = regmap_init_mem(dev_ofnode(dev), &plat->map);
891		if (ret)
892			return ret;
893	}
894
895	return ret;
896}
897
898#if CONFIG_IS_ENABLED(OF_PLATDATA)
899static int conv_of_plat(struct udevice *dev)
900{
901	struct rk3368_sdram_params *plat = dev_get_plat(dev);
902	struct dtd_rockchip_rk3368_dmc *of_plat = &plat->of_plat;
903
904	plat->ddr_freq = of_plat->rockchip_ddr_frequency;
905	plat->ddr_speed_bin = of_plat->rockchip_ddr_speed_bin;
906	plat->memory_schedule = of_plat->rockchip_memory_schedule;
907
908	return 0;
909}
910#endif
911
912static int rk3368_dmc_probe(struct udevice *dev)
913{
914#ifdef CONFIG_TPL_BUILD
915	struct rk3368_sdram_params *plat = dev_get_plat(dev);
916	struct rk3368_ddr_pctl *pctl;
917	struct rk3368_ddrphy *ddrphy;
918	struct rk3368_cru *cru;
919	struct rk3368_grf *grf;
920	struct rk3368_msch *msch;
921	int ret;
922	struct udevice *dev_clk;
923#endif
924	struct dram_info *priv = dev_get_priv(dev);
925
926#if CONFIG_IS_ENABLED(OF_PLATDATA)
927	ret = conv_of_plat(dev);
928	if (ret)
929		return ret;
930#endif
931
932	priv->pmugrf = syscon_get_first_range(ROCKCHIP_SYSCON_PMUGRF);
933	debug("%s: pmugrf=%p\n", __func__, priv->pmugrf);
934
935#ifdef CONFIG_TPL_BUILD
936	pctl = (struct rk3368_ddr_pctl *)plat->of_plat.reg[0];
937	ddrphy = (struct rk3368_ddrphy *)plat->of_plat.reg[2];
938	msch = syscon_get_first_range(ROCKCHIP_SYSCON_MSCH);
939	grf = syscon_get_first_range(ROCKCHIP_SYSCON_GRF);
940
941	priv->pctl = pctl;
942	priv->phy = ddrphy;
943	priv->msch = msch;
944	priv->grf = grf;
945
946	ret = rockchip_get_clk(&dev_clk);
947	if (ret)
948		return ret;
949	priv->ddr_clk.id = CLK_DDR;
950	ret = clk_request(dev_clk, &priv->ddr_clk);
951	if (ret)
952		return ret;
953
954	cru = rockchip_get_cru();
955	priv->cru = cru;
956	if (IS_ERR(priv->cru))
957		return PTR_ERR(priv->cru);
958
959	ret = setup_sdram(dev);
960	if (ret)
961		return ret;
962#endif
963
964	priv->info.base = 0;
965	priv->info.size =
966		rockchip_sdram_size((phys_addr_t)&priv->pmugrf->os_reg[2]);
967
968	/*
969	* we use the 0x00000000~0xfdffffff space since 0xff000000~0xffffffff
970	* is SoC register space (i.e. reserved), and 0xfe000000~0xfeffffff is
971	* inaccessible for some IP controller.
972	*/
973	priv->info.size = min(priv->info.size, (size_t)0xfe000000);
974
975	return 0;
976}
977
978static int rk3368_dmc_get_info(struct udevice *dev, struct ram_info *info)
979{
980	struct dram_info *priv = dev_get_priv(dev);
981
982	*info = priv->info;
983	return 0;
984}
985
986static struct ram_ops rk3368_dmc_ops = {
987	.get_info = rk3368_dmc_get_info,
988};
989
990
991static const struct udevice_id rk3368_dmc_ids[] = {
992	{ .compatible = "rockchip,rk3368-dmc" },
993	{ }
994};
995
996U_BOOT_DRIVER(rockchip_rk3368_dmc) = {
997	.name = "rockchip_rk3368_dmc",
998	.id = UCLASS_RAM,
999	.of_match = rk3368_dmc_ids,
1000	.ops = &rk3368_dmc_ops,
1001	.probe = rk3368_dmc_probe,
1002	.priv_auto	= sizeof(struct dram_info),
1003	.of_to_plat = rk3368_dmc_of_to_plat,
1004	.probe = rk3368_dmc_probe,
1005	.priv_auto	= sizeof(struct dram_info),
1006	.plat_auto	= sizeof(struct rk3368_sdram_params),
1007};
1008