]> git.ipfire.org Git - people/ms/u-boot.git/blob - drivers/ddr/altera/sequencer.c
e17e6012b0052749d59acac3e27e8274f5b0f13c
[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 <errno.h>
11 #include "sequencer.h"
12 #include "sequencer_auto.h"
13 #include "sequencer_auto_ac_init.h"
14 #include "sequencer_auto_inst_init.h"
15 #include "sequencer_defines.h"
16
17 static struct socfpga_sdr_rw_load_manager *sdr_rw_load_mgr_regs =
18 (struct socfpga_sdr_rw_load_manager *)(SDR_PHYGRP_RWMGRGRP_ADDRESS | 0x800);
19
20 static struct socfpga_sdr_rw_load_jump_manager *sdr_rw_load_jump_mgr_regs =
21 (struct socfpga_sdr_rw_load_jump_manager *)(SDR_PHYGRP_RWMGRGRP_ADDRESS | 0xC00);
22
23 static struct socfpga_sdr_reg_file *sdr_reg_file =
24 (struct socfpga_sdr_reg_file *)SDR_PHYGRP_REGFILEGRP_ADDRESS;
25
26 static struct socfpga_sdr_scc_mgr *sdr_scc_mgr =
27 (struct socfpga_sdr_scc_mgr *)(SDR_PHYGRP_SCCGRP_ADDRESS | 0xe00);
28
29 static struct socfpga_phy_mgr_cmd *phy_mgr_cmd =
30 (struct socfpga_phy_mgr_cmd *)SDR_PHYGRP_PHYMGRGRP_ADDRESS;
31
32 static struct socfpga_phy_mgr_cfg *phy_mgr_cfg =
33 (struct socfpga_phy_mgr_cfg *)(SDR_PHYGRP_PHYMGRGRP_ADDRESS | 0x40);
34
35 static struct socfpga_data_mgr *data_mgr =
36 (struct socfpga_data_mgr *)SDR_PHYGRP_DATAMGRGRP_ADDRESS;
37
38 static struct socfpga_sdr_ctrl *sdr_ctrl =
39 (struct socfpga_sdr_ctrl *)SDR_CTRLGRP_ADDRESS;
40
41 #define DELTA_D 1
42
43 /*
44 * In order to reduce ROM size, most of the selectable calibration steps are
45 * decided at compile time based on the user's calibration mode selection,
46 * as captured by the STATIC_CALIB_STEPS selection below.
47 *
48 * However, to support simulation-time selection of fast simulation mode, where
49 * we skip everything except the bare minimum, we need a few of the steps to
50 * be dynamic. In those cases, we either use the DYNAMIC_CALIB_STEPS for the
51 * check, which is based on the rtl-supplied value, or we dynamically compute
52 * the value to use based on the dynamically-chosen calibration mode
53 */
54
55 #define DLEVEL 0
56 #define STATIC_IN_RTL_SIM 0
57 #define STATIC_SKIP_DELAY_LOOPS 0
58
59 #define STATIC_CALIB_STEPS (STATIC_IN_RTL_SIM | CALIB_SKIP_FULL_TEST | \
60 STATIC_SKIP_DELAY_LOOPS)
61
62 /* calibration steps requested by the rtl */
63 uint16_t dyn_calib_steps;
64
65 /*
66 * To make CALIB_SKIP_DELAY_LOOPS a dynamic conditional option
67 * instead of static, we use boolean logic to select between
68 * non-skip and skip values
69 *
70 * The mask is set to include all bits when not-skipping, but is
71 * zero when skipping
72 */
73
74 uint16_t skip_delay_mask; /* mask off bits when skipping/not-skipping */
75
76 #define SKIP_DELAY_LOOP_VALUE_OR_ZERO(non_skip_value) \
77 ((non_skip_value) & skip_delay_mask)
78
79 struct gbl_type *gbl;
80 struct param_type *param;
81 uint32_t curr_shadow_reg;
82
83 static uint32_t rw_mgr_mem_calibrate_write_test(uint32_t rank_bgn,
84 uint32_t write_group, uint32_t use_dm,
85 uint32_t all_correct, uint32_t *bit_chk, uint32_t all_ranks);
86
87 static void set_failing_group_stage(uint32_t group, uint32_t stage,
88 uint32_t substage)
89 {
90 /*
91 * Only set the global stage if there was not been any other
92 * failing group
93 */
94 if (gbl->error_stage == CAL_STAGE_NIL) {
95 gbl->error_substage = substage;
96 gbl->error_stage = stage;
97 gbl->error_group = group;
98 }
99 }
100
101 static void reg_file_set_group(u16 set_group)
102 {
103 clrsetbits_le32(&sdr_reg_file->cur_stage, 0xffff0000, set_group << 16);
104 }
105
106 static void reg_file_set_stage(u8 set_stage)
107 {
108 clrsetbits_le32(&sdr_reg_file->cur_stage, 0xffff, set_stage & 0xff);
109 }
110
111 static void reg_file_set_sub_stage(u8 set_sub_stage)
112 {
113 set_sub_stage &= 0xff;
114 clrsetbits_le32(&sdr_reg_file->cur_stage, 0xff00, set_sub_stage << 8);
115 }
116
117 /**
118 * phy_mgr_initialize() - Initialize PHY Manager
119 *
120 * Initialize PHY Manager.
121 */
122 static void phy_mgr_initialize(void)
123 {
124 u32 ratio;
125
126 debug("%s:%d\n", __func__, __LINE__);
127 /* Calibration has control over path to memory */
128 /*
129 * In Hard PHY this is a 2-bit control:
130 * 0: AFI Mux Select
131 * 1: DDIO Mux Select
132 */
133 writel(0x3, &phy_mgr_cfg->mux_sel);
134
135 /* USER memory clock is not stable we begin initialization */
136 writel(0, &phy_mgr_cfg->reset_mem_stbl);
137
138 /* USER calibration status all set to zero */
139 writel(0, &phy_mgr_cfg->cal_status);
140
141 writel(0, &phy_mgr_cfg->cal_debug_info);
142
143 /* Init params only if we do NOT skip calibration. */
144 if ((dyn_calib_steps & CALIB_SKIP_ALL) == CALIB_SKIP_ALL)
145 return;
146
147 ratio = RW_MGR_MEM_DQ_PER_READ_DQS /
148 RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS;
149 param->read_correct_mask_vg = (1 << ratio) - 1;
150 param->write_correct_mask_vg = (1 << ratio) - 1;
151 param->read_correct_mask = (1 << RW_MGR_MEM_DQ_PER_READ_DQS) - 1;
152 param->write_correct_mask = (1 << RW_MGR_MEM_DQ_PER_WRITE_DQS) - 1;
153 ratio = RW_MGR_MEM_DATA_WIDTH /
154 RW_MGR_MEM_DATA_MASK_WIDTH;
155 param->dm_correct_mask = (1 << ratio) - 1;
156 }
157
158 /**
159 * set_rank_and_odt_mask() - Set Rank and ODT mask
160 * @rank: Rank mask
161 * @odt_mode: ODT mode, OFF or READ_WRITE
162 *
163 * Set Rank and ODT mask (On-Die Termination).
164 */
165 static void set_rank_and_odt_mask(const u32 rank, const u32 odt_mode)
166 {
167 u32 odt_mask_0 = 0;
168 u32 odt_mask_1 = 0;
169 u32 cs_and_odt_mask;
170
171 if (odt_mode == RW_MGR_ODT_MODE_OFF) {
172 odt_mask_0 = 0x0;
173 odt_mask_1 = 0x0;
174 } else { /* RW_MGR_ODT_MODE_READ_WRITE */
175 switch (RW_MGR_MEM_NUMBER_OF_RANKS) {
176 case 1: /* 1 Rank */
177 /* Read: ODT = 0 ; Write: ODT = 1 */
178 odt_mask_0 = 0x0;
179 odt_mask_1 = 0x1;
180 break;
181 case 2: /* 2 Ranks */
182 if (RW_MGR_MEM_NUMBER_OF_CS_PER_DIMM == 1) {
183 /*
184 * - Dual-Slot , Single-Rank (1 CS per DIMM)
185 * OR
186 * - RDIMM, 4 total CS (2 CS per DIMM, 2 DIMM)
187 *
188 * Since MEM_NUMBER_OF_RANKS is 2, they
189 * are both single rank with 2 CS each
190 * (special for RDIMM).
191 *
192 * Read: Turn on ODT on the opposite rank
193 * Write: Turn on ODT on all ranks
194 */
195 odt_mask_0 = 0x3 & ~(1 << rank);
196 odt_mask_1 = 0x3;
197 } else {
198 /*
199 * - Single-Slot , Dual-Rank (2 CS per DIMM)
200 *
201 * Read: Turn on ODT off on all ranks
202 * Write: Turn on ODT on active rank
203 */
204 odt_mask_0 = 0x0;
205 odt_mask_1 = 0x3 & (1 << rank);
206 }
207 break;
208 case 4: /* 4 Ranks */
209 /* Read:
210 * ----------+-----------------------+
211 * | ODT |
212 * Read From +-----------------------+
213 * Rank | 3 | 2 | 1 | 0 |
214 * ----------+-----+-----+-----+-----+
215 * 0 | 0 | 1 | 0 | 0 |
216 * 1 | 1 | 0 | 0 | 0 |
217 * 2 | 0 | 0 | 0 | 1 |
218 * 3 | 0 | 0 | 1 | 0 |
219 * ----------+-----+-----+-----+-----+
220 *
221 * Write:
222 * ----------+-----------------------+
223 * | ODT |
224 * Write To +-----------------------+
225 * Rank | 3 | 2 | 1 | 0 |
226 * ----------+-----+-----+-----+-----+
227 * 0 | 0 | 1 | 0 | 1 |
228 * 1 | 1 | 0 | 1 | 0 |
229 * 2 | 0 | 1 | 0 | 1 |
230 * 3 | 1 | 0 | 1 | 0 |
231 * ----------+-----+-----+-----+-----+
232 */
233 switch (rank) {
234 case 0:
235 odt_mask_0 = 0x4;
236 odt_mask_1 = 0x5;
237 break;
238 case 1:
239 odt_mask_0 = 0x8;
240 odt_mask_1 = 0xA;
241 break;
242 case 2:
243 odt_mask_0 = 0x1;
244 odt_mask_1 = 0x5;
245 break;
246 case 3:
247 odt_mask_0 = 0x2;
248 odt_mask_1 = 0xA;
249 break;
250 }
251 break;
252 }
253 }
254
255 cs_and_odt_mask = (0xFF & ~(1 << rank)) |
256 ((0xFF & odt_mask_0) << 8) |
257 ((0xFF & odt_mask_1) << 16);
258 writel(cs_and_odt_mask, SDR_PHYGRP_RWMGRGRP_ADDRESS |
259 RW_MGR_SET_CS_AND_ODT_MASK_OFFSET);
260 }
261
262 /**
263 * scc_mgr_set() - Set SCC Manager register
264 * @off: Base offset in SCC Manager space
265 * @grp: Read/Write group
266 * @val: Value to be set
267 *
268 * This function sets the SCC Manager (Scan Chain Control Manager) register.
269 */
270 static void scc_mgr_set(u32 off, u32 grp, u32 val)
271 {
272 writel(val, SDR_PHYGRP_SCCGRP_ADDRESS | off | (grp << 2));
273 }
274
275 /**
276 * scc_mgr_initialize() - Initialize SCC Manager registers
277 *
278 * Initialize SCC Manager registers.
279 */
280 static void scc_mgr_initialize(void)
281 {
282 /*
283 * Clear register file for HPS. 16 (2^4) is the size of the
284 * full register file in the scc mgr:
285 * RFILE_DEPTH = 1 + log2(MEM_DQ_PER_DQS + 1 + MEM_DM_PER_DQS +
286 * MEM_IF_READ_DQS_WIDTH - 1);
287 */
288 int i;
289
290 for (i = 0; i < 16; i++) {
291 debug_cond(DLEVEL == 1, "%s:%d: Clearing SCC RFILE index %u\n",
292 __func__, __LINE__, i);
293 scc_mgr_set(SCC_MGR_HHP_RFILE_OFFSET, 0, i);
294 }
295 }
296
297 static void scc_mgr_set_dqdqs_output_phase(uint32_t write_group, uint32_t phase)
298 {
299 scc_mgr_set(SCC_MGR_DQDQS_OUT_PHASE_OFFSET, write_group, phase);
300 }
301
302 static void scc_mgr_set_dqs_bus_in_delay(uint32_t read_group, uint32_t delay)
303 {
304 scc_mgr_set(SCC_MGR_DQS_IN_DELAY_OFFSET, read_group, delay);
305 }
306
307 static void scc_mgr_set_dqs_en_phase(uint32_t read_group, uint32_t phase)
308 {
309 scc_mgr_set(SCC_MGR_DQS_EN_PHASE_OFFSET, read_group, phase);
310 }
311
312 static void scc_mgr_set_dqs_en_delay(uint32_t read_group, uint32_t delay)
313 {
314 scc_mgr_set(SCC_MGR_DQS_EN_DELAY_OFFSET, read_group, delay);
315 }
316
317 static void scc_mgr_set_dqs_io_in_delay(uint32_t delay)
318 {
319 scc_mgr_set(SCC_MGR_IO_IN_DELAY_OFFSET, RW_MGR_MEM_DQ_PER_WRITE_DQS,
320 delay);
321 }
322
323 static void scc_mgr_set_dq_in_delay(uint32_t dq_in_group, uint32_t delay)
324 {
325 scc_mgr_set(SCC_MGR_IO_IN_DELAY_OFFSET, dq_in_group, delay);
326 }
327
328 static void scc_mgr_set_dq_out1_delay(uint32_t dq_in_group, uint32_t delay)
329 {
330 scc_mgr_set(SCC_MGR_IO_OUT1_DELAY_OFFSET, dq_in_group, delay);
331 }
332
333 static void scc_mgr_set_dqs_out1_delay(uint32_t delay)
334 {
335 scc_mgr_set(SCC_MGR_IO_OUT1_DELAY_OFFSET, RW_MGR_MEM_DQ_PER_WRITE_DQS,
336 delay);
337 }
338
339 static void scc_mgr_set_dm_out1_delay(uint32_t dm, uint32_t delay)
340 {
341 scc_mgr_set(SCC_MGR_IO_OUT1_DELAY_OFFSET,
342 RW_MGR_MEM_DQ_PER_WRITE_DQS + 1 + dm,
343 delay);
344 }
345
346 /* load up dqs config settings */
347 static void scc_mgr_load_dqs(uint32_t dqs)
348 {
349 writel(dqs, &sdr_scc_mgr->dqs_ena);
350 }
351
352 /* load up dqs io config settings */
353 static void scc_mgr_load_dqs_io(void)
354 {
355 writel(0, &sdr_scc_mgr->dqs_io_ena);
356 }
357
358 /* load up dq config settings */
359 static void scc_mgr_load_dq(uint32_t dq_in_group)
360 {
361 writel(dq_in_group, &sdr_scc_mgr->dq_ena);
362 }
363
364 /* load up dm config settings */
365 static void scc_mgr_load_dm(uint32_t dm)
366 {
367 writel(dm, &sdr_scc_mgr->dm_ena);
368 }
369
370 /**
371 * scc_mgr_set_all_ranks() - Set SCC Manager register for all ranks
372 * @off: Base offset in SCC Manager space
373 * @grp: Read/Write group
374 * @val: Value to be set
375 * @update: If non-zero, trigger SCC Manager update for all ranks
376 *
377 * This function sets the SCC Manager (Scan Chain Control Manager) register
378 * and optionally triggers the SCC update for all ranks.
379 */
380 static void scc_mgr_set_all_ranks(const u32 off, const u32 grp, const u32 val,
381 const int update)
382 {
383 u32 r;
384
385 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
386 r += NUM_RANKS_PER_SHADOW_REG) {
387 scc_mgr_set(off, grp, val);
388
389 if (update || (r == 0)) {
390 writel(grp, &sdr_scc_mgr->dqs_ena);
391 writel(0, &sdr_scc_mgr->update);
392 }
393 }
394 }
395
396 static void scc_mgr_set_dqs_en_phase_all_ranks(u32 read_group, u32 phase)
397 {
398 /*
399 * USER although the h/w doesn't support different phases per
400 * shadow register, for simplicity our scc manager modeling
401 * keeps different phase settings per shadow reg, and it's
402 * important for us to keep them in sync to match h/w.
403 * for efficiency, the scan chain update should occur only
404 * once to sr0.
405 */
406 scc_mgr_set_all_ranks(SCC_MGR_DQS_EN_PHASE_OFFSET,
407 read_group, phase, 0);
408 }
409
410 static void scc_mgr_set_dqdqs_output_phase_all_ranks(uint32_t write_group,
411 uint32_t phase)
412 {
413 /*
414 * USER although the h/w doesn't support different phases per
415 * shadow register, for simplicity our scc manager modeling
416 * keeps different phase settings per shadow reg, and it's
417 * important for us to keep them in sync to match h/w.
418 * for efficiency, the scan chain update should occur only
419 * once to sr0.
420 */
421 scc_mgr_set_all_ranks(SCC_MGR_DQDQS_OUT_PHASE_OFFSET,
422 write_group, phase, 0);
423 }
424
425 static void scc_mgr_set_dqs_en_delay_all_ranks(uint32_t read_group,
426 uint32_t delay)
427 {
428 /*
429 * In shadow register mode, the T11 settings are stored in
430 * registers in the core, which are updated by the DQS_ENA
431 * signals. Not issuing the SCC_MGR_UPD command allows us to
432 * save lots of rank switching overhead, by calling
433 * select_shadow_regs_for_update with update_scan_chains
434 * set to 0.
435 */
436 scc_mgr_set_all_ranks(SCC_MGR_DQS_EN_DELAY_OFFSET,
437 read_group, delay, 1);
438 writel(0, &sdr_scc_mgr->update);
439 }
440
441 /**
442 * scc_mgr_set_oct_out1_delay() - Set OCT output delay
443 * @write_group: Write group
444 * @delay: Delay value
445 *
446 * This function sets the OCT output delay in SCC manager.
447 */
448 static void scc_mgr_set_oct_out1_delay(const u32 write_group, const u32 delay)
449 {
450 const int ratio = RW_MGR_MEM_IF_READ_DQS_WIDTH /
451 RW_MGR_MEM_IF_WRITE_DQS_WIDTH;
452 const int base = write_group * ratio;
453 int i;
454 /*
455 * Load the setting in the SCC manager
456 * Although OCT affects only write data, the OCT delay is controlled
457 * by the DQS logic block which is instantiated once per read group.
458 * For protocols where a write group consists of multiple read groups,
459 * the setting must be set multiple times.
460 */
461 for (i = 0; i < ratio; i++)
462 scc_mgr_set(SCC_MGR_OCT_OUT1_DELAY_OFFSET, base + i, delay);
463 }
464
465 /**
466 * scc_mgr_set_hhp_extras() - Set HHP extras.
467 *
468 * Load the fixed setting in the SCC manager HHP extras.
469 */
470 static void scc_mgr_set_hhp_extras(void)
471 {
472 /*
473 * Load the fixed setting in the SCC manager
474 * bits: 0:0 = 1'b1 - DQS bypass
475 * bits: 1:1 = 1'b1 - DQ bypass
476 * bits: 4:2 = 3'b001 - rfifo_mode
477 * bits: 6:5 = 2'b01 - rfifo clock_select
478 * bits: 7:7 = 1'b0 - separate gating from ungating setting
479 * bits: 8:8 = 1'b0 - separate OE from Output delay setting
480 */
481 const u32 value = (0 << 8) | (0 << 7) | (1 << 5) |
482 (1 << 2) | (1 << 1) | (1 << 0);
483 const u32 addr = SDR_PHYGRP_SCCGRP_ADDRESS |
484 SCC_MGR_HHP_GLOBALS_OFFSET |
485 SCC_MGR_HHP_EXTRAS_OFFSET;
486
487 debug_cond(DLEVEL == 1, "%s:%d Setting HHP Extras\n",
488 __func__, __LINE__);
489 writel(value, addr);
490 debug_cond(DLEVEL == 1, "%s:%d Done Setting HHP Extras\n",
491 __func__, __LINE__);
492 }
493
494 /**
495 * scc_mgr_zero_all() - Zero all DQS config
496 *
497 * Zero all DQS config.
498 */
499 static void scc_mgr_zero_all(void)
500 {
501 int i, r;
502
503 /*
504 * USER Zero all DQS config settings, across all groups and all
505 * shadow registers
506 */
507 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
508 r += NUM_RANKS_PER_SHADOW_REG) {
509 for (i = 0; i < RW_MGR_MEM_IF_READ_DQS_WIDTH; i++) {
510 /*
511 * The phases actually don't exist on a per-rank basis,
512 * but there's no harm updating them several times, so
513 * let's keep the code simple.
514 */
515 scc_mgr_set_dqs_bus_in_delay(i, IO_DQS_IN_RESERVE);
516 scc_mgr_set_dqs_en_phase(i, 0);
517 scc_mgr_set_dqs_en_delay(i, 0);
518 }
519
520 for (i = 0; i < RW_MGR_MEM_IF_WRITE_DQS_WIDTH; i++) {
521 scc_mgr_set_dqdqs_output_phase(i, 0);
522 /* Arria V/Cyclone V don't have out2. */
523 scc_mgr_set_oct_out1_delay(i, IO_DQS_OUT_RESERVE);
524 }
525 }
526
527 /* Multicast to all DQS group enables. */
528 writel(0xff, &sdr_scc_mgr->dqs_ena);
529 writel(0, &sdr_scc_mgr->update);
530 }
531
532 /**
533 * scc_set_bypass_mode() - Set bypass mode and trigger SCC update
534 * @write_group: Write group
535 *
536 * Set bypass mode and trigger SCC update.
537 */
538 static void scc_set_bypass_mode(const u32 write_group)
539 {
540 /* Multicast to all DQ enables. */
541 writel(0xff, &sdr_scc_mgr->dq_ena);
542 writel(0xff, &sdr_scc_mgr->dm_ena);
543
544 /* Update current DQS IO enable. */
545 writel(0, &sdr_scc_mgr->dqs_io_ena);
546
547 /* Update the DQS logic. */
548 writel(write_group, &sdr_scc_mgr->dqs_ena);
549
550 /* Hit update. */
551 writel(0, &sdr_scc_mgr->update);
552 }
553
554 /**
555 * scc_mgr_load_dqs_for_write_group() - Load DQS settings for Write Group
556 * @write_group: Write group
557 *
558 * Load DQS settings for Write Group, do not trigger SCC update.
559 */
560 static void scc_mgr_load_dqs_for_write_group(const u32 write_group)
561 {
562 const int ratio = RW_MGR_MEM_IF_READ_DQS_WIDTH /
563 RW_MGR_MEM_IF_WRITE_DQS_WIDTH;
564 const int base = write_group * ratio;
565 int i;
566 /*
567 * Load the setting in the SCC manager
568 * Although OCT affects only write data, the OCT delay is controlled
569 * by the DQS logic block which is instantiated once per read group.
570 * For protocols where a write group consists of multiple read groups,
571 * the setting must be set multiple times.
572 */
573 for (i = 0; i < ratio; i++)
574 writel(base + i, &sdr_scc_mgr->dqs_ena);
575 }
576
577 /**
578 * scc_mgr_zero_group() - Zero all configs for a group
579 *
580 * Zero DQ, DM, DQS and OCT configs for a group.
581 */
582 static void scc_mgr_zero_group(const u32 write_group, const int out_only)
583 {
584 int i, r;
585
586 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
587 r += NUM_RANKS_PER_SHADOW_REG) {
588 /* Zero all DQ config settings. */
589 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
590 scc_mgr_set_dq_out1_delay(i, 0);
591 if (!out_only)
592 scc_mgr_set_dq_in_delay(i, 0);
593 }
594
595 /* Multicast to all DQ enables. */
596 writel(0xff, &sdr_scc_mgr->dq_ena);
597
598 /* Zero all DM config settings. */
599 for (i = 0; i < RW_MGR_NUM_DM_PER_WRITE_GROUP; i++)
600 scc_mgr_set_dm_out1_delay(i, 0);
601
602 /* Multicast to all DM enables. */
603 writel(0xff, &sdr_scc_mgr->dm_ena);
604
605 /* Zero all DQS IO settings. */
606 if (!out_only)
607 scc_mgr_set_dqs_io_in_delay(0);
608
609 /* Arria V/Cyclone V don't have out2. */
610 scc_mgr_set_dqs_out1_delay(IO_DQS_OUT_RESERVE);
611 scc_mgr_set_oct_out1_delay(write_group, IO_DQS_OUT_RESERVE);
612 scc_mgr_load_dqs_for_write_group(write_group);
613
614 /* Multicast to all DQS IO enables (only 1 in total). */
615 writel(0, &sdr_scc_mgr->dqs_io_ena);
616
617 /* Hit update to zero everything. */
618 writel(0, &sdr_scc_mgr->update);
619 }
620 }
621
622 /*
623 * apply and load a particular input delay for the DQ pins in a group
624 * group_bgn is the index of the first dq pin (in the write group)
625 */
626 static void scc_mgr_apply_group_dq_in_delay(uint32_t group_bgn, uint32_t delay)
627 {
628 uint32_t i, p;
629
630 for (i = 0, p = group_bgn; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++, p++) {
631 scc_mgr_set_dq_in_delay(p, delay);
632 scc_mgr_load_dq(p);
633 }
634 }
635
636 /**
637 * scc_mgr_apply_group_dq_out1_delay() - Apply and load an output delay for the DQ pins in a group
638 * @delay: Delay value
639 *
640 * Apply and load a particular output delay for the DQ pins in a group.
641 */
642 static void scc_mgr_apply_group_dq_out1_delay(const u32 delay)
643 {
644 int i;
645
646 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
647 scc_mgr_set_dq_out1_delay(i, delay);
648 scc_mgr_load_dq(i);
649 }
650 }
651
652 /* apply and load a particular output delay for the DM pins in a group */
653 static void scc_mgr_apply_group_dm_out1_delay(uint32_t delay1)
654 {
655 uint32_t i;
656
657 for (i = 0; i < RW_MGR_NUM_DM_PER_WRITE_GROUP; i++) {
658 scc_mgr_set_dm_out1_delay(i, delay1);
659 scc_mgr_load_dm(i);
660 }
661 }
662
663
664 /* apply and load delay on both DQS and OCT out1 */
665 static void scc_mgr_apply_group_dqs_io_and_oct_out1(uint32_t write_group,
666 uint32_t delay)
667 {
668 scc_mgr_set_dqs_out1_delay(delay);
669 scc_mgr_load_dqs_io();
670
671 scc_mgr_set_oct_out1_delay(write_group, delay);
672 scc_mgr_load_dqs_for_write_group(write_group);
673 }
674
675 /**
676 * scc_mgr_apply_group_all_out_delay_add() - Apply a delay to the entire output side: DQ, DM, DQS, OCT
677 * @write_group: Write group
678 * @delay: Delay value
679 *
680 * Apply a delay to the entire output side: DQ, DM, DQS, OCT.
681 */
682 static void scc_mgr_apply_group_all_out_delay_add(const u32 write_group,
683 const u32 delay)
684 {
685 u32 i, new_delay;
686
687 /* DQ shift */
688 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++)
689 scc_mgr_load_dq(i);
690
691 /* DM shift */
692 for (i = 0; i < RW_MGR_NUM_DM_PER_WRITE_GROUP; i++)
693 scc_mgr_load_dm(i);
694
695 /* DQS shift */
696 new_delay = READ_SCC_DQS_IO_OUT2_DELAY + delay;
697 if (new_delay > IO_IO_OUT2_DELAY_MAX) {
698 debug_cond(DLEVEL == 1,
699 "%s:%d (%u, %u) DQS: %u > %d; adding %u to OUT1\n",
700 __func__, __LINE__, write_group, delay, new_delay,
701 IO_IO_OUT2_DELAY_MAX,
702 new_delay - IO_IO_OUT2_DELAY_MAX);
703 new_delay -= IO_IO_OUT2_DELAY_MAX;
704 scc_mgr_set_dqs_out1_delay(new_delay);
705 }
706
707 scc_mgr_load_dqs_io();
708
709 /* OCT shift */
710 new_delay = READ_SCC_OCT_OUT2_DELAY + delay;
711 if (new_delay > IO_IO_OUT2_DELAY_MAX) {
712 debug_cond(DLEVEL == 1,
713 "%s:%d (%u, %u) DQS: %u > %d; adding %u to OUT1\n",
714 __func__, __LINE__, write_group, delay,
715 new_delay, IO_IO_OUT2_DELAY_MAX,
716 new_delay - IO_IO_OUT2_DELAY_MAX);
717 new_delay -= IO_IO_OUT2_DELAY_MAX;
718 scc_mgr_set_oct_out1_delay(write_group, new_delay);
719 }
720
721 scc_mgr_load_dqs_for_write_group(write_group);
722 }
723
724 /**
725 * scc_mgr_apply_group_all_out_delay_add() - Apply a delay to the entire output side to all ranks
726 * @write_group: Write group
727 * @delay: Delay value
728 *
729 * Apply a delay to the entire output side (DQ, DM, DQS, OCT) to all ranks.
730 */
731 static void
732 scc_mgr_apply_group_all_out_delay_add_all_ranks(const u32 write_group,
733 const u32 delay)
734 {
735 int r;
736
737 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
738 r += NUM_RANKS_PER_SHADOW_REG) {
739 scc_mgr_apply_group_all_out_delay_add(write_group, delay);
740 writel(0, &sdr_scc_mgr->update);
741 }
742 }
743
744 /**
745 * set_jump_as_return() - Return instruction optimization
746 *
747 * Optimization used to recover some slots in ddr3 inst_rom could be
748 * applied to other protocols if we wanted to
749 */
750 static void set_jump_as_return(void)
751 {
752 /*
753 * To save space, we replace return with jump to special shared
754 * RETURN instruction so we set the counter to large value so that
755 * we always jump.
756 */
757 writel(0xff, &sdr_rw_load_mgr_regs->load_cntr0);
758 writel(RW_MGR_RETURN, &sdr_rw_load_jump_mgr_regs->load_jump_add0);
759 }
760
761 /*
762 * should always use constants as argument to ensure all computations are
763 * performed at compile time
764 */
765 static void delay_for_n_mem_clocks(const uint32_t clocks)
766 {
767 uint32_t afi_clocks;
768 uint8_t inner = 0;
769 uint8_t outer = 0;
770 uint16_t c_loop = 0;
771
772 debug("%s:%d: clocks=%u ... start\n", __func__, __LINE__, clocks);
773
774
775 afi_clocks = (clocks + AFI_RATE_RATIO-1) / AFI_RATE_RATIO;
776 /* scale (rounding up) to get afi clocks */
777
778 /*
779 * Note, we don't bother accounting for being off a little bit
780 * because of a few extra instructions in outer loops
781 * Note, the loops have a test at the end, and do the test before
782 * the decrement, and so always perform the loop
783 * 1 time more than the counter value
784 */
785 if (afi_clocks == 0) {
786 ;
787 } else if (afi_clocks <= 0x100) {
788 inner = afi_clocks-1;
789 outer = 0;
790 c_loop = 0;
791 } else if (afi_clocks <= 0x10000) {
792 inner = 0xff;
793 outer = (afi_clocks-1) >> 8;
794 c_loop = 0;
795 } else {
796 inner = 0xff;
797 outer = 0xff;
798 c_loop = (afi_clocks-1) >> 16;
799 }
800
801 /*
802 * rom instructions are structured as follows:
803 *
804 * IDLE_LOOP2: jnz cntr0, TARGET_A
805 * IDLE_LOOP1: jnz cntr1, TARGET_B
806 * return
807 *
808 * so, when doing nested loops, TARGET_A is set to IDLE_LOOP2, and
809 * TARGET_B is set to IDLE_LOOP2 as well
810 *
811 * if we have no outer loop, though, then we can use IDLE_LOOP1 only,
812 * and set TARGET_B to IDLE_LOOP1 and we skip IDLE_LOOP2 entirely
813 *
814 * a little confusing, but it helps save precious space in the inst_rom
815 * and sequencer rom and keeps the delays more accurate and reduces
816 * overhead
817 */
818 if (afi_clocks <= 0x100) {
819 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(inner),
820 &sdr_rw_load_mgr_regs->load_cntr1);
821
822 writel(RW_MGR_IDLE_LOOP1,
823 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
824
825 writel(RW_MGR_IDLE_LOOP1, SDR_PHYGRP_RWMGRGRP_ADDRESS |
826 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
827 } else {
828 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(inner),
829 &sdr_rw_load_mgr_regs->load_cntr0);
830
831 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(outer),
832 &sdr_rw_load_mgr_regs->load_cntr1);
833
834 writel(RW_MGR_IDLE_LOOP2,
835 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
836
837 writel(RW_MGR_IDLE_LOOP2,
838 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
839
840 /* hack to get around compiler not being smart enough */
841 if (afi_clocks <= 0x10000) {
842 /* only need to run once */
843 writel(RW_MGR_IDLE_LOOP2, SDR_PHYGRP_RWMGRGRP_ADDRESS |
844 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
845 } else {
846 do {
847 writel(RW_MGR_IDLE_LOOP2,
848 SDR_PHYGRP_RWMGRGRP_ADDRESS |
849 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
850 } while (c_loop-- != 0);
851 }
852 }
853 debug("%s:%d clocks=%u ... end\n", __func__, __LINE__, clocks);
854 }
855
856 /**
857 * rw_mgr_mem_init_load_regs() - Load instruction registers
858 * @cntr0: Counter 0 value
859 * @cntr1: Counter 1 value
860 * @cntr2: Counter 2 value
861 * @jump: Jump instruction value
862 *
863 * Load instruction registers.
864 */
865 static void rw_mgr_mem_init_load_regs(u32 cntr0, u32 cntr1, u32 cntr2, u32 jump)
866 {
867 uint32_t grpaddr = SDR_PHYGRP_RWMGRGRP_ADDRESS |
868 RW_MGR_RUN_SINGLE_GROUP_OFFSET;
869
870 /* Load counters */
871 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(cntr0),
872 &sdr_rw_load_mgr_regs->load_cntr0);
873 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(cntr1),
874 &sdr_rw_load_mgr_regs->load_cntr1);
875 writel(SKIP_DELAY_LOOP_VALUE_OR_ZERO(cntr2),
876 &sdr_rw_load_mgr_regs->load_cntr2);
877
878 /* Load jump address */
879 writel(jump, &sdr_rw_load_jump_mgr_regs->load_jump_add0);
880 writel(jump, &sdr_rw_load_jump_mgr_regs->load_jump_add1);
881 writel(jump, &sdr_rw_load_jump_mgr_regs->load_jump_add2);
882
883 /* Execute count instruction */
884 writel(jump, grpaddr);
885 }
886
887 /**
888 * rw_mgr_mem_load_user() - Load user calibration values
889 * @fin1: Final instruction 1
890 * @fin2: Final instruction 2
891 * @precharge: If 1, precharge the banks at the end
892 *
893 * Load user calibration values and optionally precharge the banks.
894 */
895 static void rw_mgr_mem_load_user(const u32 fin1, const u32 fin2,
896 const int precharge)
897 {
898 u32 grpaddr = SDR_PHYGRP_RWMGRGRP_ADDRESS |
899 RW_MGR_RUN_SINGLE_GROUP_OFFSET;
900 u32 r;
901
902 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS; r++) {
903 if (param->skip_ranks[r]) {
904 /* request to skip the rank */
905 continue;
906 }
907
908 /* set rank */
909 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_OFF);
910
911 /* precharge all banks ... */
912 if (precharge)
913 writel(RW_MGR_PRECHARGE_ALL, grpaddr);
914
915 /*
916 * USER Use Mirror-ed commands for odd ranks if address
917 * mirrorring is on
918 */
919 if ((RW_MGR_MEM_ADDRESS_MIRRORING >> r) & 0x1) {
920 set_jump_as_return();
921 writel(RW_MGR_MRS2_MIRR, grpaddr);
922 delay_for_n_mem_clocks(4);
923 set_jump_as_return();
924 writel(RW_MGR_MRS3_MIRR, grpaddr);
925 delay_for_n_mem_clocks(4);
926 set_jump_as_return();
927 writel(RW_MGR_MRS1_MIRR, grpaddr);
928 delay_for_n_mem_clocks(4);
929 set_jump_as_return();
930 writel(fin1, grpaddr);
931 } else {
932 set_jump_as_return();
933 writel(RW_MGR_MRS2, grpaddr);
934 delay_for_n_mem_clocks(4);
935 set_jump_as_return();
936 writel(RW_MGR_MRS3, grpaddr);
937 delay_for_n_mem_clocks(4);
938 set_jump_as_return();
939 writel(RW_MGR_MRS1, grpaddr);
940 set_jump_as_return();
941 writel(fin2, grpaddr);
942 }
943
944 if (precharge)
945 continue;
946
947 set_jump_as_return();
948 writel(RW_MGR_ZQCL, grpaddr);
949
950 /* tZQinit = tDLLK = 512 ck cycles */
951 delay_for_n_mem_clocks(512);
952 }
953 }
954
955 /**
956 * rw_mgr_mem_initialize() - Initialize RW Manager
957 *
958 * Initialize RW Manager.
959 */
960 static void rw_mgr_mem_initialize(void)
961 {
962 debug("%s:%d\n", __func__, __LINE__);
963
964 /* The reset / cke part of initialization is broadcasted to all ranks */
965 writel(RW_MGR_RANK_ALL, SDR_PHYGRP_RWMGRGRP_ADDRESS |
966 RW_MGR_SET_CS_AND_ODT_MASK_OFFSET);
967
968 /*
969 * Here's how you load register for a loop
970 * Counters are located @ 0x800
971 * Jump address are located @ 0xC00
972 * For both, registers 0 to 3 are selected using bits 3 and 2, like
973 * in 0x800, 0x804, 0x808, 0x80C and 0xC00, 0xC04, 0xC08, 0xC0C
974 * I know this ain't pretty, but Avalon bus throws away the 2 least
975 * significant bits
976 */
977
978 /* Start with memory RESET activated */
979
980 /* tINIT = 200us */
981
982 /*
983 * 200us @ 266MHz (3.75 ns) ~ 54000 clock cycles
984 * If a and b are the number of iteration in 2 nested loops
985 * it takes the following number of cycles to complete the operation:
986 * number_of_cycles = ((2 + n) * a + 2) * b
987 * where n is the number of instruction in the inner loop
988 * One possible solution is n = 0 , a = 256 , b = 106 => a = FF,
989 * b = 6A
990 */
991 rw_mgr_mem_init_load_regs(SEQ_TINIT_CNTR0_VAL, SEQ_TINIT_CNTR1_VAL,
992 SEQ_TINIT_CNTR2_VAL,
993 RW_MGR_INIT_RESET_0_CKE_0);
994
995 /* Indicate that memory is stable. */
996 writel(1, &phy_mgr_cfg->reset_mem_stbl);
997
998 /*
999 * transition the RESET to high
1000 * Wait for 500us
1001 */
1002
1003 /*
1004 * 500us @ 266MHz (3.75 ns) ~ 134000 clock cycles
1005 * If a and b are the number of iteration in 2 nested loops
1006 * it takes the following number of cycles to complete the operation
1007 * number_of_cycles = ((2 + n) * a + 2) * b
1008 * where n is the number of instruction in the inner loop
1009 * One possible solution is n = 2 , a = 131 , b = 256 => a = 83,
1010 * b = FF
1011 */
1012 rw_mgr_mem_init_load_regs(SEQ_TRESET_CNTR0_VAL, SEQ_TRESET_CNTR1_VAL,
1013 SEQ_TRESET_CNTR2_VAL,
1014 RW_MGR_INIT_RESET_1_CKE_0);
1015
1016 /* Bring up clock enable. */
1017
1018 /* tXRP < 250 ck cycles */
1019 delay_for_n_mem_clocks(250);
1020
1021 rw_mgr_mem_load_user(RW_MGR_MRS0_DLL_RESET_MIRR, RW_MGR_MRS0_DLL_RESET,
1022 0);
1023 }
1024
1025 /*
1026 * At the end of calibration we have to program the user settings in, and
1027 * USER hand off the memory to the user.
1028 */
1029 static void rw_mgr_mem_handoff(void)
1030 {
1031 rw_mgr_mem_load_user(RW_MGR_MRS0_USER_MIRR, RW_MGR_MRS0_USER, 1);
1032 /*
1033 * USER need to wait tMOD (12CK or 15ns) time before issuing
1034 * other commands, but we will have plenty of NIOS cycles before
1035 * actual handoff so its okay.
1036 */
1037 }
1038
1039 /*
1040 * performs a guaranteed read on the patterns we are going to use during a
1041 * read test to ensure memory works
1042 */
1043 static uint32_t rw_mgr_mem_calibrate_read_test_patterns(uint32_t rank_bgn,
1044 uint32_t group, uint32_t num_tries, uint32_t *bit_chk,
1045 uint32_t all_ranks)
1046 {
1047 uint32_t r, vg;
1048 uint32_t correct_mask_vg;
1049 uint32_t tmp_bit_chk;
1050 uint32_t rank_end = all_ranks ? RW_MGR_MEM_NUMBER_OF_RANKS :
1051 (rank_bgn + NUM_RANKS_PER_SHADOW_REG);
1052 uint32_t addr;
1053 uint32_t base_rw_mgr;
1054
1055 *bit_chk = param->read_correct_mask;
1056 correct_mask_vg = param->read_correct_mask_vg;
1057
1058 for (r = rank_bgn; r < rank_end; r++) {
1059 if (param->skip_ranks[r])
1060 /* request to skip the rank */
1061 continue;
1062
1063 /* set rank */
1064 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_READ_WRITE);
1065
1066 /* Load up a constant bursts of read commands */
1067 writel(0x20, &sdr_rw_load_mgr_regs->load_cntr0);
1068 writel(RW_MGR_GUARANTEED_READ,
1069 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
1070
1071 writel(0x20, &sdr_rw_load_mgr_regs->load_cntr1);
1072 writel(RW_MGR_GUARANTEED_READ_CONT,
1073 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
1074
1075 tmp_bit_chk = 0;
1076 for (vg = RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS-1; ; vg--) {
1077 /* reset the fifos to get pointers to known state */
1078
1079 writel(0, &phy_mgr_cmd->fifo_reset);
1080 writel(0, SDR_PHYGRP_RWMGRGRP_ADDRESS |
1081 RW_MGR_RESET_READ_DATAPATH_OFFSET);
1082
1083 tmp_bit_chk = tmp_bit_chk << (RW_MGR_MEM_DQ_PER_READ_DQS
1084 / RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS);
1085
1086 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
1087 writel(RW_MGR_GUARANTEED_READ, addr +
1088 ((group * RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS +
1089 vg) << 2));
1090
1091 base_rw_mgr = readl(SDR_PHYGRP_RWMGRGRP_ADDRESS);
1092 tmp_bit_chk = tmp_bit_chk | (correct_mask_vg & (~base_rw_mgr));
1093
1094 if (vg == 0)
1095 break;
1096 }
1097 *bit_chk &= tmp_bit_chk;
1098 }
1099
1100 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
1101 writel(RW_MGR_CLEAR_DQS_ENABLE, addr + (group << 2));
1102
1103 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
1104 debug_cond(DLEVEL == 1, "%s:%d test_load_patterns(%u,ALL) => (%u == %u) =>\
1105 %lu\n", __func__, __LINE__, group, *bit_chk, param->read_correct_mask,
1106 (long unsigned int)(*bit_chk == param->read_correct_mask));
1107 return *bit_chk == param->read_correct_mask;
1108 }
1109
1110 /**
1111 * rw_mgr_mem_calibrate_read_load_patterns() - Load up the patterns for read test
1112 * @rank_bgn: Rank number
1113 * @all_ranks: Test all ranks
1114 *
1115 * Load up the patterns we are going to use during a read test.
1116 */
1117 static void rw_mgr_mem_calibrate_read_load_patterns(const u32 rank_bgn,
1118 const int all_ranks)
1119 {
1120 const u32 rank_end = all_ranks ?
1121 RW_MGR_MEM_NUMBER_OF_RANKS :
1122 (rank_bgn + NUM_RANKS_PER_SHADOW_REG);
1123 u32 r;
1124
1125 debug("%s:%d\n", __func__, __LINE__);
1126
1127 for (r = rank_bgn; r < rank_end; r++) {
1128 if (param->skip_ranks[r])
1129 /* request to skip the rank */
1130 continue;
1131
1132 /* set rank */
1133 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_READ_WRITE);
1134
1135 /* Load up a constant bursts */
1136 writel(0x20, &sdr_rw_load_mgr_regs->load_cntr0);
1137
1138 writel(RW_MGR_GUARANTEED_WRITE_WAIT0,
1139 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
1140
1141 writel(0x20, &sdr_rw_load_mgr_regs->load_cntr1);
1142
1143 writel(RW_MGR_GUARANTEED_WRITE_WAIT1,
1144 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
1145
1146 writel(0x04, &sdr_rw_load_mgr_regs->load_cntr2);
1147
1148 writel(RW_MGR_GUARANTEED_WRITE_WAIT2,
1149 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
1150
1151 writel(0x04, &sdr_rw_load_mgr_regs->load_cntr3);
1152
1153 writel(RW_MGR_GUARANTEED_WRITE_WAIT3,
1154 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
1155
1156 writel(RW_MGR_GUARANTEED_WRITE, SDR_PHYGRP_RWMGRGRP_ADDRESS |
1157 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
1158 }
1159
1160 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
1161 }
1162
1163 /*
1164 * try a read and see if it returns correct data back. has dummy reads
1165 * inserted into the mix used to align dqs enable. has more thorough checks
1166 * than the regular read test.
1167 */
1168 static uint32_t rw_mgr_mem_calibrate_read_test(uint32_t rank_bgn, uint32_t group,
1169 uint32_t num_tries, uint32_t all_correct, uint32_t *bit_chk,
1170 uint32_t all_groups, uint32_t all_ranks)
1171 {
1172 uint32_t r, vg;
1173 uint32_t correct_mask_vg;
1174 uint32_t tmp_bit_chk;
1175 uint32_t rank_end = all_ranks ? RW_MGR_MEM_NUMBER_OF_RANKS :
1176 (rank_bgn + NUM_RANKS_PER_SHADOW_REG);
1177 uint32_t addr;
1178 uint32_t base_rw_mgr;
1179
1180 *bit_chk = param->read_correct_mask;
1181 correct_mask_vg = param->read_correct_mask_vg;
1182
1183 uint32_t quick_read_mode = (((STATIC_CALIB_STEPS) &
1184 CALIB_SKIP_DELAY_SWEEPS) && ENABLE_SUPER_QUICK_CALIBRATION);
1185
1186 for (r = rank_bgn; r < rank_end; r++) {
1187 if (param->skip_ranks[r])
1188 /* request to skip the rank */
1189 continue;
1190
1191 /* set rank */
1192 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_READ_WRITE);
1193
1194 writel(0x10, &sdr_rw_load_mgr_regs->load_cntr1);
1195
1196 writel(RW_MGR_READ_B2B_WAIT1,
1197 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
1198
1199 writel(0x10, &sdr_rw_load_mgr_regs->load_cntr2);
1200 writel(RW_MGR_READ_B2B_WAIT2,
1201 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
1202
1203 if (quick_read_mode)
1204 writel(0x1, &sdr_rw_load_mgr_regs->load_cntr0);
1205 /* need at least two (1+1) reads to capture failures */
1206 else if (all_groups)
1207 writel(0x06, &sdr_rw_load_mgr_regs->load_cntr0);
1208 else
1209 writel(0x32, &sdr_rw_load_mgr_regs->load_cntr0);
1210
1211 writel(RW_MGR_READ_B2B,
1212 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
1213 if (all_groups)
1214 writel(RW_MGR_MEM_IF_READ_DQS_WIDTH *
1215 RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS - 1,
1216 &sdr_rw_load_mgr_regs->load_cntr3);
1217 else
1218 writel(0x0, &sdr_rw_load_mgr_regs->load_cntr3);
1219
1220 writel(RW_MGR_READ_B2B,
1221 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
1222
1223 tmp_bit_chk = 0;
1224 for (vg = RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS-1; ; vg--) {
1225 /* reset the fifos to get pointers to known state */
1226 writel(0, &phy_mgr_cmd->fifo_reset);
1227 writel(0, SDR_PHYGRP_RWMGRGRP_ADDRESS |
1228 RW_MGR_RESET_READ_DATAPATH_OFFSET);
1229
1230 tmp_bit_chk = tmp_bit_chk << (RW_MGR_MEM_DQ_PER_READ_DQS
1231 / RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS);
1232
1233 if (all_groups)
1234 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_ALL_GROUPS_OFFSET;
1235 else
1236 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
1237
1238 writel(RW_MGR_READ_B2B, addr +
1239 ((group * RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS +
1240 vg) << 2));
1241
1242 base_rw_mgr = readl(SDR_PHYGRP_RWMGRGRP_ADDRESS);
1243 tmp_bit_chk = tmp_bit_chk | (correct_mask_vg & ~(base_rw_mgr));
1244
1245 if (vg == 0)
1246 break;
1247 }
1248 *bit_chk &= tmp_bit_chk;
1249 }
1250
1251 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
1252 writel(RW_MGR_CLEAR_DQS_ENABLE, addr + (group << 2));
1253
1254 if (all_correct) {
1255 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
1256 debug_cond(DLEVEL == 2, "%s:%d read_test(%u,ALL,%u) =>\
1257 (%u == %u) => %lu", __func__, __LINE__, group,
1258 all_groups, *bit_chk, param->read_correct_mask,
1259 (long unsigned int)(*bit_chk ==
1260 param->read_correct_mask));
1261 return *bit_chk == param->read_correct_mask;
1262 } else {
1263 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
1264 debug_cond(DLEVEL == 2, "%s:%d read_test(%u,ONE,%u) =>\
1265 (%u != %lu) => %lu\n", __func__, __LINE__,
1266 group, all_groups, *bit_chk, (long unsigned int)0,
1267 (long unsigned int)(*bit_chk != 0x00));
1268 return *bit_chk != 0x00;
1269 }
1270 }
1271
1272 static uint32_t rw_mgr_mem_calibrate_read_test_all_ranks(uint32_t group,
1273 uint32_t num_tries, uint32_t all_correct, uint32_t *bit_chk,
1274 uint32_t all_groups)
1275 {
1276 return rw_mgr_mem_calibrate_read_test(0, group, num_tries, all_correct,
1277 bit_chk, all_groups, 1);
1278 }
1279
1280 static void rw_mgr_incr_vfifo(uint32_t grp, uint32_t *v)
1281 {
1282 writel(grp, &phy_mgr_cmd->inc_vfifo_hard_phy);
1283 (*v)++;
1284 }
1285
1286 static void rw_mgr_decr_vfifo(uint32_t grp, uint32_t *v)
1287 {
1288 uint32_t i;
1289
1290 for (i = 0; i < VFIFO_SIZE-1; i++)
1291 rw_mgr_incr_vfifo(grp, v);
1292 }
1293
1294 static int find_vfifo_read(uint32_t grp, uint32_t *bit_chk)
1295 {
1296 uint32_t v;
1297 uint32_t fail_cnt = 0;
1298 uint32_t test_status;
1299
1300 for (v = 0; v < VFIFO_SIZE; ) {
1301 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: vfifo %u\n",
1302 __func__, __LINE__, v);
1303 test_status = rw_mgr_mem_calibrate_read_test_all_ranks
1304 (grp, 1, PASS_ONE_BIT, bit_chk, 0);
1305 if (!test_status) {
1306 fail_cnt++;
1307
1308 if (fail_cnt == 2)
1309 break;
1310 }
1311
1312 /* fiddle with FIFO */
1313 rw_mgr_incr_vfifo(grp, &v);
1314 }
1315
1316 if (v >= VFIFO_SIZE) {
1317 /* no failing read found!! Something must have gone wrong */
1318 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: vfifo failed\n",
1319 __func__, __LINE__);
1320 return 0;
1321 } else {
1322 return v;
1323 }
1324 }
1325
1326 static int find_working_phase(uint32_t *grp, uint32_t *bit_chk,
1327 uint32_t dtaps_per_ptap, uint32_t *work_bgn,
1328 uint32_t *v, uint32_t *d, uint32_t *p,
1329 uint32_t *i, uint32_t *max_working_cnt)
1330 {
1331 uint32_t found_begin = 0;
1332 uint32_t tmp_delay = 0;
1333 uint32_t test_status;
1334
1335 for (*d = 0; *d <= dtaps_per_ptap; (*d)++, tmp_delay +=
1336 IO_DELAY_PER_DQS_EN_DCHAIN_TAP) {
1337 *work_bgn = tmp_delay;
1338 scc_mgr_set_dqs_en_delay_all_ranks(*grp, *d);
1339
1340 for (*i = 0; *i < VFIFO_SIZE; (*i)++) {
1341 for (*p = 0; *p <= IO_DQS_EN_PHASE_MAX; (*p)++, *work_bgn +=
1342 IO_DELAY_PER_OPA_TAP) {
1343 scc_mgr_set_dqs_en_phase_all_ranks(*grp, *p);
1344
1345 test_status =
1346 rw_mgr_mem_calibrate_read_test_all_ranks
1347 (*grp, 1, PASS_ONE_BIT, bit_chk, 0);
1348
1349 if (test_status) {
1350 *max_working_cnt = 1;
1351 found_begin = 1;
1352 break;
1353 }
1354 }
1355
1356 if (found_begin)
1357 break;
1358
1359 if (*p > IO_DQS_EN_PHASE_MAX)
1360 /* fiddle with FIFO */
1361 rw_mgr_incr_vfifo(*grp, v);
1362 }
1363
1364 if (found_begin)
1365 break;
1366 }
1367
1368 if (*i >= VFIFO_SIZE) {
1369 /* cannot find working solution */
1370 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: no vfifo/\
1371 ptap/dtap\n", __func__, __LINE__);
1372 return 0;
1373 } else {
1374 return 1;
1375 }
1376 }
1377
1378 static void sdr_backup_phase(uint32_t *grp, uint32_t *bit_chk,
1379 uint32_t *work_bgn, uint32_t *v, uint32_t *d,
1380 uint32_t *p, uint32_t *max_working_cnt)
1381 {
1382 uint32_t found_begin = 0;
1383 uint32_t tmp_delay;
1384
1385 /* Special case code for backing up a phase */
1386 if (*p == 0) {
1387 *p = IO_DQS_EN_PHASE_MAX;
1388 rw_mgr_decr_vfifo(*grp, v);
1389 } else {
1390 (*p)--;
1391 }
1392 tmp_delay = *work_bgn - IO_DELAY_PER_OPA_TAP;
1393 scc_mgr_set_dqs_en_phase_all_ranks(*grp, *p);
1394
1395 for (*d = 0; *d <= IO_DQS_EN_DELAY_MAX && tmp_delay < *work_bgn;
1396 (*d)++, tmp_delay += IO_DELAY_PER_DQS_EN_DCHAIN_TAP) {
1397 scc_mgr_set_dqs_en_delay_all_ranks(*grp, *d);
1398
1399 if (rw_mgr_mem_calibrate_read_test_all_ranks(*grp, 1,
1400 PASS_ONE_BIT,
1401 bit_chk, 0)) {
1402 found_begin = 1;
1403 *work_bgn = tmp_delay;
1404 break;
1405 }
1406 }
1407
1408 /* We have found a working dtap before the ptap found above */
1409 if (found_begin == 1)
1410 (*max_working_cnt)++;
1411
1412 /*
1413 * Restore VFIFO to old state before we decremented it
1414 * (if needed).
1415 */
1416 (*p)++;
1417 if (*p > IO_DQS_EN_PHASE_MAX) {
1418 *p = 0;
1419 rw_mgr_incr_vfifo(*grp, v);
1420 }
1421
1422 scc_mgr_set_dqs_en_delay_all_ranks(*grp, 0);
1423 }
1424
1425 static int sdr_nonworking_phase(uint32_t *grp, uint32_t *bit_chk,
1426 uint32_t *work_bgn, uint32_t *v, uint32_t *d,
1427 uint32_t *p, uint32_t *i, uint32_t *max_working_cnt,
1428 uint32_t *work_end)
1429 {
1430 uint32_t found_end = 0;
1431
1432 (*p)++;
1433 *work_end += IO_DELAY_PER_OPA_TAP;
1434 if (*p > IO_DQS_EN_PHASE_MAX) {
1435 /* fiddle with FIFO */
1436 *p = 0;
1437 rw_mgr_incr_vfifo(*grp, v);
1438 }
1439
1440 for (; *i < VFIFO_SIZE + 1; (*i)++) {
1441 for (; *p <= IO_DQS_EN_PHASE_MAX; (*p)++, *work_end
1442 += IO_DELAY_PER_OPA_TAP) {
1443 scc_mgr_set_dqs_en_phase_all_ranks(*grp, *p);
1444
1445 if (!rw_mgr_mem_calibrate_read_test_all_ranks
1446 (*grp, 1, PASS_ONE_BIT, bit_chk, 0)) {
1447 found_end = 1;
1448 break;
1449 } else {
1450 (*max_working_cnt)++;
1451 }
1452 }
1453
1454 if (found_end)
1455 break;
1456
1457 if (*p > IO_DQS_EN_PHASE_MAX) {
1458 /* fiddle with FIFO */
1459 rw_mgr_incr_vfifo(*grp, v);
1460 *p = 0;
1461 }
1462 }
1463
1464 if (*i >= VFIFO_SIZE + 1) {
1465 /* cannot see edge of failing read */
1466 debug_cond(DLEVEL == 2, "%s:%d sdr_nonworking_phase: end:\
1467 failed\n", __func__, __LINE__);
1468 return 0;
1469 } else {
1470 return 1;
1471 }
1472 }
1473
1474 static int sdr_find_window_centre(uint32_t *grp, uint32_t *bit_chk,
1475 uint32_t *work_bgn, uint32_t *v, uint32_t *d,
1476 uint32_t *p, uint32_t *work_mid,
1477 uint32_t *work_end)
1478 {
1479 int i;
1480 int tmp_delay = 0;
1481
1482 *work_mid = (*work_bgn + *work_end) / 2;
1483
1484 debug_cond(DLEVEL == 2, "work_bgn=%d work_end=%d work_mid=%d\n",
1485 *work_bgn, *work_end, *work_mid);
1486 /* Get the middle delay to be less than a VFIFO delay */
1487 for (*p = 0; *p <= IO_DQS_EN_PHASE_MAX;
1488 (*p)++, tmp_delay += IO_DELAY_PER_OPA_TAP)
1489 ;
1490 debug_cond(DLEVEL == 2, "vfifo ptap delay %d\n", tmp_delay);
1491 while (*work_mid > tmp_delay)
1492 *work_mid -= tmp_delay;
1493 debug_cond(DLEVEL == 2, "new work_mid %d\n", *work_mid);
1494
1495 tmp_delay = 0;
1496 for (*p = 0; *p <= IO_DQS_EN_PHASE_MAX && tmp_delay < *work_mid;
1497 (*p)++, tmp_delay += IO_DELAY_PER_OPA_TAP)
1498 ;
1499 tmp_delay -= IO_DELAY_PER_OPA_TAP;
1500 debug_cond(DLEVEL == 2, "new p %d, tmp_delay=%d\n", (*p) - 1, tmp_delay);
1501 for (*d = 0; *d <= IO_DQS_EN_DELAY_MAX && tmp_delay < *work_mid; (*d)++,
1502 tmp_delay += IO_DELAY_PER_DQS_EN_DCHAIN_TAP)
1503 ;
1504 debug_cond(DLEVEL == 2, "new d %d, tmp_delay=%d\n", *d, tmp_delay);
1505
1506 scc_mgr_set_dqs_en_phase_all_ranks(*grp, (*p) - 1);
1507 scc_mgr_set_dqs_en_delay_all_ranks(*grp, *d);
1508
1509 /*
1510 * push vfifo until we can successfully calibrate. We can do this
1511 * because the largest possible margin in 1 VFIFO cycle.
1512 */
1513 for (i = 0; i < VFIFO_SIZE; i++) {
1514 debug_cond(DLEVEL == 2, "find_dqs_en_phase: center: vfifo=%u\n",
1515 *v);
1516 if (rw_mgr_mem_calibrate_read_test_all_ranks(*grp, 1,
1517 PASS_ONE_BIT,
1518 bit_chk, 0)) {
1519 break;
1520 }
1521
1522 /* fiddle with FIFO */
1523 rw_mgr_incr_vfifo(*grp, v);
1524 }
1525
1526 if (i >= VFIFO_SIZE) {
1527 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: center: \
1528 failed\n", __func__, __LINE__);
1529 return 0;
1530 } else {
1531 return 1;
1532 }
1533 }
1534
1535 /* find a good dqs enable to use */
1536 static uint32_t rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase(uint32_t grp)
1537 {
1538 uint32_t v, d, p, i;
1539 uint32_t max_working_cnt;
1540 uint32_t bit_chk;
1541 uint32_t dtaps_per_ptap;
1542 uint32_t work_bgn, work_mid, work_end;
1543 uint32_t found_passing_read, found_failing_read, initial_failing_dtap;
1544
1545 debug("%s:%d %u\n", __func__, __LINE__, grp);
1546
1547 reg_file_set_sub_stage(CAL_SUBSTAGE_VFIFO_CENTER);
1548
1549 scc_mgr_set_dqs_en_delay_all_ranks(grp, 0);
1550 scc_mgr_set_dqs_en_phase_all_ranks(grp, 0);
1551
1552 /* ************************************************************** */
1553 /* * Step 0 : Determine number of delay taps for each phase tap * */
1554 dtaps_per_ptap = IO_DELAY_PER_OPA_TAP/IO_DELAY_PER_DQS_EN_DCHAIN_TAP;
1555
1556 /* ********************************************************* */
1557 /* * Step 1 : First push vfifo until we get a failing read * */
1558 v = find_vfifo_read(grp, &bit_chk);
1559
1560 max_working_cnt = 0;
1561
1562 /* ******************************************************** */
1563 /* * step 2: find first working phase, increment in ptaps * */
1564 work_bgn = 0;
1565 if (find_working_phase(&grp, &bit_chk, dtaps_per_ptap, &work_bgn, &v, &d,
1566 &p, &i, &max_working_cnt) == 0)
1567 return 0;
1568
1569 work_end = work_bgn;
1570
1571 /*
1572 * If d is 0 then the working window covers a phase tap and
1573 * we can follow the old procedure otherwise, we've found the beginning,
1574 * and we need to increment the dtaps until we find the end.
1575 */
1576 if (d == 0) {
1577 /* ********************************************************* */
1578 /* * step 3a: if we have room, back off by one and
1579 increment in dtaps * */
1580
1581 sdr_backup_phase(&grp, &bit_chk, &work_bgn, &v, &d, &p,
1582 &max_working_cnt);
1583
1584 /* ********************************************************* */
1585 /* * step 4a: go forward from working phase to non working
1586 phase, increment in ptaps * */
1587 if (sdr_nonworking_phase(&grp, &bit_chk, &work_bgn, &v, &d, &p,
1588 &i, &max_working_cnt, &work_end) == 0)
1589 return 0;
1590
1591 /* ********************************************************* */
1592 /* * step 5a: back off one from last, increment in dtaps * */
1593
1594 /* Special case code for backing up a phase */
1595 if (p == 0) {
1596 p = IO_DQS_EN_PHASE_MAX;
1597 rw_mgr_decr_vfifo(grp, &v);
1598 } else {
1599 p = p - 1;
1600 }
1601
1602 work_end -= IO_DELAY_PER_OPA_TAP;
1603 scc_mgr_set_dqs_en_phase_all_ranks(grp, p);
1604
1605 /* * The actual increment of dtaps is done outside of
1606 the if/else loop to share code */
1607 d = 0;
1608
1609 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: v/p: \
1610 vfifo=%u ptap=%u\n", __func__, __LINE__,
1611 v, p);
1612 } else {
1613 /* ******************************************************* */
1614 /* * step 3-5b: Find the right edge of the window using
1615 delay taps * */
1616 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase:vfifo=%u \
1617 ptap=%u dtap=%u bgn=%u\n", __func__, __LINE__,
1618 v, p, d, work_bgn);
1619
1620 work_end = work_bgn;
1621
1622 /* * The actual increment of dtaps is done outside of the
1623 if/else loop to share code */
1624
1625 /* Only here to counterbalance a subtract later on which is
1626 not needed if this branch of the algorithm is taken */
1627 max_working_cnt++;
1628 }
1629
1630 /* The dtap increment to find the failing edge is done here */
1631 for (; d <= IO_DQS_EN_DELAY_MAX; d++, work_end +=
1632 IO_DELAY_PER_DQS_EN_DCHAIN_TAP) {
1633 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: \
1634 end-2: dtap=%u\n", __func__, __LINE__, d);
1635 scc_mgr_set_dqs_en_delay_all_ranks(grp, d);
1636
1637 if (!rw_mgr_mem_calibrate_read_test_all_ranks(grp, 1,
1638 PASS_ONE_BIT,
1639 &bit_chk, 0)) {
1640 break;
1641 }
1642 }
1643
1644 /* Go back to working dtap */
1645 if (d != 0)
1646 work_end -= IO_DELAY_PER_DQS_EN_DCHAIN_TAP;
1647
1648 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: v/p/d: vfifo=%u \
1649 ptap=%u dtap=%u end=%u\n", __func__, __LINE__,
1650 v, p, d-1, work_end);
1651
1652 if (work_end < work_bgn) {
1653 /* nil range */
1654 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: end-2: \
1655 failed\n", __func__, __LINE__);
1656 return 0;
1657 }
1658
1659 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: found range [%u,%u]\n",
1660 __func__, __LINE__, work_bgn, work_end);
1661
1662 /* *************************************************************** */
1663 /*
1664 * * We need to calculate the number of dtaps that equal a ptap
1665 * * To do that we'll back up a ptap and re-find the edge of the
1666 * * window using dtaps
1667 */
1668
1669 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: calculate dtaps_per_ptap \
1670 for tracking\n", __func__, __LINE__);
1671
1672 /* Special case code for backing up a phase */
1673 if (p == 0) {
1674 p = IO_DQS_EN_PHASE_MAX;
1675 rw_mgr_decr_vfifo(grp, &v);
1676 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: backedup \
1677 cycle/phase: v=%u p=%u\n", __func__, __LINE__,
1678 v, p);
1679 } else {
1680 p = p - 1;
1681 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: backedup \
1682 phase only: v=%u p=%u", __func__, __LINE__,
1683 v, p);
1684 }
1685
1686 scc_mgr_set_dqs_en_phase_all_ranks(grp, p);
1687
1688 /*
1689 * Increase dtap until we first see a passing read (in case the
1690 * window is smaller than a ptap),
1691 * and then a failing read to mark the edge of the window again
1692 */
1693
1694 /* Find a passing read */
1695 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: find passing read\n",
1696 __func__, __LINE__);
1697 found_passing_read = 0;
1698 found_failing_read = 0;
1699 initial_failing_dtap = d;
1700 for (; d <= IO_DQS_EN_DELAY_MAX; d++) {
1701 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: testing \
1702 read d=%u\n", __func__, __LINE__, d);
1703 scc_mgr_set_dqs_en_delay_all_ranks(grp, d);
1704
1705 if (rw_mgr_mem_calibrate_read_test_all_ranks(grp, 1,
1706 PASS_ONE_BIT,
1707 &bit_chk, 0)) {
1708 found_passing_read = 1;
1709 break;
1710 }
1711 }
1712
1713 if (found_passing_read) {
1714 /* Find a failing read */
1715 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: find failing \
1716 read\n", __func__, __LINE__);
1717 for (d = d + 1; d <= IO_DQS_EN_DELAY_MAX; d++) {
1718 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: \
1719 testing read d=%u\n", __func__, __LINE__, d);
1720 scc_mgr_set_dqs_en_delay_all_ranks(grp, d);
1721
1722 if (!rw_mgr_mem_calibrate_read_test_all_ranks
1723 (grp, 1, PASS_ONE_BIT, &bit_chk, 0)) {
1724 found_failing_read = 1;
1725 break;
1726 }
1727 }
1728 } else {
1729 debug_cond(DLEVEL == 1, "%s:%d find_dqs_en_phase: failed to \
1730 calculate dtaps", __func__, __LINE__);
1731 debug_cond(DLEVEL == 1, "per ptap. Fall back on static value\n");
1732 }
1733
1734 /*
1735 * The dynamically calculated dtaps_per_ptap is only valid if we
1736 * found a passing/failing read. If we didn't, it means d hit the max
1737 * (IO_DQS_EN_DELAY_MAX). Otherwise, dtaps_per_ptap retains its
1738 * statically calculated value.
1739 */
1740 if (found_passing_read && found_failing_read)
1741 dtaps_per_ptap = d - initial_failing_dtap;
1742
1743 writel(dtaps_per_ptap, &sdr_reg_file->dtaps_per_ptap);
1744 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: dtaps_per_ptap=%u \
1745 - %u = %u", __func__, __LINE__, d,
1746 initial_failing_dtap, dtaps_per_ptap);
1747
1748 /* ******************************************** */
1749 /* * step 6: Find the centre of the window * */
1750 if (sdr_find_window_centre(&grp, &bit_chk, &work_bgn, &v, &d, &p,
1751 &work_mid, &work_end) == 0)
1752 return 0;
1753
1754 debug_cond(DLEVEL == 2, "%s:%d find_dqs_en_phase: center found: \
1755 vfifo=%u ptap=%u dtap=%u\n", __func__, __LINE__,
1756 v, p-1, d);
1757 return 1;
1758 }
1759
1760 /*
1761 * Try rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase across different
1762 * dq_in_delay values
1763 */
1764 static uint32_t
1765 rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase_sweep_dq_in_delay
1766 (uint32_t write_group, uint32_t read_group, uint32_t test_bgn)
1767 {
1768 uint32_t found;
1769 uint32_t i;
1770 uint32_t p;
1771 uint32_t d;
1772 uint32_t r;
1773
1774 const uint32_t delay_step = IO_IO_IN_DELAY_MAX /
1775 (RW_MGR_MEM_DQ_PER_READ_DQS-1);
1776 /* we start at zero, so have one less dq to devide among */
1777
1778 debug("%s:%d (%u,%u,%u)", __func__, __LINE__, write_group, read_group,
1779 test_bgn);
1780
1781 /* try different dq_in_delays since the dq path is shorter than dqs */
1782
1783 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
1784 r += NUM_RANKS_PER_SHADOW_REG) {
1785 for (i = 0, p = test_bgn, d = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++, p++, d += delay_step) {
1786 debug_cond(DLEVEL == 1, "%s:%d rw_mgr_mem_calibrate_\
1787 vfifo_find_dqs_", __func__, __LINE__);
1788 debug_cond(DLEVEL == 1, "en_phase_sweep_dq_in_delay: g=%u/%u ",
1789 write_group, read_group);
1790 debug_cond(DLEVEL == 1, "r=%u, i=%u p=%u d=%u\n", r, i , p, d);
1791 scc_mgr_set_dq_in_delay(p, d);
1792 scc_mgr_load_dq(p);
1793 }
1794 writel(0, &sdr_scc_mgr->update);
1795 }
1796
1797 found = rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase(read_group);
1798
1799 debug_cond(DLEVEL == 1, "%s:%d rw_mgr_mem_calibrate_vfifo_find_dqs_\
1800 en_phase_sweep_dq", __func__, __LINE__);
1801 debug_cond(DLEVEL == 1, "_in_delay: g=%u/%u found=%u; Reseting delay \
1802 chain to zero\n", write_group, read_group, found);
1803
1804 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
1805 r += NUM_RANKS_PER_SHADOW_REG) {
1806 for (i = 0, p = test_bgn; i < RW_MGR_MEM_DQ_PER_READ_DQS;
1807 i++, p++) {
1808 scc_mgr_set_dq_in_delay(p, 0);
1809 scc_mgr_load_dq(p);
1810 }
1811 writel(0, &sdr_scc_mgr->update);
1812 }
1813
1814 return found;
1815 }
1816
1817 /* per-bit deskew DQ and center */
1818 static uint32_t rw_mgr_mem_calibrate_vfifo_center(uint32_t rank_bgn,
1819 uint32_t write_group, uint32_t read_group, uint32_t test_bgn,
1820 uint32_t use_read_test, uint32_t update_fom)
1821 {
1822 uint32_t i, p, d, min_index;
1823 /*
1824 * Store these as signed since there are comparisons with
1825 * signed numbers.
1826 */
1827 uint32_t bit_chk;
1828 uint32_t sticky_bit_chk;
1829 int32_t left_edge[RW_MGR_MEM_DQ_PER_READ_DQS];
1830 int32_t right_edge[RW_MGR_MEM_DQ_PER_READ_DQS];
1831 int32_t final_dq[RW_MGR_MEM_DQ_PER_READ_DQS];
1832 int32_t mid;
1833 int32_t orig_mid_min, mid_min;
1834 int32_t new_dqs, start_dqs, start_dqs_en, shift_dq, final_dqs,
1835 final_dqs_en;
1836 int32_t dq_margin, dqs_margin;
1837 uint32_t stop;
1838 uint32_t temp_dq_in_delay1, temp_dq_in_delay2;
1839 uint32_t addr;
1840
1841 debug("%s:%d: %u %u", __func__, __LINE__, read_group, test_bgn);
1842
1843 addr = SDR_PHYGRP_SCCGRP_ADDRESS | SCC_MGR_DQS_IN_DELAY_OFFSET;
1844 start_dqs = readl(addr + (read_group << 2));
1845 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS)
1846 start_dqs_en = readl(addr + ((read_group << 2)
1847 - IO_DQS_EN_DELAY_OFFSET));
1848
1849 /* set the left and right edge of each bit to an illegal value */
1850 /* use (IO_IO_IN_DELAY_MAX + 1) as an illegal value */
1851 sticky_bit_chk = 0;
1852 for (i = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
1853 left_edge[i] = IO_IO_IN_DELAY_MAX + 1;
1854 right_edge[i] = IO_IO_IN_DELAY_MAX + 1;
1855 }
1856
1857 /* Search for the left edge of the window for each bit */
1858 for (d = 0; d <= IO_IO_IN_DELAY_MAX; d++) {
1859 scc_mgr_apply_group_dq_in_delay(write_group, test_bgn, d);
1860
1861 writel(0, &sdr_scc_mgr->update);
1862
1863 /*
1864 * Stop searching when the read test doesn't pass AND when
1865 * we've seen a passing read on every bit.
1866 */
1867 if (use_read_test) {
1868 stop = !rw_mgr_mem_calibrate_read_test(rank_bgn,
1869 read_group, NUM_READ_PB_TESTS, PASS_ONE_BIT,
1870 &bit_chk, 0, 0);
1871 } else {
1872 rw_mgr_mem_calibrate_write_test(rank_bgn, write_group,
1873 0, PASS_ONE_BIT,
1874 &bit_chk, 0);
1875 bit_chk = bit_chk >> (RW_MGR_MEM_DQ_PER_READ_DQS *
1876 (read_group - (write_group *
1877 RW_MGR_MEM_IF_READ_DQS_WIDTH /
1878 RW_MGR_MEM_IF_WRITE_DQS_WIDTH)));
1879 stop = (bit_chk == 0);
1880 }
1881 sticky_bit_chk = sticky_bit_chk | bit_chk;
1882 stop = stop && (sticky_bit_chk == param->read_correct_mask);
1883 debug_cond(DLEVEL == 2, "%s:%d vfifo_center(left): dtap=%u => %u == %u \
1884 && %u", __func__, __LINE__, d,
1885 sticky_bit_chk,
1886 param->read_correct_mask, stop);
1887
1888 if (stop == 1) {
1889 break;
1890 } else {
1891 for (i = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
1892 if (bit_chk & 1) {
1893 /* Remember a passing test as the
1894 left_edge */
1895 left_edge[i] = d;
1896 } else {
1897 /* If a left edge has not been seen yet,
1898 then a future passing test will mark
1899 this edge as the right edge */
1900 if (left_edge[i] ==
1901 IO_IO_IN_DELAY_MAX + 1) {
1902 right_edge[i] = -(d + 1);
1903 }
1904 }
1905 bit_chk = bit_chk >> 1;
1906 }
1907 }
1908 }
1909
1910 /* Reset DQ delay chains to 0 */
1911 scc_mgr_apply_group_dq_in_delay(test_bgn, 0);
1912 sticky_bit_chk = 0;
1913 for (i = RW_MGR_MEM_DQ_PER_READ_DQS - 1;; i--) {
1914 debug_cond(DLEVEL == 2, "%s:%d vfifo_center: left_edge[%u]: \
1915 %d right_edge[%u]: %d\n", __func__, __LINE__,
1916 i, left_edge[i], i, right_edge[i]);
1917
1918 /*
1919 * Check for cases where we haven't found the left edge,
1920 * which makes our assignment of the the right edge invalid.
1921 * Reset it to the illegal value.
1922 */
1923 if ((left_edge[i] == IO_IO_IN_DELAY_MAX + 1) && (
1924 right_edge[i] != IO_IO_IN_DELAY_MAX + 1)) {
1925 right_edge[i] = IO_IO_IN_DELAY_MAX + 1;
1926 debug_cond(DLEVEL == 2, "%s:%d vfifo_center: reset \
1927 right_edge[%u]: %d\n", __func__, __LINE__,
1928 i, right_edge[i]);
1929 }
1930
1931 /*
1932 * Reset sticky bit (except for bits where we have seen
1933 * both the left and right edge).
1934 */
1935 sticky_bit_chk = sticky_bit_chk << 1;
1936 if ((left_edge[i] != IO_IO_IN_DELAY_MAX + 1) &&
1937 (right_edge[i] != IO_IO_IN_DELAY_MAX + 1)) {
1938 sticky_bit_chk = sticky_bit_chk | 1;
1939 }
1940
1941 if (i == 0)
1942 break;
1943 }
1944
1945 /* Search for the right edge of the window for each bit */
1946 for (d = 0; d <= IO_DQS_IN_DELAY_MAX - start_dqs; d++) {
1947 scc_mgr_set_dqs_bus_in_delay(read_group, d + start_dqs);
1948 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS) {
1949 uint32_t delay = d + start_dqs_en;
1950 if (delay > IO_DQS_EN_DELAY_MAX)
1951 delay = IO_DQS_EN_DELAY_MAX;
1952 scc_mgr_set_dqs_en_delay(read_group, delay);
1953 }
1954 scc_mgr_load_dqs(read_group);
1955
1956 writel(0, &sdr_scc_mgr->update);
1957
1958 /*
1959 * Stop searching when the read test doesn't pass AND when
1960 * we've seen a passing read on every bit.
1961 */
1962 if (use_read_test) {
1963 stop = !rw_mgr_mem_calibrate_read_test(rank_bgn,
1964 read_group, NUM_READ_PB_TESTS, PASS_ONE_BIT,
1965 &bit_chk, 0, 0);
1966 } else {
1967 rw_mgr_mem_calibrate_write_test(rank_bgn, write_group,
1968 0, PASS_ONE_BIT,
1969 &bit_chk, 0);
1970 bit_chk = bit_chk >> (RW_MGR_MEM_DQ_PER_READ_DQS *
1971 (read_group - (write_group *
1972 RW_MGR_MEM_IF_READ_DQS_WIDTH /
1973 RW_MGR_MEM_IF_WRITE_DQS_WIDTH)));
1974 stop = (bit_chk == 0);
1975 }
1976 sticky_bit_chk = sticky_bit_chk | bit_chk;
1977 stop = stop && (sticky_bit_chk == param->read_correct_mask);
1978
1979 debug_cond(DLEVEL == 2, "%s:%d vfifo_center(right): dtap=%u => %u == \
1980 %u && %u", __func__, __LINE__, d,
1981 sticky_bit_chk, param->read_correct_mask, stop);
1982
1983 if (stop == 1) {
1984 break;
1985 } else {
1986 for (i = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
1987 if (bit_chk & 1) {
1988 /* Remember a passing test as
1989 the right_edge */
1990 right_edge[i] = d;
1991 } else {
1992 if (d != 0) {
1993 /* If a right edge has not been
1994 seen yet, then a future passing
1995 test will mark this edge as the
1996 left edge */
1997 if (right_edge[i] ==
1998 IO_IO_IN_DELAY_MAX + 1) {
1999 left_edge[i] = -(d + 1);
2000 }
2001 } else {
2002 /* d = 0 failed, but it passed
2003 when testing the left edge,
2004 so it must be marginal,
2005 set it to -1 */
2006 if (right_edge[i] ==
2007 IO_IO_IN_DELAY_MAX + 1 &&
2008 left_edge[i] !=
2009 IO_IO_IN_DELAY_MAX
2010 + 1) {
2011 right_edge[i] = -1;
2012 }
2013 /* If a right edge has not been
2014 seen yet, then a future passing
2015 test will mark this edge as the
2016 left edge */
2017 else if (right_edge[i] ==
2018 IO_IO_IN_DELAY_MAX +
2019 1) {
2020 left_edge[i] = -(d + 1);
2021 }
2022 }
2023 }
2024
2025 debug_cond(DLEVEL == 2, "%s:%d vfifo_center[r,\
2026 d=%u]: ", __func__, __LINE__, d);
2027 debug_cond(DLEVEL == 2, "bit_chk_test=%d left_edge[%u]: %d ",
2028 (int)(bit_chk & 1), i, left_edge[i]);
2029 debug_cond(DLEVEL == 2, "right_edge[%u]: %d\n", i,
2030 right_edge[i]);
2031 bit_chk = bit_chk >> 1;
2032 }
2033 }
2034 }
2035
2036 /* Check that all bits have a window */
2037 for (i = 0; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
2038 debug_cond(DLEVEL == 2, "%s:%d vfifo_center: left_edge[%u]: \
2039 %d right_edge[%u]: %d", __func__, __LINE__,
2040 i, left_edge[i], i, right_edge[i]);
2041 if ((left_edge[i] == IO_IO_IN_DELAY_MAX + 1) || (right_edge[i]
2042 == IO_IO_IN_DELAY_MAX + 1)) {
2043 /*
2044 * Restore delay chain settings before letting the loop
2045 * in rw_mgr_mem_calibrate_vfifo to retry different
2046 * dqs/ck relationships.
2047 */
2048 scc_mgr_set_dqs_bus_in_delay(read_group, start_dqs);
2049 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS) {
2050 scc_mgr_set_dqs_en_delay(read_group,
2051 start_dqs_en);
2052 }
2053 scc_mgr_load_dqs(read_group);
2054 writel(0, &sdr_scc_mgr->update);
2055
2056 debug_cond(DLEVEL == 1, "%s:%d vfifo_center: failed to \
2057 find edge [%u]: %d %d", __func__, __LINE__,
2058 i, left_edge[i], right_edge[i]);
2059 if (use_read_test) {
2060 set_failing_group_stage(read_group *
2061 RW_MGR_MEM_DQ_PER_READ_DQS + i,
2062 CAL_STAGE_VFIFO,
2063 CAL_SUBSTAGE_VFIFO_CENTER);
2064 } else {
2065 set_failing_group_stage(read_group *
2066 RW_MGR_MEM_DQ_PER_READ_DQS + i,
2067 CAL_STAGE_VFIFO_AFTER_WRITES,
2068 CAL_SUBSTAGE_VFIFO_CENTER);
2069 }
2070 return 0;
2071 }
2072 }
2073
2074 /* Find middle of window for each DQ bit */
2075 mid_min = left_edge[0] - right_edge[0];
2076 min_index = 0;
2077 for (i = 1; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++) {
2078 mid = left_edge[i] - right_edge[i];
2079 if (mid < mid_min) {
2080 mid_min = mid;
2081 min_index = i;
2082 }
2083 }
2084
2085 /*
2086 * -mid_min/2 represents the amount that we need to move DQS.
2087 * If mid_min is odd and positive we'll need to add one to
2088 * make sure the rounding in further calculations is correct
2089 * (always bias to the right), so just add 1 for all positive values.
2090 */
2091 if (mid_min > 0)
2092 mid_min++;
2093
2094 mid_min = mid_min / 2;
2095
2096 debug_cond(DLEVEL == 1, "%s:%d vfifo_center: mid_min=%d (index=%u)\n",
2097 __func__, __LINE__, mid_min, min_index);
2098
2099 /* Determine the amount we can change DQS (which is -mid_min) */
2100 orig_mid_min = mid_min;
2101 new_dqs = start_dqs - mid_min;
2102 if (new_dqs > IO_DQS_IN_DELAY_MAX)
2103 new_dqs = IO_DQS_IN_DELAY_MAX;
2104 else if (new_dqs < 0)
2105 new_dqs = 0;
2106
2107 mid_min = start_dqs - new_dqs;
2108 debug_cond(DLEVEL == 1, "vfifo_center: new mid_min=%d new_dqs=%d\n",
2109 mid_min, new_dqs);
2110
2111 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS) {
2112 if (start_dqs_en - mid_min > IO_DQS_EN_DELAY_MAX)
2113 mid_min += start_dqs_en - mid_min - IO_DQS_EN_DELAY_MAX;
2114 else if (start_dqs_en - mid_min < 0)
2115 mid_min += start_dqs_en - mid_min;
2116 }
2117 new_dqs = start_dqs - mid_min;
2118
2119 debug_cond(DLEVEL == 1, "vfifo_center: start_dqs=%d start_dqs_en=%d \
2120 new_dqs=%d mid_min=%d\n", start_dqs,
2121 IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS ? start_dqs_en : -1,
2122 new_dqs, mid_min);
2123
2124 /* Initialize data for export structures */
2125 dqs_margin = IO_IO_IN_DELAY_MAX + 1;
2126 dq_margin = IO_IO_IN_DELAY_MAX + 1;
2127
2128 /* add delay to bring centre of all DQ windows to the same "level" */
2129 for (i = 0, p = test_bgn; i < RW_MGR_MEM_DQ_PER_READ_DQS; i++, p++) {
2130 /* Use values before divide by 2 to reduce round off error */
2131 shift_dq = (left_edge[i] - right_edge[i] -
2132 (left_edge[min_index] - right_edge[min_index]))/2 +
2133 (orig_mid_min - mid_min);
2134
2135 debug_cond(DLEVEL == 2, "vfifo_center: before: \
2136 shift_dq[%u]=%d\n", i, shift_dq);
2137
2138 addr = SDR_PHYGRP_SCCGRP_ADDRESS | SCC_MGR_IO_IN_DELAY_OFFSET;
2139 temp_dq_in_delay1 = readl(addr + (p << 2));
2140 temp_dq_in_delay2 = readl(addr + (i << 2));
2141
2142 if (shift_dq + (int32_t)temp_dq_in_delay1 >
2143 (int32_t)IO_IO_IN_DELAY_MAX) {
2144 shift_dq = (int32_t)IO_IO_IN_DELAY_MAX - temp_dq_in_delay2;
2145 } else if (shift_dq + (int32_t)temp_dq_in_delay1 < 0) {
2146 shift_dq = -(int32_t)temp_dq_in_delay1;
2147 }
2148 debug_cond(DLEVEL == 2, "vfifo_center: after: \
2149 shift_dq[%u]=%d\n", i, shift_dq);
2150 final_dq[i] = temp_dq_in_delay1 + shift_dq;
2151 scc_mgr_set_dq_in_delay(p, final_dq[i]);
2152 scc_mgr_load_dq(p);
2153
2154 debug_cond(DLEVEL == 2, "vfifo_center: margin[%u]=[%d,%d]\n", i,
2155 left_edge[i] - shift_dq + (-mid_min),
2156 right_edge[i] + shift_dq - (-mid_min));
2157 /* To determine values for export structures */
2158 if (left_edge[i] - shift_dq + (-mid_min) < dq_margin)
2159 dq_margin = left_edge[i] - shift_dq + (-mid_min);
2160
2161 if (right_edge[i] + shift_dq - (-mid_min) < dqs_margin)
2162 dqs_margin = right_edge[i] + shift_dq - (-mid_min);
2163 }
2164
2165 final_dqs = new_dqs;
2166 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS)
2167 final_dqs_en = start_dqs_en - mid_min;
2168
2169 /* Move DQS-en */
2170 if (IO_SHIFT_DQS_EN_WHEN_SHIFT_DQS) {
2171 scc_mgr_set_dqs_en_delay(read_group, final_dqs_en);
2172 scc_mgr_load_dqs(read_group);
2173 }
2174
2175 /* Move DQS */
2176 scc_mgr_set_dqs_bus_in_delay(read_group, final_dqs);
2177 scc_mgr_load_dqs(read_group);
2178 debug_cond(DLEVEL == 2, "%s:%d vfifo_center: dq_margin=%d \
2179 dqs_margin=%d", __func__, __LINE__,
2180 dq_margin, dqs_margin);
2181
2182 /*
2183 * Do not remove this line as it makes sure all of our decisions
2184 * have been applied. Apply the update bit.
2185 */
2186 writel(0, &sdr_scc_mgr->update);
2187
2188 return (dq_margin >= 0) && (dqs_margin >= 0);
2189 }
2190
2191 /**
2192 * rw_mgr_mem_calibrate_guaranteed_write() - Perform guaranteed write into the device
2193 * @rw_group: Read/Write Group
2194 * @phase: DQ/DQS phase
2195 *
2196 * Because initially no communication ca be reliably performed with the memory
2197 * device, the sequencer uses a guaranteed write mechanism to write data into
2198 * the memory device.
2199 */
2200 static int rw_mgr_mem_calibrate_guaranteed_write(const u32 rw_group,
2201 const u32 phase)
2202 {
2203 u32 bit_chk;
2204 int ret;
2205
2206 /* Set a particular DQ/DQS phase. */
2207 scc_mgr_set_dqdqs_output_phase_all_ranks(rw_group, phase);
2208
2209 debug_cond(DLEVEL == 1, "%s:%d guaranteed write: g=%u p=%u\n",
2210 __func__, __LINE__, rw_group, phase);
2211
2212 /*
2213 * Altera EMI_RM 2015.05.04 :: Figure 1-25
2214 * Load up the patterns used by read calibration using the
2215 * current DQDQS phase.
2216 */
2217 rw_mgr_mem_calibrate_read_load_patterns(0, 1);
2218
2219 if (gbl->phy_debug_mode_flags & PHY_DEBUG_DISABLE_GUARANTEED_READ)
2220 return 0;
2221
2222 /*
2223 * Altera EMI_RM 2015.05.04 :: Figure 1-26
2224 * Back-to-Back reads of the patterns used for calibration.
2225 */
2226 ret = rw_mgr_mem_calibrate_read_test_patterns(0, rw_group, 1,
2227 &bit_chk, 1);
2228 if (!ret) { /* FIXME: 0 means failure in this old code :-( */
2229 debug_cond(DLEVEL == 1,
2230 "%s:%d Guaranteed read test failed: g=%u p=%u\n",
2231 __func__, __LINE__, rw_group, phase);
2232 return -EIO;
2233 }
2234
2235 return 0;
2236 }
2237
2238 /**
2239 * rw_mgr_mem_calibrate_dqs_enable_calibration() - DQS Enable Calibration
2240 * @rw_group: Read/Write Group
2241 * @test_bgn: Rank at which the test begins
2242 *
2243 * DQS enable calibration ensures reliable capture of the DQ signal without
2244 * glitches on the DQS line.
2245 */
2246 static int rw_mgr_mem_calibrate_dqs_enable_calibration(const u32 rw_group,
2247 const u32 test_bgn)
2248 {
2249 int ret;
2250
2251 /*
2252 * Altera EMI_RM 2015.05.04 :: Figure 1-27
2253 * DQS and DQS Eanble Signal Relationships.
2254 */
2255 ret = rw_mgr_mem_calibrate_vfifo_find_dqs_en_phase_sweep_dq_in_delay(
2256 rw_group, rw_group, test_bgn);
2257 if (!ret) /* FIXME: 0 means failure in this old code :-( */
2258 return -EIO;
2259
2260 return 0;
2261 }
2262
2263 /**
2264 * rw_mgr_mem_calibrate_dq_dqs_centering() - Centering DQ/DQS
2265 * @rw_group: Read/Write Group
2266 * @test_bgn: Rank at which the test begins
2267 * @use_read_test: Perform a read test
2268 * @update_fom: Update FOM
2269 *
2270 * The centerin DQ/DQS stage attempts to align DQ and DQS signals on reads
2271 * within a group.
2272 */
2273 static int
2274 rw_mgr_mem_calibrate_dq_dqs_centering(const u32 rw_group, const u32 test_bgn,
2275 const int use_read_test,
2276 const int update_fom)
2277
2278 {
2279 int ret, grp_calibrated;
2280 u32 rank_bgn, sr;
2281
2282 /*
2283 * Altera EMI_RM 2015.05.04 :: Figure 1-28
2284 * Read per-bit deskew can be done on a per shadow register basis.
2285 */
2286 grp_calibrated = 1;
2287 for (rank_bgn = 0, sr = 0;
2288 rank_bgn < RW_MGR_MEM_NUMBER_OF_RANKS;
2289 rank_bgn += NUM_RANKS_PER_SHADOW_REG, sr++) {
2290 /* Check if this set of ranks should be skipped entirely. */
2291 if (param->skip_shadow_regs[sr])
2292 continue;
2293
2294 ret = rw_mgr_mem_calibrate_vfifo_center(rank_bgn, rw_group,
2295 rw_group, test_bgn,
2296 use_read_test,
2297 update_fom);
2298 if (ret)
2299 continue;
2300
2301 grp_calibrated = 0;
2302 }
2303
2304 if (!grp_calibrated)
2305 return -EIO;
2306
2307 return 0;
2308 }
2309
2310 /**
2311 * rw_mgr_mem_calibrate_vfifo() - Calibrate the read valid prediction FIFO
2312 * @rw_group: Read/Write Group
2313 * @test_bgn: Rank at which the test begins
2314 *
2315 * Stage 1: Calibrate the read valid prediction FIFO.
2316 *
2317 * This function implements UniPHY calibration Stage 1, as explained in
2318 * detail in Altera EMI_RM 2015.05.04 , "UniPHY Calibration Stages".
2319 *
2320 * - read valid prediction will consist of finding:
2321 * - DQS enable phase and DQS enable delay (DQS Enable Calibration)
2322 * - DQS input phase and DQS input delay (DQ/DQS Centering)
2323 * - we also do a per-bit deskew on the DQ lines.
2324 */
2325 static int rw_mgr_mem_calibrate_vfifo(const u32 rw_group, const u32 test_bgn)
2326 {
2327 uint32_t p, d;
2328 uint32_t dtaps_per_ptap;
2329 uint32_t failed_substage;
2330
2331 int ret;
2332
2333 debug("%s:%d: %u %u\n", __func__, __LINE__, rw_group, test_bgn);
2334
2335 /* Update info for sims */
2336 reg_file_set_group(rw_group);
2337 reg_file_set_stage(CAL_STAGE_VFIFO);
2338 reg_file_set_sub_stage(CAL_SUBSTAGE_GUARANTEED_READ);
2339
2340 failed_substage = CAL_SUBSTAGE_GUARANTEED_READ;
2341
2342 /* USER Determine number of delay taps for each phase tap. */
2343 dtaps_per_ptap = DIV_ROUND_UP(IO_DELAY_PER_OPA_TAP,
2344 IO_DELAY_PER_DQS_EN_DCHAIN_TAP) - 1;
2345
2346 for (d = 0; d <= dtaps_per_ptap; d += 2) {
2347 /*
2348 * In RLDRAMX we may be messing the delay of pins in
2349 * the same write rw_group but outside of the current read
2350 * the rw_group, but that's ok because we haven't calibrated
2351 * output side yet.
2352 */
2353 if (d > 0) {
2354 scc_mgr_apply_group_all_out_delay_add_all_ranks(
2355 rw_group, d);
2356 }
2357
2358 for (p = 0; p <= IO_DQDQS_OUT_PHASE_MAX; p++) {
2359 /* 1) Guaranteed Write */
2360 ret = rw_mgr_mem_calibrate_guaranteed_write(rw_group, p);
2361 if (ret)
2362 break;
2363
2364 /* 2) DQS Enable Calibration */
2365 ret = rw_mgr_mem_calibrate_dqs_enable_calibration(rw_group,
2366 test_bgn);
2367 if (ret) {
2368 failed_substage = CAL_SUBSTAGE_DQS_EN_PHASE;
2369 continue;
2370 }
2371
2372 /* 3) Centering DQ/DQS */
2373 /*
2374 * If doing read after write calibration, do not update
2375 * FOM now. Do it then.
2376 */
2377 ret = rw_mgr_mem_calibrate_dq_dqs_centering(rw_group,
2378 test_bgn, 1, 0);
2379 if (ret) {
2380 failed_substage = CAL_SUBSTAGE_VFIFO_CENTER;
2381 continue;
2382 }
2383
2384 /* All done. */
2385 goto cal_done_ok;
2386 }
2387 }
2388
2389 /* Calibration Stage 1 failed. */
2390 set_failing_group_stage(rw_group, CAL_STAGE_VFIFO, failed_substage);
2391 return 0;
2392
2393 /* Calibration Stage 1 completed OK. */
2394 cal_done_ok:
2395 /*
2396 * Reset the delay chains back to zero if they have moved > 1
2397 * (check for > 1 because loop will increase d even when pass in
2398 * first case).
2399 */
2400 if (d > 2)
2401 scc_mgr_zero_group(rw_group, 1);
2402
2403 return 1;
2404 }
2405
2406 /* VFIFO Calibration -- Read Deskew Calibration after write deskew */
2407 static uint32_t rw_mgr_mem_calibrate_vfifo_end(uint32_t read_group,
2408 uint32_t test_bgn)
2409 {
2410 uint32_t rank_bgn, sr;
2411 uint32_t grp_calibrated;
2412 uint32_t write_group;
2413
2414 debug("%s:%d %u %u", __func__, __LINE__, read_group, test_bgn);
2415
2416 /* update info for sims */
2417
2418 reg_file_set_stage(CAL_STAGE_VFIFO_AFTER_WRITES);
2419 reg_file_set_sub_stage(CAL_SUBSTAGE_VFIFO_CENTER);
2420
2421 write_group = read_group;
2422
2423 /* update info for sims */
2424 reg_file_set_group(read_group);
2425
2426 grp_calibrated = 1;
2427 /* Read per-bit deskew can be done on a per shadow register basis */
2428 for (rank_bgn = 0, sr = 0; rank_bgn < RW_MGR_MEM_NUMBER_OF_RANKS;
2429 rank_bgn += NUM_RANKS_PER_SHADOW_REG, ++sr) {
2430 /* Determine if this set of ranks should be skipped entirely */
2431 if (!param->skip_shadow_regs[sr]) {
2432 /* This is the last calibration round, update FOM here */
2433 if (!rw_mgr_mem_calibrate_vfifo_center(rank_bgn,
2434 write_group,
2435 read_group,
2436 test_bgn, 0,
2437 1)) {
2438 grp_calibrated = 0;
2439 }
2440 }
2441 }
2442
2443
2444 if (grp_calibrated == 0) {
2445 set_failing_group_stage(write_group,
2446 CAL_STAGE_VFIFO_AFTER_WRITES,
2447 CAL_SUBSTAGE_VFIFO_CENTER);
2448 return 0;
2449 }
2450
2451 return 1;
2452 }
2453
2454 /* Calibrate LFIFO to find smallest read latency */
2455 static uint32_t rw_mgr_mem_calibrate_lfifo(void)
2456 {
2457 uint32_t found_one;
2458 uint32_t bit_chk;
2459
2460 debug("%s:%d\n", __func__, __LINE__);
2461
2462 /* update info for sims */
2463 reg_file_set_stage(CAL_STAGE_LFIFO);
2464 reg_file_set_sub_stage(CAL_SUBSTAGE_READ_LATENCY);
2465
2466 /* Load up the patterns used by read calibration for all ranks */
2467 rw_mgr_mem_calibrate_read_load_patterns(0, 1);
2468 found_one = 0;
2469
2470 do {
2471 writel(gbl->curr_read_lat, &phy_mgr_cfg->phy_rlat);
2472 debug_cond(DLEVEL == 2, "%s:%d lfifo: read_lat=%u",
2473 __func__, __LINE__, gbl->curr_read_lat);
2474
2475 if (!rw_mgr_mem_calibrate_read_test_all_ranks(0,
2476 NUM_READ_TESTS,
2477 PASS_ALL_BITS,
2478 &bit_chk, 1)) {
2479 break;
2480 }
2481
2482 found_one = 1;
2483 /* reduce read latency and see if things are working */
2484 /* correctly */
2485 gbl->curr_read_lat--;
2486 } while (gbl->curr_read_lat > 0);
2487
2488 /* reset the fifos to get pointers to known state */
2489
2490 writel(0, &phy_mgr_cmd->fifo_reset);
2491
2492 if (found_one) {
2493 /* add a fudge factor to the read latency that was determined */
2494 gbl->curr_read_lat += 2;
2495 writel(gbl->curr_read_lat, &phy_mgr_cfg->phy_rlat);
2496 debug_cond(DLEVEL == 2, "%s:%d lfifo: success: using \
2497 read_lat=%u\n", __func__, __LINE__,
2498 gbl->curr_read_lat);
2499 return 1;
2500 } else {
2501 set_failing_group_stage(0xff, CAL_STAGE_LFIFO,
2502 CAL_SUBSTAGE_READ_LATENCY);
2503
2504 debug_cond(DLEVEL == 2, "%s:%d lfifo: failed at initial \
2505 read_lat=%u\n", __func__, __LINE__,
2506 gbl->curr_read_lat);
2507 return 0;
2508 }
2509 }
2510
2511 /*
2512 * issue write test command.
2513 * two variants are provided. one that just tests a write pattern and
2514 * another that tests datamask functionality.
2515 */
2516 static void rw_mgr_mem_calibrate_write_test_issue(uint32_t group,
2517 uint32_t test_dm)
2518 {
2519 uint32_t mcc_instruction;
2520 uint32_t quick_write_mode = (((STATIC_CALIB_STEPS) & CALIB_SKIP_WRITES) &&
2521 ENABLE_SUPER_QUICK_CALIBRATION);
2522 uint32_t rw_wl_nop_cycles;
2523 uint32_t addr;
2524
2525 /*
2526 * Set counter and jump addresses for the right
2527 * number of NOP cycles.
2528 * The number of supported NOP cycles can range from -1 to infinity
2529 * Three different cases are handled:
2530 *
2531 * 1. For a number of NOP cycles greater than 0, the RW Mgr looping
2532 * mechanism will be used to insert the right number of NOPs
2533 *
2534 * 2. For a number of NOP cycles equals to 0, the micro-instruction
2535 * issuing the write command will jump straight to the
2536 * micro-instruction that turns on DQS (for DDRx), or outputs write
2537 * data (for RLD), skipping
2538 * the NOP micro-instruction all together
2539 *
2540 * 3. A number of NOP cycles equal to -1 indicates that DQS must be
2541 * turned on in the same micro-instruction that issues the write
2542 * command. Then we need
2543 * to directly jump to the micro-instruction that sends out the data
2544 *
2545 * NOTE: Implementing this mechanism uses 2 RW Mgr jump-counters
2546 * (2 and 3). One jump-counter (0) is used to perform multiple
2547 * write-read operations.
2548 * one counter left to issue this command in "multiple-group" mode
2549 */
2550
2551 rw_wl_nop_cycles = gbl->rw_wl_nop_cycles;
2552
2553 if (rw_wl_nop_cycles == -1) {
2554 /*
2555 * CNTR 2 - We want to execute the special write operation that
2556 * turns on DQS right away and then skip directly to the
2557 * instruction that sends out the data. We set the counter to a
2558 * large number so that the jump is always taken.
2559 */
2560 writel(0xFF, &sdr_rw_load_mgr_regs->load_cntr2);
2561
2562 /* CNTR 3 - Not used */
2563 if (test_dm) {
2564 mcc_instruction = RW_MGR_LFSR_WR_RD_DM_BANK_0_WL_1;
2565 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_DATA,
2566 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2567 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_NOP,
2568 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
2569 } else {
2570 mcc_instruction = RW_MGR_LFSR_WR_RD_BANK_0_WL_1;
2571 writel(RW_MGR_LFSR_WR_RD_BANK_0_DATA,
2572 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2573 writel(RW_MGR_LFSR_WR_RD_BANK_0_NOP,
2574 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
2575 }
2576 } else if (rw_wl_nop_cycles == 0) {
2577 /*
2578 * CNTR 2 - We want to skip the NOP operation and go straight
2579 * to the DQS enable instruction. We set the counter to a large
2580 * number so that the jump is always taken.
2581 */
2582 writel(0xFF, &sdr_rw_load_mgr_regs->load_cntr2);
2583
2584 /* CNTR 3 - Not used */
2585 if (test_dm) {
2586 mcc_instruction = RW_MGR_LFSR_WR_RD_DM_BANK_0;
2587 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_DQS,
2588 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2589 } else {
2590 mcc_instruction = RW_MGR_LFSR_WR_RD_BANK_0;
2591 writel(RW_MGR_LFSR_WR_RD_BANK_0_DQS,
2592 &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2593 }
2594 } else {
2595 /*
2596 * CNTR 2 - In this case we want to execute the next instruction
2597 * and NOT take the jump. So we set the counter to 0. The jump
2598 * address doesn't count.
2599 */
2600 writel(0x0, &sdr_rw_load_mgr_regs->load_cntr2);
2601 writel(0x0, &sdr_rw_load_jump_mgr_regs->load_jump_add2);
2602
2603 /*
2604 * CNTR 3 - Set the nop counter to the number of cycles we
2605 * need to loop for, minus 1.
2606 */
2607 writel(rw_wl_nop_cycles - 1, &sdr_rw_load_mgr_regs->load_cntr3);
2608 if (test_dm) {
2609 mcc_instruction = RW_MGR_LFSR_WR_RD_DM_BANK_0;
2610 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_NOP,
2611 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
2612 } else {
2613 mcc_instruction = RW_MGR_LFSR_WR_RD_BANK_0;
2614 writel(RW_MGR_LFSR_WR_RD_BANK_0_NOP,
2615 &sdr_rw_load_jump_mgr_regs->load_jump_add3);
2616 }
2617 }
2618
2619 writel(0, SDR_PHYGRP_RWMGRGRP_ADDRESS |
2620 RW_MGR_RESET_READ_DATAPATH_OFFSET);
2621
2622 if (quick_write_mode)
2623 writel(0x08, &sdr_rw_load_mgr_regs->load_cntr0);
2624 else
2625 writel(0x40, &sdr_rw_load_mgr_regs->load_cntr0);
2626
2627 writel(mcc_instruction, &sdr_rw_load_jump_mgr_regs->load_jump_add0);
2628
2629 /*
2630 * CNTR 1 - This is used to ensure enough time elapses
2631 * for read data to come back.
2632 */
2633 writel(0x30, &sdr_rw_load_mgr_regs->load_cntr1);
2634
2635 if (test_dm) {
2636 writel(RW_MGR_LFSR_WR_RD_DM_BANK_0_WAIT,
2637 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
2638 } else {
2639 writel(RW_MGR_LFSR_WR_RD_BANK_0_WAIT,
2640 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
2641 }
2642
2643 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_RUN_SINGLE_GROUP_OFFSET;
2644 writel(mcc_instruction, addr + (group << 2));
2645 }
2646
2647 /* Test writes, can check for a single bit pass or multiple bit pass */
2648 static uint32_t rw_mgr_mem_calibrate_write_test(uint32_t rank_bgn,
2649 uint32_t write_group, uint32_t use_dm, uint32_t all_correct,
2650 uint32_t *bit_chk, uint32_t all_ranks)
2651 {
2652 uint32_t r;
2653 uint32_t correct_mask_vg;
2654 uint32_t tmp_bit_chk;
2655 uint32_t vg;
2656 uint32_t rank_end = all_ranks ? RW_MGR_MEM_NUMBER_OF_RANKS :
2657 (rank_bgn + NUM_RANKS_PER_SHADOW_REG);
2658 uint32_t addr_rw_mgr;
2659 uint32_t base_rw_mgr;
2660
2661 *bit_chk = param->write_correct_mask;
2662 correct_mask_vg = param->write_correct_mask_vg;
2663
2664 for (r = rank_bgn; r < rank_end; r++) {
2665 if (param->skip_ranks[r]) {
2666 /* request to skip the rank */
2667 continue;
2668 }
2669
2670 /* set rank */
2671 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_READ_WRITE);
2672
2673 tmp_bit_chk = 0;
2674 addr_rw_mgr = SDR_PHYGRP_RWMGRGRP_ADDRESS;
2675 for (vg = RW_MGR_MEM_VIRTUAL_GROUPS_PER_WRITE_DQS-1; ; vg--) {
2676 /* reset the fifos to get pointers to known state */
2677 writel(0, &phy_mgr_cmd->fifo_reset);
2678
2679 tmp_bit_chk = tmp_bit_chk <<
2680 (RW_MGR_MEM_DQ_PER_WRITE_DQS /
2681 RW_MGR_MEM_VIRTUAL_GROUPS_PER_WRITE_DQS);
2682 rw_mgr_mem_calibrate_write_test_issue(write_group *
2683 RW_MGR_MEM_VIRTUAL_GROUPS_PER_WRITE_DQS+vg,
2684 use_dm);
2685
2686 base_rw_mgr = readl(addr_rw_mgr);
2687 tmp_bit_chk = tmp_bit_chk | (correct_mask_vg & ~(base_rw_mgr));
2688 if (vg == 0)
2689 break;
2690 }
2691 *bit_chk &= tmp_bit_chk;
2692 }
2693
2694 if (all_correct) {
2695 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
2696 debug_cond(DLEVEL == 2, "write_test(%u,%u,ALL) : %u == \
2697 %u => %lu", write_group, use_dm,
2698 *bit_chk, param->write_correct_mask,
2699 (long unsigned int)(*bit_chk ==
2700 param->write_correct_mask));
2701 return *bit_chk == param->write_correct_mask;
2702 } else {
2703 set_rank_and_odt_mask(0, RW_MGR_ODT_MODE_OFF);
2704 debug_cond(DLEVEL == 2, "write_test(%u,%u,ONE) : %u != ",
2705 write_group, use_dm, *bit_chk);
2706 debug_cond(DLEVEL == 2, "%lu" " => %lu", (long unsigned int)0,
2707 (long unsigned int)(*bit_chk != 0));
2708 return *bit_chk != 0x00;
2709 }
2710 }
2711
2712 /*
2713 * center all windows. do per-bit-deskew to possibly increase size of
2714 * certain windows.
2715 */
2716 static uint32_t rw_mgr_mem_calibrate_writes_center(uint32_t rank_bgn,
2717 uint32_t write_group, uint32_t test_bgn)
2718 {
2719 uint32_t i, p, min_index;
2720 int32_t d;
2721 /*
2722 * Store these as signed since there are comparisons with
2723 * signed numbers.
2724 */
2725 uint32_t bit_chk;
2726 uint32_t sticky_bit_chk;
2727 int32_t left_edge[RW_MGR_MEM_DQ_PER_WRITE_DQS];
2728 int32_t right_edge[RW_MGR_MEM_DQ_PER_WRITE_DQS];
2729 int32_t mid;
2730 int32_t mid_min, orig_mid_min;
2731 int32_t new_dqs, start_dqs, shift_dq;
2732 int32_t dq_margin, dqs_margin, dm_margin;
2733 uint32_t stop;
2734 uint32_t temp_dq_out1_delay;
2735 uint32_t addr;
2736
2737 debug("%s:%d %u %u", __func__, __LINE__, write_group, test_bgn);
2738
2739 dm_margin = 0;
2740
2741 addr = SDR_PHYGRP_SCCGRP_ADDRESS | SCC_MGR_IO_OUT1_DELAY_OFFSET;
2742 start_dqs = readl(addr +
2743 (RW_MGR_MEM_DQ_PER_WRITE_DQS << 2));
2744
2745 /* per-bit deskew */
2746
2747 /*
2748 * set the left and right edge of each bit to an illegal value
2749 * use (IO_IO_OUT1_DELAY_MAX + 1) as an illegal value.
2750 */
2751 sticky_bit_chk = 0;
2752 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2753 left_edge[i] = IO_IO_OUT1_DELAY_MAX + 1;
2754 right_edge[i] = IO_IO_OUT1_DELAY_MAX + 1;
2755 }
2756
2757 /* Search for the left edge of the window for each bit */
2758 for (d = 0; d <= IO_IO_OUT1_DELAY_MAX; d++) {
2759 scc_mgr_apply_group_dq_out1_delay(write_group, d);
2760
2761 writel(0, &sdr_scc_mgr->update);
2762
2763 /*
2764 * Stop searching when the read test doesn't pass AND when
2765 * we've seen a passing read on every bit.
2766 */
2767 stop = !rw_mgr_mem_calibrate_write_test(rank_bgn, write_group,
2768 0, PASS_ONE_BIT, &bit_chk, 0);
2769 sticky_bit_chk = sticky_bit_chk | bit_chk;
2770 stop = stop && (sticky_bit_chk == param->write_correct_mask);
2771 debug_cond(DLEVEL == 2, "write_center(left): dtap=%d => %u \
2772 == %u && %u [bit_chk= %u ]\n",
2773 d, sticky_bit_chk, param->write_correct_mask,
2774 stop, bit_chk);
2775
2776 if (stop == 1) {
2777 break;
2778 } else {
2779 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2780 if (bit_chk & 1) {
2781 /*
2782 * Remember a passing test as the
2783 * left_edge.
2784 */
2785 left_edge[i] = d;
2786 } else {
2787 /*
2788 * If a left edge has not been seen
2789 * yet, then a future passing test will
2790 * mark this edge as the right edge.
2791 */
2792 if (left_edge[i] ==
2793 IO_IO_OUT1_DELAY_MAX + 1) {
2794 right_edge[i] = -(d + 1);
2795 }
2796 }
2797 debug_cond(DLEVEL == 2, "write_center[l,d=%d):", d);
2798 debug_cond(DLEVEL == 2, "bit_chk_test=%d left_edge[%u]: %d",
2799 (int)(bit_chk & 1), i, left_edge[i]);
2800 debug_cond(DLEVEL == 2, "right_edge[%u]: %d\n", i,
2801 right_edge[i]);
2802 bit_chk = bit_chk >> 1;
2803 }
2804 }
2805 }
2806
2807 /* Reset DQ delay chains to 0 */
2808 scc_mgr_apply_group_dq_out1_delay(0);
2809 sticky_bit_chk = 0;
2810 for (i = RW_MGR_MEM_DQ_PER_WRITE_DQS - 1;; i--) {
2811 debug_cond(DLEVEL == 2, "%s:%d write_center: left_edge[%u]: \
2812 %d right_edge[%u]: %d\n", __func__, __LINE__,
2813 i, left_edge[i], i, right_edge[i]);
2814
2815 /*
2816 * Check for cases where we haven't found the left edge,
2817 * which makes our assignment of the the right edge invalid.
2818 * Reset it to the illegal value.
2819 */
2820 if ((left_edge[i] == IO_IO_OUT1_DELAY_MAX + 1) &&
2821 (right_edge[i] != IO_IO_OUT1_DELAY_MAX + 1)) {
2822 right_edge[i] = IO_IO_OUT1_DELAY_MAX + 1;
2823 debug_cond(DLEVEL == 2, "%s:%d write_center: reset \
2824 right_edge[%u]: %d\n", __func__, __LINE__,
2825 i, right_edge[i]);
2826 }
2827
2828 /*
2829 * Reset sticky bit (except for bits where we have
2830 * seen the left edge).
2831 */
2832 sticky_bit_chk = sticky_bit_chk << 1;
2833 if ((left_edge[i] != IO_IO_OUT1_DELAY_MAX + 1))
2834 sticky_bit_chk = sticky_bit_chk | 1;
2835
2836 if (i == 0)
2837 break;
2838 }
2839
2840 /* Search for the right edge of the window for each bit */
2841 for (d = 0; d <= IO_IO_OUT1_DELAY_MAX - start_dqs; d++) {
2842 scc_mgr_apply_group_dqs_io_and_oct_out1(write_group,
2843 d + start_dqs);
2844
2845 writel(0, &sdr_scc_mgr->update);
2846
2847 /*
2848 * Stop searching when the read test doesn't pass AND when
2849 * we've seen a passing read on every bit.
2850 */
2851 stop = !rw_mgr_mem_calibrate_write_test(rank_bgn, write_group,
2852 0, PASS_ONE_BIT, &bit_chk, 0);
2853
2854 sticky_bit_chk = sticky_bit_chk | bit_chk;
2855 stop = stop && (sticky_bit_chk == param->write_correct_mask);
2856
2857 debug_cond(DLEVEL == 2, "write_center (right): dtap=%u => %u == \
2858 %u && %u\n", d, sticky_bit_chk,
2859 param->write_correct_mask, stop);
2860
2861 if (stop == 1) {
2862 if (d == 0) {
2863 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS;
2864 i++) {
2865 /* d = 0 failed, but it passed when
2866 testing the left edge, so it must be
2867 marginal, set it to -1 */
2868 if (right_edge[i] ==
2869 IO_IO_OUT1_DELAY_MAX + 1 &&
2870 left_edge[i] !=
2871 IO_IO_OUT1_DELAY_MAX + 1) {
2872 right_edge[i] = -1;
2873 }
2874 }
2875 }
2876 break;
2877 } else {
2878 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2879 if (bit_chk & 1) {
2880 /*
2881 * Remember a passing test as
2882 * the right_edge.
2883 */
2884 right_edge[i] = d;
2885 } else {
2886 if (d != 0) {
2887 /*
2888 * If a right edge has not
2889 * been seen yet, then a future
2890 * passing test will mark this
2891 * edge as the left edge.
2892 */
2893 if (right_edge[i] ==
2894 IO_IO_OUT1_DELAY_MAX + 1)
2895 left_edge[i] = -(d + 1);
2896 } else {
2897 /*
2898 * d = 0 failed, but it passed
2899 * when testing the left edge,
2900 * so it must be marginal, set
2901 * it to -1.
2902 */
2903 if (right_edge[i] ==
2904 IO_IO_OUT1_DELAY_MAX + 1 &&
2905 left_edge[i] !=
2906 IO_IO_OUT1_DELAY_MAX + 1)
2907 right_edge[i] = -1;
2908 /*
2909 * If a right edge has not been
2910 * seen yet, then a future
2911 * passing test will mark this
2912 * edge as the left edge.
2913 */
2914 else if (right_edge[i] ==
2915 IO_IO_OUT1_DELAY_MAX +
2916 1)
2917 left_edge[i] = -(d + 1);
2918 }
2919 }
2920 debug_cond(DLEVEL == 2, "write_center[r,d=%d):", d);
2921 debug_cond(DLEVEL == 2, "bit_chk_test=%d left_edge[%u]: %d",
2922 (int)(bit_chk & 1), i, left_edge[i]);
2923 debug_cond(DLEVEL == 2, "right_edge[%u]: %d\n", i,
2924 right_edge[i]);
2925 bit_chk = bit_chk >> 1;
2926 }
2927 }
2928 }
2929
2930 /* Check that all bits have a window */
2931 for (i = 0; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2932 debug_cond(DLEVEL == 2, "%s:%d write_center: left_edge[%u]: \
2933 %d right_edge[%u]: %d", __func__, __LINE__,
2934 i, left_edge[i], i, right_edge[i]);
2935 if ((left_edge[i] == IO_IO_OUT1_DELAY_MAX + 1) ||
2936 (right_edge[i] == IO_IO_OUT1_DELAY_MAX + 1)) {
2937 set_failing_group_stage(test_bgn + i,
2938 CAL_STAGE_WRITES,
2939 CAL_SUBSTAGE_WRITES_CENTER);
2940 return 0;
2941 }
2942 }
2943
2944 /* Find middle of window for each DQ bit */
2945 mid_min = left_edge[0] - right_edge[0];
2946 min_index = 0;
2947 for (i = 1; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++) {
2948 mid = left_edge[i] - right_edge[i];
2949 if (mid < mid_min) {
2950 mid_min = mid;
2951 min_index = i;
2952 }
2953 }
2954
2955 /*
2956 * -mid_min/2 represents the amount that we need to move DQS.
2957 * If mid_min is odd and positive we'll need to add one to
2958 * make sure the rounding in further calculations is correct
2959 * (always bias to the right), so just add 1 for all positive values.
2960 */
2961 if (mid_min > 0)
2962 mid_min++;
2963 mid_min = mid_min / 2;
2964 debug_cond(DLEVEL == 1, "%s:%d write_center: mid_min=%d\n", __func__,
2965 __LINE__, mid_min);
2966
2967 /* Determine the amount we can change DQS (which is -mid_min) */
2968 orig_mid_min = mid_min;
2969 new_dqs = start_dqs;
2970 mid_min = 0;
2971 debug_cond(DLEVEL == 1, "%s:%d write_center: start_dqs=%d new_dqs=%d \
2972 mid_min=%d\n", __func__, __LINE__, start_dqs, new_dqs, mid_min);
2973 /* Initialize data for export structures */
2974 dqs_margin = IO_IO_OUT1_DELAY_MAX + 1;
2975 dq_margin = IO_IO_OUT1_DELAY_MAX + 1;
2976
2977 /* add delay to bring centre of all DQ windows to the same "level" */
2978 for (i = 0, p = test_bgn; i < RW_MGR_MEM_DQ_PER_WRITE_DQS; i++, p++) {
2979 /* Use values before divide by 2 to reduce round off error */
2980 shift_dq = (left_edge[i] - right_edge[i] -
2981 (left_edge[min_index] - right_edge[min_index]))/2 +
2982 (orig_mid_min - mid_min);
2983
2984 debug_cond(DLEVEL == 2, "%s:%d write_center: before: shift_dq \
2985 [%u]=%d\n", __func__, __LINE__, i, shift_dq);
2986
2987 addr = SDR_PHYGRP_SCCGRP_ADDRESS | SCC_MGR_IO_OUT1_DELAY_OFFSET;
2988 temp_dq_out1_delay = readl(addr + (i << 2));
2989 if (shift_dq + (int32_t)temp_dq_out1_delay >
2990 (int32_t)IO_IO_OUT1_DELAY_MAX) {
2991 shift_dq = (int32_t)IO_IO_OUT1_DELAY_MAX - temp_dq_out1_delay;
2992 } else if (shift_dq + (int32_t)temp_dq_out1_delay < 0) {
2993 shift_dq = -(int32_t)temp_dq_out1_delay;
2994 }
2995 debug_cond(DLEVEL == 2, "write_center: after: shift_dq[%u]=%d\n",
2996 i, shift_dq);
2997 scc_mgr_set_dq_out1_delay(i, temp_dq_out1_delay + shift_dq);
2998 scc_mgr_load_dq(i);
2999
3000 debug_cond(DLEVEL == 2, "write_center: margin[%u]=[%d,%d]\n", i,
3001 left_edge[i] - shift_dq + (-mid_min),
3002 right_edge[i] + shift_dq - (-mid_min));
3003 /* To determine values for export structures */
3004 if (left_edge[i] - shift_dq + (-mid_min) < dq_margin)
3005 dq_margin = left_edge[i] - shift_dq + (-mid_min);
3006
3007 if (right_edge[i] + shift_dq - (-mid_min) < dqs_margin)
3008 dqs_margin = right_edge[i] + shift_dq - (-mid_min);
3009 }
3010
3011 /* Move DQS */
3012 scc_mgr_apply_group_dqs_io_and_oct_out1(write_group, new_dqs);
3013 writel(0, &sdr_scc_mgr->update);
3014
3015 /* Centre DM */
3016 debug_cond(DLEVEL == 2, "%s:%d write_center: DM\n", __func__, __LINE__);
3017
3018 /*
3019 * set the left and right edge of each bit to an illegal value,
3020 * use (IO_IO_OUT1_DELAY_MAX + 1) as an illegal value,
3021 */
3022 left_edge[0] = IO_IO_OUT1_DELAY_MAX + 1;
3023 right_edge[0] = IO_IO_OUT1_DELAY_MAX + 1;
3024 int32_t bgn_curr = IO_IO_OUT1_DELAY_MAX + 1;
3025 int32_t end_curr = IO_IO_OUT1_DELAY_MAX + 1;
3026 int32_t bgn_best = IO_IO_OUT1_DELAY_MAX + 1;
3027 int32_t end_best = IO_IO_OUT1_DELAY_MAX + 1;
3028 int32_t win_best = 0;
3029
3030 /* Search for the/part of the window with DM shift */
3031 for (d = IO_IO_OUT1_DELAY_MAX; d >= 0; d -= DELTA_D) {
3032 scc_mgr_apply_group_dm_out1_delay(d);
3033 writel(0, &sdr_scc_mgr->update);
3034
3035 if (rw_mgr_mem_calibrate_write_test(rank_bgn, write_group, 1,
3036 PASS_ALL_BITS, &bit_chk,
3037 0)) {
3038 /* USE Set current end of the window */
3039 end_curr = -d;
3040 /*
3041 * If a starting edge of our window has not been seen
3042 * this is our current start of the DM window.
3043 */
3044 if (bgn_curr == IO_IO_OUT1_DELAY_MAX + 1)
3045 bgn_curr = -d;
3046
3047 /*
3048 * If current window is bigger than best seen.
3049 * Set best seen to be current window.
3050 */
3051 if ((end_curr-bgn_curr+1) > win_best) {
3052 win_best = end_curr-bgn_curr+1;
3053 bgn_best = bgn_curr;
3054 end_best = end_curr;
3055 }
3056 } else {
3057 /* We just saw a failing test. Reset temp edge */
3058 bgn_curr = IO_IO_OUT1_DELAY_MAX + 1;
3059 end_curr = IO_IO_OUT1_DELAY_MAX + 1;
3060 }
3061 }
3062
3063
3064 /* Reset DM delay chains to 0 */
3065 scc_mgr_apply_group_dm_out1_delay(0);
3066
3067 /*
3068 * Check to see if the current window nudges up aganist 0 delay.
3069 * If so we need to continue the search by shifting DQS otherwise DQS
3070 * search begins as a new search. */
3071 if (end_curr != 0) {
3072 bgn_curr = IO_IO_OUT1_DELAY_MAX + 1;
3073 end_curr = IO_IO_OUT1_DELAY_MAX + 1;
3074 }
3075
3076 /* Search for the/part of the window with DQS shifts */
3077 for (d = 0; d <= IO_IO_OUT1_DELAY_MAX - new_dqs; d += DELTA_D) {
3078 /*
3079 * Note: This only shifts DQS, so are we limiting ourselve to
3080 * width of DQ unnecessarily.
3081 */
3082 scc_mgr_apply_group_dqs_io_and_oct_out1(write_group,
3083 d + new_dqs);
3084
3085 writel(0, &sdr_scc_mgr->update);
3086 if (rw_mgr_mem_calibrate_write_test(rank_bgn, write_group, 1,
3087 PASS_ALL_BITS, &bit_chk,
3088 0)) {
3089 /* USE Set current end of the window */
3090 end_curr = d;
3091 /*
3092 * If a beginning edge of our window has not been seen
3093 * this is our current begin of the DM window.
3094 */
3095 if (bgn_curr == IO_IO_OUT1_DELAY_MAX + 1)
3096 bgn_curr = d;
3097
3098 /*
3099 * If current window is bigger than best seen. Set best
3100 * seen to be current window.
3101 */
3102 if ((end_curr-bgn_curr+1) > win_best) {
3103 win_best = end_curr-bgn_curr+1;
3104 bgn_best = bgn_curr;
3105 end_best = end_curr;
3106 }
3107 } else {
3108 /* We just saw a failing test. Reset temp edge */
3109 bgn_curr = IO_IO_OUT1_DELAY_MAX + 1;
3110 end_curr = IO_IO_OUT1_DELAY_MAX + 1;
3111
3112 /* Early exit optimization: if ther remaining delay
3113 chain space is less than already seen largest window
3114 we can exit */
3115 if ((win_best-1) >
3116 (IO_IO_OUT1_DELAY_MAX - new_dqs - d)) {
3117 break;
3118 }
3119 }
3120 }
3121
3122 /* assign left and right edge for cal and reporting; */
3123 left_edge[0] = -1*bgn_best;
3124 right_edge[0] = end_best;
3125
3126 debug_cond(DLEVEL == 2, "%s:%d dm_calib: left=%d right=%d\n", __func__,
3127 __LINE__, left_edge[0], right_edge[0]);
3128
3129 /* Move DQS (back to orig) */
3130 scc_mgr_apply_group_dqs_io_and_oct_out1(write_group, new_dqs);
3131
3132 /* Move DM */
3133
3134 /* Find middle of window for the DM bit */
3135 mid = (left_edge[0] - right_edge[0]) / 2;
3136
3137 /* only move right, since we are not moving DQS/DQ */
3138 if (mid < 0)
3139 mid = 0;
3140
3141 /* dm_marign should fail if we never find a window */
3142 if (win_best == 0)
3143 dm_margin = -1;
3144 else
3145 dm_margin = left_edge[0] - mid;
3146
3147 scc_mgr_apply_group_dm_out1_delay(mid);
3148 writel(0, &sdr_scc_mgr->update);
3149
3150 debug_cond(DLEVEL == 2, "%s:%d dm_calib: left=%d right=%d mid=%d \
3151 dm_margin=%d\n", __func__, __LINE__, left_edge[0],
3152 right_edge[0], mid, dm_margin);
3153 /* Export values */
3154 gbl->fom_out += dq_margin + dqs_margin;
3155
3156 debug_cond(DLEVEL == 2, "%s:%d write_center: dq_margin=%d \
3157 dqs_margin=%d dm_margin=%d\n", __func__, __LINE__,
3158 dq_margin, dqs_margin, dm_margin);
3159
3160 /*
3161 * Do not remove this line as it makes sure all of our
3162 * decisions have been applied.
3163 */
3164 writel(0, &sdr_scc_mgr->update);
3165 return (dq_margin >= 0) && (dqs_margin >= 0) && (dm_margin >= 0);
3166 }
3167
3168 /* calibrate the write operations */
3169 static uint32_t rw_mgr_mem_calibrate_writes(uint32_t rank_bgn, uint32_t g,
3170 uint32_t test_bgn)
3171 {
3172 /* update info for sims */
3173 debug("%s:%d %u %u\n", __func__, __LINE__, g, test_bgn);
3174
3175 reg_file_set_stage(CAL_STAGE_WRITES);
3176 reg_file_set_sub_stage(CAL_SUBSTAGE_WRITES_CENTER);
3177
3178 reg_file_set_group(g);
3179
3180 if (!rw_mgr_mem_calibrate_writes_center(rank_bgn, g, test_bgn)) {
3181 set_failing_group_stage(g, CAL_STAGE_WRITES,
3182 CAL_SUBSTAGE_WRITES_CENTER);
3183 return 0;
3184 }
3185
3186 return 1;
3187 }
3188
3189 /**
3190 * mem_precharge_and_activate() - Precharge all banks and activate
3191 *
3192 * Precharge all banks and activate row 0 in bank "000..." and bank "111...".
3193 */
3194 static void mem_precharge_and_activate(void)
3195 {
3196 int r;
3197
3198 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS; r++) {
3199 /* Test if the rank should be skipped. */
3200 if (param->skip_ranks[r])
3201 continue;
3202
3203 /* Set rank. */
3204 set_rank_and_odt_mask(r, RW_MGR_ODT_MODE_OFF);
3205
3206 /* Precharge all banks. */
3207 writel(RW_MGR_PRECHARGE_ALL, SDR_PHYGRP_RWMGRGRP_ADDRESS |
3208 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
3209
3210 writel(0x0F, &sdr_rw_load_mgr_regs->load_cntr0);
3211 writel(RW_MGR_ACTIVATE_0_AND_1_WAIT1,
3212 &sdr_rw_load_jump_mgr_regs->load_jump_add0);
3213
3214 writel(0x0F, &sdr_rw_load_mgr_regs->load_cntr1);
3215 writel(RW_MGR_ACTIVATE_0_AND_1_WAIT2,
3216 &sdr_rw_load_jump_mgr_regs->load_jump_add1);
3217
3218 /* Activate rows. */
3219 writel(RW_MGR_ACTIVATE_0_AND_1, SDR_PHYGRP_RWMGRGRP_ADDRESS |
3220 RW_MGR_RUN_SINGLE_GROUP_OFFSET);
3221 }
3222 }
3223
3224 /**
3225 * mem_init_latency() - Configure memory RLAT and WLAT settings
3226 *
3227 * Configure memory RLAT and WLAT parameters.
3228 */
3229 static void mem_init_latency(void)
3230 {
3231 /*
3232 * For AV/CV, LFIFO is hardened and always runs at full rate
3233 * so max latency in AFI clocks, used here, is correspondingly
3234 * smaller.
3235 */
3236 const u32 max_latency = (1 << MAX_LATENCY_COUNT_WIDTH) - 1;
3237 u32 rlat, wlat;
3238
3239 debug("%s:%d\n", __func__, __LINE__);
3240
3241 /*
3242 * Read in write latency.
3243 * WL for Hard PHY does not include additive latency.
3244 */
3245 wlat = readl(&data_mgr->t_wl_add);
3246 wlat += readl(&data_mgr->mem_t_add);
3247
3248 gbl->rw_wl_nop_cycles = wlat - 1;
3249
3250 /* Read in readl latency. */
3251 rlat = readl(&data_mgr->t_rl_add);
3252
3253 /* Set a pretty high read latency initially. */
3254 gbl->curr_read_lat = rlat + 16;
3255 if (gbl->curr_read_lat > max_latency)
3256 gbl->curr_read_lat = max_latency;
3257
3258 writel(gbl->curr_read_lat, &phy_mgr_cfg->phy_rlat);
3259
3260 /* Advertise write latency. */
3261 writel(wlat, &phy_mgr_cfg->afi_wlat);
3262 }
3263
3264 /**
3265 * @mem_skip_calibrate() - Set VFIFO and LFIFO to instant-on settings
3266 *
3267 * Set VFIFO and LFIFO to instant-on settings in skip calibration mode.
3268 */
3269 static void mem_skip_calibrate(void)
3270 {
3271 uint32_t vfifo_offset;
3272 uint32_t i, j, r;
3273
3274 debug("%s:%d\n", __func__, __LINE__);
3275 /* Need to update every shadow register set used by the interface */
3276 for (r = 0; r < RW_MGR_MEM_NUMBER_OF_RANKS;
3277 r += NUM_RANKS_PER_SHADOW_REG) {
3278 /*
3279 * Set output phase alignment settings appropriate for
3280 * skip calibration.
3281 */
3282 for (i = 0; i < RW_MGR_MEM_IF_READ_DQS_WIDTH; i++) {
3283 scc_mgr_set_dqs_en_phase(i, 0);
3284 #if IO_DLL_CHAIN_LENGTH == 6
3285 scc_mgr_set_dqdqs_output_phase(i, 6);
3286 #else
3287 scc_mgr_set_dqdqs_output_phase(i, 7);
3288 #endif
3289 /*
3290 * Case:33398
3291 *
3292 * Write data arrives to the I/O two cycles before write
3293 * latency is reached (720 deg).
3294 * -> due to bit-slip in a/c bus
3295 * -> to allow board skew where dqs is longer than ck
3296 * -> how often can this happen!?
3297 * -> can claim back some ptaps for high freq
3298 * support if we can relax this, but i digress...
3299 *
3300 * The write_clk leads mem_ck by 90 deg
3301 * The minimum ptap of the OPA is 180 deg
3302 * Each ptap has (360 / IO_DLL_CHAIN_LENGH) deg of delay
3303 * The write_clk is always delayed by 2 ptaps
3304 *
3305 * Hence, to make DQS aligned to CK, we need to delay
3306 * DQS by:
3307 * (720 - 90 - 180 - 2 * (360 / IO_DLL_CHAIN_LENGTH))
3308 *
3309 * Dividing the above by (360 / IO_DLL_CHAIN_LENGTH)
3310 * gives us the number of ptaps, which simplies to:
3311 *
3312 * (1.25 * IO_DLL_CHAIN_LENGTH - 2)
3313 */
3314 scc_mgr_set_dqdqs_output_phase(i,
3315 1.25 * IO_DLL_CHAIN_LENGTH - 2);
3316 }
3317 writel(0xff, &sdr_scc_mgr->dqs_ena);
3318 writel(0xff, &sdr_scc_mgr->dqs_io_ena);
3319
3320 for (i = 0; i < RW_MGR_MEM_IF_WRITE_DQS_WIDTH; i++) {
3321 writel(i, SDR_PHYGRP_SCCGRP_ADDRESS |
3322 SCC_MGR_GROUP_COUNTER_OFFSET);
3323 }
3324 writel(0xff, &sdr_scc_mgr->dq_ena);
3325 writel(0xff, &sdr_scc_mgr->dm_ena);
3326 writel(0, &sdr_scc_mgr->update);
3327 }
3328
3329 /* Compensate for simulation model behaviour */
3330 for (i = 0; i < RW_MGR_MEM_IF_READ_DQS_WIDTH; i++) {
3331 scc_mgr_set_dqs_bus_in_delay(i, 10);
3332 scc_mgr_load_dqs(i);
3333 }
3334 writel(0, &sdr_scc_mgr->update);
3335
3336 /*
3337 * ArriaV has hard FIFOs that can only be initialized by incrementing
3338 * in sequencer.
3339 */
3340 vfifo_offset = CALIB_VFIFO_OFFSET;
3341 for (j = 0; j < vfifo_offset; j++)
3342 writel(0xff, &phy_mgr_cmd->inc_vfifo_hard_phy);
3343 writel(0, &phy_mgr_cmd->fifo_reset);
3344
3345 /*
3346 * For Arria V and Cyclone V with hard LFIFO, we get the skip-cal
3347 * setting from generation-time constant.
3348 */
3349 gbl->curr_read_lat = CALIB_LFIFO_OFFSET;
3350 writel(gbl->curr_read_lat, &phy_mgr_cfg->phy_rlat);
3351 }
3352
3353 /**
3354 * mem_calibrate() - Memory calibration entry point.
3355 *
3356 * Perform memory calibration.
3357 */
3358 static uint32_t mem_calibrate(void)
3359 {
3360 uint32_t i;
3361 uint32_t rank_bgn, sr;
3362 uint32_t write_group, write_test_bgn;
3363 uint32_t read_group, read_test_bgn;
3364 uint32_t run_groups, current_run;
3365 uint32_t failing_groups = 0;
3366 uint32_t group_failed = 0;
3367
3368 const u32 rwdqs_ratio = RW_MGR_MEM_IF_READ_DQS_WIDTH /
3369 RW_MGR_MEM_IF_WRITE_DQS_WIDTH;
3370
3371 debug("%s:%d\n", __func__, __LINE__);
3372
3373 /* Initialize the data settings */
3374 gbl->error_substage = CAL_SUBSTAGE_NIL;
3375 gbl->error_stage = CAL_STAGE_NIL;
3376 gbl->error_group = 0xff;
3377 gbl->fom_in = 0;
3378 gbl->fom_out = 0;
3379
3380 /* Initialize WLAT and RLAT. */
3381 mem_init_latency();
3382
3383 /* Initialize bit slips. */
3384 mem_precharge_and_activate();
3385
3386 for (i = 0; i < RW_MGR_MEM_IF_READ_DQS_WIDTH; i++) {
3387 writel(i, SDR_PHYGRP_SCCGRP_ADDRESS |
3388 SCC_MGR_GROUP_COUNTER_OFFSET);
3389 /* Only needed once to set all groups, pins, DQ, DQS, DM. */
3390 if (i == 0)
3391 scc_mgr_set_hhp_extras();
3392
3393 scc_set_bypass_mode(i);
3394 }
3395
3396 /* Calibration is skipped. */
3397 if ((dyn_calib_steps & CALIB_SKIP_ALL) == CALIB_SKIP_ALL) {
3398 /*
3399 * Set VFIFO and LFIFO to instant-on settings in skip
3400 * calibration mode.
3401 */
3402 mem_skip_calibrate();
3403
3404 /*
3405 * Do not remove this line as it makes sure all of our
3406 * decisions have been applied.
3407 */
3408 writel(0, &sdr_scc_mgr->update);
3409 return 1;
3410 }
3411
3412 /* Calibration is not skipped. */
3413 for (i = 0; i < NUM_CALIB_REPEAT; i++) {
3414 /*
3415 * Zero all delay chain/phase settings for all
3416 * groups and all shadow register sets.
3417 */
3418 scc_mgr_zero_all();
3419
3420 run_groups = ~param->skip_groups;
3421
3422 for (write_group = 0, write_test_bgn = 0; write_group
3423 < RW_MGR_MEM_IF_WRITE_DQS_WIDTH; write_group++,
3424 write_test_bgn += RW_MGR_MEM_DQ_PER_WRITE_DQS) {
3425
3426 /* Initialize the group failure */
3427 group_failed = 0;
3428
3429 current_run = run_groups & ((1 <<
3430 RW_MGR_NUM_DQS_PER_WRITE_GROUP) - 1);
3431 run_groups = run_groups >>
3432 RW_MGR_NUM_DQS_PER_WRITE_GROUP;
3433
3434 if (current_run == 0)
3435 continue;
3436
3437 writel(write_group, SDR_PHYGRP_SCCGRP_ADDRESS |
3438 SCC_MGR_GROUP_COUNTER_OFFSET);
3439 scc_mgr_zero_group(write_group, 0);
3440
3441 for (read_group = write_group * rwdqs_ratio,
3442 read_test_bgn = 0;
3443 read_group < (write_group + 1) * rwdqs_ratio;
3444 read_group++,
3445 read_test_bgn += RW_MGR_MEM_DQ_PER_READ_DQS) {
3446 if (STATIC_CALIB_STEPS & CALIB_SKIP_VFIFO)
3447 continue;
3448
3449 /* Calibrate the VFIFO */
3450 if (rw_mgr_mem_calibrate_vfifo(read_group,
3451 read_test_bgn))
3452 continue;
3453
3454 if (!(gbl->phy_debug_mode_flags & PHY_DEBUG_SWEEP_ALL_GROUPS))
3455 return 0;
3456
3457 /* The group failed, we're done. */
3458 goto grp_failed;
3459 }
3460
3461 /* Calibrate the output side */
3462 for (rank_bgn = 0, sr = 0;
3463 rank_bgn < RW_MGR_MEM_NUMBER_OF_RANKS;
3464 rank_bgn += NUM_RANKS_PER_SHADOW_REG, sr++) {
3465 if (STATIC_CALIB_STEPS & CALIB_SKIP_WRITES)
3466 continue;
3467
3468 /* Not needed in quick mode! */
3469 if (STATIC_CALIB_STEPS & CALIB_SKIP_DELAY_SWEEPS)
3470 continue;
3471
3472 /*
3473 * Determine if this set of ranks
3474 * should be skipped entirely.
3475 */
3476 if (param->skip_shadow_regs[sr])
3477 continue;
3478
3479 /* Calibrate WRITEs */
3480 if (rw_mgr_mem_calibrate_writes(rank_bgn,
3481 write_group, write_test_bgn))
3482 continue;
3483
3484 group_failed = 1;
3485 if (!(gbl->phy_debug_mode_flags & PHY_DEBUG_SWEEP_ALL_GROUPS))
3486 return 0;
3487 }
3488
3489 /* Some group failed, we're done. */
3490 if (group_failed)
3491 goto grp_failed;
3492
3493 for (read_group = write_group * rwdqs_ratio,
3494 read_test_bgn = 0;
3495 read_group < (write_group + 1) * rwdqs_ratio;
3496 read_group++,
3497 read_test_bgn += RW_MGR_MEM_DQ_PER_READ_DQS) {
3498 if (STATIC_CALIB_STEPS & CALIB_SKIP_WRITES)
3499 continue;
3500
3501 if (rw_mgr_mem_calibrate_vfifo_end(read_group,
3502 read_test_bgn))
3503 continue;
3504
3505 if (!(gbl->phy_debug_mode_flags & PHY_DEBUG_SWEEP_ALL_GROUPS))
3506 return 0;
3507
3508 /* The group failed, we're done. */
3509 goto grp_failed;
3510 }
3511
3512 /* No group failed, continue as usual. */
3513 continue;
3514
3515 grp_failed: /* A group failed, increment the counter. */
3516 failing_groups++;
3517 }
3518
3519 /*
3520 * USER If there are any failing groups then report
3521 * the failure.
3522 */
3523 if (failing_groups != 0)
3524 return 0;
3525
3526 if (STATIC_CALIB_STEPS & CALIB_SKIP_LFIFO)
3527 continue;
3528
3529 /*
3530 * If we're skipping groups as part of debug,
3531 * don't calibrate LFIFO.
3532 */
3533 if (param->skip_groups != 0)
3534 continue;
3535
3536 /* Calibrate the LFIFO */
3537 if (!rw_mgr_mem_calibrate_lfifo())
3538 return 0;
3539 }
3540
3541 /*
3542 * Do not remove this line as it makes sure all of our decisions
3543 * have been applied.
3544 */
3545 writel(0, &sdr_scc_mgr->update);
3546 return 1;
3547 }
3548
3549 /**
3550 * run_mem_calibrate() - Perform memory calibration
3551 *
3552 * This function triggers the entire memory calibration procedure.
3553 */
3554 static int run_mem_calibrate(void)
3555 {
3556 int pass;
3557
3558 debug("%s:%d\n", __func__, __LINE__);
3559
3560 /* Reset pass/fail status shown on afi_cal_success/fail */
3561 writel(PHY_MGR_CAL_RESET, &phy_mgr_cfg->cal_status);
3562
3563 /* Stop tracking manager. */
3564 clrbits_le32(&sdr_ctrl->ctrl_cfg, 1 << 22);
3565
3566 phy_mgr_initialize();
3567 rw_mgr_mem_initialize();
3568
3569 /* Perform the actual memory calibration. */
3570 pass = mem_calibrate();
3571
3572 mem_precharge_and_activate();
3573 writel(0, &phy_mgr_cmd->fifo_reset);
3574
3575 /* Handoff. */
3576 rw_mgr_mem_handoff();
3577 /*
3578 * In Hard PHY this is a 2-bit control:
3579 * 0: AFI Mux Select
3580 * 1: DDIO Mux Select
3581 */
3582 writel(0x2, &phy_mgr_cfg->mux_sel);
3583
3584 /* Start tracking manager. */
3585 setbits_le32(&sdr_ctrl->ctrl_cfg, 1 << 22);
3586
3587 return pass;
3588 }
3589
3590 /**
3591 * debug_mem_calibrate() - Report result of memory calibration
3592 * @pass: Value indicating whether calibration passed or failed
3593 *
3594 * This function reports the results of the memory calibration
3595 * and writes debug information into the register file.
3596 */
3597 static void debug_mem_calibrate(int pass)
3598 {
3599 uint32_t debug_info;
3600
3601 if (pass) {
3602 printf("%s: CALIBRATION PASSED\n", __FILE__);
3603
3604 gbl->fom_in /= 2;
3605 gbl->fom_out /= 2;
3606
3607 if (gbl->fom_in > 0xff)
3608 gbl->fom_in = 0xff;
3609
3610 if (gbl->fom_out > 0xff)
3611 gbl->fom_out = 0xff;
3612
3613 /* Update the FOM in the register file */
3614 debug_info = gbl->fom_in;
3615 debug_info |= gbl->fom_out << 8;
3616 writel(debug_info, &sdr_reg_file->fom);
3617
3618 writel(debug_info, &phy_mgr_cfg->cal_debug_info);
3619 writel(PHY_MGR_CAL_SUCCESS, &phy_mgr_cfg->cal_status);
3620 } else {
3621 printf("%s: CALIBRATION FAILED\n", __FILE__);
3622
3623 debug_info = gbl->error_stage;
3624 debug_info |= gbl->error_substage << 8;
3625 debug_info |= gbl->error_group << 16;
3626
3627 writel(debug_info, &sdr_reg_file->failing_stage);
3628 writel(debug_info, &phy_mgr_cfg->cal_debug_info);
3629 writel(PHY_MGR_CAL_FAIL, &phy_mgr_cfg->cal_status);
3630
3631 /* Update the failing group/stage in the register file */
3632 debug_info = gbl->error_stage;
3633 debug_info |= gbl->error_substage << 8;
3634 debug_info |= gbl->error_group << 16;
3635 writel(debug_info, &sdr_reg_file->failing_stage);
3636 }
3637
3638 printf("%s: Calibration complete\n", __FILE__);
3639 }
3640
3641 /**
3642 * hc_initialize_rom_data() - Initialize ROM data
3643 *
3644 * Initialize ROM data.
3645 */
3646 static void hc_initialize_rom_data(void)
3647 {
3648 u32 i, addr;
3649
3650 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_INST_ROM_WRITE_OFFSET;
3651 for (i = 0; i < ARRAY_SIZE(inst_rom_init); i++)
3652 writel(inst_rom_init[i], addr + (i << 2));
3653
3654 addr = SDR_PHYGRP_RWMGRGRP_ADDRESS | RW_MGR_AC_ROM_WRITE_OFFSET;
3655 for (i = 0; i < ARRAY_SIZE(ac_rom_init); i++)
3656 writel(ac_rom_init[i], addr + (i << 2));
3657 }
3658
3659 /**
3660 * initialize_reg_file() - Initialize SDR register file
3661 *
3662 * Initialize SDR register file.
3663 */
3664 static void initialize_reg_file(void)
3665 {
3666 /* Initialize the register file with the correct data */
3667 writel(REG_FILE_INIT_SEQ_SIGNATURE, &sdr_reg_file->signature);
3668 writel(0, &sdr_reg_file->debug_data_addr);
3669 writel(0, &sdr_reg_file->cur_stage);
3670 writel(0, &sdr_reg_file->fom);
3671 writel(0, &sdr_reg_file->failing_stage);
3672 writel(0, &sdr_reg_file->debug1);
3673 writel(0, &sdr_reg_file->debug2);
3674 }
3675
3676 /**
3677 * initialize_hps_phy() - Initialize HPS PHY
3678 *
3679 * Initialize HPS PHY.
3680 */
3681 static void initialize_hps_phy(void)
3682 {
3683 uint32_t reg;
3684 /*
3685 * Tracking also gets configured here because it's in the
3686 * same register.
3687 */
3688 uint32_t trk_sample_count = 7500;
3689 uint32_t trk_long_idle_sample_count = (10 << 16) | 100;
3690 /*
3691 * Format is number of outer loops in the 16 MSB, sample
3692 * count in 16 LSB.
3693 */
3694
3695 reg = 0;
3696 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_ACDELAYEN_SET(2);
3697 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_DQDELAYEN_SET(1);
3698 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_DQSDELAYEN_SET(1);
3699 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_DQSLOGICDELAYEN_SET(1);
3700 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_RESETDELAYEN_SET(0);
3701 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_LPDDRDIS_SET(1);
3702 /*
3703 * This field selects the intrinsic latency to RDATA_EN/FULL path.
3704 * 00-bypass, 01- add 5 cycles, 10- add 10 cycles, 11- add 15 cycles.
3705 */
3706 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_ADDLATSEL_SET(0);
3707 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_SAMPLECOUNT_19_0_SET(
3708 trk_sample_count);
3709 writel(reg, &sdr_ctrl->phy_ctrl0);
3710
3711 reg = 0;
3712 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_1_SAMPLECOUNT_31_20_SET(
3713 trk_sample_count >>
3714 SDR_CTRLGRP_PHYCTRL_PHYCTRL_0_SAMPLECOUNT_19_0_WIDTH);
3715 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_1_LONGIDLESAMPLECOUNT_19_0_SET(
3716 trk_long_idle_sample_count);
3717 writel(reg, &sdr_ctrl->phy_ctrl1);
3718
3719 reg = 0;
3720 reg |= SDR_CTRLGRP_PHYCTRL_PHYCTRL_2_LONGIDLESAMPLECOUNT_31_20_SET(
3721 trk_long_idle_sample_count >>
3722 SDR_CTRLGRP_PHYCTRL_PHYCTRL_1_LONGIDLESAMPLECOUNT_19_0_WIDTH);
3723 writel(reg, &sdr_ctrl->phy_ctrl2);
3724 }
3725
3726 /**
3727 * initialize_tracking() - Initialize tracking
3728 *
3729 * Initialize the register file with usable initial data.
3730 */
3731 static void initialize_tracking(void)
3732 {
3733 /*
3734 * Initialize the register file with the correct data.
3735 * Compute usable version of value in case we skip full
3736 * computation later.
3737 */
3738 writel(DIV_ROUND_UP(IO_DELAY_PER_OPA_TAP, IO_DELAY_PER_DCHAIN_TAP) - 1,
3739 &sdr_reg_file->dtaps_per_ptap);
3740
3741 /* trk_sample_count */
3742 writel(7500, &sdr_reg_file->trk_sample_count);
3743
3744 /* longidle outer loop [15:0] */
3745 writel((10 << 16) | (100 << 0), &sdr_reg_file->trk_longidle);
3746
3747 /*
3748 * longidle sample count [31:24]
3749 * trfc, worst case of 933Mhz 4Gb [23:16]
3750 * trcd, worst case [15:8]
3751 * vfifo wait [7:0]
3752 */
3753 writel((243 << 24) | (14 << 16) | (10 << 8) | (4 << 0),
3754 &sdr_reg_file->delays);
3755
3756 /* mux delay */
3757 writel((RW_MGR_IDLE << 24) | (RW_MGR_ACTIVATE_1 << 16) |
3758 (RW_MGR_SGLE_READ << 8) | (RW_MGR_PRECHARGE_ALL << 0),
3759 &sdr_reg_file->trk_rw_mgr_addr);
3760
3761 writel(RW_MGR_MEM_IF_READ_DQS_WIDTH,
3762 &sdr_reg_file->trk_read_dqs_width);
3763
3764 /* trefi [7:0] */
3765 writel((RW_MGR_REFRESH_ALL << 24) | (1000 << 0),
3766 &sdr_reg_file->trk_rfsh);
3767 }
3768
3769 int sdram_calibration_full(void)
3770 {
3771 struct param_type my_param;
3772 struct gbl_type my_gbl;
3773 uint32_t pass;
3774
3775 memset(&my_param, 0, sizeof(my_param));
3776 memset(&my_gbl, 0, sizeof(my_gbl));
3777
3778 param = &my_param;
3779 gbl = &my_gbl;
3780
3781 /* Set the calibration enabled by default */
3782 gbl->phy_debug_mode_flags |= PHY_DEBUG_ENABLE_CAL_RPT;
3783 /*
3784 * Only sweep all groups (regardless of fail state) by default
3785 * Set enabled read test by default.
3786 */
3787 #if DISABLE_GUARANTEED_READ
3788 gbl->phy_debug_mode_flags |= PHY_DEBUG_DISABLE_GUARANTEED_READ;
3789 #endif
3790 /* Initialize the register file */
3791 initialize_reg_file();
3792
3793 /* Initialize any PHY CSR */
3794 initialize_hps_phy();
3795
3796 scc_mgr_initialize();
3797
3798 initialize_tracking();
3799
3800 printf("%s: Preparing to start memory calibration\n", __FILE__);
3801
3802 debug("%s:%d\n", __func__, __LINE__);
3803 debug_cond(DLEVEL == 1,
3804 "DDR3 FULL_RATE ranks=%u cs/dimm=%u dq/dqs=%u,%u vg/dqs=%u,%u ",
3805 RW_MGR_MEM_NUMBER_OF_RANKS, RW_MGR_MEM_NUMBER_OF_CS_PER_DIMM,
3806 RW_MGR_MEM_DQ_PER_READ_DQS, RW_MGR_MEM_DQ_PER_WRITE_DQS,
3807 RW_MGR_MEM_VIRTUAL_GROUPS_PER_READ_DQS,
3808 RW_MGR_MEM_VIRTUAL_GROUPS_PER_WRITE_DQS);
3809 debug_cond(DLEVEL == 1,
3810 "dqs=%u,%u dq=%u dm=%u ptap_delay=%u dtap_delay=%u ",
3811 RW_MGR_MEM_IF_READ_DQS_WIDTH, RW_MGR_MEM_IF_WRITE_DQS_WIDTH,
3812 RW_MGR_MEM_DATA_WIDTH, RW_MGR_MEM_DATA_MASK_WIDTH,
3813 IO_DELAY_PER_OPA_TAP, IO_DELAY_PER_DCHAIN_TAP);
3814 debug_cond(DLEVEL == 1, "dtap_dqsen_delay=%u, dll=%u",
3815 IO_DELAY_PER_DQS_EN_DCHAIN_TAP, IO_DLL_CHAIN_LENGTH);
3816 debug_cond(DLEVEL == 1, "max values: en_p=%u dqdqs_p=%u en_d=%u dqs_in_d=%u ",
3817 IO_DQS_EN_PHASE_MAX, IO_DQDQS_OUT_PHASE_MAX,
3818 IO_DQS_EN_DELAY_MAX, IO_DQS_IN_DELAY_MAX);
3819 debug_cond(DLEVEL == 1, "io_in_d=%u io_out1_d=%u io_out2_d=%u ",
3820 IO_IO_IN_DELAY_MAX, IO_IO_OUT1_DELAY_MAX,
3821 IO_IO_OUT2_DELAY_MAX);
3822 debug_cond(DLEVEL == 1, "dqs_in_reserve=%u dqs_out_reserve=%u\n",
3823 IO_DQS_IN_RESERVE, IO_DQS_OUT_RESERVE);
3824
3825 hc_initialize_rom_data();
3826
3827 /* update info for sims */
3828 reg_file_set_stage(CAL_STAGE_NIL);
3829 reg_file_set_group(0);
3830
3831 /*
3832 * Load global needed for those actions that require
3833 * some dynamic calibration support.
3834 */
3835 dyn_calib_steps = STATIC_CALIB_STEPS;
3836 /*
3837 * Load global to allow dynamic selection of delay loop settings
3838 * based on calibration mode.
3839 */
3840 if (!(dyn_calib_steps & CALIB_SKIP_DELAY_LOOPS))
3841 skip_delay_mask = 0xff;
3842 else
3843 skip_delay_mask = 0x0;
3844
3845 pass = run_mem_calibrate();
3846 debug_mem_calibrate(pass);
3847 return pass;
3848 }