Loading...
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/kernel.h>
14#include <linux/module.h>
15#include <linux/platform_device.h>
16#include <linux/err.h>
17#include <linux/iopoll.h>
18#include <linux/mtd/rawnand.h>
19#include <linux/mtd/partitions.h>
20#include <linux/slab.h>
21#include <linux/of.h>
22
23#include <linux/platform_data/mtd-davinci.h>
24#include <linux/platform_data/mtd-davinci-aemif.h>
25
26/*
27 * This is a device driver for the NAND flash controller found on the
28 * various DaVinci family chips. It handles up to four SoC chipselects,
29 * and some flavors of secondary chipselect (e.g. based on A12) as used
30 * with multichip packages.
31 *
32 * The 1-bit ECC hardware is supported, as well as the newer 4-bit ECC
33 * available on chips like the DM355 and OMAP-L137 and needed with the
34 * more error-prone MLC NAND chips.
35 *
36 * This driver assumes EM_WAIT connects all the NAND devices' RDY/nBUSY
37 * outputs in a "wire-AND" configuration, with no per-chip signals.
38 */
39struct davinci_nand_info {
40 struct nand_controller controller;
41 struct nand_chip chip;
42
43 struct platform_device *pdev;
44
45 bool is_readmode;
46
47 void __iomem *base;
48 void __iomem *vaddr;
49
50 void __iomem *current_cs;
51
52 uint32_t mask_chipsel;
53 uint32_t mask_ale;
54 uint32_t mask_cle;
55
56 uint32_t core_chipsel;
57
58 struct davinci_aemif_timing *timing;
59};
60
61static DEFINE_SPINLOCK(davinci_nand_lock);
62static bool ecc4_busy;
63
64static inline struct davinci_nand_info *to_davinci_nand(struct mtd_info *mtd)
65{
66 return container_of(mtd_to_nand(mtd), struct davinci_nand_info, chip);
67}
68
69static inline unsigned int davinci_nand_readl(struct davinci_nand_info *info,
70 int offset)
71{
72 return __raw_readl(info->base + offset);
73}
74
75static inline void davinci_nand_writel(struct davinci_nand_info *info,
76 int offset, unsigned long value)
77{
78 __raw_writel(value, info->base + offset);
79}
80
81/*----------------------------------------------------------------------*/
82
83/*
84 * 1-bit hardware ECC ... context maintained for each core chipselect
85 */
86
87static inline uint32_t nand_davinci_readecc_1bit(struct mtd_info *mtd)
88{
89 struct davinci_nand_info *info = to_davinci_nand(mtd);
90
91 return davinci_nand_readl(info, NANDF1ECC_OFFSET
92 + 4 * info->core_chipsel);
93}
94
95static void nand_davinci_hwctl_1bit(struct nand_chip *chip, int mode)
96{
97 struct davinci_nand_info *info;
98 uint32_t nandcfr;
99 unsigned long flags;
100
101 info = to_davinci_nand(nand_to_mtd(chip));
102
103 /* Reset ECC hardware */
104 nand_davinci_readecc_1bit(nand_to_mtd(chip));
105
106 spin_lock_irqsave(&davinci_nand_lock, flags);
107
108 /* Restart ECC hardware */
109 nandcfr = davinci_nand_readl(info, NANDFCR_OFFSET);
110 nandcfr |= BIT(8 + info->core_chipsel);
111 davinci_nand_writel(info, NANDFCR_OFFSET, nandcfr);
112
113 spin_unlock_irqrestore(&davinci_nand_lock, flags);
114}
115
116/*
117 * Read hardware ECC value and pack into three bytes
118 */
119static int nand_davinci_calculate_1bit(struct nand_chip *chip,
120 const u_char *dat, u_char *ecc_code)
121{
122 unsigned int ecc_val = nand_davinci_readecc_1bit(nand_to_mtd(chip));
123 unsigned int ecc24 = (ecc_val & 0x0fff) | ((ecc_val & 0x0fff0000) >> 4);
124
125 /* invert so that erased block ecc is correct */
126 ecc24 = ~ecc24;
127 ecc_code[0] = (u_char)(ecc24);
128 ecc_code[1] = (u_char)(ecc24 >> 8);
129 ecc_code[2] = (u_char)(ecc24 >> 16);
130
131 return 0;
132}
133
134static int nand_davinci_correct_1bit(struct nand_chip *chip, u_char *dat,
135 u_char *read_ecc, u_char *calc_ecc)
136{
137 uint32_t eccNand = read_ecc[0] | (read_ecc[1] << 8) |
138 (read_ecc[2] << 16);
139 uint32_t eccCalc = calc_ecc[0] | (calc_ecc[1] << 8) |
140 (calc_ecc[2] << 16);
141 uint32_t diff = eccCalc ^ eccNand;
142
143 if (diff) {
144 if ((((diff >> 12) ^ diff) & 0xfff) == 0xfff) {
145 /* Correctable error */
146 if ((diff >> (12 + 3)) < chip->ecc.size) {
147 dat[diff >> (12 + 3)] ^= BIT((diff >> 12) & 7);
148 return 1;
149 } else {
150 return -EBADMSG;
151 }
152 } else if (!(diff & (diff - 1))) {
153 /* Single bit ECC error in the ECC itself,
154 * nothing to fix */
155 return 1;
156 } else {
157 /* Uncorrectable error */
158 return -EBADMSG;
159 }
160
161 }
162 return 0;
163}
164
165/*----------------------------------------------------------------------*/
166
167/*
168 * 4-bit hardware ECC ... context maintained over entire AEMIF
169 *
170 * This is a syndrome engine, but we avoid NAND_ECC_PLACEMENT_INTERLEAVED
171 * since that forces use of a problematic "infix OOB" layout.
172 * Among other things, it trashes manufacturer bad block markers.
173 * Also, and specific to this hardware, it ECC-protects the "prepad"
174 * in the OOB ... while having ECC protection for parts of OOB would
175 * seem useful, the current MTD stack sometimes wants to update the
176 * OOB without recomputing ECC.
177 */
178
179static void nand_davinci_hwctl_4bit(struct nand_chip *chip, int mode)
180{
181 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip));
182 unsigned long flags;
183 u32 val;
184
185 /* Reset ECC hardware */
186 davinci_nand_readl(info, NAND_4BIT_ECC1_OFFSET);
187
188 spin_lock_irqsave(&davinci_nand_lock, flags);
189
190 /* Start 4-bit ECC calculation for read/write */
191 val = davinci_nand_readl(info, NANDFCR_OFFSET);
192 val &= ~(0x03 << 4);
193 val |= (info->core_chipsel << 4) | BIT(12);
194 davinci_nand_writel(info, NANDFCR_OFFSET, val);
195
196 info->is_readmode = (mode == NAND_ECC_READ);
197
198 spin_unlock_irqrestore(&davinci_nand_lock, flags);
199}
200
201/* Read raw ECC code after writing to NAND. */
202static void
203nand_davinci_readecc_4bit(struct davinci_nand_info *info, u32 code[4])
204{
205 const u32 mask = 0x03ff03ff;
206
207 code[0] = davinci_nand_readl(info, NAND_4BIT_ECC1_OFFSET) & mask;
208 code[1] = davinci_nand_readl(info, NAND_4BIT_ECC2_OFFSET) & mask;
209 code[2] = davinci_nand_readl(info, NAND_4BIT_ECC3_OFFSET) & mask;
210 code[3] = davinci_nand_readl(info, NAND_4BIT_ECC4_OFFSET) & mask;
211}
212
213/* Terminate read ECC; or return ECC (as bytes) of data written to NAND. */
214static int nand_davinci_calculate_4bit(struct nand_chip *chip,
215 const u_char *dat, u_char *ecc_code)
216{
217 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip));
218 u32 raw_ecc[4], *p;
219 unsigned i;
220
221 /* After a read, terminate ECC calculation by a dummy read
222 * of some 4-bit ECC register. ECC covers everything that
223 * was read; correct() just uses the hardware state, so
224 * ecc_code is not needed.
225 */
226 if (info->is_readmode) {
227 davinci_nand_readl(info, NAND_4BIT_ECC1_OFFSET);
228 return 0;
229 }
230
231 /* Pack eight raw 10-bit ecc values into ten bytes, making
232 * two passes which each convert four values (in upper and
233 * lower halves of two 32-bit words) into five bytes. The
234 * ROM boot loader uses this same packing scheme.
235 */
236 nand_davinci_readecc_4bit(info, raw_ecc);
237 for (i = 0, p = raw_ecc; i < 2; i++, p += 2) {
238 *ecc_code++ = p[0] & 0xff;
239 *ecc_code++ = ((p[0] >> 8) & 0x03) | ((p[0] >> 14) & 0xfc);
240 *ecc_code++ = ((p[0] >> 22) & 0x0f) | ((p[1] << 4) & 0xf0);
241 *ecc_code++ = ((p[1] >> 4) & 0x3f) | ((p[1] >> 10) & 0xc0);
242 *ecc_code++ = (p[1] >> 18) & 0xff;
243 }
244
245 return 0;
246}
247
248/* Correct up to 4 bits in data we just read, using state left in the
249 * hardware plus the ecc_code computed when it was first written.
250 */
251static int nand_davinci_correct_4bit(struct nand_chip *chip, u_char *data,
252 u_char *ecc_code, u_char *null)
253{
254 int i;
255 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip));
256 unsigned short ecc10[8];
257 unsigned short *ecc16;
258 u32 syndrome[4];
259 u32 ecc_state;
260 unsigned num_errors, corrected;
261 unsigned long timeo;
262
263 /* Unpack ten bytes into eight 10 bit values. We know we're
264 * little-endian, and use type punning for less shifting/masking.
265 */
266 if (WARN_ON(0x01 & (uintptr_t)ecc_code))
267 return -EINVAL;
268 ecc16 = (unsigned short *)ecc_code;
269
270 ecc10[0] = (ecc16[0] >> 0) & 0x3ff;
271 ecc10[1] = ((ecc16[0] >> 10) & 0x3f) | ((ecc16[1] << 6) & 0x3c0);
272 ecc10[2] = (ecc16[1] >> 4) & 0x3ff;
273 ecc10[3] = ((ecc16[1] >> 14) & 0x3) | ((ecc16[2] << 2) & 0x3fc);
274 ecc10[4] = (ecc16[2] >> 8) | ((ecc16[3] << 8) & 0x300);
275 ecc10[5] = (ecc16[3] >> 2) & 0x3ff;
276 ecc10[6] = ((ecc16[3] >> 12) & 0xf) | ((ecc16[4] << 4) & 0x3f0);
277 ecc10[7] = (ecc16[4] >> 6) & 0x3ff;
278
279 /* Tell ECC controller about the expected ECC codes. */
280 for (i = 7; i >= 0; i--)
281 davinci_nand_writel(info, NAND_4BIT_ECC_LOAD_OFFSET, ecc10[i]);
282
283 /* Allow time for syndrome calculation ... then read it.
284 * A syndrome of all zeroes 0 means no detected errors.
285 */
286 davinci_nand_readl(info, NANDFSR_OFFSET);
287 nand_davinci_readecc_4bit(info, syndrome);
288 if (!(syndrome[0] | syndrome[1] | syndrome[2] | syndrome[3]))
289 return 0;
290
291 /*
292 * Clear any previous address calculation by doing a dummy read of an
293 * error address register.
294 */
295 davinci_nand_readl(info, NAND_ERR_ADD1_OFFSET);
296
297 /* Start address calculation, and wait for it to complete.
298 * We _could_ start reading more data while this is working,
299 * to speed up the overall page read.
300 */
301 davinci_nand_writel(info, NANDFCR_OFFSET,
302 davinci_nand_readl(info, NANDFCR_OFFSET) | BIT(13));
303
304 /*
305 * ECC_STATE field reads 0x3 (Error correction complete) immediately
306 * after setting the 4BITECC_ADD_CALC_START bit. So if you immediately
307 * begin trying to poll for the state, you may fall right out of your
308 * loop without any of the correction calculations having taken place.
309 * The recommendation from the hardware team is to initially delay as
310 * long as ECC_STATE reads less than 4. After that, ECC HW has entered
311 * correction state.
312 */
313 timeo = jiffies + usecs_to_jiffies(100);
314 do {
315 ecc_state = (davinci_nand_readl(info,
316 NANDFSR_OFFSET) >> 8) & 0x0f;
317 cpu_relax();
318 } while ((ecc_state < 4) && time_before(jiffies, timeo));
319
320 for (;;) {
321 u32 fsr = davinci_nand_readl(info, NANDFSR_OFFSET);
322
323 switch ((fsr >> 8) & 0x0f) {
324 case 0: /* no error, should not happen */
325 davinci_nand_readl(info, NAND_ERR_ERRVAL1_OFFSET);
326 return 0;
327 case 1: /* five or more errors detected */
328 davinci_nand_readl(info, NAND_ERR_ERRVAL1_OFFSET);
329 return -EBADMSG;
330 case 2: /* error addresses computed */
331 case 3:
332 num_errors = 1 + ((fsr >> 16) & 0x03);
333 goto correct;
334 default: /* still working on it */
335 cpu_relax();
336 continue;
337 }
338 }
339
340correct:
341 /* correct each error */
342 for (i = 0, corrected = 0; i < num_errors; i++) {
343 int error_address, error_value;
344
345 if (i > 1) {
346 error_address = davinci_nand_readl(info,
347 NAND_ERR_ADD2_OFFSET);
348 error_value = davinci_nand_readl(info,
349 NAND_ERR_ERRVAL2_OFFSET);
350 } else {
351 error_address = davinci_nand_readl(info,
352 NAND_ERR_ADD1_OFFSET);
353 error_value = davinci_nand_readl(info,
354 NAND_ERR_ERRVAL1_OFFSET);
355 }
356
357 if (i & 1) {
358 error_address >>= 16;
359 error_value >>= 16;
360 }
361 error_address &= 0x3ff;
362 error_address = (512 + 7) - error_address;
363
364 if (error_address < 512) {
365 data[error_address] ^= error_value;
366 corrected++;
367 }
368 }
369
370 return corrected;
371}
372
373/*----------------------------------------------------------------------*/
374
375/* An ECC layout for using 4-bit ECC with small-page flash, storing
376 * ten ECC bytes plus the manufacturer's bad block marker byte, and
377 * and not overlapping the default BBT markers.
378 */
379static int hwecc4_ooblayout_small_ecc(struct mtd_info *mtd, int section,
380 struct mtd_oob_region *oobregion)
381{
382 if (section > 2)
383 return -ERANGE;
384
385 if (!section) {
386 oobregion->offset = 0;
387 oobregion->length = 5;
388 } else if (section == 1) {
389 oobregion->offset = 6;
390 oobregion->length = 2;
391 } else {
392 oobregion->offset = 13;
393 oobregion->length = 3;
394 }
395
396 return 0;
397}
398
399static int hwecc4_ooblayout_small_free(struct mtd_info *mtd, int section,
400 struct mtd_oob_region *oobregion)
401{
402 if (section > 1)
403 return -ERANGE;
404
405 if (!section) {
406 oobregion->offset = 8;
407 oobregion->length = 5;
408 } else {
409 oobregion->offset = 16;
410 oobregion->length = mtd->oobsize - 16;
411 }
412
413 return 0;
414}
415
416static const struct mtd_ooblayout_ops hwecc4_small_ooblayout_ops = {
417 .ecc = hwecc4_ooblayout_small_ecc,
418 .free = hwecc4_ooblayout_small_free,
419};
420
421#if defined(CONFIG_OF)
422static const struct of_device_id davinci_nand_of_match[] = {
423 {.compatible = "ti,davinci-nand", },
424 {.compatible = "ti,keystone-nand", },
425 {},
426};
427MODULE_DEVICE_TABLE(of, davinci_nand_of_match);
428
429static struct davinci_nand_pdata
430 *nand_davinci_get_pdata(struct platform_device *pdev)
431{
432 if (!dev_get_platdata(&pdev->dev) && pdev->dev.of_node) {
433 struct davinci_nand_pdata *pdata;
434 const char *mode;
435 u32 prop;
436
437 pdata = devm_kzalloc(&pdev->dev,
438 sizeof(struct davinci_nand_pdata),
439 GFP_KERNEL);
440 pdev->dev.platform_data = pdata;
441 if (!pdata)
442 return ERR_PTR(-ENOMEM);
443 if (!of_property_read_u32(pdev->dev.of_node,
444 "ti,davinci-chipselect", &prop))
445 pdata->core_chipsel = prop;
446 else
447 return ERR_PTR(-EINVAL);
448
449 if (!of_property_read_u32(pdev->dev.of_node,
450 "ti,davinci-mask-ale", &prop))
451 pdata->mask_ale = prop;
452 if (!of_property_read_u32(pdev->dev.of_node,
453 "ti,davinci-mask-cle", &prop))
454 pdata->mask_cle = prop;
455 if (!of_property_read_u32(pdev->dev.of_node,
456 "ti,davinci-mask-chipsel", &prop))
457 pdata->mask_chipsel = prop;
458 if (!of_property_read_string(pdev->dev.of_node,
459 "ti,davinci-ecc-mode", &mode)) {
460 if (!strncmp("none", mode, 4))
461 pdata->engine_type = NAND_ECC_ENGINE_TYPE_NONE;
462 if (!strncmp("soft", mode, 4))
463 pdata->engine_type = NAND_ECC_ENGINE_TYPE_SOFT;
464 if (!strncmp("hw", mode, 2))
465 pdata->engine_type = NAND_ECC_ENGINE_TYPE_ON_HOST;
466 }
467 if (!of_property_read_u32(pdev->dev.of_node,
468 "ti,davinci-ecc-bits", &prop))
469 pdata->ecc_bits = prop;
470
471 if (!of_property_read_u32(pdev->dev.of_node,
472 "ti,davinci-nand-buswidth", &prop) && prop == 16)
473 pdata->options |= NAND_BUSWIDTH_16;
474
475 if (of_property_read_bool(pdev->dev.of_node,
476 "ti,davinci-nand-use-bbt"))
477 pdata->bbt_options = NAND_BBT_USE_FLASH;
478
479 /*
480 * Since kernel v4.8, this driver has been fixed to enable
481 * use of 4-bit hardware ECC with subpages and verified on
482 * TI's keystone EVMs (K2L, K2HK and K2E).
483 * However, in the interest of not breaking systems using
484 * existing UBI partitions, sub-page writes are not being
485 * (re)enabled. If you want to use subpage writes on Keystone
486 * platforms (i.e. do not have any existing UBI partitions),
487 * then use "ti,davinci-nand" as the compatible in your
488 * device-tree file.
489 */
490 if (of_device_is_compatible(pdev->dev.of_node,
491 "ti,keystone-nand")) {
492 pdata->options |= NAND_NO_SUBPAGE_WRITE;
493 }
494 }
495
496 return dev_get_platdata(&pdev->dev);
497}
498#else
499static struct davinci_nand_pdata
500 *nand_davinci_get_pdata(struct platform_device *pdev)
501{
502 return dev_get_platdata(&pdev->dev);
503}
504#endif
505
506static int davinci_nand_attach_chip(struct nand_chip *chip)
507{
508 struct mtd_info *mtd = nand_to_mtd(chip);
509 struct davinci_nand_info *info = to_davinci_nand(mtd);
510 struct davinci_nand_pdata *pdata = nand_davinci_get_pdata(info->pdev);
511 int ret = 0;
512
513 if (IS_ERR(pdata))
514 return PTR_ERR(pdata);
515
516 /* Use board-specific ECC config */
517 chip->ecc.engine_type = pdata->engine_type;
518 chip->ecc.placement = pdata->ecc_placement;
519
520 switch (chip->ecc.engine_type) {
521 case NAND_ECC_ENGINE_TYPE_NONE:
522 pdata->ecc_bits = 0;
523 break;
524 case NAND_ECC_ENGINE_TYPE_SOFT:
525 pdata->ecc_bits = 0;
526 /*
527 * This driver expects Hamming based ECC when engine_type is set
528 * to NAND_ECC_ENGINE_TYPE_SOFT. Force ecc.algo to
529 * NAND_ECC_ALGO_HAMMING to avoid adding an extra ->ecc_algo
530 * field to davinci_nand_pdata.
531 */
532 chip->ecc.algo = NAND_ECC_ALGO_HAMMING;
533 break;
534 case NAND_ECC_ENGINE_TYPE_ON_HOST:
535 if (pdata->ecc_bits == 4) {
536 int chunks = mtd->writesize / 512;
537
538 if (!chunks || mtd->oobsize < 16) {
539 dev_dbg(&info->pdev->dev, "too small\n");
540 return -EINVAL;
541 }
542
543 /*
544 * No sanity checks: CPUs must support this,
545 * and the chips may not use NAND_BUSWIDTH_16.
546 */
547
548 /* No sharing 4-bit hardware between chipselects yet */
549 spin_lock_irq(&davinci_nand_lock);
550 if (ecc4_busy)
551 ret = -EBUSY;
552 else
553 ecc4_busy = true;
554 spin_unlock_irq(&davinci_nand_lock);
555
556 if (ret == -EBUSY)
557 return ret;
558
559 chip->ecc.calculate = nand_davinci_calculate_4bit;
560 chip->ecc.correct = nand_davinci_correct_4bit;
561 chip->ecc.hwctl = nand_davinci_hwctl_4bit;
562 chip->ecc.bytes = 10;
563 chip->ecc.options = NAND_ECC_GENERIC_ERASED_CHECK;
564 chip->ecc.algo = NAND_ECC_ALGO_BCH;
565
566 /*
567 * Update ECC layout if needed ... for 1-bit HW ECC, the
568 * default is OK, but it allocates 6 bytes when only 3
569 * are needed (for each 512 bytes). For 4-bit HW ECC,
570 * the default is not usable: 10 bytes needed, not 6.
571 *
572 * For small page chips, preserve the manufacturer's
573 * badblock marking data ... and make sure a flash BBT
574 * table marker fits in the free bytes.
575 */
576 if (chunks == 1) {
577 mtd_set_ooblayout(mtd,
578 &hwecc4_small_ooblayout_ops);
579 } else if (chunks == 4 || chunks == 8) {
580 mtd_set_ooblayout(mtd,
581 nand_get_large_page_ooblayout());
582 chip->ecc.read_page = nand_read_page_hwecc_oob_first;
583 } else {
584 return -EIO;
585 }
586 } else {
587 /* 1bit ecc hamming */
588 chip->ecc.calculate = nand_davinci_calculate_1bit;
589 chip->ecc.correct = nand_davinci_correct_1bit;
590 chip->ecc.hwctl = nand_davinci_hwctl_1bit;
591 chip->ecc.bytes = 3;
592 chip->ecc.algo = NAND_ECC_ALGO_HAMMING;
593 }
594 chip->ecc.size = 512;
595 chip->ecc.strength = pdata->ecc_bits;
596 break;
597 default:
598 return -EINVAL;
599 }
600
601 return ret;
602}
603
604static void nand_davinci_data_in(struct davinci_nand_info *info, void *buf,
605 unsigned int len, bool force_8bit)
606{
607 u32 alignment = ((uintptr_t)buf | len) & 3;
608
609 if (force_8bit || (alignment & 1))
610 ioread8_rep(info->current_cs, buf, len);
611 else if (alignment & 3)
612 ioread16_rep(info->current_cs, buf, len >> 1);
613 else
614 ioread32_rep(info->current_cs, buf, len >> 2);
615}
616
617static void nand_davinci_data_out(struct davinci_nand_info *info,
618 const void *buf, unsigned int len,
619 bool force_8bit)
620{
621 u32 alignment = ((uintptr_t)buf | len) & 3;
622
623 if (force_8bit || (alignment & 1))
624 iowrite8_rep(info->current_cs, buf, len);
625 else if (alignment & 3)
626 iowrite16_rep(info->current_cs, buf, len >> 1);
627 else
628 iowrite32_rep(info->current_cs, buf, len >> 2);
629}
630
631static int davinci_nand_exec_instr(struct davinci_nand_info *info,
632 const struct nand_op_instr *instr)
633{
634 unsigned int i, timeout_us;
635 u32 status;
636 int ret;
637
638 switch (instr->type) {
639 case NAND_OP_CMD_INSTR:
640 iowrite8(instr->ctx.cmd.opcode,
641 info->current_cs + info->mask_cle);
642 break;
643
644 case NAND_OP_ADDR_INSTR:
645 for (i = 0; i < instr->ctx.addr.naddrs; i++) {
646 iowrite8(instr->ctx.addr.addrs[i],
647 info->current_cs + info->mask_ale);
648 }
649 break;
650
651 case NAND_OP_DATA_IN_INSTR:
652 nand_davinci_data_in(info, instr->ctx.data.buf.in,
653 instr->ctx.data.len,
654 instr->ctx.data.force_8bit);
655 break;
656
657 case NAND_OP_DATA_OUT_INSTR:
658 nand_davinci_data_out(info, instr->ctx.data.buf.out,
659 instr->ctx.data.len,
660 instr->ctx.data.force_8bit);
661 break;
662
663 case NAND_OP_WAITRDY_INSTR:
664 timeout_us = instr->ctx.waitrdy.timeout_ms * 1000;
665 ret = readl_relaxed_poll_timeout(info->base + NANDFSR_OFFSET,
666 status, status & BIT(0), 100,
667 timeout_us);
668 if (ret)
669 return ret;
670
671 break;
672 }
673
674 if (instr->delay_ns)
675 ndelay(instr->delay_ns);
676
677 return 0;
678}
679
680static int davinci_nand_exec_op(struct nand_chip *chip,
681 const struct nand_operation *op,
682 bool check_only)
683{
684 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip));
685 unsigned int i;
686
687 if (check_only)
688 return 0;
689
690 info->current_cs = info->vaddr + (op->cs * info->mask_chipsel);
691
692 for (i = 0; i < op->ninstrs; i++) {
693 int ret;
694
695 ret = davinci_nand_exec_instr(info, &op->instrs[i]);
696 if (ret)
697 return ret;
698 }
699
700 return 0;
701}
702
703static const struct nand_controller_ops davinci_nand_controller_ops = {
704 .attach_chip = davinci_nand_attach_chip,
705 .exec_op = davinci_nand_exec_op,
706};
707
708static int nand_davinci_probe(struct platform_device *pdev)
709{
710 struct davinci_nand_pdata *pdata;
711 struct davinci_nand_info *info;
712 struct resource *res1;
713 struct resource *res2;
714 void __iomem *vaddr;
715 void __iomem *base;
716 int ret;
717 uint32_t val;
718 struct mtd_info *mtd;
719
720 pdata = nand_davinci_get_pdata(pdev);
721 if (IS_ERR(pdata))
722 return PTR_ERR(pdata);
723
724 /* insist on board-specific configuration */
725 if (!pdata)
726 return -ENODEV;
727
728 /* which external chipselect will we be managing? */
729 if (pdata->core_chipsel > 3)
730 return -ENODEV;
731
732 info = devm_kzalloc(&pdev->dev, sizeof(*info), GFP_KERNEL);
733 if (!info)
734 return -ENOMEM;
735
736 platform_set_drvdata(pdev, info);
737
738 res1 = platform_get_resource(pdev, IORESOURCE_MEM, 0);
739 res2 = platform_get_resource(pdev, IORESOURCE_MEM, 1);
740 if (!res1 || !res2) {
741 dev_err(&pdev->dev, "resource missing\n");
742 return -EINVAL;
743 }
744
745 vaddr = devm_ioremap_resource(&pdev->dev, res1);
746 if (IS_ERR(vaddr))
747 return PTR_ERR(vaddr);
748
749 /*
750 * This registers range is used to setup NAND settings. In case with
751 * TI AEMIF driver, the same memory address range is requested already
752 * by AEMIF, so we cannot request it twice, just ioremap.
753 * The AEMIF and NAND drivers not use the same registers in this range.
754 */
755 base = devm_ioremap(&pdev->dev, res2->start, resource_size(res2));
756 if (!base) {
757 dev_err(&pdev->dev, "ioremap failed for resource %pR\n", res2);
758 return -EADDRNOTAVAIL;
759 }
760
761 info->pdev = pdev;
762 info->base = base;
763 info->vaddr = vaddr;
764
765 mtd = nand_to_mtd(&info->chip);
766 mtd->dev.parent = &pdev->dev;
767 nand_set_flash_node(&info->chip, pdev->dev.of_node);
768
769 /* options such as NAND_BBT_USE_FLASH */
770 info->chip.bbt_options = pdata->bbt_options;
771 /* options such as 16-bit widths */
772 info->chip.options = pdata->options;
773 info->chip.bbt_td = pdata->bbt_td;
774 info->chip.bbt_md = pdata->bbt_md;
775 info->timing = pdata->timing;
776
777 info->current_cs = info->vaddr;
778 info->core_chipsel = pdata->core_chipsel;
779 info->mask_chipsel = pdata->mask_chipsel;
780
781 /* use nandboot-capable ALE/CLE masks by default */
782 info->mask_ale = pdata->mask_ale ? : MASK_ALE;
783 info->mask_cle = pdata->mask_cle ? : MASK_CLE;
784
785 spin_lock_irq(&davinci_nand_lock);
786
787 /* put CSxNAND into NAND mode */
788 val = davinci_nand_readl(info, NANDFCR_OFFSET);
789 val |= BIT(info->core_chipsel);
790 davinci_nand_writel(info, NANDFCR_OFFSET, val);
791
792 spin_unlock_irq(&davinci_nand_lock);
793
794 /* Scan to find existence of the device(s) */
795 nand_controller_init(&info->controller);
796 info->controller.ops = &davinci_nand_controller_ops;
797 info->chip.controller = &info->controller;
798 ret = nand_scan(&info->chip, pdata->mask_chipsel ? 2 : 1);
799 if (ret < 0) {
800 dev_dbg(&pdev->dev, "no NAND chip(s) found\n");
801 return ret;
802 }
803
804 if (pdata->parts)
805 ret = mtd_device_register(mtd, pdata->parts, pdata->nr_parts);
806 else
807 ret = mtd_device_register(mtd, NULL, 0);
808 if (ret < 0)
809 goto err_cleanup_nand;
810
811 val = davinci_nand_readl(info, NRCSR_OFFSET);
812 dev_info(&pdev->dev, "controller rev. %d.%d\n",
813 (val >> 8) & 0xff, val & 0xff);
814
815 return 0;
816
817err_cleanup_nand:
818 nand_cleanup(&info->chip);
819
820 return ret;
821}
822
823static void nand_davinci_remove(struct platform_device *pdev)
824{
825 struct davinci_nand_info *info = platform_get_drvdata(pdev);
826 struct nand_chip *chip = &info->chip;
827 int ret;
828
829 spin_lock_irq(&davinci_nand_lock);
830 if (chip->ecc.placement == NAND_ECC_PLACEMENT_INTERLEAVED)
831 ecc4_busy = false;
832 spin_unlock_irq(&davinci_nand_lock);
833
834 ret = mtd_device_unregister(nand_to_mtd(chip));
835 WARN_ON(ret);
836 nand_cleanup(chip);
837}
838
839static struct platform_driver nand_davinci_driver = {
840 .probe = nand_davinci_probe,
841 .remove_new = nand_davinci_remove,
842 .driver = {
843 .name = "davinci_nand",
844 .of_match_table = of_match_ptr(davinci_nand_of_match),
845 },
846};
847MODULE_ALIAS("platform:davinci_nand");
848
849module_platform_driver(nand_davinci_driver);
850
851MODULE_LICENSE("GPL");
852MODULE_AUTHOR("Texas Instruments");
853MODULE_DESCRIPTION("Davinci NAND flash driver");
854
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/err.h>
14#include <linux/iopoll.h>
15#include <linux/kernel.h>
16#include <linux/module.h>
17#include <linux/mtd/partitions.h>
18#include <linux/mtd/rawnand.h>
19#include <linux/platform_device.h>
20#include <linux/property.h>
21#include <linux/slab.h>
22
23#define NRCSR_OFFSET 0x00
24#define NANDFCR_OFFSET 0x60
25#define NANDFSR_OFFSET 0x64
26#define NANDF1ECC_OFFSET 0x70
27
28/* 4-bit ECC syndrome registers */
29#define NAND_4BIT_ECC_LOAD_OFFSET 0xbc
30#define NAND_4BIT_ECC1_OFFSET 0xc0
31#define NAND_4BIT_ECC2_OFFSET 0xc4
32#define NAND_4BIT_ECC3_OFFSET 0xc8
33#define NAND_4BIT_ECC4_OFFSET 0xcc
34#define NAND_ERR_ADD1_OFFSET 0xd0
35#define NAND_ERR_ADD2_OFFSET 0xd4
36#define NAND_ERR_ERRVAL1_OFFSET 0xd8
37#define NAND_ERR_ERRVAL2_OFFSET 0xdc
38
39/* NOTE: boards don't need to use these address bits
40 * for ALE/CLE unless they support booting from NAND.
41 * They're used unless platform data overrides them.
42 */
43#define MASK_ALE 0x08
44#define MASK_CLE 0x10
45
46struct davinci_nand_pdata {
47 uint32_t mask_ale;
48 uint32_t mask_cle;
49
50 /*
51 * 0-indexed chip-select number of the asynchronous
52 * interface to which the NAND device has been connected.
53 *
54 * So, if you have NAND connected to CS3 of DA850, you
55 * will pass '1' here. Since the asynchronous interface
56 * on DA850 starts from CS2.
57 */
58 uint32_t core_chipsel;
59
60 /* for packages using two chipselects */
61 uint32_t mask_chipsel;
62
63 /* board's default static partition info */
64 struct mtd_partition *parts;
65 unsigned int nr_parts;
66
67 /* none == NAND_ECC_ENGINE_TYPE_NONE (strongly *not* advised!!)
68 * soft == NAND_ECC_ENGINE_TYPE_SOFT
69 * else == NAND_ECC_ENGINE_TYPE_ON_HOST, according to ecc_bits
70 *
71 * All DaVinci-family chips support 1-bit hardware ECC.
72 * Newer ones also support 4-bit ECC, but are awkward
73 * using it with large page chips.
74 */
75 enum nand_ecc_engine_type engine_type;
76 enum nand_ecc_placement ecc_placement;
77 u8 ecc_bits;
78
79 /* e.g. NAND_BUSWIDTH_16 */
80 unsigned int options;
81 /* e.g. NAND_BBT_USE_FLASH */
82 unsigned int bbt_options;
83
84 /* Main and mirror bbt descriptor overrides */
85 struct nand_bbt_descr *bbt_td;
86 struct nand_bbt_descr *bbt_md;
87};
88
89/*
90 * This is a device driver for the NAND flash controller found on the
91 * various DaVinci family chips. It handles up to four SoC chipselects,
92 * and some flavors of secondary chipselect (e.g. based on A12) as used
93 * with multichip packages.
94 *
95 * The 1-bit ECC hardware is supported, as well as the newer 4-bit ECC
96 * available on chips like the DM355 and OMAP-L137 and needed with the
97 * more error-prone MLC NAND chips.
98 *
99 * This driver assumes EM_WAIT connects all the NAND devices' RDY/nBUSY
100 * outputs in a "wire-AND" configuration, with no per-chip signals.
101 */
102struct davinci_nand_info {
103 struct nand_controller controller;
104 struct nand_chip chip;
105
106 struct platform_device *pdev;
107
108 bool is_readmode;
109
110 void __iomem *base;
111 void __iomem *vaddr;
112
113 void __iomem *current_cs;
114
115 uint32_t mask_chipsel;
116 uint32_t mask_ale;
117 uint32_t mask_cle;
118
119 uint32_t core_chipsel;
120};
121
122static DEFINE_SPINLOCK(davinci_nand_lock);
123static bool ecc4_busy;
124
125static inline struct davinci_nand_info *to_davinci_nand(struct mtd_info *mtd)
126{
127 return container_of(mtd_to_nand(mtd), struct davinci_nand_info, chip);
128}
129
130static inline unsigned int davinci_nand_readl(struct davinci_nand_info *info,
131 int offset)
132{
133 return __raw_readl(info->base + offset);
134}
135
136static inline void davinci_nand_writel(struct davinci_nand_info *info,
137 int offset, unsigned long value)
138{
139 __raw_writel(value, info->base + offset);
140}
141
142/*----------------------------------------------------------------------*/
143
144/*
145 * 1-bit hardware ECC ... context maintained for each core chipselect
146 */
147
148static inline uint32_t nand_davinci_readecc_1bit(struct mtd_info *mtd)
149{
150 struct davinci_nand_info *info = to_davinci_nand(mtd);
151
152 return davinci_nand_readl(info, NANDF1ECC_OFFSET
153 + 4 * info->core_chipsel);
154}
155
156static void nand_davinci_hwctl_1bit(struct nand_chip *chip, int mode)
157{
158 struct davinci_nand_info *info;
159 uint32_t nandcfr;
160 unsigned long flags;
161
162 info = to_davinci_nand(nand_to_mtd(chip));
163
164 /* Reset ECC hardware */
165 nand_davinci_readecc_1bit(nand_to_mtd(chip));
166
167 spin_lock_irqsave(&davinci_nand_lock, flags);
168
169 /* Restart ECC hardware */
170 nandcfr = davinci_nand_readl(info, NANDFCR_OFFSET);
171 nandcfr |= BIT(8 + info->core_chipsel);
172 davinci_nand_writel(info, NANDFCR_OFFSET, nandcfr);
173
174 spin_unlock_irqrestore(&davinci_nand_lock, flags);
175}
176
177/*
178 * Read hardware ECC value and pack into three bytes
179 */
180static int nand_davinci_calculate_1bit(struct nand_chip *chip,
181 const u_char *dat, u_char *ecc_code)
182{
183 unsigned int ecc_val = nand_davinci_readecc_1bit(nand_to_mtd(chip));
184 unsigned int ecc24 = (ecc_val & 0x0fff) | ((ecc_val & 0x0fff0000) >> 4);
185
186 /* invert so that erased block ecc is correct */
187 ecc24 = ~ecc24;
188 ecc_code[0] = (u_char)(ecc24);
189 ecc_code[1] = (u_char)(ecc24 >> 8);
190 ecc_code[2] = (u_char)(ecc24 >> 16);
191
192 return 0;
193}
194
195static int nand_davinci_correct_1bit(struct nand_chip *chip, u_char *dat,
196 u_char *read_ecc, u_char *calc_ecc)
197{
198 uint32_t eccNand = read_ecc[0] | (read_ecc[1] << 8) |
199 (read_ecc[2] << 16);
200 uint32_t eccCalc = calc_ecc[0] | (calc_ecc[1] << 8) |
201 (calc_ecc[2] << 16);
202 uint32_t diff = eccCalc ^ eccNand;
203
204 if (diff) {
205 if ((((diff >> 12) ^ diff) & 0xfff) == 0xfff) {
206 /* Correctable error */
207 if ((diff >> (12 + 3)) < chip->ecc.size) {
208 dat[diff >> (12 + 3)] ^= BIT((diff >> 12) & 7);
209 return 1;
210 } else {
211 return -EBADMSG;
212 }
213 } else if (!(diff & (diff - 1))) {
214 /* Single bit ECC error in the ECC itself,
215 * nothing to fix */
216 return 1;
217 } else {
218 /* Uncorrectable error */
219 return -EBADMSG;
220 }
221
222 }
223 return 0;
224}
225
226/*----------------------------------------------------------------------*/
227
228/*
229 * 4-bit hardware ECC ... context maintained over entire AEMIF
230 *
231 * This is a syndrome engine, but we avoid NAND_ECC_PLACEMENT_INTERLEAVED
232 * since that forces use of a problematic "infix OOB" layout.
233 * Among other things, it trashes manufacturer bad block markers.
234 * Also, and specific to this hardware, it ECC-protects the "prepad"
235 * in the OOB ... while having ECC protection for parts of OOB would
236 * seem useful, the current MTD stack sometimes wants to update the
237 * OOB without recomputing ECC.
238 */
239
240static void nand_davinci_hwctl_4bit(struct nand_chip *chip, int mode)
241{
242 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip));
243 unsigned long flags;
244 u32 val;
245
246 /* Reset ECC hardware */
247 davinci_nand_readl(info, NAND_4BIT_ECC1_OFFSET);
248
249 spin_lock_irqsave(&davinci_nand_lock, flags);
250
251 /* Start 4-bit ECC calculation for read/write */
252 val = davinci_nand_readl(info, NANDFCR_OFFSET);
253 val &= ~(0x03 << 4);
254 val |= (info->core_chipsel << 4) | BIT(12);
255 davinci_nand_writel(info, NANDFCR_OFFSET, val);
256
257 info->is_readmode = (mode == NAND_ECC_READ);
258
259 spin_unlock_irqrestore(&davinci_nand_lock, flags);
260}
261
262/* Read raw ECC code after writing to NAND. */
263static void
264nand_davinci_readecc_4bit(struct davinci_nand_info *info, u32 code[4])
265{
266 const u32 mask = 0x03ff03ff;
267
268 code[0] = davinci_nand_readl(info, NAND_4BIT_ECC1_OFFSET) & mask;
269 code[1] = davinci_nand_readl(info, NAND_4BIT_ECC2_OFFSET) & mask;
270 code[2] = davinci_nand_readl(info, NAND_4BIT_ECC3_OFFSET) & mask;
271 code[3] = davinci_nand_readl(info, NAND_4BIT_ECC4_OFFSET) & mask;
272}
273
274/* Terminate read ECC; or return ECC (as bytes) of data written to NAND. */
275static int nand_davinci_calculate_4bit(struct nand_chip *chip,
276 const u_char *dat, u_char *ecc_code)
277{
278 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip));
279 u32 raw_ecc[4], *p;
280 unsigned i;
281
282 /* After a read, terminate ECC calculation by a dummy read
283 * of some 4-bit ECC register. ECC covers everything that
284 * was read; correct() just uses the hardware state, so
285 * ecc_code is not needed.
286 */
287 if (info->is_readmode) {
288 davinci_nand_readl(info, NAND_4BIT_ECC1_OFFSET);
289 return 0;
290 }
291
292 /* Pack eight raw 10-bit ecc values into ten bytes, making
293 * two passes which each convert four values (in upper and
294 * lower halves of two 32-bit words) into five bytes. The
295 * ROM boot loader uses this same packing scheme.
296 */
297 nand_davinci_readecc_4bit(info, raw_ecc);
298 for (i = 0, p = raw_ecc; i < 2; i++, p += 2) {
299 *ecc_code++ = p[0] & 0xff;
300 *ecc_code++ = ((p[0] >> 8) & 0x03) | ((p[0] >> 14) & 0xfc);
301 *ecc_code++ = ((p[0] >> 22) & 0x0f) | ((p[1] << 4) & 0xf0);
302 *ecc_code++ = ((p[1] >> 4) & 0x3f) | ((p[1] >> 10) & 0xc0);
303 *ecc_code++ = (p[1] >> 18) & 0xff;
304 }
305
306 return 0;
307}
308
309/* Correct up to 4 bits in data we just read, using state left in the
310 * hardware plus the ecc_code computed when it was first written.
311 */
312static int nand_davinci_correct_4bit(struct nand_chip *chip, u_char *data,
313 u_char *ecc_code, u_char *null)
314{
315 int i;
316 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip));
317 unsigned short ecc10[8];
318 unsigned short *ecc16;
319 u32 syndrome[4];
320 u32 ecc_state;
321 unsigned num_errors, corrected;
322 unsigned long timeo;
323
324 /* Unpack ten bytes into eight 10 bit values. We know we're
325 * little-endian, and use type punning for less shifting/masking.
326 */
327 if (WARN_ON(0x01 & (uintptr_t)ecc_code))
328 return -EINVAL;
329 ecc16 = (unsigned short *)ecc_code;
330
331 ecc10[0] = (ecc16[0] >> 0) & 0x3ff;
332 ecc10[1] = ((ecc16[0] >> 10) & 0x3f) | ((ecc16[1] << 6) & 0x3c0);
333 ecc10[2] = (ecc16[1] >> 4) & 0x3ff;
334 ecc10[3] = ((ecc16[1] >> 14) & 0x3) | ((ecc16[2] << 2) & 0x3fc);
335 ecc10[4] = (ecc16[2] >> 8) | ((ecc16[3] << 8) & 0x300);
336 ecc10[5] = (ecc16[3] >> 2) & 0x3ff;
337 ecc10[6] = ((ecc16[3] >> 12) & 0xf) | ((ecc16[4] << 4) & 0x3f0);
338 ecc10[7] = (ecc16[4] >> 6) & 0x3ff;
339
340 /* Tell ECC controller about the expected ECC codes. */
341 for (i = 7; i >= 0; i--)
342 davinci_nand_writel(info, NAND_4BIT_ECC_LOAD_OFFSET, ecc10[i]);
343
344 /* Allow time for syndrome calculation ... then read it.
345 * A syndrome of all zeroes 0 means no detected errors.
346 */
347 davinci_nand_readl(info, NANDFSR_OFFSET);
348 nand_davinci_readecc_4bit(info, syndrome);
349 if (!(syndrome[0] | syndrome[1] | syndrome[2] | syndrome[3]))
350 return 0;
351
352 /*
353 * Clear any previous address calculation by doing a dummy read of an
354 * error address register.
355 */
356 davinci_nand_readl(info, NAND_ERR_ADD1_OFFSET);
357
358 /* Start address calculation, and wait for it to complete.
359 * We _could_ start reading more data while this is working,
360 * to speed up the overall page read.
361 */
362 davinci_nand_writel(info, NANDFCR_OFFSET,
363 davinci_nand_readl(info, NANDFCR_OFFSET) | BIT(13));
364
365 /*
366 * ECC_STATE field reads 0x3 (Error correction complete) immediately
367 * after setting the 4BITECC_ADD_CALC_START bit. So if you immediately
368 * begin trying to poll for the state, you may fall right out of your
369 * loop without any of the correction calculations having taken place.
370 * The recommendation from the hardware team is to initially delay as
371 * long as ECC_STATE reads less than 4. After that, ECC HW has entered
372 * correction state.
373 */
374 timeo = jiffies + usecs_to_jiffies(100);
375 do {
376 ecc_state = (davinci_nand_readl(info,
377 NANDFSR_OFFSET) >> 8) & 0x0f;
378 cpu_relax();
379 } while ((ecc_state < 4) && time_before(jiffies, timeo));
380
381 for (;;) {
382 u32 fsr = davinci_nand_readl(info, NANDFSR_OFFSET);
383
384 switch ((fsr >> 8) & 0x0f) {
385 case 0: /* no error, should not happen */
386 davinci_nand_readl(info, NAND_ERR_ERRVAL1_OFFSET);
387 return 0;
388 case 1: /* five or more errors detected */
389 davinci_nand_readl(info, NAND_ERR_ERRVAL1_OFFSET);
390 return -EBADMSG;
391 case 2: /* error addresses computed */
392 case 3:
393 num_errors = 1 + ((fsr >> 16) & 0x03);
394 goto correct;
395 default: /* still working on it */
396 cpu_relax();
397 continue;
398 }
399 }
400
401correct:
402 /* correct each error */
403 for (i = 0, corrected = 0; i < num_errors; i++) {
404 int error_address, error_value;
405
406 if (i > 1) {
407 error_address = davinci_nand_readl(info,
408 NAND_ERR_ADD2_OFFSET);
409 error_value = davinci_nand_readl(info,
410 NAND_ERR_ERRVAL2_OFFSET);
411 } else {
412 error_address = davinci_nand_readl(info,
413 NAND_ERR_ADD1_OFFSET);
414 error_value = davinci_nand_readl(info,
415 NAND_ERR_ERRVAL1_OFFSET);
416 }
417
418 if (i & 1) {
419 error_address >>= 16;
420 error_value >>= 16;
421 }
422 error_address &= 0x3ff;
423 error_address = (512 + 7) - error_address;
424
425 if (error_address < 512) {
426 data[error_address] ^= error_value;
427 corrected++;
428 }
429 }
430
431 return corrected;
432}
433
434/*----------------------------------------------------------------------*/
435
436/* An ECC layout for using 4-bit ECC with small-page flash, storing
437 * ten ECC bytes plus the manufacturer's bad block marker byte, and
438 * and not overlapping the default BBT markers.
439 */
440static int hwecc4_ooblayout_small_ecc(struct mtd_info *mtd, int section,
441 struct mtd_oob_region *oobregion)
442{
443 if (section > 2)
444 return -ERANGE;
445
446 if (!section) {
447 oobregion->offset = 0;
448 oobregion->length = 5;
449 } else if (section == 1) {
450 oobregion->offset = 6;
451 oobregion->length = 2;
452 } else {
453 oobregion->offset = 13;
454 oobregion->length = 3;
455 }
456
457 return 0;
458}
459
460static int hwecc4_ooblayout_small_free(struct mtd_info *mtd, int section,
461 struct mtd_oob_region *oobregion)
462{
463 if (section > 1)
464 return -ERANGE;
465
466 if (!section) {
467 oobregion->offset = 8;
468 oobregion->length = 5;
469 } else {
470 oobregion->offset = 16;
471 oobregion->length = mtd->oobsize - 16;
472 }
473
474 return 0;
475}
476
477static const struct mtd_ooblayout_ops hwecc4_small_ooblayout_ops = {
478 .ecc = hwecc4_ooblayout_small_ecc,
479 .free = hwecc4_ooblayout_small_free,
480};
481
482#if defined(CONFIG_OF)
483static const struct of_device_id davinci_nand_of_match[] = {
484 {.compatible = "ti,davinci-nand", },
485 {.compatible = "ti,keystone-nand", },
486 {},
487};
488MODULE_DEVICE_TABLE(of, davinci_nand_of_match);
489
490static struct davinci_nand_pdata *
491nand_davinci_get_pdata(struct platform_device *pdev)
492{
493 if (!dev_get_platdata(&pdev->dev)) {
494 struct davinci_nand_pdata *pdata;
495 const char *mode;
496 u32 prop;
497
498 pdata = devm_kzalloc(&pdev->dev,
499 sizeof(struct davinci_nand_pdata),
500 GFP_KERNEL);
501 pdev->dev.platform_data = pdata;
502 if (!pdata)
503 return ERR_PTR(-ENOMEM);
504 if (!device_property_read_u32(&pdev->dev,
505 "ti,davinci-chipselect", &prop))
506 pdata->core_chipsel = prop;
507 else
508 return ERR_PTR(-EINVAL);
509
510 if (!device_property_read_u32(&pdev->dev,
511 "ti,davinci-mask-ale", &prop))
512 pdata->mask_ale = prop;
513 if (!device_property_read_u32(&pdev->dev,
514 "ti,davinci-mask-cle", &prop))
515 pdata->mask_cle = prop;
516 if (!device_property_read_u32(&pdev->dev,
517 "ti,davinci-mask-chipsel", &prop))
518 pdata->mask_chipsel = prop;
519 if (!device_property_read_string(&pdev->dev,
520 "ti,davinci-ecc-mode",
521 &mode)) {
522 if (!strncmp("none", mode, 4))
523 pdata->engine_type = NAND_ECC_ENGINE_TYPE_NONE;
524 if (!strncmp("soft", mode, 4))
525 pdata->engine_type = NAND_ECC_ENGINE_TYPE_SOFT;
526 if (!strncmp("hw", mode, 2))
527 pdata->engine_type = NAND_ECC_ENGINE_TYPE_ON_HOST;
528 }
529 if (!device_property_read_u32(&pdev->dev,
530 "ti,davinci-ecc-bits", &prop))
531 pdata->ecc_bits = prop;
532
533 if (!device_property_read_u32(&pdev->dev,
534 "ti,davinci-nand-buswidth",
535 &prop) && prop == 16)
536 pdata->options |= NAND_BUSWIDTH_16;
537
538 if (device_property_read_bool(&pdev->dev,
539 "ti,davinci-nand-use-bbt"))
540 pdata->bbt_options = NAND_BBT_USE_FLASH;
541
542 /*
543 * Since kernel v4.8, this driver has been fixed to enable
544 * use of 4-bit hardware ECC with subpages and verified on
545 * TI's keystone EVMs (K2L, K2HK and K2E).
546 * However, in the interest of not breaking systems using
547 * existing UBI partitions, sub-page writes are not being
548 * (re)enabled. If you want to use subpage writes on Keystone
549 * platforms (i.e. do not have any existing UBI partitions),
550 * then use "ti,davinci-nand" as the compatible in your
551 * device-tree file.
552 */
553 if (device_is_compatible(&pdev->dev, "ti,keystone-nand"))
554 pdata->options |= NAND_NO_SUBPAGE_WRITE;
555 }
556
557 return dev_get_platdata(&pdev->dev);
558}
559#else
560static struct davinci_nand_pdata *
561nand_davinci_get_pdata(struct platform_device *pdev)
562{
563 return dev_get_platdata(&pdev->dev);
564}
565#endif
566
567static int davinci_nand_attach_chip(struct nand_chip *chip)
568{
569 struct mtd_info *mtd = nand_to_mtd(chip);
570 struct davinci_nand_info *info = to_davinci_nand(mtd);
571 struct davinci_nand_pdata *pdata = nand_davinci_get_pdata(info->pdev);
572 int ret = 0;
573
574 if (IS_ERR(pdata))
575 return PTR_ERR(pdata);
576
577 /* Use board-specific ECC config */
578 chip->ecc.engine_type = pdata->engine_type;
579 chip->ecc.placement = pdata->ecc_placement;
580
581 switch (chip->ecc.engine_type) {
582 case NAND_ECC_ENGINE_TYPE_NONE:
583 pdata->ecc_bits = 0;
584 break;
585 case NAND_ECC_ENGINE_TYPE_SOFT:
586 pdata->ecc_bits = 0;
587 /*
588 * This driver expects Hamming based ECC when engine_type is set
589 * to NAND_ECC_ENGINE_TYPE_SOFT. Force ecc.algo to
590 * NAND_ECC_ALGO_HAMMING to avoid adding an extra ->ecc_algo
591 * field to davinci_nand_pdata.
592 */
593 chip->ecc.algo = NAND_ECC_ALGO_HAMMING;
594 break;
595 case NAND_ECC_ENGINE_TYPE_ON_HOST:
596 if (pdata->ecc_bits == 4) {
597 int chunks = mtd->writesize / 512;
598
599 if (!chunks || mtd->oobsize < 16) {
600 dev_dbg(&info->pdev->dev, "too small\n");
601 return -EINVAL;
602 }
603
604 /*
605 * No sanity checks: CPUs must support this,
606 * and the chips may not use NAND_BUSWIDTH_16.
607 */
608
609 /* No sharing 4-bit hardware between chipselects yet */
610 spin_lock_irq(&davinci_nand_lock);
611 if (ecc4_busy)
612 ret = -EBUSY;
613 else
614 ecc4_busy = true;
615 spin_unlock_irq(&davinci_nand_lock);
616
617 if (ret == -EBUSY)
618 return ret;
619
620 chip->ecc.calculate = nand_davinci_calculate_4bit;
621 chip->ecc.correct = nand_davinci_correct_4bit;
622 chip->ecc.hwctl = nand_davinci_hwctl_4bit;
623 chip->ecc.bytes = 10;
624 chip->ecc.options = NAND_ECC_GENERIC_ERASED_CHECK;
625 chip->ecc.algo = NAND_ECC_ALGO_BCH;
626
627 /*
628 * Update ECC layout if needed ... for 1-bit HW ECC, the
629 * default is OK, but it allocates 6 bytes when only 3
630 * are needed (for each 512 bytes). For 4-bit HW ECC,
631 * the default is not usable: 10 bytes needed, not 6.
632 *
633 * For small page chips, preserve the manufacturer's
634 * badblock marking data ... and make sure a flash BBT
635 * table marker fits in the free bytes.
636 */
637 if (chunks == 1) {
638 mtd_set_ooblayout(mtd,
639 &hwecc4_small_ooblayout_ops);
640 } else if (chunks == 4 || chunks == 8) {
641 mtd_set_ooblayout(mtd,
642 nand_get_large_page_ooblayout());
643 chip->ecc.read_page = nand_read_page_hwecc_oob_first;
644 } else {
645 return -EIO;
646 }
647 } else {
648 /* 1bit ecc hamming */
649 chip->ecc.calculate = nand_davinci_calculate_1bit;
650 chip->ecc.correct = nand_davinci_correct_1bit;
651 chip->ecc.hwctl = nand_davinci_hwctl_1bit;
652 chip->ecc.bytes = 3;
653 chip->ecc.algo = NAND_ECC_ALGO_HAMMING;
654 }
655 chip->ecc.size = 512;
656 chip->ecc.strength = pdata->ecc_bits;
657 break;
658 default:
659 return -EINVAL;
660 }
661
662 return ret;
663}
664
665static void nand_davinci_data_in(struct davinci_nand_info *info, void *buf,
666 unsigned int len, bool force_8bit)
667{
668 u32 alignment = ((uintptr_t)buf | len) & 3;
669
670 if (force_8bit || (alignment & 1))
671 ioread8_rep(info->current_cs, buf, len);
672 else if (alignment & 3)
673 ioread16_rep(info->current_cs, buf, len >> 1);
674 else
675 ioread32_rep(info->current_cs, buf, len >> 2);
676}
677
678static void nand_davinci_data_out(struct davinci_nand_info *info,
679 const void *buf, unsigned int len,
680 bool force_8bit)
681{
682 u32 alignment = ((uintptr_t)buf | len) & 3;
683
684 if (force_8bit || (alignment & 1))
685 iowrite8_rep(info->current_cs, buf, len);
686 else if (alignment & 3)
687 iowrite16_rep(info->current_cs, buf, len >> 1);
688 else
689 iowrite32_rep(info->current_cs, buf, len >> 2);
690}
691
692static int davinci_nand_exec_instr(struct davinci_nand_info *info,
693 const struct nand_op_instr *instr)
694{
695 unsigned int i, timeout_us;
696 u32 status;
697 int ret;
698
699 switch (instr->type) {
700 case NAND_OP_CMD_INSTR:
701 iowrite8(instr->ctx.cmd.opcode,
702 info->current_cs + info->mask_cle);
703 break;
704
705 case NAND_OP_ADDR_INSTR:
706 for (i = 0; i < instr->ctx.addr.naddrs; i++) {
707 iowrite8(instr->ctx.addr.addrs[i],
708 info->current_cs + info->mask_ale);
709 }
710 break;
711
712 case NAND_OP_DATA_IN_INSTR:
713 nand_davinci_data_in(info, instr->ctx.data.buf.in,
714 instr->ctx.data.len,
715 instr->ctx.data.force_8bit);
716 break;
717
718 case NAND_OP_DATA_OUT_INSTR:
719 nand_davinci_data_out(info, instr->ctx.data.buf.out,
720 instr->ctx.data.len,
721 instr->ctx.data.force_8bit);
722 break;
723
724 case NAND_OP_WAITRDY_INSTR:
725 timeout_us = instr->ctx.waitrdy.timeout_ms * 1000;
726 ret = readl_relaxed_poll_timeout(info->base + NANDFSR_OFFSET,
727 status, status & BIT(0), 100,
728 timeout_us);
729 if (ret)
730 return ret;
731
732 break;
733 }
734
735 if (instr->delay_ns) {
736 /* Dummy read to be sure that command is sent before ndelay starts */
737 davinci_nand_readl(info, 0);
738 ndelay(instr->delay_ns);
739 }
740
741 return 0;
742}
743
744static int davinci_nand_exec_op(struct nand_chip *chip,
745 const struct nand_operation *op,
746 bool check_only)
747{
748 struct davinci_nand_info *info = to_davinci_nand(nand_to_mtd(chip));
749 unsigned int i;
750
751 if (check_only)
752 return 0;
753
754 info->current_cs = info->vaddr + (op->cs * info->mask_chipsel);
755
756 for (i = 0; i < op->ninstrs; i++) {
757 int ret;
758
759 ret = davinci_nand_exec_instr(info, &op->instrs[i]);
760 if (ret)
761 return ret;
762 }
763
764 return 0;
765}
766
767static const struct nand_controller_ops davinci_nand_controller_ops = {
768 .attach_chip = davinci_nand_attach_chip,
769 .exec_op = davinci_nand_exec_op,
770};
771
772static int nand_davinci_probe(struct platform_device *pdev)
773{
774 struct davinci_nand_pdata *pdata;
775 struct davinci_nand_info *info;
776 struct resource *res1;
777 struct resource *res2;
778 void __iomem *vaddr;
779 void __iomem *base;
780 int ret;
781 uint32_t val;
782 struct mtd_info *mtd;
783
784 pdata = nand_davinci_get_pdata(pdev);
785 if (IS_ERR(pdata))
786 return PTR_ERR(pdata);
787
788 /* insist on board-specific configuration */
789 if (!pdata)
790 return -ENODEV;
791
792 /* which external chipselect will we be managing? */
793 if (pdata->core_chipsel > 3)
794 return -ENODEV;
795
796 info = devm_kzalloc(&pdev->dev, sizeof(*info), GFP_KERNEL);
797 if (!info)
798 return -ENOMEM;
799
800 platform_set_drvdata(pdev, info);
801
802 res1 = platform_get_resource(pdev, IORESOURCE_MEM, 0);
803 res2 = platform_get_resource(pdev, IORESOURCE_MEM, 1);
804 if (!res1 || !res2) {
805 dev_err(&pdev->dev, "resource missing\n");
806 return -EINVAL;
807 }
808
809 vaddr = devm_ioremap_resource(&pdev->dev, res1);
810 if (IS_ERR(vaddr))
811 return PTR_ERR(vaddr);
812
813 /*
814 * This registers range is used to setup NAND settings. In case with
815 * TI AEMIF driver, the same memory address range is requested already
816 * by AEMIF, so we cannot request it twice, just ioremap.
817 * The AEMIF and NAND drivers not use the same registers in this range.
818 */
819 base = devm_ioremap(&pdev->dev, res2->start, resource_size(res2));
820 if (!base) {
821 dev_err(&pdev->dev, "ioremap failed for resource %pR\n", res2);
822 return -EADDRNOTAVAIL;
823 }
824
825 info->pdev = pdev;
826 info->base = base;
827 info->vaddr = vaddr;
828
829 mtd = nand_to_mtd(&info->chip);
830 mtd->dev.parent = &pdev->dev;
831 nand_set_flash_node(&info->chip, pdev->dev.of_node);
832
833 /* options such as NAND_BBT_USE_FLASH */
834 info->chip.bbt_options = pdata->bbt_options;
835 /* options such as 16-bit widths */
836 info->chip.options = pdata->options;
837 info->chip.bbt_td = pdata->bbt_td;
838 info->chip.bbt_md = pdata->bbt_md;
839
840 info->current_cs = info->vaddr;
841 info->core_chipsel = pdata->core_chipsel;
842 info->mask_chipsel = pdata->mask_chipsel;
843
844 /* use nandboot-capable ALE/CLE masks by default */
845 info->mask_ale = pdata->mask_ale ? : MASK_ALE;
846 info->mask_cle = pdata->mask_cle ? : MASK_CLE;
847
848 spin_lock_irq(&davinci_nand_lock);
849
850 /* put CSxNAND into NAND mode */
851 val = davinci_nand_readl(info, NANDFCR_OFFSET);
852 val |= BIT(info->core_chipsel);
853 davinci_nand_writel(info, NANDFCR_OFFSET, val);
854
855 spin_unlock_irq(&davinci_nand_lock);
856
857 /* Scan to find existence of the device(s) */
858 nand_controller_init(&info->controller);
859 info->controller.ops = &davinci_nand_controller_ops;
860 info->chip.controller = &info->controller;
861 ret = nand_scan(&info->chip, pdata->mask_chipsel ? 2 : 1);
862 if (ret < 0) {
863 dev_dbg(&pdev->dev, "no NAND chip(s) found\n");
864 return ret;
865 }
866
867 if (pdata->parts)
868 ret = mtd_device_register(mtd, pdata->parts, pdata->nr_parts);
869 else
870 ret = mtd_device_register(mtd, NULL, 0);
871 if (ret < 0)
872 goto err_cleanup_nand;
873
874 val = davinci_nand_readl(info, NRCSR_OFFSET);
875 dev_info(&pdev->dev, "controller rev. %d.%d\n",
876 (val >> 8) & 0xff, val & 0xff);
877
878 return 0;
879
880err_cleanup_nand:
881 nand_cleanup(&info->chip);
882
883 return ret;
884}
885
886static void nand_davinci_remove(struct platform_device *pdev)
887{
888 struct davinci_nand_info *info = platform_get_drvdata(pdev);
889 struct nand_chip *chip = &info->chip;
890 int ret;
891
892 spin_lock_irq(&davinci_nand_lock);
893 if (chip->ecc.placement == NAND_ECC_PLACEMENT_INTERLEAVED)
894 ecc4_busy = false;
895 spin_unlock_irq(&davinci_nand_lock);
896
897 ret = mtd_device_unregister(nand_to_mtd(chip));
898 WARN_ON(ret);
899 nand_cleanup(chip);
900}
901
902static struct platform_driver nand_davinci_driver = {
903 .probe = nand_davinci_probe,
904 .remove = nand_davinci_remove,
905 .driver = {
906 .name = "davinci_nand",
907 .of_match_table = of_match_ptr(davinci_nand_of_match),
908 },
909};
910MODULE_ALIAS("platform:davinci_nand");
911
912module_platform_driver(nand_davinci_driver);
913
914MODULE_LICENSE("GPL");
915MODULE_AUTHOR("Texas Instruments");
916MODULE_DESCRIPTION("Davinci NAND flash driver");
917