1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * davinci_nand.c - NAND Flash Driver for DaVinci family chips 4 * 5 * Copyright © 2006 Texas Instruments. 6 * 7 * Port to 2.6.23 Copyright © 2008 by: 8 * Sander Huijsen <Shuijsen@optelecom-nkf.com> 9 * Troy Kisky <troy.kisky@boundarydevices.com> 10 * Dirk Behme <Dirk.Behme@gmail.com> 11 */ 12 13 #include <linux/clk.h> 14 #include <linux/err.h> 15 #include <linux/iopoll.h> 16 #include <linux/kernel.h> 17 #include <linux/memory/ti-aemif.h> 18 #include <linux/module.h> 19 #include <linux/mtd/partitions.h> 20 #include <linux/mtd/rawnand.h> 21 #include <linux/platform_device.h> 22 #include <linux/property.h> 23 #include <linux/slab.h> 24 25 #define NRCSR_OFFSET 0x00 26 #define NANDFCR_OFFSET 0x60 27 #define NANDFSR_OFFSET 0x64 28 #define NANDF1ECC_OFFSET 0x70 29 30 /* 4-bit ECC syndrome registers */ 31 #define NAND_4BIT_ECC_LOAD_OFFSET 0xbc 32 #define NAND_4BIT_ECC1_OFFSET 0xc0 33 #define NAND_4BIT_ECC2_OFFSET 0xc4 34 #define NAND_4BIT_ECC3_OFFSET 0xc8 35 #define NAND_4BIT_ECC4_OFFSET 0xcc 36 #define NAND_ERR_ADD1_OFFSET 0xd0 37 #define NAND_ERR_ADD2_OFFSET 0xd4 38 #define NAND_ERR_ERRVAL1_OFFSET 0xd8 39 #define NAND_ERR_ERRVAL2_OFFSET 0xdc 40 41 /* NOTE: boards don't need to use these address bits 42 * for ALE/CLE unless they support booting from NAND. 43 * They're used unless platform data overrides them. 44 */ 45 #define MASK_ALE 0x08 46 #define MASK_CLE 0x10 47 48 #define MAX_TSU_PS 3000 /* Input setup time in ps */ 49 #define MAX_TH_PS 1600 /* Input hold time in ps */ 50 51 struct davinci_nand_pdata { 52 uint32_t mask_ale; 53 uint32_t mask_cle; 54 55 /* 56 * 0-indexed chip-select number of the asynchronous 57 * interface to which the NAND device has been connected. 58 * 59 * So, if you have NAND connected to CS3 of DA850, you 60 * will pass '1' here. Since the asynchronous interface 61 * on DA850 starts from CS2. 62 */ 63 uint32_t core_chipsel; 64 65 /* for packages using two chipselects */ 66 uint32_t mask_chipsel; 67 68 /* board's default static partition info */ 69 struct mtd_partition *parts; 70 unsigned int nr_parts; 71 72 /* none == NAND_ECC_ENGINE_TYPE_NONE (strongly *not* advised!!) 73 * soft == NAND_ECC_ENGINE_TYPE_SOFT 74 * on-die == NAND_ECC_ENGINE_TYPE_ON_DIE 75 * else == NAND_ECC_ENGINE_TYPE_ON_HOST, according to ecc_bits 76 * 77 * All DaVinci-family chips support 1-bit hardware ECC. 78 * Newer ones also support 4-bit ECC, but are awkward 79 * using it with large page chips. 80 */ 81 enum nand_ecc_engine_type engine_type; 82 enum nand_ecc_placement ecc_placement; 83 u8 ecc_bits; 84 85 /* e.g. NAND_BUSWIDTH_16 */ 86 unsigned int options; 87 /* e.g. NAND_BBT_USE_FLASH */ 88 unsigned int bbt_options; 89 90 /* Main and mirror bbt descriptor overrides */ 91 struct nand_bbt_descr *bbt_td; 92 struct nand_bbt_descr *bbt_md; 93 }; 94 95 /* 96 * This is a device driver for the NAND flash controller found on the 97 * various DaVinci family chips. It handles up to four SoC chipselects, 98 * and some flavors of secondary chipselect (e.g. based on A12) as used 99 * with multichip packages. 100 * 101 * The 1-bit ECC hardware is supported, as well as the newer 4-bit ECC 102 * available on chips like the DM355 and OMAP-L137 and needed with the 103 * more error-prone MLC NAND chips. 104 * 105 * This driver assumes EM_WAIT connects all the NAND devices' RDY/nBUSY 106 * outputs in a "wire-AND" configuration, with no per-chip signals. 107 */ 108 struct davinci_nand_info { 109 struct nand_controller controller; 110 struct nand_chip chip; 111 112 struct platform_device *pdev; 113 114 bool is_readmode; 115 116 void __iomem *base; 117 void __iomem *vaddr; 118 119 void __iomem *current_cs; 120 121 uint32_t mask_chipsel; 122 uint32_t mask_ale; 123 uint32_t mask_cle; 124 125 uint32_t core_chipsel; 126 127 struct clk *clk; 128 struct aemif_device *aemif; 129 }; 130 131 static DEFINE_SPINLOCK(davinci_nand_lock); 132 static bool ecc4_busy; 133 134 static inline struct davinci_nand_info *to_davinci_nand(struct mtd_info *mtd) 135 { 136 return container_of(mtd_to_nand(mtd), struct davinci_nand_info, chip); 137 } 138 139 static inline unsigned int davinci_nand_readl(struct davinci_nand_info *info, 140 int offset) 141 { 142 return __raw_readl(info->base + offset); 143 } 144 145 static inline void davinci_nand_writel(struct davinci_nand_info *info, 146 int offset, unsigned long value) 147 { 148 __raw_writel(value, info->base + offset); 149 } 150 151 /*----------------------------------------------------------------------*/ 152 153 /* 154 * 1-bit hardware ECC ... context maintained for each core chipselect 155 */ 156 157 static inline uint32_t nand_davinci_readecc_1bit(struct mtd_info *mtd) 158 { 159 struct davinci_nand_info *info = to_davinci_nand(mtd); 160 161 return davinci_nand_readl(info, NANDF1ECC_OFFSET 162 + 4 * info->core_chipsel); 163 } 164 165 static void nand_davinci_hwctl_1bit(struct nand_chip *chip, int mode) 166 { 167 struct davinci_nand_info *info; 168 uint32_t nandcfr; 169 unsigned long flags; 170 171 info = to_davinci_nand(nand_to_mtd(chip)); 172 173 /* Reset ECC hardware */ 174 nand_davinci_readecc_1bit(nand_to_mtd(chip)); 175 176 spin_lock_irqsave(&davinci_nand_lock, flags); 177 178 /* Restart ECC hardware */ 179 nandcfr = davinci_nand_readl(info, NANDFCR_OFFSET); 180 nandcfr |= BIT(8 + info->core_chipsel); 181 davinci_nand_writel(info, NANDFCR_OFFSET, nandcfr); 182 183 spin_unlock_irqrestore(&davinci_nand_lock, flags); 184 } 185 186 /* 187 * Read hardware ECC value and pack into three bytes 188 */ 189 static int nand_davinci_calculate_1bit(struct nand_chip *chip, 190 const u_char *dat, u_char *ecc_code) 191 { 192 unsigned int ecc_val = nand_davinci_readecc_1bit(nand_to_mtd(chip)); 193 unsigned int ecc24 = (ecc_val & 0x0fff) | ((ecc_val & 0x0fff0000) >> 4); 194 195 /* invert so that erased block ecc is correct */ 196 ecc24 = ~ecc24; 197 ecc_code[0] = (u_char)(ecc24); 198 ecc_code[1] = (u_char)(ecc24 >> 8); 199 ecc_code[2] = (u_char)(ecc24 >> 16); 200 201 return 0; 202 } 203 204 static int nand_davinci_correct_1bit(struct nand_chip *chip, u_char *dat, 205 u_char *read_ecc, u_char *calc_ecc) 206 { 207 uint32_t eccNand = read_ecc[0] | (read_ecc[1] << 8) | 208 (read_ecc[2] << 16); 209 uint32_t eccCalc = calc_ecc[0] | (calc_ecc[1] << 8) | 210 (calc_ecc[2] << 16); 211 uint32_t diff = eccCalc ^ eccNand; 212 213 if (diff) { 214 if ((((diff >> 12) ^ diff) & 0xfff) == 0xfff) { 215 /* Correctable error */ 216 if ((diff >> (12 + 3)) < chip->ecc.size) { 217 dat[diff >> (12 + 3)] ^= BIT((diff >> 12) & 7); 218 return 1; 219 } else { 220 return -EBADMSG; 221 } 222 } else if (!(diff & (diff - 1))) { 223 /* Single bit ECC error in the ECC itself, 224 * nothing to fix */ 225 return 1; 226 } else { 227 /* Uncorrectable error */ 228 return -EBADMSG; 229 } 230 231 } 232 return 0; 233 } 234 235 /*----------------------------------------------------------------------*/ 236 237 /* 238 * 4-bit hardware ECC ... context maintained over entire AEMIF 239 * 240 * This is a syndrome engine, but we avoid NAND_ECC_PLACEMENT_INTERLEAVED 241 * since that forces use of a problematic "infix OOB" layout. 242 * Among other things, it trashes manufacturer bad block markers. 243 * Also, and specific to this hardware, it ECC-protects the "prepad" 244 * in the OOB ... while having ECC protection for parts of OOB would 245 * seem useful, the current MTD stack sometimes wants to update the 246 * OOB without recomputing ECC. 247 */ 248 249 static void nand_davinci_hwctl_4bit(struct nand_chip *chip, int mode) 250 { 251 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip)); 252 unsigned long flags; 253 u32 val; 254 255 /* Reset ECC hardware */ 256 davinci_nand_readl(info, NAND_4BIT_ECC1_OFFSET); 257 258 spin_lock_irqsave(&davinci_nand_lock, flags); 259 260 /* Start 4-bit ECC calculation for read/write */ 261 val = davinci_nand_readl(info, NANDFCR_OFFSET); 262 val &= ~(0x03 << 4); 263 val |= (info->core_chipsel << 4) | BIT(12); 264 davinci_nand_writel(info, NANDFCR_OFFSET, val); 265 266 info->is_readmode = (mode == NAND_ECC_READ); 267 268 spin_unlock_irqrestore(&davinci_nand_lock, flags); 269 } 270 271 /* Read raw ECC code after writing to NAND. */ 272 static void 273 nand_davinci_readecc_4bit(struct davinci_nand_info *info, u32 code[4]) 274 { 275 const u32 mask = 0x03ff03ff; 276 277 code[0] = davinci_nand_readl(info, NAND_4BIT_ECC1_OFFSET) & mask; 278 code[1] = davinci_nand_readl(info, NAND_4BIT_ECC2_OFFSET) & mask; 279 code[2] = davinci_nand_readl(info, NAND_4BIT_ECC3_OFFSET) & mask; 280 code[3] = davinci_nand_readl(info, NAND_4BIT_ECC4_OFFSET) & mask; 281 } 282 283 /* Terminate read ECC; or return ECC (as bytes) of data written to NAND. */ 284 static int nand_davinci_calculate_4bit(struct nand_chip *chip, 285 const u_char *dat, u_char *ecc_code) 286 { 287 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip)); 288 u32 raw_ecc[4], *p; 289 unsigned i; 290 291 /* After a read, terminate ECC calculation by a dummy read 292 * of some 4-bit ECC register. ECC covers everything that 293 * was read; correct() just uses the hardware state, so 294 * ecc_code is not needed. 295 */ 296 if (info->is_readmode) { 297 davinci_nand_readl(info, NAND_4BIT_ECC1_OFFSET); 298 return 0; 299 } 300 301 /* Pack eight raw 10-bit ecc values into ten bytes, making 302 * two passes which each convert four values (in upper and 303 * lower halves of two 32-bit words) into five bytes. The 304 * ROM boot loader uses this same packing scheme. 305 */ 306 nand_davinci_readecc_4bit(info, raw_ecc); 307 for (i = 0, p = raw_ecc; i < 2; i++, p += 2) { 308 *ecc_code++ = p[0] & 0xff; 309 *ecc_code++ = ((p[0] >> 8) & 0x03) | ((p[0] >> 14) & 0xfc); 310 *ecc_code++ = ((p[0] >> 22) & 0x0f) | ((p[1] << 4) & 0xf0); 311 *ecc_code++ = ((p[1] >> 4) & 0x3f) | ((p[1] >> 10) & 0xc0); 312 *ecc_code++ = (p[1] >> 18) & 0xff; 313 } 314 315 return 0; 316 } 317 318 /* Correct up to 4 bits in data we just read, using state left in the 319 * hardware plus the ecc_code computed when it was first written. 320 */ 321 static int nand_davinci_correct_4bit(struct nand_chip *chip, u_char *data, 322 u_char *ecc_code, u_char *null) 323 { 324 int i; 325 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip)); 326 unsigned short ecc10[8]; 327 unsigned short *ecc16; 328 u32 syndrome[4]; 329 u32 ecc_state; 330 unsigned num_errors, corrected; 331 unsigned long timeo; 332 333 /* Unpack ten bytes into eight 10 bit values. We know we're 334 * little-endian, and use type punning for less shifting/masking. 335 */ 336 if (WARN_ON(0x01 & (uintptr_t)ecc_code)) 337 return -EINVAL; 338 ecc16 = (unsigned short *)ecc_code; 339 340 ecc10[0] = (ecc16[0] >> 0) & 0x3ff; 341 ecc10[1] = ((ecc16[0] >> 10) & 0x3f) | ((ecc16[1] << 6) & 0x3c0); 342 ecc10[2] = (ecc16[1] >> 4) & 0x3ff; 343 ecc10[3] = ((ecc16[1] >> 14) & 0x3) | ((ecc16[2] << 2) & 0x3fc); 344 ecc10[4] = (ecc16[2] >> 8) | ((ecc16[3] << 8) & 0x300); 345 ecc10[5] = (ecc16[3] >> 2) & 0x3ff; 346 ecc10[6] = ((ecc16[3] >> 12) & 0xf) | ((ecc16[4] << 4) & 0x3f0); 347 ecc10[7] = (ecc16[4] >> 6) & 0x3ff; 348 349 /* Tell ECC controller about the expected ECC codes. */ 350 for (i = 7; i >= 0; i--) 351 davinci_nand_writel(info, NAND_4BIT_ECC_LOAD_OFFSET, ecc10[i]); 352 353 /* Allow time for syndrome calculation ... then read it. 354 * A syndrome of all zeroes 0 means no detected errors. 355 */ 356 davinci_nand_readl(info, NANDFSR_OFFSET); 357 nand_davinci_readecc_4bit(info, syndrome); 358 if (!(syndrome[0] | syndrome[1] | syndrome[2] | syndrome[3])) 359 return 0; 360 361 /* 362 * Clear any previous address calculation by doing a dummy read of an 363 * error address register. 364 */ 365 davinci_nand_readl(info, NAND_ERR_ADD1_OFFSET); 366 367 /* Start address calculation, and wait for it to complete. 368 * We _could_ start reading more data while this is working, 369 * to speed up the overall page read. 370 */ 371 davinci_nand_writel(info, NANDFCR_OFFSET, 372 davinci_nand_readl(info, NANDFCR_OFFSET) | BIT(13)); 373 374 /* 375 * ECC_STATE field reads 0x3 (Error correction complete) immediately 376 * after setting the 4BITECC_ADD_CALC_START bit. So if you immediately 377 * begin trying to poll for the state, you may fall right out of your 378 * loop without any of the correction calculations having taken place. 379 * The recommendation from the hardware team is to initially delay as 380 * long as ECC_STATE reads less than 4. After that, ECC HW has entered 381 * correction state. 382 */ 383 timeo = jiffies + usecs_to_jiffies(100); 384 do { 385 ecc_state = (davinci_nand_readl(info, 386 NANDFSR_OFFSET) >> 8) & 0x0f; 387 cpu_relax(); 388 } while ((ecc_state < 4) && time_before(jiffies, timeo)); 389 390 for (;;) { 391 u32 fsr = davinci_nand_readl(info, NANDFSR_OFFSET); 392 393 switch ((fsr >> 8) & 0x0f) { 394 case 0: /* no error, should not happen */ 395 davinci_nand_readl(info, NAND_ERR_ERRVAL1_OFFSET); 396 return 0; 397 case 1: /* five or more errors detected */ 398 davinci_nand_readl(info, NAND_ERR_ERRVAL1_OFFSET); 399 return -EBADMSG; 400 case 2: /* error addresses computed */ 401 case 3: 402 num_errors = 1 + ((fsr >> 16) & 0x03); 403 goto correct; 404 default: /* still working on it */ 405 cpu_relax(); 406 continue; 407 } 408 } 409 410 correct: 411 /* correct each error */ 412 for (i = 0, corrected = 0; i < num_errors; i++) { 413 int error_address, error_value; 414 415 if (i > 1) { 416 error_address = davinci_nand_readl(info, 417 NAND_ERR_ADD2_OFFSET); 418 error_value = davinci_nand_readl(info, 419 NAND_ERR_ERRVAL2_OFFSET); 420 } else { 421 error_address = davinci_nand_readl(info, 422 NAND_ERR_ADD1_OFFSET); 423 error_value = davinci_nand_readl(info, 424 NAND_ERR_ERRVAL1_OFFSET); 425 } 426 427 if (i & 1) { 428 error_address >>= 16; 429 error_value >>= 16; 430 } 431 error_address &= 0x3ff; 432 error_address = (512 + 7) - error_address; 433 434 if (error_address < 512) { 435 data[error_address] ^= error_value; 436 corrected++; 437 } 438 } 439 440 return corrected; 441 } 442 443 /*----------------------------------------------------------------------*/ 444 445 /* An ECC layout for using 4-bit ECC with small-page flash, storing 446 * ten ECC bytes plus the manufacturer's bad block marker byte, and 447 * and not overlapping the default BBT markers. 448 */ 449 static int hwecc4_ooblayout_small_ecc(struct mtd_info *mtd, int section, 450 struct mtd_oob_region *oobregion) 451 { 452 if (section > 2) 453 return -ERANGE; 454 455 if (!section) { 456 oobregion->offset = 0; 457 oobregion->length = 5; 458 } else if (section == 1) { 459 oobregion->offset = 6; 460 oobregion->length = 2; 461 } else { 462 oobregion->offset = 13; 463 oobregion->length = 3; 464 } 465 466 return 0; 467 } 468 469 static int hwecc4_ooblayout_small_free(struct mtd_info *mtd, int section, 470 struct mtd_oob_region *oobregion) 471 { 472 if (section > 1) 473 return -ERANGE; 474 475 if (!section) { 476 oobregion->offset = 8; 477 oobregion->length = 5; 478 } else { 479 oobregion->offset = 16; 480 oobregion->length = mtd->oobsize - 16; 481 } 482 483 return 0; 484 } 485 486 static const struct mtd_ooblayout_ops hwecc4_small_ooblayout_ops = { 487 .ecc = hwecc4_ooblayout_small_ecc, 488 .free = hwecc4_ooblayout_small_free, 489 }; 490 491 static int hwecc4_ooblayout_large_ecc(struct mtd_info *mtd, int section, 492 struct mtd_oob_region *oobregion) 493 { 494 struct nand_device *nand = mtd_to_nanddev(mtd); 495 unsigned int total_ecc_bytes = nand->ecc.ctx.total; 496 int nregions = total_ecc_bytes / 10; /* 10 bytes per chunk */ 497 498 if (section >= nregions) 499 return -ERANGE; 500 501 oobregion->offset = (section * 16) + 6; 502 oobregion->length = 10; 503 504 return 0; 505 } 506 507 static int hwecc4_ooblayout_large_free(struct mtd_info *mtd, int section, 508 struct mtd_oob_region *oobregion) 509 { 510 struct nand_device *nand = mtd_to_nanddev(mtd); 511 unsigned int total_ecc_bytes = nand->ecc.ctx.total; 512 int nregions = total_ecc_bytes / 10; /* 10 bytes per chunk */ 513 514 /* First region is used for BBT */ 515 if (section >= (nregions - 1)) 516 return -ERANGE; 517 518 oobregion->offset = ((section + 1) * 16); 519 oobregion->length = 6; 520 521 return 0; 522 } 523 524 static const struct mtd_ooblayout_ops hwecc4_large_ooblayout_ops = { 525 .ecc = hwecc4_ooblayout_large_ecc, 526 .free = hwecc4_ooblayout_large_free, 527 }; 528 529 #if defined(CONFIG_OF) 530 static const struct of_device_id davinci_nand_of_match[] = { 531 {.compatible = "ti,davinci-nand", }, 532 {.compatible = "ti,keystone-nand", }, 533 {}, 534 }; 535 MODULE_DEVICE_TABLE(of, davinci_nand_of_match); 536 537 static struct davinci_nand_pdata * 538 nand_davinci_get_pdata(struct platform_device *pdev) 539 { 540 if (!dev_get_platdata(&pdev->dev)) { 541 struct davinci_nand_pdata *pdata; 542 const char *mode; 543 u32 prop; 544 545 pdata = devm_kzalloc(&pdev->dev, 546 sizeof(struct davinci_nand_pdata), 547 GFP_KERNEL); 548 pdev->dev.platform_data = pdata; 549 if (!pdata) 550 return ERR_PTR(-ENOMEM); 551 if (!device_property_read_u32(&pdev->dev, 552 "ti,davinci-chipselect", &prop)) 553 pdata->core_chipsel = prop; 554 else 555 return ERR_PTR(-EINVAL); 556 557 if (!device_property_read_u32(&pdev->dev, 558 "ti,davinci-mask-ale", &prop)) 559 pdata->mask_ale = prop; 560 if (!device_property_read_u32(&pdev->dev, 561 "ti,davinci-mask-cle", &prop)) 562 pdata->mask_cle = prop; 563 if (!device_property_read_u32(&pdev->dev, 564 "ti,davinci-mask-chipsel", &prop)) 565 pdata->mask_chipsel = prop; 566 if (!device_property_read_string(&pdev->dev, 567 "ti,davinci-ecc-mode", 568 &mode)) { 569 if (!strncmp("none", mode, 4)) 570 pdata->engine_type = NAND_ECC_ENGINE_TYPE_NONE; 571 if (!strncmp("soft", mode, 4)) 572 pdata->engine_type = NAND_ECC_ENGINE_TYPE_SOFT; 573 if (!strncmp("hw", mode, 2)) 574 pdata->engine_type = NAND_ECC_ENGINE_TYPE_ON_HOST; 575 if (!strncmp("on-die", mode, 6)) 576 pdata->engine_type = NAND_ECC_ENGINE_TYPE_ON_DIE; 577 } 578 if (!device_property_read_u32(&pdev->dev, 579 "ti,davinci-ecc-bits", &prop)) 580 pdata->ecc_bits = prop; 581 582 if (!device_property_read_u32(&pdev->dev, 583 "ti,davinci-nand-buswidth", 584 &prop) && prop == 16) 585 pdata->options |= NAND_BUSWIDTH_16; 586 587 if (device_property_read_bool(&pdev->dev, 588 "ti,davinci-nand-use-bbt")) 589 pdata->bbt_options = NAND_BBT_USE_FLASH; 590 591 /* 592 * Since kernel v4.8, this driver has been fixed to enable 593 * use of 4-bit hardware ECC with subpages and verified on 594 * TI's keystone EVMs (K2L, K2HK and K2E). 595 * However, in the interest of not breaking systems using 596 * existing UBI partitions, sub-page writes are not being 597 * (re)enabled. If you want to use subpage writes on Keystone 598 * platforms (i.e. do not have any existing UBI partitions), 599 * then use "ti,davinci-nand" as the compatible in your 600 * device-tree file. 601 */ 602 if (device_is_compatible(&pdev->dev, "ti,keystone-nand")) 603 pdata->options |= NAND_NO_SUBPAGE_WRITE; 604 } 605 606 return dev_get_platdata(&pdev->dev); 607 } 608 #else 609 static struct davinci_nand_pdata * 610 nand_davinci_get_pdata(struct platform_device *pdev) 611 { 612 return dev_get_platdata(&pdev->dev); 613 } 614 #endif 615 616 static int davinci_nand_attach_chip(struct nand_chip *chip) 617 { 618 struct mtd_info *mtd = nand_to_mtd(chip); 619 struct davinci_nand_info *info = to_davinci_nand(mtd); 620 struct davinci_nand_pdata *pdata = nand_davinci_get_pdata(info->pdev); 621 int ret = 0; 622 623 if (IS_ERR(pdata)) 624 return PTR_ERR(pdata); 625 626 /* Use board-specific ECC config */ 627 chip->ecc.engine_type = pdata->engine_type; 628 chip->ecc.placement = pdata->ecc_placement; 629 630 switch (chip->ecc.engine_type) { 631 case NAND_ECC_ENGINE_TYPE_NONE: 632 case NAND_ECC_ENGINE_TYPE_ON_DIE: 633 pdata->ecc_bits = 0; 634 break; 635 case NAND_ECC_ENGINE_TYPE_SOFT: 636 pdata->ecc_bits = 0; 637 /* 638 * This driver expects Hamming based ECC when engine_type is set 639 * to NAND_ECC_ENGINE_TYPE_SOFT. Force ecc.algo to 640 * NAND_ECC_ALGO_HAMMING to avoid adding an extra ->ecc_algo 641 * field to davinci_nand_pdata. 642 */ 643 chip->ecc.algo = NAND_ECC_ALGO_HAMMING; 644 break; 645 case NAND_ECC_ENGINE_TYPE_ON_HOST: 646 if (pdata->ecc_bits == 4) { 647 int chunks = mtd->writesize / 512; 648 649 if (!chunks || mtd->oobsize < 16) { 650 dev_dbg(&info->pdev->dev, "too small\n"); 651 return -EINVAL; 652 } 653 654 /* 655 * No sanity checks: CPUs must support this, 656 * and the chips may not use NAND_BUSWIDTH_16. 657 */ 658 659 /* No sharing 4-bit hardware between chipselects yet */ 660 spin_lock_irq(&davinci_nand_lock); 661 if (ecc4_busy) 662 ret = -EBUSY; 663 else 664 ecc4_busy = true; 665 spin_unlock_irq(&davinci_nand_lock); 666 667 if (ret == -EBUSY) 668 return ret; 669 670 chip->ecc.calculate = nand_davinci_calculate_4bit; 671 chip->ecc.correct = nand_davinci_correct_4bit; 672 chip->ecc.hwctl = nand_davinci_hwctl_4bit; 673 chip->ecc.bytes = 10; 674 chip->ecc.options = NAND_ECC_GENERIC_ERASED_CHECK; 675 chip->ecc.algo = NAND_ECC_ALGO_BCH; 676 677 /* 678 * Update ECC layout if needed ... for 1-bit HW ECC, the 679 * default is OK, but it allocates 6 bytes when only 3 680 * are needed (for each 512 bytes). For 4-bit HW ECC, 681 * the default is not usable: 10 bytes needed, not 6. 682 * 683 * For small page chips, preserve the manufacturer's 684 * badblock marking data ... and make sure a flash BBT 685 * table marker fits in the free bytes. 686 */ 687 if (chunks == 1) { 688 mtd_set_ooblayout(mtd, 689 &hwecc4_small_ooblayout_ops); 690 } else if (chunks == 4 || chunks == 8) { 691 chip->ecc.read_page = nand_read_page_hwecc_oob_first; 692 693 if (chip->options & NAND_IS_BOOT_MEDIUM) 694 mtd_set_ooblayout(mtd, &hwecc4_large_ooblayout_ops); 695 else 696 mtd_set_ooblayout(mtd, nand_get_large_page_ooblayout()); 697 } else { 698 return -EIO; 699 } 700 } else { 701 /* 1bit ecc hamming */ 702 chip->ecc.calculate = nand_davinci_calculate_1bit; 703 chip->ecc.correct = nand_davinci_correct_1bit; 704 chip->ecc.hwctl = nand_davinci_hwctl_1bit; 705 chip->ecc.bytes = 3; 706 chip->ecc.algo = NAND_ECC_ALGO_HAMMING; 707 } 708 chip->ecc.size = 512; 709 chip->ecc.strength = pdata->ecc_bits; 710 break; 711 default: 712 return -EINVAL; 713 } 714 715 return ret; 716 } 717 718 static void nand_davinci_data_in(struct davinci_nand_info *info, void *buf, 719 unsigned int len, bool force_8bit) 720 { 721 u32 alignment = ((uintptr_t)buf | len) & 3; 722 723 if (force_8bit || (alignment & 1)) 724 ioread8_rep(info->current_cs, buf, len); 725 else if (alignment & 3) 726 ioread16_rep(info->current_cs, buf, len >> 1); 727 else 728 ioread32_rep(info->current_cs, buf, len >> 2); 729 } 730 731 static void nand_davinci_data_out(struct davinci_nand_info *info, 732 const void *buf, unsigned int len, 733 bool force_8bit) 734 { 735 u32 alignment = ((uintptr_t)buf | len) & 3; 736 737 if (force_8bit || (alignment & 1)) 738 iowrite8_rep(info->current_cs, buf, len); 739 else if (alignment & 3) 740 iowrite16_rep(info->current_cs, buf, len >> 1); 741 else 742 iowrite32_rep(info->current_cs, buf, len >> 2); 743 } 744 745 static int davinci_nand_exec_instr(struct davinci_nand_info *info, 746 const struct nand_op_instr *instr) 747 { 748 unsigned int i, timeout_us; 749 u32 status; 750 int ret; 751 752 switch (instr->type) { 753 case NAND_OP_CMD_INSTR: 754 iowrite8(instr->ctx.cmd.opcode, 755 info->current_cs + info->mask_cle); 756 break; 757 758 case NAND_OP_ADDR_INSTR: 759 for (i = 0; i < instr->ctx.addr.naddrs; i++) { 760 iowrite8(instr->ctx.addr.addrs[i], 761 info->current_cs + info->mask_ale); 762 } 763 break; 764 765 case NAND_OP_DATA_IN_INSTR: 766 nand_davinci_data_in(info, instr->ctx.data.buf.in, 767 instr->ctx.data.len, 768 instr->ctx.data.force_8bit); 769 break; 770 771 case NAND_OP_DATA_OUT_INSTR: 772 nand_davinci_data_out(info, instr->ctx.data.buf.out, 773 instr->ctx.data.len, 774 instr->ctx.data.force_8bit); 775 break; 776 777 case NAND_OP_WAITRDY_INSTR: 778 timeout_us = instr->ctx.waitrdy.timeout_ms * 1000; 779 ret = readl_relaxed_poll_timeout(info->base + NANDFSR_OFFSET, 780 status, status & BIT(0), 5, 781 timeout_us); 782 if (ret) 783 return ret; 784 785 break; 786 } 787 788 if (instr->delay_ns) { 789 /* Dummy read to be sure that command is sent before ndelay starts */ 790 davinci_nand_readl(info, 0); 791 ndelay(instr->delay_ns); 792 } 793 794 return 0; 795 } 796 797 static int davinci_nand_exec_op(struct nand_chip *chip, 798 const struct nand_operation *op, 799 bool check_only) 800 { 801 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip)); 802 unsigned int i; 803 804 if (check_only) 805 return 0; 806 807 info->current_cs = info->vaddr + (op->cs * info->mask_chipsel); 808 809 for (i = 0; i < op->ninstrs; i++) { 810 int ret; 811 812 ret = davinci_nand_exec_instr(info, &op->instrs[i]); 813 if (ret) 814 return ret; 815 } 816 817 return 0; 818 } 819 820 #define TO_CYCLES(ps, period_ns) (DIV_ROUND_UP((ps) / 1000, (period_ns))) 821 822 static int davinci_nand_setup_interface(struct nand_chip *chip, int chipnr, 823 const struct nand_interface_config *conf) 824 { 825 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip)); 826 const struct nand_sdr_timings *sdr; 827 struct aemif_cs_timings timings; 828 s32 cfg, min, cyc_ns; 829 int ret; 830 831 cyc_ns = 1000000000 / clk_get_rate(info->clk); 832 833 sdr = nand_get_sdr_timings(conf); 834 if (IS_ERR(sdr)) 835 return PTR_ERR(sdr); 836 837 cfg = TO_CYCLES(sdr->tCLR_min, cyc_ns) - 1; 838 timings.rsetup = cfg > 0 ? cfg : 0; 839 840 cfg = max_t(s32, TO_CYCLES(sdr->tREA_max + MAX_TSU_PS, cyc_ns), 841 TO_CYCLES(sdr->tRP_min, cyc_ns)) - 1; 842 timings.rstrobe = cfg > 0 ? cfg : 0; 843 844 min = TO_CYCLES(sdr->tCEA_max + MAX_TSU_PS, cyc_ns) - 2; 845 while ((s32)(timings.rsetup + timings.rstrobe) < min) 846 timings.rstrobe++; 847 848 cfg = TO_CYCLES((s32)(MAX_TH_PS - sdr->tCHZ_max), cyc_ns) - 1; 849 timings.rhold = cfg > 0 ? cfg : 0; 850 851 min = TO_CYCLES(sdr->tRC_min, cyc_ns) - 3; 852 while ((s32)(timings.rsetup + timings.rstrobe + timings.rhold) < min) 853 timings.rhold++; 854 855 cfg = TO_CYCLES((s32)(sdr->tRHZ_max - (timings.rhold + 1) * cyc_ns * 1000), cyc_ns); 856 cfg = max_t(s32, cfg, TO_CYCLES(sdr->tCHZ_max, cyc_ns)) - 1; 857 timings.ta = cfg > 0 ? cfg : 0; 858 859 cfg = TO_CYCLES(sdr->tWP_min, cyc_ns) - 1; 860 timings.wstrobe = cfg > 0 ? cfg : 0; 861 862 cfg = max_t(s32, TO_CYCLES(sdr->tCLS_min, cyc_ns), TO_CYCLES(sdr->tALS_min, cyc_ns)); 863 cfg = max_t(s32, cfg, TO_CYCLES(sdr->tCS_min, cyc_ns)) - 1; 864 timings.wsetup = cfg > 0 ? cfg : 0; 865 866 min = TO_CYCLES(sdr->tDS_min, cyc_ns) - 2; 867 while ((s32)(timings.wsetup + timings.wstrobe) < min) 868 timings.wstrobe++; 869 870 cfg = max_t(s32, TO_CYCLES(sdr->tCLH_min, cyc_ns), TO_CYCLES(sdr->tALH_min, cyc_ns)); 871 cfg = max_t(s32, cfg, TO_CYCLES(sdr->tCH_min, cyc_ns)); 872 cfg = max_t(s32, cfg, TO_CYCLES(sdr->tDH_min, cyc_ns)) - 1; 873 timings.whold = cfg > 0 ? cfg : 0; 874 875 min = TO_CYCLES(sdr->tWC_min, cyc_ns) - 2; 876 while ((s32)(timings.wsetup + timings.wstrobe + timings.whold) < min) 877 timings.whold++; 878 879 dev_dbg(&info->pdev->dev, "RSETUP %x RSTROBE %x RHOLD %x\n", 880 timings.rsetup, timings.rstrobe, timings.rhold); 881 dev_dbg(&info->pdev->dev, "TA %x\n", timings.ta); 882 dev_dbg(&info->pdev->dev, "WSETUP %x WSTROBE %x WHOLD %x\n", 883 timings.wsetup, timings.wstrobe, timings.whold); 884 885 ret = aemif_check_cs_timings(&timings); 886 if (ret || chipnr == NAND_DATA_IFACE_CHECK_ONLY) 887 return ret; 888 889 return aemif_set_cs_timings(info->aemif, info->core_chipsel, &timings); 890 } 891 892 static const struct nand_controller_ops davinci_nand_controller_ops = { 893 .attach_chip = davinci_nand_attach_chip, 894 .exec_op = davinci_nand_exec_op, 895 .setup_interface = davinci_nand_setup_interface, 896 }; 897 898 static int nand_davinci_probe(struct platform_device *pdev) 899 { 900 struct davinci_nand_pdata *pdata; 901 struct davinci_nand_info *info; 902 struct resource *res1; 903 struct resource *res2; 904 void __iomem *vaddr; 905 void __iomem *base; 906 int ret; 907 uint32_t val; 908 struct mtd_info *mtd; 909 910 pdata = nand_davinci_get_pdata(pdev); 911 if (IS_ERR(pdata)) 912 return PTR_ERR(pdata); 913 914 /* insist on board-specific configuration */ 915 if (!pdata) 916 return -ENODEV; 917 918 /* which external chipselect will we be managing? */ 919 if (pdata->core_chipsel > 3) 920 return -ENODEV; 921 922 info = devm_kzalloc(&pdev->dev, sizeof(*info), GFP_KERNEL); 923 if (!info) 924 return -ENOMEM; 925 926 platform_set_drvdata(pdev, info); 927 928 res1 = platform_get_resource(pdev, IORESOURCE_MEM, 0); 929 res2 = platform_get_resource(pdev, IORESOURCE_MEM, 1); 930 if (!res1 || !res2) { 931 dev_err(&pdev->dev, "resource missing\n"); 932 return -EINVAL; 933 } 934 935 vaddr = devm_ioremap_resource(&pdev->dev, res1); 936 if (IS_ERR(vaddr)) 937 return PTR_ERR(vaddr); 938 939 /* 940 * This registers range is used to setup NAND settings. In case with 941 * TI AEMIF driver, the same memory address range is requested already 942 * by AEMIF, so we cannot request it twice, just ioremap. 943 * The AEMIF and NAND drivers not use the same registers in this range. 944 */ 945 base = devm_ioremap(&pdev->dev, res2->start, resource_size(res2)); 946 if (!base) { 947 dev_err(&pdev->dev, "ioremap failed for resource %pR\n", res2); 948 return -EADDRNOTAVAIL; 949 } 950 951 info->clk = devm_clk_get_enabled(&pdev->dev, "aemif"); 952 if (IS_ERR(info->clk)) 953 return dev_err_probe(&pdev->dev, PTR_ERR(info->clk), "failed to get clock"); 954 955 info->pdev = pdev; 956 info->base = base; 957 info->vaddr = vaddr; 958 info->aemif = dev_get_drvdata(pdev->dev.parent); 959 960 mtd = nand_to_mtd(&info->chip); 961 mtd->dev.parent = &pdev->dev; 962 nand_set_flash_node(&info->chip, pdev->dev.of_node); 963 964 /* options such as NAND_BBT_USE_FLASH */ 965 info->chip.bbt_options = pdata->bbt_options; 966 /* options such as 16-bit widths */ 967 info->chip.options = pdata->options; 968 info->chip.bbt_td = pdata->bbt_td; 969 info->chip.bbt_md = pdata->bbt_md; 970 971 info->current_cs = info->vaddr; 972 info->core_chipsel = pdata->core_chipsel; 973 info->mask_chipsel = pdata->mask_chipsel; 974 975 /* use nandboot-capable ALE/CLE masks by default */ 976 info->mask_ale = pdata->mask_ale ? : MASK_ALE; 977 info->mask_cle = pdata->mask_cle ? : MASK_CLE; 978 979 spin_lock_irq(&davinci_nand_lock); 980 981 /* put CSxNAND into NAND mode */ 982 val = davinci_nand_readl(info, NANDFCR_OFFSET); 983 val |= BIT(info->core_chipsel); 984 davinci_nand_writel(info, NANDFCR_OFFSET, val); 985 986 spin_unlock_irq(&davinci_nand_lock); 987 988 /* Scan to find existence of the device(s) */ 989 nand_controller_init(&info->controller); 990 info->controller.ops = &davinci_nand_controller_ops; 991 info->chip.controller = &info->controller; 992 ret = nand_scan(&info->chip, pdata->mask_chipsel ? 2 : 1); 993 if (ret < 0) { 994 dev_dbg(&pdev->dev, "no NAND chip(s) found\n"); 995 return ret; 996 } 997 998 if (pdata->parts) 999 ret = mtd_device_register(mtd, pdata->parts, pdata->nr_parts); 1000 else 1001 ret = mtd_device_register(mtd, NULL, 0); 1002 if (ret < 0) 1003 goto err_cleanup_nand; 1004 1005 val = davinci_nand_readl(info, NRCSR_OFFSET); 1006 dev_info(&pdev->dev, "controller rev. %d.%d\n", 1007 (val >> 8) & 0xff, val & 0xff); 1008 1009 return 0; 1010 1011 err_cleanup_nand: 1012 nand_cleanup(&info->chip); 1013 1014 return ret; 1015 } 1016 1017 static void nand_davinci_remove(struct platform_device *pdev) 1018 { 1019 struct davinci_nand_info *info = platform_get_drvdata(pdev); 1020 struct nand_chip *chip = &info->chip; 1021 int ret; 1022 1023 spin_lock_irq(&davinci_nand_lock); 1024 if (chip->ecc.placement == NAND_ECC_PLACEMENT_INTERLEAVED) 1025 ecc4_busy = false; 1026 spin_unlock_irq(&davinci_nand_lock); 1027 1028 ret = mtd_device_unregister(nand_to_mtd(chip)); 1029 WARN_ON(ret); 1030 nand_cleanup(chip); 1031 } 1032 1033 static struct platform_driver nand_davinci_driver = { 1034 .probe = nand_davinci_probe, 1035 .remove = nand_davinci_remove, 1036 .driver = { 1037 .name = "davinci_nand", 1038 .of_match_table = of_match_ptr(davinci_nand_of_match), 1039 }, 1040 }; 1041 MODULE_ALIAS("platform:davinci_nand"); 1042 1043 module_platform_driver(nand_davinci_driver); 1044 1045 MODULE_LICENSE("GPL"); 1046 MODULE_AUTHOR("Texas Instruments"); 1047 MODULE_DESCRIPTION("Davinci NAND flash driver"); 1048 1049