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
855