1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright 2008-2016 Freescale Semiconductor, Inc. 4 * Copyright 2017-2018 NXP Semiconductor 5 */ 6 7 #include <common.h> 8 #include <fsl_ddr_sdram.h> 9 10 #include <fsl_ddr.h> 11 12 #if defined(CONFIG_SYS_FSL_DDR3) || defined(CONFIG_SYS_FSL_DDR4) 13 static unsigned int 14 compute_cas_latency(const unsigned int ctrl_num, 15 const dimm_params_t *dimm_params, 16 common_timing_params_t *outpdimm, 17 unsigned int number_of_dimms) 18 { 19 unsigned int i; 20 unsigned int common_caslat; 21 unsigned int caslat_actual; 22 unsigned int retry = 16; 23 unsigned int tmp = ~0; 24 const unsigned int mclk_ps = get_memory_clk_period_ps(ctrl_num); 25 #ifdef CONFIG_SYS_FSL_DDR3 26 const unsigned int taamax = 20000; 27 #else 28 const unsigned int taamax = 18000; 29 #endif 30 31 /* compute the common CAS latency supported between slots */ 32 for (i = 0; i < number_of_dimms; i++) { 33 if (dimm_params[i].n_ranks) 34 tmp &= dimm_params[i].caslat_x; 35 } 36 common_caslat = tmp; 37 38 /* validate if the memory clk is in the range of dimms */ 39 if (mclk_ps < outpdimm->tckmin_x_ps) { 40 printf("DDR clock (MCLK cycle %u ps) is faster than " 41 "the slowest DIMM(s) (tCKmin %u ps) can support.\n", 42 mclk_ps, outpdimm->tckmin_x_ps); 43 } 44 #ifdef CONFIG_SYS_FSL_DDR4 45 if (mclk_ps > outpdimm->tckmax_ps) { 46 printf("DDR clock (MCLK cycle %u ps) is slower than DIMM(s) (tCKmax %u ps) can support.\n", 47 mclk_ps, outpdimm->tckmax_ps); 48 } 49 #endif 50 /* determine the acutal cas latency */ 51 caslat_actual = (outpdimm->taamin_ps + mclk_ps - 1) / mclk_ps; 52 /* check if the dimms support the CAS latency */ 53 while (!(common_caslat & (1 << caslat_actual)) && retry > 0) { 54 caslat_actual++; 55 retry--; 56 } 57 /* once the caculation of caslat_actual is completed 58 * we must verify that this CAS latency value does not 59 * exceed tAAmax, which is 20 ns for all DDR3 speed grades, 60 * 18ns for all DDR4 speed grades. 61 */ 62 if (caslat_actual * mclk_ps > taamax) { 63 printf("The chosen cas latency %d is too large\n", 64 caslat_actual); 65 } 66 outpdimm->lowest_common_spd_caslat = caslat_actual; 67 debug("lowest_common_spd_caslat is 0x%x\n", caslat_actual); 68 69 return 0; 70 } 71 #else /* for DDR1 and DDR2 */ 72 static unsigned int 73 compute_cas_latency(const unsigned int ctrl_num, 74 const dimm_params_t *dimm_params, 75 common_timing_params_t *outpdimm, 76 unsigned int number_of_dimms) 77 { 78 int i; 79 const unsigned int mclk_ps = get_memory_clk_period_ps(ctrl_num); 80 unsigned int lowest_good_caslat; 81 unsigned int not_ok; 82 unsigned int temp1, temp2; 83 84 debug("using mclk_ps = %u\n", mclk_ps); 85 if (mclk_ps > outpdimm->tckmax_ps) { 86 printf("Warning: DDR clock (%u ps) is slower than DIMM(s) (tCKmax %u ps)\n", 87 mclk_ps, outpdimm->tckmax_ps); 88 } 89 90 /* 91 * Compute a CAS latency suitable for all DIMMs 92 * 93 * Strategy for SPD-defined latencies: compute only 94 * CAS latency defined by all DIMMs. 95 */ 96 97 /* 98 * Step 1: find CAS latency common to all DIMMs using bitwise 99 * operation. 100 */ 101 temp1 = 0xFF; 102 for (i = 0; i < number_of_dimms; i++) { 103 if (dimm_params[i].n_ranks) { 104 temp2 = 0; 105 temp2 |= 1 << dimm_params[i].caslat_x; 106 temp2 |= 1 << dimm_params[i].caslat_x_minus_1; 107 temp2 |= 1 << dimm_params[i].caslat_x_minus_2; 108 /* 109 * If there was no entry for X-2 (X-1) in 110 * the SPD, then caslat_x_minus_2 111 * (caslat_x_minus_1) contains either 255 or 112 * 0xFFFFFFFF because that's what the glorious 113 * __ilog2 function returns for an input of 0. 114 * On 32-bit PowerPC, left shift counts with bit 115 * 26 set (that the value of 255 or 0xFFFFFFFF 116 * will have), cause the destination register to 117 * be 0. That is why this works. 118 */ 119 temp1 &= temp2; 120 } 121 } 122 123 /* 124 * Step 2: check each common CAS latency against tCK of each 125 * DIMM's SPD. 126 */ 127 lowest_good_caslat = 0; 128 temp2 = 0; 129 while (temp1) { 130 not_ok = 0; 131 temp2 = __ilog2(temp1); 132 debug("checking common caslat = %u\n", temp2); 133 134 /* Check if this CAS latency will work on all DIMMs at tCK. */ 135 for (i = 0; i < number_of_dimms; i++) { 136 if (!dimm_params[i].n_ranks) 137 continue; 138 139 if (dimm_params[i].caslat_x == temp2) { 140 if (mclk_ps >= dimm_params[i].tckmin_x_ps) { 141 debug("CL = %u ok on DIMM %u at tCK=%u ps with tCKmin_X_ps of %u\n", 142 temp2, i, mclk_ps, 143 dimm_params[i].tckmin_x_ps); 144 continue; 145 } else { 146 not_ok++; 147 } 148 } 149 150 if (dimm_params[i].caslat_x_minus_1 == temp2) { 151 unsigned int tckmin_x_minus_1_ps 152 = dimm_params[i].tckmin_x_minus_1_ps; 153 if (mclk_ps >= tckmin_x_minus_1_ps) { 154 debug("CL = %u ok on DIMM %u at tCK=%u ps with tckmin_x_minus_1_ps of %u\n", 155 temp2, i, mclk_ps, 156 tckmin_x_minus_1_ps); 157 continue; 158 } else { 159 not_ok++; 160 } 161 } 162 163 if (dimm_params[i].caslat_x_minus_2 == temp2) { 164 unsigned int tckmin_x_minus_2_ps 165 = dimm_params[i].tckmin_x_minus_2_ps; 166 if (mclk_ps >= tckmin_x_minus_2_ps) { 167 debug("CL = %u ok on DIMM %u at tCK=%u ps with tckmin_x_minus_2_ps of %u\n", 168 temp2, i, mclk_ps, 169 tckmin_x_minus_2_ps); 170 continue; 171 } else { 172 not_ok++; 173 } 174 } 175 } 176 177 if (!not_ok) 178 lowest_good_caslat = temp2; 179 180 temp1 &= ~(1 << temp2); 181 } 182 183 debug("lowest common SPD-defined CAS latency = %u\n", 184 lowest_good_caslat); 185 outpdimm->lowest_common_spd_caslat = lowest_good_caslat; 186 187 188 /* 189 * Compute a common 'de-rated' CAS latency. 190 * 191 * The strategy here is to find the *highest* dereated cas latency 192 * with the assumption that all of the DIMMs will support a dereated 193 * CAS latency higher than or equal to their lowest dereated value. 194 */ 195 temp1 = 0; 196 for (i = 0; i < number_of_dimms; i++) 197 temp1 = max(temp1, dimm_params[i].caslat_lowest_derated); 198 199 outpdimm->highest_common_derated_caslat = temp1; 200 debug("highest common dereated CAS latency = %u\n", temp1); 201 202 return 0; 203 } 204 #endif 205 206 /* 207 * compute_lowest_common_dimm_parameters() 208 * 209 * Determine the worst-case DIMM timing parameters from the set of DIMMs 210 * whose parameters have been computed into the array pointed to 211 * by dimm_params. 212 */ 213 unsigned int 214 compute_lowest_common_dimm_parameters(const unsigned int ctrl_num, 215 const dimm_params_t *dimm_params, 216 common_timing_params_t *outpdimm, 217 const unsigned int number_of_dimms) 218 { 219 unsigned int i, j; 220 221 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; 228 #endif 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; 237 unsigned int trfc_slr_ps = 0; 238 #else 239 unsigned int twr_ps = 0; 240 unsigned int twtr_ps = 0; 241 unsigned int trfc_ps = 0; 242 unsigned int trrd_ps = 0; 243 unsigned int trtp_ps = 0; 244 #endif 245 unsigned int trc_ps = 0; 246 unsigned int refresh_rate_ps = 0; 247 unsigned int extended_op_srt = 1; 248 #if defined(CONFIG_SYS_FSL_DDR1) || defined(CONFIG_SYS_FSL_DDR2) 249 unsigned int tis_ps = 0; 250 unsigned int tih_ps = 0; 251 unsigned int tds_ps = 0; 252 unsigned int tdh_ps = 0; 253 unsigned int tdqsq_max_ps = 0; 254 unsigned int tqhs_ps = 0; 255 #endif 256 unsigned int temp1, temp2; 257 unsigned int additive_latency = 0; 258 259 temp1 = 0; 260 for (i = 0; i < number_of_dimms; i++) { 261 /* 262 * If there are no ranks on this DIMM, 263 * it probably doesn't exist, so skip it. 264 */ 265 if (dimm_params[i].n_ranks == 0) { 266 temp1++; 267 continue; 268 } 269 if (dimm_params[i].n_ranks == 4 && i != 0) { 270 printf("Found Quad-rank DIMM in wrong bank, ignored." 271 " Software may not run as expected.\n"); 272 temp1++; 273 continue; 274 } 275 276 /* 277 * check if quad-rank DIMM is plugged if 278 * CONFIG_CHIP_SELECT_QUAD_CAPABLE is not defined 279 * Only the board with proper design is capable 280 */ 281 #ifndef CONFIG_FSL_DDR_FIRST_SLOT_QUAD_CAPABLE 282 if (dimm_params[i].n_ranks == 4 && \ 283 CONFIG_CHIP_SELECTS_PER_CTRL/CONFIG_DIMM_SLOTS_PER_CTLR < 4) { 284 printf("Found Quad-rank DIMM, not able to support."); 285 temp1++; 286 continue; 287 } 288 #endif 289 /* 290 * Find minimum tckmax_ps to find fastest slow speed, 291 * i.e., this is the slowest the whole system can go. 292 */ 293 tckmax_ps = min(tckmax_ps, 294 (unsigned int)dimm_params[i].tckmax_ps); 295 #if defined(CONFIG_SYS_FSL_DDR3) || defined(CONFIG_SYS_FSL_DDR4) 296 taamin_ps = max(taamin_ps, 297 (unsigned int)dimm_params[i].taa_ps); 298 #endif 299 tckmin_x_ps = max(tckmin_x_ps, 300 (unsigned int)dimm_params[i].tckmin_x_ps); 301 trcd_ps = max(trcd_ps, (unsigned int)dimm_params[i].trcd_ps); 302 trp_ps = max(trp_ps, (unsigned int)dimm_params[i].trp_ps); 303 tras_ps = max(tras_ps, (unsigned int)dimm_params[i].tras_ps); 304 #ifdef CONFIG_SYS_FSL_DDR4 305 trfc1_ps = max(trfc1_ps, 306 (unsigned int)dimm_params[i].trfc1_ps); 307 trfc2_ps = max(trfc2_ps, 308 (unsigned int)dimm_params[i].trfc2_ps); 309 trfc4_ps = max(trfc4_ps, 310 (unsigned int)dimm_params[i].trfc4_ps); 311 trrds_ps = max(trrds_ps, 312 (unsigned int)dimm_params[i].trrds_ps); 313 trrdl_ps = max(trrdl_ps, 314 (unsigned int)dimm_params[i].trrdl_ps); 315 tccdl_ps = max(tccdl_ps, 316 (unsigned int)dimm_params[i].tccdl_ps); 317 trfc_slr_ps = max(trfc_slr_ps, 318 (unsigned int)dimm_params[i].trfc_slr_ps); 319 #else 320 twr_ps = max(twr_ps, (unsigned int)dimm_params[i].twr_ps); 321 twtr_ps = max(twtr_ps, (unsigned int)dimm_params[i].twtr_ps); 322 trfc_ps = max(trfc_ps, (unsigned int)dimm_params[i].trfc_ps); 323 trrd_ps = max(trrd_ps, (unsigned int)dimm_params[i].trrd_ps); 324 trtp_ps = max(trtp_ps, (unsigned int)dimm_params[i].trtp_ps); 325 #endif 326 trc_ps = max(trc_ps, (unsigned int)dimm_params[i].trc_ps); 327 #if defined(CONFIG_SYS_FSL_DDR1) || defined(CONFIG_SYS_FSL_DDR2) 328 tis_ps = max(tis_ps, (unsigned int)dimm_params[i].tis_ps); 329 tih_ps = max(tih_ps, (unsigned int)dimm_params[i].tih_ps); 330 tds_ps = max(tds_ps, (unsigned int)dimm_params[i].tds_ps); 331 tdh_ps = max(tdh_ps, (unsigned int)dimm_params[i].tdh_ps); 332 tqhs_ps = max(tqhs_ps, (unsigned int)dimm_params[i].tqhs_ps); 333 /* 334 * Find maximum tdqsq_max_ps to find slowest. 335 * 336 * FIXME: is finding the slowest value the correct 337 * strategy for this parameter? 338 */ 339 tdqsq_max_ps = max(tdqsq_max_ps, 340 (unsigned int)dimm_params[i].tdqsq_max_ps); 341 #endif 342 refresh_rate_ps = max(refresh_rate_ps, 343 (unsigned int)dimm_params[i].refresh_rate_ps); 344 /* extended_op_srt is either 0 or 1, 0 having priority */ 345 extended_op_srt = min(extended_op_srt, 346 (unsigned int)dimm_params[i].extended_op_srt); 347 } 348 349 outpdimm->ndimms_present = number_of_dimms - temp1; 350 351 if (temp1 == number_of_dimms) { 352 debug("no dimms this memory controller\n"); 353 return 0; 354 } 355 356 outpdimm->tckmin_x_ps = tckmin_x_ps; 357 outpdimm->tckmax_ps = tckmax_ps; 358 #if defined(CONFIG_SYS_FSL_DDR3) || defined(CONFIG_SYS_FSL_DDR4) 359 outpdimm->taamin_ps = taamin_ps; 360 #endif 361 outpdimm->trcd_ps = trcd_ps; 362 outpdimm->trp_ps = trp_ps; 363 outpdimm->tras_ps = tras_ps; 364 #ifdef CONFIG_SYS_FSL_DDR4 365 outpdimm->trfc1_ps = trfc1_ps; 366 outpdimm->trfc2_ps = trfc2_ps; 367 outpdimm->trfc4_ps = trfc4_ps; 368 outpdimm->trrds_ps = trrds_ps; 369 outpdimm->trrdl_ps = trrdl_ps; 370 outpdimm->tccdl_ps = tccdl_ps; 371 outpdimm->trfc_slr_ps = trfc_slr_ps; 372 #else 373 outpdimm->twtr_ps = twtr_ps; 374 outpdimm->trfc_ps = trfc_ps; 375 outpdimm->trrd_ps = trrd_ps; 376 outpdimm->trtp_ps = trtp_ps; 377 #endif 378 outpdimm->twr_ps = twr_ps; 379 outpdimm->trc_ps = trc_ps; 380 outpdimm->refresh_rate_ps = refresh_rate_ps; 381 outpdimm->extended_op_srt = extended_op_srt; 382 #if defined(CONFIG_SYS_FSL_DDR1) || defined(CONFIG_SYS_FSL_DDR2) 383 outpdimm->tis_ps = tis_ps; 384 outpdimm->tih_ps = tih_ps; 385 outpdimm->tds_ps = tds_ps; 386 outpdimm->tdh_ps = tdh_ps; 387 outpdimm->tdqsq_max_ps = tdqsq_max_ps; 388 outpdimm->tqhs_ps = tqhs_ps; 389 #endif 390 391 /* Determine common burst length for all DIMMs. */ 392 temp1 = 0xff; 393 for (i = 0; i < number_of_dimms; i++) { 394 if (dimm_params[i].n_ranks) { 395 temp1 &= dimm_params[i].burst_lengths_bitmask; 396 } 397 } 398 outpdimm->all_dimms_burst_lengths_bitmask = temp1; 399 400 /* Determine if all DIMMs registered buffered. */ 401 temp1 = temp2 = 0; 402 for (i = 0; i < number_of_dimms; i++) { 403 if (dimm_params[i].n_ranks) { 404 if (dimm_params[i].registered_dimm) { 405 temp1 = 1; 406 #ifndef CONFIG_SPL_BUILD 407 printf("Detected RDIMM %s\n", 408 dimm_params[i].mpart); 409 #endif 410 } else { 411 temp2 = 1; 412 #ifndef CONFIG_SPL_BUILD 413 printf("Detected UDIMM %s\n", 414 dimm_params[i].mpart); 415 #endif 416 } 417 } 418 } 419 420 outpdimm->all_dimms_registered = 0; 421 outpdimm->all_dimms_unbuffered = 0; 422 if (temp1 && !temp2) { 423 outpdimm->all_dimms_registered = 1; 424 } else if (!temp1 && temp2) { 425 outpdimm->all_dimms_unbuffered = 1; 426 } else { 427 printf("ERROR: Mix of registered buffered and unbuffered " 428 "DIMMs detected!\n"); 429 } 430 431 temp1 = 0; 432 if (outpdimm->all_dimms_registered) 433 for (j = 0; j < 16; j++) { 434 outpdimm->rcw[j] = dimm_params[0].rcw[j]; 435 for (i = 1; i < number_of_dimms; i++) { 436 if (!dimm_params[i].n_ranks) 437 continue; 438 if (dimm_params[i].rcw[j] != dimm_params[0].rcw[j]) { 439 temp1 = 1; 440 break; 441 } 442 } 443 } 444 445 if (temp1 != 0) 446 printf("ERROR: Mix different RDIMM detected!\n"); 447 448 /* calculate cas latency for all DDR types */ 449 if (compute_cas_latency(ctrl_num, dimm_params, 450 outpdimm, number_of_dimms)) 451 return 1; 452 453 /* Determine if all DIMMs ECC capable. */ 454 temp1 = 1; 455 for (i = 0; i < number_of_dimms; i++) { 456 if (dimm_params[i].n_ranks && 457 !(dimm_params[i].edc_config & EDC_ECC)) { 458 temp1 = 0; 459 break; 460 } 461 } 462 if (temp1) { 463 debug("all DIMMs ECC capable\n"); 464 } else { 465 debug("Warning: not all DIMMs ECC capable, cant enable ECC\n"); 466 } 467 outpdimm->all_dimms_ecc_capable = temp1; 468 469 /* 470 * Compute additive latency. 471 * 472 * For DDR1, additive latency should be 0. 473 * 474 * For DDR2, with ODT enabled, use "a value" less than ACTTORW, 475 * which comes from Trcd, and also note that: 476 * add_lat + caslat must be >= 4 477 * 478 * For DDR3, we use the AL=0 479 * 480 * When to use additive latency for DDR2: 481 * 482 * I. Because you are using CL=3 and need to do ODT on writes and 483 * want functionality. 484 * 1. Are you going to use ODT? (Does your board not have 485 * additional termination circuitry for DQ, DQS, DQS_, 486 * DM, RDQS, RDQS_ for x4/x8 configs?) 487 * 2. If so, is your lowest supported CL going to be 3? 488 * 3. If so, then you must set AL=1 because 489 * 490 * WL >= 3 for ODT on writes 491 * RL = AL + CL 492 * WL = RL - 1 493 * -> 494 * WL = AL + CL - 1 495 * AL + CL - 1 >= 3 496 * AL + CL >= 4 497 * QED 498 * 499 * RL >= 3 for ODT on reads 500 * RL = AL + CL 501 * 502 * Since CL aren't usually less than 2, AL=0 is a minimum, 503 * so the WL-derived AL should be the -- FIXME? 504 * 505 * II. Because you are using auto-precharge globally and want to 506 * use additive latency (posted CAS) to get more bandwidth. 507 * 1. Are you going to use auto-precharge mode globally? 508 * 509 * Use addtivie latency and compute AL to be 1 cycle less than 510 * tRCD, i.e. the READ or WRITE command is in the cycle 511 * immediately following the ACTIVATE command.. 512 * 513 * III. Because you feel like it or want to do some sort of 514 * degraded-performance experiment. 515 * 1. Do you just want to use additive latency because you feel 516 * like it? 517 * 518 * Validation: AL is less than tRCD, and within the other 519 * read-to-precharge constraints. 520 */ 521 522 additive_latency = 0; 523 524 #if defined(CONFIG_SYS_FSL_DDR2) 525 if ((outpdimm->lowest_common_spd_caslat < 4) && 526 (picos_to_mclk(ctrl_num, trcd_ps) > 527 outpdimm->lowest_common_spd_caslat)) { 528 additive_latency = picos_to_mclk(ctrl_num, trcd_ps) - 529 outpdimm->lowest_common_spd_caslat; 530 if (mclk_to_picos(ctrl_num, additive_latency) > trcd_ps) { 531 additive_latency = picos_to_mclk(ctrl_num, trcd_ps); 532 debug("setting additive_latency to %u because it was " 533 " greater than tRCD_ps\n", additive_latency); 534 } 535 } 536 #endif 537 538 /* 539 * Validate additive latency 540 * 541 * AL <= tRCD(min) 542 */ 543 if (mclk_to_picos(ctrl_num, additive_latency) > trcd_ps) { 544 printf("Error: invalid additive latency exceeds tRCD(min).\n"); 545 return 1; 546 } 547 548 /* 549 * RL = CL + AL; RL >= 3 for ODT_RD_CFG to be enabled 550 * WL = RL - 1; WL >= 3 for ODT_WL_CFG to be enabled 551 * ADD_LAT (the register) must be set to a value less 552 * than ACTTORW if WL = 1, then AL must be set to 1 553 * RD_TO_PRE (the register) must be set to a minimum 554 * tRTP + AL if AL is nonzero 555 */ 556 557 /* 558 * Additive latency will be applied only if the memctl option to 559 * use it. 560 */ 561 outpdimm->additive_latency = additive_latency; 562 563 debug("tCKmin_ps = %u\n", outpdimm->tckmin_x_ps); 564 debug("trcd_ps = %u\n", outpdimm->trcd_ps); 565 debug("trp_ps = %u\n", outpdimm->trp_ps); 566 debug("tras_ps = %u\n", outpdimm->tras_ps); 567 #ifdef CONFIG_SYS_FSL_DDR4 568 debug("trfc1_ps = %u\n", trfc1_ps); 569 debug("trfc2_ps = %u\n", trfc2_ps); 570 debug("trfc4_ps = %u\n", trfc4_ps); 571 debug("trrds_ps = %u\n", trrds_ps); 572 debug("trrdl_ps = %u\n", trrdl_ps); 573 debug("tccdl_ps = %u\n", tccdl_ps); 574 debug("trfc_slr_ps = %u\n", trfc_slr_ps); 575 #else 576 debug("twtr_ps = %u\n", outpdimm->twtr_ps); 577 debug("trfc_ps = %u\n", outpdimm->trfc_ps); 578 debug("trrd_ps = %u\n", outpdimm->trrd_ps); 579 #endif 580 debug("twr_ps = %u\n", outpdimm->twr_ps); 581 debug("trc_ps = %u\n", outpdimm->trc_ps); 582 583 return 0; 584 } 585