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