2
* Copyright 2008-2014 Freescale Semiconductor, Inc.
4
* This program is free software; you can redistribute it and/or
5
* modify it under the terms of the GNU General Public License
6
* Version 2 as published by the Free Software Foundation.
10
#include <fsl_ddr_sdram.h>
14
#if defined(CONFIG_SYS_FSL_DDR3) || defined(CONFIG_SYS_FSL_DDR4)
16
compute_cas_latency(const dimm_params_t *dimm_params,
17
common_timing_params_t *outpdimm,
18
unsigned int number_of_dimms)
21
unsigned int common_caslat;
22
unsigned int caslat_actual;
23
unsigned int retry = 16;
25
const unsigned int mclk_ps = get_memory_clk_period_ps();
26
#ifdef CONFIG_SYS_FSL_DDR3
27
const unsigned int taamax = 20000;
29
const unsigned int taamax = 18000;
32
/* compute the common CAS latency supported between slots */
33
tmp = dimm_params[0].caslat_x;
34
for (i = 1; i < number_of_dimms; i++) {
35
if (dimm_params[i].n_ranks)
36
tmp &= dimm_params[i].caslat_x;
40
/* validate if the memory clk is in the range of dimms */
41
if (mclk_ps < outpdimm->tckmin_x_ps) {
42
printf("DDR clock (MCLK cycle %u ps) is faster than "
43
"the slowest DIMM(s) (tCKmin %u ps) can support.\n",
44
mclk_ps, outpdimm->tckmin_x_ps);
46
#ifdef CONFIG_SYS_FSL_DDR4
47
if (mclk_ps > outpdimm->tckmax_ps) {
48
printf("DDR clock (MCLK cycle %u ps) is slower than DIMM(s) (tCKmax %u ps) can support.\n",
49
mclk_ps, outpdimm->tckmax_ps);
52
/* determine the acutal cas latency */
53
caslat_actual = (outpdimm->taamin_ps + mclk_ps - 1) / mclk_ps;
54
/* check if the dimms support the CAS latency */
55
while (!(common_caslat & (1 << caslat_actual)) && retry > 0) {
59
/* once the caculation of caslat_actual is completed
60
* we must verify that this CAS latency value does not
61
* exceed tAAmax, which is 20 ns for all DDR3 speed grades,
62
* 18ns for all DDR4 speed grades.
64
if (caslat_actual * mclk_ps > taamax) {
65
printf("The choosen cas latency %d is too large\n",
68
outpdimm->lowest_common_spd_caslat = caslat_actual;
69
debug("lowest_common_spd_caslat is 0x%x\n", caslat_actual);
73
#else /* for DDR1 and DDR2 */
75
compute_cas_latency(const dimm_params_t *dimm_params,
76
common_timing_params_t *outpdimm,
77
unsigned int number_of_dimms)
80
const unsigned int mclk_ps = get_memory_clk_period_ps();
81
unsigned int lowest_good_caslat;
83
unsigned int temp1, temp2;
85
debug("using mclk_ps = %u\n", mclk_ps);
86
if (mclk_ps > outpdimm->tckmax_ps) {
87
printf("Warning: DDR clock (%u ps) is slower than DIMM(s) (tCKmax %u ps)\n",
88
mclk_ps, outpdimm->tckmax_ps);
92
* Compute a CAS latency suitable for all DIMMs
94
* Strategy for SPD-defined latencies: compute only
95
* CAS latency defined by all DIMMs.
99
* Step 1: find CAS latency common to all DIMMs using bitwise
103
for (i = 0; i < number_of_dimms; i++) {
104
if (dimm_params[i].n_ranks) {
106
temp2 |= 1 << dimm_params[i].caslat_x;
107
temp2 |= 1 << dimm_params[i].caslat_x_minus_1;
108
temp2 |= 1 << dimm_params[i].caslat_x_minus_2;
110
* If there was no entry for X-2 (X-1) in
111
* the SPD, then caslat_x_minus_2
112
* (caslat_x_minus_1) contains either 255 or
113
* 0xFFFFFFFF because that's what the glorious
114
* __ilog2 function returns for an input of 0.
115
* On 32-bit PowerPC, left shift counts with bit
116
* 26 set (that the value of 255 or 0xFFFFFFFF
117
* will have), cause the destination register to
118
* be 0. That is why this works.
125
* Step 2: check each common CAS latency against tCK of each
128
lowest_good_caslat = 0;
132
temp2 = __ilog2(temp1);
133
debug("checking common caslat = %u\n", temp2);
135
/* Check if this CAS latency will work on all DIMMs at tCK. */
136
for (i = 0; i < number_of_dimms; i++) {
137
if (!dimm_params[i].n_ranks)
140
if (dimm_params[i].caslat_x == temp2) {
141
if (mclk_ps >= dimm_params[i].tckmin_x_ps) {
142
debug("CL = %u ok on DIMM %u at tCK=%u ps with tCKmin_X_ps of %u\n",
144
dimm_params[i].tckmin_x_ps);
151
if (dimm_params[i].caslat_x_minus_1 == temp2) {
152
unsigned int tckmin_x_minus_1_ps
153
= dimm_params[i].tckmin_x_minus_1_ps;
154
if (mclk_ps >= tckmin_x_minus_1_ps) {
155
debug("CL = %u ok on DIMM %u at tCK=%u ps with tckmin_x_minus_1_ps of %u\n",
157
tckmin_x_minus_1_ps);
164
if (dimm_params[i].caslat_x_minus_2 == temp2) {
165
unsigned int tckmin_x_minus_2_ps
166
= dimm_params[i].tckmin_x_minus_2_ps;
167
if (mclk_ps >= tckmin_x_minus_2_ps) {
168
debug("CL = %u ok on DIMM %u at tCK=%u ps with tckmin_x_minus_2_ps of %u\n",
170
tckmin_x_minus_2_ps);
179
lowest_good_caslat = temp2;
181
temp1 &= ~(1 << temp2);
184
debug("lowest common SPD-defined CAS latency = %u\n",
186
outpdimm->lowest_common_spd_caslat = lowest_good_caslat;
190
* Compute a common 'de-rated' CAS latency.
192
* The strategy here is to find the *highest* dereated cas latency
193
* with the assumption that all of the DIMMs will support a dereated
194
* CAS latency higher than or equal to their lowest dereated value.
197
for (i = 0; i < number_of_dimms; i++)
198
temp1 = max(temp1, dimm_params[i].caslat_lowest_derated);
200
outpdimm->highest_common_derated_caslat = temp1;
201
debug("highest common dereated CAS latency = %u\n", temp1);
208
* compute_lowest_common_dimm_parameters()
210
* Determine the worst-case DIMM timing parameters from the set of DIMMs
211
* whose parameters have been computed into the array pointed to
215
compute_lowest_common_dimm_parameters(const dimm_params_t *dimm_params,
216
common_timing_params_t *outpdimm,
217
const unsigned int number_of_dimms)
221
unsigned int tckmin_x_ps = 0;
222
unsigned int tckmax_ps = 0xFFFFFFFF;
223
unsigned int trcd_ps = 0;
224
unsigned int trp_ps = 0;
225
unsigned int tras_ps = 0;
226
#if defined(CONFIG_SYS_FSL_DDR3) || defined(CONFIG_SYS_FSL_DDR4)
227
unsigned int taamin_ps = 0;
229
#ifdef CONFIG_SYS_FSL_DDR4
230
unsigned int twr_ps = 15000;
231
unsigned int trfc1_ps = 0;
232
unsigned int trfc2_ps = 0;
233
unsigned int trfc4_ps = 0;
234
unsigned int trrds_ps = 0;
235
unsigned int trrdl_ps = 0;
236
unsigned int tccdl_ps = 0;
238
unsigned int twr_ps = 0;
239
unsigned int twtr_ps = 0;
240
unsigned int trfc_ps = 0;
241
unsigned int trrd_ps = 0;
242
unsigned int trtp_ps = 0;
244
unsigned int trc_ps = 0;
245
unsigned int refresh_rate_ps = 0;
246
unsigned int extended_op_srt = 1;
247
#if defined(CONFIG_SYS_FSL_DDR1) || defined(CONFIG_SYS_FSL_DDR2)
248
unsigned int tis_ps = 0;
249
unsigned int tih_ps = 0;
250
unsigned int tds_ps = 0;
251
unsigned int tdh_ps = 0;
252
unsigned int tdqsq_max_ps = 0;
253
unsigned int tqhs_ps = 0;
255
unsigned int temp1, temp2;
256
unsigned int additive_latency = 0;
259
for (i = 0; i < number_of_dimms; i++) {
261
* If there are no ranks on this DIMM,
262
* it probably doesn't exist, so skip it.
264
if (dimm_params[i].n_ranks == 0) {
268
if (dimm_params[i].n_ranks == 4 && i != 0) {
269
printf("Found Quad-rank DIMM in wrong bank, ignored."
270
" Software may not run as expected.\n");
276
* check if quad-rank DIMM is plugged if
277
* CONFIG_CHIP_SELECT_QUAD_CAPABLE is not defined
278
* Only the board with proper design is capable
280
#ifndef CONFIG_FSL_DDR_FIRST_SLOT_QUAD_CAPABLE
281
if (dimm_params[i].n_ranks == 4 && \
282
CONFIG_CHIP_SELECTS_PER_CTRL/CONFIG_DIMM_SLOTS_PER_CTLR < 4) {
283
printf("Found Quad-rank DIMM, not able to support.");
289
* Find minimum tckmax_ps to find fastest slow speed,
290
* i.e., this is the slowest the whole system can go.
292
tckmax_ps = min(tckmax_ps, dimm_params[i].tckmax_ps);
293
#if defined(CONFIG_SYS_FSL_DDR3) || defined(CONFIG_SYS_FSL_DDR4)
294
taamin_ps = max(taamin_ps, dimm_params[i].taa_ps);
296
tckmin_x_ps = max(tckmin_x_ps, dimm_params[i].tckmin_x_ps);
297
trcd_ps = max(trcd_ps, dimm_params[i].trcd_ps);
298
trp_ps = max(trp_ps, dimm_params[i].trp_ps);
299
tras_ps = max(tras_ps, dimm_params[i].tras_ps);
300
#ifdef CONFIG_SYS_FSL_DDR4
301
trfc1_ps = max(trfc1_ps, dimm_params[i].trfc1_ps);
302
trfc2_ps = max(trfc2_ps, dimm_params[i].trfc2_ps);
303
trfc4_ps = max(trfc4_ps, dimm_params[i].trfc4_ps);
304
trrds_ps = max(trrds_ps, dimm_params[i].trrds_ps);
305
trrdl_ps = max(trrdl_ps, dimm_params[i].trrdl_ps);
306
tccdl_ps = max(tccdl_ps, dimm_params[i].tccdl_ps);
308
twr_ps = max(twr_ps, dimm_params[i].twr_ps);
309
twtr_ps = max(twtr_ps, dimm_params[i].twtr_ps);
310
trfc_ps = max(trfc_ps, dimm_params[i].trfc_ps);
311
trrd_ps = max(trrd_ps, dimm_params[i].trrd_ps);
312
trtp_ps = max(trtp_ps, dimm_params[i].trtp_ps);
314
trc_ps = max(trc_ps, dimm_params[i].trc_ps);
315
#if defined(CONFIG_SYS_FSL_DDR1) || defined(CONFIG_SYS_FSL_DDR2)
316
tis_ps = max(tis_ps, dimm_params[i].tis_ps);
317
tih_ps = max(tih_ps, dimm_params[i].tih_ps);
318
tds_ps = max(tds_ps, dimm_params[i].tds_ps);
319
tdh_ps = max(tdh_ps, dimm_params[i].tdh_ps);
320
tqhs_ps = max(tqhs_ps, dimm_params[i].tqhs_ps);
322
* Find maximum tdqsq_max_ps to find slowest.
324
* FIXME: is finding the slowest value the correct
325
* strategy for this parameter?
327
tdqsq_max_ps = max(tdqsq_max_ps, dimm_params[i].tdqsq_max_ps);
329
refresh_rate_ps = max(refresh_rate_ps,
330
dimm_params[i].refresh_rate_ps);
331
/* extended_op_srt is either 0 or 1, 0 having priority */
332
extended_op_srt = min(extended_op_srt,
333
dimm_params[i].extended_op_srt);
336
outpdimm->ndimms_present = number_of_dimms - temp1;
338
if (temp1 == number_of_dimms) {
339
debug("no dimms this memory controller\n");
343
outpdimm->tckmin_x_ps = tckmin_x_ps;
344
outpdimm->tckmax_ps = tckmax_ps;
345
#if defined(CONFIG_SYS_FSL_DDR3) || defined(CONFIG_SYS_FSL_DDR4)
346
outpdimm->taamin_ps = taamin_ps;
348
outpdimm->trcd_ps = trcd_ps;
349
outpdimm->trp_ps = trp_ps;
350
outpdimm->tras_ps = tras_ps;
351
#ifdef CONFIG_SYS_FSL_DDR4
352
outpdimm->trfc1_ps = trfc1_ps;
353
outpdimm->trfc2_ps = trfc2_ps;
354
outpdimm->trfc4_ps = trfc4_ps;
355
outpdimm->trrds_ps = trrds_ps;
356
outpdimm->trrdl_ps = trrdl_ps;
357
outpdimm->tccdl_ps = tccdl_ps;
359
outpdimm->twtr_ps = twtr_ps;
360
outpdimm->trfc_ps = trfc_ps;
361
outpdimm->trrd_ps = trrd_ps;
362
outpdimm->trtp_ps = trtp_ps;
364
outpdimm->twr_ps = twr_ps;
365
outpdimm->trc_ps = trc_ps;
366
outpdimm->refresh_rate_ps = refresh_rate_ps;
367
outpdimm->extended_op_srt = extended_op_srt;
368
#if defined(CONFIG_SYS_FSL_DDR1) || defined(CONFIG_SYS_FSL_DDR2)
369
outpdimm->tis_ps = tis_ps;
370
outpdimm->tih_ps = tih_ps;
371
outpdimm->tds_ps = tds_ps;
372
outpdimm->tdh_ps = tdh_ps;
373
outpdimm->tdqsq_max_ps = tdqsq_max_ps;
374
outpdimm->tqhs_ps = tqhs_ps;
377
/* Determine common burst length for all DIMMs. */
379
for (i = 0; i < number_of_dimms; i++) {
380
if (dimm_params[i].n_ranks) {
381
temp1 &= dimm_params[i].burst_lengths_bitmask;
384
outpdimm->all_dimms_burst_lengths_bitmask = temp1;
386
/* Determine if all DIMMs registered buffered. */
388
for (i = 0; i < number_of_dimms; i++) {
389
if (dimm_params[i].n_ranks) {
390
if (dimm_params[i].registered_dimm) {
392
#ifndef CONFIG_SPL_BUILD
393
printf("Detected RDIMM %s\n",
394
dimm_params[i].mpart);
398
#ifndef CONFIG_SPL_BUILD
399
printf("Detected UDIMM %s\n",
400
dimm_params[i].mpart);
406
outpdimm->all_dimms_registered = 0;
407
outpdimm->all_dimms_unbuffered = 0;
408
if (temp1 && !temp2) {
409
outpdimm->all_dimms_registered = 1;
410
} else if (!temp1 && temp2) {
411
outpdimm->all_dimms_unbuffered = 1;
413
printf("ERROR: Mix of registered buffered and unbuffered "
414
"DIMMs detected!\n");
418
if (outpdimm->all_dimms_registered)
419
for (j = 0; j < 16; j++) {
420
outpdimm->rcw[j] = dimm_params[0].rcw[j];
421
for (i = 1; i < number_of_dimms; i++) {
422
if (!dimm_params[i].n_ranks)
424
if (dimm_params[i].rcw[j] != dimm_params[0].rcw[j]) {
432
printf("ERROR: Mix different RDIMM detected!\n");
434
/* calculate cas latency for all DDR types */
435
if (compute_cas_latency(dimm_params, outpdimm, number_of_dimms))
438
/* Determine if all DIMMs ECC capable. */
440
for (i = 0; i < number_of_dimms; i++) {
441
if (dimm_params[i].n_ranks &&
442
!(dimm_params[i].edc_config & EDC_ECC)) {
448
debug("all DIMMs ECC capable\n");
450
debug("Warning: not all DIMMs ECC capable, cant enable ECC\n");
452
outpdimm->all_dimms_ecc_capable = temp1;
455
* Compute additive latency.
457
* For DDR1, additive latency should be 0.
459
* For DDR2, with ODT enabled, use "a value" less than ACTTORW,
460
* which comes from Trcd, and also note that:
461
* add_lat + caslat must be >= 4
463
* For DDR3, we use the AL=0
465
* When to use additive latency for DDR2:
467
* I. Because you are using CL=3 and need to do ODT on writes and
468
* want functionality.
469
* 1. Are you going to use ODT? (Does your board not have
470
* additional termination circuitry for DQ, DQS, DQS_,
471
* DM, RDQS, RDQS_ for x4/x8 configs?)
472
* 2. If so, is your lowest supported CL going to be 3?
473
* 3. If so, then you must set AL=1 because
475
* WL >= 3 for ODT on writes
484
* RL >= 3 for ODT on reads
487
* Since CL aren't usually less than 2, AL=0 is a minimum,
488
* so the WL-derived AL should be the -- FIXME?
490
* II. Because you are using auto-precharge globally and want to
491
* use additive latency (posted CAS) to get more bandwidth.
492
* 1. Are you going to use auto-precharge mode globally?
494
* Use addtivie latency and compute AL to be 1 cycle less than
495
* tRCD, i.e. the READ or WRITE command is in the cycle
496
* immediately following the ACTIVATE command..
498
* III. Because you feel like it or want to do some sort of
499
* degraded-performance experiment.
500
* 1. Do you just want to use additive latency because you feel
503
* Validation: AL is less than tRCD, and within the other
504
* read-to-precharge constraints.
507
additive_latency = 0;
509
#if defined(CONFIG_SYS_FSL_DDR2)
510
if ((outpdimm->lowest_common_spd_caslat < 4) &&
511
(picos_to_mclk(trcd_ps) > outpdimm->lowest_common_spd_caslat)) {
512
additive_latency = picos_to_mclk(trcd_ps) -
513
outpdimm->lowest_common_spd_caslat;
514
if (mclk_to_picos(additive_latency) > trcd_ps) {
515
additive_latency = picos_to_mclk(trcd_ps);
516
debug("setting additive_latency to %u because it was "
517
" greater than tRCD_ps\n", additive_latency);
523
* Validate additive latency
527
if (mclk_to_picos(additive_latency) > trcd_ps) {
528
printf("Error: invalid additive latency exceeds tRCD(min).\n");
533
* RL = CL + AL; RL >= 3 for ODT_RD_CFG to be enabled
534
* WL = RL - 1; WL >= 3 for ODT_WL_CFG to be enabled
535
* ADD_LAT (the register) must be set to a value less
536
* than ACTTORW if WL = 1, then AL must be set to 1
537
* RD_TO_PRE (the register) must be set to a minimum
538
* tRTP + AL if AL is nonzero
542
* Additive latency will be applied only if the memctl option to
545
outpdimm->additive_latency = additive_latency;
547
debug("tCKmin_ps = %u\n", outpdimm->tckmin_x_ps);
548
debug("trcd_ps = %u\n", outpdimm->trcd_ps);
549
debug("trp_ps = %u\n", outpdimm->trp_ps);
550
debug("tras_ps = %u\n", outpdimm->tras_ps);
551
#ifdef CONFIG_SYS_FSL_DDR4
552
debug("trfc1_ps = %u\n", trfc1_ps);
553
debug("trfc2_ps = %u\n", trfc2_ps);
554
debug("trfc4_ps = %u\n", trfc4_ps);
555
debug("trrds_ps = %u\n", trrds_ps);
556
debug("trrdl_ps = %u\n", trrdl_ps);
557
debug("tccdl_ps = %u\n", tccdl_ps);
559
debug("twtr_ps = %u\n", outpdimm->twtr_ps);
560
debug("trfc_ps = %u\n", outpdimm->trfc_ps);
561
debug("trrd_ps = %u\n", outpdimm->trrd_ps);
563
debug("twr_ps = %u\n", outpdimm->twr_ps);
564
debug("trc_ps = %u\n", outpdimm->trc_ps);