xref: /openbmc/u-boot/arch/x86/cpu/quark/smc.c (revision 9c0e2f6e)
1 /*
2  * Copyright (C) 2013, Intel Corporation
3  * Copyright (C) 2015, Bin Meng <bmeng.cn@gmail.com>
4  *
5  * Ported from Intel released Quark UEFI BIOS
6  * QuarkSocPkg/QuarkNorthCluster/MemoryInit/Pei
7  *
8  * SPDX-License-Identifier:	Intel
9  */
10 
11 #include <common.h>
12 #include <pci.h>
13 #include <asm/arch/device.h>
14 #include <asm/arch/mrc.h>
15 #include <asm/arch/msg_port.h>
16 #include "mrc_util.h"
17 #include "hte.h"
18 #include "smc.h"
19 
20 /* t_ck clock period in picoseconds per speed index 800, 1066, 1333 */
21 static const uint32_t t_ck[3] = {
22 	2500,
23 	1875,
24 	1500
25 };
26 
27 /* Global variables */
28 static const uint16_t ddr_wclk[] = {193, 158};
29 #ifdef BACKUP_WCTL
30 static const uint16_t ddr_wctl[] = {1, 217};
31 #endif
32 #ifdef BACKUP_WCMD
33 static const uint16_t ddr_wcmd[] = {1, 220};
34 #endif
35 
36 #ifdef BACKUP_RCVN
37 static const uint16_t ddr_rcvn[] = {129, 498};
38 #endif
39 
40 #ifdef BACKUP_WDQS
41 static const uint16_t ddr_wdqs[] = {65, 289};
42 #endif
43 
44 #ifdef BACKUP_RDQS
45 static const uint8_t ddr_rdqs[] = {32, 24};
46 #endif
47 
48 #ifdef BACKUP_WDQ
49 static const uint16_t ddr_wdq[] = {32, 257};
50 #endif
51 
52 /* Stop self refresh driven by MCU */
53 void clear_self_refresh(struct mrc_params *mrc_params)
54 {
55 	ENTERFN();
56 
57 	/* clear the PMSTS Channel Self Refresh bits */
58 	mrc_write_mask(MEM_CTLR, PMSTS, PMSTS_DISR, PMSTS_DISR);
59 
60 	LEAVEFN();
61 }
62 
63 /* It will initialize timing registers in the MCU (DTR0..DTR4) */
64 void prog_ddr_timing_control(struct mrc_params *mrc_params)
65 {
66 	uint8_t tcl, wl;
67 	uint8_t trp, trcd, tras, twr, twtr, trrd, trtp, tfaw;
68 	uint32_t tck;
69 	u32 dtr0, dtr1, dtr2, dtr3, dtr4;
70 	u32 tmp1, tmp2;
71 
72 	ENTERFN();
73 
74 	/* mcu_init starts */
75 	mrc_post_code(0x02, 0x00);
76 
77 	dtr0 = msg_port_read(MEM_CTLR, DTR0);
78 	dtr1 = msg_port_read(MEM_CTLR, DTR1);
79 	dtr2 = msg_port_read(MEM_CTLR, DTR2);
80 	dtr3 = msg_port_read(MEM_CTLR, DTR3);
81 	dtr4 = msg_port_read(MEM_CTLR, DTR4);
82 
83 	tck = t_ck[mrc_params->ddr_speed];	/* Clock in picoseconds */
84 	tcl = mrc_params->params.cl;		/* CAS latency in clocks */
85 	trp = tcl;	/* Per CAT MRC */
86 	trcd = tcl;	/* Per CAT MRC */
87 	tras = MCEIL(mrc_params->params.ras, tck);
88 
89 	/* Per JEDEC: tWR=15000ps DDR2/3 from 800-1600 */
90 	twr = MCEIL(15000, tck);
91 
92 	twtr = MCEIL(mrc_params->params.wtr, tck);
93 	trrd = MCEIL(mrc_params->params.rrd, tck);
94 	trtp = 4;	/* Valid for 800 and 1066, use 5 for 1333 */
95 	tfaw = MCEIL(mrc_params->params.faw, tck);
96 
97 	wl = 5 + mrc_params->ddr_speed;
98 
99 	dtr0 &= ~DTR0_DFREQ_MASK;
100 	dtr0 |= mrc_params->ddr_speed;
101 	dtr0 &= ~DTR0_TCL_MASK;
102 	tmp1 = tcl - 5;
103 	dtr0 |= ((tcl - 5) << 12);
104 	dtr0 &= ~DTR0_TRP_MASK;
105 	dtr0 |= ((trp - 5) << 4);	/* 5 bit DRAM Clock */
106 	dtr0 &= ~DTR0_TRCD_MASK;
107 	dtr0 |= ((trcd - 5) << 8);	/* 5 bit DRAM Clock */
108 
109 	dtr1 &= ~DTR1_TWCL_MASK;
110 	tmp2 = wl - 3;
111 	dtr1 |= (wl - 3);
112 	dtr1 &= ~DTR1_TWTP_MASK;
113 	dtr1 |= ((wl + 4 + twr - 14) << 8);	/* Change to tWTP */
114 	dtr1 &= ~DTR1_TRTP_MASK;
115 	dtr1 |= ((MMAX(trtp, 4) - 3) << 28);	/* 4 bit DRAM Clock */
116 	dtr1 &= ~DTR1_TRRD_MASK;
117 	dtr1 |= ((trrd - 4) << 24);		/* 4 bit DRAM Clock */
118 	dtr1 &= ~DTR1_TCMD_MASK;
119 	dtr1 |= (1 << 4);
120 	dtr1 &= ~DTR1_TRAS_MASK;
121 	dtr1 |= ((tras - 14) << 20);		/* 6 bit DRAM Clock */
122 	dtr1 &= ~DTR1_TFAW_MASK;
123 	dtr1 |= ((((tfaw + 1) >> 1) - 5) << 16);/* 4 bit DRAM Clock */
124 	/* Set 4 Clock CAS to CAS delay (multi-burst) */
125 	dtr1 &= ~DTR1_TCCD_MASK;
126 
127 	dtr2 &= ~DTR2_TRRDR_MASK;
128 	dtr2 |= 1;
129 	dtr2 &= ~DTR2_TWWDR_MASK;
130 	dtr2 |= (2 << 8);
131 	dtr2 &= ~DTR2_TRWDR_MASK;
132 	dtr2 |= (2 << 16);
133 
134 	dtr3 &= ~DTR3_TWRDR_MASK;
135 	dtr3 |= 2;
136 	dtr3 &= ~DTR3_TXXXX_MASK;
137 	dtr3 |= (2 << 4);
138 
139 	dtr3 &= ~DTR3_TRWSR_MASK;
140 	if (mrc_params->ddr_speed == DDRFREQ_800) {
141 		/* Extended RW delay (+1) */
142 		dtr3 |= ((tcl - 5 + 1) << 8);
143 	} else if (mrc_params->ddr_speed == DDRFREQ_1066) {
144 		/* Extended RW delay (+1) */
145 		dtr3 |= ((tcl - 5 + 1) << 8);
146 	}
147 
148 	dtr3 &= ~DTR3_TWRSR_MASK;
149 	dtr3 |= ((4 + wl + twtr - 11) << 13);
150 
151 	dtr3 &= ~DTR3_TXP_MASK;
152 	if (mrc_params->ddr_speed == DDRFREQ_800)
153 		dtr3 |= ((MMAX(0, 1 - 1)) << 22);
154 	else
155 		dtr3 |= ((MMAX(0, 2 - 1)) << 22);
156 
157 	dtr4 &= ~DTR4_WRODTSTRT_MASK;
158 	dtr4 |= 1;
159 	dtr4 &= ~DTR4_WRODTSTOP_MASK;
160 	dtr4 |= (1 << 4);
161 	dtr4 &= ~DTR4_XXXX1_MASK;
162 	dtr4 |= ((1 + tmp1 - tmp2 + 2) << 8);
163 	dtr4 &= ~DTR4_XXXX2_MASK;
164 	dtr4 |= ((1 + tmp1 - tmp2 + 2) << 12);
165 	dtr4 &= ~(DTR4_ODTDIS | DTR4_TRGSTRDIS);
166 
167 	msg_port_write(MEM_CTLR, DTR0, dtr0);
168 	msg_port_write(MEM_CTLR, DTR1, dtr1);
169 	msg_port_write(MEM_CTLR, DTR2, dtr2);
170 	msg_port_write(MEM_CTLR, DTR3, dtr3);
171 	msg_port_write(MEM_CTLR, DTR4, dtr4);
172 
173 	LEAVEFN();
174 }
175 
176 /* Configure MCU before jedec init sequence */
177 void prog_decode_before_jedec(struct mrc_params *mrc_params)
178 {
179 	u32 drp;
180 	u32 drfc;
181 	u32 dcal;
182 	u32 dsch;
183 	u32 dpmc0;
184 
185 	ENTERFN();
186 
187 	/* Disable power saving features */
188 	dpmc0 = msg_port_read(MEM_CTLR, DPMC0);
189 	dpmc0 |= (DPMC0_CLKGTDIS | DPMC0_DISPWRDN);
190 	dpmc0 &= ~DPMC0_PCLSTO_MASK;
191 	dpmc0 &= ~DPMC0_DYNSREN;
192 	msg_port_write(MEM_CTLR, DPMC0, dpmc0);
193 
194 	/* Disable out of order transactions */
195 	dsch = msg_port_read(MEM_CTLR, DSCH);
196 	dsch |= (DSCH_OOODIS | DSCH_NEWBYPDIS);
197 	msg_port_write(MEM_CTLR, DSCH, dsch);
198 
199 	/* Disable issuing the REF command */
200 	drfc = msg_port_read(MEM_CTLR, DRFC);
201 	drfc &= ~DRFC_TREFI_MASK;
202 	msg_port_write(MEM_CTLR, DRFC, drfc);
203 
204 	/* Disable ZQ calibration short */
205 	dcal = msg_port_read(MEM_CTLR, DCAL);
206 	dcal &= ~DCAL_ZQCINT_MASK;
207 	dcal &= ~DCAL_SRXZQCL_MASK;
208 	msg_port_write(MEM_CTLR, DCAL, dcal);
209 
210 	/*
211 	 * Training performed in address mode 0, rank population has limited
212 	 * impact, however simulator complains if enabled non-existing rank.
213 	 */
214 	drp = 0;
215 	if (mrc_params->rank_enables & 1)
216 		drp |= DRP_RKEN0;
217 	if (mrc_params->rank_enables & 2)
218 		drp |= DRP_RKEN1;
219 	msg_port_write(MEM_CTLR, DRP, drp);
220 
221 	LEAVEFN();
222 }
223 
224 /*
225  * After Cold Reset, BIOS should set COLDWAKE bit to 1 before
226  * sending the WAKE message to the Dunit.
227  *
228  * For Standby Exit, or any other mode in which the DRAM is in
229  * SR, this bit must be set to 0.
230  */
231 void perform_ddr_reset(struct mrc_params *mrc_params)
232 {
233 	ENTERFN();
234 
235 	/* Set COLDWAKE bit before sending the WAKE message */
236 	mrc_write_mask(MEM_CTLR, DRMC, DRMC_COLDWAKE, DRMC_COLDWAKE);
237 
238 	/* Send wake command to DUNIT (MUST be done before JEDEC) */
239 	dram_wake_command();
240 
241 	/* Set default value */
242 	msg_port_write(MEM_CTLR, DRMC,
243 		       mrc_params->rd_odt_value == 0 ? DRMC_ODTMODE : 0);
244 
245 	LEAVEFN();
246 }
247 
248 
249 /*
250  * This function performs some initialization on the DDRIO unit.
251  * This function is dependent on BOARD_ID, DDR_SPEED, and CHANNEL_ENABLES.
252  */
253 void ddrphy_init(struct mrc_params *mrc_params)
254 {
255 	uint32_t temp;
256 	uint8_t ch;	/* channel counter */
257 	uint8_t rk;	/* rank counter */
258 	uint8_t bl_grp;	/*  byte lane group counter (2 BLs per module) */
259 	uint8_t bl_divisor = 1;	/* byte lane divisor */
260 	/* For DDR3 --> 0 == 800, 1 == 1066, 2 == 1333 */
261 	uint8_t speed = mrc_params->ddr_speed & 3;
262 	uint8_t cas;
263 	uint8_t cwl;
264 
265 	ENTERFN();
266 
267 	cas = mrc_params->params.cl;
268 	cwl = 5 + mrc_params->ddr_speed;
269 
270 	/* ddrphy_init starts */
271 	mrc_post_code(0x03, 0x00);
272 
273 	/*
274 	 * HSD#231531
275 	 * Make sure IOBUFACT is deasserted before initializing the DDR PHY
276 	 *
277 	 * HSD#234845
278 	 * Make sure WRPTRENABLE is deasserted before initializing the DDR PHY
279 	 */
280 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
281 		if (mrc_params->channel_enables & (1 << ch)) {
282 			/* Deassert DDRPHY Initialization Complete */
283 			mrc_alt_write_mask(DDRPHY,
284 				CMDPMCONFIG0 + ch * DDRIOCCC_CH_OFFSET,
285 				~(1 << 20), 1 << 20);	/* SPID_INIT_COMPLETE=0 */
286 			/* Deassert IOBUFACT */
287 			mrc_alt_write_mask(DDRPHY,
288 				CMDCFGREG0 + ch * DDRIOCCC_CH_OFFSET,
289 				~(1 << 2), 1 << 2);	/* IOBUFACTRST_N=0 */
290 			/* Disable WRPTR */
291 			mrc_alt_write_mask(DDRPHY,
292 				CMDPTRREG + ch * DDRIOCCC_CH_OFFSET,
293 				~(1 << 0), 1 << 0);	/* WRPTRENABLE=0 */
294 		}
295 	}
296 
297 	/* Put PHY in reset */
298 	mrc_alt_write_mask(DDRPHY, MASTERRSTN, 0, 1);
299 
300 	/* Initialize DQ01, DQ23, CMD, CLK-CTL, COMP modules */
301 
302 	/* STEP0 */
303 	mrc_post_code(0x03, 0x10);
304 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
305 		if (mrc_params->channel_enables & (1 << ch)) {
306 			/* DQ01-DQ23 */
307 			for (bl_grp = 0;
308 			     bl_grp < (NUM_BYTE_LANES / bl_divisor) / 2;
309 			     bl_grp++) {
310 				/* Analog MUX select - IO2xCLKSEL */
311 				mrc_alt_write_mask(DDRPHY,
312 					DQOBSCKEBBCTL +
313 					bl_grp * DDRIODQ_BL_OFFSET +
314 					ch * DDRIODQ_CH_OFFSET,
315 					bl_grp ? 0 : (1 << 22), 1 << 22);
316 
317 				/* ODT Strength */
318 				switch (mrc_params->rd_odt_value) {
319 				case 1:
320 					temp = 0x3;
321 					break;	/* 60 ohm */
322 				case 2:
323 					temp = 0x3;
324 					break;	/* 120 ohm */
325 				case 3:
326 					temp = 0x3;
327 					break;	/* 180 ohm */
328 				default:
329 					temp = 0x3;
330 					break;	/* 120 ohm */
331 				}
332 
333 				/* ODT strength */
334 				mrc_alt_write_mask(DDRPHY,
335 					B0RXIOBUFCTL +
336 					bl_grp * DDRIODQ_BL_OFFSET +
337 					ch * DDRIODQ_CH_OFFSET,
338 					temp << 5, 0x60);
339 				/* ODT strength */
340 				mrc_alt_write_mask(DDRPHY,
341 					B1RXIOBUFCTL +
342 					bl_grp * DDRIODQ_BL_OFFSET +
343 					ch * DDRIODQ_CH_OFFSET,
344 					temp << 5, 0x60);
345 
346 				/* Dynamic ODT/DIFFAMP */
347 				temp = (cas << 24) | (cas << 16) |
348 					(cas << 8) | (cas << 0);
349 				switch (speed) {
350 				case 0:
351 					temp -= 0x01010101;
352 					break;	/* 800 */
353 				case 1:
354 					temp -= 0x02020202;
355 					break;	/* 1066 */
356 				case 2:
357 					temp -= 0x03030303;
358 					break;	/* 1333 */
359 				case 3:
360 					temp -= 0x04040404;
361 					break;	/* 1600 */
362 				}
363 
364 				/* Launch Time: ODT, DIFFAMP, ODT, DIFFAMP */
365 				mrc_alt_write_mask(DDRPHY,
366 					B01LATCTL1 +
367 					bl_grp * DDRIODQ_BL_OFFSET +
368 					ch * DDRIODQ_CH_OFFSET,
369 					temp, 0x1f1f1f1f);
370 				switch (speed) {
371 				/* HSD#234715 */
372 				case 0:
373 					temp = (0x06 << 16) | (0x07 << 8);
374 					break;	/* 800 */
375 				case 1:
376 					temp = (0x07 << 16) | (0x08 << 8);
377 					break;	/* 1066 */
378 				case 2:
379 					temp = (0x09 << 16) | (0x0a << 8);
380 					break;	/* 1333 */
381 				case 3:
382 					temp = (0x0a << 16) | (0x0b << 8);
383 					break;	/* 1600 */
384 				}
385 
386 				/* On Duration: ODT, DIFFAMP */
387 				mrc_alt_write_mask(DDRPHY,
388 					B0ONDURCTL +
389 					bl_grp * DDRIODQ_BL_OFFSET +
390 					ch * DDRIODQ_CH_OFFSET,
391 					temp, 0x003f3f00);
392 				/* On Duration: ODT, DIFFAMP */
393 				mrc_alt_write_mask(DDRPHY,
394 					B1ONDURCTL +
395 					bl_grp * DDRIODQ_BL_OFFSET +
396 					ch * DDRIODQ_CH_OFFSET,
397 					temp, 0x003f3f00);
398 
399 				switch (mrc_params->rd_odt_value) {
400 				case 0:
401 					/* override DIFFAMP=on, ODT=off */
402 					temp = (0x3f << 16) | (0x3f << 10);
403 					break;
404 				default:
405 					/* override DIFFAMP=on, ODT=on */
406 					temp = (0x3f << 16) | (0x2a << 10);
407 					break;
408 				}
409 
410 				/* Override: DIFFAMP, ODT */
411 				mrc_alt_write_mask(DDRPHY,
412 					B0OVRCTL +
413 					bl_grp * DDRIODQ_BL_OFFSET +
414 					ch * DDRIODQ_CH_OFFSET,
415 					temp, 0x003ffc00);
416 				/* Override: DIFFAMP, ODT */
417 				mrc_alt_write_mask(DDRPHY,
418 					B1OVRCTL +
419 					bl_grp * DDRIODQ_BL_OFFSET +
420 					ch * DDRIODQ_CH_OFFSET,
421 					temp, 0x003ffc00);
422 
423 				/* DLL Setup */
424 
425 				/* 1xCLK Domain Timings: tEDP,RCVEN,WDQS (PO) */
426 				mrc_alt_write_mask(DDRPHY,
427 					B0LATCTL0 +
428 					bl_grp * DDRIODQ_BL_OFFSET +
429 					ch * DDRIODQ_CH_OFFSET,
430 					((cas + 7) << 16) | ((cas - 4) << 8) |
431 					((cwl - 2) << 0), 0x003f1f1f);
432 				mrc_alt_write_mask(DDRPHY,
433 					B1LATCTL0 +
434 					bl_grp * DDRIODQ_BL_OFFSET +
435 					ch * DDRIODQ_CH_OFFSET,
436 					((cas + 7) << 16) | ((cas - 4) << 8) |
437 					((cwl - 2) << 0), 0x003f1f1f);
438 
439 				/* RCVEN Bypass (PO) */
440 				mrc_alt_write_mask(DDRPHY,
441 					B0RXIOBUFCTL +
442 					bl_grp * DDRIODQ_BL_OFFSET +
443 					ch * DDRIODQ_CH_OFFSET,
444 					0, 0x81);
445 				mrc_alt_write_mask(DDRPHY,
446 					B1RXIOBUFCTL +
447 					bl_grp * DDRIODQ_BL_OFFSET +
448 					ch * DDRIODQ_CH_OFFSET,
449 					0, 0x81);
450 
451 				/* TX */
452 				mrc_alt_write_mask(DDRPHY,
453 					DQCTL +
454 					bl_grp * DDRIODQ_BL_OFFSET +
455 					ch * DDRIODQ_CH_OFFSET,
456 					1 << 16, 1 << 16);
457 				mrc_alt_write_mask(DDRPHY,
458 					B01PTRCTL1 +
459 					bl_grp * DDRIODQ_BL_OFFSET +
460 					ch * DDRIODQ_CH_OFFSET,
461 					1 << 8, 1 << 8);
462 
463 				/* RX (PO) */
464 				/* Internal Vref Code, Enable#, Ext_or_Int (1=Ext) */
465 				mrc_alt_write_mask(DDRPHY,
466 					B0VREFCTL +
467 					bl_grp * DDRIODQ_BL_OFFSET +
468 					ch * DDRIODQ_CH_OFFSET,
469 					(0x03 << 2) | (0x0 << 1) | (0x0 << 0),
470 					0xff);
471 				/* Internal Vref Code, Enable#, Ext_or_Int (1=Ext) */
472 				mrc_alt_write_mask(DDRPHY,
473 					B1VREFCTL +
474 					bl_grp * DDRIODQ_BL_OFFSET +
475 					ch * DDRIODQ_CH_OFFSET,
476 					(0x03 << 2) | (0x0 << 1) | (0x0 << 0),
477 					0xff);
478 				/* Per-Bit De-Skew Enable */
479 				mrc_alt_write_mask(DDRPHY,
480 					B0RXIOBUFCTL +
481 					bl_grp * DDRIODQ_BL_OFFSET +
482 					ch * DDRIODQ_CH_OFFSET,
483 					0, 0x10);
484 				/* Per-Bit De-Skew Enable */
485 				mrc_alt_write_mask(DDRPHY,
486 					B1RXIOBUFCTL +
487 					bl_grp * DDRIODQ_BL_OFFSET +
488 					ch * DDRIODQ_CH_OFFSET,
489 					0, 0x10);
490 			}
491 
492 			/* CLKEBB */
493 			mrc_alt_write_mask(DDRPHY,
494 				CMDOBSCKEBBCTL + ch * DDRIOCCC_CH_OFFSET,
495 				0, 1 << 23);
496 
497 			/* Enable tristate control of cmd/address bus */
498 			mrc_alt_write_mask(DDRPHY,
499 				CMDCFGREG0 + ch * DDRIOCCC_CH_OFFSET,
500 				0, 0x03);
501 
502 			/* ODT RCOMP */
503 			mrc_alt_write_mask(DDRPHY,
504 				CMDRCOMPODT + ch * DDRIOCCC_CH_OFFSET,
505 				(0x03 << 5) | (0x03 << 0), 0x3ff);
506 
507 			/* CMDPM* registers must be programmed in this order */
508 
509 			/* Turn On Delays: SFR (regulator), MPLL */
510 			mrc_alt_write_mask(DDRPHY,
511 				CMDPMDLYREG4 + ch * DDRIOCCC_CH_OFFSET,
512 				0xffffffff, 0xffffffff);
513 			/*
514 			 * Delays: ASSERT_IOBUFACT_to_ALLON0_for_PM_MSG_3,
515 			 * VREG (MDLL) Turn On, ALLON0_to_DEASSERT_IOBUFACT
516 			 * for_PM_MSG_gt0, MDLL Turn On
517 			 */
518 			mrc_alt_write_mask(DDRPHY,
519 				CMDPMDLYREG3 + ch * DDRIOCCC_CH_OFFSET,
520 				0xfffff616, 0xffffffff);
521 			/* MPLL Divider Reset Delays */
522 			mrc_alt_write_mask(DDRPHY,
523 				CMDPMDLYREG2 + ch * DDRIOCCC_CH_OFFSET,
524 				0xffffffff, 0xffffffff);
525 			/* Turn Off Delays: VREG, Staggered MDLL, MDLL, PI */
526 			mrc_alt_write_mask(DDRPHY,
527 				CMDPMDLYREG1 + ch * DDRIOCCC_CH_OFFSET,
528 				0xffffffff, 0xffffffff);
529 			/* Turn On Delays: MPLL, Staggered MDLL, PI, IOBUFACT */
530 			mrc_alt_write_mask(DDRPHY,
531 				CMDPMDLYREG0 + ch * DDRIOCCC_CH_OFFSET,
532 				0xffffffff, 0xffffffff);
533 			/* Allow PUnit signals */
534 			mrc_alt_write_mask(DDRPHY,
535 				CMDPMCONFIG0 + ch * DDRIOCCC_CH_OFFSET,
536 				(0x6 << 8) | (0x1 << 6) | (0x4 << 0),
537 				0xffe00f4f);
538 			/* DLL_VREG Bias Trim, VREF Tuning for DLL_VREG */
539 			mrc_alt_write_mask(DDRPHY,
540 				CMDMDLLCTL + ch * DDRIOCCC_CH_OFFSET,
541 				(0x3 << 4) | (0x7 << 0), 0x7f);
542 
543 			/* CLK-CTL */
544 			mrc_alt_write_mask(DDRPHY,
545 				CCOBSCKEBBCTL + ch * DDRIOCCC_CH_OFFSET,
546 				0, 1 << 24);	/* CLKEBB */
547 			/* Buffer Enable: CS,CKE,ODT,CLK */
548 			mrc_alt_write_mask(DDRPHY,
549 				CCCFGREG0 + ch * DDRIOCCC_CH_OFFSET,
550 				0x1f, 0x000ffff1);
551 			/* ODT RCOMP */
552 			mrc_alt_write_mask(DDRPHY,
553 				CCRCOMPODT + ch * DDRIOCCC_CH_OFFSET,
554 				(0x03 << 8) | (0x03 << 0), 0x00001f1f);
555 			/* DLL_VREG Bias Trim, VREF Tuning for DLL_VREG */
556 			mrc_alt_write_mask(DDRPHY,
557 				CCMDLLCTL + ch * DDRIOCCC_CH_OFFSET,
558 				(0x3 << 4) | (0x7 << 0), 0x7f);
559 
560 			/*
561 			 * COMP (RON channel specific)
562 			 * - DQ/DQS/DM RON: 32 Ohm
563 			 * - CTRL/CMD RON: 27 Ohm
564 			 * - CLK RON: 26 Ohm
565 			 */
566 			/* RCOMP Vref PU/PD */
567 			mrc_alt_write_mask(DDRPHY,
568 				DQVREFCH0 +  ch * DDRCOMP_CH_OFFSET,
569 				(0x08 << 24) | (0x03 << 16), 0x3f3f0000);
570 			/* RCOMP Vref PU/PD */
571 			mrc_alt_write_mask(DDRPHY,
572 				CMDVREFCH0 + ch * DDRCOMP_CH_OFFSET,
573 				(0x0C << 24) | (0x03 << 16), 0x3f3f0000);
574 			/* RCOMP Vref PU/PD */
575 			mrc_alt_write_mask(DDRPHY,
576 				CLKVREFCH0 + ch * DDRCOMP_CH_OFFSET,
577 				(0x0F << 24) | (0x03 << 16), 0x3f3f0000);
578 			/* RCOMP Vref PU/PD */
579 			mrc_alt_write_mask(DDRPHY,
580 				DQSVREFCH0 + ch * DDRCOMP_CH_OFFSET,
581 				(0x08 << 24) | (0x03 << 16), 0x3f3f0000);
582 			/* RCOMP Vref PU/PD */
583 			mrc_alt_write_mask(DDRPHY,
584 				CTLVREFCH0 + ch * DDRCOMP_CH_OFFSET,
585 				(0x0C << 24) | (0x03 << 16), 0x3f3f0000);
586 
587 			/* DQS Swapped Input Enable */
588 			mrc_alt_write_mask(DDRPHY,
589 				COMPEN1CH0 + ch * DDRCOMP_CH_OFFSET,
590 				(1 << 19) | (1 << 17), 0xc00ac000);
591 
592 			/* ODT VREF = 1.5 x 274/360+274 = 0.65V (code of ~50) */
593 			/* ODT Vref PU/PD */
594 			mrc_alt_write_mask(DDRPHY,
595 				DQVREFCH0 + ch * DDRCOMP_CH_OFFSET,
596 				(0x32 << 8) | (0x03 << 0), 0x00003f3f);
597 			/* ODT Vref PU/PD */
598 			mrc_alt_write_mask(DDRPHY,
599 				DQSVREFCH0 + ch * DDRCOMP_CH_OFFSET,
600 				(0x32 << 8) | (0x03 << 0), 0x00003f3f);
601 			/* ODT Vref PU/PD */
602 			mrc_alt_write_mask(DDRPHY,
603 				CLKVREFCH0 + ch * DDRCOMP_CH_OFFSET,
604 				(0x0E << 8) | (0x05 << 0), 0x00003f3f);
605 
606 			/*
607 			 * Slew rate settings are frequency specific,
608 			 * numbers below are for 800Mhz (speed == 0)
609 			 * - DQ/DQS/DM/CLK SR: 4V/ns,
610 			 * - CTRL/CMD SR: 1.5V/ns
611 			 */
612 			temp = (0x0e << 16) | (0x0e << 12) | (0x08 << 8) |
613 				(0x0b << 4) | (0x0b << 0);
614 			/* DCOMP Delay Select: CTL,CMD,CLK,DQS,DQ */
615 			mrc_alt_write_mask(DDRPHY,
616 				DLYSELCH0 + ch * DDRCOMP_CH_OFFSET,
617 				temp, 0x000fffff);
618 			/* TCO Vref CLK,DQS,DQ */
619 			mrc_alt_write_mask(DDRPHY,
620 				TCOVREFCH0 + ch * DDRCOMP_CH_OFFSET,
621 				(0x05 << 16) | (0x05 << 8) | (0x05 << 0),
622 				0x003f3f3f);
623 			/* ODTCOMP CMD/CTL PU/PD */
624 			mrc_alt_write_mask(DDRPHY,
625 				CCBUFODTCH0 + ch * DDRCOMP_CH_OFFSET,
626 				(0x03 << 8) | (0x03 << 0),
627 				0x00001f1f);
628 			/* COMP */
629 			mrc_alt_write_mask(DDRPHY,
630 				COMPEN0CH0 + ch * DDRCOMP_CH_OFFSET,
631 				0, 0xc0000100);
632 
633 #ifdef BACKUP_COMPS
634 			/* DQ COMP Overrides */
635 			/* RCOMP PU */
636 			mrc_alt_write_mask(DDRPHY,
637 				DQDRVPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
638 				(1 << 31) | (0x0a << 16),
639 				0x801f0000);
640 			/* RCOMP PD */
641 			mrc_alt_write_mask(DDRPHY,
642 				DQDRVPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
643 				(1 << 31) | (0x0a << 16),
644 				0x801f0000);
645 			/* DCOMP PU */
646 			mrc_alt_write_mask(DDRPHY,
647 				DQDLYPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
648 				(1 << 31) | (0x10 << 16),
649 				0x801f0000);
650 			/* DCOMP PD */
651 			mrc_alt_write_mask(DDRPHY,
652 				DQDLYPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
653 				(1 << 31) | (0x10 << 16),
654 				0x801f0000);
655 			/* ODTCOMP PU */
656 			mrc_alt_write_mask(DDRPHY,
657 				DQODTPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
658 				(1 << 31) | (0x0b << 16),
659 				0x801f0000);
660 			/* ODTCOMP PD */
661 			mrc_alt_write_mask(DDRPHY,
662 				DQODTPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
663 				(1 << 31) | (0x0b << 16),
664 				0x801f0000);
665 			/* TCOCOMP PU */
666 			mrc_alt_write_mask(DDRPHY,
667 				DQTCOPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
668 				1 << 31, 1 << 31);
669 			/* TCOCOMP PD */
670 			mrc_alt_write_mask(DDRPHY,
671 				DQTCOPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
672 				1 << 31, 1 << 31);
673 
674 			/* DQS COMP Overrides */
675 			/* RCOMP PU */
676 			mrc_alt_write_mask(DDRPHY,
677 				DQSDRVPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
678 				(1 << 31) | (0x0a << 16),
679 				0x801f0000);
680 			/* RCOMP PD */
681 			mrc_alt_write_mask(DDRPHY,
682 				DQSDRVPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
683 				(1 << 31) | (0x0a << 16),
684 				0x801f0000);
685 			/* DCOMP PU */
686 			mrc_alt_write_mask(DDRPHY,
687 				DQSDLYPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
688 				(1 << 31) | (0x10 << 16),
689 				0x801f0000);
690 			/* DCOMP PD */
691 			mrc_alt_write_mask(DDRPHY,
692 				DQSDLYPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
693 				(1 << 31) | (0x10 << 16),
694 				0x801f0000);
695 			/* ODTCOMP PU */
696 			mrc_alt_write_mask(DDRPHY,
697 				DQSODTPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
698 				(1 << 31) | (0x0b << 16),
699 				0x801f0000);
700 			/* ODTCOMP PD */
701 			mrc_alt_write_mask(DDRPHY,
702 				DQSODTPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
703 				(1 << 31) | (0x0b << 16),
704 				0x801f0000);
705 			/* TCOCOMP PU */
706 			mrc_alt_write_mask(DDRPHY,
707 				DQSTCOPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
708 				1 << 31, 1 << 31);
709 			/* TCOCOMP PD */
710 			mrc_alt_write_mask(DDRPHY,
711 				DQSTCOPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
712 				1 << 31, 1 << 31);
713 
714 			/* CLK COMP Overrides */
715 			/* RCOMP PU */
716 			mrc_alt_write_mask(DDRPHY,
717 				CLKDRVPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
718 				(1 << 31) | (0x0c << 16),
719 				0x801f0000);
720 			/* RCOMP PD */
721 			mrc_alt_write_mask(DDRPHY,
722 				CLKDRVPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
723 				(1 << 31) | (0x0c << 16),
724 				0x801f0000);
725 			/* DCOMP PU */
726 			mrc_alt_write_mask(DDRPHY,
727 				CLKDLYPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
728 				(1 << 31) | (0x07 << 16),
729 				0x801f0000);
730 			/* DCOMP PD */
731 			mrc_alt_write_mask(DDRPHY,
732 				CLKDLYPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
733 				(1 << 31) | (0x07 << 16),
734 				0x801f0000);
735 			/* ODTCOMP PU */
736 			mrc_alt_write_mask(DDRPHY,
737 				CLKODTPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
738 				(1 << 31) | (0x0b << 16),
739 				0x801f0000);
740 			/* ODTCOMP PD */
741 			mrc_alt_write_mask(DDRPHY,
742 				CLKODTPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
743 				(1 << 31) | (0x0b << 16),
744 				0x801f0000);
745 			/* TCOCOMP PU */
746 			mrc_alt_write_mask(DDRPHY,
747 				CLKTCOPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
748 				1 << 31, 1 << 31);
749 			/* TCOCOMP PD */
750 			mrc_alt_write_mask(DDRPHY,
751 				CLKTCOPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
752 				1 << 31, 1 << 31);
753 
754 			/* CMD COMP Overrides */
755 			/* RCOMP PU */
756 			mrc_alt_write_mask(DDRPHY,
757 				CMDDRVPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
758 				(1 << 31) | (0x0d << 16),
759 				0x803f0000);
760 			/* RCOMP PD */
761 			mrc_alt_write_mask(DDRPHY,
762 				CMDDRVPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
763 				(1 << 31) | (0x0d << 16),
764 				0x803f0000);
765 			/* DCOMP PU */
766 			mrc_alt_write_mask(DDRPHY,
767 				CMDDLYPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
768 				(1 << 31) | (0x0a << 16),
769 				0x801f0000);
770 			/* DCOMP PD */
771 			mrc_alt_write_mask(DDRPHY,
772 				CMDDLYPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
773 				(1 << 31) | (0x0a << 16),
774 				0x801f0000);
775 
776 			/* CTL COMP Overrides */
777 			/* RCOMP PU */
778 			mrc_alt_write_mask(DDRPHY,
779 				CTLDRVPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
780 				(1 << 31) | (0x0d << 16),
781 				0x803f0000);
782 			/* RCOMP PD */
783 			mrc_alt_write_mask(DDRPHY,
784 				CTLDRVPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
785 				(1 << 31) | (0x0d << 16),
786 				0x803f0000);
787 			/* DCOMP PU */
788 			mrc_alt_write_mask(DDRPHY,
789 				CTLDLYPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
790 				(1 << 31) | (0x0a << 16),
791 				0x801f0000);
792 			/* DCOMP PD */
793 			mrc_alt_write_mask(DDRPHY,
794 				CTLDLYPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
795 				(1 << 31) | (0x0a << 16),
796 				0x801f0000);
797 #else
798 			/* DQ TCOCOMP Overrides */
799 			/* TCOCOMP PU */
800 			mrc_alt_write_mask(DDRPHY,
801 				DQTCOPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
802 				(1 << 31) | (0x1f << 16),
803 				0x801f0000);
804 			/* TCOCOMP PD */
805 			mrc_alt_write_mask(DDRPHY,
806 				DQTCOPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
807 				(1 << 31) | (0x1f << 16),
808 				0x801f0000);
809 
810 			/* DQS TCOCOMP Overrides */
811 			/* TCOCOMP PU */
812 			mrc_alt_write_mask(DDRPHY,
813 				DQSTCOPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
814 				(1 << 31) | (0x1f << 16),
815 				0x801f0000);
816 			/* TCOCOMP PD */
817 			mrc_alt_write_mask(DDRPHY,
818 				DQSTCOPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
819 				(1 << 31) | (0x1f << 16),
820 				0x801f0000);
821 
822 			/* CLK TCOCOMP Overrides */
823 			/* TCOCOMP PU */
824 			mrc_alt_write_mask(DDRPHY,
825 				CLKTCOPUCTLCH0 + ch * DDRCOMP_CH_OFFSET,
826 				(1 << 31) | (0x1f << 16),
827 				0x801f0000);
828 			/* TCOCOMP PD */
829 			mrc_alt_write_mask(DDRPHY,
830 				CLKTCOPDCTLCH0 + ch * DDRCOMP_CH_OFFSET,
831 				(1 << 31) | (0x1f << 16),
832 				0x801f0000);
833 #endif
834 
835 			/* program STATIC delays */
836 #ifdef BACKUP_WCMD
837 			set_wcmd(ch, ddr_wcmd[PLATFORM_ID]);
838 #else
839 			set_wcmd(ch, ddr_wclk[PLATFORM_ID] + HALF_CLK);
840 #endif
841 
842 			for (rk = 0; rk < NUM_RANKS; rk++) {
843 				if (mrc_params->rank_enables & (1 << rk)) {
844 					set_wclk(ch, rk, ddr_wclk[PLATFORM_ID]);
845 #ifdef BACKUP_WCTL
846 					set_wctl(ch, rk, ddr_wctl[PLATFORM_ID]);
847 #else
848 					set_wctl(ch, rk, ddr_wclk[PLATFORM_ID] + HALF_CLK);
849 #endif
850 				}
851 			}
852 		}
853 	}
854 
855 	/* COMP (non channel specific) */
856 	/* RCOMP: Dither PU Enable */
857 	mrc_alt_write_mask(DDRPHY, DQANADRVPUCTL, 1 << 30, 1 << 30);
858 	/* RCOMP: Dither PD Enable */
859 	mrc_alt_write_mask(DDRPHY, DQANADRVPDCTL, 1 << 30, 1 << 30);
860 	/* RCOMP: Dither PU Enable */
861 	mrc_alt_write_mask(DDRPHY, CMDANADRVPUCTL, 1 << 30, 1 << 30);
862 	/* RCOMP: Dither PD Enable */
863 	mrc_alt_write_mask(DDRPHY, CMDANADRVPDCTL, 1 << 30, 1 << 30);
864 	/* RCOMP: Dither PU Enable */
865 	mrc_alt_write_mask(DDRPHY, CLKANADRVPUCTL, 1 << 30, 1 << 30);
866 	/* RCOMP: Dither PD Enable */
867 	mrc_alt_write_mask(DDRPHY, CLKANADRVPDCTL, 1 << 30, 1 << 30);
868 	/* RCOMP: Dither PU Enable */
869 	mrc_alt_write_mask(DDRPHY, DQSANADRVPUCTL, 1 << 30, 1 << 30);
870 	/* RCOMP: Dither PD Enable */
871 	mrc_alt_write_mask(DDRPHY, DQSANADRVPDCTL, 1 << 30, 1 << 30);
872 	/* RCOMP: Dither PU Enable */
873 	mrc_alt_write_mask(DDRPHY, CTLANADRVPUCTL, 1 << 30, 1 << 30);
874 	/* RCOMP: Dither PD Enable */
875 	mrc_alt_write_mask(DDRPHY, CTLANADRVPDCTL, 1 << 30, 1 << 30);
876 	/* ODT: Dither PU Enable */
877 	mrc_alt_write_mask(DDRPHY, DQANAODTPUCTL, 1 << 30, 1 << 30);
878 	/* ODT: Dither PD Enable */
879 	mrc_alt_write_mask(DDRPHY, DQANAODTPDCTL, 1 << 30, 1 << 30);
880 	/* ODT: Dither PU Enable */
881 	mrc_alt_write_mask(DDRPHY, CLKANAODTPUCTL, 1 << 30, 1 << 30);
882 	/* ODT: Dither PD Enable */
883 	mrc_alt_write_mask(DDRPHY, CLKANAODTPDCTL, 1 << 30, 1 << 30);
884 	/* ODT: Dither PU Enable */
885 	mrc_alt_write_mask(DDRPHY, DQSANAODTPUCTL, 1 << 30, 1 << 30);
886 	/* ODT: Dither PD Enable */
887 	mrc_alt_write_mask(DDRPHY, DQSANAODTPDCTL, 1 << 30, 1 << 30);
888 	/* DCOMP: Dither PU Enable */
889 	mrc_alt_write_mask(DDRPHY, DQANADLYPUCTL, 1 << 30, 1 << 30);
890 	/* DCOMP: Dither PD Enable */
891 	mrc_alt_write_mask(DDRPHY, DQANADLYPDCTL, 1 << 30, 1 << 30);
892 	/* DCOMP: Dither PU Enable */
893 	mrc_alt_write_mask(DDRPHY, CMDANADLYPUCTL, 1 << 30, 1 << 30);
894 	/* DCOMP: Dither PD Enable */
895 	mrc_alt_write_mask(DDRPHY, CMDANADLYPDCTL, 1 << 30, 1 << 30);
896 	/* DCOMP: Dither PU Enable */
897 	mrc_alt_write_mask(DDRPHY, CLKANADLYPUCTL, 1 << 30, 1 << 30);
898 	/* DCOMP: Dither PD Enable */
899 	mrc_alt_write_mask(DDRPHY, CLKANADLYPDCTL, 1 << 30, 1 << 30);
900 	/* DCOMP: Dither PU Enable */
901 	mrc_alt_write_mask(DDRPHY, DQSANADLYPUCTL, 1 << 30, 1 << 30);
902 	/* DCOMP: Dither PD Enable */
903 	mrc_alt_write_mask(DDRPHY, DQSANADLYPDCTL, 1 << 30, 1 << 30);
904 	/* DCOMP: Dither PU Enable */
905 	mrc_alt_write_mask(DDRPHY, CTLANADLYPUCTL, 1 << 30, 1 << 30);
906 	/* DCOMP: Dither PD Enable */
907 	mrc_alt_write_mask(DDRPHY, CTLANADLYPDCTL, 1 << 30, 1 << 30);
908 	/* TCO: Dither PU Enable */
909 	mrc_alt_write_mask(DDRPHY, DQANATCOPUCTL, 1 << 30, 1 << 30);
910 	/* TCO: Dither PD Enable */
911 	mrc_alt_write_mask(DDRPHY, DQANATCOPDCTL, 1 << 30, 1 << 30);
912 	/* TCO: Dither PU Enable */
913 	mrc_alt_write_mask(DDRPHY, CLKANATCOPUCTL, 1 << 30, 1 << 30);
914 	/* TCO: Dither PD Enable */
915 	mrc_alt_write_mask(DDRPHY, CLKANATCOPDCTL, 1 << 30, 1 << 30);
916 	/* TCO: Dither PU Enable */
917 	mrc_alt_write_mask(DDRPHY, DQSANATCOPUCTL, 1 << 30, 1 << 30);
918 	/* TCO: Dither PD Enable */
919 	mrc_alt_write_mask(DDRPHY, DQSANATCOPDCTL, 1 << 30, 1 << 30);
920 	/* TCOCOMP: Pulse Count */
921 	mrc_alt_write_mask(DDRPHY, TCOCNTCTRL, 1, 3);
922 	/* ODT: CMD/CTL PD/PU */
923 	mrc_alt_write_mask(DDRPHY, CHNLBUFSTATIC,
924 		(0x03 << 24) | (0x03 << 16), 0x1f1f0000);
925 	/* Set 1us counter */
926 	mrc_alt_write_mask(DDRPHY, MSCNTR, 0x64, 0xff);
927 	mrc_alt_write_mask(DDRPHY, LATCH1CTL, 0x1 << 28, 0x70000000);
928 
929 	/* Release PHY from reset */
930 	mrc_alt_write_mask(DDRPHY, MASTERRSTN, 1, 1);
931 
932 	/* STEP1 */
933 	mrc_post_code(0x03, 0x11);
934 
935 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
936 		if (mrc_params->channel_enables & (1 << ch)) {
937 			/* DQ01-DQ23 */
938 			for (bl_grp = 0;
939 			     bl_grp < (NUM_BYTE_LANES / bl_divisor) / 2;
940 			     bl_grp++) {
941 				mrc_alt_write_mask(DDRPHY,
942 					DQMDLLCTL +
943 					bl_grp * DDRIODQ_BL_OFFSET +
944 					ch * DDRIODQ_CH_OFFSET,
945 					1 << 13,
946 					1 << 13);	/* Enable VREG */
947 				delay_n(3);
948 			}
949 
950 			/* ECC */
951 			mrc_alt_write_mask(DDRPHY, ECCMDLLCTL,
952 				1 << 13, 1 << 13);	/* Enable VREG */
953 			delay_n(3);
954 			/* CMD */
955 			mrc_alt_write_mask(DDRPHY,
956 				CMDMDLLCTL + ch * DDRIOCCC_CH_OFFSET,
957 				1 << 13, 1 << 13);	/* Enable VREG */
958 			delay_n(3);
959 			/* CLK-CTL */
960 			mrc_alt_write_mask(DDRPHY,
961 				CCMDLLCTL + ch * DDRIOCCC_CH_OFFSET,
962 				1 << 13, 1 << 13);	/* Enable VREG */
963 			delay_n(3);
964 		}
965 	}
966 
967 	/* STEP2 */
968 	mrc_post_code(0x03, 0x12);
969 	delay_n(200);
970 
971 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
972 		if (mrc_params->channel_enables & (1 << ch)) {
973 			/* DQ01-DQ23 */
974 			for (bl_grp = 0;
975 			     bl_grp < (NUM_BYTE_LANES / bl_divisor) / 2;
976 			     bl_grp++) {
977 				mrc_alt_write_mask(DDRPHY,
978 					DQMDLLCTL +
979 					bl_grp * DDRIODQ_BL_OFFSET +
980 					ch * DDRIODQ_CH_OFFSET,
981 					1 << 17,
982 					1 << 17);	/* Enable MCDLL */
983 				delay_n(50);
984 			}
985 
986 		/* ECC */
987 		mrc_alt_write_mask(DDRPHY, ECCMDLLCTL,
988 			1 << 17, 1 << 17);	/* Enable MCDLL */
989 		delay_n(50);
990 		/* CMD */
991 		mrc_alt_write_mask(DDRPHY,
992 			CMDMDLLCTL + ch * DDRIOCCC_CH_OFFSET,
993 			1 << 18, 1 << 18);	/* Enable MCDLL */
994 		delay_n(50);
995 		/* CLK-CTL */
996 		mrc_alt_write_mask(DDRPHY,
997 			CCMDLLCTL + ch * DDRIOCCC_CH_OFFSET,
998 			1 << 18, 1 << 18);	/* Enable MCDLL */
999 		delay_n(50);
1000 		}
1001 	}
1002 
1003 	/* STEP3: */
1004 	mrc_post_code(0x03, 0x13);
1005 	delay_n(100);
1006 
1007 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
1008 		if (mrc_params->channel_enables & (1 << ch)) {
1009 			/* DQ01-DQ23 */
1010 			for (bl_grp = 0;
1011 			     bl_grp < (NUM_BYTE_LANES / bl_divisor) / 2;
1012 			     bl_grp++) {
1013 #ifdef FORCE_16BIT_DDRIO
1014 				temp = (bl_grp &&
1015 					(mrc_params->channel_width == X16)) ?
1016 					0x11ff : 0xffff;
1017 #else
1018 				temp = 0xffff;
1019 #endif
1020 				/* Enable TXDLL */
1021 				mrc_alt_write_mask(DDRPHY,
1022 					DQDLLTXCTL +
1023 					bl_grp * DDRIODQ_BL_OFFSET +
1024 					ch * DDRIODQ_CH_OFFSET,
1025 					temp, 0xffff);
1026 				delay_n(3);
1027 				/* Enable RXDLL */
1028 				mrc_alt_write_mask(DDRPHY,
1029 					DQDLLRXCTL +
1030 					bl_grp * DDRIODQ_BL_OFFSET +
1031 					ch * DDRIODQ_CH_OFFSET,
1032 					0xf, 0xf);
1033 				delay_n(3);
1034 				/* Enable RXDLL Overrides BL0 */
1035 				mrc_alt_write_mask(DDRPHY,
1036 					B0OVRCTL +
1037 					bl_grp * DDRIODQ_BL_OFFSET +
1038 					ch * DDRIODQ_CH_OFFSET,
1039 					0xf, 0xf);
1040 			}
1041 
1042 			/* ECC */
1043 			temp = 0xffff;
1044 			mrc_alt_write_mask(DDRPHY, ECCDLLTXCTL,
1045 				temp, 0xffff);
1046 			delay_n(3);
1047 
1048 			/* CMD (PO) */
1049 			mrc_alt_write_mask(DDRPHY,
1050 				CMDDLLTXCTL + ch * DDRIOCCC_CH_OFFSET,
1051 				temp, 0xffff);
1052 			delay_n(3);
1053 		}
1054 	}
1055 
1056 	/* STEP4 */
1057 	mrc_post_code(0x03, 0x14);
1058 
1059 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
1060 		if (mrc_params->channel_enables & (1 << ch)) {
1061 			/* Host To Memory Clock Alignment (HMC) for 800/1066 */
1062 			for (bl_grp = 0;
1063 			     bl_grp < (NUM_BYTE_LANES / bl_divisor) / 2;
1064 			     bl_grp++) {
1065 				/* CLK_ALIGN_MOD_ID */
1066 				mrc_alt_write_mask(DDRPHY,
1067 					DQCLKALIGNREG2 +
1068 					bl_grp * DDRIODQ_BL_OFFSET +
1069 					ch * DDRIODQ_CH_OFFSET,
1070 					bl_grp ? 3 : 1,
1071 					0xf);
1072 			}
1073 
1074 			mrc_alt_write_mask(DDRPHY,
1075 				ECCCLKALIGNREG2 + ch * DDRIODQ_CH_OFFSET,
1076 				0x2, 0xf);
1077 			mrc_alt_write_mask(DDRPHY,
1078 				CMDCLKALIGNREG2 + ch * DDRIODQ_CH_OFFSET,
1079 				0x0, 0xf);
1080 			mrc_alt_write_mask(DDRPHY,
1081 				CCCLKALIGNREG2 + ch * DDRIODQ_CH_OFFSET,
1082 				0x2, 0xf);
1083 			mrc_alt_write_mask(DDRPHY,
1084 				CMDCLKALIGNREG0 + ch * DDRIOCCC_CH_OFFSET,
1085 				0x20, 0x30);
1086 			/*
1087 			 * NUM_SAMPLES, MAX_SAMPLES,
1088 			 * MACRO_PI_STEP, MICRO_PI_STEP
1089 			 */
1090 			mrc_alt_write_mask(DDRPHY,
1091 				CMDCLKALIGNREG1 + ch * DDRIOCCC_CH_OFFSET,
1092 				(0x18 << 16) | (0x10 << 8) |
1093 				(0x8 << 2) | (0x1 << 0),
1094 				0x007f7fff);
1095 			/* TOTAL_NUM_MODULES, FIRST_U_PARTITION */
1096 			mrc_alt_write_mask(DDRPHY,
1097 				CMDCLKALIGNREG2 + ch * DDRIOCCC_CH_OFFSET,
1098 				(0x10 << 16) | (0x4 << 8) | (0x2 << 4),
1099 				0x001f0ff0);
1100 #ifdef HMC_TEST
1101 			/* START_CLK_ALIGN=1 */
1102 			mrc_alt_write_mask(DDRPHY,
1103 				CMDCLKALIGNREG0 + ch * DDRIOCCC_CH_OFFSET,
1104 				1 << 24, 1 << 24);
1105 			while (msg_port_alt_read(DDRPHY,
1106 				CMDCLKALIGNREG0 + ch * DDRIOCCC_CH_OFFSET) &
1107 				(1 << 24))
1108 				;	/* wait for START_CLK_ALIGN=0 */
1109 #endif
1110 
1111 			/* Set RD/WR Pointer Seperation & COUNTEN & FIFOPTREN */
1112 			mrc_alt_write_mask(DDRPHY,
1113 				CMDPTRREG + ch * DDRIOCCC_CH_OFFSET,
1114 				1, 1);	/* WRPTRENABLE=1 */
1115 
1116 			/* COMP initial */
1117 			/* enable bypass for CLK buffer (PO) */
1118 			mrc_alt_write_mask(DDRPHY,
1119 				COMPEN0CH0 + ch * DDRCOMP_CH_OFFSET,
1120 				1 << 5, 1 << 5);
1121 			/* Initial COMP Enable */
1122 			mrc_alt_write_mask(DDRPHY, CMPCTRL, 1, 1);
1123 			/* wait for Initial COMP Enable = 0 */
1124 			while (msg_port_alt_read(DDRPHY, CMPCTRL) & 1)
1125 				;
1126 			/* disable bypass for CLK buffer (PO) */
1127 			mrc_alt_write_mask(DDRPHY,
1128 				COMPEN0CH0 + ch * DDRCOMP_CH_OFFSET,
1129 				~(1 << 5), 1 << 5);
1130 
1131 			/* IOBUFACT */
1132 
1133 			/* STEP4a */
1134 			mrc_alt_write_mask(DDRPHY,
1135 				CMDCFGREG0 + ch * DDRIOCCC_CH_OFFSET,
1136 				1 << 2, 1 << 2);	/* IOBUFACTRST_N=1 */
1137 
1138 			/* DDRPHY initialization complete */
1139 			mrc_alt_write_mask(DDRPHY,
1140 				CMDPMCONFIG0 + ch * DDRIOCCC_CH_OFFSET,
1141 				1 << 20, 1 << 20);	/* SPID_INIT_COMPLETE=1 */
1142 		}
1143 	}
1144 
1145 	LEAVEFN();
1146 }
1147 
1148 /* This function performs JEDEC initialization on all enabled channels */
1149 void perform_jedec_init(struct mrc_params *mrc_params)
1150 {
1151 	uint8_t twr, wl, rank;
1152 	uint32_t tck;
1153 	u32 dtr0;
1154 	u32 drp;
1155 	u32 drmc;
1156 	u32 mrs0_cmd = 0;
1157 	u32 emrs1_cmd = 0;
1158 	u32 emrs2_cmd = 0;
1159 	u32 emrs3_cmd = 0;
1160 
1161 	ENTERFN();
1162 
1163 	/* jedec_init starts */
1164 	mrc_post_code(0x04, 0x00);
1165 
1166 	/* DDR3_RESET_SET=0, DDR3_RESET_RESET=1 */
1167 	mrc_alt_write_mask(DDRPHY, CCDDR3RESETCTL, 2, 0x102);
1168 
1169 	/* Assert RESET# for 200us */
1170 	delay_u(200);
1171 
1172 	/* DDR3_RESET_SET=1, DDR3_RESET_RESET=0 */
1173 	mrc_alt_write_mask(DDRPHY, CCDDR3RESETCTL, 0x100, 0x102);
1174 
1175 	dtr0 = msg_port_read(MEM_CTLR, DTR0);
1176 
1177 	/*
1178 	 * Set CKEVAL for populated ranks
1179 	 * then send NOP to each rank (#4550197)
1180 	 */
1181 
1182 	drp = msg_port_read(MEM_CTLR, DRP);
1183 	drp &= 0x3;
1184 
1185 	drmc = msg_port_read(MEM_CTLR, DRMC);
1186 	drmc &= 0xfffffffc;
1187 	drmc |= (DRMC_CKEMODE | drp);
1188 
1189 	msg_port_write(MEM_CTLR, DRMC, drmc);
1190 
1191 	for (rank = 0; rank < NUM_RANKS; rank++) {
1192 		/* Skip to next populated rank */
1193 		if ((mrc_params->rank_enables & (1 << rank)) == 0)
1194 			continue;
1195 
1196 		dram_init_command(DCMD_NOP(rank));
1197 	}
1198 
1199 	msg_port_write(MEM_CTLR, DRMC,
1200 		(mrc_params->rd_odt_value == 0 ? DRMC_ODTMODE : 0));
1201 
1202 	/*
1203 	 * setup for emrs 2
1204 	 * BIT[15:11] --> Always "0"
1205 	 * BIT[10:09] --> Rtt_WR: want "Dynamic ODT Off" (0)
1206 	 * BIT[08]    --> Always "0"
1207 	 * BIT[07]    --> SRT: use sr_temp_range
1208 	 * BIT[06]    --> ASR: want "Manual SR Reference" (0)
1209 	 * BIT[05:03] --> CWL: use oem_tCWL
1210 	 * BIT[02:00] --> PASR: want "Full Array" (0)
1211 	 */
1212 	emrs2_cmd |= (2 << 3);
1213 	wl = 5 + mrc_params->ddr_speed;
1214 	emrs2_cmd |= ((wl - 5) << 9);
1215 	emrs2_cmd |= (mrc_params->sr_temp_range << 13);
1216 
1217 	/*
1218 	 * setup for emrs 3
1219 	 * BIT[15:03] --> Always "0"
1220 	 * BIT[02]    --> MPR: want "Normal Operation" (0)
1221 	 * BIT[01:00] --> MPR_Loc: want "Predefined Pattern" (0)
1222 	 */
1223 	emrs3_cmd |= (3 << 3);
1224 
1225 	/*
1226 	 * setup for emrs 1
1227 	 * BIT[15:13]     --> Always "0"
1228 	 * BIT[12:12]     --> Qoff: want "Output Buffer Enabled" (0)
1229 	 * BIT[11:11]     --> TDQS: want "Disabled" (0)
1230 	 * BIT[10:10]     --> Always "0"
1231 	 * BIT[09,06,02]  --> Rtt_nom: use rtt_nom_value
1232 	 * BIT[08]        --> Always "0"
1233 	 * BIT[07]        --> WR_LVL: want "Disabled" (0)
1234 	 * BIT[05,01]     --> DIC: use ron_value
1235 	 * BIT[04:03]     --> AL: additive latency want "0" (0)
1236 	 * BIT[00]        --> DLL: want "Enable" (0)
1237 	 *
1238 	 * (BIT5|BIT1) set Ron value
1239 	 * 00 --> RZQ/6 (40ohm)
1240 	 * 01 --> RZQ/7 (34ohm)
1241 	 * 1* --> RESERVED
1242 	 *
1243 	 * (BIT9|BIT6|BIT2) set Rtt_nom value
1244 	 * 000 --> Disabled
1245 	 * 001 --> RZQ/4 ( 60ohm)
1246 	 * 010 --> RZQ/2 (120ohm)
1247 	 * 011 --> RZQ/6 ( 40ohm)
1248 	 * 1** --> RESERVED
1249 	 */
1250 	emrs1_cmd |= (1 << 3);
1251 	emrs1_cmd &= ~(1 << 6);
1252 
1253 	if (mrc_params->ron_value == 0)
1254 		emrs1_cmd |= (1 << 7);
1255 	else
1256 		emrs1_cmd &= ~(1 << 7);
1257 
1258 	if (mrc_params->rtt_nom_value == 0)
1259 		emrs1_cmd |= (DDR3_EMRS1_RTTNOM_40 << 6);
1260 	else if (mrc_params->rtt_nom_value == 1)
1261 		emrs1_cmd |= (DDR3_EMRS1_RTTNOM_60 << 6);
1262 	else if (mrc_params->rtt_nom_value == 2)
1263 		emrs1_cmd |= (DDR3_EMRS1_RTTNOM_120 << 6);
1264 
1265 	/* save MRS1 value (excluding control fields) */
1266 	mrc_params->mrs1 = emrs1_cmd >> 6;
1267 
1268 	/*
1269 	 * setup for mrs 0
1270 	 * BIT[15:13]     --> Always "0"
1271 	 * BIT[12]        --> PPD: for Quark (1)
1272 	 * BIT[11:09]     --> WR: use oem_tWR
1273 	 * BIT[08]        --> DLL: want "Reset" (1, self clearing)
1274 	 * BIT[07]        --> MODE: want "Normal" (0)
1275 	 * BIT[06:04,02]  --> CL: use oem_tCAS
1276 	 * BIT[03]        --> RD_BURST_TYPE: want "Interleave" (1)
1277 	 * BIT[01:00]     --> BL: want "8 Fixed" (0)
1278 	 * WR:
1279 	 * 0 --> 16
1280 	 * 1 --> 5
1281 	 * 2 --> 6
1282 	 * 3 --> 7
1283 	 * 4 --> 8
1284 	 * 5 --> 10
1285 	 * 6 --> 12
1286 	 * 7 --> 14
1287 	 * CL:
1288 	 * BIT[02:02] "0" if oem_tCAS <= 11 (1866?)
1289 	 * BIT[06:04] use oem_tCAS-4
1290 	 */
1291 	mrs0_cmd |= (1 << 14);
1292 	mrs0_cmd |= (1 << 18);
1293 	mrs0_cmd |= ((((dtr0 >> 12) & 7) + 1) << 10);
1294 
1295 	tck = t_ck[mrc_params->ddr_speed];
1296 	/* Per JEDEC: tWR=15000ps DDR2/3 from 800-1600 */
1297 	twr = MCEIL(15000, tck);
1298 	mrs0_cmd |= ((twr - 4) << 15);
1299 
1300 	for (rank = 0; rank < NUM_RANKS; rank++) {
1301 		/* Skip to next populated rank */
1302 		if ((mrc_params->rank_enables & (1 << rank)) == 0)
1303 			continue;
1304 
1305 		emrs2_cmd |= (rank << 22);
1306 		dram_init_command(emrs2_cmd);
1307 
1308 		emrs3_cmd |= (rank << 22);
1309 		dram_init_command(emrs3_cmd);
1310 
1311 		emrs1_cmd |= (rank << 22);
1312 		dram_init_command(emrs1_cmd);
1313 
1314 		mrs0_cmd |= (rank << 22);
1315 		dram_init_command(mrs0_cmd);
1316 
1317 		dram_init_command(DCMD_ZQCL(rank));
1318 	}
1319 
1320 	LEAVEFN();
1321 }
1322 
1323 /*
1324  * Dunit Initialization Complete
1325  *
1326  * Indicates that initialization of the Dunit has completed.
1327  *
1328  * Memory accesses are permitted and maintenance operation begins.
1329  * Until this bit is set to a 1, the memory controller will not accept
1330  * DRAM requests from the MEMORY_MANAGER or HTE.
1331  */
1332 void set_ddr_init_complete(struct mrc_params *mrc_params)
1333 {
1334 	u32 dco;
1335 
1336 	ENTERFN();
1337 
1338 	dco = msg_port_read(MEM_CTLR, DCO);
1339 	dco &= ~DCO_PMICTL;
1340 	dco |= DCO_IC;
1341 	msg_port_write(MEM_CTLR, DCO, dco);
1342 
1343 	LEAVEFN();
1344 }
1345 
1346 /*
1347  * This function will retrieve relevant timing data
1348  *
1349  * This data will be used on subsequent boots to speed up boot times
1350  * and is required for Suspend To RAM capabilities.
1351  */
1352 void restore_timings(struct mrc_params *mrc_params)
1353 {
1354 	uint8_t ch, rk, bl;
1355 	const struct mrc_timings *mt = &mrc_params->timings;
1356 
1357 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
1358 		for (rk = 0; rk < NUM_RANKS; rk++) {
1359 			for (bl = 0; bl < NUM_BYTE_LANES; bl++) {
1360 				set_rcvn(ch, rk, bl, mt->rcvn[ch][rk][bl]);
1361 				set_rdqs(ch, rk, bl, mt->rdqs[ch][rk][bl]);
1362 				set_wdqs(ch, rk, bl, mt->wdqs[ch][rk][bl]);
1363 				set_wdq(ch, rk, bl, mt->wdq[ch][rk][bl]);
1364 				if (rk == 0) {
1365 					/* VREF (RANK0 only) */
1366 					set_vref(ch, bl, mt->vref[ch][bl]);
1367 				}
1368 			}
1369 			set_wctl(ch, rk, mt->wctl[ch][rk]);
1370 		}
1371 		set_wcmd(ch, mt->wcmd[ch]);
1372 	}
1373 }
1374 
1375 /*
1376  * Configure default settings normally set as part of read training
1377  *
1378  * Some defaults have to be set earlier as they may affect earlier
1379  * training steps.
1380  */
1381 void default_timings(struct mrc_params *mrc_params)
1382 {
1383 	uint8_t ch, rk, bl;
1384 
1385 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
1386 		for (rk = 0; rk < NUM_RANKS; rk++) {
1387 			for (bl = 0; bl < NUM_BYTE_LANES; bl++) {
1388 				set_rdqs(ch, rk, bl, 24);
1389 				if (rk == 0) {
1390 					/* VREF (RANK0 only) */
1391 					set_vref(ch, bl, 32);
1392 				}
1393 			}
1394 		}
1395 	}
1396 }
1397 
1398 /*
1399  * This function will perform our RCVEN Calibration Algorithm.
1400  * We will only use the 2xCLK domain timings to perform RCVEN Calibration.
1401  * All byte lanes will be calibrated "simultaneously" per channel per rank.
1402  */
1403 void rcvn_cal(struct mrc_params *mrc_params)
1404 {
1405 	uint8_t ch;	/* channel counter */
1406 	uint8_t rk;	/* rank counter */
1407 	uint8_t bl;	/* byte lane counter */
1408 	uint8_t bl_divisor = (mrc_params->channel_width == X16) ? 2 : 1;
1409 
1410 #ifdef R2R_SHARING
1411 	/* used to find placement for rank2rank sharing configs */
1412 	uint32_t final_delay[NUM_CHANNELS][NUM_BYTE_LANES];
1413 #ifndef BACKUP_RCVN
1414 	/* used to find placement for rank2rank sharing configs */
1415 	uint32_t num_ranks_enabled = 0;
1416 #endif
1417 #endif
1418 
1419 #ifdef BACKUP_RCVN
1420 #else
1421 	uint32_t temp;
1422 	/* absolute PI value to be programmed on the byte lane */
1423 	uint32_t delay[NUM_BYTE_LANES];
1424 	u32 dtr1, dtr1_save;
1425 #endif
1426 
1427 	ENTERFN();
1428 
1429 	/* rcvn_cal starts */
1430 	mrc_post_code(0x05, 0x00);
1431 
1432 #ifndef BACKUP_RCVN
1433 	/* need separate burst to sample DQS preamble */
1434 	dtr1 = msg_port_read(MEM_CTLR, DTR1);
1435 	dtr1_save = dtr1;
1436 	dtr1 |= DTR1_TCCD_12CLK;
1437 	msg_port_write(MEM_CTLR, DTR1, dtr1);
1438 #endif
1439 
1440 #ifdef R2R_SHARING
1441 	/* need to set "final_delay[][]" elements to "0" */
1442 	memset((void *)(final_delay), 0x00, (size_t)sizeof(final_delay));
1443 #endif
1444 
1445 	/* loop through each enabled channel */
1446 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
1447 		if (mrc_params->channel_enables & (1 << ch)) {
1448 			/* perform RCVEN Calibration on a per rank basis */
1449 			for (rk = 0; rk < NUM_RANKS; rk++) {
1450 				if (mrc_params->rank_enables & (1 << rk)) {
1451 					/*
1452 					 * POST_CODE here indicates the current
1453 					 * channel and rank being calibrated
1454 					 */
1455 					mrc_post_code(0x05, 0x10 + ((ch << 4) | rk));
1456 
1457 #ifdef BACKUP_RCVN
1458 					/* et hard-coded timing values */
1459 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++)
1460 						set_rcvn(ch, rk, bl, ddr_rcvn[PLATFORM_ID]);
1461 #else
1462 					/* enable FIFORST */
1463 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl += 2) {
1464 						mrc_alt_write_mask(DDRPHY,
1465 							B01PTRCTL1 +
1466 							(bl >> 1) * DDRIODQ_BL_OFFSET +
1467 							ch * DDRIODQ_CH_OFFSET,
1468 							0, 1 << 8);
1469 					}
1470 					/* initialize the starting delay to 128 PI (cas +1 CLK) */
1471 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
1472 						/* 1x CLK domain timing is cas-4 */
1473 						delay[bl] = (4 + 1) * FULL_CLK;
1474 
1475 						set_rcvn(ch, rk, bl, delay[bl]);
1476 					}
1477 
1478 					/* now find the rising edge */
1479 					find_rising_edge(mrc_params, delay, ch, rk, true);
1480 
1481 					/* Now increase delay by 32 PI (1/4 CLK) to place in center of high pulse */
1482 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
1483 						delay[bl] += QRTR_CLK;
1484 						set_rcvn(ch, rk, bl, delay[bl]);
1485 					}
1486 					/* Now decrement delay by 128 PI (1 CLK) until we sample a "0" */
1487 					do {
1488 						temp = sample_dqs(mrc_params, ch, rk, true);
1489 						for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
1490 							if (temp & (1 << bl)) {
1491 								if (delay[bl] >= FULL_CLK) {
1492 									delay[bl] -= FULL_CLK;
1493 									set_rcvn(ch, rk, bl, delay[bl]);
1494 								} else {
1495 									/* not enough delay */
1496 									training_message(ch, rk, bl);
1497 									mrc_post_code(0xee, 0x50);
1498 								}
1499 							}
1500 						}
1501 					} while (temp & 0xff);
1502 
1503 #ifdef R2R_SHARING
1504 					/* increment "num_ranks_enabled" */
1505 					num_ranks_enabled++;
1506 					/* Finally increment delay by 32 PI (1/4 CLK) to place in center of preamble */
1507 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
1508 						delay[bl] += QRTR_CLK;
1509 						/* add "delay[]" values to "final_delay[][]" for rolling average */
1510 						final_delay[ch][bl] += delay[bl];
1511 						/* set timing based on rolling average values */
1512 						set_rcvn(ch, rk, bl, final_delay[ch][bl] / num_ranks_enabled);
1513 					}
1514 #else
1515 					/* Finally increment delay by 32 PI (1/4 CLK) to place in center of preamble */
1516 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
1517 						delay[bl] += QRTR_CLK;
1518 						set_rcvn(ch, rk, bl, delay[bl]);
1519 					}
1520 #endif
1521 
1522 					/* disable FIFORST */
1523 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl += 2) {
1524 						mrc_alt_write_mask(DDRPHY,
1525 							B01PTRCTL1 +
1526 							(bl >> 1) * DDRIODQ_BL_OFFSET +
1527 							ch * DDRIODQ_CH_OFFSET,
1528 							1 << 8, 1 << 8);
1529 					}
1530 #endif
1531 				}
1532 			}
1533 		}
1534 	}
1535 
1536 #ifndef BACKUP_RCVN
1537 	/* restore original */
1538 	msg_port_write(MEM_CTLR, DTR1, dtr1_save);
1539 #endif
1540 
1541 	LEAVEFN();
1542 }
1543 
1544 /*
1545  * This function will perform the Write Levelling algorithm
1546  * (align WCLK and WDQS).
1547  *
1548  * This algorithm will act on each rank in each channel separately.
1549  */
1550 void wr_level(struct mrc_params *mrc_params)
1551 {
1552 	uint8_t ch;	/* channel counter */
1553 	uint8_t rk;	/* rank counter */
1554 	uint8_t bl;	/* byte lane counter */
1555 	uint8_t bl_divisor = (mrc_params->channel_width == X16) ? 2 : 1;
1556 
1557 #ifdef R2R_SHARING
1558 	/* used to find placement for rank2rank sharing configs */
1559 	uint32_t final_delay[NUM_CHANNELS][NUM_BYTE_LANES];
1560 #ifndef BACKUP_WDQS
1561 	/* used to find placement for rank2rank sharing configs */
1562 	uint32_t num_ranks_enabled = 0;
1563 #endif
1564 #endif
1565 
1566 #ifdef BACKUP_WDQS
1567 #else
1568 	/* determines stop condition for CRS_WR_LVL */
1569 	bool all_edges_found;
1570 	/* absolute PI value to be programmed on the byte lane */
1571 	uint32_t delay[NUM_BYTE_LANES];
1572 	/*
1573 	 * static makes it so the data is loaded in the heap once by shadow(),
1574 	 * where non-static copies the data onto the stack every time this
1575 	 * function is called
1576 	 */
1577 	uint32_t address;	/* address to be checked during COARSE_WR_LVL */
1578 	u32 dtr4, dtr4_save;
1579 #endif
1580 
1581 	ENTERFN();
1582 
1583 	/* wr_level starts */
1584 	mrc_post_code(0x06, 0x00);
1585 
1586 #ifdef R2R_SHARING
1587 	/* need to set "final_delay[][]" elements to "0" */
1588 	memset((void *)(final_delay), 0x00, (size_t)sizeof(final_delay));
1589 #endif
1590 
1591 	/* loop through each enabled channel */
1592 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
1593 		if (mrc_params->channel_enables & (1 << ch)) {
1594 			/* perform WRITE LEVELING algorithm on a per rank basis */
1595 			for (rk = 0; rk < NUM_RANKS; rk++) {
1596 				if (mrc_params->rank_enables & (1 << rk)) {
1597 					/*
1598 					 * POST_CODE here indicates the current
1599 					 * rank and channel being calibrated
1600 					 */
1601 					mrc_post_code(0x06, 0x10 + ((ch << 4) | rk));
1602 
1603 #ifdef BACKUP_WDQS
1604 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
1605 						set_wdqs(ch, rk, bl, ddr_wdqs[PLATFORM_ID]);
1606 						set_wdq(ch, rk, bl, ddr_wdqs[PLATFORM_ID] - QRTR_CLK);
1607 					}
1608 #else
1609 					/*
1610 					 * perform a single PRECHARGE_ALL command to
1611 					 * make DRAM state machine go to IDLE state
1612 					 */
1613 					dram_init_command(DCMD_PREA(rk));
1614 
1615 					/*
1616 					 * enable Write Levelling Mode
1617 					 * (EMRS1 w/ Write Levelling Mode Enable)
1618 					 */
1619 					dram_init_command(DCMD_MRS1(rk, 0x82));
1620 
1621 					/*
1622 					 * set ODT DRAM Full Time Termination
1623 					 * disable in MCU
1624 					 */
1625 
1626 					dtr4 = msg_port_read(MEM_CTLR, DTR4);
1627 					dtr4_save = dtr4;
1628 					dtr4 |= DTR4_ODTDIS;
1629 					msg_port_write(MEM_CTLR, DTR4, dtr4);
1630 
1631 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor) / 2; bl++) {
1632 						/*
1633 						 * Enable Sandy Bridge Mode (WDQ Tri-State) &
1634 						 * Ensure 5 WDQS pulses during Write Leveling
1635 						 */
1636 						mrc_alt_write_mask(DDRPHY,
1637 							DQCTL + DDRIODQ_BL_OFFSET * bl + DDRIODQ_CH_OFFSET * ch,
1638 							0x10000154,
1639 							0x100003fc);
1640 					}
1641 
1642 					/* Write Leveling Mode enabled in IO */
1643 					mrc_alt_write_mask(DDRPHY,
1644 						CCDDR3RESETCTL + DDRIOCCC_CH_OFFSET * ch,
1645 						1 << 16, 1 << 16);
1646 
1647 					/* Initialize the starting delay to WCLK */
1648 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
1649 						/*
1650 						 * CLK0 --> RK0
1651 						 * CLK1 --> RK1
1652 						 */
1653 						delay[bl] = get_wclk(ch, rk);
1654 
1655 						set_wdqs(ch, rk, bl, delay[bl]);
1656 					}
1657 
1658 					/* now find the rising edge */
1659 					find_rising_edge(mrc_params, delay, ch, rk, false);
1660 
1661 					/* disable Write Levelling Mode */
1662 					mrc_alt_write_mask(DDRPHY,
1663 						CCDDR3RESETCTL + DDRIOCCC_CH_OFFSET * ch,
1664 						0, 1 << 16);
1665 
1666 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor) / 2; bl++) {
1667 						/* Disable Sandy Bridge Mode & Ensure 4 WDQS pulses during normal operation */
1668 						mrc_alt_write_mask(DDRPHY,
1669 							DQCTL + DDRIODQ_BL_OFFSET * bl + DDRIODQ_CH_OFFSET * ch,
1670 							0x00000154,
1671 							0x100003fc);
1672 					}
1673 
1674 					/* restore original DTR4 */
1675 					msg_port_write(MEM_CTLR, DTR4, dtr4_save);
1676 
1677 					/*
1678 					 * restore original value
1679 					 * (Write Levelling Mode Disable)
1680 					 */
1681 					dram_init_command(DCMD_MRS1(rk, mrc_params->mrs1));
1682 
1683 					/*
1684 					 * perform a single PRECHARGE_ALL command to
1685 					 * make DRAM state machine go to IDLE state
1686 					 */
1687 					dram_init_command(DCMD_PREA(rk));
1688 
1689 					mrc_post_code(0x06, 0x30 + ((ch << 4) | rk));
1690 
1691 					/*
1692 					 * COARSE WRITE LEVEL:
1693 					 * check that we're on the correct clock edge
1694 					 */
1695 
1696 					/* hte reconfiguration request */
1697 					mrc_params->hte_setup = 1;
1698 
1699 					/* start CRS_WR_LVL with WDQS = WDQS + 128 PI */
1700 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
1701 						delay[bl] = get_wdqs(ch, rk, bl) + FULL_CLK;
1702 						set_wdqs(ch, rk, bl, delay[bl]);
1703 						/*
1704 						 * program WDQ timings based on WDQS
1705 						 * (WDQ = WDQS - 32 PI)
1706 						 */
1707 						set_wdq(ch, rk, bl, (delay[bl] - QRTR_CLK));
1708 					}
1709 
1710 					/* get an address in the targeted channel/rank */
1711 					address = get_addr(ch, rk);
1712 					do {
1713 						uint32_t coarse_result = 0x00;
1714 						uint32_t coarse_result_mask = byte_lane_mask(mrc_params);
1715 						/* assume pass */
1716 						all_edges_found = true;
1717 
1718 						mrc_params->hte_setup = 1;
1719 						coarse_result = check_rw_coarse(mrc_params, address);
1720 
1721 						/* check for failures and margin the byte lane back 128 PI (1 CLK) */
1722 						for (bl = 0; bl < NUM_BYTE_LANES / bl_divisor; bl++) {
1723 							if (coarse_result & (coarse_result_mask << bl)) {
1724 								all_edges_found = false;
1725 								delay[bl] -= FULL_CLK;
1726 								set_wdqs(ch, rk, bl, delay[bl]);
1727 								/* program WDQ timings based on WDQS (WDQ = WDQS - 32 PI) */
1728 								set_wdq(ch, rk, bl, delay[bl] - QRTR_CLK);
1729 							}
1730 						}
1731 					} while (!all_edges_found);
1732 
1733 #ifdef R2R_SHARING
1734 					/* increment "num_ranks_enabled" */
1735 					 num_ranks_enabled++;
1736 					/* accumulate "final_delay[][]" values from "delay[]" values for rolling average */
1737 					for (bl = 0; bl < NUM_BYTE_LANES / bl_divisor; bl++) {
1738 						final_delay[ch][bl] += delay[bl];
1739 						set_wdqs(ch, rk, bl, final_delay[ch][bl] / num_ranks_enabled);
1740 						/* program WDQ timings based on WDQS (WDQ = WDQS - 32 PI) */
1741 						set_wdq(ch, rk, bl, final_delay[ch][bl] / num_ranks_enabled - QRTR_CLK);
1742 					}
1743 #endif
1744 #endif
1745 				}
1746 			}
1747 		}
1748 	}
1749 
1750 	LEAVEFN();
1751 }
1752 
1753 void prog_page_ctrl(struct mrc_params *mrc_params)
1754 {
1755 	u32 dpmc0;
1756 
1757 	ENTERFN();
1758 
1759 	dpmc0 = msg_port_read(MEM_CTLR, DPMC0);
1760 	dpmc0 &= ~DPMC0_PCLSTO_MASK;
1761 	dpmc0 |= (4 << 16);
1762 	dpmc0 |= DPMC0_PREAPWDEN;
1763 	msg_port_write(MEM_CTLR, DPMC0, dpmc0);
1764 }
1765 
1766 /*
1767  * This function will perform the READ TRAINING Algorithm on all
1768  * channels/ranks/byte_lanes simultaneously to minimize execution time.
1769  *
1770  * The idea here is to train the VREF and RDQS (and eventually RDQ) values
1771  * to achieve maximum READ margins. The algorithm will first determine the
1772  * X coordinate (RDQS setting). This is done by collapsing the VREF eye
1773  * until we find a minimum required RDQS eye for VREF_MIN and VREF_MAX.
1774  * Then we take the averages of the RDQS eye at VREF_MIN and VREF_MAX,
1775  * then average those; this will be the final X coordinate. The algorithm
1776  * will then determine the Y coordinate (VREF setting). This is done by
1777  * collapsing the RDQS eye until we find a minimum required VREF eye for
1778  * RDQS_MIN and RDQS_MAX. Then we take the averages of the VREF eye at
1779  * RDQS_MIN and RDQS_MAX, then average those; this will be the final Y
1780  * coordinate.
1781  *
1782  * NOTE: this algorithm assumes the eye curves have a one-to-one relationship,
1783  * meaning for each X the curve has only one Y and vice-a-versa.
1784  */
1785 void rd_train(struct mrc_params *mrc_params)
1786 {
1787 	uint8_t ch;	/* channel counter */
1788 	uint8_t rk;	/* rank counter */
1789 	uint8_t bl;	/* byte lane counter */
1790 	uint8_t bl_divisor = (mrc_params->channel_width == X16) ? 2 : 1;
1791 #ifdef BACKUP_RDQS
1792 #else
1793 	uint8_t side_x;	/* tracks LEFT/RIGHT approach vectors */
1794 	uint8_t side_y;	/* tracks BOTTOM/TOP approach vectors */
1795 	/* X coordinate data (passing RDQS values) for approach vectors */
1796 	uint8_t x_coordinate[2][2][NUM_CHANNELS][NUM_RANKS][NUM_BYTE_LANES];
1797 	/* Y coordinate data (passing VREF values) for approach vectors */
1798 	uint8_t y_coordinate[2][2][NUM_CHANNELS][NUM_BYTE_LANES];
1799 	/* centered X (RDQS) */
1800 	uint8_t x_center[NUM_CHANNELS][NUM_RANKS][NUM_BYTE_LANES];
1801 	/* centered Y (VREF) */
1802 	uint8_t y_center[NUM_CHANNELS][NUM_BYTE_LANES];
1803 	uint32_t address;	/* target address for check_bls_ex() */
1804 	uint32_t result;	/* result of check_bls_ex() */
1805 	uint32_t bl_mask;	/* byte lane mask for result checking */
1806 #ifdef R2R_SHARING
1807 	/* used to find placement for rank2rank sharing configs */
1808 	uint32_t final_delay[NUM_CHANNELS][NUM_BYTE_LANES];
1809 	/* used to find placement for rank2rank sharing configs */
1810 	uint32_t num_ranks_enabled = 0;
1811 #endif
1812 #endif
1813 
1814 	/* rd_train starts */
1815 	mrc_post_code(0x07, 0x00);
1816 
1817 	ENTERFN();
1818 
1819 #ifdef BACKUP_RDQS
1820 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
1821 		if (mrc_params->channel_enables & (1 << ch)) {
1822 			for (rk = 0; rk < NUM_RANKS; rk++) {
1823 				if (mrc_params->rank_enables & (1 << rk)) {
1824 					for (bl = 0;
1825 					     bl < NUM_BYTE_LANES / bl_divisor;
1826 					     bl++) {
1827 						set_rdqs(ch, rk, bl, ddr_rdqs[PLATFORM_ID]);
1828 					}
1829 				}
1830 			}
1831 		}
1832 	}
1833 #else
1834 	/* initialize x/y_coordinate arrays */
1835 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
1836 		if (mrc_params->channel_enables & (1 << ch)) {
1837 			for (rk = 0; rk < NUM_RANKS; rk++) {
1838 				if (mrc_params->rank_enables & (1 << rk)) {
1839 					for (bl = 0;
1840 					     bl < NUM_BYTE_LANES / bl_divisor;
1841 					     bl++) {
1842 						/* x_coordinate */
1843 						x_coordinate[L][B][ch][rk][bl] = RDQS_MIN;
1844 						x_coordinate[R][B][ch][rk][bl] = RDQS_MAX;
1845 						x_coordinate[L][T][ch][rk][bl] = RDQS_MIN;
1846 						x_coordinate[R][T][ch][rk][bl] = RDQS_MAX;
1847 						/* y_coordinate */
1848 						y_coordinate[L][B][ch][bl] = VREF_MIN;
1849 						y_coordinate[R][B][ch][bl] = VREF_MIN;
1850 						y_coordinate[L][T][ch][bl] = VREF_MAX;
1851 						y_coordinate[R][T][ch][bl] = VREF_MAX;
1852 					}
1853 				}
1854 			}
1855 		}
1856 	}
1857 
1858 	/* initialize other variables */
1859 	bl_mask = byte_lane_mask(mrc_params);
1860 	address = get_addr(0, 0);
1861 
1862 #ifdef R2R_SHARING
1863 	/* need to set "final_delay[][]" elements to "0" */
1864 	memset((void *)(final_delay), 0x00, (size_t)sizeof(final_delay));
1865 #endif
1866 
1867 	/* look for passing coordinates */
1868 	for (side_y = B; side_y <= T; side_y++) {
1869 		for (side_x = L; side_x <= R; side_x++) {
1870 			mrc_post_code(0x07, 0x10 + side_y * 2 + side_x);
1871 
1872 			/* find passing values */
1873 			for (ch = 0; ch < NUM_CHANNELS; ch++) {
1874 				if (mrc_params->channel_enables & (0x1 << ch)) {
1875 					for (rk = 0; rk < NUM_RANKS; rk++) {
1876 						if (mrc_params->rank_enables &
1877 							(0x1 << rk)) {
1878 							/* set x/y_coordinate search starting settings */
1879 							for (bl = 0;
1880 							     bl < NUM_BYTE_LANES / bl_divisor;
1881 							     bl++) {
1882 								set_rdqs(ch, rk, bl,
1883 									 x_coordinate[side_x][side_y][ch][rk][bl]);
1884 								set_vref(ch, bl,
1885 									 y_coordinate[side_x][side_y][ch][bl]);
1886 							}
1887 
1888 							/* get an address in the target channel/rank */
1889 							address = get_addr(ch, rk);
1890 
1891 							/* request HTE reconfiguration */
1892 							mrc_params->hte_setup = 1;
1893 
1894 							/* test the settings */
1895 							do {
1896 								/* result[07:00] == failing byte lane (MAX 8) */
1897 								result = check_bls_ex(mrc_params, address);
1898 
1899 								/* check for failures */
1900 								if (result & 0xff) {
1901 									/* at least 1 byte lane failed */
1902 									for (bl = 0; bl < NUM_BYTE_LANES / bl_divisor; bl++) {
1903 										if (result &
1904 											(bl_mask << bl)) {
1905 											/* adjust the RDQS values accordingly */
1906 											if (side_x == L)
1907 												x_coordinate[L][side_y][ch][rk][bl] += RDQS_STEP;
1908 											else
1909 												x_coordinate[R][side_y][ch][rk][bl] -= RDQS_STEP;
1910 
1911 											/* check that we haven't closed the RDQS_EYE too much */
1912 											if ((x_coordinate[L][side_y][ch][rk][bl] > (RDQS_MAX - MIN_RDQS_EYE)) ||
1913 												(x_coordinate[R][side_y][ch][rk][bl] < (RDQS_MIN + MIN_RDQS_EYE)) ||
1914 												(x_coordinate[L][side_y][ch][rk][bl] ==
1915 												x_coordinate[R][side_y][ch][rk][bl])) {
1916 												/*
1917 												 * not enough RDQS margin available at this VREF
1918 												 * update VREF values accordingly
1919 												 */
1920 												if (side_y == B)
1921 													y_coordinate[side_x][B][ch][bl] += VREF_STEP;
1922 												else
1923 													y_coordinate[side_x][T][ch][bl] -= VREF_STEP;
1924 
1925 												/* check that we haven't closed the VREF_EYE too much */
1926 												if ((y_coordinate[side_x][B][ch][bl] > (VREF_MAX - MIN_VREF_EYE)) ||
1927 													(y_coordinate[side_x][T][ch][bl] < (VREF_MIN + MIN_VREF_EYE)) ||
1928 													(y_coordinate[side_x][B][ch][bl] == y_coordinate[side_x][T][ch][bl])) {
1929 													/* VREF_EYE collapsed below MIN_VREF_EYE */
1930 													training_message(ch, rk, bl);
1931 													mrc_post_code(0xEE, 0x70 + side_y * 2 + side_x);
1932 												} else {
1933 													/* update the VREF setting */
1934 													set_vref(ch, bl, y_coordinate[side_x][side_y][ch][bl]);
1935 													/* reset the X coordinate to begin the search at the new VREF */
1936 													x_coordinate[side_x][side_y][ch][rk][bl] =
1937 														(side_x == L) ? RDQS_MIN : RDQS_MAX;
1938 												}
1939 											}
1940 
1941 											/* update the RDQS setting */
1942 											set_rdqs(ch, rk, bl, x_coordinate[side_x][side_y][ch][rk][bl]);
1943 										}
1944 									}
1945 								}
1946 							} while (result & 0xff);
1947 						}
1948 					}
1949 				}
1950 			}
1951 		}
1952 	}
1953 
1954 	mrc_post_code(0x07, 0x20);
1955 
1956 	/* find final RDQS (X coordinate) & final VREF (Y coordinate) */
1957 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
1958 		if (mrc_params->channel_enables & (1 << ch)) {
1959 			for (rk = 0; rk < NUM_RANKS; rk++) {
1960 				if (mrc_params->rank_enables & (1 << rk)) {
1961 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
1962 						uint32_t temp1;
1963 						uint32_t temp2;
1964 
1965 						/* x_coordinate */
1966 						DPF(D_INFO,
1967 						    "RDQS T/B eye rank%d lane%d : %d-%d %d-%d\n",
1968 						    rk, bl,
1969 						    x_coordinate[L][T][ch][rk][bl],
1970 						    x_coordinate[R][T][ch][rk][bl],
1971 						    x_coordinate[L][B][ch][rk][bl],
1972 						    x_coordinate[R][B][ch][rk][bl]);
1973 
1974 						/* average the TOP side LEFT & RIGHT values */
1975 						temp1 = (x_coordinate[R][T][ch][rk][bl] + x_coordinate[L][T][ch][rk][bl]) / 2;
1976 						/* average the BOTTOM side LEFT & RIGHT values */
1977 						temp2 = (x_coordinate[R][B][ch][rk][bl] + x_coordinate[L][B][ch][rk][bl]) / 2;
1978 						/* average the above averages */
1979 						x_center[ch][rk][bl] = (uint8_t) ((temp1 + temp2) / 2);
1980 
1981 						/* y_coordinate */
1982 						DPF(D_INFO,
1983 						    "VREF R/L eye lane%d : %d-%d %d-%d\n",
1984 						    bl,
1985 						    y_coordinate[R][B][ch][bl],
1986 						    y_coordinate[R][T][ch][bl],
1987 						    y_coordinate[L][B][ch][bl],
1988 						    y_coordinate[L][T][ch][bl]);
1989 
1990 						/* average the RIGHT side TOP & BOTTOM values */
1991 						temp1 = (y_coordinate[R][T][ch][bl] + y_coordinate[R][B][ch][bl]) / 2;
1992 						/* average the LEFT side TOP & BOTTOM values */
1993 						temp2 = (y_coordinate[L][T][ch][bl] + y_coordinate[L][B][ch][bl]) / 2;
1994 						/* average the above averages */
1995 						y_center[ch][bl] = (uint8_t) ((temp1 + temp2) / 2);
1996 					}
1997 				}
1998 			}
1999 		}
2000 	}
2001 
2002 #ifdef RX_EYE_CHECK
2003 	/* perform an eye check */
2004 	for (side_y = B; side_y <= T; side_y++) {
2005 		for (side_x = L; side_x <= R; side_x++) {
2006 			mrc_post_code(0x07, 0x30 + side_y * 2 + side_x);
2007 
2008 			/* update the settings for the eye check */
2009 			for (ch = 0; ch < NUM_CHANNELS; ch++) {
2010 				if (mrc_params->channel_enables & (1 << ch)) {
2011 					for (rk = 0; rk < NUM_RANKS; rk++) {
2012 						if (mrc_params->rank_enables & (1 << rk)) {
2013 							for (bl = 0; bl < NUM_BYTE_LANES / bl_divisor; bl++) {
2014 								if (side_x == L)
2015 									set_rdqs(ch, rk, bl, x_center[ch][rk][bl] - (MIN_RDQS_EYE / 2));
2016 								else
2017 									set_rdqs(ch, rk, bl, x_center[ch][rk][bl] + (MIN_RDQS_EYE / 2));
2018 
2019 								if (side_y == B)
2020 									set_vref(ch, bl, y_center[ch][bl] - (MIN_VREF_EYE / 2));
2021 								else
2022 									set_vref(ch, bl, y_center[ch][bl] + (MIN_VREF_EYE / 2));
2023 							}
2024 						}
2025 					}
2026 				}
2027 			}
2028 
2029 			/* request HTE reconfiguration */
2030 			mrc_params->hte_setup = 1;
2031 
2032 			/* check the eye */
2033 			if (check_bls_ex(mrc_params, address) & 0xff) {
2034 				/* one or more byte lanes failed */
2035 				mrc_post_code(0xee, 0x74 + side_x * 2 + side_y);
2036 			}
2037 		}
2038 	}
2039 #endif
2040 
2041 	mrc_post_code(0x07, 0x40);
2042 
2043 	/* set final placements */
2044 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
2045 		if (mrc_params->channel_enables & (1 << ch)) {
2046 			for (rk = 0; rk < NUM_RANKS; rk++) {
2047 				if (mrc_params->rank_enables & (1 << rk)) {
2048 #ifdef R2R_SHARING
2049 					/* increment "num_ranks_enabled" */
2050 					num_ranks_enabled++;
2051 #endif
2052 					for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor); bl++) {
2053 						/* x_coordinate */
2054 #ifdef R2R_SHARING
2055 						final_delay[ch][bl] += x_center[ch][rk][bl];
2056 						set_rdqs(ch, rk, bl, final_delay[ch][bl] / num_ranks_enabled);
2057 #else
2058 						set_rdqs(ch, rk, bl, x_center[ch][rk][bl]);
2059 #endif
2060 						/* y_coordinate */
2061 						set_vref(ch, bl, y_center[ch][bl]);
2062 					}
2063 				}
2064 			}
2065 		}
2066 	}
2067 #endif
2068 
2069 	LEAVEFN();
2070 }
2071 
2072 /*
2073  * This function will perform the WRITE TRAINING Algorithm on all
2074  * channels/ranks/byte_lanes simultaneously to minimize execution time.
2075  *
2076  * The idea here is to train the WDQ timings to achieve maximum WRITE margins.
2077  * The algorithm will start with WDQ at the current WDQ setting (tracks WDQS
2078  * in WR_LVL) +/- 32 PIs (+/- 1/4 CLK) and collapse the eye until all data
2079  * patterns pass. This is because WDQS will be aligned to WCLK by the
2080  * Write Leveling algorithm and WDQ will only ever have a 1/2 CLK window
2081  * of validity.
2082  */
2083 void wr_train(struct mrc_params *mrc_params)
2084 {
2085 	uint8_t ch;	/* channel counter */
2086 	uint8_t rk;	/* rank counter */
2087 	uint8_t bl;	/* byte lane counter */
2088 	uint8_t bl_divisor = (mrc_params->channel_width == X16) ? 2 : 1;
2089 #ifdef BACKUP_WDQ
2090 #else
2091 	uint8_t side;		/* LEFT/RIGHT side indicator (0=L, 1=R) */
2092 	uint32_t temp;		/* temporary DWORD */
2093 	/* 2 arrays, for L & R side passing delays */
2094 	uint32_t delay[2][NUM_CHANNELS][NUM_RANKS][NUM_BYTE_LANES];
2095 	uint32_t address;	/* target address for check_bls_ex() */
2096 	uint32_t result;	/* result of check_bls_ex() */
2097 	uint32_t bl_mask;	/* byte lane mask for result checking */
2098 #ifdef R2R_SHARING
2099 	/* used to find placement for rank2rank sharing configs */
2100 	uint32_t final_delay[NUM_CHANNELS][NUM_BYTE_LANES];
2101 	/* used to find placement for rank2rank sharing configs */
2102 	uint32_t num_ranks_enabled = 0;
2103 #endif
2104 #endif
2105 
2106 	/* wr_train starts */
2107 	mrc_post_code(0x08, 0x00);
2108 
2109 	ENTERFN();
2110 
2111 #ifdef BACKUP_WDQ
2112 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
2113 		if (mrc_params->channel_enables & (1 << ch)) {
2114 			for (rk = 0; rk < NUM_RANKS; rk++) {
2115 				if (mrc_params->rank_enables & (1 << rk)) {
2116 					for (bl = 0;
2117 					     bl < NUM_BYTE_LANES / bl_divisor;
2118 					     bl++) {
2119 						set_wdq(ch, rk, bl, ddr_wdq[PLATFORM_ID]);
2120 					}
2121 				}
2122 			}
2123 		}
2124 	}
2125 #else
2126 	/* initialize "delay" */
2127 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
2128 		if (mrc_params->channel_enables & (1 << ch)) {
2129 			for (rk = 0; rk < NUM_RANKS; rk++) {
2130 				if (mrc_params->rank_enables & (1 << rk)) {
2131 					for (bl = 0;
2132 					     bl < NUM_BYTE_LANES / bl_divisor;
2133 					     bl++) {
2134 						/*
2135 						 * want to start with
2136 						 * WDQ = (WDQS - QRTR_CLK)
2137 						 * +/- QRTR_CLK
2138 						 */
2139 						temp = get_wdqs(ch, rk, bl) - QRTR_CLK;
2140 						delay[L][ch][rk][bl] = temp - QRTR_CLK;
2141 						delay[R][ch][rk][bl] = temp + QRTR_CLK;
2142 					}
2143 				}
2144 			}
2145 		}
2146 	}
2147 
2148 	/* initialize other variables */
2149 	bl_mask = byte_lane_mask(mrc_params);
2150 	address = get_addr(0, 0);
2151 
2152 #ifdef R2R_SHARING
2153 	/* need to set "final_delay[][]" elements to "0" */
2154 	memset((void *)(final_delay), 0x00, (size_t)sizeof(final_delay));
2155 #endif
2156 
2157 	/*
2158 	 * start algorithm on the LEFT side and train each channel/bl
2159 	 * until no failures are observed, then repeat for the RIGHT side.
2160 	 */
2161 	for (side = L; side <= R; side++) {
2162 		mrc_post_code(0x08, 0x10 + side);
2163 
2164 		/* set starting values */
2165 		for (ch = 0; ch < NUM_CHANNELS; ch++) {
2166 			if (mrc_params->channel_enables & (1 << ch)) {
2167 				for (rk = 0; rk < NUM_RANKS; rk++) {
2168 					if (mrc_params->rank_enables &
2169 						(1 << rk)) {
2170 						for (bl = 0;
2171 						     bl < NUM_BYTE_LANES / bl_divisor;
2172 						     bl++) {
2173 							set_wdq(ch, rk, bl, delay[side][ch][rk][bl]);
2174 						}
2175 					}
2176 				}
2177 			}
2178 		}
2179 
2180 		/* find passing values */
2181 		for (ch = 0; ch < NUM_CHANNELS; ch++) {
2182 			if (mrc_params->channel_enables & (1 << ch)) {
2183 				for (rk = 0; rk < NUM_RANKS; rk++) {
2184 					if (mrc_params->rank_enables &
2185 						(1 << rk)) {
2186 						/* get an address in the target channel/rank */
2187 						address = get_addr(ch, rk);
2188 
2189 						/* request HTE reconfiguration */
2190 						mrc_params->hte_setup = 1;
2191 
2192 						/* check the settings */
2193 						do {
2194 							/* result[07:00] == failing byte lane (MAX 8) */
2195 							result = check_bls_ex(mrc_params, address);
2196 							/* check for failures */
2197 							if (result & 0xff) {
2198 								/* at least 1 byte lane failed */
2199 								for (bl = 0; bl < NUM_BYTE_LANES / bl_divisor; bl++) {
2200 									if (result &
2201 										(bl_mask << bl)) {
2202 										if (side == L)
2203 											delay[L][ch][rk][bl] += WDQ_STEP;
2204 										else
2205 											delay[R][ch][rk][bl] -= WDQ_STEP;
2206 
2207 										/* check for algorithm failure */
2208 										if (delay[L][ch][rk][bl] != delay[R][ch][rk][bl]) {
2209 											/*
2210 											 * margin available
2211 											 * update delay setting
2212 											 */
2213 											set_wdq(ch, rk, bl,
2214 												delay[side][ch][rk][bl]);
2215 										} else {
2216 											/*
2217 											 * no margin available
2218 											 * notify the user and halt
2219 											 */
2220 											training_message(ch, rk, bl);
2221 											mrc_post_code(0xee, 0x80 + side);
2222 										}
2223 									}
2224 								}
2225 							}
2226 						/* stop when all byte lanes pass */
2227 						} while (result & 0xff);
2228 					}
2229 				}
2230 			}
2231 		}
2232 	}
2233 
2234 	/* program WDQ to the middle of passing window */
2235 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
2236 		if (mrc_params->channel_enables & (1 << ch)) {
2237 			for (rk = 0; rk < NUM_RANKS; rk++) {
2238 				if (mrc_params->rank_enables & (1 << rk)) {
2239 #ifdef R2R_SHARING
2240 					/* increment "num_ranks_enabled" */
2241 					num_ranks_enabled++;
2242 #endif
2243 					for (bl = 0; bl < NUM_BYTE_LANES / bl_divisor; bl++) {
2244 						DPF(D_INFO,
2245 						    "WDQ eye rank%d lane%d : %d-%d\n",
2246 						    rk, bl,
2247 						    delay[L][ch][rk][bl],
2248 						    delay[R][ch][rk][bl]);
2249 
2250 						temp = (delay[R][ch][rk][bl] + delay[L][ch][rk][bl]) / 2;
2251 
2252 #ifdef R2R_SHARING
2253 						final_delay[ch][bl] += temp;
2254 						set_wdq(ch, rk, bl,
2255 							final_delay[ch][bl] / num_ranks_enabled);
2256 #else
2257 						set_wdq(ch, rk, bl, temp);
2258 #endif
2259 					}
2260 				}
2261 			}
2262 		}
2263 	}
2264 #endif
2265 
2266 	LEAVEFN();
2267 }
2268 
2269 /*
2270  * This function will store relevant timing data
2271  *
2272  * This data will be used on subsequent boots to speed up boot times
2273  * and is required for Suspend To RAM capabilities.
2274  */
2275 void store_timings(struct mrc_params *mrc_params)
2276 {
2277 	uint8_t ch, rk, bl;
2278 	struct mrc_timings *mt = &mrc_params->timings;
2279 
2280 	for (ch = 0; ch < NUM_CHANNELS; ch++) {
2281 		for (rk = 0; rk < NUM_RANKS; rk++) {
2282 			for (bl = 0; bl < NUM_BYTE_LANES; bl++) {
2283 				mt->rcvn[ch][rk][bl] = get_rcvn(ch, rk, bl);
2284 				mt->rdqs[ch][rk][bl] = get_rdqs(ch, rk, bl);
2285 				mt->wdqs[ch][rk][bl] = get_wdqs(ch, rk, bl);
2286 				mt->wdq[ch][rk][bl] = get_wdq(ch, rk, bl);
2287 
2288 				if (rk == 0)
2289 					mt->vref[ch][bl] = get_vref(ch, bl);
2290 			}
2291 
2292 			mt->wctl[ch][rk] = get_wctl(ch, rk);
2293 		}
2294 
2295 		mt->wcmd[ch] = get_wcmd(ch);
2296 	}
2297 
2298 	/* need to save for a case of changing frequency after warm reset */
2299 	mt->ddr_speed = mrc_params->ddr_speed;
2300 }
2301 
2302 /*
2303  * The purpose of this function is to ensure the SEC comes out of reset
2304  * and IA initiates the SEC enabling Memory Scrambling.
2305  */
2306 void enable_scrambling(struct mrc_params *mrc_params)
2307 {
2308 	uint32_t lfsr = 0;
2309 	uint8_t i;
2310 
2311 	if (mrc_params->scrambling_enables == 0)
2312 		return;
2313 
2314 	ENTERFN();
2315 
2316 	/* 32 bit seed is always stored in BIOS NVM */
2317 	lfsr = mrc_params->timings.scrambler_seed;
2318 
2319 	if (mrc_params->boot_mode == BM_COLD) {
2320 		/*
2321 		 * factory value is 0 and in first boot,
2322 		 * a clock based seed is loaded.
2323 		 */
2324 		if (lfsr == 0) {
2325 			/*
2326 			 * get seed from system clock
2327 			 * and make sure it is not all 1's
2328 			 */
2329 			lfsr = rdtsc() & 0x0fffffff;
2330 		} else {
2331 			/*
2332 			 * Need to replace scrambler
2333 			 *
2334 			 * get next 32bit LFSR 16 times which is the last
2335 			 * part of the previous scrambler vector
2336 			 */
2337 			for (i = 0; i < 16; i++)
2338 				lfsr32(&lfsr);
2339 		}
2340 
2341 		/* save new seed */
2342 		mrc_params->timings.scrambler_seed = lfsr;
2343 	}
2344 
2345 	/*
2346 	 * In warm boot or S3 exit, we have the previous seed.
2347 	 * In cold boot, we have the last 32bit LFSR which is the new seed.
2348 	 */
2349 	lfsr32(&lfsr);	/* shift to next value */
2350 	msg_port_write(MEM_CTLR, SCRMSEED, (lfsr & 0x0003ffff));
2351 
2352 	for (i = 0; i < 2; i++)
2353 		msg_port_write(MEM_CTLR, SCRMLO + i, (lfsr & 0xaaaaaaaa));
2354 
2355 	LEAVEFN();
2356 }
2357 
2358 /*
2359  * Configure MCU Power Management Control Register
2360  * and Scheduler Control Register
2361  */
2362 void prog_ddr_control(struct mrc_params *mrc_params)
2363 {
2364 	u32 dsch;
2365 	u32 dpmc0;
2366 
2367 	ENTERFN();
2368 
2369 	dsch = msg_port_read(MEM_CTLR, DSCH);
2370 	dsch &= ~(DSCH_OOODIS | DSCH_OOOST3DIS | DSCH_NEWBYPDIS);
2371 	msg_port_write(MEM_CTLR, DSCH, dsch);
2372 
2373 	dpmc0 = msg_port_read(MEM_CTLR, DPMC0);
2374 	dpmc0 &= ~DPMC0_DISPWRDN;
2375 	dpmc0 |= (mrc_params->power_down_disable << 25);
2376 	dpmc0 &= ~DPMC0_CLKGTDIS;
2377 	dpmc0 &= ~DPMC0_PCLSTO_MASK;
2378 	dpmc0 |= (4 << 16);
2379 	dpmc0 |= DPMC0_PREAPWDEN;
2380 	msg_port_write(MEM_CTLR, DPMC0, dpmc0);
2381 
2382 	/* CMDTRIST = 2h - CMD/ADDR are tristated when no valid command */
2383 	mrc_write_mask(MEM_CTLR, DPMC1, 0x20, 0x30);
2384 
2385 	LEAVEFN();
2386 }
2387 
2388 /*
2389  * After training complete configure MCU Rank Population Register
2390  * specifying: ranks enabled, device width, density, address mode
2391  */
2392 void prog_dra_drb(struct mrc_params *mrc_params)
2393 {
2394 	u32 drp;
2395 	u32 dco;
2396 	u8 density = mrc_params->params.density;
2397 
2398 	ENTERFN();
2399 
2400 	dco = msg_port_read(MEM_CTLR, DCO);
2401 	dco &= ~DCO_IC;
2402 	msg_port_write(MEM_CTLR, DCO, dco);
2403 
2404 	drp = 0;
2405 	if (mrc_params->rank_enables & 1)
2406 		drp |= DRP_RKEN0;
2407 	if (mrc_params->rank_enables & 2)
2408 		drp |= DRP_RKEN1;
2409 	if (mrc_params->dram_width == X16) {
2410 		drp |= (1 << 4);
2411 		drp |= (1 << 9);
2412 	}
2413 
2414 	/*
2415 	 * Density encoding in struct dram_params: 0=512Mb, 1=Gb, 2=2Gb, 3=4Gb
2416 	 * has to be mapped RANKDENSx encoding (0=1Gb)
2417 	 */
2418 	if (density == 0)
2419 		density = 4;
2420 
2421 	drp |= ((density - 1) << 6);
2422 	drp |= ((density - 1) << 11);
2423 
2424 	/* Address mode can be overwritten if ECC enabled */
2425 	drp |= (mrc_params->address_mode << 14);
2426 
2427 	msg_port_write(MEM_CTLR, DRP, drp);
2428 
2429 	dco &= ~DCO_PMICTL;
2430 	dco |= DCO_IC;
2431 	msg_port_write(MEM_CTLR, DCO, dco);
2432 
2433 	LEAVEFN();
2434 }
2435 
2436 /* Send DRAM wake command */
2437 void perform_wake(struct mrc_params *mrc_params)
2438 {
2439 	ENTERFN();
2440 
2441 	dram_wake_command();
2442 
2443 	LEAVEFN();
2444 }
2445 
2446 /*
2447  * Configure refresh rate and short ZQ calibration interval
2448  * Activate dynamic self refresh
2449  */
2450 void change_refresh_period(struct mrc_params *mrc_params)
2451 {
2452 	u32 drfc;
2453 	u32 dcal;
2454 	u32 dpmc0;
2455 
2456 	ENTERFN();
2457 
2458 	drfc = msg_port_read(MEM_CTLR, DRFC);
2459 	drfc &= ~DRFC_TREFI_MASK;
2460 	drfc |= (mrc_params->refresh_rate << 12);
2461 	drfc |= DRFC_REFDBTCLR;
2462 	msg_port_write(MEM_CTLR, DRFC, drfc);
2463 
2464 	dcal = msg_port_read(MEM_CTLR, DCAL);
2465 	dcal &= ~DCAL_ZQCINT_MASK;
2466 	dcal |= (3 << 8);	/* 63ms */
2467 	msg_port_write(MEM_CTLR, DCAL, dcal);
2468 
2469 	dpmc0 = msg_port_read(MEM_CTLR, DPMC0);
2470 	dpmc0 |= (DPMC0_DYNSREN | DPMC0_ENPHYCLKGATE);
2471 	msg_port_write(MEM_CTLR, DPMC0, dpmc0);
2472 
2473 	LEAVEFN();
2474 }
2475 
2476 /*
2477  * Configure DDRPHY for Auto-Refresh, Periodic Compensations,
2478  * Dynamic Diff-Amp, ZQSPERIOD, Auto-Precharge, CKE Power-Down
2479  */
2480 void set_auto_refresh(struct mrc_params *mrc_params)
2481 {
2482 	uint32_t channel;
2483 	uint32_t rank;
2484 	uint32_t bl;
2485 	uint32_t bl_divisor = 1;
2486 	uint32_t temp;
2487 
2488 	ENTERFN();
2489 
2490 	/*
2491 	 * Enable Auto-Refresh, Periodic Compensations, Dynamic Diff-Amp,
2492 	 * ZQSPERIOD, Auto-Precharge, CKE Power-Down
2493 	 */
2494 	for (channel = 0; channel < NUM_CHANNELS; channel++) {
2495 		if (mrc_params->channel_enables & (1 << channel)) {
2496 			/* Enable Periodic RCOMPS */
2497 			mrc_alt_write_mask(DDRPHY, CMPCTRL, 2, 2);
2498 
2499 			/* Enable Dynamic DiffAmp & Set Read ODT Value */
2500 			switch (mrc_params->rd_odt_value) {
2501 			case 0:
2502 				temp = 0x3f;	/* OFF */
2503 				break;
2504 			default:
2505 				temp = 0x00;	/* Auto */
2506 				break;
2507 			}
2508 
2509 			for (bl = 0; bl < (NUM_BYTE_LANES / bl_divisor) / 2; bl++) {
2510 				/* Override: DIFFAMP, ODT */
2511 				mrc_alt_write_mask(DDRPHY,
2512 					B0OVRCTL + bl * DDRIODQ_BL_OFFSET +
2513 					channel * DDRIODQ_CH_OFFSET,
2514 					temp << 10,
2515 					0x003ffc00);
2516 
2517 				/* Override: DIFFAMP, ODT */
2518 				mrc_alt_write_mask(DDRPHY,
2519 					B1OVRCTL + bl * DDRIODQ_BL_OFFSET +
2520 					channel * DDRIODQ_CH_OFFSET,
2521 					temp << 10,
2522 					0x003ffc00);
2523 			}
2524 
2525 			/* Issue ZQCS command */
2526 			for (rank = 0; rank < NUM_RANKS; rank++) {
2527 				if (mrc_params->rank_enables & (1 << rank))
2528 					dram_init_command(DCMD_ZQCS(rank));
2529 			}
2530 		}
2531 	}
2532 
2533 	clear_pointers();
2534 
2535 	LEAVEFN();
2536 }
2537 
2538 /*
2539  * Depending on configuration enables ECC support
2540  *
2541  * Available memory size is decreased, and updated with 0s
2542  * in order to clear error status. Address mode 2 forced.
2543  */
2544 void ecc_enable(struct mrc_params *mrc_params)
2545 {
2546 	u32 drp;
2547 	u32 dsch;
2548 	u32 ecc_ctrl;
2549 
2550 	if (mrc_params->ecc_enables == 0)
2551 		return;
2552 
2553 	ENTERFN();
2554 
2555 	/* Configuration required in ECC mode */
2556 	drp = msg_port_read(MEM_CTLR, DRP);
2557 	drp &= ~DRP_ADDRMAP_MASK;
2558 	drp |= DRP_ADDRMAP_MAP1;
2559 	drp |= DRP_PRI64BSPLITEN;
2560 	msg_port_write(MEM_CTLR, DRP, drp);
2561 
2562 	/* Disable new request bypass */
2563 	dsch = msg_port_read(MEM_CTLR, DSCH);
2564 	dsch |= DSCH_NEWBYPDIS;
2565 	msg_port_write(MEM_CTLR, DSCH, dsch);
2566 
2567 	/* Enable ECC */
2568 	ecc_ctrl = (DECCCTRL_SBEEN | DECCCTRL_DBEEN | DECCCTRL_ENCBGEN);
2569 	msg_port_write(MEM_CTLR, DECCCTRL, ecc_ctrl);
2570 
2571 	/* Assume 8 bank memory, one bank is gone for ECC */
2572 	mrc_params->mem_size -= mrc_params->mem_size / 8;
2573 
2574 	/* For S3 resume memory content has to be preserved */
2575 	if (mrc_params->boot_mode != BM_S3) {
2576 		select_hte();
2577 		hte_mem_init(mrc_params, MRC_MEM_INIT);
2578 		select_mem_mgr();
2579 	}
2580 
2581 	LEAVEFN();
2582 }
2583 
2584 /*
2585  * Execute memory test
2586  * if error detected it is indicated in mrc_params->status
2587  */
2588 void memory_test(struct mrc_params *mrc_params)
2589 {
2590 	uint32_t result = 0;
2591 
2592 	ENTERFN();
2593 
2594 	select_hte();
2595 	result = hte_mem_init(mrc_params, MRC_MEM_TEST);
2596 	select_mem_mgr();
2597 
2598 	DPF(D_INFO, "Memory test result %x\n", result);
2599 	mrc_params->status = ((result == 0) ? MRC_SUCCESS : MRC_E_MEMTEST);
2600 	LEAVEFN();
2601 }
2602 
2603 /* Lock MCU registers at the end of initialization sequence */
2604 void lock_registers(struct mrc_params *mrc_params)
2605 {
2606 	u32 dco;
2607 
2608 	ENTERFN();
2609 
2610 	dco = msg_port_read(MEM_CTLR, DCO);
2611 	dco &= ~(DCO_PMICTL | DCO_PMIDIS);
2612 	dco |= (DCO_DRPLOCK | DCO_CPGCLOCK);
2613 	msg_port_write(MEM_CTLR, DCO, dco);
2614 
2615 	LEAVEFN();
2616 }
2617