1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * Freescale GPMI NAND Flash Driver
4 *
5 * Copyright (C) 2010-2015 Freescale Semiconductor, Inc.
6 * Copyright (C) 2008 Embedded Alley Solutions, Inc.
7 */
8#include <linux/clk.h>
9#include <linux/delay.h>
10#include <linux/slab.h>
11#include <linux/sched/task_stack.h>
12#include <linux/interrupt.h>
13#include <linux/module.h>
14#include <linux/mtd/partitions.h>
15#include <linux/of.h>
16#include <linux/platform_device.h>
17#include <linux/pm_runtime.h>
18#include <linux/dma/mxs-dma.h>
19#include "gpmi-nand.h"
20#include "gpmi-regs.h"
21#include "bch-regs.h"
22
23/* Resource names for the GPMI NAND driver. */
24#define GPMI_NAND_GPMI_REGS_ADDR_RES_NAME  "gpmi-nand"
25#define GPMI_NAND_BCH_REGS_ADDR_RES_NAME   "bch"
26#define GPMI_NAND_BCH_INTERRUPT_RES_NAME   "bch"
27
28/* Converts time to clock cycles */
29#define TO_CYCLES(duration, period) DIV_ROUND_UP_ULL(duration, period)
30
31#define MXS_SET_ADDR		0x4
32#define MXS_CLR_ADDR		0x8
33/*
34 * Clear the bit and poll it cleared.  This is usually called with
35 * a reset address and mask being either SFTRST(bit 31) or CLKGATE
36 * (bit 30).
37 */
38static int clear_poll_bit(void __iomem *addr, u32 mask)
39{
40	int timeout = 0x400;
41
42	/* clear the bit */
43	writel(mask, addr + MXS_CLR_ADDR);
44
45	/*
46	 * SFTRST needs 3 GPMI clocks to settle, the reference manual
47	 * recommends to wait 1us.
48	 */
49	udelay(1);
50
51	/* poll the bit becoming clear */
52	while ((readl(addr) & mask) && --timeout)
53		/* nothing */;
54
55	return !timeout;
56}
57
58#define MODULE_CLKGATE		(1 << 30)
59#define MODULE_SFTRST		(1 << 31)
60/*
61 * The current mxs_reset_block() will do two things:
62 *  [1] enable the module.
63 *  [2] reset the module.
64 *
65 * In most of the cases, it's ok.
66 * But in MX23, there is a hardware bug in the BCH block (see erratum #2847).
67 * If you try to soft reset the BCH block, it becomes unusable until
68 * the next hard reset. This case occurs in the NAND boot mode. When the board
69 * boots by NAND, the ROM of the chip will initialize the BCH blocks itself.
70 * So If the driver tries to reset the BCH again, the BCH will not work anymore.
71 * You will see a DMA timeout in this case. The bug has been fixed
72 * in the following chips, such as MX28.
73 *
74 * To avoid this bug, just add a new parameter `just_enable` for
75 * the mxs_reset_block(), and rewrite it here.
76 */
77static int gpmi_reset_block(void __iomem *reset_addr, bool just_enable)
78{
79	int ret;
80	int timeout = 0x400;
81
82	/* clear and poll SFTRST */
83	ret = clear_poll_bit(reset_addr, MODULE_SFTRST);
84	if (unlikely(ret))
85		goto error;
86
87	/* clear CLKGATE */
88	writel(MODULE_CLKGATE, reset_addr + MXS_CLR_ADDR);
89
90	if (!just_enable) {
91		/* set SFTRST to reset the block */
92		writel(MODULE_SFTRST, reset_addr + MXS_SET_ADDR);
93		udelay(1);
94
95		/* poll CLKGATE becoming set */
96		while ((!(readl(reset_addr) & MODULE_CLKGATE)) && --timeout)
97			/* nothing */;
98		if (unlikely(!timeout))
99			goto error;
100	}
101
102	/* clear and poll SFTRST */
103	ret = clear_poll_bit(reset_addr, MODULE_SFTRST);
104	if (unlikely(ret))
105		goto error;
106
107	/* clear and poll CLKGATE */
108	ret = clear_poll_bit(reset_addr, MODULE_CLKGATE);
109	if (unlikely(ret))
110		goto error;
111
112	return 0;
113
114error:
115	pr_err("%s(%p): module reset timeout\n", __func__, reset_addr);
116	return -ETIMEDOUT;
117}
118
119static int __gpmi_enable_clk(struct gpmi_nand_data *this, bool v)
120{
121	struct clk *clk;
122	int ret;
123	int i;
124
125	for (i = 0; i < GPMI_CLK_MAX; i++) {
126		clk = this->resources.clock[i];
127		if (!clk)
128			break;
129
130		if (v) {
131			ret = clk_prepare_enable(clk);
132			if (ret)
133				goto err_clk;
134		} else {
135			clk_disable_unprepare(clk);
136		}
137	}
138	return 0;
139
140err_clk:
141	for (; i > 0; i--)
142		clk_disable_unprepare(this->resources.clock[i - 1]);
143	return ret;
144}
145
146static int gpmi_init(struct gpmi_nand_data *this)
147{
148	struct resources *r = &this->resources;
149	int ret;
150
151	ret = pm_runtime_resume_and_get(this->dev);
152	if (ret < 0)
153		return ret;
154
155	ret = gpmi_reset_block(r->gpmi_regs, false);
156	if (ret)
157		goto err_out;
158
159	/*
160	 * Reset BCH here, too. We got failures otherwise :(
161	 * See later BCH reset for explanation of MX23 and MX28 handling
162	 */
163	ret = gpmi_reset_block(r->bch_regs, GPMI_IS_MXS(this));
164	if (ret)
165		goto err_out;
166
167	/* Choose NAND mode. */
168	writel(BM_GPMI_CTRL1_GPMI_MODE, r->gpmi_regs + HW_GPMI_CTRL1_CLR);
169
170	/* Set the IRQ polarity. */
171	writel(BM_GPMI_CTRL1_ATA_IRQRDY_POLARITY,
172				r->gpmi_regs + HW_GPMI_CTRL1_SET);
173
174	/* Disable Write-Protection. */
175	writel(BM_GPMI_CTRL1_DEV_RESET, r->gpmi_regs + HW_GPMI_CTRL1_SET);
176
177	/* Select BCH ECC. */
178	writel(BM_GPMI_CTRL1_BCH_MODE, r->gpmi_regs + HW_GPMI_CTRL1_SET);
179
180	/*
181	 * Decouple the chip select from dma channel. We use dma0 for all
182	 * the chips, force all NAND RDY_BUSY inputs to be sourced from
183	 * RDY_BUSY0.
184	 */
185	writel(BM_GPMI_CTRL1_DECOUPLE_CS | BM_GPMI_CTRL1_GANGED_RDYBUSY,
186	       r->gpmi_regs + HW_GPMI_CTRL1_SET);
187
188err_out:
189	pm_runtime_mark_last_busy(this->dev);
190	pm_runtime_put_autosuspend(this->dev);
191	return ret;
192}
193
194/* This function is very useful. It is called only when the bug occur. */
195static void gpmi_dump_info(struct gpmi_nand_data *this)
196{
197	struct resources *r = &this->resources;
198	struct bch_geometry *geo = &this->bch_geometry;
199	u32 reg;
200	int i;
201
202	dev_err(this->dev, "Show GPMI registers :\n");
203	for (i = 0; i <= HW_GPMI_DEBUG / 0x10 + 1; i++) {
204		reg = readl(r->gpmi_regs + i * 0x10);
205		dev_err(this->dev, "offset 0x%.3x : 0x%.8x\n", i * 0x10, reg);
206	}
207
208	/* start to print out the BCH info */
209	dev_err(this->dev, "Show BCH registers :\n");
210	for (i = 0; i <= HW_BCH_VERSION / 0x10 + 1; i++) {
211		reg = readl(r->bch_regs + i * 0x10);
212		dev_err(this->dev, "offset 0x%.3x : 0x%.8x\n", i * 0x10, reg);
213	}
214	dev_err(this->dev, "BCH Geometry :\n"
215		"GF length              : %u\n"
216		"ECC Strength           : %u\n"
217		"Page Size in Bytes     : %u\n"
218		"Metadata Size in Bytes : %u\n"
219		"ECC0 Chunk Size in Bytes: %u\n"
220		"ECCn Chunk Size in Bytes: %u\n"
221		"ECC Chunk Count        : %u\n"
222		"Payload Size in Bytes  : %u\n"
223		"Auxiliary Size in Bytes: %u\n"
224		"Auxiliary Status Offset: %u\n"
225		"Block Mark Byte Offset : %u\n"
226		"Block Mark Bit Offset  : %u\n",
227		geo->gf_len,
228		geo->ecc_strength,
229		geo->page_size,
230		geo->metadata_size,
231		geo->ecc0_chunk_size,
232		geo->eccn_chunk_size,
233		geo->ecc_chunk_count,
234		geo->payload_size,
235		geo->auxiliary_size,
236		geo->auxiliary_status_offset,
237		geo->block_mark_byte_offset,
238		geo->block_mark_bit_offset);
239}
240
241static bool gpmi_check_ecc(struct gpmi_nand_data *this)
242{
243	struct nand_chip *chip = &this->nand;
244	struct bch_geometry *geo = &this->bch_geometry;
245	struct nand_device *nand = &chip->base;
246	struct nand_ecc_props *conf = &nand->ecc.ctx.conf;
247
248	conf->step_size = geo->eccn_chunk_size;
249	conf->strength = geo->ecc_strength;
250
251	/* Do the sanity check. */
252	if (GPMI_IS_MXS(this)) {
253		/* The mx23/mx28 only support the GF13. */
254		if (geo->gf_len == 14)
255			return false;
256	}
257
258	if (geo->ecc_strength > this->devdata->bch_max_ecc_strength)
259		return false;
260
261	if (!nand_ecc_is_strong_enough(nand))
262		return false;
263
264	return true;
265}
266
267/* check if bbm locates in data chunk rather than ecc chunk */
268static bool bbm_in_data_chunk(struct gpmi_nand_data *this,
269			unsigned int *chunk_num)
270{
271	struct bch_geometry *geo = &this->bch_geometry;
272	struct nand_chip *chip = &this->nand;
273	struct mtd_info *mtd = nand_to_mtd(chip);
274	unsigned int i, j;
275
276	if (geo->ecc0_chunk_size != geo->eccn_chunk_size) {
277		dev_err(this->dev,
278			"The size of ecc0_chunk must equal to eccn_chunk\n");
279		return false;
280	}
281
282	i = (mtd->writesize * 8 - geo->metadata_size * 8) /
283		(geo->gf_len * geo->ecc_strength +
284			geo->eccn_chunk_size * 8);
285
286	j = (mtd->writesize * 8 - geo->metadata_size * 8) -
287		(geo->gf_len * geo->ecc_strength +
288			geo->eccn_chunk_size * 8) * i;
289
290	if (j < geo->eccn_chunk_size * 8) {
291		*chunk_num = i+1;
292		dev_dbg(this->dev, "Set ecc to %d and bbm in chunk %d\n",
293				geo->ecc_strength, *chunk_num);
294		return true;
295	}
296
297	return false;
298}
299
300/*
301 * If we can get the ECC information from the nand chip, we do not
302 * need to calculate them ourselves.
303 *
304 * We may have available oob space in this case.
305 */
306static int set_geometry_by_ecc_info(struct gpmi_nand_data *this,
307				    unsigned int ecc_strength,
308				    unsigned int ecc_step)
309{
310	struct bch_geometry *geo = &this->bch_geometry;
311	struct nand_chip *chip = &this->nand;
312	struct mtd_info *mtd = nand_to_mtd(chip);
313	unsigned int block_mark_bit_offset;
314
315	switch (ecc_step) {
316	case SZ_512:
317		geo->gf_len = 13;
318		break;
319	case SZ_1K:
320		geo->gf_len = 14;
321		break;
322	default:
323		dev_err(this->dev,
324			"unsupported nand chip. ecc bits : %d, ecc size : %d\n",
325			nanddev_get_ecc_requirements(&chip->base)->strength,
326			nanddev_get_ecc_requirements(&chip->base)->step_size);
327		return -EINVAL;
328	}
329	geo->ecc0_chunk_size = ecc_step;
330	geo->eccn_chunk_size = ecc_step;
331	geo->ecc_strength = round_up(ecc_strength, 2);
332	if (!gpmi_check_ecc(this))
333		return -EINVAL;
334
335	/* Keep the C >= O */
336	if (geo->eccn_chunk_size < mtd->oobsize) {
337		dev_err(this->dev,
338			"unsupported nand chip. ecc size: %d, oob size : %d\n",
339			ecc_step, mtd->oobsize);
340		return -EINVAL;
341	}
342
343	/* The default value, see comment in the legacy_set_geometry(). */
344	geo->metadata_size = 10;
345
346	geo->ecc_chunk_count = mtd->writesize / geo->eccn_chunk_size;
347
348	/*
349	 * Now, the NAND chip with 2K page(data chunk is 512byte) shows below:
350	 *
351	 *    |                          P                            |
352	 *    |<----------------------------------------------------->|
353	 *    |                                                       |
354	 *    |                                        (Block Mark)   |
355	 *    |                      P'                      |      | |     |
356	 *    |<-------------------------------------------->|  D   | |  O' |
357	 *    |                                              |<---->| |<--->|
358	 *    V                                              V      V V     V
359	 *    +---+----------+-+----------+-+----------+-+----------+-+-----+
360	 *    | M |   data   |E|   data   |E|   data   |E|   data   |E|     |
361	 *    +---+----------+-+----------+-+----------+-+----------+-+-----+
362	 *                                                   ^              ^
363	 *                                                   |      O       |
364	 *                                                   |<------------>|
365	 *                                                   |              |
366	 *
367	 *	P : the page size for BCH module.
368	 *	E : The ECC strength.
369	 *	G : the length of Galois Field.
370	 *	N : The chunk count of per page.
371	 *	M : the metasize of per page.
372	 *	C : the ecc chunk size, aka the "data" above.
373	 *	P': the nand chip's page size.
374	 *	O : the nand chip's oob size.
375	 *	O': the free oob.
376	 *
377	 *	The formula for P is :
378	 *
379	 *	            E * G * N
380	 *	       P = ------------ + P' + M
381	 *                      8
382	 *
383	 * The position of block mark moves forward in the ECC-based view
384	 * of page, and the delta is:
385	 *
386	 *                   E * G * (N - 1)
387	 *             D = (---------------- + M)
388	 *                          8
389	 *
390	 * Please see the comment in legacy_set_geometry().
391	 * With the condition C >= O , we still can get same result.
392	 * So the bit position of the physical block mark within the ECC-based
393	 * view of the page is :
394	 *             (P' - D) * 8
395	 */
396	geo->page_size = mtd->writesize + geo->metadata_size +
397		(geo->gf_len * geo->ecc_strength * geo->ecc_chunk_count) / 8;
398
399	geo->payload_size = mtd->writesize;
400
401	geo->auxiliary_status_offset = ALIGN(geo->metadata_size, 4);
402	geo->auxiliary_size = ALIGN(geo->metadata_size, 4)
403				+ ALIGN(geo->ecc_chunk_count, 4);
404
405	if (!this->swap_block_mark)
406		return 0;
407
408	/* For bit swap. */
409	block_mark_bit_offset = mtd->writesize * 8 -
410		(geo->ecc_strength * geo->gf_len * (geo->ecc_chunk_count - 1)
411				+ geo->metadata_size * 8);
412
413	geo->block_mark_byte_offset = block_mark_bit_offset / 8;
414	geo->block_mark_bit_offset  = block_mark_bit_offset % 8;
415	return 0;
416}
417
418/*
419 *  Calculate the ECC strength by hand:
420 *	E : The ECC strength.
421 *	G : the length of Galois Field.
422 *	N : The chunk count of per page.
423 *	O : the oobsize of the NAND chip.
424 *	M : the metasize of per page.
425 *
426 *	The formula is :
427 *		E * G * N
428 *	      ------------ <= (O - M)
429 *                  8
430 *
431 *      So, we get E by:
432 *                    (O - M) * 8
433 *              E <= -------------
434 *                       G * N
435 */
436static inline int get_ecc_strength(struct gpmi_nand_data *this)
437{
438	struct bch_geometry *geo = &this->bch_geometry;
439	struct mtd_info	*mtd = nand_to_mtd(&this->nand);
440	int ecc_strength;
441
442	ecc_strength = ((mtd->oobsize - geo->metadata_size) * 8)
443			/ (geo->gf_len * geo->ecc_chunk_count);
444
445	/* We need the minor even number. */
446	return round_down(ecc_strength, 2);
447}
448
449static int set_geometry_for_large_oob(struct gpmi_nand_data *this)
450{
451	struct bch_geometry *geo = &this->bch_geometry;
452	struct nand_chip *chip = &this->nand;
453	struct mtd_info *mtd = nand_to_mtd(chip);
454	const struct nand_ecc_props *requirements =
455		nanddev_get_ecc_requirements(&chip->base);
456	unsigned int block_mark_bit_offset;
457	unsigned int max_ecc;
458	unsigned int bbm_chunk;
459	unsigned int i;
460
461	/* sanity check for the minimum ecc nand required */
462	if (!(requirements->strength > 0 &&
463	      requirements->step_size > 0))
464		return -EINVAL;
465	geo->ecc_strength = requirements->strength;
466
467	/* check if platform can support this nand */
468	if (!gpmi_check_ecc(this)) {
469		dev_err(this->dev,
470			"unsupported NAND chip, minimum ecc required %d\n",
471			geo->ecc_strength);
472		return -EINVAL;
473	}
474
475	/* calculate the maximum ecc platform can support*/
476	geo->metadata_size = 10;
477	geo->gf_len = 14;
478	geo->ecc0_chunk_size = 1024;
479	geo->eccn_chunk_size = 1024;
480	geo->ecc_chunk_count = mtd->writesize / geo->eccn_chunk_size;
481	max_ecc = min(get_ecc_strength(this),
482		      this->devdata->bch_max_ecc_strength);
483
484	/*
485	 * search a supported ecc strength that makes bbm
486	 * located in data chunk
487	 */
488	geo->ecc_strength = max_ecc;
489	while (!(geo->ecc_strength < requirements->strength)) {
490		if (bbm_in_data_chunk(this, &bbm_chunk))
491			goto geo_setting;
492		geo->ecc_strength -= 2;
493	}
494
495	/* if none of them works, keep using the minimum ecc */
496	/* nand required but changing ecc page layout  */
497	geo->ecc_strength = requirements->strength;
498	/* add extra ecc for meta data */
499	geo->ecc0_chunk_size = 0;
500	geo->ecc_chunk_count = (mtd->writesize / geo->eccn_chunk_size) + 1;
501	geo->ecc_for_meta = 1;
502	/* check if oob can afford this extra ecc chunk */
503	if (mtd->oobsize * 8 < geo->metadata_size * 8 +
504	    geo->gf_len * geo->ecc_strength * geo->ecc_chunk_count) {
505		dev_err(this->dev, "unsupported NAND chip with new layout\n");
506		return -EINVAL;
507	}
508
509	/* calculate in which chunk bbm located */
510	bbm_chunk = (mtd->writesize * 8 - geo->metadata_size * 8 -
511		     geo->gf_len * geo->ecc_strength) /
512		     (geo->gf_len * geo->ecc_strength +
513		     geo->eccn_chunk_size * 8) + 1;
514
515geo_setting:
516
517	geo->page_size = mtd->writesize + geo->metadata_size +
518		(geo->gf_len * geo->ecc_strength * geo->ecc_chunk_count) / 8;
519	geo->payload_size = mtd->writesize;
520
521	/*
522	 * The auxiliary buffer contains the metadata and the ECC status. The
523	 * metadata is padded to the nearest 32-bit boundary. The ECC status
524	 * contains one byte for every ECC chunk, and is also padded to the
525	 * nearest 32-bit boundary.
526	 */
527	geo->auxiliary_status_offset = ALIGN(geo->metadata_size, 4);
528	geo->auxiliary_size = ALIGN(geo->metadata_size, 4)
529				    + ALIGN(geo->ecc_chunk_count, 4);
530
531	if (!this->swap_block_mark)
532		return 0;
533
534	/* calculate the number of ecc chunk behind the bbm */
535	i = (mtd->writesize / geo->eccn_chunk_size) - bbm_chunk + 1;
536
537	block_mark_bit_offset = mtd->writesize * 8 -
538		(geo->ecc_strength * geo->gf_len * (geo->ecc_chunk_count - i)
539		+ geo->metadata_size * 8);
540
541	geo->block_mark_byte_offset = block_mark_bit_offset / 8;
542	geo->block_mark_bit_offset  = block_mark_bit_offset % 8;
543
544	dev_dbg(this->dev, "BCH Geometry :\n"
545		"GF length              : %u\n"
546		"ECC Strength           : %u\n"
547		"Page Size in Bytes     : %u\n"
548		"Metadata Size in Bytes : %u\n"
549		"ECC0 Chunk Size in Bytes: %u\n"
550		"ECCn Chunk Size in Bytes: %u\n"
551		"ECC Chunk Count        : %u\n"
552		"Payload Size in Bytes  : %u\n"
553		"Auxiliary Size in Bytes: %u\n"
554		"Auxiliary Status Offset: %u\n"
555		"Block Mark Byte Offset : %u\n"
556		"Block Mark Bit Offset  : %u\n"
557		"Block Mark in chunk	: %u\n"
558		"Ecc for Meta data	: %u\n",
559		geo->gf_len,
560		geo->ecc_strength,
561		geo->page_size,
562		geo->metadata_size,
563		geo->ecc0_chunk_size,
564		geo->eccn_chunk_size,
565		geo->ecc_chunk_count,
566		geo->payload_size,
567		geo->auxiliary_size,
568		geo->auxiliary_status_offset,
569		geo->block_mark_byte_offset,
570		geo->block_mark_bit_offset,
571		bbm_chunk,
572		geo->ecc_for_meta);
573
574	return 0;
575}
576
577static int legacy_set_geometry(struct gpmi_nand_data *this)
578{
579	struct bch_geometry *geo = &this->bch_geometry;
580	struct mtd_info *mtd = nand_to_mtd(&this->nand);
581	unsigned int metadata_size;
582	unsigned int status_size;
583	unsigned int block_mark_bit_offset;
584
585	/*
586	 * The size of the metadata can be changed, though we set it to 10
587	 * bytes now. But it can't be too large, because we have to save
588	 * enough space for BCH.
589	 */
590	geo->metadata_size = 10;
591
592	/* The default for the length of Galois Field. */
593	geo->gf_len = 13;
594
595	/* The default for chunk size. */
596	geo->ecc0_chunk_size = 512;
597	geo->eccn_chunk_size = 512;
598	while (geo->eccn_chunk_size < mtd->oobsize) {
599		geo->ecc0_chunk_size *= 2; /* keep C >= O */
600		geo->eccn_chunk_size *= 2; /* keep C >= O */
601		geo->gf_len = 14;
602	}
603
604	geo->ecc_chunk_count = mtd->writesize / geo->eccn_chunk_size;
605
606	/* We use the same ECC strength for all chunks. */
607	geo->ecc_strength = get_ecc_strength(this);
608	if (!gpmi_check_ecc(this)) {
609		dev_err(this->dev,
610			"ecc strength: %d cannot be supported by the controller (%d)\n"
611			"try to use minimum ecc strength that NAND chip required\n",
612			geo->ecc_strength,
613			this->devdata->bch_max_ecc_strength);
614		return -EINVAL;
615	}
616
617	geo->page_size = mtd->writesize + geo->metadata_size +
618		(geo->gf_len * geo->ecc_strength * geo->ecc_chunk_count) / 8;
619	geo->payload_size = mtd->writesize;
620
621	/*
622	 * The auxiliary buffer contains the metadata and the ECC status. The
623	 * metadata is padded to the nearest 32-bit boundary. The ECC status
624	 * contains one byte for every ECC chunk, and is also padded to the
625	 * nearest 32-bit boundary.
626	 */
627	metadata_size = ALIGN(geo->metadata_size, 4);
628	status_size   = ALIGN(geo->ecc_chunk_count, 4);
629
630	geo->auxiliary_size = metadata_size + status_size;
631	geo->auxiliary_status_offset = metadata_size;
632
633	if (!this->swap_block_mark)
634		return 0;
635
636	/*
637	 * We need to compute the byte and bit offsets of
638	 * the physical block mark within the ECC-based view of the page.
639	 *
640	 * NAND chip with 2K page shows below:
641	 *                                             (Block Mark)
642	 *                                                   |      |
643	 *                                                   |  D   |
644	 *                                                   |<---->|
645	 *                                                   V      V
646	 *    +---+----------+-+----------+-+----------+-+----------+-+
647	 *    | M |   data   |E|   data   |E|   data   |E|   data   |E|
648	 *    +---+----------+-+----------+-+----------+-+----------+-+
649	 *
650	 * The position of block mark moves forward in the ECC-based view
651	 * of page, and the delta is:
652	 *
653	 *                   E * G * (N - 1)
654	 *             D = (---------------- + M)
655	 *                          8
656	 *
657	 * With the formula to compute the ECC strength, and the condition
658	 *       : C >= O         (C is the ecc chunk size)
659	 *
660	 * It's easy to deduce to the following result:
661	 *
662	 *         E * G       (O - M)      C - M         C - M
663	 *      ----------- <= ------- <=  --------  <  ---------
664	 *           8            N           N          (N - 1)
665	 *
666	 *  So, we get:
667	 *
668	 *                   E * G * (N - 1)
669	 *             D = (---------------- + M) < C
670	 *                          8
671	 *
672	 *  The above inequality means the position of block mark
673	 *  within the ECC-based view of the page is still in the data chunk,
674	 *  and it's NOT in the ECC bits of the chunk.
675	 *
676	 *  Use the following to compute the bit position of the
677	 *  physical block mark within the ECC-based view of the page:
678	 *          (page_size - D) * 8
679	 *
680	 *  --Huang Shijie
681	 */
682	block_mark_bit_offset = mtd->writesize * 8 -
683		(geo->ecc_strength * geo->gf_len * (geo->ecc_chunk_count - 1)
684				+ geo->metadata_size * 8);
685
686	geo->block_mark_byte_offset = block_mark_bit_offset / 8;
687	geo->block_mark_bit_offset  = block_mark_bit_offset % 8;
688	return 0;
689}
690
691static int common_nfc_set_geometry(struct gpmi_nand_data *this)
692{
693	struct nand_chip *chip = &this->nand;
694	struct mtd_info *mtd = nand_to_mtd(&this->nand);
695	const struct nand_ecc_props *requirements =
696		nanddev_get_ecc_requirements(&chip->base);
697	bool use_minimun_ecc;
698	int err;
699
700	use_minimun_ecc = of_property_read_bool(this->dev->of_node,
701						"fsl,use-minimum-ecc");
702
703	/* use legacy bch geometry settings by default*/
704	if ((!use_minimun_ecc && mtd->oobsize < 1024) ||
705	    !(requirements->strength > 0 && requirements->step_size > 0)) {
706		dev_dbg(this->dev, "use legacy bch geometry\n");
707		err = legacy_set_geometry(this);
708		if (!err)
709			return 0;
710	}
711
712	/* for large oob nand */
713	if (mtd->oobsize > 1024) {
714		dev_dbg(this->dev, "use large oob bch geometry\n");
715		err = set_geometry_for_large_oob(this);
716		if (!err)
717			return 0;
718	}
719
720	/* otherwise use the minimum ecc nand chip required */
721	dev_dbg(this->dev, "use minimum ecc bch geometry\n");
722	err = set_geometry_by_ecc_info(this, requirements->strength,
723					requirements->step_size);
724	if (err)
725		dev_err(this->dev, "none of the bch geometry setting works\n");
726
727	return err;
728}
729
730/* Configures the geometry for BCH.  */
731static int bch_set_geometry(struct gpmi_nand_data *this)
732{
733	struct resources *r = &this->resources;
734	int ret;
735
736	ret = common_nfc_set_geometry(this);
737	if (ret)
738		return ret;
739
740	ret = pm_runtime_get_sync(this->dev);
741	if (ret < 0) {
742		pm_runtime_put_autosuspend(this->dev);
743		return ret;
744	}
745
746	/*
747	* Due to erratum #2847 of the MX23, the BCH cannot be soft reset on this
748	* chip, otherwise it will lock up. So we skip resetting BCH on the MX23.
749	* and MX28.
750	*/
751	ret = gpmi_reset_block(r->bch_regs, GPMI_IS_MXS(this));
752	if (ret)
753		goto err_out;
754
755	/* Set *all* chip selects to use layout 0. */
756	writel(0, r->bch_regs + HW_BCH_LAYOUTSELECT);
757
758	ret = 0;
759err_out:
760	pm_runtime_mark_last_busy(this->dev);
761	pm_runtime_put_autosuspend(this->dev);
762
763	return ret;
764}
765
766/*
767 * <1> Firstly, we should know what's the GPMI-clock means.
768 *     The GPMI-clock is the internal clock in the gpmi nand controller.
769 *     If you set 100MHz to gpmi nand controller, the GPMI-clock's period
770 *     is 10ns. Mark the GPMI-clock's period as GPMI-clock-period.
771 *
772 * <2> Secondly, we should know what's the frequency on the nand chip pins.
773 *     The frequency on the nand chip pins is derived from the GPMI-clock.
774 *     We can get it from the following equation:
775 *
776 *         F = G / (DS + DH)
777 *
778 *         F  : the frequency on the nand chip pins.
779 *         G  : the GPMI clock, such as 100MHz.
780 *         DS : GPMI_HW_GPMI_TIMING0:DATA_SETUP
781 *         DH : GPMI_HW_GPMI_TIMING0:DATA_HOLD
782 *
783 * <3> Thirdly, when the frequency on the nand chip pins is above 33MHz,
784 *     the nand EDO(extended Data Out) timing could be applied.
785 *     The GPMI implements a feedback read strobe to sample the read data.
786 *     The feedback read strobe can be delayed to support the nand EDO timing
787 *     where the read strobe may deasserts before the read data is valid, and
788 *     read data is valid for some time after read strobe.
789 *
790 *     The following figure illustrates some aspects of a NAND Flash read:
791 *
792 *                   |<---tREA---->|
793 *                   |             |
794 *                   |         |   |
795 *                   |<--tRP-->|   |
796 *                   |         |   |
797 *                  __          ___|__________________________________
798 *     RDN            \________/   |
799 *                                 |
800 *                                 /---------\
801 *     Read Data    --------------<           >---------
802 *                                 \---------/
803 *                                |     |
804 *                                |<-D->|
805 *     FeedbackRDN  ________             ____________
806 *                          \___________/
807 *
808 *          D stands for delay, set in the HW_GPMI_CTRL1:RDN_DELAY.
809 *
810 *
811 * <4> Now, we begin to describe how to compute the right RDN_DELAY.
812 *
813 *  4.1) From the aspect of the nand chip pins:
814 *        Delay = (tREA + C - tRP)               {1}
815 *
816 *        tREA : the maximum read access time.
817 *        C    : a constant to adjust the delay. default is 4000ps.
818 *        tRP  : the read pulse width, which is exactly:
819 *                   tRP = (GPMI-clock-period) * DATA_SETUP
820 *
821 *  4.2) From the aspect of the GPMI nand controller:
822 *         Delay = RDN_DELAY * 0.125 * RP        {2}
823 *
824 *         RP   : the DLL reference period.
825 *            if (GPMI-clock-period > DLL_THRETHOLD)
826 *                   RP = GPMI-clock-period / 2;
827 *            else
828 *                   RP = GPMI-clock-period;
829 *
830 *            Set the HW_GPMI_CTRL1:HALF_PERIOD if GPMI-clock-period
831 *            is greater DLL_THRETHOLD. In other SOCs, the DLL_THRETHOLD
832 *            is 16000ps, but in mx6q, we use 12000ps.
833 *
834 *  4.3) since {1} equals {2}, we get:
835 *
836 *                     (tREA + 4000 - tRP) * 8
837 *         RDN_DELAY = -----------------------     {3}
838 *                           RP
839 */
840static int gpmi_nfc_compute_timings(struct gpmi_nand_data *this,
841				    const struct nand_sdr_timings *sdr)
842{
843	struct gpmi_nfc_hardware_timing *hw = &this->hw;
844	struct resources *r = &this->resources;
845	unsigned int dll_threshold_ps = this->devdata->max_chain_delay;
846	unsigned int period_ps, reference_period_ps;
847	unsigned int data_setup_cycles, data_hold_cycles, addr_setup_cycles;
848	unsigned int tRP_ps;
849	bool use_half_period;
850	int sample_delay_ps, sample_delay_factor;
851	unsigned int busy_timeout_cycles;
852	u8 wrn_dly_sel;
853	unsigned long clk_rate, min_rate;
854	u64 busy_timeout_ps;
855
856	if (sdr->tRC_min >= 30000) {
857		/* ONFI non-EDO modes [0-3] */
858		hw->clk_rate = 22000000;
859		min_rate = 0;
860		wrn_dly_sel = BV_GPMI_CTRL1_WRN_DLY_SEL_4_TO_8NS;
861	} else if (sdr->tRC_min >= 25000) {
862		/* ONFI EDO mode 4 */
863		hw->clk_rate = 80000000;
864		min_rate = 22000000;
865		wrn_dly_sel = BV_GPMI_CTRL1_WRN_DLY_SEL_NO_DELAY;
866	} else {
867		/* ONFI EDO mode 5 */
868		hw->clk_rate = 100000000;
869		min_rate = 80000000;
870		wrn_dly_sel = BV_GPMI_CTRL1_WRN_DLY_SEL_NO_DELAY;
871	}
872
873	clk_rate = clk_round_rate(r->clock[0], hw->clk_rate);
874	if (clk_rate <= min_rate) {
875		dev_err(this->dev, "clock setting: expected %ld, got %ld\n",
876			hw->clk_rate, clk_rate);
877		return -ENOTSUPP;
878	}
879
880	hw->clk_rate = clk_rate;
881	/* SDR core timings are given in picoseconds */
882	period_ps = div_u64((u64)NSEC_PER_SEC * 1000, hw->clk_rate);
883
884	addr_setup_cycles = TO_CYCLES(sdr->tALS_min, period_ps);
885	data_setup_cycles = TO_CYCLES(sdr->tDS_min, period_ps);
886	data_hold_cycles = TO_CYCLES(sdr->tDH_min, period_ps);
887	busy_timeout_ps = max(sdr->tBERS_max, sdr->tPROG_max);
888	busy_timeout_cycles = TO_CYCLES(busy_timeout_ps, period_ps);
889
890	hw->timing0 = BF_GPMI_TIMING0_ADDRESS_SETUP(addr_setup_cycles) |
891		      BF_GPMI_TIMING0_DATA_HOLD(data_hold_cycles) |
892		      BF_GPMI_TIMING0_DATA_SETUP(data_setup_cycles);
893	hw->timing1 = BF_GPMI_TIMING1_BUSY_TIMEOUT(DIV_ROUND_UP(busy_timeout_cycles, 4096));
894
895	/*
896	 * Derive NFC ideal delay from {3}:
897	 *
898	 *                     (tREA + 4000 - tRP) * 8
899	 *         RDN_DELAY = -----------------------
900	 *                                RP
901	 */
902	if (period_ps > dll_threshold_ps) {
903		use_half_period = true;
904		reference_period_ps = period_ps / 2;
905	} else {
906		use_half_period = false;
907		reference_period_ps = period_ps;
908	}
909
910	tRP_ps = data_setup_cycles * period_ps;
911	sample_delay_ps = (sdr->tREA_max + 4000 - tRP_ps) * 8;
912	if (sample_delay_ps > 0)
913		sample_delay_factor = sample_delay_ps / reference_period_ps;
914	else
915		sample_delay_factor = 0;
916
917	hw->ctrl1n = BF_GPMI_CTRL1_WRN_DLY_SEL(wrn_dly_sel);
918	if (sample_delay_factor)
919		hw->ctrl1n |= BF_GPMI_CTRL1_RDN_DELAY(sample_delay_factor) |
920			      BM_GPMI_CTRL1_DLL_ENABLE |
921			      (use_half_period ? BM_GPMI_CTRL1_HALF_PERIOD : 0);
922	return 0;
923}
924
925static int gpmi_nfc_apply_timings(struct gpmi_nand_data *this)
926{
927	struct gpmi_nfc_hardware_timing *hw = &this->hw;
928	struct resources *r = &this->resources;
929	void __iomem *gpmi_regs = r->gpmi_regs;
930	unsigned int dll_wait_time_us;
931	int ret;
932
933	/* Clock dividers do NOT guarantee a clean clock signal on its output
934	 * during the change of the divide factor on i.MX6Q/UL/SX. On i.MX7/8,
935	 * all clock dividers provide these guarantee.
936	 */
937	if (GPMI_IS_MX6Q(this) || GPMI_IS_MX6SX(this))
938		clk_disable_unprepare(r->clock[0]);
939
940	ret = clk_set_rate(r->clock[0], hw->clk_rate);
941	if (ret) {
942		dev_err(this->dev, "cannot set clock rate to %lu Hz: %d\n", hw->clk_rate, ret);
943		return ret;
944	}
945
946	if (GPMI_IS_MX6Q(this) || GPMI_IS_MX6SX(this)) {
947		ret = clk_prepare_enable(r->clock[0]);
948		if (ret)
949			return ret;
950	}
951
952	writel(hw->timing0, gpmi_regs + HW_GPMI_TIMING0);
953	writel(hw->timing1, gpmi_regs + HW_GPMI_TIMING1);
954
955	/*
956	 * Clear several CTRL1 fields, DLL must be disabled when setting
957	 * RDN_DELAY or HALF_PERIOD.
958	 */
959	writel(BM_GPMI_CTRL1_CLEAR_MASK, gpmi_regs + HW_GPMI_CTRL1_CLR);
960	writel(hw->ctrl1n, gpmi_regs + HW_GPMI_CTRL1_SET);
961
962	/* Wait 64 clock cycles before using the GPMI after enabling the DLL */
963	dll_wait_time_us = USEC_PER_SEC / hw->clk_rate * 64;
964	if (!dll_wait_time_us)
965		dll_wait_time_us = 1;
966
967	/* Wait for the DLL to settle. */
968	udelay(dll_wait_time_us);
969
970	return 0;
971}
972
973static int gpmi_setup_interface(struct nand_chip *chip, int chipnr,
974				const struct nand_interface_config *conf)
975{
976	struct gpmi_nand_data *this = nand_get_controller_data(chip);
977	const struct nand_sdr_timings *sdr;
978	int ret;
979
980	/* Retrieve required NAND timings */
981	sdr = nand_get_sdr_timings(conf);
982	if (IS_ERR(sdr))
983		return PTR_ERR(sdr);
984
985	/* Only MX28/MX6 GPMI controller can reach EDO timings */
986	if (sdr->tRC_min <= 25000 && !GPMI_IS_MX28(this) && !GPMI_IS_MX6(this))
987		return -ENOTSUPP;
988
989	/* Stop here if this call was just a check */
990	if (chipnr < 0)
991		return 0;
992
993	/* Do the actual derivation of the controller timings */
994	ret = gpmi_nfc_compute_timings(this, sdr);
995	if (ret)
996		return ret;
997
998	this->hw.must_apply_timings = true;
999
1000	return 0;
1001}
1002
1003/* Clears a BCH interrupt. */
1004static void gpmi_clear_bch(struct gpmi_nand_data *this)
1005{
1006	struct resources *r = &this->resources;
1007	writel(BM_BCH_CTRL_COMPLETE_IRQ, r->bch_regs + HW_BCH_CTRL_CLR);
1008}
1009
1010static struct dma_chan *get_dma_chan(struct gpmi_nand_data *this)
1011{
1012	/* We use the DMA channel 0 to access all the nand chips. */
1013	return this->dma_chans[0];
1014}
1015
1016/* This will be called after the DMA operation is finished. */
1017static void dma_irq_callback(void *param)
1018{
1019	struct gpmi_nand_data *this = param;
1020	struct completion *dma_c = &this->dma_done;
1021
1022	complete(dma_c);
1023}
1024
1025static irqreturn_t bch_irq(int irq, void *cookie)
1026{
1027	struct gpmi_nand_data *this = cookie;
1028
1029	gpmi_clear_bch(this);
1030	complete(&this->bch_done);
1031	return IRQ_HANDLED;
1032}
1033
1034static int gpmi_raw_len_to_len(struct gpmi_nand_data *this, int raw_len)
1035{
1036	/*
1037	 * raw_len is the length to read/write including bch data which
1038	 * we are passed in exec_op. Calculate the data length from it.
1039	 */
1040	if (this->bch)
1041		return ALIGN_DOWN(raw_len, this->bch_geometry.eccn_chunk_size);
1042	else
1043		return raw_len;
1044}
1045
1046/* Can we use the upper's buffer directly for DMA? */
1047static bool prepare_data_dma(struct gpmi_nand_data *this, const void *buf,
1048			     int raw_len, struct scatterlist *sgl,
1049			     enum dma_data_direction dr)
1050{
1051	int ret;
1052	int len = gpmi_raw_len_to_len(this, raw_len);
1053
1054	/* first try to map the upper buffer directly */
1055	if (virt_addr_valid(buf) && !object_is_on_stack(buf)) {
1056		sg_init_one(sgl, buf, len);
1057		ret = dma_map_sg(this->dev, sgl, 1, dr);
1058		if (ret == 0)
1059			goto map_fail;
1060
1061		return true;
1062	}
1063
1064map_fail:
1065	/* We have to use our own DMA buffer. */
1066	sg_init_one(sgl, this->data_buffer_dma, len);
1067
1068	if (dr == DMA_TO_DEVICE && buf != this->data_buffer_dma)
1069		memcpy(this->data_buffer_dma, buf, len);
1070
1071	dma_map_sg(this->dev, sgl, 1, dr);
1072
1073	return false;
1074}
1075
1076/* add our owner bbt descriptor */
1077static uint8_t scan_ff_pattern[] = { 0xff };
1078static struct nand_bbt_descr gpmi_bbt_descr = {
1079	.options	= 0,
1080	.offs		= 0,
1081	.len		= 1,
1082	.pattern	= scan_ff_pattern
1083};
1084
1085/*
1086 * We may change the layout if we can get the ECC info from the datasheet,
1087 * else we will use all the (page + OOB).
1088 */
1089static int gpmi_ooblayout_ecc(struct mtd_info *mtd, int section,
1090			      struct mtd_oob_region *oobregion)
1091{
1092	struct nand_chip *chip = mtd_to_nand(mtd);
1093	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1094	struct bch_geometry *geo = &this->bch_geometry;
1095
1096	if (section)
1097		return -ERANGE;
1098
1099	oobregion->offset = 0;
1100	oobregion->length = geo->page_size - mtd->writesize;
1101
1102	return 0;
1103}
1104
1105static int gpmi_ooblayout_free(struct mtd_info *mtd, int section,
1106			       struct mtd_oob_region *oobregion)
1107{
1108	struct nand_chip *chip = mtd_to_nand(mtd);
1109	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1110	struct bch_geometry *geo = &this->bch_geometry;
1111
1112	if (section)
1113		return -ERANGE;
1114
1115	/* The available oob size we have. */
1116	if (geo->page_size < mtd->writesize + mtd->oobsize) {
1117		oobregion->offset = geo->page_size - mtd->writesize;
1118		oobregion->length = mtd->oobsize - oobregion->offset;
1119	}
1120
1121	return 0;
1122}
1123
1124static const char * const gpmi_clks_for_mx2x[] = {
1125	"gpmi_io",
1126};
1127
1128static const struct mtd_ooblayout_ops gpmi_ooblayout_ops = {
1129	.ecc = gpmi_ooblayout_ecc,
1130	.free = gpmi_ooblayout_free,
1131};
1132
1133static const struct gpmi_devdata gpmi_devdata_imx23 = {
1134	.type = IS_MX23,
1135	.bch_max_ecc_strength = 20,
1136	.max_chain_delay = 16000,
1137	.clks = gpmi_clks_for_mx2x,
1138	.clks_count = ARRAY_SIZE(gpmi_clks_for_mx2x),
1139};
1140
1141static const struct gpmi_devdata gpmi_devdata_imx28 = {
1142	.type = IS_MX28,
1143	.bch_max_ecc_strength = 20,
1144	.max_chain_delay = 16000,
1145	.clks = gpmi_clks_for_mx2x,
1146	.clks_count = ARRAY_SIZE(gpmi_clks_for_mx2x),
1147};
1148
1149static const char * const gpmi_clks_for_mx6[] = {
1150	"gpmi_io", "gpmi_apb", "gpmi_bch", "gpmi_bch_apb", "per1_bch",
1151};
1152
1153static const struct gpmi_devdata gpmi_devdata_imx6q = {
1154	.type = IS_MX6Q,
1155	.bch_max_ecc_strength = 40,
1156	.max_chain_delay = 12000,
1157	.clks = gpmi_clks_for_mx6,
1158	.clks_count = ARRAY_SIZE(gpmi_clks_for_mx6),
1159};
1160
1161static const struct gpmi_devdata gpmi_devdata_imx6sx = {
1162	.type = IS_MX6SX,
1163	.bch_max_ecc_strength = 62,
1164	.max_chain_delay = 12000,
1165	.clks = gpmi_clks_for_mx6,
1166	.clks_count = ARRAY_SIZE(gpmi_clks_for_mx6),
1167};
1168
1169static const char * const gpmi_clks_for_mx7d[] = {
1170	"gpmi_io", "gpmi_bch_apb",
1171};
1172
1173static const struct gpmi_devdata gpmi_devdata_imx7d = {
1174	.type = IS_MX7D,
1175	.bch_max_ecc_strength = 62,
1176	.max_chain_delay = 12000,
1177	.clks = gpmi_clks_for_mx7d,
1178	.clks_count = ARRAY_SIZE(gpmi_clks_for_mx7d),
1179};
1180
1181static int acquire_register_block(struct gpmi_nand_data *this,
1182				  const char *res_name)
1183{
1184	struct platform_device *pdev = this->pdev;
1185	struct resources *res = &this->resources;
1186	void __iomem *p;
1187
1188	p = devm_platform_ioremap_resource_byname(pdev, res_name);
1189	if (IS_ERR(p))
1190		return PTR_ERR(p);
1191
1192	if (!strcmp(res_name, GPMI_NAND_GPMI_REGS_ADDR_RES_NAME))
1193		res->gpmi_regs = p;
1194	else if (!strcmp(res_name, GPMI_NAND_BCH_REGS_ADDR_RES_NAME))
1195		res->bch_regs = p;
1196	else
1197		dev_err(this->dev, "unknown resource name : %s\n", res_name);
1198
1199	return 0;
1200}
1201
1202static int acquire_bch_irq(struct gpmi_nand_data *this, irq_handler_t irq_h)
1203{
1204	struct platform_device *pdev = this->pdev;
1205	const char *res_name = GPMI_NAND_BCH_INTERRUPT_RES_NAME;
1206	int err;
1207
1208	err = platform_get_irq_byname(pdev, res_name);
1209	if (err < 0)
1210		return err;
1211
1212	err = devm_request_irq(this->dev, err, irq_h, 0, res_name, this);
1213	if (err)
1214		dev_err(this->dev, "error requesting BCH IRQ\n");
1215
1216	return err;
1217}
1218
1219static void release_dma_channels(struct gpmi_nand_data *this)
1220{
1221	unsigned int i;
1222	for (i = 0; i < DMA_CHANS; i++)
1223		if (this->dma_chans[i]) {
1224			dma_release_channel(this->dma_chans[i]);
1225			this->dma_chans[i] = NULL;
1226		}
1227}
1228
1229static int acquire_dma_channels(struct gpmi_nand_data *this)
1230{
1231	struct platform_device *pdev = this->pdev;
1232	struct dma_chan *dma_chan;
1233	int ret = 0;
1234
1235	/* request dma channel */
1236	dma_chan = dma_request_chan(&pdev->dev, "rx-tx");
1237	if (IS_ERR(dma_chan)) {
1238		ret = dev_err_probe(this->dev, PTR_ERR(dma_chan),
1239				    "DMA channel request failed\n");
1240		release_dma_channels(this);
1241	} else {
1242		this->dma_chans[0] = dma_chan;
1243	}
1244
1245	return ret;
1246}
1247
1248static int gpmi_get_clks(struct gpmi_nand_data *this)
1249{
1250	struct resources *r = &this->resources;
1251	struct clk *clk;
1252	int err, i;
1253
1254	for (i = 0; i < this->devdata->clks_count; i++) {
1255		clk = devm_clk_get(this->dev, this->devdata->clks[i]);
1256		if (IS_ERR(clk)) {
1257			err = PTR_ERR(clk);
1258			goto err_clock;
1259		}
1260
1261		r->clock[i] = clk;
1262	}
1263
1264	return 0;
1265
1266err_clock:
1267	dev_dbg(this->dev, "failed in finding the clocks.\n");
1268	return err;
1269}
1270
1271static int acquire_resources(struct gpmi_nand_data *this)
1272{
1273	int ret;
1274
1275	ret = acquire_register_block(this, GPMI_NAND_GPMI_REGS_ADDR_RES_NAME);
1276	if (ret)
1277		goto exit_regs;
1278
1279	ret = acquire_register_block(this, GPMI_NAND_BCH_REGS_ADDR_RES_NAME);
1280	if (ret)
1281		goto exit_regs;
1282
1283	ret = acquire_bch_irq(this, bch_irq);
1284	if (ret)
1285		goto exit_regs;
1286
1287	ret = acquire_dma_channels(this);
1288	if (ret)
1289		goto exit_regs;
1290
1291	ret = gpmi_get_clks(this);
1292	if (ret)
1293		goto exit_clock;
1294	return 0;
1295
1296exit_clock:
1297	release_dma_channels(this);
1298exit_regs:
1299	return ret;
1300}
1301
1302static void release_resources(struct gpmi_nand_data *this)
1303{
1304	release_dma_channels(this);
1305}
1306
1307static void gpmi_free_dma_buffer(struct gpmi_nand_data *this)
1308{
1309	struct device *dev = this->dev;
1310	struct bch_geometry *geo = &this->bch_geometry;
1311
1312	if (this->auxiliary_virt && virt_addr_valid(this->auxiliary_virt))
1313		dma_free_coherent(dev, geo->auxiliary_size,
1314					this->auxiliary_virt,
1315					this->auxiliary_phys);
1316	kfree(this->data_buffer_dma);
1317	kfree(this->raw_buffer);
1318
1319	this->data_buffer_dma	= NULL;
1320	this->raw_buffer	= NULL;
1321}
1322
1323/* Allocate the DMA buffers */
1324static int gpmi_alloc_dma_buffer(struct gpmi_nand_data *this)
1325{
1326	struct bch_geometry *geo = &this->bch_geometry;
1327	struct device *dev = this->dev;
1328	struct mtd_info *mtd = nand_to_mtd(&this->nand);
1329
1330	/*
1331	 * [2] Allocate a read/write data buffer.
1332	 *     The gpmi_alloc_dma_buffer can be called twice.
1333	 *     We allocate a PAGE_SIZE length buffer if gpmi_alloc_dma_buffer
1334	 *     is called before the NAND identification; and we allocate a
1335	 *     buffer of the real NAND page size when the gpmi_alloc_dma_buffer
1336	 *     is called after.
1337	 */
1338	this->data_buffer_dma = kzalloc(mtd->writesize ?: PAGE_SIZE,
1339					GFP_DMA | GFP_KERNEL);
1340	if (this->data_buffer_dma == NULL)
1341		goto error_alloc;
1342
1343	this->auxiliary_virt = dma_alloc_coherent(dev, geo->auxiliary_size,
1344					&this->auxiliary_phys, GFP_DMA);
1345	if (!this->auxiliary_virt)
1346		goto error_alloc;
1347
1348	this->raw_buffer = kzalloc((mtd->writesize ?: PAGE_SIZE) + mtd->oobsize, GFP_KERNEL);
1349	if (!this->raw_buffer)
1350		goto error_alloc;
1351
1352	return 0;
1353
1354error_alloc:
1355	gpmi_free_dma_buffer(this);
1356	return -ENOMEM;
1357}
1358
1359/*
1360 * Handles block mark swapping.
1361 * It can be called in swapping the block mark, or swapping it back,
1362 * because the operations are the same.
1363 */
1364static void block_mark_swapping(struct gpmi_nand_data *this,
1365				void *payload, void *auxiliary)
1366{
1367	struct bch_geometry *nfc_geo = &this->bch_geometry;
1368	unsigned char *p;
1369	unsigned char *a;
1370	unsigned int  bit;
1371	unsigned char mask;
1372	unsigned char from_data;
1373	unsigned char from_oob;
1374
1375	if (!this->swap_block_mark)
1376		return;
1377
1378	/*
1379	 * If control arrives here, we're swapping. Make some convenience
1380	 * variables.
1381	 */
1382	bit = nfc_geo->block_mark_bit_offset;
1383	p   = payload + nfc_geo->block_mark_byte_offset;
1384	a   = auxiliary;
1385
1386	/*
1387	 * Get the byte from the data area that overlays the block mark. Since
1388	 * the ECC engine applies its own view to the bits in the page, the
1389	 * physical block mark won't (in general) appear on a byte boundary in
1390	 * the data.
1391	 */
1392	from_data = (p[0] >> bit) | (p[1] << (8 - bit));
1393
1394	/* Get the byte from the OOB. */
1395	from_oob = a[0];
1396
1397	/* Swap them. */
1398	a[0] = from_data;
1399
1400	mask = (0x1 << bit) - 1;
1401	p[0] = (p[0] & mask) | (from_oob << bit);
1402
1403	mask = ~0 << bit;
1404	p[1] = (p[1] & mask) | (from_oob >> (8 - bit));
1405}
1406
1407static int gpmi_count_bitflips(struct nand_chip *chip, void *buf, int first,
1408			       int last, int meta)
1409{
1410	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1411	struct bch_geometry *nfc_geo = &this->bch_geometry;
1412	struct mtd_info *mtd = nand_to_mtd(chip);
1413	int i;
1414	unsigned char *status;
1415	unsigned int max_bitflips = 0;
1416
1417	/* Loop over status bytes, accumulating ECC status. */
1418	status = this->auxiliary_virt + ALIGN(meta, 4);
1419
1420	for (i = first; i < last; i++, status++) {
1421		if ((*status == STATUS_GOOD) || (*status == STATUS_ERASED))
1422			continue;
1423
1424		if (*status == STATUS_UNCORRECTABLE) {
1425			int eccbits = nfc_geo->ecc_strength * nfc_geo->gf_len;
1426			u8 *eccbuf = this->raw_buffer;
1427			int offset, bitoffset;
1428			int eccbytes;
1429			int flips;
1430
1431			/* Read ECC bytes into our internal raw_buffer */
1432			offset = nfc_geo->metadata_size * 8;
1433			offset += ((8 * nfc_geo->eccn_chunk_size) + eccbits) * (i + 1);
1434			offset -= eccbits;
1435			bitoffset = offset % 8;
1436			eccbytes = DIV_ROUND_UP(offset + eccbits, 8);
1437			offset /= 8;
1438			eccbytes -= offset;
1439			nand_change_read_column_op(chip, offset, eccbuf,
1440						   eccbytes, false);
1441
1442			/*
1443			 * ECC data are not byte aligned and we may have
1444			 * in-band data in the first and last byte of
1445			 * eccbuf. Set non-eccbits to one so that
1446			 * nand_check_erased_ecc_chunk() does not count them
1447			 * as bitflips.
1448			 */
1449			if (bitoffset)
1450				eccbuf[0] |= GENMASK(bitoffset - 1, 0);
1451
1452			bitoffset = (bitoffset + eccbits) % 8;
1453			if (bitoffset)
1454				eccbuf[eccbytes - 1] |= GENMASK(7, bitoffset);
1455
1456			/*
1457			 * The ECC hardware has an uncorrectable ECC status
1458			 * code in case we have bitflips in an erased page. As
1459			 * nothing was written into this subpage the ECC is
1460			 * obviously wrong and we can not trust it. We assume
1461			 * at this point that we are reading an erased page and
1462			 * try to correct the bitflips in buffer up to
1463			 * ecc_strength bitflips. If this is a page with random
1464			 * data, we exceed this number of bitflips and have a
1465			 * ECC failure. Otherwise we use the corrected buffer.
1466			 */
1467			if (i == 0) {
1468				/* The first block includes metadata */
1469				flips = nand_check_erased_ecc_chunk(
1470						buf + i * nfc_geo->eccn_chunk_size,
1471						nfc_geo->eccn_chunk_size,
1472						eccbuf, eccbytes,
1473						this->auxiliary_virt,
1474						nfc_geo->metadata_size,
1475						nfc_geo->ecc_strength);
1476			} else {
1477				flips = nand_check_erased_ecc_chunk(
1478						buf + i * nfc_geo->eccn_chunk_size,
1479						nfc_geo->eccn_chunk_size,
1480						eccbuf, eccbytes,
1481						NULL, 0,
1482						nfc_geo->ecc_strength);
1483			}
1484
1485			if (flips > 0) {
1486				max_bitflips = max_t(unsigned int, max_bitflips,
1487						     flips);
1488				mtd->ecc_stats.corrected += flips;
1489				continue;
1490			}
1491
1492			mtd->ecc_stats.failed++;
1493			continue;
1494		}
1495
1496		mtd->ecc_stats.corrected += *status;
1497		max_bitflips = max_t(unsigned int, max_bitflips, *status);
1498	}
1499
1500	return max_bitflips;
1501}
1502
1503static void gpmi_bch_layout_std(struct gpmi_nand_data *this)
1504{
1505	struct bch_geometry *geo = &this->bch_geometry;
1506	unsigned int ecc_strength = geo->ecc_strength >> 1;
1507	unsigned int gf_len = geo->gf_len;
1508	unsigned int block0_size = geo->ecc0_chunk_size;
1509	unsigned int blockn_size = geo->eccn_chunk_size;
1510
1511	this->bch_flashlayout0 =
1512		BF_BCH_FLASH0LAYOUT0_NBLOCKS(geo->ecc_chunk_count - 1) |
1513		BF_BCH_FLASH0LAYOUT0_META_SIZE(geo->metadata_size) |
1514		BF_BCH_FLASH0LAYOUT0_ECC0(ecc_strength, this) |
1515		BF_BCH_FLASH0LAYOUT0_GF(gf_len, this) |
1516		BF_BCH_FLASH0LAYOUT0_DATA0_SIZE(block0_size, this);
1517
1518	this->bch_flashlayout1 =
1519		BF_BCH_FLASH0LAYOUT1_PAGE_SIZE(geo->page_size) |
1520		BF_BCH_FLASH0LAYOUT1_ECCN(ecc_strength, this) |
1521		BF_BCH_FLASH0LAYOUT1_GF(gf_len, this) |
1522		BF_BCH_FLASH0LAYOUT1_DATAN_SIZE(blockn_size, this);
1523}
1524
1525static int gpmi_ecc_read_page(struct nand_chip *chip, uint8_t *buf,
1526			      int oob_required, int page)
1527{
1528	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1529	struct mtd_info *mtd = nand_to_mtd(chip);
1530	struct bch_geometry *geo = &this->bch_geometry;
1531	unsigned int max_bitflips;
1532	int ret;
1533
1534	gpmi_bch_layout_std(this);
1535	this->bch = true;
1536
1537	ret = nand_read_page_op(chip, page, 0, buf, geo->page_size);
1538	if (ret)
1539		return ret;
1540
1541	max_bitflips = gpmi_count_bitflips(chip, buf, 0,
1542					   geo->ecc_chunk_count,
1543					   geo->auxiliary_status_offset);
1544
1545	/* handle the block mark swapping */
1546	block_mark_swapping(this, buf, this->auxiliary_virt);
1547
1548	if (oob_required) {
1549		/*
1550		 * It's time to deliver the OOB bytes. See gpmi_ecc_read_oob()
1551		 * for details about our policy for delivering the OOB.
1552		 *
1553		 * We fill the caller's buffer with set bits, and then copy the
1554		 * block mark to th caller's buffer. Note that, if block mark
1555		 * swapping was necessary, it has already been done, so we can
1556		 * rely on the first byte of the auxiliary buffer to contain
1557		 * the block mark.
1558		 */
1559		memset(chip->oob_poi, ~0, mtd->oobsize);
1560		chip->oob_poi[0] = ((uint8_t *)this->auxiliary_virt)[0];
1561	}
1562
1563	return max_bitflips;
1564}
1565
1566/* Fake a virtual small page for the subpage read */
1567static int gpmi_ecc_read_subpage(struct nand_chip *chip, uint32_t offs,
1568				 uint32_t len, uint8_t *buf, int page)
1569{
1570	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1571	struct bch_geometry *geo = &this->bch_geometry;
1572	int size = chip->ecc.size; /* ECC chunk size */
1573	int meta, n, page_size;
1574	unsigned int max_bitflips;
1575	unsigned int ecc_strength;
1576	int first, last, marker_pos;
1577	int ecc_parity_size;
1578	int col = 0;
1579	int ret;
1580
1581	/* The size of ECC parity */
1582	ecc_parity_size = geo->gf_len * geo->ecc_strength / 8;
1583
1584	/* Align it with the chunk size */
1585	first = offs / size;
1586	last = (offs + len - 1) / size;
1587
1588	if (this->swap_block_mark) {
1589		/*
1590		 * Find the chunk which contains the Block Marker.
1591		 * If this chunk is in the range of [first, last],
1592		 * we have to read out the whole page.
1593		 * Why? since we had swapped the data at the position of Block
1594		 * Marker to the metadata which is bound with the chunk 0.
1595		 */
1596		marker_pos = geo->block_mark_byte_offset / size;
1597		if (last >= marker_pos && first <= marker_pos) {
1598			dev_dbg(this->dev,
1599				"page:%d, first:%d, last:%d, marker at:%d\n",
1600				page, first, last, marker_pos);
1601			return gpmi_ecc_read_page(chip, buf, 0, page);
1602		}
1603	}
1604
1605	/*
1606	 * if there is an ECC dedicate for meta:
1607	 * - need to add an extra ECC size when calculating col and page_size,
1608	 *   if the meta size is NOT zero.
1609	 * - ecc0_chunk size need to set to the same size as other chunks,
1610	 *   if the meta size is zero.
1611	 */
1612
1613	meta = geo->metadata_size;
1614	if (first) {
1615		if (geo->ecc_for_meta)
1616			col = meta + ecc_parity_size
1617				+ (size + ecc_parity_size) * first;
1618		else
1619			col = meta + (size + ecc_parity_size) * first;
1620
1621		meta = 0;
1622		buf = buf + first * size;
1623	}
1624
1625	ecc_parity_size = geo->gf_len * geo->ecc_strength / 8;
1626	n = last - first + 1;
1627
1628	if (geo->ecc_for_meta && meta)
1629		page_size = meta + ecc_parity_size
1630			    + (size + ecc_parity_size) * n;
1631	else
1632		page_size = meta + (size + ecc_parity_size) * n;
1633
1634	ecc_strength = geo->ecc_strength >> 1;
1635
1636	this->bch_flashlayout0 = BF_BCH_FLASH0LAYOUT0_NBLOCKS(
1637		(geo->ecc_for_meta ? n : n - 1)) |
1638		BF_BCH_FLASH0LAYOUT0_META_SIZE(meta) |
1639		BF_BCH_FLASH0LAYOUT0_ECC0(ecc_strength, this) |
1640		BF_BCH_FLASH0LAYOUT0_GF(geo->gf_len, this) |
1641		BF_BCH_FLASH0LAYOUT0_DATA0_SIZE((geo->ecc_for_meta ?
1642		0 : geo->ecc0_chunk_size), this);
1643
1644	this->bch_flashlayout1 = BF_BCH_FLASH0LAYOUT1_PAGE_SIZE(page_size) |
1645		BF_BCH_FLASH0LAYOUT1_ECCN(ecc_strength, this) |
1646		BF_BCH_FLASH0LAYOUT1_GF(geo->gf_len, this) |
1647		BF_BCH_FLASH0LAYOUT1_DATAN_SIZE(geo->eccn_chunk_size, this);
1648
1649	this->bch = true;
1650
1651	ret = nand_read_page_op(chip, page, col, buf, page_size);
1652	if (ret)
1653		return ret;
1654
1655	dev_dbg(this->dev, "page:%d(%d:%d)%d, chunk:(%d:%d), BCH PG size:%d\n",
1656		page, offs, len, col, first, n, page_size);
1657
1658	max_bitflips = gpmi_count_bitflips(chip, buf, first, last, meta);
1659
1660	return max_bitflips;
1661}
1662
1663static int gpmi_ecc_write_page(struct nand_chip *chip, const uint8_t *buf,
1664			       int oob_required, int page)
1665{
1666	struct mtd_info *mtd = nand_to_mtd(chip);
1667	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1668	struct bch_geometry *nfc_geo = &this->bch_geometry;
1669
1670	dev_dbg(this->dev, "ecc write page.\n");
1671
1672	gpmi_bch_layout_std(this);
1673	this->bch = true;
1674
1675	memcpy(this->auxiliary_virt, chip->oob_poi, nfc_geo->auxiliary_size);
1676
1677	if (this->swap_block_mark) {
1678		/*
1679		 * When doing bad block marker swapping we must always copy the
1680		 * input buffer as we can't modify the const buffer.
1681		 */
1682		memcpy(this->data_buffer_dma, buf, mtd->writesize);
1683		buf = this->data_buffer_dma;
1684		block_mark_swapping(this, this->data_buffer_dma,
1685				    this->auxiliary_virt);
1686	}
1687
1688	return nand_prog_page_op(chip, page, 0, buf, nfc_geo->page_size);
1689}
1690
1691/*
1692 * There are several places in this driver where we have to handle the OOB and
1693 * block marks. This is the function where things are the most complicated, so
1694 * this is where we try to explain it all. All the other places refer back to
1695 * here.
1696 *
1697 * These are the rules, in order of decreasing importance:
1698 *
1699 * 1) Nothing the caller does can be allowed to imperil the block mark.
1700 *
1701 * 2) In read operations, the first byte of the OOB we return must reflect the
1702 *    true state of the block mark, no matter where that block mark appears in
1703 *    the physical page.
1704 *
1705 * 3) ECC-based read operations return an OOB full of set bits (since we never
1706 *    allow ECC-based writes to the OOB, it doesn't matter what ECC-based reads
1707 *    return).
1708 *
1709 * 4) "Raw" read operations return a direct view of the physical bytes in the
1710 *    page, using the conventional definition of which bytes are data and which
1711 *    are OOB. This gives the caller a way to see the actual, physical bytes
1712 *    in the page, without the distortions applied by our ECC engine.
1713 *
1714 *
1715 * What we do for this specific read operation depends on two questions:
1716 *
1717 * 1) Are we doing a "raw" read, or an ECC-based read?
1718 *
1719 * 2) Are we using block mark swapping or transcription?
1720 *
1721 * There are four cases, illustrated by the following Karnaugh map:
1722 *
1723 *                    |           Raw           |         ECC-based       |
1724 *       -------------+-------------------------+-------------------------+
1725 *                    | Read the conventional   |                         |
1726 *                    | OOB at the end of the   |                         |
1727 *       Swapping     | page and return it. It  |                         |
1728 *                    | contains exactly what   |                         |
1729 *                    | we want.                | Read the block mark and |
1730 *       -------------+-------------------------+ return it in a buffer   |
1731 *                    | Read the conventional   | full of set bits.       |
1732 *                    | OOB at the end of the   |                         |
1733 *                    | page and also the block |                         |
1734 *       Transcribing | mark in the metadata.   |                         |
1735 *                    | Copy the block mark     |                         |
1736 *                    | into the first byte of  |                         |
1737 *                    | the OOB.                |                         |
1738 *       -------------+-------------------------+-------------------------+
1739 *
1740 * Note that we break rule #4 in the Transcribing/Raw case because we're not
1741 * giving an accurate view of the actual, physical bytes in the page (we're
1742 * overwriting the block mark). That's OK because it's more important to follow
1743 * rule #2.
1744 *
1745 * It turns out that knowing whether we want an "ECC-based" or "raw" read is not
1746 * easy. When reading a page, for example, the NAND Flash MTD code calls our
1747 * ecc.read_page or ecc.read_page_raw function. Thus, the fact that MTD wants an
1748 * ECC-based or raw view of the page is implicit in which function it calls
1749 * (there is a similar pair of ECC-based/raw functions for writing).
1750 */
1751static int gpmi_ecc_read_oob(struct nand_chip *chip, int page)
1752{
1753	struct mtd_info *mtd = nand_to_mtd(chip);
1754	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1755	int ret;
1756
1757	/* clear the OOB buffer */
1758	memset(chip->oob_poi, ~0, mtd->oobsize);
1759
1760	/* Read out the conventional OOB. */
1761	ret = nand_read_page_op(chip, page, mtd->writesize, chip->oob_poi,
1762				mtd->oobsize);
1763	if (ret)
1764		return ret;
1765
1766	/*
1767	 * Now, we want to make sure the block mark is correct. In the
1768	 * non-transcribing case (!GPMI_IS_MX23()), we already have it.
1769	 * Otherwise, we need to explicitly read it.
1770	 */
1771	if (GPMI_IS_MX23(this)) {
1772		/* Read the block mark into the first byte of the OOB buffer. */
1773		ret = nand_read_page_op(chip, page, 0, chip->oob_poi, 1);
1774		if (ret)
1775			return ret;
1776	}
1777
1778	return 0;
1779}
1780
1781static int gpmi_ecc_write_oob(struct nand_chip *chip, int page)
1782{
1783	struct mtd_info *mtd = nand_to_mtd(chip);
1784	struct mtd_oob_region of = { };
1785
1786	/* Do we have available oob area? */
1787	mtd_ooblayout_free(mtd, 0, &of);
1788	if (!of.length)
1789		return -EPERM;
1790
1791	if (!nand_is_slc(chip))
1792		return -EPERM;
1793
1794	return nand_prog_page_op(chip, page, mtd->writesize + of.offset,
1795				 chip->oob_poi + of.offset, of.length);
1796}
1797
1798/*
1799 * This function reads a NAND page without involving the ECC engine (no HW
1800 * ECC correction).
1801 * The tricky part in the GPMI/BCH controller is that it stores ECC bits
1802 * inline (interleaved with payload DATA), and do not align data chunk on
1803 * byte boundaries.
1804 * We thus need to take care moving the payload data and ECC bits stored in the
1805 * page into the provided buffers, which is why we're using nand_extract_bits().
1806 *
1807 * See set_geometry_by_ecc_info inline comments to have a full description
1808 * of the layout used by the GPMI controller.
1809 */
1810static int gpmi_ecc_read_page_raw(struct nand_chip *chip, uint8_t *buf,
1811				  int oob_required, int page)
1812{
1813	struct mtd_info *mtd = nand_to_mtd(chip);
1814	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1815	struct bch_geometry *nfc_geo = &this->bch_geometry;
1816	int eccsize = nfc_geo->eccn_chunk_size;
1817	int eccbits = nfc_geo->ecc_strength * nfc_geo->gf_len;
1818	u8 *tmp_buf = this->raw_buffer;
1819	size_t src_bit_off;
1820	size_t oob_bit_off;
1821	size_t oob_byte_off;
1822	uint8_t *oob = chip->oob_poi;
1823	int step;
1824	int ret;
1825
1826	ret = nand_read_page_op(chip, page, 0, tmp_buf,
1827				mtd->writesize + mtd->oobsize);
1828	if (ret)
1829		return ret;
1830
1831	/*
1832	 * If required, swap the bad block marker and the data stored in the
1833	 * metadata section, so that we don't wrongly consider a block as bad.
1834	 *
1835	 * See the layout description for a detailed explanation on why this
1836	 * is needed.
1837	 */
1838	if (this->swap_block_mark)
1839		swap(tmp_buf[0], tmp_buf[mtd->writesize]);
1840
1841	/*
1842	 * Copy the metadata section into the oob buffer (this section is
1843	 * guaranteed to be aligned on a byte boundary).
1844	 */
1845	if (oob_required)
1846		memcpy(oob, tmp_buf, nfc_geo->metadata_size);
1847
1848	oob_bit_off = nfc_geo->metadata_size * 8;
1849	src_bit_off = oob_bit_off;
1850
1851	/* Extract interleaved payload data and ECC bits */
1852	for (step = 0; step < nfc_geo->ecc_chunk_count; step++) {
1853		if (buf)
1854			nand_extract_bits(buf, step * eccsize * 8, tmp_buf,
1855					  src_bit_off, eccsize * 8);
1856		src_bit_off += eccsize * 8;
1857
1858		/* Align last ECC block to align a byte boundary */
1859		if (step == nfc_geo->ecc_chunk_count - 1 &&
1860		    (oob_bit_off + eccbits) % 8)
1861			eccbits += 8 - ((oob_bit_off + eccbits) % 8);
1862
1863		if (oob_required)
1864			nand_extract_bits(oob, oob_bit_off, tmp_buf,
1865					  src_bit_off, eccbits);
1866
1867		src_bit_off += eccbits;
1868		oob_bit_off += eccbits;
1869	}
1870
1871	if (oob_required) {
1872		oob_byte_off = oob_bit_off / 8;
1873
1874		if (oob_byte_off < mtd->oobsize)
1875			memcpy(oob + oob_byte_off,
1876			       tmp_buf + mtd->writesize + oob_byte_off,
1877			       mtd->oobsize - oob_byte_off);
1878	}
1879
1880	return 0;
1881}
1882
1883/*
1884 * This function writes a NAND page without involving the ECC engine (no HW
1885 * ECC generation).
1886 * The tricky part in the GPMI/BCH controller is that it stores ECC bits
1887 * inline (interleaved with payload DATA), and do not align data chunk on
1888 * byte boundaries.
1889 * We thus need to take care moving the OOB area at the right place in the
1890 * final page, which is why we're using nand_extract_bits().
1891 *
1892 * See set_geometry_by_ecc_info inline comments to have a full description
1893 * of the layout used by the GPMI controller.
1894 */
1895static int gpmi_ecc_write_page_raw(struct nand_chip *chip, const uint8_t *buf,
1896				   int oob_required, int page)
1897{
1898	struct mtd_info *mtd = nand_to_mtd(chip);
1899	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1900	struct bch_geometry *nfc_geo = &this->bch_geometry;
1901	int eccsize = nfc_geo->eccn_chunk_size;
1902	int eccbits = nfc_geo->ecc_strength * nfc_geo->gf_len;
1903	u8 *tmp_buf = this->raw_buffer;
1904	uint8_t *oob = chip->oob_poi;
1905	size_t dst_bit_off;
1906	size_t oob_bit_off;
1907	size_t oob_byte_off;
1908	int step;
1909
1910	/*
1911	 * Initialize all bits to 1 in case we don't have a buffer for the
1912	 * payload or oob data in order to leave unspecified bits of data
1913	 * to their initial state.
1914	 */
1915	if (!buf || !oob_required)
1916		memset(tmp_buf, 0xff, mtd->writesize + mtd->oobsize);
1917
1918	/*
1919	 * First copy the metadata section (stored in oob buffer) at the
1920	 * beginning of the page, as imposed by the GPMI layout.
1921	 */
1922	memcpy(tmp_buf, oob, nfc_geo->metadata_size);
1923	oob_bit_off = nfc_geo->metadata_size * 8;
1924	dst_bit_off = oob_bit_off;
1925
1926	/* Interleave payload data and ECC bits */
1927	for (step = 0; step < nfc_geo->ecc_chunk_count; step++) {
1928		if (buf)
1929			nand_extract_bits(tmp_buf, dst_bit_off, buf,
1930					  step * eccsize * 8, eccsize * 8);
1931		dst_bit_off += eccsize * 8;
1932
1933		/* Align last ECC block to align a byte boundary */
1934		if (step == nfc_geo->ecc_chunk_count - 1 &&
1935		    (oob_bit_off + eccbits) % 8)
1936			eccbits += 8 - ((oob_bit_off + eccbits) % 8);
1937
1938		if (oob_required)
1939			nand_extract_bits(tmp_buf, dst_bit_off, oob,
1940					  oob_bit_off, eccbits);
1941
1942		dst_bit_off += eccbits;
1943		oob_bit_off += eccbits;
1944	}
1945
1946	oob_byte_off = oob_bit_off / 8;
1947
1948	if (oob_required && oob_byte_off < mtd->oobsize)
1949		memcpy(tmp_buf + mtd->writesize + oob_byte_off,
1950		       oob + oob_byte_off, mtd->oobsize - oob_byte_off);
1951
1952	/*
1953	 * If required, swap the bad block marker and the first byte of the
1954	 * metadata section, so that we don't modify the bad block marker.
1955	 *
1956	 * See the layout description for a detailed explanation on why this
1957	 * is needed.
1958	 */
1959	if (this->swap_block_mark)
1960		swap(tmp_buf[0], tmp_buf[mtd->writesize]);
1961
1962	return nand_prog_page_op(chip, page, 0, tmp_buf,
1963				 mtd->writesize + mtd->oobsize);
1964}
1965
1966static int gpmi_ecc_read_oob_raw(struct nand_chip *chip, int page)
1967{
1968	return gpmi_ecc_read_page_raw(chip, NULL, 1, page);
1969}
1970
1971static int gpmi_ecc_write_oob_raw(struct nand_chip *chip, int page)
1972{
1973	return gpmi_ecc_write_page_raw(chip, NULL, 1, page);
1974}
1975
1976static int gpmi_block_markbad(struct nand_chip *chip, loff_t ofs)
1977{
1978	struct mtd_info *mtd = nand_to_mtd(chip);
1979	struct gpmi_nand_data *this = nand_get_controller_data(chip);
1980	int ret = 0;
1981	uint8_t *block_mark;
1982	int column, page, chipnr;
1983
1984	chipnr = (int)(ofs >> chip->chip_shift);
1985	nand_select_target(chip, chipnr);
1986
1987	column = !GPMI_IS_MX23(this) ? mtd->writesize : 0;
1988
1989	/* Write the block mark. */
1990	block_mark = this->data_buffer_dma;
1991	block_mark[0] = 0; /* bad block marker */
1992
1993	/* Shift to get page */
1994	page = (int)(ofs >> chip->page_shift);
1995
1996	ret = nand_prog_page_op(chip, page, column, block_mark, 1);
1997
1998	nand_deselect_target(chip);
1999
2000	return ret;
2001}
2002
2003static int nand_boot_set_geometry(struct gpmi_nand_data *this)
2004{
2005	struct boot_rom_geometry *geometry = &this->rom_geometry;
2006
2007	/*
2008	 * Set the boot block stride size.
2009	 *
2010	 * In principle, we should be reading this from the OTP bits, since
2011	 * that's where the ROM is going to get it. In fact, we don't have any
2012	 * way to read the OTP bits, so we go with the default and hope for the
2013	 * best.
2014	 */
2015	geometry->stride_size_in_pages = 64;
2016
2017	/*
2018	 * Set the search area stride exponent.
2019	 *
2020	 * In principle, we should be reading this from the OTP bits, since
2021	 * that's where the ROM is going to get it. In fact, we don't have any
2022	 * way to read the OTP bits, so we go with the default and hope for the
2023	 * best.
2024	 */
2025	geometry->search_area_stride_exponent = 2;
2026	return 0;
2027}
2028
2029static const char  *fingerprint = "STMP";
2030static int mx23_check_transcription_stamp(struct gpmi_nand_data *this)
2031{
2032	struct boot_rom_geometry *rom_geo = &this->rom_geometry;
2033	struct device *dev = this->dev;
2034	struct nand_chip *chip = &this->nand;
2035	unsigned int search_area_size_in_strides;
2036	unsigned int stride;
2037	unsigned int page;
2038	u8 *buffer = nand_get_data_buf(chip);
2039	int found_an_ncb_fingerprint = false;
2040	int ret;
2041
2042	/* Compute the number of strides in a search area. */
2043	search_area_size_in_strides = 1 << rom_geo->search_area_stride_exponent;
2044
2045	nand_select_target(chip, 0);
2046
2047	/*
2048	 * Loop through the first search area, looking for the NCB fingerprint.
2049	 */
2050	dev_dbg(dev, "Scanning for an NCB fingerprint...\n");
2051
2052	for (stride = 0; stride < search_area_size_in_strides; stride++) {
2053		/* Compute the page addresses. */
2054		page = stride * rom_geo->stride_size_in_pages;
2055
2056		dev_dbg(dev, "Looking for a fingerprint in page 0x%x\n", page);
2057
2058		/*
2059		 * Read the NCB fingerprint. The fingerprint is four bytes long
2060		 * and starts in the 12th byte of the page.
2061		 */
2062		ret = nand_read_page_op(chip, page, 12, buffer,
2063					strlen(fingerprint));
2064		if (ret)
2065			continue;
2066
2067		/* Look for the fingerprint. */
2068		if (!memcmp(buffer, fingerprint, strlen(fingerprint))) {
2069			found_an_ncb_fingerprint = true;
2070			break;
2071		}
2072
2073	}
2074
2075	nand_deselect_target(chip);
2076
2077	if (found_an_ncb_fingerprint)
2078		dev_dbg(dev, "\tFound a fingerprint\n");
2079	else
2080		dev_dbg(dev, "\tNo fingerprint found\n");
2081	return found_an_ncb_fingerprint;
2082}
2083
2084/* Writes a transcription stamp. */
2085static int mx23_write_transcription_stamp(struct gpmi_nand_data *this)
2086{
2087	struct device *dev = this->dev;
2088	struct boot_rom_geometry *rom_geo = &this->rom_geometry;
2089	struct nand_chip *chip = &this->nand;
2090	struct mtd_info *mtd = nand_to_mtd(chip);
2091	unsigned int block_size_in_pages;
2092	unsigned int search_area_size_in_strides;
2093	unsigned int search_area_size_in_pages;
2094	unsigned int search_area_size_in_blocks;
2095	unsigned int block;
2096	unsigned int stride;
2097	unsigned int page;
2098	u8 *buffer = nand_get_data_buf(chip);
2099	int status;
2100
2101	/* Compute the search area geometry. */
2102	block_size_in_pages = mtd->erasesize / mtd->writesize;
2103	search_area_size_in_strides = 1 << rom_geo->search_area_stride_exponent;
2104	search_area_size_in_pages = search_area_size_in_strides *
2105					rom_geo->stride_size_in_pages;
2106	search_area_size_in_blocks =
2107		  (search_area_size_in_pages + (block_size_in_pages - 1)) /
2108				    block_size_in_pages;
2109
2110	dev_dbg(dev, "Search Area Geometry :\n");
2111	dev_dbg(dev, "\tin Blocks : %u\n", search_area_size_in_blocks);
2112	dev_dbg(dev, "\tin Strides: %u\n", search_area_size_in_strides);
2113	dev_dbg(dev, "\tin Pages  : %u\n", search_area_size_in_pages);
2114
2115	nand_select_target(chip, 0);
2116
2117	/* Loop over blocks in the first search area, erasing them. */
2118	dev_dbg(dev, "Erasing the search area...\n");
2119
2120	for (block = 0; block < search_area_size_in_blocks; block++) {
2121		/* Erase this block. */
2122		dev_dbg(dev, "\tErasing block 0x%x\n", block);
2123		status = nand_erase_op(chip, block);
2124		if (status)
2125			dev_err(dev, "[%s] Erase failed.\n", __func__);
2126	}
2127
2128	/* Write the NCB fingerprint into the page buffer. */
2129	memset(buffer, ~0, mtd->writesize);
2130	memcpy(buffer + 12, fingerprint, strlen(fingerprint));
2131
2132	/* Loop through the first search area, writing NCB fingerprints. */
2133	dev_dbg(dev, "Writing NCB fingerprints...\n");
2134	for (stride = 0; stride < search_area_size_in_strides; stride++) {
2135		/* Compute the page addresses. */
2136		page = stride * rom_geo->stride_size_in_pages;
2137
2138		/* Write the first page of the current stride. */
2139		dev_dbg(dev, "Writing an NCB fingerprint in page 0x%x\n", page);
2140
2141		status = chip->ecc.write_page_raw(chip, buffer, 0, page);
2142		if (status)
2143			dev_err(dev, "[%s] Write failed.\n", __func__);
2144	}
2145
2146	nand_deselect_target(chip);
2147
2148	return 0;
2149}
2150
2151static int mx23_boot_init(struct gpmi_nand_data  *this)
2152{
2153	struct device *dev = this->dev;
2154	struct nand_chip *chip = &this->nand;
2155	struct mtd_info *mtd = nand_to_mtd(chip);
2156	unsigned int block_count;
2157	unsigned int block;
2158	int     chipnr;
2159	int     page;
2160	loff_t  byte;
2161	uint8_t block_mark;
2162	int     ret = 0;
2163
2164	/*
2165	 * If control arrives here, we can't use block mark swapping, which
2166	 * means we're forced to use transcription. First, scan for the
2167	 * transcription stamp. If we find it, then we don't have to do
2168	 * anything -- the block marks are already transcribed.
2169	 */
2170	if (mx23_check_transcription_stamp(this))
2171		return 0;
2172
2173	/*
2174	 * If control arrives here, we couldn't find a transcription stamp, so
2175	 * so we presume the block marks are in the conventional location.
2176	 */
2177	dev_dbg(dev, "Transcribing bad block marks...\n");
2178
2179	/* Compute the number of blocks in the entire medium. */
2180	block_count = nanddev_eraseblocks_per_target(&chip->base);
2181
2182	/*
2183	 * Loop over all the blocks in the medium, transcribing block marks as
2184	 * we go.
2185	 */
2186	for (block = 0; block < block_count; block++) {
2187		/*
2188		 * Compute the chip, page and byte addresses for this block's
2189		 * conventional mark.
2190		 */
2191		chipnr = block >> (chip->chip_shift - chip->phys_erase_shift);
2192		page = block << (chip->phys_erase_shift - chip->page_shift);
2193		byte = block <<  chip->phys_erase_shift;
2194
2195		/* Send the command to read the conventional block mark. */
2196		nand_select_target(chip, chipnr);
2197		ret = nand_read_page_op(chip, page, mtd->writesize, &block_mark,
2198					1);
2199		nand_deselect_target(chip);
2200
2201		if (ret)
2202			continue;
2203
2204		/*
2205		 * Check if the block is marked bad. If so, we need to mark it
2206		 * again, but this time the result will be a mark in the
2207		 * location where we transcribe block marks.
2208		 */
2209		if (block_mark != 0xff) {
2210			dev_dbg(dev, "Transcribing mark in block %u\n", block);
2211			ret = chip->legacy.block_markbad(chip, byte);
2212			if (ret)
2213				dev_err(dev,
2214					"Failed to mark block bad with ret %d\n",
2215					ret);
2216		}
2217	}
2218
2219	/* Write the stamp that indicates we've transcribed the block marks. */
2220	mx23_write_transcription_stamp(this);
2221	return 0;
2222}
2223
2224static int nand_boot_init(struct gpmi_nand_data  *this)
2225{
2226	nand_boot_set_geometry(this);
2227
2228	/* This is ROM arch-specific initilization before the BBT scanning. */
2229	if (GPMI_IS_MX23(this))
2230		return mx23_boot_init(this);
2231	return 0;
2232}
2233
2234static int gpmi_set_geometry(struct gpmi_nand_data *this)
2235{
2236	int ret;
2237
2238	/* Free the temporary DMA memory for reading ID. */
2239	gpmi_free_dma_buffer(this);
2240
2241	/* Set up the NFC geometry which is used by BCH. */
2242	ret = bch_set_geometry(this);
2243	if (ret) {
2244		dev_err(this->dev, "Error setting BCH geometry : %d\n", ret);
2245		return ret;
2246	}
2247
2248	/* Alloc the new DMA buffers according to the pagesize and oobsize */
2249	return gpmi_alloc_dma_buffer(this);
2250}
2251
2252static int gpmi_init_last(struct gpmi_nand_data *this)
2253{
2254	struct nand_chip *chip = &this->nand;
2255	struct mtd_info *mtd = nand_to_mtd(chip);
2256	struct nand_ecc_ctrl *ecc = &chip->ecc;
2257	struct bch_geometry *bch_geo = &this->bch_geometry;
2258	int ret;
2259
2260	/* Set up the medium geometry */
2261	ret = gpmi_set_geometry(this);
2262	if (ret)
2263		return ret;
2264
2265	/* Init the nand_ecc_ctrl{} */
2266	ecc->read_page	= gpmi_ecc_read_page;
2267	ecc->write_page	= gpmi_ecc_write_page;
2268	ecc->read_oob	= gpmi_ecc_read_oob;
2269	ecc->write_oob	= gpmi_ecc_write_oob;
2270	ecc->read_page_raw = gpmi_ecc_read_page_raw;
2271	ecc->write_page_raw = gpmi_ecc_write_page_raw;
2272	ecc->read_oob_raw = gpmi_ecc_read_oob_raw;
2273	ecc->write_oob_raw = gpmi_ecc_write_oob_raw;
2274	ecc->engine_type = NAND_ECC_ENGINE_TYPE_ON_HOST;
2275	ecc->size	= bch_geo->eccn_chunk_size;
2276	ecc->strength	= bch_geo->ecc_strength;
2277	mtd_set_ooblayout(mtd, &gpmi_ooblayout_ops);
2278
2279	/*
2280	 * We only enable the subpage read when:
2281	 *  (1) the chip is imx6, and
2282	 *  (2) the size of the ECC parity is byte aligned.
2283	 */
2284	if (GPMI_IS_MX6(this) &&
2285		((bch_geo->gf_len * bch_geo->ecc_strength) % 8) == 0) {
2286		ecc->read_subpage = gpmi_ecc_read_subpage;
2287		chip->options |= NAND_SUBPAGE_READ;
2288	}
2289
2290	return 0;
2291}
2292
2293static int gpmi_nand_attach_chip(struct nand_chip *chip)
2294{
2295	struct gpmi_nand_data *this = nand_get_controller_data(chip);
2296	int ret;
2297
2298	if (chip->bbt_options & NAND_BBT_USE_FLASH) {
2299		chip->bbt_options |= NAND_BBT_NO_OOB;
2300
2301		if (of_property_read_bool(this->dev->of_node,
2302					  "fsl,no-blockmark-swap"))
2303			this->swap_block_mark = false;
2304	}
2305	dev_dbg(this->dev, "Blockmark swapping %sabled\n",
2306		this->swap_block_mark ? "en" : "dis");
2307
2308	ret = gpmi_init_last(this);
2309	if (ret)
2310		return ret;
2311
2312	chip->options |= NAND_SKIP_BBTSCAN;
2313
2314	return 0;
2315}
2316
2317static struct gpmi_transfer *get_next_transfer(struct gpmi_nand_data *this)
2318{
2319	struct gpmi_transfer *transfer = &this->transfers[this->ntransfers];
2320
2321	this->ntransfers++;
2322
2323	if (this->ntransfers == GPMI_MAX_TRANSFERS)
2324		return NULL;
2325
2326	return transfer;
2327}
2328
2329static struct dma_async_tx_descriptor *gpmi_chain_command(
2330	struct gpmi_nand_data *this, u8 cmd, const u8 *addr, int naddr)
2331{
2332	struct dma_chan *channel = get_dma_chan(this);
2333	struct dma_async_tx_descriptor *desc;
2334	struct gpmi_transfer *transfer;
2335	int chip = this->nand.cur_cs;
2336	u32 pio[3];
2337
2338	/* [1] send out the PIO words */
2339	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(BV_GPMI_CTRL0_COMMAND_MODE__WRITE)
2340		| BM_GPMI_CTRL0_WORD_LENGTH
2341		| BF_GPMI_CTRL0_CS(chip, this)
2342		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
2343		| BF_GPMI_CTRL0_ADDRESS(BV_GPMI_CTRL0_ADDRESS__NAND_CLE)
2344		| BM_GPMI_CTRL0_ADDRESS_INCREMENT
2345		| BF_GPMI_CTRL0_XFER_COUNT(naddr + 1);
2346	pio[1] = 0;
2347	pio[2] = 0;
2348	desc = mxs_dmaengine_prep_pio(channel, pio, ARRAY_SIZE(pio),
2349				      DMA_TRANS_NONE, 0);
2350	if (!desc)
2351		return NULL;
2352
2353	transfer = get_next_transfer(this);
2354	if (!transfer)
2355		return NULL;
2356
2357	transfer->cmdbuf[0] = cmd;
2358	if (naddr)
2359		memcpy(&transfer->cmdbuf[1], addr, naddr);
2360
2361	sg_init_one(&transfer->sgl, transfer->cmdbuf, naddr + 1);
2362	dma_map_sg(this->dev, &transfer->sgl, 1, DMA_TO_DEVICE);
2363
2364	transfer->direction = DMA_TO_DEVICE;
2365
2366	desc = dmaengine_prep_slave_sg(channel, &transfer->sgl, 1, DMA_MEM_TO_DEV,
2367				       MXS_DMA_CTRL_WAIT4END);
2368	return desc;
2369}
2370
2371static struct dma_async_tx_descriptor *gpmi_chain_wait_ready(
2372	struct gpmi_nand_data *this)
2373{
2374	struct dma_chan *channel = get_dma_chan(this);
2375	u32 pio[2];
2376
2377	pio[0] =  BF_GPMI_CTRL0_COMMAND_MODE(BV_GPMI_CTRL0_COMMAND_MODE__WAIT_FOR_READY)
2378		| BM_GPMI_CTRL0_WORD_LENGTH
2379		| BF_GPMI_CTRL0_CS(this->nand.cur_cs, this)
2380		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
2381		| BF_GPMI_CTRL0_ADDRESS(BV_GPMI_CTRL0_ADDRESS__NAND_DATA)
2382		| BF_GPMI_CTRL0_XFER_COUNT(0);
2383	pio[1] = 0;
2384
2385	return mxs_dmaengine_prep_pio(channel, pio, 2, DMA_TRANS_NONE,
2386				MXS_DMA_CTRL_WAIT4END | MXS_DMA_CTRL_WAIT4RDY);
2387}
2388
2389static struct dma_async_tx_descriptor *gpmi_chain_data_read(
2390	struct gpmi_nand_data *this, void *buf, int raw_len, bool *direct)
2391{
2392	struct dma_async_tx_descriptor *desc;
2393	struct dma_chan *channel = get_dma_chan(this);
2394	struct gpmi_transfer *transfer;
2395	u32 pio[6] = {};
2396
2397	transfer = get_next_transfer(this);
2398	if (!transfer)
2399		return NULL;
2400
2401	transfer->direction = DMA_FROM_DEVICE;
2402
2403	*direct = prepare_data_dma(this, buf, raw_len, &transfer->sgl,
2404				   DMA_FROM_DEVICE);
2405
2406	pio[0] =  BF_GPMI_CTRL0_COMMAND_MODE(BV_GPMI_CTRL0_COMMAND_MODE__READ)
2407		| BM_GPMI_CTRL0_WORD_LENGTH
2408		| BF_GPMI_CTRL0_CS(this->nand.cur_cs, this)
2409		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
2410		| BF_GPMI_CTRL0_ADDRESS(BV_GPMI_CTRL0_ADDRESS__NAND_DATA)
2411		| BF_GPMI_CTRL0_XFER_COUNT(raw_len);
2412
2413	if (this->bch) {
2414		pio[2] =  BM_GPMI_ECCCTRL_ENABLE_ECC
2415			| BF_GPMI_ECCCTRL_ECC_CMD(BV_GPMI_ECCCTRL_ECC_CMD__BCH_DECODE)
2416			| BF_GPMI_ECCCTRL_BUFFER_MASK(BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_PAGE
2417				| BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_AUXONLY);
2418		pio[3] = raw_len;
2419		pio[4] = transfer->sgl.dma_address;
2420		pio[5] = this->auxiliary_phys;
2421	}
2422
2423	desc = mxs_dmaengine_prep_pio(channel, pio, ARRAY_SIZE(pio),
2424				      DMA_TRANS_NONE, 0);
2425	if (!desc)
2426		return NULL;
2427
2428	if (!this->bch)
2429		desc = dmaengine_prep_slave_sg(channel, &transfer->sgl, 1,
2430					     DMA_DEV_TO_MEM,
2431					     MXS_DMA_CTRL_WAIT4END);
2432
2433	return desc;
2434}
2435
2436static struct dma_async_tx_descriptor *gpmi_chain_data_write(
2437	struct gpmi_nand_data *this, const void *buf, int raw_len)
2438{
2439	struct dma_chan *channel = get_dma_chan(this);
2440	struct dma_async_tx_descriptor *desc;
2441	struct gpmi_transfer *transfer;
2442	u32 pio[6] = {};
2443
2444	transfer = get_next_transfer(this);
2445	if (!transfer)
2446		return NULL;
2447
2448	transfer->direction = DMA_TO_DEVICE;
2449
2450	prepare_data_dma(this, buf, raw_len, &transfer->sgl, DMA_TO_DEVICE);
2451
2452	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(BV_GPMI_CTRL0_COMMAND_MODE__WRITE)
2453		| BM_GPMI_CTRL0_WORD_LENGTH
2454		| BF_GPMI_CTRL0_CS(this->nand.cur_cs, this)
2455		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
2456		| BF_GPMI_CTRL0_ADDRESS(BV_GPMI_CTRL0_ADDRESS__NAND_DATA)
2457		| BF_GPMI_CTRL0_XFER_COUNT(raw_len);
2458
2459	if (this->bch) {
2460		pio[2] = BM_GPMI_ECCCTRL_ENABLE_ECC
2461			| BF_GPMI_ECCCTRL_ECC_CMD(BV_GPMI_ECCCTRL_ECC_CMD__BCH_ENCODE)
2462			| BF_GPMI_ECCCTRL_BUFFER_MASK(BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_PAGE |
2463					BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_AUXONLY);
2464		pio[3] = raw_len;
2465		pio[4] = transfer->sgl.dma_address;
2466		pio[5] = this->auxiliary_phys;
2467	}
2468
2469	desc = mxs_dmaengine_prep_pio(channel, pio, ARRAY_SIZE(pio),
2470				      DMA_TRANS_NONE,
2471				      (this->bch ? MXS_DMA_CTRL_WAIT4END : 0));
2472	if (!desc)
2473		return NULL;
2474
2475	if (!this->bch)
2476		desc = dmaengine_prep_slave_sg(channel, &transfer->sgl, 1,
2477					       DMA_MEM_TO_DEV,
2478					       MXS_DMA_CTRL_WAIT4END);
2479
2480	return desc;
2481}
2482
2483static int gpmi_nfc_exec_op(struct nand_chip *chip,
2484			     const struct nand_operation *op,
2485			     bool check_only)
2486{
2487	const struct nand_op_instr *instr;
2488	struct gpmi_nand_data *this = nand_get_controller_data(chip);
2489	struct dma_async_tx_descriptor *desc = NULL;
2490	int i, ret, buf_len = 0, nbufs = 0;
2491	u8 cmd = 0;
2492	void *buf_read = NULL;
2493	const void *buf_write = NULL;
2494	bool direct = false;
2495	struct completion *dma_completion, *bch_completion;
2496	unsigned long to;
2497
2498	if (check_only)
2499		return 0;
2500
2501	this->ntransfers = 0;
2502	for (i = 0; i < GPMI_MAX_TRANSFERS; i++)
2503		this->transfers[i].direction = DMA_NONE;
2504
2505	ret = pm_runtime_resume_and_get(this->dev);
2506	if (ret < 0)
2507		return ret;
2508
2509	/*
2510	 * This driver currently supports only one NAND chip. Plus, dies share
2511	 * the same configuration. So once timings have been applied on the
2512	 * controller side, they will not change anymore. When the time will
2513	 * come, the check on must_apply_timings will have to be dropped.
2514	 */
2515	if (this->hw.must_apply_timings) {
2516		this->hw.must_apply_timings = false;
2517		ret = gpmi_nfc_apply_timings(this);
2518		if (ret)
2519			goto out_pm;
2520	}
2521
2522	dev_dbg(this->dev, "%s: %d instructions\n", __func__, op->ninstrs);
2523
2524	for (i = 0; i < op->ninstrs; i++) {
2525		instr = &op->instrs[i];
2526
2527		nand_op_trace("  ", instr);
2528
2529		switch (instr->type) {
2530		case NAND_OP_WAITRDY_INSTR:
2531			desc = gpmi_chain_wait_ready(this);
2532			break;
2533		case NAND_OP_CMD_INSTR:
2534			cmd = instr->ctx.cmd.opcode;
2535
2536			/*
2537			 * When this command has an address cycle chain it
2538			 * together with the address cycle
2539			 */
2540			if (i + 1 != op->ninstrs &&
2541			    op->instrs[i + 1].type == NAND_OP_ADDR_INSTR)
2542				continue;
2543
2544			desc = gpmi_chain_command(this, cmd, NULL, 0);
2545
2546			break;
2547		case NAND_OP_ADDR_INSTR:
2548			desc = gpmi_chain_command(this, cmd, instr->ctx.addr.addrs,
2549						  instr->ctx.addr.naddrs);
2550			break;
2551		case NAND_OP_DATA_OUT_INSTR:
2552			buf_write = instr->ctx.data.buf.out;
2553			buf_len = instr->ctx.data.len;
2554			nbufs++;
2555
2556			desc = gpmi_chain_data_write(this, buf_write, buf_len);
2557
2558			break;
2559		case NAND_OP_DATA_IN_INSTR:
2560			if (!instr->ctx.data.len)
2561				break;
2562			buf_read = instr->ctx.data.buf.in;
2563			buf_len = instr->ctx.data.len;
2564			nbufs++;
2565
2566			desc = gpmi_chain_data_read(this, buf_read, buf_len,
2567						   &direct);
2568			break;
2569		}
2570
2571		if (!desc) {
2572			ret = -ENXIO;
2573			goto unmap;
2574		}
2575	}
2576
2577	dev_dbg(this->dev, "%s setup done\n", __func__);
2578
2579	if (nbufs > 1) {
2580		dev_err(this->dev, "Multiple data instructions not supported\n");
2581		ret = -EINVAL;
2582		goto unmap;
2583	}
2584
2585	if (this->bch) {
2586		writel(this->bch_flashlayout0,
2587		       this->resources.bch_regs + HW_BCH_FLASH0LAYOUT0);
2588		writel(this->bch_flashlayout1,
2589		       this->resources.bch_regs + HW_BCH_FLASH0LAYOUT1);
2590	}
2591
2592	desc->callback = dma_irq_callback;
2593	desc->callback_param = this;
2594	dma_completion = &this->dma_done;
2595	bch_completion = NULL;
2596
2597	init_completion(dma_completion);
2598
2599	if (this->bch && buf_read) {
2600		writel(BM_BCH_CTRL_COMPLETE_IRQ_EN,
2601		       this->resources.bch_regs + HW_BCH_CTRL_SET);
2602		bch_completion = &this->bch_done;
2603		init_completion(bch_completion);
2604	}
2605
2606	dmaengine_submit(desc);
2607	dma_async_issue_pending(get_dma_chan(this));
2608
2609	to = wait_for_completion_timeout(dma_completion, msecs_to_jiffies(1000));
2610	if (!to) {
2611		dev_err(this->dev, "DMA timeout, last DMA\n");
2612		gpmi_dump_info(this);
2613		ret = -ETIMEDOUT;
2614		goto unmap;
2615	}
2616
2617	if (this->bch && buf_read) {
2618		to = wait_for_completion_timeout(bch_completion, msecs_to_jiffies(1000));
2619		if (!to) {
2620			dev_err(this->dev, "BCH timeout, last DMA\n");
2621			gpmi_dump_info(this);
2622			ret = -ETIMEDOUT;
2623			goto unmap;
2624		}
2625	}
2626
2627	writel(BM_BCH_CTRL_COMPLETE_IRQ_EN,
2628	       this->resources.bch_regs + HW_BCH_CTRL_CLR);
2629	gpmi_clear_bch(this);
2630
2631	ret = 0;
2632
2633unmap:
2634	for (i = 0; i < this->ntransfers; i++) {
2635		struct gpmi_transfer *transfer = &this->transfers[i];
2636
2637		if (transfer->direction != DMA_NONE)
2638			dma_unmap_sg(this->dev, &transfer->sgl, 1,
2639				     transfer->direction);
2640	}
2641
2642	if (!ret && buf_read && !direct)
2643		memcpy(buf_read, this->data_buffer_dma,
2644		       gpmi_raw_len_to_len(this, buf_len));
2645
2646	this->bch = false;
2647
2648out_pm:
2649	pm_runtime_mark_last_busy(this->dev);
2650	pm_runtime_put_autosuspend(this->dev);
2651
2652	return ret;
2653}
2654
2655static const struct nand_controller_ops gpmi_nand_controller_ops = {
2656	.attach_chip = gpmi_nand_attach_chip,
2657	.setup_interface = gpmi_setup_interface,
2658	.exec_op = gpmi_nfc_exec_op,
2659};
2660
2661static int gpmi_nand_init(struct gpmi_nand_data *this)
2662{
2663	struct nand_chip *chip = &this->nand;
2664	struct mtd_info  *mtd = nand_to_mtd(chip);
2665	int ret;
2666
2667	/* init the MTD data structures */
2668	mtd->name		= "gpmi-nand";
2669	mtd->dev.parent		= this->dev;
2670
2671	/* init the nand_chip{}, we don't support a 16-bit NAND Flash bus. */
2672	nand_set_controller_data(chip, this);
2673	nand_set_flash_node(chip, this->pdev->dev.of_node);
2674	chip->legacy.block_markbad = gpmi_block_markbad;
2675	chip->badblock_pattern	= &gpmi_bbt_descr;
2676	chip->options		|= NAND_NO_SUBPAGE_WRITE;
2677
2678	/* Set up swap_block_mark, must be set before the gpmi_set_geometry() */
2679	this->swap_block_mark = !GPMI_IS_MX23(this);
2680
2681	/*
2682	 * Allocate a temporary DMA buffer for reading ID in the
2683	 * nand_scan_ident().
2684	 */
2685	this->bch_geometry.payload_size = 1024;
2686	this->bch_geometry.auxiliary_size = 128;
2687	ret = gpmi_alloc_dma_buffer(this);
2688	if (ret)
2689		return ret;
2690
2691	nand_controller_init(&this->base);
2692	this->base.ops = &gpmi_nand_controller_ops;
2693	chip->controller = &this->base;
2694
2695	ret = nand_scan(chip, GPMI_IS_MX6(this) ? 2 : 1);
2696	if (ret)
2697		goto err_out;
2698
2699	ret = nand_boot_init(this);
2700	if (ret)
2701		goto err_nand_cleanup;
2702	ret = nand_create_bbt(chip);
2703	if (ret)
2704		goto err_nand_cleanup;
2705
2706	ret = mtd_device_register(mtd, NULL, 0);
2707	if (ret)
2708		goto err_nand_cleanup;
2709	return 0;
2710
2711err_nand_cleanup:
2712	nand_cleanup(chip);
2713err_out:
2714	gpmi_free_dma_buffer(this);
2715	return ret;
2716}
2717
2718static const struct of_device_id gpmi_nand_id_table[] = {
2719	{ .compatible = "fsl,imx23-gpmi-nand", .data = &gpmi_devdata_imx23, },
2720	{ .compatible = "fsl,imx28-gpmi-nand", .data = &gpmi_devdata_imx28, },
2721	{ .compatible = "fsl,imx6q-gpmi-nand", .data = &gpmi_devdata_imx6q, },
2722	{ .compatible = "fsl,imx6sx-gpmi-nand", .data = &gpmi_devdata_imx6sx, },
2723	{ .compatible = "fsl,imx7d-gpmi-nand", .data = &gpmi_devdata_imx7d,},
2724	{}
2725};
2726MODULE_DEVICE_TABLE(of, gpmi_nand_id_table);
2727
2728static int gpmi_nand_probe(struct platform_device *pdev)
2729{
2730	struct gpmi_nand_data *this;
2731	int ret;
2732
2733	this = devm_kzalloc(&pdev->dev, sizeof(*this), GFP_KERNEL);
2734	if (!this)
2735		return -ENOMEM;
2736
2737	this->devdata = of_device_get_match_data(&pdev->dev);
2738	platform_set_drvdata(pdev, this);
2739	this->pdev  = pdev;
2740	this->dev   = &pdev->dev;
2741
2742	ret = acquire_resources(this);
2743	if (ret)
2744		goto exit_acquire_resources;
2745
2746	ret = __gpmi_enable_clk(this, true);
2747	if (ret)
2748		goto exit_acquire_resources;
2749
2750	pm_runtime_set_autosuspend_delay(&pdev->dev, 500);
2751	pm_runtime_use_autosuspend(&pdev->dev);
2752	pm_runtime_set_active(&pdev->dev);
2753	pm_runtime_enable(&pdev->dev);
2754	pm_runtime_get_sync(&pdev->dev);
2755
2756	ret = gpmi_init(this);
2757	if (ret)
2758		goto exit_nfc_init;
2759
2760	ret = gpmi_nand_init(this);
2761	if (ret)
2762		goto exit_nfc_init;
2763
2764	pm_runtime_mark_last_busy(&pdev->dev);
2765	pm_runtime_put_autosuspend(&pdev->dev);
2766
2767	dev_info(this->dev, "driver registered.\n");
2768
2769	return 0;
2770
2771exit_nfc_init:
2772	pm_runtime_put(&pdev->dev);
2773	pm_runtime_disable(&pdev->dev);
2774	release_resources(this);
2775exit_acquire_resources:
2776
2777	return ret;
2778}
2779
2780static void gpmi_nand_remove(struct platform_device *pdev)
2781{
2782	struct gpmi_nand_data *this = platform_get_drvdata(pdev);
2783	struct nand_chip *chip = &this->nand;
2784	int ret;
2785
2786	pm_runtime_put_sync(&pdev->dev);
2787	pm_runtime_disable(&pdev->dev);
2788
2789	ret = mtd_device_unregister(nand_to_mtd(chip));
2790	WARN_ON(ret);
2791	nand_cleanup(chip);
2792	gpmi_free_dma_buffer(this);
2793	release_resources(this);
2794}
2795
2796#ifdef CONFIG_PM_SLEEP
2797static int gpmi_pm_suspend(struct device *dev)
2798{
2799	struct gpmi_nand_data *this = dev_get_drvdata(dev);
2800
2801	release_dma_channels(this);
2802	return 0;
2803}
2804
2805static int gpmi_pm_resume(struct device *dev)
2806{
2807	struct gpmi_nand_data *this = dev_get_drvdata(dev);
2808	int ret;
2809
2810	ret = acquire_dma_channels(this);
2811	if (ret < 0)
2812		return ret;
2813
2814	/* re-init the GPMI registers */
2815	ret = gpmi_init(this);
2816	if (ret) {
2817		dev_err(this->dev, "Error setting GPMI : %d\n", ret);
2818		return ret;
2819	}
2820
2821	/* Set flag to get timing setup restored for next exec_op */
2822	if (this->hw.clk_rate)
2823		this->hw.must_apply_timings = true;
2824
2825	/* re-init the BCH registers */
2826	ret = bch_set_geometry(this);
2827	if (ret) {
2828		dev_err(this->dev, "Error setting BCH : %d\n", ret);
2829		return ret;
2830	}
2831
2832	return 0;
2833}
2834#endif /* CONFIG_PM_SLEEP */
2835
2836static int __maybe_unused gpmi_runtime_suspend(struct device *dev)
2837{
2838	struct gpmi_nand_data *this = dev_get_drvdata(dev);
2839
2840	return __gpmi_enable_clk(this, false);
2841}
2842
2843static int __maybe_unused gpmi_runtime_resume(struct device *dev)
2844{
2845	struct gpmi_nand_data *this = dev_get_drvdata(dev);
2846
2847	return __gpmi_enable_clk(this, true);
2848}
2849
2850static const struct dev_pm_ops gpmi_pm_ops = {
2851	SET_SYSTEM_SLEEP_PM_OPS(gpmi_pm_suspend, gpmi_pm_resume)
2852	SET_RUNTIME_PM_OPS(gpmi_runtime_suspend, gpmi_runtime_resume, NULL)
2853};
2854
2855static struct platform_driver gpmi_nand_driver = {
2856	.driver = {
2857		.name = "gpmi-nand",
2858		.pm = &gpmi_pm_ops,
2859		.of_match_table = gpmi_nand_id_table,
2860	},
2861	.probe   = gpmi_nand_probe,
2862	.remove_new = gpmi_nand_remove,
2863};
2864module_platform_driver(gpmi_nand_driver);
2865
2866MODULE_AUTHOR("Freescale Semiconductor, Inc.");
2867MODULE_DESCRIPTION("i.MX GPMI NAND Flash Controller Driver");
2868MODULE_LICENSE("GPL");
2869