]> git.ipfire.org Git - people/ms/u-boot.git/blob - drivers/ddr/altera/sequencer.c
ddr: altera: Clean up set_rank_and_odt_mask() part 3
[people/ms/u-boot.git] / drivers / ddr / altera / sequencer.c
1 /*
2 * Copyright Altera Corporation (C) 2012-2015
3 *
4 * SPDX-License-Identifier: BSD-3-Clause
5 */
6
7 #include <common.h>
8 #include <asm/io.h>
9 #include <asm/arch/sdram.h>
10 #include "sequencer.h"
11 #include "sequencer_auto.h"
12 #include "sequencer_auto_ac_init.h"
13 #include "sequencer_auto_inst_init.h"
14 #include "sequencer_defines.h"
15
16 static struct socfpga_sdr_rw_load_manager *sdr_rw_load_mgr_regs =
17 (struct socfpga_sdr_rw_load_manager *)(SDR_PHYGRP_RWMGRGRP_ADDRESS | 0x800);
18
19 static struct socfpga_sdr_rw_load_jump_manager *sdr_rw_load_jump_mgr_regs =
20 (struct socfpga_sdr_rw_load_jump_manager *)(SDR_PHYGRP_RWMGRGRP_ADDRESS | 0xC00);
21
22 static struct socfpga_sdr_reg_file *sdr_reg_file =
23 (struct socfpga_sdr_reg_file *)SDR_PHYGRP_REGFILEGRP_ADDRESS;
24
25 static struct socfpga_sdr_scc_mgr *sdr_scc_mgr =
26 (struct socfpga_sdr_scc_mgr *)(SDR_PHYGRP_SCCGRP_ADDRESS | 0xe00);
27
28 static struct socfpga_phy_mgr_cmd *phy_mgr_cmd =
29 (struct socfpga_phy_mgr_cmd *)SDR_PHYGRP_PHYMGRGRP_ADDRESS;
30
31 static struct socfpga_phy_mgr_cfg *phy_mgr_cfg =
32 (struct socfpga_phy_mgr_cfg *)(SDR_PHYGRP_PHYMGRGRP_ADDRESS | 0x40);
33
34 static struct socfpga_data_mgr *data_mgr =
35 (struct socfpga_data_mgr *)SDR_PHYGRP_DATAMGRGRP_ADDRESS;
36
37 static struct socfpga_sdr_ctrl *sdr_ctrl =
38 (struct socfpga_sdr_ctrl *)SDR_CTRLGRP_ADDRESS;
39
40 #define DELTA_D 1
41
42 /*
43 * In order to reduce ROM size, most of the selectable calibration steps are
44 * decided at compile time based on the user's calibration mode selection,
45 * as captured by the STATIC_CALIB_STEPS selection below.
46 *
47 * However, to support simulation-time selection of fast simulation mode, where
48 * we skip everything except the bare minimum, we need a few of the steps to
49 * be dynamic. In those cases, we either use the DYNAMIC_CALIB_STEPS for the
50 * check, which is based on the rtl-supplied value, or we dynamically compute
51 * the value to use based on the dynamically-chosen calibration mode
52 */
53
54 #define DLEVEL 0
55 #define STATIC_IN_RTL_SIM 0
56 #define STATIC_SKIP_DELAY_LOOPS 0
57
58 #define STATIC_CALIB_STEPS (STATIC_IN_RTL_SIM | CALIB_SKIP_FULL_TEST | \
59 STATIC_SKIP_DELAY_LOOPS)
60
61 /* calibration steps requested by the rtl */
62 uint16_t dyn_calib_steps;
63
64 /*
65 * To make CALIB_SKIP_DELAY_LOOPS a dynamic conditional option
66 * instead of static, we use boolean logic to select between
67 * non-skip and skip values
68 *
69 * The mask is set to include all bits when not-skipping, but is
70 * zero when skipping
71 */
72
73 uint16_t skip_delay_mask; /* mask off bits when skipping/not-skipping */
74
75 #define SKIP_DELAY_LOOP_VALUE_OR_ZERO(non_skip_value) \
76 ((non_skip_value) & skip_delay_mask)
77
78 struct gbl_type *gbl;
79 struct param_type *param;
80 uint32_t curr_shadow_reg;
81
82 static uint32_t rw_mgr_mem_calibrate_write_test(uint32_t rank_bgn,
83 uint32_t write_group, uint32_t use_dm,
84 uint32_t all_correct, uint32_t *bit_chk, uint32_t all_ranks);
85
86 static void set_failing_group_stage(uint32_t group, uint32_t stage,
87 uint32_t substage)
88 {
89 /*
90 * Only set the global stage if there was not been any other
91 * failing group
92 */
93 if (gbl->error_stage == CAL_STAGE_NIL) {
94 gbl->error_substage = substage;
95 gbl->error_stage = stage;
96 gbl->error_group = group;
97 }
98 }
99
100 static void reg_file_set_group(u16 set_group)
101 {
102 clrsetbits_le32(&sdr_reg_file->cur_stage, 0xffff0000, set_group << 16);
103 }
104
105 static void reg_file_set_stage(u8 set_stage)
106 {
107 clrsetbits_le32(&sdr_reg_file->cur_stage, 0xffff, set_stage & 0xff);
108 }
109
110 static void reg_file_set_sub_stage(u8 set_sub_stage)
111 {
112 set_sub_stage &= 0xff;
113 clrsetbits_le32(&sdr_reg_file->cur_stage, 0xff00, set_sub_stage << 8);
114 }
115
116 /**
117 * phy_mgr_initialize() - Initialize PHY Manager
118 *
119 * Initialize PHY Manager.
120 */
121 static void phy_mgr_initialize(void)
122 {
123 u32 ratio;
124
125 debug("%s:%d\n", __func__, __LINE__);
126 /* Calibration has control over path to memory */
127 /*
128 * In Hard PHY this is a 2-bit control:
129 * 0: AFI Mux Select
130 * 1: DDIO Mux Select
131 */
132 writel(0x3, &phy_mgr_cfg->mux_sel);
133
134 /* USER memory clock is not stable we begin initialization */
135 writel(0, &phy_mgr_cfg->reset_mem_stbl);
136
137 /* USER calibration status all set to zero */
138 writel(0, &phy_mgr_cfg->cal_status);
139
140 writel(0, &phy_mgr_cfg->cal_debug_info);
141
142 /* Init params only if we do NOT skip calibration. */
143 if ((dyn_calib_steps & CALIB_SKIP_ALL) == CALIB_SKIP_ALL)
144 return;
145
146 ratio = RW_MGR_MEM_DQ_PER_READ_DQS /
147 RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS;
148 param->read_correct_mask_vg = (1 << ratio) - 1;
149 param->write_correct_mask_vg = (1 << ratio) - 1;
150 param->read_correct_mask = (1 << RW_MGR_MEM_DQ_PER_READ_DQS) - 1;
151 param->write_correct_mask = (1 << RW_MGR_MEM_DQ_PER_WRITE_DQS) - 1;
152 ratio = RW_MGR_MEM_DATA_WIDTH /
153 RW_MGR_MEM_DATA_MASK_WIDTH;
154 param->dm_correct_mask = (1 << ratio) - 1;
155 }
156
157 /**
158 * set_rank_and_odt_mask() - Set Rank and ODT mask
159 * @rank: Rank mask
160 * @odt_mode: ODT mode, OFF or READ_WRITE
161 *
162 * Set Rank and ODT mask (On-Die Termination).
163 */
164 static void set_rank_and_odt_mask(const u32 rank, const u32 odt_mode)
165 {
166 u32 odt_mask_0 = 0;
167 u32 odt_mask_1 = 0;
168 u32 cs_and_odt_mask;
169
170 if (odt_mode == RW_MGR_ODT_MODE_OFF) {
171 odt_mask_0 = 0x0;
172 odt_mask_1 = 0x0;
173 } else { /* RW_MGR_ODT_MODE_READ_WRITE */
174 switch (RW_MGR_MEM_NUMBER_OF_RANKS) {
175 case 1: /* 1 Rank */
176 /* Read: ODT = 0 ; Write: ODT = 1 */
177 odt_mask_0 = 0x0;
178 odt_mask_1 = 0x1;
179 break;
180 case 2: /* 2 Ranks */
181 if (RW_MGR_MEM_NUMBER_OF_CS_PER_DIMM == 1) {
182 /*
183 * - Dual-Slot , Single-Rank (1 CS per DIMM)
184 * OR
185 * - RDIMM, 4 total CS (2 CS per DIMM, 2 DIMM)
186 *
187 * Since MEM_NUMBER_OF_RANKS is 2, they
188 * are both single rank with 2 CS each
189 * (special for RDIMM).
190 *
191 * Read: Turn on ODT on the opposite rank
192 * Write: Turn on ODT on all ranks
193 */
194 odt_mask_0 = 0x3 & ~(1 << rank);
195 odt_mask_1 = 0x3;
196 } else {
197 /*
198 * - Single-Slot , Dual-Rank (2 CS per DIMM)
199 *
200 * Read: Turn on ODT off on all ranks
201 * Write: Turn on ODT on active rank
202 */
203 odt_mask_0 = 0x0;
204 odt_mask_1 = 0x3 & (1 << rank);
205 }
206 break;
207 case 4: /* 4 Ranks */
208 /* Read:
209 * ----------+-----------------------+
210 * | ODT |
211 * Read From +-----------------------+
212 * Rank | 3 | 2 | 1 | 0 |
213 * ----------+-----+-----+-----+-----+
214 * 0 | 0 | 1 | 0 | 0 |
215 * 1 | 1 | 0 | 0 | 0 |
216 * 2 | 0 | 0 | 0 | 1 |
217 * 3 | 0 | 0 | 1 | 0 |
218 * ----------+-----+-----+-----+-----+
219 *
220 * Write:
221 * ----------+-----------------------+
222 * | ODT |
223 * Write To +-----------------------+
224 * Rank | 3 | 2 | 1 | 0 |
225 * ----------+-----+-----+-----+-----+
226 * 0 | 0 | 1 | 0 | 1 |
227 * 1 | 1 | 0 | 1 | 0 |
228 * 2 | 0 | 1 | 0 | 1 |
229 * 3 | 1 | 0 | 1 | 0 |
230 * ----------+-----+-----+-----+-----+
231 */
232 switch (rank) {
233 case 0:
234 odt_mask_0 = 0x4;
235 odt_mask_1 = 0x5;
236 break;
237 case 1:
238 odt_mask_0 = 0x8;
239 odt_mask_1 = 0xA;
240 break;
241 case 2:
242 odt_mask_0 = 0x1;
243 odt_mask_1 = 0x5;
244 break;
245 case 3:
246 odt_mask_0 = 0x2;
247 odt_mask_1 = 0xA;
248 break;
249 }
250 break;
251 }
252 }
253
254 cs_and_odt_mask = (0xFF & ~(1 << rank)) |
255 ((0xFF & odt_mask_0) << 8) |
256 ((0xFF & odt_mask_1) << 16);
257 writel(cs_and_odt_mask, SDR_PHYGRP_RWMGRGRP_ADDRESS |
258 RW_MGR_SET_CS_AND_ODT_MASK_OFFSET);
259 }
260
261 /**
262 * scc_mgr_set() - Set SCC Manager register
263 * @off: Base offset in SCC Manager space
264 * @grp: Read/Write group
265 * @val: Value to be set
266 *
267 * This function sets the SCC Manager (Scan Chain Control Manager) register.
268 */
269 static void scc_mgr_set(u32 off, u32 grp, u32 val)
270 {
271 writel(val, SDR_PHYGRP_SCCGRP_ADDRESS | off | (grp << 2));
272 }
273
274 /**
275 * scc_mgr_initialize() - Initialize SCC Manager registers
276 *
277 * Initialize SCC Manager registers.
278 */
279 static void scc_mgr_initialize(void)
280 {
281 /*
282 * Clear register file for HPS. 16 (2^4) is the size of the
283 * full register file in the scc mgr:
284 * RFILE_DEPTH = 1 + log2(MEM_DQ_PER_DQS + 1 + MEM_DM_PER_DQS +
285 * MEM_IF_READ_DQS_WIDTH - 1);
286 */
287 int i;
288
289 for (i = 0; i < 16; i++) {
290 debug_cond(DLEVEL == 1, "%s:%d: Clearing SCC RFILE index %u\n",
291 __func__, __LINE__, i);
292 scc_mgr_set(SCC_MGR_HHP_RFILE_OFFSET, 0, i);
293 }
294 }
295
296 static void scc_mgr_set_dqdqs_output_phase(uint32_t write_group, uint32_t phase)
297 {
298 scc_mgr_set(SCC_MGR_DQDQS_OUT_PHASE_OFFSET, write_group, phase);
299 }
300
301 static void scc_mgr_set_dqs_bus_in_delay(uint32_t read_group, uint32_t delay)
302 {
303 scc_mgr_set(SCC_MGR_DQS_IN_DELAY_OFFSET, read_group, delay);
304 }
305
306 static void scc_mgr_set_dqs_en_phase(uint32_t read_group, uint32_t phase)
307 {
308 scc_mgr_set(SCC_MGR_DQS_EN_PHASE_OFFSET, read_group, phase);
309 }
310
311 static void scc_mgr_set_dqs_en_delay(uint32_t read_group, uint32_t delay)
312 {
313 scc_mgr_set(SCC_MGR_DQS_EN_DELAY_OFFSET, read_group, delay);
314 }
315
316 static void scc_mgr_set_dqs_io_in_delay(uint32_t delay)
317 {
318 scc_mgr_set(SCC_MGR_IO_IN_DELAY_OFFSET, RW_MGR_MEM_DQ_PER_WRITE_DQS,
319 delay);
320 }
321
322 static void scc_mgr_set_dq_in_delay(uint32_t dq_in_group, uint32_t delay)
323 {
324 scc_mgr_set(SCC_MGR_IO_IN_DELAY_OFFSET, dq_in_group, delay);
325 }
326
327 static void scc_mgr_set_dq_out1_delay(uint32_t dq_in_group, uint32_t delay)
328 {
329 scc_mgr_set(SCC_MGR_IO_OUT1_DELAY_OFFSET, dq_in_group, delay);
330 }
331
332 static void scc_mgr_set_dqs_out1_delay(uint32_t delay)
333 {
334 scc_mgr_set(SCC_MGR_IO_OUT1_DELAY_OFFSET, RW_MGR_MEM_DQ_PER_WRITE_DQS,
335 delay);
336 }
337
338 static void scc_mgr_set_dm_out1_delay(uint32_t dm, uint32_t delay)
339 {
340 scc_mgr_set(SCC_MGR_IO_OUT1_DELAY_OFFSET,
341 RW_MGR_MEM_DQ_PER_WRITE_DQS + 1 + dm,
342 delay);
343 }
344
345 /* load up dqs config settings */
346 static void scc_mgr_load_dqs(uint32_t dqs)
347 {
348 writel(dqs, &sdr_scc_mgr->dqs_ena);
349 }
350
351 /* load up dqs io config settings */
352 static void scc_mgr_load_dqs_io(void)
353 {
354 writel(0, &sdr_scc_mgr->dqs_io_ena);
355 }
356
357 /* load up dq config settings */
358 static void scc_mgr_load_dq(uint32_t dq_in_group)
359 {
360 writel(dq_in_group, &sdr_scc_mgr->dq_ena);
361 }
362
363 /* load up dm config settings */
364 static void scc_mgr_load_dm(uint32_t dm)
365 {
366 writel(dm, &sdr_scc_mgr->dm_ena);
367 }
368
369 /**
370 * scc_mgr_set_all_ranks() - Set SCC Manager register for all ranks
371 * @off: Base offset in SCC Manager space
372 * @grp: Read/Write group
373 * @val: Value to be set
374 * @update: If non-zero, trigger SCC Manager update for all ranks
375 *
376 * This function sets the SCC Manager (Scan Chain Control Manager) register
377 * and optionally triggers the SCC update for all ranks.
378 */
379 static void scc_mgr_set_all_ranks(const u32 off, const u32 grp, const u32 val,
380 const int update)
381 {
382 u32 r;
383
384 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
385 r += NUM_RANKS_PER_SHADOW_REG) {
386 scc_mgr_set(off, grp, val);
387
388 if (update || (r == 0)) {
389 writel(grp, &sdr_scc_mgr->dqs_ena);
390 writel(0, &sdr_scc_mgr->update);
391 }
392 }
393 }
394
395 static void scc_mgr_set_dqs_en_phase_all_ranks(u32 read_group, u32 phase)
396 {
397 /*
398 * USER although the h/w doesn't support different phases per
399 * shadow register, for simplicity our scc manager modeling
400 * keeps different phase settings per shadow reg, and it's
401 * important for us to keep them in sync to match h/w.
402 * for efficiency, the scan chain update should occur only
403 * once to sr0.
404 */
405 scc_mgr_set_all_ranks(SCC_MGR_DQS_EN_PHASE_OFFSET,
406 read_group, phase, 0);
407 }
408
409 static void scc_mgr_set_dqdqs_output_phase_all_ranks(uint32_t write_group,
410 uint32_t phase)
411 {
412 /*
413 * USER although the h/w doesn't support different phases per
414 * shadow register, for simplicity our scc manager modeling
415 * keeps different phase settings per shadow reg, and it's
416 * important for us to keep them in sync to match h/w.
417 * for efficiency, the scan chain update should occur only
418 * once to sr0.
419 */
420 scc_mgr_set_all_ranks(SCC_MGR_DQDQS_OUT_PHASE_OFFSET,
421 write_group, phase, 0);
422 }
423
424 static void scc_mgr_set_dqs_en_delay_all_ranks(uint32_t read_group,
425 uint32_t delay)
426 {
427 /*
428 * In shadow register mode, the T11 settings are stored in
429 * registers in the core, which are updated by the DQS_ENA
430 * signals. Not issuing the SCC_MGR_UPD command allows us to
431 * save lots of rank switching overhead, by calling
432 * select_shadow_regs_for_update with update_scan_chains
433 * set to 0.
434 */
435 scc_mgr_set_all_ranks(SCC_MGR_DQS_EN_DELAY_OFFSET,
436 read_group, delay, 1);
437 writel(0, &sdr_scc_mgr->update);
438 }
439
440 /**
441 * scc_mgr_set_oct_out1_delay() - Set OCT output delay
442 * @write_group: Write group
443 * @delay: Delay value
444 *
445 * This function sets the OCT output delay in SCC manager.
446 */
447 static void scc_mgr_set_oct_out1_delay(const u32 write_group, const u32 delay)
448 {
449 const int ratio = RW_MGR_MEM_IF_READ_DQS_WIDTH /
450 RW_MGR_MEM_IF_WRITE_DQS_WIDTH;
451 const int base = write_group * ratio;
452 int i;
453 /*
454 * Load the setting in the SCC manager
455 * Although OCT affects only write data, the OCT delay is controlled
456 * by the DQS logic block which is instantiated once per read group.
457 * For protocols where a write group consists of multiple read groups,
458 * the setting must be set multiple times.
459 */
460 for (i = 0; i < ratio; i++)
461 scc_mgr_set(SCC_MGR_OCT_OUT1_DELAY_OFFSET, base + i, delay);
462 }
463
464 /**
465 * scc_mgr_set_hhp_extras() - Set HHP extras.
466 *
467 * Load the fixed setting in the SCC manager HHP extras.
468 */
469 static void scc_mgr_set_hhp_extras(void)
470 {
471 /*
472 * Load the fixed setting in the SCC manager
473 * bits: 0:0 = 1'b1 - DQS bypass
474 * bits: 1:1 = 1'b1 - DQ bypass
475 * bits: 4:2 = 3'b001 - rfifo_mode
476 * bits: 6:5 = 2'b01 - rfifo clock_select
477 * bits: 7:7 = 1'b0 - separate gating from ungating setting
478 * bits: 8:8 = 1'b0 - separate OE from Output delay setting
479 */
480 const u32 value = (0 << 8) | (0 << 7) | (1 << 5) |
481 (1 << 2) | (1 << 1) | (1 << 0);
482 const u32 addr = SDR_PHYGRP_SCCGRP_ADDRESS |
483 SCC_MGR_HHP_GLOBALS_OFFSET |
484 SCC_MGR_HHP_EXTRAS_OFFSET;
485
486 debug_cond(DLEVEL == 1, "%s:%d Setting HHP Extras\n",
487 __func__, __LINE__);
488 writel(value, addr);
489 debug_cond(DLEVEL == 1, "%s:%d Done Setting HHP Extras\n",
490 __func__, __LINE__);
491 }
492
493 /**
494 * scc_mgr_zero_all() - Zero all DQS config
495 *
496 * Zero all DQS config.
497 */
498 static void scc_mgr_zero_all(void)
499 {
500 int i, r;
501
502 /*
503 * USER Zero all DQS config settings, across all groups and all
504 * shadow registers
505 */
506 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
507 r += NUM_RANKS_PER_SHADOW_REG) {
508 for (i = 0; i < RW_MGR_MEM_IF_READ_DQS_WIDTH; i++) {
509 /*
510 * The phases actually don't exist on a per-rank basis,
511 * but there's no harm updating them several times, so
512 * let's keep the code simple.
513 */
514 scc_mgr_set_dqs_bus_in_delay(i, IO_DQS_IN_RESERVE);
515 scc_mgr_set_dqs_en_phase(i, 0);
516 scc_mgr_set_dqs_en_delay(i, 0);
517 }
518
519 for (i = 0; i < RW_MGR_MEM_IF_WRITE_DQS_WIDTH; i++) {
520 scc_mgr_set_dqdqs_output_phase(i, 0);
521 /* Arria V/Cyclone V don't have out2. */
522 scc_mgr_set_oct_out1_delay(i, IO_DQS_OUT_RESERVE);
523 }
524 }
525
526 /* Multicast to all DQS group enables. */
527 writel(0xff, &sdr_scc_mgr->dqs_ena);
528 writel(0, &sdr_scc_mgr->update);
529 }
530
531 /**
532 * scc_set_bypass_mode() - Set bypass mode and trigger SCC update
533 * @write_group: Write group
534 *
535 * Set bypass mode and trigger SCC update.
536 */
537 static void scc_set_bypass_mode(const u32 write_group)
538 {
539 /* Multicast to all DQ enables. */
540 writel(0xff, &sdr_scc_mgr->dq_ena);
541 writel(0xff, &sdr_scc_mgr->dm_ena);
542
543 /* Update current DQS IO enable. */
544 writel(0, &sdr_scc_mgr->dqs_io_ena);
545
546 /* Update the DQS logic. */
547 writel(write_group, &sdr_scc_mgr->dqs_ena);
548
549 /* Hit update. */
550 writel(0, &sdr_scc_mgr->update);
551 }
552
553 /**
554 * scc_mgr_load_dqs_for_write_group() - Load DQS settings for Write Group
555 * @write_group: Write group
556 *
557 * Load DQS settings for Write Group, do not trigger SCC update.
558 */
559 static void scc_mgr_load_dqs_for_write_group(const u32 write_group)
560 {
561 const int ratio = RW_MGR_MEM_IF_READ_DQS_WIDTH /
562 RW_MGR_MEM_IF_WRITE_DQS_WIDTH;
563 const int base = write_group * ratio;
564 int i;
565 /*
566 * Load the setting in the SCC manager
567 * Although OCT affects only write data, the OCT delay is controlled
568 * by the DQS logic block which is instantiated once per read group.
569 * For protocols where a write group consists of multiple read groups,
570 * the setting must be set multiple times.
571 */
572 for (i = 0; i < ratio; i++)
573 writel(base + i, &sdr_scc_mgr->dqs_ena);
574 }
575
576 /**
577 * scc_mgr_zero_group() - Zero all configs for a group
578 *
579 * Zero DQ, DM, DQS and OCT configs for a group.
580 */
581 static void scc_mgr_zero_group(const u32 write_group, const int out_only)
582 {
583 int i, r;
584
585 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
586 r += NUM_RANKS_PER_SHADOW_REG) {
587 /* Zero all DQ config settings. */
588 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
589 scc_mgr_set_dq_out1_delay(i, 0);
590 if (!out_only)
591 scc_mgr_set_dq_in_delay(i, 0);
592 }
593
594 /* Multicast to all DQ enables. */
595 writel(0xff, &sdr_scc_mgr->dq_ena);
596
597 /* Zero all DM config settings. */
598 for (i = 0; i < RW_MGR_NUM_DM_PER_WRITE_GROUP; i++)
599 scc_mgr_set_dm_out1_delay(i, 0);
600
601 /* Multicast to all DM enables. */
602 writel(0xff, &sdr_scc_mgr->dm_ena);
603
604 /* Zero all DQS IO settings. */
605 if (!out_only)
606 scc_mgr_set_dqs_io_in_delay(0);
607
608 /* Arria V/Cyclone V don't have out2. */
609 scc_mgr_set_dqs_out1_delay(IO_DQS_OUT_RESERVE);
610 scc_mgr_set_oct_out1_delay(write_group, IO_DQS_OUT_RESERVE);
611 scc_mgr_load_dqs_for_write_group(write_group);
612
613 /* Multicast to all DQS IO enables (only 1 in total). */
614 writel(0, &sdr_scc_mgr->dqs_io_ena);
615
616 /* Hit update to zero everything. */
617 writel(0, &sdr_scc_mgr->update);
618 }
619 }
620
621 /*
622 * apply and load a particular input delay for the DQ pins in a group
623 * group_bgn is the index of the first dq pin (in the write group)
624 */
625 static void scc_mgr_apply_group_dq_in_delay(uint32_t group_bgn, uint32_t delay)
626 {
627 uint32_t i, p;
628
629 for (i = 0, p = group_bgn; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++, p++) {
630 scc_mgr_set_dq_in_delay(p, delay);
631 scc_mgr_load_dq(p);
632 }
633 }
634
635 /**
636 * scc_mgr_apply_group_dq_out1_delay() - Apply and load an output delay for the DQ pins in a group
637 * @delay: Delay value
638 *
639 * Apply and load a particular output delay for the DQ pins in a group.
640 */
641 static void scc_mgr_apply_group_dq_out1_delay(const u32 delay)
642 {
643 int i;
644
645 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
646 scc_mgr_set_dq_out1_delay(i, delay);
647 scc_mgr_load_dq(i);
648 }
649 }
650
651 /* apply and load a particular output delay for the DM pins in a group */
652 static void scc_mgr_apply_group_dm_out1_delay(uint32_t delay1)
653 {
654 uint32_t i;
655
656 for (i = 0; i < RW_MGR_NUM_DM_PER_WRITE_GROUP; i++) {
657 scc_mgr_set_dm_out1_delay(i, delay1);
658 scc_mgr_load_dm(i);
659 }
660 }
661
662
663 /* apply and load delay on both DQS and OCT out1 */
664 static void scc_mgr_apply_group_dqs_io_and_oct_out1(uint32_t write_group,
665 uint32_t delay)
666 {
667 scc_mgr_set_dqs_out1_delay(delay);
668 scc_mgr_load_dqs_io();
669
670 scc_mgr_set_oct_out1_delay(write_group, delay);
671 scc_mgr_load_dqs_for_write_group(write_group);
672 }
673
674 /**
675 * scc_mgr_apply_group_all_out_delay_add() - Apply a delay to the entire output side: DQ, DM, DQS, OCT
676 * @write_group: Write group
677 * @delay: Delay value
678 *
679 * Apply a delay to the entire output side: DQ, DM, DQS, OCT.
680 */
681 static void scc_mgr_apply_group_all_out_delay_add(const u32 write_group,
682 const u32 delay)
683 {
684 u32 i, new_delay;
685
686 /* DQ shift */
687 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++)
688 scc_mgr_load_dq(i);
689
690 /* DM shift */
691 for (i = 0; i < RW_MGR_NUM_DM_PER_WRITE_GROUP; i++)
692 scc_mgr_load_dm(i);
693
694 /* DQS shift */
695 new_delay = READ_SCC_DQS_IO_OUT2_DELAY + delay;
696 if (new_delay > IO_IO_OUT2_DELAY_MAX) {
697 debug_cond(DLEVEL == 1,
698 "%s:%d (%u, %u) DQS: %u > %d; adding %u to OUT1\n",
699 __func__, __LINE__, write_group, delay, new_delay,
700 IO_IO_OUT2_DELAY_MAX,
701 new_delay - IO_IO_OUT2_DELAY_MAX);
702 new_delay -= IO_IO_OUT2_DELAY_MAX;
703 scc_mgr_set_dqs_out1_delay(new_delay);
704 }
705
706 scc_mgr_load_dqs_io();
707
708 /* OCT shift */
709 new_delay = READ_SCC_OCT_OUT2_DELAY + delay;
710 if (new_delay > IO_IO_OUT2_DELAY_MAX) {
711 debug_cond(DLEVEL == 1,
712 "%s:%d (%u, %u) DQS: %u > %d; adding %u to OUT1\n",
713 __func__, __LINE__, write_group, delay,
714 new_delay, IO_IO_OUT2_DELAY_MAX,
715 new_delay - IO_IO_OUT2_DELAY_MAX);
716 new_delay -= IO_IO_OUT2_DELAY_MAX;
717 scc_mgr_set_oct_out1_delay(write_group, new_delay);
718 }
719
720 scc_mgr_load_dqs_for_write_group(write_group);
721 }
722
723 /**
724 * scc_mgr_apply_group_all_out_delay_add() - Apply a delay to the entire output side to all ranks
725 * @write_group: Write group
726 * @delay: Delay value
727 *
728 * Apply a delay to the entire output side (DQ, DM, DQS, OCT) to all ranks.
729 */
730 static void
731 scc_mgr_apply_group_all_out_delay_add_all_ranks(const u32 write_group,
732 const u32 delay)
733 {
734 int r;
735
736 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
737 r += NUM_RANKS_PER_SHADOW_REG) {
738 scc_mgr_apply_group_all_out_delay_add(write_group, delay);
739 writel(0, &sdr_scc_mgr->update);
740 }
741 }
742
743 /**
744 * set_jump_as_return() - Return instruction optimization
745 *
746 * Optimization used to recover some slots in ddr3 inst_rom could be
747 * applied to other protocols if we wanted to
748 */
749 static void set_jump_as_return(void)
750 {
751 /*
752 * To save space, we replace return with jump to special shared
753 * RETURN instruction so we set the counter to large value so that
754 * we always jump.
755 */
756 writel(0xff, &sdr_rw_load_mgr_regs->load_cntr0);
757 writel(RW_MGR_RETURN, &sdr_rw_load_jump_mgr_regs->load_jump_add0);
758 }
759
760 /*
761 * should always use constants as argument to ensure all computations are
762 * performed at compile time
763 */
764 static void delay_for_n_mem_clocks(const uint32_t clocks)
765 {
766 uint32_t afi_clocks;
767 uint8_t inner = 0;
768 uint8_t outer = 0;
769 uint16_t c_loop = 0;
770
771 debug("%s:%d: clocks=%u ... start\n", __func__, __LINE__, clocks);
772
773
774 afi_clocks = (clocks + AFI_RATE_RATIO-1) / AFI_RATE_RATIO;
775 /* scale (rounding up) to get afi clocks */
776
777 /*
778 * Note, we don't bother accounting for being off a little bit
779 * because of a few extra instructions in outer loops
780 * Note, the loops have a test at the end, and do the test before
781 * the decrement, and so always perform the loop
782 * 1 time more than the counter value
783 */
784 if (afi_clocks == 0) {
785 ;
786 } else if (afi_clocks <= 0x100) {
787 inner = afi_clocks-1;
788 outer = 0;
789 c_loop = 0;
790 } else if (afi_clocks <= 0x10000) {
791 inner = 0xff;
792 outer = (afi_clocks-1) >> 8;
793 c_loop = 0;
794 } else {
795 inner = 0xff;
796 outer = 0xff;
797 c_loop = (afi_clocks-1) >> 16;
798 }
799
800 /*
801 * rom instructions are structured as follows:
802 *
803 * IDLE_LOOP2: jnz cntr0, TARGET_A
804 * IDLE_LOOP1: jnz cntr1, TARGET_B
805 * return
806 *
807 * so, when doing nested loops, TARGET_A is set to IDLE_LOOP2, and
808 * TARGET_B is set to IDLE_LOOP2 as well
809 *
810 * if we have no outer loop, though, then we can use IDLE_LOOP1 only,
811 * and set TARGET_B to IDLE_LOOP1 and we skip IDLE_LOOP2 entirely
812 *
813 * a little confusing, but it helps save precious space in the inst_rom
814 * and sequencer rom and keeps the delays more accurate and reduces
815 * overhead
816 */
817 if (afi_clocks <= 0x100) {
818 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(inner),
819 &sdr_rw_load_mgr_regs->load_cntr1);
820
821 writel(RW_MGR_IDLE_LOOP1,
822 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
823
824 writel(RW_MGR_IDLE_LOOP1, SDR_PHYGRP_RWMGRGRP_ADDRESS |
825 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
826 } else {
827 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(inner),
828 &sdr_rw_load_mgr_regs->load_cntr0);
829
830 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(outer),
831 &sdr_rw_load_mgr_regs->load_cntr1);
832
833 writel(RW_MGR_IDLE_LOOP2,
834 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
835
836 writel(RW_MGR_IDLE_LOOP2,
837 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
838
839 /* hack to get around compiler not being smart enough */
840 if (afi_clocks <= 0x10000) {
841 /* only need to run once */
842 writel(RW_MGR_IDLE_LOOP2, SDR_PHYGRP_RWMGRGRP_ADDRESS |
843 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
844 } else {
845 do {
846 writel(RW_MGR_IDLE_LOOP2,
847 SDR_PHYGRP_RWMGRGRP_ADDRESS |
848 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
849 } while (c_loop-- != 0);
850 }
851 }
852 debug("%s:%d clocks=%u ... end\n", __func__, __LINE__, clocks);
853 }
854
855 /**
856 * rw_mgr_mem_init_load_regs() - Load instruction registers
857 * @cntr0: Counter 0 value
858 * @cntr1: Counter 1 value
859 * @cntr2: Counter 2 value
860 * @jump: Jump instruction value
861 *
862 * Load instruction registers.
863 */
864 static void rw_mgr_mem_init_load_regs(u32 cntr0, u32 cntr1, u32 cntr2, u32 jump)
865 {
866 uint32_t grpaddr = SDR_PHYGRP_RWMGRGRP_ADDRESS |
867 RW_MGR_RUN_SINGLE_GROUP_OFFSET;
868
869 /* Load counters */
870 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(cntr0),
871 &sdr_rw_load_mgr_regs->load_cntr0);
872 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(cntr1),
873 &sdr_rw_load_mgr_regs->load_cntr1);
874 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(cntr2),
875 &sdr_rw_load_mgr_regs->load_cntr2);
876
877 /* Load jump address */
878 writel(jump, &sdr_rw_load_jump_mgr_regs->load_jump_add0);
879 writel(jump, &sdr_rw_load_jump_mgr_regs->load_jump_add1);
880 writel(jump, &sdr_rw_load_jump_mgr_regs->load_jump_add2);
881
882 /* Execute count instruction */
883 writel(jump, grpaddr);
884 }
885
886 /**
887 * rw_mgr_mem_load_user() - Load user calibration values
888 * @fin1: Final instruction 1
889 * @fin2: Final instruction 2
890 * @precharge: If 1, precharge the banks at the end
891 *
892 * Load user calibration values and optionally precharge the banks.
893 */
894 static void rw_mgr_mem_load_user(const u32 fin1, const u32 fin2,
895 const int precharge)
896 {
897 u32 grpaddr = SDR_PHYGRP_RWMGRGRP_ADDRESS |
898 RW_MGR_RUN_SINGLE_GROUP_OFFSET;
899 u32 r;
900
901 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS; r++) {
902 if (param->skip_ranks[r]) {
903 /* request to skip the rank */
904 continue;
905 }
906
907 /* set rank */
908 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_OFF);
909
910 /* precharge all banks ... */
911 if (precharge)
912 writel(RW_MGR_PRECHARGE_ALL, grpaddr);
913
914 /*
915 * USER Use Mirror-ed commands for odd ranks if address
916 * mirrorring is on
917 */
918 if ((RW_MGR_MEM_ADDRESS_MIRRORING >> r) & 0x1) {
919 set_jump_as_return();
920 writel(RW_MGR_MRS2_MIRR, grpaddr);
921 delay_for_n_mem_clocks(4);
922 set_jump_as_return();
923 writel(RW_MGR_MRS3_MIRR, grpaddr);
924 delay_for_n_mem_clocks(4);
925 set_jump_as_return();
926 writel(RW_MGR_MRS1_MIRR, grpaddr);
927 delay_for_n_mem_clocks(4);
928 set_jump_as_return();
929 writel(fin1, grpaddr);
930 } else {
931 set_jump_as_return();
932 writel(RW_MGR_MRS2, grpaddr);
933 delay_for_n_mem_clocks(4);
934 set_jump_as_return();
935 writel(RW_MGR_MRS3, grpaddr);
936 delay_for_n_mem_clocks(4);
937 set_jump_as_return();
938 writel(RW_MGR_MRS1, grpaddr);
939 set_jump_as_return();
940 writel(fin2, grpaddr);
941 }
942
943 if (precharge)
944 continue;
945
946 set_jump_as_return();
947 writel(RW_MGR_ZQCL, grpaddr);
948
949 /* tZQinit = tDLLK = 512 ck cycles */
950 delay_for_n_mem_clocks(512);
951 }
952 }
953
954 static void rw_mgr_mem_initialize(void)
955 {
956 debug("%s:%d\n", __func__, __LINE__);
957
958 /* The reset / cke part of initialization is broadcasted to all ranks */
959 writel(RW_MGR_RANK_ALL, SDR_PHYGRP_RWMGRGRP_ADDRESS |
960 RW_MGR_SET_CS_AND_ODT_MASK_OFFSET);
961
962 /*
963 * Here's how you load register for a loop
964 * Counters are located @ 0x800
965 * Jump address are located @ 0xC00
966 * For both, registers 0 to 3 are selected using bits 3 and 2, like
967 * in 0x800, 0x804, 0x808, 0x80C and 0xC00, 0xC04, 0xC08, 0xC0C
968 * I know this ain't pretty, but Avalon bus throws away the 2 least
969 * significant bits
970 */
971
972 /* start with memory RESET activated */
973
974 /* tINIT = 200us */
975
976 /*
977 * 200us @ 266MHz (3.75 ns) ~ 54000 clock cycles
978 * If a and b are the number of iteration in 2 nested loops
979 * it takes the following number of cycles to complete the operation:
980 * number_of_cycles = ((2 + n) * a + 2) * b
981 * where n is the number of instruction in the inner loop
982 * One possible solution is n = 0 , a = 256 , b = 106 => a = FF,
983 * b = 6A
984 */
985 rw_mgr_mem_init_load_regs(SEQ_TINIT_CNTR0_VAL, SEQ_TINIT_CNTR1_VAL,
986 SEQ_TINIT_CNTR2_VAL,
987 RW_MGR_INIT_RESET_0_CKE_0);
988
989 /* indicate that memory is stable */
990 writel(1, &phy_mgr_cfg->reset_mem_stbl);
991
992 /*
993 * transition the RESET to high
994 * Wait for 500us
995 */
996
997 /*
998 * 500us @ 266MHz (3.75 ns) ~ 134000 clock cycles
999 * If a and b are the number of iteration in 2 nested loops
1000 * it takes the following number of cycles to complete the operation
1001 * number_of_cycles = ((2 + n) * a + 2) * b
1002 * where n is the number of instruction in the inner loop
1003 * One possible solution is n = 2 , a = 131 , b = 256 => a = 83,
1004 * b = FF
1005 */
1006 rw_mgr_mem_init_load_regs(SEQ_TRESET_CNTR0_VAL, SEQ_TRESET_CNTR1_VAL,
1007 SEQ_TRESET_CNTR2_VAL,
1008 RW_MGR_INIT_RESET_1_CKE_0);
1009
1010 /* bring up clock enable */
1011
1012 /* tXRP < 250 ck cycles */
1013 delay_for_n_mem_clocks(250);
1014
1015 rw_mgr_mem_load_user(RW_MGR_MRS0_DLL_RESET_MIRR, RW_MGR_MRS0_DLL_RESET,
1016 0);
1017 }
1018
1019 /*
1020 * At the end of calibration we have to program the user settings in, and
1021 * USER hand off the memory to the user.
1022 */
1023 static void rw_mgr_mem_handoff(void)
1024 {
1025 rw_mgr_mem_load_user(RW_MGR_MRS0_USER_MIRR, RW_MGR_MRS0_USER, 1);
1026 /*
1027 * USER need to wait tMOD (12CK or 15ns) time before issuing
1028 * other commands, but we will have plenty of NIOS cycles before
1029 * actual handoff so its okay.
1030 */
1031 }
1032
1033 /*
1034 * performs a guaranteed read on the patterns we are going to use during a
1035 * read test to ensure memory works
1036 */
1037 static uint32_t rw_mgr_mem_calibrate_read_test_patterns(uint32_t rank_bgn,
1038 uint32_t group, uint32_t num_tries, uint32_t *bit_chk,
1039 uint32_t all_ranks)
1040 {
1041 uint32_t r, vg;
1042 uint32_t correct_mask_vg;
1043 uint32_t tmp_bit_chk;
1044 uint32_t rank_end = all_ranks ? RW_MGR_MEM_NUMBER_OF_RANKS :
1045 (rank_bgn + NUM_RANKS_PER_SHADOW_REG);
1046 uint32_t addr;
1047 uint32_t base_rw_mgr;
1048
1049 *bit_chk = param->read_correct_mask;
1050 correct_mask_vg = param->read_correct_mask_vg;
1051
1052 for (r = rank_bgn; r < rank_end; r++) {
1053 if (param->skip_ranks[r])
1054 /* request to skip the rank */
1055 continue;
1056
1057 /* set rank */
1058 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_READ_WRITE);
1059
1060 /* Load up a constant bursts of read commands */
1061 writel(0x20, &sdr_rw_load_mgr_regs->load_cntr0);
1062 writel(RW_MGR_GUARANTEED_READ,
1063 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
1064
1065 writel(0x20, &sdr_rw_load_mgr_regs->load_cntr1);
1066 writel(RW_MGR_GUARANTEED_READ_CONT,
1067 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
1068
1069 tmp_bit_chk = 0;
1070 for (vg = RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS-1; ; vg--) {
1071 /* reset the fifos to get pointers to known state */
1072
1073 writel(0, &phy_mgr_cmd->fifo_reset);
1074 writel(0, SDR_PHYGRP_RWMGRGRP_ADDRESS |
1075 RW_MGR_RESET_READ_DATAPATH_OFFSET);
1076
1077 tmp_bit_chk = tmp_bit_chk << (RW_MGR_MEM_DQ_PER_READ_DQS
1078 / RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS);
1079
1080 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
1081 writel(RW_MGR_GUARANTEED_READ, addr +
1082 ((group * RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS +
1083 vg) << 2));
1084
1085 base_rw_mgr = readl(SDR_PHYGRP_RWMGRGRP_ADDRESS);
1086 tmp_bit_chk = tmp_bit_chk | (correct_mask_vg & (~base_rw_mgr));
1087
1088 if (vg == 0)
1089 break;
1090 }
1091 *bit_chk &= tmp_bit_chk;
1092 }
1093
1094 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
1095 writel(RW_MGR_CLEAR_DQS_ENABLE, addr + (group << 2));
1096
1097 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
1098 debug_cond(DLEVEL == 1, "%s:%d test_load_patterns(%u,ALL) => (%u == %u) =>\
1099 %lu\n", __func__, __LINE__, group, *bit_chk, param->read_correct_mask,
1100 (long unsigned int)(*bit_chk == param->read_correct_mask));
1101 return *bit_chk == param->read_correct_mask;
1102 }
1103
1104 static uint32_t rw_mgr_mem_calibrate_read_test_patterns_all_ranks
1105 (uint32_t group, uint32_t num_tries, uint32_t *bit_chk)
1106 {
1107 return rw_mgr_mem_calibrate_read_test_patterns(0, group,
1108 num_tries, bit_chk, 1);
1109 }
1110
1111 /* load up the patterns we are going to use during a read test */
1112 static void rw_mgr_mem_calibrate_read_load_patterns(uint32_t rank_bgn,
1113 uint32_t all_ranks)
1114 {
1115 uint32_t r;
1116 uint32_t rank_end = all_ranks ? RW_MGR_MEM_NUMBER_OF_RANKS :
1117 (rank_bgn + NUM_RANKS_PER_SHADOW_REG);
1118
1119 debug("%s:%d\n", __func__, __LINE__);
1120 for (r = rank_bgn; r < rank_end; r++) {
1121 if (param->skip_ranks[r])
1122 /* request to skip the rank */
1123 continue;
1124
1125 /* set rank */
1126 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_READ_WRITE);
1127
1128 /* Load up a constant bursts */
1129 writel(0x20, &sdr_rw_load_mgr_regs->load_cntr0);
1130
1131 writel(RW_MGR_GUARANTEED_WRITE_WAIT0,
1132 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
1133
1134 writel(0x20, &sdr_rw_load_mgr_regs->load_cntr1);
1135
1136 writel(RW_MGR_GUARANTEED_WRITE_WAIT1,
1137 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
1138
1139 writel(0x04, &sdr_rw_load_mgr_regs->load_cntr2);
1140
1141 writel(RW_MGR_GUARANTEED_WRITE_WAIT2,
1142 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
1143
1144 writel(0x04, &sdr_rw_load_mgr_regs->load_cntr3);
1145
1146 writel(RW_MGR_GUARANTEED_WRITE_WAIT3,
1147 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
1148
1149 writel(RW_MGR_GUARANTEED_WRITE, SDR_PHYGRP_RWMGRGRP_ADDRESS |
1150 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
1151 }
1152
1153 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
1154 }
1155
1156 /*
1157 * try a read and see if it returns correct data back. has dummy reads
1158 * inserted into the mix used to align dqs enable. has more thorough checks
1159 * than the regular read test.
1160 */
1161 static uint32_t rw_mgr_mem_calibrate_read_test(uint32_t rank_bgn, uint32_t group,
1162 uint32_t num_tries, uint32_t all_correct, uint32_t *bit_chk,
1163 uint32_t all_groups, uint32_t all_ranks)
1164 {
1165 uint32_t r, vg;
1166 uint32_t correct_mask_vg;
1167 uint32_t tmp_bit_chk;
1168 uint32_t rank_end = all_ranks ? RW_MGR_MEM_NUMBER_OF_RANKS :
1169 (rank_bgn + NUM_RANKS_PER_SHADOW_REG);
1170 uint32_t addr;
1171 uint32_t base_rw_mgr;
1172
1173 *bit_chk = param->read_correct_mask;
1174 correct_mask_vg = param->read_correct_mask_vg;
1175
1176 uint32_t quick_read_mode = (((STATIC_CALIB_STEPS) &
1177 CALIB_SKIP_DELAY_SWEEPS) && ENABLE_SUPER_QUICK_CALIBRATION);
1178
1179 for (r = rank_bgn; r < rank_end; r++) {
1180 if (param->skip_ranks[r])
1181 /* request to skip the rank */
1182 continue;
1183
1184 /* set rank */
1185 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_READ_WRITE);
1186
1187 writel(0x10, &sdr_rw_load_mgr_regs->load_cntr1);
1188
1189 writel(RW_MGR_READ_B2B_WAIT1,
1190 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
1191
1192 writel(0x10, &sdr_rw_load_mgr_regs->load_cntr2);
1193 writel(RW_MGR_READ_B2B_WAIT2,
1194 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
1195
1196 if (quick_read_mode)
1197 writel(0x1, &sdr_rw_load_mgr_regs->load_cntr0);
1198 /* need at least two (1+1) reads to capture failures */
1199 else if (all_groups)
1200 writel(0x06, &sdr_rw_load_mgr_regs->load_cntr0);
1201 else
1202 writel(0x32, &sdr_rw_load_mgr_regs->load_cntr0);
1203
1204 writel(RW_MGR_READ_B2B,
1205 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
1206 if (all_groups)
1207 writel(RW_MGR_MEM_IF_READ_DQS_WIDTH *
1208 RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS - 1,
1209 &sdr_rw_load_mgr_regs->load_cntr3);
1210 else
1211 writel(0x0, &sdr_rw_load_mgr_regs->load_cntr3);
1212
1213 writel(RW_MGR_READ_B2B,
1214 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
1215
1216 tmp_bit_chk = 0;
1217 for (vg = RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS-1; ; vg--) {
1218 /* reset the fifos to get pointers to known state */
1219 writel(0, &phy_mgr_cmd->fifo_reset);
1220 writel(0, SDR_PHYGRP_RWMGRGRP_ADDRESS |
1221 RW_MGR_RESET_READ_DATAPATH_OFFSET);
1222
1223 tmp_bit_chk = tmp_bit_chk << (RW_MGR_MEM_DQ_PER_READ_DQS
1224 / RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS);
1225
1226 if (all_groups)
1227 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_ALL_GROUPS_OFFSET;
1228 else
1229 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
1230
1231 writel(RW_MGR_READ_B2B, addr +
1232 ((group * RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS +
1233 vg) << 2));
1234
1235 base_rw_mgr = readl(SDR_PHYGRP_RWMGRGRP_ADDRESS);
1236 tmp_bit_chk = tmp_bit_chk | (correct_mask_vg & ~(base_rw_mgr));
1237
1238 if (vg == 0)
1239 break;
1240 }
1241 *bit_chk &= tmp_bit_chk;
1242 }
1243
1244 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
1245 writel(RW_MGR_CLEAR_DQS_ENABLE, addr + (group << 2));
1246
1247 if (all_correct) {
1248 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
1249 debug_cond(DLEVEL == 2, "%s:%d read_test(%u,ALL,%u) =>\
1250 (%u == %u) => %lu", __func__, __LINE__, group,
1251 all_groups, *bit_chk, param->read_correct_mask,
1252 (long unsigned int)(*bit_chk ==
1253 param->read_correct_mask));
1254 return *bit_chk == param->read_correct_mask;
1255 } else {
1256 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
1257 debug_cond(DLEVEL == 2, "%s:%d read_test(%u,ONE,%u) =>\
1258 (%u != %lu) => %lu\n", __func__, __LINE__,
1259 group, all_groups, *bit_chk, (long unsigned int)0,
1260 (long unsigned int)(*bit_chk != 0x00));
1261 return *bit_chk != 0x00;
1262 }
1263 }
1264
1265 static uint32_t rw_mgr_mem_calibrate_read_test_all_ranks(uint32_t group,
1266 uint32_t num_tries, uint32_t all_correct, uint32_t *bit_chk,
1267 uint32_t all_groups)
1268 {
1269 return rw_mgr_mem_calibrate_read_test(0, group, num_tries, all_correct,
1270 bit_chk, all_groups, 1);
1271 }
1272
1273 static void rw_mgr_incr_vfifo(uint32_t grp, uint32_t *v)
1274 {
1275 writel(grp, &phy_mgr_cmd->inc_vfifo_hard_phy);
1276 (*v)++;
1277 }
1278
1279 static void rw_mgr_decr_vfifo(uint32_t grp, uint32_t *v)
1280 {
1281 uint32_t i;
1282
1283 for (i = 0; i < VFIFO_SIZE-1; i++)
1284 rw_mgr_incr_vfifo(grp, v);
1285 }
1286
1287 static int find_vfifo_read(uint32_t grp, uint32_t *bit_chk)
1288 {
1289 uint32_t v;
1290 uint32_t fail_cnt = 0;
1291 uint32_t test_status;
1292
1293 for (v = 0; v < VFIFO_SIZE; ) {
1294 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: vfifo %u\n",
1295 __func__, __LINE__, v);
1296 test_status = rw_mgr_mem_calibrate_read_test_all_ranks
1297 (grp, 1, PASS_ONE_BIT, bit_chk, 0);
1298 if (!test_status) {
1299 fail_cnt++;
1300
1301 if (fail_cnt == 2)
1302 break;
1303 }
1304
1305 /* fiddle with FIFO */
1306 rw_mgr_incr_vfifo(grp, &v);
1307 }
1308
1309 if (v >= VFIFO_SIZE) {
1310 /* no failing read found!! Something must have gone wrong */
1311 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: vfifo failed\n",
1312 __func__, __LINE__);
1313 return 0;
1314 } else {
1315 return v;
1316 }
1317 }
1318
1319 static int find_working_phase(uint32_t *grp, uint32_t *bit_chk,
1320 uint32_t dtaps_per_ptap, uint32_t *work_bgn,
1321 uint32_t *v, uint32_t *d, uint32_t *p,
1322 uint32_t *i, uint32_t *max_working_cnt)
1323 {
1324 uint32_t found_begin = 0;
1325 uint32_t tmp_delay = 0;
1326 uint32_t test_status;
1327
1328 for (*d = 0; *d <= dtaps_per_ptap; (*d)++, tmp_delay +=
1329 IO_DELAY_PER_DQS_EN_DCHAIN_TAP) {
1330 *work_bgn = tmp_delay;
1331 scc_mgr_set_dqs_en_delay_all_ranks(*grp, *d);
1332
1333 for (*i = 0; *i < VFIFO_SIZE; (*i)++) {
1334 for (*p = 0; *p <= IO_DQS_EN_PHASE_MAX; (*p)++, *work_bgn +=
1335 IO_DELAY_PER_OPA_TAP) {
1336 scc_mgr_set_dqs_en_phase_all_ranks(*grp, *p);
1337
1338 test_status =
1339 rw_mgr_mem_calibrate_read_test_all_ranks
1340 (*grp, 1, PASS_ONE_BIT, bit_chk, 0);
1341
1342 if (test_status) {
1343 *max_working_cnt = 1;
1344 found_begin = 1;
1345 break;
1346 }
1347 }
1348
1349 if (found_begin)
1350 break;
1351
1352 if (*p > IO_DQS_EN_PHASE_MAX)
1353 /* fiddle with FIFO */
1354 rw_mgr_incr_vfifo(*grp, v);
1355 }
1356
1357 if (found_begin)
1358 break;
1359 }
1360
1361 if (*i >= VFIFO_SIZE) {
1362 /* cannot find working solution */
1363 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: no vfifo/\
1364 ptap/dtap\n", __func__, __LINE__);
1365 return 0;
1366 } else {
1367 return 1;
1368 }
1369 }
1370
1371 static void sdr_backup_phase(uint32_t *grp, uint32_t *bit_chk,
1372 uint32_t *work_bgn, uint32_t *v, uint32_t *d,
1373 uint32_t *p, uint32_t *max_working_cnt)
1374 {
1375 uint32_t found_begin = 0;
1376 uint32_t tmp_delay;
1377
1378 /* Special case code for backing up a phase */
1379 if (*p == 0) {
1380 *p = IO_DQS_EN_PHASE_MAX;
1381 rw_mgr_decr_vfifo(*grp, v);
1382 } else {
1383 (*p)--;
1384 }
1385 tmp_delay = *work_bgn - IO_DELAY_PER_OPA_TAP;
1386 scc_mgr_set_dqs_en_phase_all_ranks(*grp, *p);
1387
1388 for (*d = 0; *d <= IO_DQS_EN_DELAY_MAX && tmp_delay < *work_bgn;
1389 (*d)++, tmp_delay += IO_DELAY_PER_DQS_EN_DCHAIN_TAP) {
1390 scc_mgr_set_dqs_en_delay_all_ranks(*grp, *d);
1391
1392 if (rw_mgr_mem_calibrate_read_test_all_ranks(*grp, 1,
1393 PASS_ONE_BIT,
1394 bit_chk, 0)) {
1395 found_begin = 1;
1396 *work_bgn = tmp_delay;
1397 break;
1398 }
1399 }
1400
1401 /* We have found a working dtap before the ptap found above */
1402 if (found_begin == 1)
1403 (*max_working_cnt)++;
1404
1405 /*
1406 * Restore VFIFO to old state before we decremented it
1407 * (if needed).
1408 */
1409 (*p)++;
1410 if (*p > IO_DQS_EN_PHASE_MAX) {
1411 *p = 0;
1412 rw_mgr_incr_vfifo(*grp, v);
1413 }
1414
1415 scc_mgr_set_dqs_en_delay_all_ranks(*grp, 0);
1416 }
1417
1418 static int sdr_nonworking_phase(uint32_t *grp, uint32_t *bit_chk,
1419 uint32_t *work_bgn, uint32_t *v, uint32_t *d,
1420 uint32_t *p, uint32_t *i, uint32_t *max_working_cnt,
1421 uint32_t *work_end)
1422 {
1423 uint32_t found_end = 0;
1424
1425 (*p)++;
1426 *work_end += IO_DELAY_PER_OPA_TAP;
1427 if (*p > IO_DQS_EN_PHASE_MAX) {
1428 /* fiddle with FIFO */
1429 *p = 0;
1430 rw_mgr_incr_vfifo(*grp, v);
1431 }
1432
1433 for (; *i < VFIFO_SIZE + 1; (*i)++) {
1434 for (; *p <= IO_DQS_EN_PHASE_MAX; (*p)++, *work_end
1435 += IO_DELAY_PER_OPA_TAP) {
1436 scc_mgr_set_dqs_en_phase_all_ranks(*grp, *p);
1437
1438 if (!rw_mgr_mem_calibrate_read_test_all_ranks
1439 (*grp, 1, PASS_ONE_BIT, bit_chk, 0)) {
1440 found_end = 1;
1441 break;
1442 } else {
1443 (*max_working_cnt)++;
1444 }
1445 }
1446
1447 if (found_end)
1448 break;
1449
1450 if (*p > IO_DQS_EN_PHASE_MAX) {
1451 /* fiddle with FIFO */
1452 rw_mgr_incr_vfifo(*grp, v);
1453 *p = 0;
1454 }
1455 }
1456
1457 if (*i >= VFIFO_SIZE + 1) {
1458 /* cannot see edge of failing read */
1459 debug_cond(DLEVEL == 2, "%s:%d sdr_nonworking_phase: end:\
1460 failed\n", __func__, __LINE__);
1461 return 0;
1462 } else {
1463 return 1;
1464 }
1465 }
1466
1467 static int sdr_find_window_centre(uint32_t *grp, uint32_t *bit_chk,
1468 uint32_t *work_bgn, uint32_t *v, uint32_t *d,
1469 uint32_t *p, uint32_t *work_mid,
1470 uint32_t *work_end)
1471 {
1472 int i;
1473 int tmp_delay = 0;
1474
1475 *work_mid = (*work_bgn + *work_end) / 2;
1476
1477 debug_cond(DLEVEL == 2, "work_bgn=%d work_end=%d work_mid=%d\n",
1478 *work_bgn, *work_end, *work_mid);
1479 /* Get the middle delay to be less than a VFIFO delay */
1480 for (*p = 0; *p <= IO_DQS_EN_PHASE_MAX;
1481 (*p)++, tmp_delay += IO_DELAY_PER_OPA_TAP)
1482 ;
1483 debug_cond(DLEVEL == 2, "vfifo ptap delay %d\n", tmp_delay);
1484 while (*work_mid > tmp_delay)
1485 *work_mid -= tmp_delay;
1486 debug_cond(DLEVEL == 2, "new work_mid %d\n", *work_mid);
1487
1488 tmp_delay = 0;
1489 for (*p = 0; *p <= IO_DQS_EN_PHASE_MAX && tmp_delay < *work_mid;
1490 (*p)++, tmp_delay += IO_DELAY_PER_OPA_TAP)
1491 ;
1492 tmp_delay -= IO_DELAY_PER_OPA_TAP;
1493 debug_cond(DLEVEL == 2, "new p %d, tmp_delay=%d\n", (*p) - 1, tmp_delay);
1494 for (*d = 0; *d <= IO_DQS_EN_DELAY_MAX && tmp_delay < *work_mid; (*d)++,
1495 tmp_delay += IO_DELAY_PER_DQS_EN_DCHAIN_TAP)
1496 ;
1497 debug_cond(DLEVEL == 2, "new d %d, tmp_delay=%d\n", *d, tmp_delay);
1498
1499 scc_mgr_set_dqs_en_phase_all_ranks(*grp, (*p) - 1);
1500 scc_mgr_set_dqs_en_delay_all_ranks(*grp, *d);
1501
1502 /*
1503 * push vfifo until we can successfully calibrate. We can do this
1504 * because the largest possible margin in 1 VFIFO cycle.
1505 */
1506 for (i = 0; i < VFIFO_SIZE; i++) {
1507 debug_cond(DLEVEL == 2, "find_dqs_en_phase: center: vfifo=%u\n",
1508 *v);
1509 if (rw_mgr_mem_calibrate_read_test_all_ranks(*grp, 1,
1510 PASS_ONE_BIT,
1511 bit_chk, 0)) {
1512 break;
1513 }
1514
1515 /* fiddle with FIFO */
1516 rw_mgr_incr_vfifo(*grp, v);
1517 }
1518
1519 if (i >= VFIFO_SIZE) {
1520 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: center: \
1521 failed\n", __func__, __LINE__);
1522 return 0;
1523 } else {
1524 return 1;
1525 }
1526 }
1527
1528 /* find a good dqs enable to use */
1529 static uint32_t rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase(uint32_t grp)
1530 {
1531 uint32_t v, d, p, i;
1532 uint32_t max_working_cnt;
1533 uint32_t bit_chk;
1534 uint32_t dtaps_per_ptap;
1535 uint32_t work_bgn, work_mid, work_end;
1536 uint32_t found_passing_read, found_failing_read, initial_failing_dtap;
1537
1538 debug("%s:%d %u\n", __func__, __LINE__, grp);
1539
1540 reg_file_set_sub_stage(CAL_SUBSTAGE_VFIFO_CENTER);
1541
1542 scc_mgr_set_dqs_en_delay_all_ranks(grp, 0);
1543 scc_mgr_set_dqs_en_phase_all_ranks(grp, 0);
1544
1545 /* ************************************************************** */
1546 /* * Step 0 : Determine number of delay taps for each phase tap * */
1547 dtaps_per_ptap = IO_DELAY_PER_OPA_TAP/IO_DELAY_PER_DQS_EN_DCHAIN_TAP;
1548
1549 /* ********************************************************* */
1550 /* * Step 1 : First push vfifo until we get a failing read * */
1551 v = find_vfifo_read(grp, &bit_chk);
1552
1553 max_working_cnt = 0;
1554
1555 /* ******************************************************** */
1556 /* * step 2: find first working phase, increment in ptaps * */
1557 work_bgn = 0;
1558 if (find_working_phase(&grp, &bit_chk, dtaps_per_ptap, &work_bgn, &v, &d,
1559 &p, &i, &max_working_cnt) == 0)
1560 return 0;
1561
1562 work_end = work_bgn;
1563
1564 /*
1565 * If d is 0 then the working window covers a phase tap and
1566 * we can follow the old procedure otherwise, we've found the beginning,
1567 * and we need to increment the dtaps until we find the end.
1568 */
1569 if (d == 0) {
1570 /* ********************************************************* */
1571 /* * step 3a: if we have room, back off by one and
1572 increment in dtaps * */
1573
1574 sdr_backup_phase(&grp, &bit_chk, &work_bgn, &v, &d, &p,
1575 &max_working_cnt);
1576
1577 /* ********************************************************* */
1578 /* * step 4a: go forward from working phase to non working
1579 phase, increment in ptaps * */
1580 if (sdr_nonworking_phase(&grp, &bit_chk, &work_bgn, &v, &d, &p,
1581 &i, &max_working_cnt, &work_end) == 0)
1582 return 0;
1583
1584 /* ********************************************************* */
1585 /* * step 5a: back off one from last, increment in dtaps * */
1586
1587 /* Special case code for backing up a phase */
1588 if (p == 0) {
1589 p = IO_DQS_EN_PHASE_MAX;
1590 rw_mgr_decr_vfifo(grp, &v);
1591 } else {
1592 p = p - 1;
1593 }
1594
1595 work_end -= IO_DELAY_PER_OPA_TAP;
1596 scc_mgr_set_dqs_en_phase_all_ranks(grp, p);
1597
1598 /* * The actual increment of dtaps is done outside of
1599 the if/else loop to share code */
1600 d = 0;
1601
1602 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: v/p: \
1603 vfifo=%u ptap=%u\n", __func__, __LINE__,
1604 v, p);
1605 } else {
1606 /* ******************************************************* */
1607 /* * step 3-5b: Find the right edge of the window using
1608 delay taps * */
1609 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase:vfifo=%u \
1610 ptap=%u dtap=%u bgn=%u\n", __func__, __LINE__,
1611 v, p, d, work_bgn);
1612
1613 work_end = work_bgn;
1614
1615 /* * The actual increment of dtaps is done outside of the
1616 if/else loop to share code */
1617
1618 /* Only here to counterbalance a subtract later on which is
1619 not needed if this branch of the algorithm is taken */
1620 max_working_cnt++;
1621 }
1622
1623 /* The dtap increment to find the failing edge is done here */
1624 for (; d <= IO_DQS_EN_DELAY_MAX; d++, work_end +=
1625 IO_DELAY_PER_DQS_EN_DCHAIN_TAP) {
1626 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: \
1627 end-2: dtap=%u\n", __func__, __LINE__, d);
1628 scc_mgr_set_dqs_en_delay_all_ranks(grp, d);
1629
1630 if (!rw_mgr_mem_calibrate_read_test_all_ranks(grp, 1,
1631 PASS_ONE_BIT,
1632 &bit_chk, 0)) {
1633 break;
1634 }
1635 }
1636
1637 /* Go back to working dtap */
1638 if (d != 0)
1639 work_end -= IO_DELAY_PER_DQS_EN_DCHAIN_TAP;
1640
1641 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: v/p/d: vfifo=%u \
1642 ptap=%u dtap=%u end=%u\n", __func__, __LINE__,
1643 v, p, d-1, work_end);
1644
1645 if (work_end < work_bgn) {
1646 /* nil range */
1647 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: end-2: \
1648 failed\n", __func__, __LINE__);
1649 return 0;
1650 }
1651
1652 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: found range [%u,%u]\n",
1653 __func__, __LINE__, work_bgn, work_end);
1654
1655 /* *************************************************************** */
1656 /*
1657 * * We need to calculate the number of dtaps that equal a ptap
1658 * * To do that we'll back up a ptap and re-find the edge of the
1659 * * window using dtaps
1660 */
1661
1662 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: calculate dtaps_per_ptap \
1663 for tracking\n", __func__, __LINE__);
1664
1665 /* Special case code for backing up a phase */
1666 if (p == 0) {
1667 p = IO_DQS_EN_PHASE_MAX;
1668 rw_mgr_decr_vfifo(grp, &v);
1669 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: backedup \
1670 cycle/phase: v=%u p=%u\n", __func__, __LINE__,
1671 v, p);
1672 } else {
1673 p = p - 1;
1674 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: backedup \
1675 phase only: v=%u p=%u", __func__, __LINE__,
1676 v, p);
1677 }
1678
1679 scc_mgr_set_dqs_en_phase_all_ranks(grp, p);
1680
1681 /*
1682 * Increase dtap until we first see a passing read (in case the
1683 * window is smaller than a ptap),
1684 * and then a failing read to mark the edge of the window again
1685 */
1686
1687 /* Find a passing read */
1688 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: find passing read\n",
1689 __func__, __LINE__);
1690 found_passing_read = 0;
1691 found_failing_read = 0;
1692 initial_failing_dtap = d;
1693 for (; d <= IO_DQS_EN_DELAY_MAX; d++) {
1694 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: testing \
1695 read d=%u\n", __func__, __LINE__, d);
1696 scc_mgr_set_dqs_en_delay_all_ranks(grp, d);
1697
1698 if (rw_mgr_mem_calibrate_read_test_all_ranks(grp, 1,
1699 PASS_ONE_BIT,
1700 &bit_chk, 0)) {
1701 found_passing_read = 1;
1702 break;
1703 }
1704 }
1705
1706 if (found_passing_read) {
1707 /* Find a failing read */
1708 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: find failing \
1709 read\n", __func__, __LINE__);
1710 for (d = d + 1; d <= IO_DQS_EN_DELAY_MAX; d++) {
1711 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: \
1712 testing read d=%u\n", __func__, __LINE__, d);
1713 scc_mgr_set_dqs_en_delay_all_ranks(grp, d);
1714
1715 if (!rw_mgr_mem_calibrate_read_test_all_ranks
1716 (grp, 1, PASS_ONE_BIT, &bit_chk, 0)) {
1717 found_failing_read = 1;
1718 break;
1719 }
1720 }
1721 } else {
1722 debug_cond(DLEVEL == 1, "%s:%d find_dqs_en_phase: failed to \
1723 calculate dtaps", __func__, __LINE__);
1724 debug_cond(DLEVEL == 1, "per ptap. Fall back on static value\n");
1725 }
1726
1727 /*
1728 * The dynamically calculated dtaps_per_ptap is only valid if we
1729 * found a passing/failing read. If we didn't, it means d hit the max
1730 * (IO_DQS_EN_DELAY_MAX). Otherwise, dtaps_per_ptap retains its
1731 * statically calculated value.
1732 */
1733 if (found_passing_read && found_failing_read)
1734 dtaps_per_ptap = d - initial_failing_dtap;
1735
1736 writel(dtaps_per_ptap, &sdr_reg_file->dtaps_per_ptap);
1737 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: dtaps_per_ptap=%u \
1738 - %u = %u", __func__, __LINE__, d,
1739 initial_failing_dtap, dtaps_per_ptap);
1740
1741 /* ******************************************** */
1742 /* * step 6: Find the centre of the window * */
1743 if (sdr_find_window_centre(&grp, &bit_chk, &work_bgn, &v, &d, &p,
1744 &work_mid, &work_end) == 0)
1745 return 0;
1746
1747 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: center found: \
1748 vfifo=%u ptap=%u dtap=%u\n", __func__, __LINE__,
1749 v, p-1, d);
1750 return 1;
1751 }
1752
1753 /*
1754 * Try rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase across different
1755 * dq_in_delay values
1756 */
1757 static uint32_t
1758 rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase_sweep_dq_in_delay
1759 (uint32_t write_group, uint32_t read_group, uint32_t test_bgn)
1760 {
1761 uint32_t found;
1762 uint32_t i;
1763 uint32_t p;
1764 uint32_t d;
1765 uint32_t r;
1766
1767 const uint32_t delay_step = IO_IO_IN_DELAY_MAX /
1768 (RW_MGR_MEM_DQ_PER_READ_DQS-1);
1769 /* we start at zero, so have one less dq to devide among */
1770
1771 debug("%s:%d (%u,%u,%u)", __func__, __LINE__, write_group, read_group,
1772 test_bgn);
1773
1774 /* try different dq_in_delays since the dq path is shorter than dqs */
1775
1776 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
1777 r += NUM_RANKS_PER_SHADOW_REG) {
1778 for (i = 0, p = test_bgn, d = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++, p++, d += delay_step) {
1779 debug_cond(DLEVEL == 1, "%s:%d rw_mgr_mem_calibrate_\
1780 vfifo_find_dqs_", __func__, __LINE__);
1781 debug_cond(DLEVEL == 1, "en_phase_sweep_dq_in_delay: g=%u/%u ",
1782 write_group, read_group);
1783 debug_cond(DLEVEL == 1, "r=%u, i=%u p=%u d=%u\n", r, i , p, d);
1784 scc_mgr_set_dq_in_delay(p, d);
1785 scc_mgr_load_dq(p);
1786 }
1787 writel(0, &sdr_scc_mgr->update);
1788 }
1789
1790 found = rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase(read_group);
1791
1792 debug_cond(DLEVEL == 1, "%s:%d rw_mgr_mem_calibrate_vfifo_find_dqs_\
1793 en_phase_sweep_dq", __func__, __LINE__);
1794 debug_cond(DLEVEL == 1, "_in_delay: g=%u/%u found=%u; Reseting delay \
1795 chain to zero\n", write_group, read_group, found);
1796
1797 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
1798 r += NUM_RANKS_PER_SHADOW_REG) {
1799 for (i = 0, p = test_bgn; i < RW_MGR_MEM_DQ_PER_READ_DQS;
1800 i++, p++) {
1801 scc_mgr_set_dq_in_delay(p, 0);
1802 scc_mgr_load_dq(p);
1803 }
1804 writel(0, &sdr_scc_mgr->update);
1805 }
1806
1807 return found;
1808 }
1809
1810 /* per-bit deskew DQ and center */
1811 static uint32_t rw_mgr_mem_calibrate_vfifo_center(uint32_t rank_bgn,
1812 uint32_t write_group, uint32_t read_group, uint32_t test_bgn,
1813 uint32_t use_read_test, uint32_t update_fom)
1814 {
1815 uint32_t i, p, d, min_index;
1816 /*
1817 * Store these as signed since there are comparisons with
1818 * signed numbers.
1819 */
1820 uint32_t bit_chk;
1821 uint32_t sticky_bit_chk;
1822 int32_t left_edge[RW_MGR_MEM_DQ_PER_READ_DQS];
1823 int32_t right_edge[RW_MGR_MEM_DQ_PER_READ_DQS];
1824 int32_t final_dq[RW_MGR_MEM_DQ_PER_READ_DQS];
1825 int32_t mid;
1826 int32_t orig_mid_min, mid_min;
1827 int32_t new_dqs, start_dqs, start_dqs_en, shift_dq, final_dqs,
1828 final_dqs_en;
1829 int32_t dq_margin, dqs_margin;
1830 uint32_t stop;
1831 uint32_t temp_dq_in_delay1, temp_dq_in_delay2;
1832 uint32_t addr;
1833
1834 debug("%s:%d: %u %u", __func__, __LINE__, read_group, test_bgn);
1835
1836 addr = SDR_PHYGRP_SCCGRP_ADDRESS | SCC_MGR_DQS_IN_DELAY_OFFSET;
1837 start_dqs = readl(addr + (read_group << 2));
1838 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS)
1839 start_dqs_en = readl(addr + ((read_group << 2)
1840 - IO_DQS_EN_DELAY_OFFSET));
1841
1842 /* set the left and right edge of each bit to an illegal value */
1843 /* use (IO_IO_IN_DELAY_MAX + 1) as an illegal value */
1844 sticky_bit_chk = 0;
1845 for (i = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
1846 left_edge[i] = IO_IO_IN_DELAY_MAX + 1;
1847 right_edge[i] = IO_IO_IN_DELAY_MAX + 1;
1848 }
1849
1850 /* Search for the left edge of the window for each bit */
1851 for (d = 0; d <= IO_IO_IN_DELAY_MAX; d++) {
1852 scc_mgr_apply_group_dq_in_delay(write_group, test_bgn, d);
1853
1854 writel(0, &sdr_scc_mgr->update);
1855
1856 /*
1857 * Stop searching when the read test doesn't pass AND when
1858 * we've seen a passing read on every bit.
1859 */
1860 if (use_read_test) {
1861 stop = !rw_mgr_mem_calibrate_read_test(rank_bgn,
1862 read_group, NUM_READ_PB_TESTS, PASS_ONE_BIT,
1863 &bit_chk, 0, 0);
1864 } else {
1865 rw_mgr_mem_calibrate_write_test(rank_bgn, write_group,
1866 0, PASS_ONE_BIT,
1867 &bit_chk, 0);
1868 bit_chk = bit_chk >> (RW_MGR_MEM_DQ_PER_READ_DQS *
1869 (read_group - (write_group *
1870 RW_MGR_MEM_IF_READ_DQS_WIDTH /
1871 RW_MGR_MEM_IF_WRITE_DQS_WIDTH)));
1872 stop = (bit_chk == 0);
1873 }
1874 sticky_bit_chk = sticky_bit_chk | bit_chk;
1875 stop = stop && (sticky_bit_chk == param->read_correct_mask);
1876 debug_cond(DLEVEL == 2, "%s:%d vfifo_center(left): dtap=%u => %u == %u \
1877 && %u", __func__, __LINE__, d,
1878 sticky_bit_chk,
1879 param->read_correct_mask, stop);
1880
1881 if (stop == 1) {
1882 break;
1883 } else {
1884 for (i = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
1885 if (bit_chk & 1) {
1886 /* Remember a passing test as the
1887 left_edge */
1888 left_edge[i] = d;
1889 } else {
1890 /* If a left edge has not been seen yet,
1891 then a future passing test will mark
1892 this edge as the right edge */
1893 if (left_edge[i] ==
1894 IO_IO_IN_DELAY_MAX + 1) {
1895 right_edge[i] = -(d + 1);
1896 }
1897 }
1898 bit_chk = bit_chk >> 1;
1899 }
1900 }
1901 }
1902
1903 /* Reset DQ delay chains to 0 */
1904 scc_mgr_apply_group_dq_in_delay(test_bgn, 0);
1905 sticky_bit_chk = 0;
1906 for (i = RW_MGR_MEM_DQ_PER_READ_DQS - 1;; i--) {
1907 debug_cond(DLEVEL == 2, "%s:%d vfifo_center: left_edge[%u]: \
1908 %d right_edge[%u]: %d\n", __func__, __LINE__,
1909 i, left_edge[i], i, right_edge[i]);
1910
1911 /*
1912 * Check for cases where we haven't found the left edge,
1913 * which makes our assignment of the the right edge invalid.
1914 * Reset it to the illegal value.
1915 */
1916 if ((left_edge[i] == IO_IO_IN_DELAY_MAX + 1) && (
1917 right_edge[i] != IO_IO_IN_DELAY_MAX + 1)) {
1918 right_edge[i] = IO_IO_IN_DELAY_MAX + 1;
1919 debug_cond(DLEVEL == 2, "%s:%d vfifo_center: reset \
1920 right_edge[%u]: %d\n", __func__, __LINE__,
1921 i, right_edge[i]);
1922 }
1923
1924 /*
1925 * Reset sticky bit (except for bits where we have seen
1926 * both the left and right edge).
1927 */
1928 sticky_bit_chk = sticky_bit_chk << 1;
1929 if ((left_edge[i] != IO_IO_IN_DELAY_MAX + 1) &&
1930 (right_edge[i] != IO_IO_IN_DELAY_MAX + 1)) {
1931 sticky_bit_chk = sticky_bit_chk | 1;
1932 }
1933
1934 if (i == 0)
1935 break;
1936 }
1937
1938 /* Search for the right edge of the window for each bit */
1939 for (d = 0; d <= IO_DQS_IN_DELAY_MAX - start_dqs; d++) {
1940 scc_mgr_set_dqs_bus_in_delay(read_group, d + start_dqs);
1941 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS) {
1942 uint32_t delay = d + start_dqs_en;
1943 if (delay > IO_DQS_EN_DELAY_MAX)
1944 delay = IO_DQS_EN_DELAY_MAX;
1945 scc_mgr_set_dqs_en_delay(read_group, delay);
1946 }
1947 scc_mgr_load_dqs(read_group);
1948
1949 writel(0, &sdr_scc_mgr->update);
1950
1951 /*
1952 * Stop searching when the read test doesn't pass AND when
1953 * we've seen a passing read on every bit.
1954 */
1955 if (use_read_test) {
1956 stop = !rw_mgr_mem_calibrate_read_test(rank_bgn,
1957 read_group, NUM_READ_PB_TESTS, PASS_ONE_BIT,
1958 &bit_chk, 0, 0);
1959 } else {
1960 rw_mgr_mem_calibrate_write_test(rank_bgn, write_group,
1961 0, PASS_ONE_BIT,
1962 &bit_chk, 0);
1963 bit_chk = bit_chk >> (RW_MGR_MEM_DQ_PER_READ_DQS *
1964 (read_group - (write_group *
1965 RW_MGR_MEM_IF_READ_DQS_WIDTH /
1966 RW_MGR_MEM_IF_WRITE_DQS_WIDTH)));
1967 stop = (bit_chk == 0);
1968 }
1969 sticky_bit_chk = sticky_bit_chk | bit_chk;
1970 stop = stop && (sticky_bit_chk == param->read_correct_mask);
1971
1972 debug_cond(DLEVEL == 2, "%s:%d vfifo_center(right): dtap=%u => %u == \
1973 %u && %u", __func__, __LINE__, d,
1974 sticky_bit_chk, param->read_correct_mask, stop);
1975
1976 if (stop == 1) {
1977 break;
1978 } else {
1979 for (i = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
1980 if (bit_chk & 1) {
1981 /* Remember a passing test as
1982 the right_edge */
1983 right_edge[i] = d;
1984 } else {
1985 if (d != 0) {
1986 /* If a right edge has not been
1987 seen yet, then a future passing
1988 test will mark this edge as the
1989 left edge */
1990 if (right_edge[i] ==
1991 IO_IO_IN_DELAY_MAX + 1) {
1992 left_edge[i] = -(d + 1);
1993 }
1994 } else {
1995 /* d = 0 failed, but it passed
1996 when testing the left edge,
1997 so it must be marginal,
1998 set it to -1 */
1999 if (right_edge[i] ==
2000 IO_IO_IN_DELAY_MAX + 1 &&
2001 left_edge[i] !=
2002 IO_IO_IN_DELAY_MAX
2003 + 1) {
2004 right_edge[i] = -1;
2005 }
2006 /* If a right edge has not been
2007 seen yet, then a future passing
2008 test will mark this edge as the
2009 left edge */
2010 else if (right_edge[i] ==
2011 IO_IO_IN_DELAY_MAX +
2012 1) {
2013 left_edge[i] = -(d + 1);
2014 }
2015 }
2016 }
2017
2018 debug_cond(DLEVEL == 2, "%s:%d vfifo_center[r,\
2019 d=%u]: ", __func__, __LINE__, d);
2020 debug_cond(DLEVEL == 2, "bit_chk_test=%d left_edge[%u]: %d ",
2021 (int)(bit_chk & 1), i, left_edge[i]);
2022 debug_cond(DLEVEL == 2, "right_edge[%u]: %d\n", i,
2023 right_edge[i]);
2024 bit_chk = bit_chk >> 1;
2025 }
2026 }
2027 }
2028
2029 /* Check that all bits have a window */
2030 for (i = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
2031 debug_cond(DLEVEL == 2, "%s:%d vfifo_center: left_edge[%u]: \
2032 %d right_edge[%u]: %d", __func__, __LINE__,
2033 i, left_edge[i], i, right_edge[i]);
2034 if ((left_edge[i] == IO_IO_IN_DELAY_MAX + 1) || (right_edge[i]
2035 == IO_IO_IN_DELAY_MAX + 1)) {
2036 /*
2037 * Restore delay chain settings before letting the loop
2038 * in rw_mgr_mem_calibrate_vfifo to retry different
2039 * dqs/ck relationships.
2040 */
2041 scc_mgr_set_dqs_bus_in_delay(read_group, start_dqs);
2042 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS) {
2043 scc_mgr_set_dqs_en_delay(read_group,
2044 start_dqs_en);
2045 }
2046 scc_mgr_load_dqs(read_group);
2047 writel(0, &sdr_scc_mgr->update);
2048
2049 debug_cond(DLEVEL == 1, "%s:%d vfifo_center: failed to \
2050 find edge [%u]: %d %d", __func__, __LINE__,
2051 i, left_edge[i], right_edge[i]);
2052 if (use_read_test) {
2053 set_failing_group_stage(read_group *
2054 RW_MGR_MEM_DQ_PER_READ_DQS + i,
2055 CAL_STAGE_VFIFO,
2056 CAL_SUBSTAGE_VFIFO_CENTER);
2057 } else {
2058 set_failing_group_stage(read_group *
2059 RW_MGR_MEM_DQ_PER_READ_DQS + i,
2060 CAL_STAGE_VFIFO_AFTER_WRITES,
2061 CAL_SUBSTAGE_VFIFO_CENTER);
2062 }
2063 return 0;
2064 }
2065 }
2066
2067 /* Find middle of window for each DQ bit */
2068 mid_min = left_edge[0] - right_edge[0];
2069 min_index = 0;
2070 for (i = 1; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
2071 mid = left_edge[i] - right_edge[i];
2072 if (mid < mid_min) {
2073 mid_min = mid;
2074 min_index = i;
2075 }
2076 }
2077
2078 /*
2079 * -mid_min/2 represents the amount that we need to move DQS.
2080 * If mid_min is odd and positive we'll need to add one to
2081 * make sure the rounding in further calculations is correct
2082 * (always bias to the right), so just add 1 for all positive values.
2083 */
2084 if (mid_min > 0)
2085 mid_min++;
2086
2087 mid_min = mid_min / 2;
2088
2089 debug_cond(DLEVEL == 1, "%s:%d vfifo_center: mid_min=%d (index=%u)\n",
2090 __func__, __LINE__, mid_min, min_index);
2091
2092 /* Determine the amount we can change DQS (which is -mid_min) */
2093 orig_mid_min = mid_min;
2094 new_dqs = start_dqs - mid_min;
2095 if (new_dqs > IO_DQS_IN_DELAY_MAX)
2096 new_dqs = IO_DQS_IN_DELAY_MAX;
2097 else if (new_dqs < 0)
2098 new_dqs = 0;
2099
2100 mid_min = start_dqs - new_dqs;
2101 debug_cond(DLEVEL == 1, "vfifo_center: new mid_min=%d new_dqs=%d\n",
2102 mid_min, new_dqs);
2103
2104 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS) {
2105 if (start_dqs_en - mid_min > IO_DQS_EN_DELAY_MAX)
2106 mid_min += start_dqs_en - mid_min - IO_DQS_EN_DELAY_MAX;
2107 else if (start_dqs_en - mid_min < 0)
2108 mid_min += start_dqs_en - mid_min;
2109 }
2110 new_dqs = start_dqs - mid_min;
2111
2112 debug_cond(DLEVEL == 1, "vfifo_center: start_dqs=%d start_dqs_en=%d \
2113 new_dqs=%d mid_min=%d\n", start_dqs,
2114 IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS ? start_dqs_en : -1,
2115 new_dqs, mid_min);
2116
2117 /* Initialize data for export structures */
2118 dqs_margin = IO_IO_IN_DELAY_MAX + 1;
2119 dq_margin = IO_IO_IN_DELAY_MAX + 1;
2120
2121 /* add delay to bring centre of all DQ windows to the same "level" */
2122 for (i = 0, p = test_bgn; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++, p++) {
2123 /* Use values before divide by 2 to reduce round off error */
2124 shift_dq = (left_edge[i] - right_edge[i] -
2125 (left_edge[min_index] - right_edge[min_index]))/2 +
2126 (orig_mid_min - mid_min);
2127
2128 debug_cond(DLEVEL == 2, "vfifo_center: before: \
2129 shift_dq[%u]=%d\n", i, shift_dq);
2130
2131 addr = SDR_PHYGRP_SCCGRP_ADDRESS | SCC_MGR_IO_IN_DELAY_OFFSET;
2132 temp_dq_in_delay1 = readl(addr + (p << 2));
2133 temp_dq_in_delay2 = readl(addr + (i << 2));
2134
2135 if (shift_dq + (int32_t)temp_dq_in_delay1 >
2136 (int32_t)IO_IO_IN_DELAY_MAX) {
2137 shift_dq = (int32_t)IO_IO_IN_DELAY_MAX - temp_dq_in_delay2;
2138 } else if (shift_dq + (int32_t)temp_dq_in_delay1 < 0) {
2139 shift_dq = -(int32_t)temp_dq_in_delay1;
2140 }
2141 debug_cond(DLEVEL == 2, "vfifo_center: after: \
2142 shift_dq[%u]=%d\n", i, shift_dq);
2143 final_dq[i] = temp_dq_in_delay1 + shift_dq;
2144 scc_mgr_set_dq_in_delay(p, final_dq[i]);
2145 scc_mgr_load_dq(p);
2146
2147 debug_cond(DLEVEL == 2, "vfifo_center: margin[%u]=[%d,%d]\n", i,
2148 left_edge[i] - shift_dq + (-mid_min),
2149 right_edge[i] + shift_dq - (-mid_min));
2150 /* To determine values for export structures */
2151 if (left_edge[i] - shift_dq + (-mid_min) < dq_margin)
2152 dq_margin = left_edge[i] - shift_dq + (-mid_min);
2153
2154 if (right_edge[i] + shift_dq - (-mid_min) < dqs_margin)
2155 dqs_margin = right_edge[i] + shift_dq - (-mid_min);
2156 }
2157
2158 final_dqs = new_dqs;
2159 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS)
2160 final_dqs_en = start_dqs_en - mid_min;
2161
2162 /* Move DQS-en */
2163 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS) {
2164 scc_mgr_set_dqs_en_delay(read_group, final_dqs_en);
2165 scc_mgr_load_dqs(read_group);
2166 }
2167
2168 /* Move DQS */
2169 scc_mgr_set_dqs_bus_in_delay(read_group, final_dqs);
2170 scc_mgr_load_dqs(read_group);
2171 debug_cond(DLEVEL == 2, "%s:%d vfifo_center: dq_margin=%d \
2172 dqs_margin=%d", __func__, __LINE__,
2173 dq_margin, dqs_margin);
2174
2175 /*
2176 * Do not remove this line as it makes sure all of our decisions
2177 * have been applied. Apply the update bit.
2178 */
2179 writel(0, &sdr_scc_mgr->update);
2180
2181 return (dq_margin >= 0) && (dqs_margin >= 0);
2182 }
2183
2184 /*
2185 * calibrate the read valid prediction FIFO.
2186 *
2187 * - read valid prediction will consist of finding a good DQS enable phase,
2188 * DQS enable delay, DQS input phase, and DQS input delay.
2189 * - we also do a per-bit deskew on the DQ lines.
2190 */
2191 static uint32_t rw_mgr_mem_calibrate_vfifo(uint32_t read_group,
2192 uint32_t test_bgn)
2193 {
2194 uint32_t p, d, rank_bgn, sr;
2195 uint32_t dtaps_per_ptap;
2196 uint32_t bit_chk;
2197 uint32_t grp_calibrated;
2198 uint32_t write_group, write_test_bgn;
2199 uint32_t failed_substage;
2200
2201 debug("%s:%d: %u %u\n", __func__, __LINE__, read_group, test_bgn);
2202
2203 /* update info for sims */
2204 reg_file_set_stage(CAL_STAGE_VFIFO);
2205
2206 write_group = read_group;
2207 write_test_bgn = test_bgn;
2208
2209 /* USER Determine number of delay taps for each phase tap */
2210 dtaps_per_ptap = DIV_ROUND_UP(IO_DELAY_PER_OPA_TAP,
2211 IO_DELAY_PER_DQS_EN_DCHAIN_TAP) - 1;
2212
2213 /* update info for sims */
2214 reg_file_set_group(read_group);
2215
2216 grp_calibrated = 0;
2217
2218 reg_file_set_sub_stage(CAL_SUBSTAGE_GUARANTEED_READ);
2219 failed_substage = CAL_SUBSTAGE_GUARANTEED_READ;
2220
2221 for (d = 0; d <= dtaps_per_ptap && grp_calibrated == 0; d += 2) {
2222 /*
2223 * In RLDRAMX we may be messing the delay of pins in
2224 * the same write group but outside of the current read
2225 * the group, but that's ok because we haven't
2226 * calibrated output side yet.
2227 */
2228 if (d > 0) {
2229 scc_mgr_apply_group_all_out_delay_add_all_ranks(
2230 write_group, d);
2231 }
2232
2233 for (p = 0; p <= IO_DQDQS_OUT_PHASE_MAX && grp_calibrated == 0;
2234 p++) {
2235 /* set a particular dqdqs phase */
2236 scc_mgr_set_dqdqs_output_phase_all_ranks(read_group, p);
2237
2238 debug_cond(DLEVEL == 1, "%s:%d calibrate_vfifo: g=%u \
2239 p=%u d=%u\n", __func__, __LINE__,
2240 read_group, p, d);
2241
2242 /*
2243 * Load up the patterns used by read calibration
2244 * using current DQDQS phase.
2245 */
2246 rw_mgr_mem_calibrate_read_load_patterns(0, 1);
2247 if (!(gbl->phy_debug_mode_flags &
2248 PHY_DEBUG_DISABLE_GUARANTEED_READ)) {
2249 if (!rw_mgr_mem_calibrate_read_test_patterns_all_ranks
2250 (read_group, 1, &bit_chk)) {
2251 debug_cond(DLEVEL == 1, "%s:%d Guaranteed read test failed:",
2252 __func__, __LINE__);
2253 debug_cond(DLEVEL == 1, " g=%u p=%u d=%u\n",
2254 read_group, p, d);
2255 break;
2256 }
2257 }
2258
2259 /* case:56390 */
2260 grp_calibrated = 1;
2261 if (rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase_sweep_dq_in_delay
2262 (write_group, read_group, test_bgn)) {
2263 /*
2264 * USER Read per-bit deskew can be done on a
2265 * per shadow register basis.
2266 */
2267 for (rank_bgn = 0, sr = 0;
2268 rank_bgn < RW_MGR_MEM_NUMBER_OF_RANKS;
2269 rank_bgn += NUM_RANKS_PER_SHADOW_REG,
2270 ++sr) {
2271 /*
2272 * Determine if this set of ranks
2273 * should be skipped entirely.
2274 */
2275 if (!param->skip_shadow_regs[sr]) {
2276 /*
2277 * If doing read after write
2278 * calibration, do not update
2279 * FOM, now - do it then.
2280 */
2281 if (!rw_mgr_mem_calibrate_vfifo_center
2282 (rank_bgn, write_group,
2283 read_group, test_bgn, 1, 0)) {
2284 grp_calibrated = 0;
2285 failed_substage =
2286 CAL_SUBSTAGE_VFIFO_CENTER;
2287 }
2288 }
2289 }
2290 } else {
2291 grp_calibrated = 0;
2292 failed_substage = CAL_SUBSTAGE_DQS_EN_PHASE;
2293 }
2294 }
2295 }
2296
2297 if (grp_calibrated == 0) {
2298 set_failing_group_stage(write_group, CAL_STAGE_VFIFO,
2299 failed_substage);
2300 return 0;
2301 }
2302
2303 /*
2304 * Reset the delay chains back to zero if they have moved > 1
2305 * (check for > 1 because loop will increase d even when pass in
2306 * first case).
2307 */
2308 if (d > 2)
2309 scc_mgr_zero_group(write_group, 1);
2310
2311 return 1;
2312 }
2313
2314 /* VFIFO Calibration -- Read Deskew Calibration after write deskew */
2315 static uint32_t rw_mgr_mem_calibrate_vfifo_end(uint32_t read_group,
2316 uint32_t test_bgn)
2317 {
2318 uint32_t rank_bgn, sr;
2319 uint32_t grp_calibrated;
2320 uint32_t write_group;
2321
2322 debug("%s:%d %u %u", __func__, __LINE__, read_group, test_bgn);
2323
2324 /* update info for sims */
2325
2326 reg_file_set_stage(CAL_STAGE_VFIFO_AFTER_WRITES);
2327 reg_file_set_sub_stage(CAL_SUBSTAGE_VFIFO_CENTER);
2328
2329 write_group = read_group;
2330
2331 /* update info for sims */
2332 reg_file_set_group(read_group);
2333
2334 grp_calibrated = 1;
2335 /* Read per-bit deskew can be done on a per shadow register basis */
2336 for (rank_bgn = 0, sr = 0; rank_bgn < RW_MGR_MEM_NUMBER_OF_RANKS;
2337 rank_bgn += NUM_RANKS_PER_SHADOW_REG, ++sr) {
2338 /* Determine if this set of ranks should be skipped entirely */
2339 if (!param->skip_shadow_regs[sr]) {
2340 /* This is the last calibration round, update FOM here */
2341 if (!rw_mgr_mem_calibrate_vfifo_center(rank_bgn,
2342 write_group,
2343 read_group,
2344 test_bgn, 0,
2345 1)) {
2346 grp_calibrated = 0;
2347 }
2348 }
2349 }
2350
2351
2352 if (grp_calibrated == 0) {
2353 set_failing_group_stage(write_group,
2354 CAL_STAGE_VFIFO_AFTER_WRITES,
2355 CAL_SUBSTAGE_VFIFO_CENTER);
2356 return 0;
2357 }
2358
2359 return 1;
2360 }
2361
2362 /* Calibrate LFIFO to find smallest read latency */
2363 static uint32_t rw_mgr_mem_calibrate_lfifo(void)
2364 {
2365 uint32_t found_one;
2366 uint32_t bit_chk;
2367
2368 debug("%s:%d\n", __func__, __LINE__);
2369
2370 /* update info for sims */
2371 reg_file_set_stage(CAL_STAGE_LFIFO);
2372 reg_file_set_sub_stage(CAL_SUBSTAGE_READ_LATENCY);
2373
2374 /* Load up the patterns used by read calibration for all ranks */
2375 rw_mgr_mem_calibrate_read_load_patterns(0, 1);
2376 found_one = 0;
2377
2378 do {
2379 writel(gbl->curr_read_lat, &phy_mgr_cfg->phy_rlat);
2380 debug_cond(DLEVEL == 2, "%s:%d lfifo: read_lat=%u",
2381 __func__, __LINE__, gbl->curr_read_lat);
2382
2383 if (!rw_mgr_mem_calibrate_read_test_all_ranks(0,
2384 NUM_READ_TESTS,
2385 PASS_ALL_BITS,
2386 &bit_chk, 1)) {
2387 break;
2388 }
2389
2390 found_one = 1;
2391 /* reduce read latency and see if things are working */
2392 /* correctly */
2393 gbl->curr_read_lat--;
2394 } while (gbl->curr_read_lat > 0);
2395
2396 /* reset the fifos to get pointers to known state */
2397
2398 writel(0, &phy_mgr_cmd->fifo_reset);
2399
2400 if (found_one) {
2401 /* add a fudge factor to the read latency that was determined */
2402 gbl->curr_read_lat += 2;
2403 writel(gbl->curr_read_lat, &phy_mgr_cfg->phy_rlat);
2404 debug_cond(DLEVEL == 2, "%s:%d lfifo: success: using \
2405 read_lat=%u\n", __func__, __LINE__,
2406 gbl->curr_read_lat);
2407 return 1;
2408 } else {
2409 set_failing_group_stage(0xff, CAL_STAGE_LFIFO,
2410 CAL_SUBSTAGE_READ_LATENCY);
2411
2412 debug_cond(DLEVEL == 2, "%s:%d lfifo: failed at initial \
2413 read_lat=%u\n", __func__, __LINE__,
2414 gbl->curr_read_lat);
2415 return 0;
2416 }
2417 }
2418
2419 /*
2420 * issue write test command.
2421 * two variants are provided. one that just tests a write pattern and
2422 * another that tests datamask functionality.
2423 */
2424 static void rw_mgr_mem_calibrate_write_test_issue(uint32_t group,
2425 uint32_t test_dm)
2426 {
2427 uint32_t mcc_instruction;
2428 uint32_t quick_write_mode = (((STATIC_CALIB_STEPS) & CALIB_SKIP_WRITES) &&
2429 ENABLE_SUPER_QUICK_CALIBRATION);
2430 uint32_t rw_wl_nop_cycles;
2431 uint32_t addr;
2432
2433 /*
2434 * Set counter and jump addresses for the right
2435 * number of NOP cycles.
2436 * The number of supported NOP cycles can range from -1 to infinity
2437 * Three different cases are handled:
2438 *
2439 * 1. For a number of NOP cycles greater than 0, the RW Mgr looping
2440 * mechanism will be used to insert the right number of NOPs
2441 *
2442 * 2. For a number of NOP cycles equals to 0, the micro-instruction
2443 * issuing the write command will jump straight to the
2444 * micro-instruction that turns on DQS (for DDRx), or outputs write
2445 * data (for RLD), skipping
2446 * the NOP micro-instruction all together
2447 *
2448 * 3. A number of NOP cycles equal to -1 indicates that DQS must be
2449 * turned on in the same micro-instruction that issues the write
2450 * command. Then we need
2451 * to directly jump to the micro-instruction that sends out the data
2452 *
2453 * NOTE: Implementing this mechanism uses 2 RW Mgr jump-counters
2454 * (2 and 3). One jump-counter (0) is used to perform multiple
2455 * write-read operations.
2456 * one counter left to issue this command in "multiple-group" mode
2457 */
2458
2459 rw_wl_nop_cycles = gbl->rw_wl_nop_cycles;
2460
2461 if (rw_wl_nop_cycles == -1) {
2462 /*
2463 * CNTR 2 - We want to execute the special write operation that
2464 * turns on DQS right away and then skip directly to the
2465 * instruction that sends out the data. We set the counter to a
2466 * large number so that the jump is always taken.
2467 */
2468 writel(0xFF, &sdr_rw_load_mgr_regs->load_cntr2);
2469
2470 /* CNTR 3 - Not used */
2471 if (test_dm) {
2472 mcc_instruction = RW_MGR_LFSR_WR_RD_DM_BANK_0_WL_1;
2473 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_DATA,
2474 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2475 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_NOP,
2476 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
2477 } else {
2478 mcc_instruction = RW_MGR_LFSR_WR_RD_BANK_0_WL_1;
2479 writel(RW_MGR_LFSR_WR_RD_BANK_0_DATA,
2480 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2481 writel(RW_MGR_LFSR_WR_RD_BANK_0_NOP,
2482 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
2483 }
2484 } else if (rw_wl_nop_cycles == 0) {
2485 /*
2486 * CNTR 2 - We want to skip the NOP operation and go straight
2487 * to the DQS enable instruction. We set the counter to a large
2488 * number so that the jump is always taken.
2489 */
2490 writel(0xFF, &sdr_rw_load_mgr_regs->load_cntr2);
2491
2492 /* CNTR 3 - Not used */
2493 if (test_dm) {
2494 mcc_instruction = RW_MGR_LFSR_WR_RD_DM_BANK_0;
2495 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_DQS,
2496 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2497 } else {
2498 mcc_instruction = RW_MGR_LFSR_WR_RD_BANK_0;
2499 writel(RW_MGR_LFSR_WR_RD_BANK_0_DQS,
2500 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2501 }
2502 } else {
2503 /*
2504 * CNTR 2 - In this case we want to execute the next instruction
2505 * and NOT take the jump. So we set the counter to 0. The jump
2506 * address doesn't count.
2507 */
2508 writel(0x0, &sdr_rw_load_mgr_regs->load_cntr2);
2509 writel(0x0, &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2510
2511 /*
2512 * CNTR 3 - Set the nop counter to the number of cycles we
2513 * need to loop for, minus 1.
2514 */
2515 writel(rw_wl_nop_cycles - 1, &sdr_rw_load_mgr_regs->load_cntr3);
2516 if (test_dm) {
2517 mcc_instruction = RW_MGR_LFSR_WR_RD_DM_BANK_0;
2518 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_NOP,
2519 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
2520 } else {
2521 mcc_instruction = RW_MGR_LFSR_WR_RD_BANK_0;
2522 writel(RW_MGR_LFSR_WR_RD_BANK_0_NOP,
2523 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
2524 }
2525 }
2526
2527 writel(0, SDR_PHYGRP_RWMGRGRP_ADDRESS |
2528 RW_MGR_RESET_READ_DATAPATH_OFFSET);
2529
2530 if (quick_write_mode)
2531 writel(0x08, &sdr_rw_load_mgr_regs->load_cntr0);
2532 else
2533 writel(0x40, &sdr_rw_load_mgr_regs->load_cntr0);
2534
2535 writel(mcc_instruction, &sdr_rw_load_jump_mgr_regs->load_jump_add0);
2536
2537 /*
2538 * CNTR 1 - This is used to ensure enough time elapses
2539 * for read data to come back.
2540 */
2541 writel(0x30, &sdr_rw_load_mgr_regs->load_cntr1);
2542
2543 if (test_dm) {
2544 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_WAIT,
2545 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
2546 } else {
2547 writel(RW_MGR_LFSR_WR_RD_BANK_0_WAIT,
2548 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
2549 }
2550
2551 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
2552 writel(mcc_instruction, addr + (group << 2));
2553 }
2554
2555 /* Test writes, can check for a single bit pass or multiple bit pass */
2556 static uint32_t rw_mgr_mem_calibrate_write_test(uint32_t rank_bgn,
2557 uint32_t write_group, uint32_t use_dm, uint32_t all_correct,
2558 uint32_t *bit_chk, uint32_t all_ranks)
2559 {
2560 uint32_t r;
2561 uint32_t correct_mask_vg;
2562 uint32_t tmp_bit_chk;
2563 uint32_t vg;
2564 uint32_t rank_end = all_ranks ? RW_MGR_MEM_NUMBER_OF_RANKS :
2565 (rank_bgn + NUM_RANKS_PER_SHADOW_REG);
2566 uint32_t addr_rw_mgr;
2567 uint32_t base_rw_mgr;
2568
2569 *bit_chk = param->write_correct_mask;
2570 correct_mask_vg = param->write_correct_mask_vg;
2571
2572 for (r = rank_bgn; r < rank_end; r++) {
2573 if (param->skip_ranks[r]) {
2574 /* request to skip the rank */
2575 continue;
2576 }
2577
2578 /* set rank */
2579 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_READ_WRITE);
2580
2581 tmp_bit_chk = 0;
2582 addr_rw_mgr = SDR_PHYGRP_RWMGRGRP_ADDRESS;
2583 for (vg = RW_MGR_MEM_VIRTUAL_GROUPS_PER_WRITE_DQS-1; ; vg--) {
2584 /* reset the fifos to get pointers to known state */
2585 writel(0, &phy_mgr_cmd->fifo_reset);
2586
2587 tmp_bit_chk = tmp_bit_chk <<
2588 (RW_MGR_MEM_DQ_PER_WRITE_DQS /
2589 RW_MGR_MEM_VIRTUAL_GROUPS_PER_WRITE_DQS);
2590 rw_mgr_mem_calibrate_write_test_issue(write_group *
2591 RW_MGR_MEM_VIRTUAL_GROUPS_PER_WRITE_DQS+vg,
2592 use_dm);
2593
2594 base_rw_mgr = readl(addr_rw_mgr);
2595 tmp_bit_chk = tmp_bit_chk | (correct_mask_vg & ~(base_rw_mgr));
2596 if (vg == 0)
2597 break;
2598 }
2599 *bit_chk &= tmp_bit_chk;
2600 }
2601
2602 if (all_correct) {
2603 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
2604 debug_cond(DLEVEL == 2, "write_test(%u,%u,ALL) : %u == \
2605 %u => %lu", write_group, use_dm,
2606 *bit_chk, param->write_correct_mask,
2607 (long unsigned int)(*bit_chk ==
2608 param->write_correct_mask));
2609 return *bit_chk == param->write_correct_mask;
2610 } else {
2611 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
2612 debug_cond(DLEVEL == 2, "write_test(%u,%u,ONE) : %u != ",
2613 write_group, use_dm, *bit_chk);
2614 debug_cond(DLEVEL == 2, "%lu" " => %lu", (long unsigned int)0,
2615 (long unsigned int)(*bit_chk != 0));
2616 return *bit_chk != 0x00;
2617 }
2618 }
2619
2620 /*
2621 * center all windows. do per-bit-deskew to possibly increase size of
2622 * certain windows.
2623 */
2624 static uint32_t rw_mgr_mem_calibrate_writes_center(uint32_t rank_bgn,
2625 uint32_t write_group, uint32_t test_bgn)
2626 {
2627 uint32_t i, p, min_index;
2628 int32_t d;
2629 /*
2630 * Store these as signed since there are comparisons with
2631 * signed numbers.
2632 */
2633 uint32_t bit_chk;
2634 uint32_t sticky_bit_chk;
2635 int32_t left_edge[RW_MGR_MEM_DQ_PER_WRITE_DQS];
2636 int32_t right_edge[RW_MGR_MEM_DQ_PER_WRITE_DQS];
2637 int32_t mid;
2638 int32_t mid_min, orig_mid_min;
2639 int32_t new_dqs, start_dqs, shift_dq;
2640 int32_t dq_margin, dqs_margin, dm_margin;
2641 uint32_t stop;
2642 uint32_t temp_dq_out1_delay;
2643 uint32_t addr;
2644
2645 debug("%s:%d %u %u", __func__, __LINE__, write_group, test_bgn);
2646
2647 dm_margin = 0;
2648
2649 addr = SDR_PHYGRP_SCCGRP_ADDRESS | SCC_MGR_IO_OUT1_DELAY_OFFSET;
2650 start_dqs = readl(addr +
2651 (RW_MGR_MEM_DQ_PER_WRITE_DQS << 2));
2652
2653 /* per-bit deskew */
2654
2655 /*
2656 * set the left and right edge of each bit to an illegal value
2657 * use (IO_IO_OUT1_DELAY_MAX + 1) as an illegal value.
2658 */
2659 sticky_bit_chk = 0;
2660 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2661 left_edge[i] = IO_IO_OUT1_DELAY_MAX + 1;
2662 right_edge[i] = IO_IO_OUT1_DELAY_MAX + 1;
2663 }
2664
2665 /* Search for the left edge of the window for each bit */
2666 for (d = 0; d <= IO_IO_OUT1_DELAY_MAX; d++) {
2667 scc_mgr_apply_group_dq_out1_delay(write_group, d);
2668
2669 writel(0, &sdr_scc_mgr->update);
2670
2671 /*
2672 * Stop searching when the read test doesn't pass AND when
2673 * we've seen a passing read on every bit.
2674 */
2675 stop = !rw_mgr_mem_calibrate_write_test(rank_bgn, write_group,
2676 0, PASS_ONE_BIT, &bit_chk, 0);
2677 sticky_bit_chk = sticky_bit_chk | bit_chk;
2678 stop = stop && (sticky_bit_chk == param->write_correct_mask);
2679 debug_cond(DLEVEL == 2, "write_center(left): dtap=%d => %u \
2680 == %u && %u [bit_chk= %u ]\n",
2681 d, sticky_bit_chk, param->write_correct_mask,
2682 stop, bit_chk);
2683
2684 if (stop == 1) {
2685 break;
2686 } else {
2687 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2688 if (bit_chk & 1) {
2689 /*
2690 * Remember a passing test as the
2691 * left_edge.
2692 */
2693 left_edge[i] = d;
2694 } else {
2695 /*
2696 * If a left edge has not been seen
2697 * yet, then a future passing test will
2698 * mark this edge as the right edge.
2699 */
2700 if (left_edge[i] ==
2701 IO_IO_OUT1_DELAY_MAX + 1) {
2702 right_edge[i] = -(d + 1);
2703 }
2704 }
2705 debug_cond(DLEVEL == 2, "write_center[l,d=%d):", d);
2706 debug_cond(DLEVEL == 2, "bit_chk_test=%d left_edge[%u]: %d",
2707 (int)(bit_chk & 1), i, left_edge[i]);
2708 debug_cond(DLEVEL == 2, "right_edge[%u]: %d\n", i,
2709 right_edge[i]);
2710 bit_chk = bit_chk >> 1;
2711 }
2712 }
2713 }
2714
2715 /* Reset DQ delay chains to 0 */
2716 scc_mgr_apply_group_dq_out1_delay(0);
2717 sticky_bit_chk = 0;
2718 for (i = RW_MGR_MEM_DQ_PER_WRITE_DQS - 1;; i--) {
2719 debug_cond(DLEVEL == 2, "%s:%d write_center: left_edge[%u]: \
2720 %d right_edge[%u]: %d\n", __func__, __LINE__,
2721 i, left_edge[i], i, right_edge[i]);
2722
2723 /*
2724 * Check for cases where we haven't found the left edge,
2725 * which makes our assignment of the the right edge invalid.
2726 * Reset it to the illegal value.
2727 */
2728 if ((left_edge[i] == IO_IO_OUT1_DELAY_MAX + 1) &&
2729 (right_edge[i] != IO_IO_OUT1_DELAY_MAX + 1)) {
2730 right_edge[i] = IO_IO_OUT1_DELAY_MAX + 1;
2731 debug_cond(DLEVEL == 2, "%s:%d write_center: reset \
2732 right_edge[%u]: %d\n", __func__, __LINE__,
2733 i, right_edge[i]);
2734 }
2735
2736 /*
2737 * Reset sticky bit (except for bits where we have
2738 * seen the left edge).
2739 */
2740 sticky_bit_chk = sticky_bit_chk << 1;
2741 if ((left_edge[i] != IO_IO_OUT1_DELAY_MAX + 1))
2742 sticky_bit_chk = sticky_bit_chk | 1;
2743
2744 if (i == 0)
2745 break;
2746 }
2747
2748 /* Search for the right edge of the window for each bit */
2749 for (d = 0; d <= IO_IO_OUT1_DELAY_MAX - start_dqs; d++) {
2750 scc_mgr_apply_group_dqs_io_and_oct_out1(write_group,
2751 d + start_dqs);
2752
2753 writel(0, &sdr_scc_mgr->update);
2754
2755 /*
2756 * Stop searching when the read test doesn't pass AND when
2757 * we've seen a passing read on every bit.
2758 */
2759 stop = !rw_mgr_mem_calibrate_write_test(rank_bgn, write_group,
2760 0, PASS_ONE_BIT, &bit_chk, 0);
2761
2762 sticky_bit_chk = sticky_bit_chk | bit_chk;
2763 stop = stop && (sticky_bit_chk == param->write_correct_mask);
2764
2765 debug_cond(DLEVEL == 2, "write_center (right): dtap=%u => %u == \
2766 %u && %u\n", d, sticky_bit_chk,
2767 param->write_correct_mask, stop);
2768
2769 if (stop == 1) {
2770 if (d == 0) {
2771 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS;
2772 i++) {
2773 /* d = 0 failed, but it passed when
2774 testing the left edge, so it must be
2775 marginal, set it to -1 */
2776 if (right_edge[i] ==
2777 IO_IO_OUT1_DELAY_MAX + 1 &&
2778 left_edge[i] !=
2779 IO_IO_OUT1_DELAY_MAX + 1) {
2780 right_edge[i] = -1;
2781 }
2782 }
2783 }
2784 break;
2785 } else {
2786 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2787 if (bit_chk & 1) {
2788 /*
2789 * Remember a passing test as
2790 * the right_edge.
2791 */
2792 right_edge[i] = d;
2793 } else {
2794 if (d != 0) {
2795 /*
2796 * If a right edge has not
2797 * been seen yet, then a future
2798 * passing test will mark this
2799 * edge as the left edge.
2800 */
2801 if (right_edge[i] ==
2802 IO_IO_OUT1_DELAY_MAX + 1)
2803 left_edge[i] = -(d + 1);
2804 } else {
2805 /*
2806 * d = 0 failed, but it passed
2807 * when testing the left edge,
2808 * so it must be marginal, set
2809 * it to -1.
2810 */
2811 if (right_edge[i] ==
2812 IO_IO_OUT1_DELAY_MAX + 1 &&
2813 left_edge[i] !=
2814 IO_IO_OUT1_DELAY_MAX + 1)
2815 right_edge[i] = -1;
2816 /*
2817 * If a right edge has not been
2818 * seen yet, then a future
2819 * passing test will mark this
2820 * edge as the left edge.
2821 */
2822 else if (right_edge[i] ==
2823 IO_IO_OUT1_DELAY_MAX +
2824 1)
2825 left_edge[i] = -(d + 1);
2826 }
2827 }
2828 debug_cond(DLEVEL == 2, "write_center[r,d=%d):", d);
2829 debug_cond(DLEVEL == 2, "bit_chk_test=%d left_edge[%u]: %d",
2830 (int)(bit_chk & 1), i, left_edge[i]);
2831 debug_cond(DLEVEL == 2, "right_edge[%u]: %d\n", i,
2832 right_edge[i]);
2833 bit_chk = bit_chk >> 1;
2834 }
2835 }
2836 }
2837
2838 /* Check that all bits have a window */
2839 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2840 debug_cond(DLEVEL == 2, "%s:%d write_center: left_edge[%u]: \
2841 %d right_edge[%u]: %d", __func__, __LINE__,
2842 i, left_edge[i], i, right_edge[i]);
2843 if ((left_edge[i] == IO_IO_OUT1_DELAY_MAX + 1) ||
2844 (right_edge[i] == IO_IO_OUT1_DELAY_MAX + 1)) {
2845 set_failing_group_stage(test_bgn + i,
2846 CAL_STAGE_WRITES,
2847 CAL_SUBSTAGE_WRITES_CENTER);
2848 return 0;
2849 }
2850 }
2851
2852 /* Find middle of window for each DQ bit */
2853 mid_min = left_edge[0] - right_edge[0];
2854 min_index = 0;
2855 for (i = 1; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2856 mid = left_edge[i] - right_edge[i];
2857 if (mid < mid_min) {
2858 mid_min = mid;
2859 min_index = i;
2860 }
2861 }
2862
2863 /*
2864 * -mid_min/2 represents the amount that we need to move DQS.
2865 * If mid_min is odd and positive we'll need to add one to
2866 * make sure the rounding in further calculations is correct
2867 * (always bias to the right), so just add 1 for all positive values.
2868 */
2869 if (mid_min > 0)
2870 mid_min++;
2871 mid_min = mid_min / 2;
2872 debug_cond(DLEVEL == 1, "%s:%d write_center: mid_min=%d\n", __func__,
2873 __LINE__, mid_min);
2874
2875 /* Determine the amount we can change DQS (which is -mid_min) */
2876 orig_mid_min = mid_min;
2877 new_dqs = start_dqs;
2878 mid_min = 0;
2879 debug_cond(DLEVEL == 1, "%s:%d write_center: start_dqs=%d new_dqs=%d \
2880 mid_min=%d\n", __func__, __LINE__, start_dqs, new_dqs, mid_min);
2881 /* Initialize data for export structures */
2882 dqs_margin = IO_IO_OUT1_DELAY_MAX + 1;
2883 dq_margin = IO_IO_OUT1_DELAY_MAX + 1;
2884
2885 /* add delay to bring centre of all DQ windows to the same "level" */
2886 for (i = 0, p = test_bgn; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++, p++) {
2887 /* Use values before divide by 2 to reduce round off error */
2888 shift_dq = (left_edge[i] - right_edge[i] -
2889 (left_edge[min_index] - right_edge[min_index]))/2 +
2890 (orig_mid_min - mid_min);
2891
2892 debug_cond(DLEVEL == 2, "%s:%d write_center: before: shift_dq \
2893 [%u]=%d\n", __func__, __LINE__, i, shift_dq);
2894
2895 addr = SDR_PHYGRP_SCCGRP_ADDRESS | SCC_MGR_IO_OUT1_DELAY_OFFSET;
2896 temp_dq_out1_delay = readl(addr + (i << 2));
2897 if (shift_dq + (int32_t)temp_dq_out1_delay >
2898 (int32_t)IO_IO_OUT1_DELAY_MAX) {
2899 shift_dq = (int32_t)IO_IO_OUT1_DELAY_MAX - temp_dq_out1_delay;
2900 } else if (shift_dq + (int32_t)temp_dq_out1_delay < 0) {
2901 shift_dq = -(int32_t)temp_dq_out1_delay;
2902 }
2903 debug_cond(DLEVEL == 2, "write_center: after: shift_dq[%u]=%d\n",
2904 i, shift_dq);
2905 scc_mgr_set_dq_out1_delay(i, temp_dq_out1_delay + shift_dq);
2906 scc_mgr_load_dq(i);
2907
2908 debug_cond(DLEVEL == 2, "write_center: margin[%u]=[%d,%d]\n", i,
2909 left_edge[i] - shift_dq + (-mid_min),
2910 right_edge[i] + shift_dq - (-mid_min));
2911 /* To determine values for export structures */
2912 if (left_edge[i] - shift_dq + (-mid_min) < dq_margin)
2913 dq_margin = left_edge[i] - shift_dq + (-mid_min);
2914
2915 if (right_edge[i] + shift_dq - (-mid_min) < dqs_margin)
2916 dqs_margin = right_edge[i] + shift_dq - (-mid_min);
2917 }
2918
2919 /* Move DQS */
2920 scc_mgr_apply_group_dqs_io_and_oct_out1(write_group, new_dqs);
2921 writel(0, &sdr_scc_mgr->update);
2922
2923 /* Centre DM */
2924 debug_cond(DLEVEL == 2, "%s:%d write_center: DM\n", __func__, __LINE__);
2925
2926 /*
2927 * set the left and right edge of each bit to an illegal value,
2928 * use (IO_IO_OUT1_DELAY_MAX + 1) as an illegal value,
2929 */
2930 left_edge[0] = IO_IO_OUT1_DELAY_MAX + 1;
2931 right_edge[0] = IO_IO_OUT1_DELAY_MAX + 1;
2932 int32_t bgn_curr = IO_IO_OUT1_DELAY_MAX + 1;
2933 int32_t end_curr = IO_IO_OUT1_DELAY_MAX + 1;
2934 int32_t bgn_best = IO_IO_OUT1_DELAY_MAX + 1;
2935 int32_t end_best = IO_IO_OUT1_DELAY_MAX + 1;
2936 int32_t win_best = 0;
2937
2938 /* Search for the/part of the window with DM shift */
2939 for (d = IO_IO_OUT1_DELAY_MAX; d >= 0; d -= DELTA_D) {
2940 scc_mgr_apply_group_dm_out1_delay(d);
2941 writel(0, &sdr_scc_mgr->update);
2942
2943 if (rw_mgr_mem_calibrate_write_test(rank_bgn, write_group, 1,
2944 PASS_ALL_BITS, &bit_chk,
2945 0)) {
2946 /* USE Set current end of the window */
2947 end_curr = -d;
2948 /*
2949 * If a starting edge of our window has not been seen
2950 * this is our current start of the DM window.
2951 */
2952 if (bgn_curr == IO_IO_OUT1_DELAY_MAX + 1)
2953 bgn_curr = -d;
2954
2955 /*
2956 * If current window is bigger than best seen.
2957 * Set best seen to be current window.
2958 */
2959 if ((end_curr-bgn_curr+1) > win_best) {
2960 win_best = end_curr-bgn_curr+1;
2961 bgn_best = bgn_curr;
2962 end_best = end_curr;
2963 }
2964 } else {
2965 /* We just saw a failing test. Reset temp edge */
2966 bgn_curr = IO_IO_OUT1_DELAY_MAX + 1;
2967 end_curr = IO_IO_OUT1_DELAY_MAX + 1;
2968 }
2969 }
2970
2971
2972 /* Reset DM delay chains to 0 */
2973 scc_mgr_apply_group_dm_out1_delay(0);
2974
2975 /*
2976 * Check to see if the current window nudges up aganist 0 delay.
2977 * If so we need to continue the search by shifting DQS otherwise DQS
2978 * search begins as a new search. */
2979 if (end_curr != 0) {
2980 bgn_curr = IO_IO_OUT1_DELAY_MAX + 1;
2981 end_curr = IO_IO_OUT1_DELAY_MAX + 1;
2982 }
2983
2984 /* Search for the/part of the window with DQS shifts */
2985 for (d = 0; d <= IO_IO_OUT1_DELAY_MAX - new_dqs; d += DELTA_D) {
2986 /*
2987 * Note: This only shifts DQS, so are we limiting ourselve to
2988 * width of DQ unnecessarily.
2989 */
2990 scc_mgr_apply_group_dqs_io_and_oct_out1(write_group,
2991 d + new_dqs);
2992
2993 writel(0, &sdr_scc_mgr->update);
2994 if (rw_mgr_mem_calibrate_write_test(rank_bgn, write_group, 1,
2995 PASS_ALL_BITS, &bit_chk,
2996 0)) {
2997 /* USE Set current end of the window */
2998 end_curr = d;
2999 /*
3000 * If a beginning edge of our window has not been seen
3001 * this is our current begin of the DM window.
3002 */
3003 if (bgn_curr == IO_IO_OUT1_DELAY_MAX + 1)
3004 bgn_curr = d;
3005
3006 /*
3007 * If current window is bigger than best seen. Set best
3008 * seen to be current window.
3009 */
3010 if ((end_curr-bgn_curr+1) > win_best) {
3011 win_best = end_curr-bgn_curr+1;
3012 bgn_best = bgn_curr;
3013 end_best = end_curr;
3014 }
3015 } else {
3016 /* We just saw a failing test. Reset temp edge */
3017 bgn_curr = IO_IO_OUT1_DELAY_MAX + 1;
3018 end_curr = IO_IO_OUT1_DELAY_MAX + 1;
3019
3020 /* Early exit optimization: if ther remaining delay
3021 chain space is less than already seen largest window
3022 we can exit */
3023 if ((win_best-1) >
3024 (IO_IO_OUT1_DELAY_MAX - new_dqs - d)) {
3025 break;
3026 }
3027 }
3028 }
3029
3030 /* assign left and right edge for cal and reporting; */
3031 left_edge[0] = -1*bgn_best;
3032 right_edge[0] = end_best;
3033
3034 debug_cond(DLEVEL == 2, "%s:%d dm_calib: left=%d right=%d\n", __func__,
3035 __LINE__, left_edge[0], right_edge[0]);
3036
3037 /* Move DQS (back to orig) */
3038 scc_mgr_apply_group_dqs_io_and_oct_out1(write_group, new_dqs);
3039
3040 /* Move DM */
3041
3042 /* Find middle of window for the DM bit */
3043 mid = (left_edge[0] - right_edge[0]) / 2;
3044
3045 /* only move right, since we are not moving DQS/DQ */
3046 if (mid < 0)
3047 mid = 0;
3048
3049 /* dm_marign should fail if we never find a window */
3050 if (win_best == 0)
3051 dm_margin = -1;
3052 else
3053 dm_margin = left_edge[0] - mid;
3054
3055 scc_mgr_apply_group_dm_out1_delay(mid);
3056 writel(0, &sdr_scc_mgr->update);
3057
3058 debug_cond(DLEVEL == 2, "%s:%d dm_calib: left=%d right=%d mid=%d \
3059 dm_margin=%d\n", __func__, __LINE__, left_edge[0],
3060 right_edge[0], mid, dm_margin);
3061 /* Export values */
3062 gbl->fom_out += dq_margin + dqs_margin;
3063
3064 debug_cond(DLEVEL == 2, "%s:%d write_center: dq_margin=%d \
3065 dqs_margin=%d dm_margin=%d\n", __func__, __LINE__,
3066 dq_margin, dqs_margin, dm_margin);
3067
3068 /*
3069 * Do not remove this line as it makes sure all of our
3070 * decisions have been applied.
3071 */
3072 writel(0, &sdr_scc_mgr->update);
3073 return (dq_margin >= 0) && (dqs_margin >= 0) && (dm_margin >= 0);
3074 }
3075
3076 /* calibrate the write operations */
3077 static uint32_t rw_mgr_mem_calibrate_writes(uint32_t rank_bgn, uint32_t g,
3078 uint32_t test_bgn)
3079 {
3080 /* update info for sims */
3081 debug("%s:%d %u %u\n", __func__, __LINE__, g, test_bgn);
3082
3083 reg_file_set_stage(CAL_STAGE_WRITES);
3084 reg_file_set_sub_stage(CAL_SUBSTAGE_WRITES_CENTER);
3085
3086 reg_file_set_group(g);
3087
3088 if (!rw_mgr_mem_calibrate_writes_center(rank_bgn, g, test_bgn)) {
3089 set_failing_group_stage(g, CAL_STAGE_WRITES,
3090 CAL_SUBSTAGE_WRITES_CENTER);
3091 return 0;
3092 }
3093
3094 return 1;
3095 }
3096
3097 /**
3098 * mem_precharge_and_activate() - Precharge all banks and activate
3099 *
3100 * Precharge all banks and activate row 0 in bank "000..." and bank "111...".
3101 */
3102 static void mem_precharge_and_activate(void)
3103 {
3104 int r;
3105
3106 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS; r++) {
3107 /* Test if the rank should be skipped. */
3108 if (param->skip_ranks[r])
3109 continue;
3110
3111 /* Set rank. */
3112 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_OFF);
3113
3114 /* Precharge all banks. */
3115 writel(RW_MGR_PRECHARGE_ALL, SDR_PHYGRP_RWMGRGRP_ADDRESS |
3116 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
3117
3118 writel(0x0F, &sdr_rw_load_mgr_regs->load_cntr0);
3119 writel(RW_MGR_ACTIVATE_0_AND_1_WAIT1,
3120 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
3121
3122 writel(0x0F, &sdr_rw_load_mgr_regs->load_cntr1);
3123 writel(RW_MGR_ACTIVATE_0_AND_1_WAIT2,
3124 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
3125
3126 /* Activate rows. */
3127 writel(RW_MGR_ACTIVATE_0_AND_1, SDR_PHYGRP_RWMGRGRP_ADDRESS |
3128 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
3129 }
3130 }
3131
3132 /**
3133 * mem_init_latency() - Configure memory RLAT and WLAT settings
3134 *
3135 * Configure memory RLAT and WLAT parameters.
3136 */
3137 static void mem_init_latency(void)
3138 {
3139 /*
3140 * For AV/CV, LFIFO is hardened and always runs at full rate
3141 * so max latency in AFI clocks, used here, is correspondingly
3142 * smaller.
3143 */
3144 const u32 max_latency = (1 << MAX_LATENCY_COUNT_WIDTH) - 1;
3145 u32 rlat, wlat;
3146
3147 debug("%s:%d\n", __func__, __LINE__);
3148
3149 /*
3150 * Read in write latency.
3151 * WL for Hard PHY does not include additive latency.
3152 */
3153 wlat = readl(&data_mgr->t_wl_add);
3154 wlat += readl(&data_mgr->mem_t_add);
3155
3156 gbl->rw_wl_nop_cycles = wlat - 1;
3157
3158 /* Read in readl latency. */
3159 rlat = readl(&data_mgr->t_rl_add);
3160
3161 /* Set a pretty high read latency initially. */
3162 gbl->curr_read_lat = rlat + 16;
3163 if (gbl->curr_read_lat > max_latency)
3164 gbl->curr_read_lat = max_latency;
3165
3166 writel(gbl->curr_read_lat, &phy_mgr_cfg->phy_rlat);
3167
3168 /* Advertise write latency. */
3169 writel(wlat, &phy_mgr_cfg->afi_wlat);
3170 }
3171
3172 /* Set VFIFO and LFIFO to instant-on settings in skip calibration mode */
3173 static void mem_skip_calibrate(void)
3174 {
3175 uint32_t vfifo_offset;
3176 uint32_t i, j, r;
3177
3178 debug("%s:%d\n", __func__, __LINE__);
3179 /* Need to update every shadow register set used by the interface */
3180 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
3181 r += NUM_RANKS_PER_SHADOW_REG) {
3182 /*
3183 * Set output phase alignment settings appropriate for
3184 * skip calibration.
3185 */
3186 for (i = 0; i < RW_MGR_MEM_IF_READ_DQS_WIDTH; i++) {
3187 scc_mgr_set_dqs_en_phase(i, 0);
3188 #if IO_DLL_CHAIN_LENGTH == 6
3189 scc_mgr_set_dqdqs_output_phase(i, 6);
3190 #else
3191 scc_mgr_set_dqdqs_output_phase(i, 7);
3192 #endif
3193 /*
3194 * Case:33398
3195 *
3196 * Write data arrives to the I/O two cycles before write
3197 * latency is reached (720 deg).
3198 * -> due to bit-slip in a/c bus
3199 * -> to allow board skew where dqs is longer than ck
3200 * -> how often can this happen!?
3201 * -> can claim back some ptaps for high freq
3202 * support if we can relax this, but i digress...
3203 *
3204 * The write_clk leads mem_ck by 90 deg
3205 * The minimum ptap of the OPA is 180 deg
3206 * Each ptap has (360 / IO_DLL_CHAIN_LENGH) deg of delay
3207 * The write_clk is always delayed by 2 ptaps
3208 *
3209 * Hence, to make DQS aligned to CK, we need to delay
3210 * DQS by:
3211 * (720 - 90 - 180 - 2 * (360 / IO_DLL_CHAIN_LENGTH))
3212 *
3213 * Dividing the above by (360 / IO_DLL_CHAIN_LENGTH)
3214 * gives us the number of ptaps, which simplies to:
3215 *
3216 * (1.25 * IO_DLL_CHAIN_LENGTH - 2)
3217 */
3218 scc_mgr_set_dqdqs_output_phase(i, (1.25 *
3219 IO_DLL_CHAIN_LENGTH - 2));
3220 }
3221 writel(0xff, &sdr_scc_mgr->dqs_ena);
3222 writel(0xff, &sdr_scc_mgr->dqs_io_ena);
3223
3224 for (i = 0; i < RW_MGR_MEM_IF_WRITE_DQS_WIDTH; i++) {
3225 writel(i, SDR_PHYGRP_SCCGRP_ADDRESS |
3226 SCC_MGR_GROUP_COUNTER_OFFSET);
3227 }
3228 writel(0xff, &sdr_scc_mgr->dq_ena);
3229 writel(0xff, &sdr_scc_mgr->dm_ena);
3230 writel(0, &sdr_scc_mgr->update);
3231 }
3232
3233 /* Compensate for simulation model behaviour */
3234 for (i = 0; i < RW_MGR_MEM_IF_READ_DQS_WIDTH; i++) {
3235 scc_mgr_set_dqs_bus_in_delay(i, 10);
3236 scc_mgr_load_dqs(i);
3237 }
3238 writel(0, &sdr_scc_mgr->update);
3239
3240 /*
3241 * ArriaV has hard FIFOs that can only be initialized by incrementing
3242 * in sequencer.
3243 */
3244 vfifo_offset = CALIB_VFIFO_OFFSET;
3245 for (j = 0; j < vfifo_offset; j++) {
3246 writel(0xff, &phy_mgr_cmd->inc_vfifo_hard_phy);
3247 }
3248 writel(0, &phy_mgr_cmd->fifo_reset);
3249
3250 /*
3251 * For ACV with hard lfifo, we get the skip-cal setting from
3252 * generation-time constant.
3253 */
3254 gbl->curr_read_lat = CALIB_LFIFO_OFFSET;
3255 writel(gbl->curr_read_lat, &phy_mgr_cfg->phy_rlat);
3256 }
3257
3258 /* Memory calibration entry point */
3259 static uint32_t mem_calibrate(void)
3260 {
3261 uint32_t i;
3262 uint32_t rank_bgn, sr;
3263 uint32_t write_group, write_test_bgn;
3264 uint32_t read_group, read_test_bgn;
3265 uint32_t run_groups, current_run;
3266 uint32_t failing_groups = 0;
3267 uint32_t group_failed = 0;
3268 uint32_t sr_failed = 0;
3269
3270 debug("%s:%d\n", __func__, __LINE__);
3271
3272 /* Initialize the data settings */
3273 gbl->error_substage = CAL_SUBSTAGE_NIL;
3274 gbl->error_stage = CAL_STAGE_NIL;
3275 gbl->error_group = 0xff;
3276 gbl->fom_in = 0;
3277 gbl->fom_out = 0;
3278
3279 /* Initialize WLAT and RLAT. */
3280 mem_init_latency();
3281
3282 /* Initialize bit slips. */
3283 mem_precharge_and_activate();
3284
3285 for (i = 0; i < RW_MGR_MEM_IF_READ_DQS_WIDTH; i++) {
3286 writel(i, SDR_PHYGRP_SCCGRP_ADDRESS |
3287 SCC_MGR_GROUP_COUNTER_OFFSET);
3288 /* Only needed once to set all groups, pins, DQ, DQS, DM. */
3289 if (i == 0)
3290 scc_mgr_set_hhp_extras();
3291
3292 scc_set_bypass_mode(i);
3293 }
3294
3295 if ((dyn_calib_steps & CALIB_SKIP_ALL) == CALIB_SKIP_ALL) {
3296 /*
3297 * Set VFIFO and LFIFO to instant-on settings in skip
3298 * calibration mode.
3299 */
3300 mem_skip_calibrate();
3301 } else {
3302 for (i = 0; i < NUM_CALIB_REPEAT; i++) {
3303 /*
3304 * Zero all delay chain/phase settings for all
3305 * groups and all shadow register sets.
3306 */
3307 scc_mgr_zero_all();
3308
3309 run_groups = ~param->skip_groups;
3310
3311 for (write_group = 0, write_test_bgn = 0; write_group
3312 < RW_MGR_MEM_IF_WRITE_DQS_WIDTH; write_group++,
3313 write_test_bgn += RW_MGR_MEM_DQ_PER_WRITE_DQS) {
3314 /* Initialized the group failure */
3315 group_failed = 0;
3316
3317 current_run = run_groups & ((1 <<
3318 RW_MGR_NUM_DQS_PER_WRITE_GROUP) - 1);
3319 run_groups = run_groups >>
3320 RW_MGR_NUM_DQS_PER_WRITE_GROUP;
3321
3322 if (current_run == 0)
3323 continue;
3324
3325 writel(write_group, SDR_PHYGRP_SCCGRP_ADDRESS |
3326 SCC_MGR_GROUP_COUNTER_OFFSET);
3327 scc_mgr_zero_group(write_group, 0);
3328
3329 for (read_group = write_group *
3330 RW_MGR_MEM_IF_READ_DQS_WIDTH /
3331 RW_MGR_MEM_IF_WRITE_DQS_WIDTH,
3332 read_test_bgn = 0;
3333 read_group < (write_group + 1) *
3334 RW_MGR_MEM_IF_READ_DQS_WIDTH /
3335 RW_MGR_MEM_IF_WRITE_DQS_WIDTH &&
3336 group_failed == 0;
3337 read_group++, read_test_bgn +=
3338 RW_MGR_MEM_DQ_PER_READ_DQS) {
3339 /* Calibrate the VFIFO */
3340 if (!((STATIC_CALIB_STEPS) &
3341 CALIB_SKIP_VFIFO)) {
3342 if (!rw_mgr_mem_calibrate_vfifo
3343 (read_group,
3344 read_test_bgn)) {
3345 group_failed = 1;
3346
3347 if (!(gbl->
3348 phy_debug_mode_flags &
3349 PHY_DEBUG_SWEEP_ALL_GROUPS)) {
3350 return 0;
3351 }
3352 }
3353 }
3354 }
3355
3356 /* Calibrate the output side */
3357 if (group_failed == 0) {
3358 for (rank_bgn = 0, sr = 0; rank_bgn
3359 < RW_MGR_MEM_NUMBER_OF_RANKS;
3360 rank_bgn +=
3361 NUM_RANKS_PER_SHADOW_REG,
3362 ++sr) {
3363 sr_failed = 0;
3364 if (!((STATIC_CALIB_STEPS) &
3365 CALIB_SKIP_WRITES)) {
3366 if ((STATIC_CALIB_STEPS)
3367 & CALIB_SKIP_DELAY_SWEEPS) {
3368 /* not needed in quick mode! */
3369 } else {
3370 /*
3371 * Determine if this set of
3372 * ranks should be skipped
3373 * entirely.
3374 */
3375 if (!param->skip_shadow_regs[sr]) {
3376 if (!rw_mgr_mem_calibrate_writes
3377 (rank_bgn, write_group,
3378 write_test_bgn)) {
3379 sr_failed = 1;
3380 if (!(gbl->
3381 phy_debug_mode_flags &
3382 PHY_DEBUG_SWEEP_ALL_GROUPS)) {
3383 return 0;
3384 }
3385 }
3386 }
3387 }
3388 }
3389 if (sr_failed != 0)
3390 group_failed = 1;
3391 }
3392 }
3393
3394 if (group_failed == 0) {
3395 for (read_group = write_group *
3396 RW_MGR_MEM_IF_READ_DQS_WIDTH /
3397 RW_MGR_MEM_IF_WRITE_DQS_WIDTH,
3398 read_test_bgn = 0;
3399 read_group < (write_group + 1)
3400 * RW_MGR_MEM_IF_READ_DQS_WIDTH
3401 / RW_MGR_MEM_IF_WRITE_DQS_WIDTH &&
3402 group_failed == 0;
3403 read_group++, read_test_bgn +=
3404 RW_MGR_MEM_DQ_PER_READ_DQS) {
3405 if (!((STATIC_CALIB_STEPS) &
3406 CALIB_SKIP_WRITES)) {
3407 if (!rw_mgr_mem_calibrate_vfifo_end
3408 (read_group, read_test_bgn)) {
3409 group_failed = 1;
3410
3411 if (!(gbl->phy_debug_mode_flags
3412 & PHY_DEBUG_SWEEP_ALL_GROUPS)) {
3413 return 0;
3414 }
3415 }
3416 }
3417 }
3418 }
3419
3420 if (group_failed != 0)
3421 failing_groups++;
3422 }
3423
3424 /*
3425 * USER If there are any failing groups then report
3426 * the failure.
3427 */
3428 if (failing_groups != 0)
3429 return 0;
3430
3431 /* Calibrate the LFIFO */
3432 if (!((STATIC_CALIB_STEPS) & CALIB_SKIP_LFIFO)) {
3433 /*
3434 * If we're skipping groups as part of debug,
3435 * don't calibrate LFIFO.
3436 */
3437 if (param->skip_groups == 0) {
3438 if (!rw_mgr_mem_calibrate_lfifo())
3439 return 0;
3440 }
3441 }
3442 }
3443 }
3444
3445 /*
3446 * Do not remove this line as it makes sure all of our decisions
3447 * have been applied.
3448 */
3449 writel(0, &sdr_scc_mgr->update);
3450 return 1;
3451 }
3452
3453 /**
3454 * run_mem_calibrate() - Perform memory calibration
3455 *
3456 * This function triggers the entire memory calibration procedure.
3457 */
3458 static int run_mem_calibrate(void)
3459 {
3460 int pass;
3461
3462 debug("%s:%d\n", __func__, __LINE__);
3463
3464 /* Reset pass/fail status shown on afi_cal_success/fail */
3465 writel(PHY_MGR_CAL_RESET, &phy_mgr_cfg->cal_status);
3466
3467 /* Stop tracking manager. */
3468 clrbits_le32(&sdr_ctrl->ctrl_cfg, 1 << 22);
3469
3470 phy_mgr_initialize();
3471 rw_mgr_mem_initialize();
3472
3473 /* Perform the actual memory calibration. */
3474 pass = mem_calibrate();
3475
3476 mem_precharge_and_activate();
3477 writel(0, &phy_mgr_cmd->fifo_reset);
3478
3479 /* Handoff. */
3480 rw_mgr_mem_handoff();
3481 /*
3482 * In Hard PHY this is a 2-bit control:
3483 * 0: AFI Mux Select
3484 * 1: DDIO Mux Select
3485 */
3486 writel(0x2, &phy_mgr_cfg->mux_sel);
3487
3488 /* Start tracking manager. */
3489 setbits_le32(&sdr_ctrl->ctrl_cfg, 1 << 22);
3490
3491 return pass;
3492 }
3493
3494 /**
3495 * debug_mem_calibrate() - Report result of memory calibration
3496 * @pass: Value indicating whether calibration passed or failed
3497 *
3498 * This function reports the results of the memory calibration
3499 * and writes debug information into the register file.
3500 */
3501 static void debug_mem_calibrate(int pass)
3502 {
3503 uint32_t debug_info;
3504
3505 if (pass) {
3506 printf("%s: CALIBRATION PASSED\n", __FILE__);
3507
3508 gbl->fom_in /= 2;
3509 gbl->fom_out /= 2;
3510
3511 if (gbl->fom_in > 0xff)
3512 gbl->fom_in = 0xff;
3513
3514 if (gbl->fom_out > 0xff)
3515 gbl->fom_out = 0xff;
3516
3517 /* Update the FOM in the register file */
3518 debug_info = gbl->fom_in;
3519 debug_info |= gbl->fom_out << 8;
3520 writel(debug_info, &sdr_reg_file->fom);
3521
3522 writel(debug_info, &phy_mgr_cfg->cal_debug_info);
3523 writel(PHY_MGR_CAL_SUCCESS, &phy_mgr_cfg->cal_status);
3524 } else {
3525 printf("%s: CALIBRATION FAILED\n", __FILE__);
3526
3527 debug_info = gbl->error_stage;
3528 debug_info |= gbl->error_substage << 8;
3529 debug_info |= gbl->error_group << 16;
3530
3531 writel(debug_info, &sdr_reg_file->failing_stage);
3532 writel(debug_info, &phy_mgr_cfg->cal_debug_info);
3533 writel(PHY_MGR_CAL_FAIL, &phy_mgr_cfg->cal_status);
3534
3535 /* Update the failing group/stage in the register file */
3536 debug_info = gbl->error_stage;
3537 debug_info |= gbl->error_substage << 8;
3538 debug_info |= gbl->error_group << 16;
3539 writel(debug_info, &sdr_reg_file->failing_stage);
3540 }
3541
3542 printf("%s: Calibration complete\n", __FILE__);
3543 }
3544
3545 /**
3546 * hc_initialize_rom_data() - Initialize ROM data
3547 *
3548 * Initialize ROM data.
3549 */
3550 static void hc_initialize_rom_data(void)
3551 {
3552 u32 i, addr;
3553
3554 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_INST_ROM_WRITE_OFFSET;
3555 for (i = 0; i < ARRAY_SIZE(inst_rom_init); i++)
3556 writel(inst_rom_init[i], addr + (i << 2));
3557
3558 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_AC_ROM_WRITE_OFFSET;
3559 for (i = 0; i < ARRAY_SIZE(ac_rom_init); i++)
3560 writel(ac_rom_init[i], addr + (i << 2));
3561 }
3562
3563 /**
3564 * initialize_reg_file() - Initialize SDR register file
3565 *
3566 * Initialize SDR register file.
3567 */
3568 static void initialize_reg_file(void)
3569 {
3570 /* Initialize the register file with the correct data */
3571 writel(REG_FILE_INIT_SEQ_SIGNATURE, &sdr_reg_file->signature);
3572 writel(0, &sdr_reg_file->debug_data_addr);
3573 writel(0, &sdr_reg_file->cur_stage);
3574 writel(0, &sdr_reg_file->fom);
3575 writel(0, &sdr_reg_file->failing_stage);
3576 writel(0, &sdr_reg_file->debug1);
3577 writel(0, &sdr_reg_file->debug2);
3578 }
3579
3580 /**
3581 * initialize_hps_phy() - Initialize HPS PHY
3582 *
3583 * Initialize HPS PHY.
3584 */
3585 static void initialize_hps_phy(void)
3586 {
3587 uint32_t reg;
3588 /*
3589 * Tracking also gets configured here because it's in the
3590 * same register.
3591 */
3592 uint32_t trk_sample_count = 7500;
3593 uint32_t trk_long_idle_sample_count = (10 << 16) | 100;
3594 /*
3595 * Format is number of outer loops in the 16 MSB, sample
3596 * count in 16 LSB.
3597 */
3598
3599 reg = 0;
3600 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_ACDELAYEN_SET(2);
3601 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_DQDELAYEN_SET(1);
3602 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_DQSDELAYEN_SET(1);
3603 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_DQSLOGICDELAYEN_SET(1);
3604 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_RESETDELAYEN_SET(0);
3605 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_LPDDRDIS_SET(1);
3606 /*
3607 * This field selects the intrinsic latency to RDATA_EN/FULL path.
3608 * 00-bypass, 01- add 5 cycles, 10- add 10 cycles, 11- add 15 cycles.
3609 */
3610 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_ADDLATSEL_SET(0);
3611 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_SAMPLECOUNT_19_0_SET(
3612 trk_sample_count);
3613 writel(reg, &sdr_ctrl->phy_ctrl0);
3614
3615 reg = 0;
3616 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_1_SAMPLECOUNT_31_20_SET(
3617 trk_sample_count >>
3618 SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_SAMPLECOUNT_19_0_WIDTH);
3619 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_1_LONGIDLESAMPLECOUNT_19_0_SET(
3620 trk_long_idle_sample_count);
3621 writel(reg, &sdr_ctrl->phy_ctrl1);
3622
3623 reg = 0;
3624 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_2_LONGIDLESAMPLECOUNT_31_20_SET(
3625 trk_long_idle_sample_count >>
3626 SDR_CTRLGRP_PHYCTRL_PHYCTRL_1_LONGIDLESAMPLECOUNT_19_0_WIDTH);
3627 writel(reg, &sdr_ctrl->phy_ctrl2);
3628 }
3629
3630 /**
3631 * initialize_tracking() - Initialize tracking
3632 *
3633 * Initialize the register file with usable initial data.
3634 */
3635 static void initialize_tracking(void)
3636 {
3637 /*
3638 * Initialize the register file with the correct data.
3639 * Compute usable version of value in case we skip full
3640 * computation later.
3641 */
3642 writel(DIV_ROUND_UP(IO_DELAY_PER_OPA_TAP, IO_DELAY_PER_DCHAIN_TAP) - 1,
3643 &sdr_reg_file->dtaps_per_ptap);
3644
3645 /* trk_sample_count */
3646 writel(7500, &sdr_reg_file->trk_sample_count);
3647
3648 /* longidle outer loop [15:0] */
3649 writel((10 << 16) | (100 << 0), &sdr_reg_file->trk_longidle);
3650
3651 /*
3652 * longidle sample count [31:24]
3653 * trfc, worst case of 933Mhz 4Gb [23:16]
3654 * trcd, worst case [15:8]
3655 * vfifo wait [7:0]
3656 */
3657 writel((243 << 24) | (14 << 16) | (10 << 8) | (4 << 0),
3658 &sdr_reg_file->delays);
3659
3660 /* mux delay */
3661 writel((RW_MGR_IDLE << 24) | (RW_MGR_ACTIVATE_1 << 16) |
3662 (RW_MGR_SGLE_READ << 8) | (RW_MGR_PRECHARGE_ALL << 0),
3663 &sdr_reg_file->trk_rw_mgr_addr);
3664
3665 writel(RW_MGR_MEM_IF_READ_DQS_WIDTH,
3666 &sdr_reg_file->trk_read_dqs_width);
3667
3668 /* trefi [7:0] */
3669 writel((RW_MGR_REFRESH_ALL << 24) | (1000 << 0),
3670 &sdr_reg_file->trk_rfsh);
3671 }
3672
3673 int sdram_calibration_full(void)
3674 {
3675 struct param_type my_param;
3676 struct gbl_type my_gbl;
3677 uint32_t pass;
3678
3679 memset(&my_param, 0, sizeof(my_param));
3680 memset(&my_gbl, 0, sizeof(my_gbl));
3681
3682 param = &my_param;
3683 gbl = &my_gbl;
3684
3685 /* Set the calibration enabled by default */
3686 gbl->phy_debug_mode_flags |= PHY_DEBUG_ENABLE_CAL_RPT;
3687 /*
3688 * Only sweep all groups (regardless of fail state) by default
3689 * Set enabled read test by default.
3690 */
3691 #if DISABLE_GUARANTEED_READ
3692 gbl->phy_debug_mode_flags |= PHY_DEBUG_DISABLE_GUARANTEED_READ;
3693 #endif
3694 /* Initialize the register file */
3695 initialize_reg_file();
3696
3697 /* Initialize any PHY CSR */
3698 initialize_hps_phy();
3699
3700 scc_mgr_initialize();
3701
3702 initialize_tracking();
3703
3704 printf("%s: Preparing to start memory calibration\n", __FILE__);
3705
3706 debug("%s:%d\n", __func__, __LINE__);
3707 debug_cond(DLEVEL == 1,
3708 "DDR3 FULL_RATE ranks=%u cs/dimm=%u dq/dqs=%u,%u vg/dqs=%u,%u ",
3709 RW_MGR_MEM_NUMBER_OF_RANKS, RW_MGR_MEM_NUMBER_OF_CS_PER_DIMM,
3710 RW_MGR_MEM_DQ_PER_READ_DQS, RW_MGR_MEM_DQ_PER_WRITE_DQS,
3711 RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS,
3712 RW_MGR_MEM_VIRTUAL_GROUPS_PER_WRITE_DQS);
3713 debug_cond(DLEVEL == 1,
3714 "dqs=%u,%u dq=%u dm=%u ptap_delay=%u dtap_delay=%u ",
3715 RW_MGR_MEM_IF_READ_DQS_WIDTH, RW_MGR_MEM_IF_WRITE_DQS_WIDTH,
3716 RW_MGR_MEM_DATA_WIDTH, RW_MGR_MEM_DATA_MASK_WIDTH,
3717 IO_DELAY_PER_OPA_TAP, IO_DELAY_PER_DCHAIN_TAP);
3718 debug_cond(DLEVEL == 1, "dtap_dqsen_delay=%u, dll=%u",
3719 IO_DELAY_PER_DQS_EN_DCHAIN_TAP, IO_DLL_CHAIN_LENGTH);
3720 debug_cond(DLEVEL == 1, "max values: en_p=%u dqdqs_p=%u en_d=%u dqs_in_d=%u ",
3721 IO_DQS_EN_PHASE_MAX, IO_DQDQS_OUT_PHASE_MAX,
3722 IO_DQS_EN_DELAY_MAX, IO_DQS_IN_DELAY_MAX);
3723 debug_cond(DLEVEL == 1, "io_in_d=%u io_out1_d=%u io_out2_d=%u ",
3724 IO_IO_IN_DELAY_MAX, IO_IO_OUT1_DELAY_MAX,
3725 IO_IO_OUT2_DELAY_MAX);
3726 debug_cond(DLEVEL == 1, "dqs_in_reserve=%u dqs_out_reserve=%u\n",
3727 IO_DQS_IN_RESERVE, IO_DQS_OUT_RESERVE);
3728
3729 hc_initialize_rom_data();
3730
3731 /* update info for sims */
3732 reg_file_set_stage(CAL_STAGE_NIL);
3733 reg_file_set_group(0);
3734
3735 /*
3736 * Load global needed for those actions that require
3737 * some dynamic calibration support.
3738 */
3739 dyn_calib_steps = STATIC_CALIB_STEPS;
3740 /*
3741 * Load global to allow dynamic selection of delay loop settings
3742 * based on calibration mode.
3743 */
3744 if (!(dyn_calib_steps & CALIB_SKIP_DELAY_LOOPS))
3745 skip_delay_mask = 0xff;
3746 else
3747 skip_delay_mask = 0x0;
3748
3749 pass = run_mem_calibrate();
3750 debug_mem_calibrate(pass);
3751 return pass;
3752 }