1 /*
2  * sunxi DRAM controller initialization
3  * (C) Copyright 2012 Henrik Nordstrom <henrik@henriknordstrom.net>
4  * (C) Copyright 2013 Luke Kenneth Casson Leighton <lkcl@lkcl.net>
5  *
6  * Based on sun4i Linux kernel sources mach-sunxi/pm/standby/dram*.c
7  * and earlier U-Boot Allwiner A10 SPL work
8  *
9  * (C) Copyright 2007-2012
10  * Allwinner Technology Co., Ltd. <www.allwinnertech.com>
11  * Berg Xing <bergxing@allwinnertech.com>
12  * Tom Cubie <tangliang@allwinnertech.com>
13  *
14  * SPDX-License-Identifier:	GPL-2.0+
15  */
16 
17 /*
18  * Unfortunately the only documentation we have on the sun7i DRAM
19  * controller is Allwinner boot0 + boot1 code, and that code uses
20  * magic numbers & shifts with no explanations. Hence this code is
21  * rather undocumented and full of magic.
22  */
23 
24 #include <common.h>
25 #include <asm/io.h>
26 #include <asm/arch/clock.h>
27 #include <asm/arch/dram.h>
28 #include <asm/arch/timer.h>
29 #include <asm/arch/sys_proto.h>
30 
31 #define CPU_CFG_CHIP_VER(n) ((n) << 6)
32 #define CPU_CFG_CHIP_VER_MASK CPU_CFG_CHIP_VER(0x3)
33 #define CPU_CFG_CHIP_REV_A 0x0
34 #define CPU_CFG_CHIP_REV_C1 0x1
35 #define CPU_CFG_CHIP_REV_C2 0x2
36 #define CPU_CFG_CHIP_REV_B 0x3
37 
38 /*
39  * Wait up to 1s for mask to be clear in given reg.
40  */
41 static inline void await_bits_clear(u32 *reg, u32 mask)
42 {
43 	mctl_await_completion(reg, mask, 0);
44 }
45 
46 /*
47  * Wait up to 1s for mask to be set in given reg.
48  */
49 static inline void await_bits_set(u32 *reg, u32 mask)
50 {
51 	mctl_await_completion(reg, mask, mask);
52 }
53 
54 /*
55  * This performs the external DRAM reset by driving the RESET pin low and
56  * then high again. According to the DDR3 spec, the RESET pin needs to be
57  * kept low for at least 200 us.
58  */
59 static void mctl_ddr3_reset(void)
60 {
61 	struct sunxi_dram_reg *dram =
62 			(struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
63 
64 #ifdef CONFIG_MACH_SUN4I
65 	struct sunxi_timer_reg *timer =
66 			(struct sunxi_timer_reg *)SUNXI_TIMER_BASE;
67 	u32 reg_val;
68 
69 	writel(0, &timer->cpu_cfg);
70 	reg_val = readl(&timer->cpu_cfg);
71 
72 	if ((reg_val & CPU_CFG_CHIP_VER_MASK) !=
73 	    CPU_CFG_CHIP_VER(CPU_CFG_CHIP_REV_A)) {
74 		setbits_le32(&dram->mcr, DRAM_MCR_RESET);
75 		udelay(200);
76 		clrbits_le32(&dram->mcr, DRAM_MCR_RESET);
77 	} else
78 #endif
79 	{
80 		clrbits_le32(&dram->mcr, DRAM_MCR_RESET);
81 		udelay(200);
82 		setbits_le32(&dram->mcr, DRAM_MCR_RESET);
83 	}
84 	/* After the RESET pin is de-asserted, the DDR3 spec requires to wait
85 	 * for additional 500 us before driving the CKE pin (Clock Enable)
86 	 * high. The duration of this delay can be configured in the SDR_IDCR
87 	 * (Initialization Delay Configuration Register) and applied
88 	 * automatically by the DRAM controller during the DDR3 initialization
89 	 * step. But SDR_IDCR has limited range on sun4i/sun5i hardware and
90 	 * can't provide sufficient delay at DRAM clock frequencies higher than
91 	 * 524 MHz (while Allwinner A13 supports DRAM clock frequency up to
92 	 * 533 MHz according to the datasheet). Additionally, there is no
93 	 * official documentation for the SDR_IDCR register anywhere, and
94 	 * there is always a chance that we are interpreting it wrong.
95 	 * Better be safe than sorry, so add an explicit delay here. */
96 	udelay(500);
97 }
98 
99 static void mctl_set_drive(void)
100 {
101 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
102 
103 #ifdef CONFIG_MACH_SUN7I
104 	clrsetbits_le32(&dram->mcr, DRAM_MCR_MODE_NORM(0x3) | (0x3 << 28),
105 #else
106 	clrsetbits_le32(&dram->mcr, DRAM_MCR_MODE_NORM(0x3),
107 #endif
108 			DRAM_MCR_MODE_EN(0x3) |
109 			0xffc);
110 }
111 
112 static void mctl_itm_disable(void)
113 {
114 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
115 
116 	clrsetbits_le32(&dram->ccr, DRAM_CCR_INIT, DRAM_CCR_ITM_OFF);
117 }
118 
119 static void mctl_itm_enable(void)
120 {
121 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
122 
123 	clrbits_le32(&dram->ccr, DRAM_CCR_ITM_OFF);
124 }
125 
126 static void mctl_itm_reset(void)
127 {
128 	mctl_itm_disable();
129 	udelay(1); /* ITM reset needs a bit of delay */
130 	mctl_itm_enable();
131 	udelay(1);
132 }
133 
134 static void mctl_enable_dll0(u32 phase)
135 {
136 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
137 
138 	clrsetbits_le32(&dram->dllcr[0], 0x3f << 6,
139 			((phase >> 16) & 0x3f) << 6);
140 	clrsetbits_le32(&dram->dllcr[0], DRAM_DLLCR_NRESET, DRAM_DLLCR_DISABLE);
141 	udelay(2);
142 
143 	clrbits_le32(&dram->dllcr[0], DRAM_DLLCR_NRESET | DRAM_DLLCR_DISABLE);
144 	udelay(22);
145 
146 	clrsetbits_le32(&dram->dllcr[0], DRAM_DLLCR_DISABLE, DRAM_DLLCR_NRESET);
147 	udelay(22);
148 }
149 
150 /* Get the number of DDR byte lanes */
151 static u32 mctl_get_number_of_lanes(void)
152 {
153 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
154 	if ((readl(&dram->dcr) & DRAM_DCR_BUS_WIDTH_MASK) ==
155 				DRAM_DCR_BUS_WIDTH(DRAM_DCR_BUS_WIDTH_32BIT))
156 		return 4;
157 	else
158 		return 2;
159 }
160 
161 /*
162  * Note: This differs from pm/standby in that it checks the bus width
163  */
164 static void mctl_enable_dllx(u32 phase)
165 {
166 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
167 	u32 i, number_of_lanes;
168 
169 	number_of_lanes = mctl_get_number_of_lanes();
170 
171 	for (i = 1; i <= number_of_lanes; i++) {
172 		clrsetbits_le32(&dram->dllcr[i], 0xf << 14,
173 				(phase & 0xf) << 14);
174 		clrsetbits_le32(&dram->dllcr[i], DRAM_DLLCR_NRESET,
175 				DRAM_DLLCR_DISABLE);
176 		phase >>= 4;
177 	}
178 	udelay(2);
179 
180 	for (i = 1; i <= number_of_lanes; i++)
181 		clrbits_le32(&dram->dllcr[i], DRAM_DLLCR_NRESET |
182 			     DRAM_DLLCR_DISABLE);
183 	udelay(22);
184 
185 	for (i = 1; i <= number_of_lanes; i++)
186 		clrsetbits_le32(&dram->dllcr[i], DRAM_DLLCR_DISABLE,
187 				DRAM_DLLCR_NRESET);
188 	udelay(22);
189 }
190 
191 static u32 hpcr_value[32] = {
192 #ifdef CONFIG_MACH_SUN5I
193 	0, 0, 0, 0,
194 	0, 0, 0, 0,
195 	0, 0, 0, 0,
196 	0, 0, 0, 0,
197 	0x1031, 0x1031, 0x0735, 0x1035,
198 	0x1035, 0x0731, 0x1031, 0,
199 	0x0301, 0x0301, 0x0301, 0x0301,
200 	0x0301, 0x0301, 0x0301, 0
201 #endif
202 #ifdef CONFIG_MACH_SUN4I
203 	0x0301, 0x0301, 0x0301, 0x0301,
204 	0x0301, 0x0301, 0, 0,
205 	0, 0, 0, 0,
206 	0, 0, 0, 0,
207 	0x1031, 0x1031, 0x0735, 0x5031,
208 	0x1035, 0x0731, 0x1031, 0x0735,
209 	0x1035, 0x1031, 0x0731, 0x1035,
210 	0x1031, 0x0301, 0x0301, 0x0731
211 #endif
212 #ifdef CONFIG_MACH_SUN7I
213 	0x0301, 0x0301, 0x0301, 0x0301,
214 	0x0301, 0x0301, 0x0301, 0x0301,
215 	0, 0, 0, 0,
216 	0, 0, 0, 0,
217 	0x1031, 0x1031, 0x0735, 0x1035,
218 	0x1035, 0x0731, 0x1031, 0x0735,
219 	0x1035, 0x1031, 0x0731, 0x1035,
220 	0x0001, 0x1031, 0, 0x1031
221 	/* last row differs from boot0 source table
222 	 * 0x1031, 0x0301, 0x0301, 0x0731
223 	 * but boot0 code skips #28 and #30, and sets #29 and #31 to the
224 	 * value from #28 entry (0x1031)
225 	 */
226 #endif
227 };
228 
229 static void mctl_configure_hostport(void)
230 {
231 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
232 	u32 i;
233 
234 	for (i = 0; i < 32; i++)
235 		writel(hpcr_value[i], &dram->hpcr[i]);
236 }
237 
238 static void mctl_setup_dram_clock(u32 clk, u32 mbus_clk)
239 {
240 	u32 reg_val;
241 	struct sunxi_ccm_reg *ccm = (struct sunxi_ccm_reg *)SUNXI_CCM_BASE;
242 	u32 pll5p_clk, pll6x_clk;
243 	u32 pll5p_div, pll6x_div;
244 	u32 pll5p_rate, pll6x_rate;
245 
246 	/* setup DRAM PLL */
247 	reg_val = readl(&ccm->pll5_cfg);
248 	reg_val &= ~CCM_PLL5_CTRL_M_MASK;		/* set M to 0 (x1) */
249 	reg_val &= ~CCM_PLL5_CTRL_K_MASK;		/* set K to 0 (x1) */
250 	reg_val &= ~CCM_PLL5_CTRL_N_MASK;		/* set N to 0 (x0) */
251 	reg_val &= ~CCM_PLL5_CTRL_P_MASK;		/* set P to 0 (x1) */
252 #ifdef CONFIG_OLD_SUNXI_KERNEL_COMPAT
253 	/* Old kernels are hardcoded to P=1 (divide by 2) */
254 	reg_val |= CCM_PLL5_CTRL_P(1);
255 #endif
256 	if (clk >= 540 && clk < 552) {
257 		/* dram = 540MHz */
258 		reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(2));
259 		reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(3));
260 		reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(15));
261 	} else if (clk >= 512 && clk < 528) {
262 		/* dram = 512MHz */
263 		reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(3));
264 		reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(4));
265 		reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(16));
266 	} else if (clk >= 496 && clk < 504) {
267 		/* dram = 496MHz */
268 		reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(3));
269 		reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(2));
270 		reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(31));
271 	} else if (clk >= 468 && clk < 480) {
272 		/* dram = 468MHz */
273 		reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(2));
274 		reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(3));
275 		reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(13));
276 	} else if (clk >= 396 && clk < 408) {
277 		/* dram = 396MHz */
278 		reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(2));
279 		reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(3));
280 		reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(11));
281 	} else 	{
282 		/* any other frequency that is a multiple of 24 */
283 		reg_val |= CCM_PLL5_CTRL_M(CCM_PLL5_CTRL_M_X(2));
284 		reg_val |= CCM_PLL5_CTRL_K(CCM_PLL5_CTRL_K_X(2));
285 		reg_val |= CCM_PLL5_CTRL_N(CCM_PLL5_CTRL_N_X(clk / 24));
286 	}
287 	reg_val &= ~CCM_PLL5_CTRL_VCO_GAIN;		/* PLL VCO Gain off */
288 	reg_val |= CCM_PLL5_CTRL_EN;			/* PLL On */
289 	writel(reg_val, &ccm->pll5_cfg);
290 	udelay(5500);
291 
292 	setbits_le32(&ccm->pll5_cfg, CCM_PLL5_CTRL_DDR_CLK);
293 
294 #if defined(CONFIG_MACH_SUN4I) || defined(CONFIG_MACH_SUN7I)
295 	/* reset GPS */
296 	clrbits_le32(&ccm->gps_clk_cfg, CCM_GPS_CTRL_RESET | CCM_GPS_CTRL_GATE);
297 	setbits_le32(&ccm->ahb_gate0, CCM_AHB_GATE_GPS);
298 	udelay(1);
299 	clrbits_le32(&ccm->ahb_gate0, CCM_AHB_GATE_GPS);
300 #endif
301 
302 	/* setup MBUS clock */
303 	if (!mbus_clk)
304 		mbus_clk = 300;
305 
306 	/* PLL5P and PLL6 are the potential clock sources for MBUS */
307 	pll6x_clk = clock_get_pll6() / 1000000;
308 #ifdef CONFIG_MACH_SUN7I
309 	pll6x_clk *= 2; /* sun7i uses PLL6*2, sun5i uses just PLL6 */
310 #endif
311 	pll5p_clk = clock_get_pll5p() / 1000000;
312 	pll6x_div = DIV_ROUND_UP(pll6x_clk, mbus_clk);
313 	pll5p_div = DIV_ROUND_UP(pll5p_clk, mbus_clk);
314 	pll6x_rate = pll6x_clk / pll6x_div;
315 	pll5p_rate = pll5p_clk / pll5p_div;
316 
317 	if (pll6x_div <= 16 && pll6x_rate > pll5p_rate) {
318 		/* use PLL6 as the MBUS clock source */
319 		reg_val = CCM_MBUS_CTRL_GATE |
320 			  CCM_MBUS_CTRL_CLK_SRC(CCM_MBUS_CTRL_CLK_SRC_PLL6) |
321 			  CCM_MBUS_CTRL_N(CCM_MBUS_CTRL_N_X(1)) |
322 			  CCM_MBUS_CTRL_M(CCM_MBUS_CTRL_M_X(pll6x_div));
323 	} else if (pll5p_div <= 16) {
324 		/* use PLL5P as the MBUS clock source */
325 		reg_val = CCM_MBUS_CTRL_GATE |
326 			  CCM_MBUS_CTRL_CLK_SRC(CCM_MBUS_CTRL_CLK_SRC_PLL5) |
327 			  CCM_MBUS_CTRL_N(CCM_MBUS_CTRL_N_X(1)) |
328 			  CCM_MBUS_CTRL_M(CCM_MBUS_CTRL_M_X(pll5p_div));
329 	} else {
330 		panic("Bad mbus_clk\n");
331 	}
332 	writel(reg_val, &ccm->mbus_clk_cfg);
333 
334 	/*
335 	 * open DRAMC AHB & DLL register clock
336 	 * close it first
337 	 */
338 #if defined(CONFIG_MACH_SUN5I) || defined(CONFIG_MACH_SUN7I)
339 	clrbits_le32(&ccm->ahb_gate0, CCM_AHB_GATE_SDRAM | CCM_AHB_GATE_DLL);
340 #else
341 	clrbits_le32(&ccm->ahb_gate0, CCM_AHB_GATE_SDRAM);
342 #endif
343 	udelay(22);
344 
345 	/* then open it */
346 #if defined(CONFIG_MACH_SUN5I) || defined(CONFIG_MACH_SUN7I)
347 	setbits_le32(&ccm->ahb_gate0, CCM_AHB_GATE_SDRAM | CCM_AHB_GATE_DLL);
348 #else
349 	setbits_le32(&ccm->ahb_gate0, CCM_AHB_GATE_SDRAM);
350 #endif
351 	udelay(22);
352 }
353 
354 /*
355  * The data from rslrX and rdgrX registers (X=rank) is stored
356  * in a single 32-bit value using the following format:
357  *   bits [31:26] - DQS gating system latency for byte lane 3
358  *   bits [25:24] - DQS gating phase select for byte lane 3
359  *   bits [23:18] - DQS gating system latency for byte lane 2
360  *   bits [17:16] - DQS gating phase select for byte lane 2
361  *   bits [15:10] - DQS gating system latency for byte lane 1
362  *   bits [ 9:8 ] - DQS gating phase select for byte lane 1
363  *   bits [ 7:2 ] - DQS gating system latency for byte lane 0
364  *   bits [ 1:0 ] - DQS gating phase select for byte lane 0
365  */
366 static void mctl_set_dqs_gating_delay(int rank, u32 dqs_gating_delay)
367 {
368 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
369 	u32 lane, number_of_lanes = mctl_get_number_of_lanes();
370 	/* rank0 gating system latency (3 bits per lane: cycles) */
371 	u32 slr = readl(rank == 0 ? &dram->rslr0 : &dram->rslr1);
372 	/* rank0 gating phase select (2 bits per lane: 90, 180, 270, 360) */
373 	u32 dgr = readl(rank == 0 ? &dram->rdgr0 : &dram->rdgr1);
374 	for (lane = 0; lane < number_of_lanes; lane++) {
375 		u32 tmp = dqs_gating_delay >> (lane * 8);
376 		slr &= ~(7 << (lane * 3));
377 		slr |= ((tmp >> 2) & 7) << (lane * 3);
378 		dgr &= ~(3 << (lane * 2));
379 		dgr |= (tmp & 3) << (lane * 2);
380 	}
381 	writel(slr, rank == 0 ? &dram->rslr0 : &dram->rslr1);
382 	writel(dgr, rank == 0 ? &dram->rdgr0 : &dram->rdgr1);
383 }
384 
385 static int dramc_scan_readpipe(void)
386 {
387 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
388 	u32 reg_val;
389 
390 	/* data training trigger */
391 	clrbits_le32(&dram->csr, DRAM_CSR_FAILED);
392 	setbits_le32(&dram->ccr, DRAM_CCR_DATA_TRAINING);
393 
394 	/* check whether data training process has completed */
395 	await_bits_clear(&dram->ccr, DRAM_CCR_DATA_TRAINING);
396 
397 	/* check data training result */
398 	reg_val = readl(&dram->csr);
399 	if (reg_val & DRAM_CSR_FAILED)
400 		return -1;
401 
402 	return 0;
403 }
404 
405 static void dramc_clock_output_en(u32 on)
406 {
407 #if defined(CONFIG_MACH_SUN5I) || defined(CONFIG_MACH_SUN7I)
408 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
409 
410 	if (on)
411 		setbits_le32(&dram->mcr, DRAM_MCR_DCLK_OUT);
412 	else
413 		clrbits_le32(&dram->mcr, DRAM_MCR_DCLK_OUT);
414 #endif
415 #ifdef CONFIG_MACH_SUN4I
416 	struct sunxi_ccm_reg *ccm = (struct sunxi_ccm_reg *)SUNXI_CCM_BASE;
417 	if (on)
418 		setbits_le32(&ccm->dram_clk_gate, CCM_DRAM_CTRL_DCLK_OUT);
419 	else
420 		clrbits_le32(&ccm->dram_clk_gate, CCM_DRAM_CTRL_DCLK_OUT);
421 #endif
422 }
423 
424 /* tRFC in nanoseconds for different densities (from the DDR3 spec) */
425 static const u16 tRFC_DDR3_table[6] = {
426 	/* 256Mb    512Mb    1Gb      2Gb      4Gb      8Gb */
427 	   90,      90,      110,     160,     300,     350
428 };
429 
430 static void dramc_set_autorefresh_cycle(u32 clk, u32 density)
431 {
432 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
433 	u32 tRFC, tREFI;
434 
435 	tRFC = (tRFC_DDR3_table[density] * clk + 999) / 1000;
436 	tREFI = (7987 * clk) >> 10;	/* <= 7.8us */
437 
438 	writel(DRAM_DRR_TREFI(tREFI) | DRAM_DRR_TRFC(tRFC), &dram->drr);
439 }
440 
441 /* Calculate the value for A11, A10, A9 bits in MR0 (write recovery) */
442 static u32 ddr3_write_recovery(u32 clk)
443 {
444 	u32 twr_ns = 15; /* DDR3 spec says that it is 15ns for all speed bins */
445 	u32 twr_ck = (twr_ns * clk + 999) / 1000;
446 	if (twr_ck < 5)
447 		return 1;
448 	else if (twr_ck <= 8)
449 		return twr_ck - 4;
450 	else if (twr_ck <= 10)
451 		return 5;
452 	else
453 		return 6;
454 }
455 
456 /*
457  * If the dram->ppwrsctl (SDR_DPCR) register has the lowest bit set to 1, this
458  * means that DRAM is currently in self-refresh mode and retaining the old
459  * data. Since we have no idea what to do in this situation yet, just set this
460  * register to 0 and initialize DRAM in the same way as on any normal reboot
461  * (discarding whatever was stored there).
462  *
463  * Note: on sun7i hardware, the highest 16 bits need to be set to 0x1651 magic
464  * value for this write operation to have any effect. On sun5i hadware this
465  * magic value is not necessary. And on sun4i hardware the writes to this
466  * register seem to have no effect at all.
467  */
468 static void mctl_disable_power_save(void)
469 {
470 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
471 	writel(0x16510000, &dram->ppwrsctl);
472 }
473 
474 /*
475  * After the DRAM is powered up or reset, the DDR3 spec requires to wait at
476  * least 500 us before driving the CKE pin (Clock Enable) high. The dram->idct
477  * (SDR_IDCR) register appears to configure this delay, which gets applied
478  * right at the time when the DRAM initialization is activated in the
479  * 'mctl_ddr3_initialize' function.
480  */
481 static void mctl_set_cke_delay(void)
482 {
483 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
484 
485 	/* The CKE delay is represented in DRAM clock cycles, multiplied by N
486 	 * (where N=2 for sun4i/sun5i and N=3 for sun7i). Here it is set to
487 	 * the maximum possible value 0x1ffff, just like in the Allwinner's
488 	 * boot0 bootloader. The resulting delay value is somewhere between
489 	 * ~0.4 ms (sun5i with 648 MHz DRAM clock speed) and ~1.1 ms (sun7i
490 	 * with 360 MHz DRAM clock speed). */
491 	setbits_le32(&dram->idcr, 0x1ffff);
492 }
493 
494 /*
495  * This triggers the DRAM initialization. It performs sending the mode registers
496  * to the DRAM among other things. Very likely the ZQCL command is also getting
497  * executed (to do the initial impedance calibration on the DRAM side of the
498  * wire). The memory controller and the PHY must be already configured before
499  * calling this function.
500  */
501 static void mctl_ddr3_initialize(void)
502 {
503 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
504 	setbits_le32(&dram->ccr, DRAM_CCR_INIT);
505 	await_bits_clear(&dram->ccr, DRAM_CCR_INIT);
506 }
507 
508 /*
509  * Perform impedance calibration on the DRAM controller side of the wire.
510  */
511 static void mctl_set_impedance(u32 zq, bool odt_en)
512 {
513 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
514 	u32 reg_val;
515 	u32 zprog = zq & 0xFF, zdata = (zq >> 8) & 0xFFFFF;
516 
517 #ifndef CONFIG_MACH_SUN7I
518 	/* Appears that some kind of automatically initiated default
519 	 * ZQ calibration is already in progress at this point on sun4i/sun5i
520 	 * hardware, but not on sun7i. So it is reasonable to wait for its
521 	 * completion before doing anything else. */
522 	await_bits_set(&dram->zqsr, DRAM_ZQSR_ZDONE);
523 #endif
524 
525 	/* ZQ calibration is not really useful unless ODT is enabled */
526 	if (!odt_en)
527 		return;
528 
529 #ifdef CONFIG_MACH_SUN7I
530 	/* Enabling ODT in SDR_IOCR on sun7i hardware results in a deadlock
531 	 * unless bit 24 is set in SDR_ZQCR1. Not much is known about the
532 	 * SDR_ZQCR1 register, but there are hints indicating that it might
533 	 * be related to periodic impedance re-calibration. This particular
534 	 * magic value is borrowed from the Allwinner boot0 bootloader, and
535 	 * using it helps to avoid troubles */
536 	writel((1 << 24) | (1 << 1), &dram->zqcr1);
537 #endif
538 
539 	/* Needed at least for sun5i, because it does not self clear there */
540 	clrbits_le32(&dram->zqcr0, DRAM_ZQCR0_ZCAL);
541 
542 	if (zdata) {
543 		/* Set the user supplied impedance data */
544 		reg_val = DRAM_ZQCR0_ZDEN | zdata;
545 		writel(reg_val, &dram->zqcr0);
546 		/* no need to wait, this takes effect immediately */
547 	} else {
548 		/* Do the calibration using the external resistor */
549 		reg_val = DRAM_ZQCR0_ZCAL | DRAM_ZQCR0_IMP_DIV(zprog);
550 		writel(reg_val, &dram->zqcr0);
551 		/* Wait for the new impedance configuration to settle */
552 		await_bits_set(&dram->zqsr, DRAM_ZQSR_ZDONE);
553 	}
554 
555 	/* Needed at least for sun5i, because it does not self clear there */
556 	clrbits_le32(&dram->zqcr0, DRAM_ZQCR0_ZCAL);
557 
558 	/* Set I/O configure register */
559 	writel(DRAM_IOCR_ODT_EN, &dram->iocr);
560 }
561 
562 static unsigned long dramc_init_helper(struct dram_para *para)
563 {
564 	struct sunxi_dram_reg *dram = (struct sunxi_dram_reg *)SUNXI_DRAMC_BASE;
565 	u32 reg_val;
566 	u32 density;
567 	int ret_val;
568 
569 	/*
570 	 * only single rank DDR3 is supported by this code even though the
571 	 * hardware can theoretically support DDR2 and up to two ranks
572 	 */
573 	if (para->type != DRAM_MEMORY_TYPE_DDR3 || para->rank_num != 1)
574 		return 0;
575 
576 	/* setup DRAM relative clock */
577 	mctl_setup_dram_clock(para->clock, para->mbus_clock);
578 
579 	/* Disable any pad power save control */
580 	mctl_disable_power_save();
581 
582 	mctl_set_drive();
583 
584 	/* dram clock off */
585 	dramc_clock_output_en(0);
586 
587 #ifdef CONFIG_MACH_SUN4I
588 	/* select dram controller 1 */
589 	writel(DRAM_CSEL_MAGIC, &dram->csel);
590 #endif
591 
592 	mctl_itm_disable();
593 	mctl_enable_dll0(para->tpr3);
594 
595 	/* configure external DRAM */
596 	reg_val = DRAM_DCR_TYPE_DDR3;
597 	reg_val |= DRAM_DCR_IO_WIDTH(para->io_width >> 3);
598 
599 	if (para->density == 256)
600 		density = DRAM_DCR_CHIP_DENSITY_256M;
601 	else if (para->density == 512)
602 		density = DRAM_DCR_CHIP_DENSITY_512M;
603 	else if (para->density == 1024)
604 		density = DRAM_DCR_CHIP_DENSITY_1024M;
605 	else if (para->density == 2048)
606 		density = DRAM_DCR_CHIP_DENSITY_2048M;
607 	else if (para->density == 4096)
608 		density = DRAM_DCR_CHIP_DENSITY_4096M;
609 	else if (para->density == 8192)
610 		density = DRAM_DCR_CHIP_DENSITY_8192M;
611 	else
612 		density = DRAM_DCR_CHIP_DENSITY_256M;
613 
614 	reg_val |= DRAM_DCR_CHIP_DENSITY(density);
615 	reg_val |= DRAM_DCR_BUS_WIDTH((para->bus_width >> 3) - 1);
616 	reg_val |= DRAM_DCR_RANK_SEL(para->rank_num - 1);
617 	reg_val |= DRAM_DCR_CMD_RANK_ALL;
618 	reg_val |= DRAM_DCR_MODE(DRAM_DCR_MODE_INTERLEAVE);
619 	writel(reg_val, &dram->dcr);
620 
621 	dramc_clock_output_en(1);
622 
623 	mctl_set_impedance(para->zq, para->odt_en);
624 
625 	mctl_set_cke_delay();
626 
627 	mctl_ddr3_reset();
628 
629 	udelay(1);
630 
631 	await_bits_clear(&dram->ccr, DRAM_CCR_INIT);
632 
633 	mctl_enable_dllx(para->tpr3);
634 
635 	/* set refresh period */
636 	dramc_set_autorefresh_cycle(para->clock, density);
637 
638 	/* set timing parameters */
639 	writel(para->tpr0, &dram->tpr0);
640 	writel(para->tpr1, &dram->tpr1);
641 	writel(para->tpr2, &dram->tpr2);
642 
643 	reg_val = DRAM_MR_BURST_LENGTH(0x0);
644 #if (defined(CONFIG_MACH_SUN5I) || defined(CONFIG_MACH_SUN7I))
645 	reg_val |= DRAM_MR_POWER_DOWN;
646 #endif
647 	reg_val |= DRAM_MR_CAS_LAT(para->cas - 4);
648 	reg_val |= DRAM_MR_WRITE_RECOVERY(ddr3_write_recovery(para->clock));
649 	writel(reg_val, &dram->mr);
650 
651 	writel(para->emr1, &dram->emr);
652 	writel(para->emr2, &dram->emr2);
653 	writel(para->emr3, &dram->emr3);
654 
655 	/* disable drift compensation and set passive DQS window mode */
656 	clrsetbits_le32(&dram->ccr, DRAM_CCR_DQS_DRIFT_COMP, DRAM_CCR_DQS_GATE);
657 
658 #ifdef CONFIG_MACH_SUN7I
659 	/* Command rate timing mode 2T & 1T */
660 	if (para->tpr4 & 0x1)
661 		setbits_le32(&dram->ccr, DRAM_CCR_COMMAND_RATE_1T);
662 #endif
663 	/* initialize external DRAM */
664 	mctl_ddr3_initialize();
665 
666 	/* scan read pipe value */
667 	mctl_itm_enable();
668 
669 	/* Hardware DQS gate training */
670 	ret_val = dramc_scan_readpipe();
671 
672 	if (ret_val < 0)
673 		return 0;
674 
675 	/* allow to override the DQS training results with a custom delay */
676 	if (para->dqs_gating_delay)
677 		mctl_set_dqs_gating_delay(0, para->dqs_gating_delay);
678 
679 	/* set the DQS gating window type */
680 	if (para->active_windowing)
681 		clrbits_le32(&dram->ccr, DRAM_CCR_DQS_GATE);
682 	else
683 		setbits_le32(&dram->ccr, DRAM_CCR_DQS_GATE);
684 
685 	mctl_itm_reset();
686 
687 	/* configure all host port */
688 	mctl_configure_hostport();
689 
690 	return get_ram_size((long *)PHYS_SDRAM_0, PHYS_SDRAM_0_SIZE);
691 }
692 
693 unsigned long dramc_init(struct dram_para *para)
694 {
695 	unsigned long dram_size, actual_density;
696 
697 	/* If the dram configuration is not provided, use a default */
698 	if (!para)
699 		return 0;
700 
701 	/* if everything is known, then autodetection is not necessary */
702 	if (para->io_width && para->bus_width && para->density)
703 		return dramc_init_helper(para);
704 
705 	/* try to autodetect the DRAM bus width and density */
706 	para->io_width  = 16;
707 	para->bus_width = 32;
708 #if defined(CONFIG_MACH_SUN4I) || defined(CONFIG_MACH_SUN5I)
709 	/* only A0-A14 address lines on A10/A13, limiting max density to 4096 */
710 	para->density = 4096;
711 #else
712 	/* all A0-A15 address lines on A20, which allow density 8192 */
713 	para->density = 8192;
714 #endif
715 
716 	dram_size = dramc_init_helper(para);
717 	if (!dram_size) {
718 		/* if 32-bit bus width failed, try 16-bit bus width instead */
719 		para->bus_width = 16;
720 		dram_size = dramc_init_helper(para);
721 		if (!dram_size) {
722 			/* if 16-bit bus width also failed, then bail out */
723 			return dram_size;
724 		}
725 	}
726 
727 	/* check if we need to adjust the density */
728 	actual_density = (dram_size >> 17) * para->io_width / para->bus_width;
729 
730 	if (actual_density != para->density) {
731 		/* update the density and re-initialize DRAM again */
732 		para->density = actual_density;
733 		dram_size = dramc_init_helper(para);
734 	}
735 
736 	return dram_size;
737 }
738