1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Freescale GPMI NAND Flash Driver
4  *
5  * Copyright (C) 2008-2011 Freescale Semiconductor, Inc.
6  * Copyright (C) 2008 Embedded Alley Solutions, Inc.
7  */
8 #include <linux/delay.h>
9 #include <linux/clk.h>
10 #include <linux/slab.h>
11 
12 #include "gpmi-nand.h"
13 #include "gpmi-regs.h"
14 #include "bch-regs.h"
15 
16 /* Converts time to clock cycles */
17 #define TO_CYCLES(duration, period) DIV_ROUND_UP_ULL(duration, period)
18 
19 #define MXS_SET_ADDR		0x4
20 #define MXS_CLR_ADDR		0x8
21 /*
22  * Clear the bit and poll it cleared.  This is usually called with
23  * a reset address and mask being either SFTRST(bit 31) or CLKGATE
24  * (bit 30).
25  */
clear_poll_bit(void __iomem * addr,u32 mask)26 static int clear_poll_bit(void __iomem *addr, u32 mask)
27 {
28 	int timeout = 0x400;
29 
30 	/* clear the bit */
31 	writel(mask, addr + MXS_CLR_ADDR);
32 
33 	/*
34 	 * SFTRST needs 3 GPMI clocks to settle, the reference manual
35 	 * recommends to wait 1us.
36 	 */
37 	udelay(1);
38 
39 	/* poll the bit becoming clear */
40 	while ((readl(addr) & mask) && --timeout)
41 		/* nothing */;
42 
43 	return !timeout;
44 }
45 
46 #define MODULE_CLKGATE		(1 << 30)
47 #define MODULE_SFTRST		(1 << 31)
48 /*
49  * The current mxs_reset_block() will do two things:
50  *  [1] enable the module.
51  *  [2] reset the module.
52  *
53  * In most of the cases, it's ok.
54  * But in MX23, there is a hardware bug in the BCH block (see erratum #2847).
55  * If you try to soft reset the BCH block, it becomes unusable until
56  * the next hard reset. This case occurs in the NAND boot mode. When the board
57  * boots by NAND, the ROM of the chip will initialize the BCH blocks itself.
58  * So If the driver tries to reset the BCH again, the BCH will not work anymore.
59  * You will see a DMA timeout in this case. The bug has been fixed
60  * in the following chips, such as MX28.
61  *
62  * To avoid this bug, just add a new parameter `just_enable` for
63  * the mxs_reset_block(), and rewrite it here.
64  */
gpmi_reset_block(void __iomem * reset_addr,bool just_enable)65 static int gpmi_reset_block(void __iomem *reset_addr, bool just_enable)
66 {
67 	int ret;
68 	int timeout = 0x400;
69 
70 	/* clear and poll SFTRST */
71 	ret = clear_poll_bit(reset_addr, MODULE_SFTRST);
72 	if (unlikely(ret))
73 		goto error;
74 
75 	/* clear CLKGATE */
76 	writel(MODULE_CLKGATE, reset_addr + MXS_CLR_ADDR);
77 
78 	if (!just_enable) {
79 		/* set SFTRST to reset the block */
80 		writel(MODULE_SFTRST, reset_addr + MXS_SET_ADDR);
81 		udelay(1);
82 
83 		/* poll CLKGATE becoming set */
84 		while ((!(readl(reset_addr) & MODULE_CLKGATE)) && --timeout)
85 			/* nothing */;
86 		if (unlikely(!timeout))
87 			goto error;
88 	}
89 
90 	/* clear and poll SFTRST */
91 	ret = clear_poll_bit(reset_addr, MODULE_SFTRST);
92 	if (unlikely(ret))
93 		goto error;
94 
95 	/* clear and poll CLKGATE */
96 	ret = clear_poll_bit(reset_addr, MODULE_CLKGATE);
97 	if (unlikely(ret))
98 		goto error;
99 
100 	return 0;
101 
102 error:
103 	pr_err("%s(%p): module reset timeout\n", __func__, reset_addr);
104 	return -ETIMEDOUT;
105 }
106 
__gpmi_enable_clk(struct gpmi_nand_data * this,bool v)107 static int __gpmi_enable_clk(struct gpmi_nand_data *this, bool v)
108 {
109 	struct clk *clk;
110 	int ret;
111 	int i;
112 
113 	for (i = 0; i < GPMI_CLK_MAX; i++) {
114 		clk = this->resources.clock[i];
115 		if (!clk)
116 			break;
117 
118 		if (v) {
119 			ret = clk_prepare_enable(clk);
120 			if (ret)
121 				goto err_clk;
122 		} else {
123 			clk_disable_unprepare(clk);
124 		}
125 	}
126 	return 0;
127 
128 err_clk:
129 	for (; i > 0; i--)
130 		clk_disable_unprepare(this->resources.clock[i - 1]);
131 	return ret;
132 }
133 
gpmi_enable_clk(struct gpmi_nand_data * this)134 int gpmi_enable_clk(struct gpmi_nand_data *this)
135 {
136 	return __gpmi_enable_clk(this, true);
137 }
138 
gpmi_disable_clk(struct gpmi_nand_data * this)139 int gpmi_disable_clk(struct gpmi_nand_data *this)
140 {
141 	return __gpmi_enable_clk(this, false);
142 }
143 
gpmi_init(struct gpmi_nand_data * this)144 int gpmi_init(struct gpmi_nand_data *this)
145 {
146 	struct resources *r = &this->resources;
147 	int ret;
148 
149 	ret = gpmi_enable_clk(this);
150 	if (ret)
151 		return ret;
152 	ret = gpmi_reset_block(r->gpmi_regs, false);
153 	if (ret)
154 		goto err_out;
155 
156 	/*
157 	 * Reset BCH here, too. We got failures otherwise :(
158 	 * See later BCH reset for explanation of MX23 handling
159 	 */
160 	ret = gpmi_reset_block(r->bch_regs, GPMI_IS_MX23(this));
161 	if (ret)
162 		goto err_out;
163 
164 	/* Choose NAND mode. */
165 	writel(BM_GPMI_CTRL1_GPMI_MODE, r->gpmi_regs + HW_GPMI_CTRL1_CLR);
166 
167 	/* Set the IRQ polarity. */
168 	writel(BM_GPMI_CTRL1_ATA_IRQRDY_POLARITY,
169 				r->gpmi_regs + HW_GPMI_CTRL1_SET);
170 
171 	/* Disable Write-Protection. */
172 	writel(BM_GPMI_CTRL1_DEV_RESET, r->gpmi_regs + HW_GPMI_CTRL1_SET);
173 
174 	/* Select BCH ECC. */
175 	writel(BM_GPMI_CTRL1_BCH_MODE, r->gpmi_regs + HW_GPMI_CTRL1_SET);
176 
177 	/*
178 	 * Decouple the chip select from dma channel. We use dma0 for all
179 	 * the chips.
180 	 */
181 	writel(BM_GPMI_CTRL1_DECOUPLE_CS, r->gpmi_regs + HW_GPMI_CTRL1_SET);
182 
183 	gpmi_disable_clk(this);
184 	return 0;
185 err_out:
186 	gpmi_disable_clk(this);
187 	return ret;
188 }
189 
190 /* This function is very useful. It is called only when the bug occur. */
gpmi_dump_info(struct gpmi_nand_data * this)191 void gpmi_dump_info(struct gpmi_nand_data *this)
192 {
193 	struct resources *r = &this->resources;
194 	struct bch_geometry *geo = &this->bch_geometry;
195 	u32 reg;
196 	int i;
197 
198 	dev_err(this->dev, "Show GPMI registers :\n");
199 	for (i = 0; i <= HW_GPMI_DEBUG / 0x10 + 1; i++) {
200 		reg = readl(r->gpmi_regs + i * 0x10);
201 		dev_err(this->dev, "offset 0x%.3x : 0x%.8x\n", i * 0x10, reg);
202 	}
203 
204 	/* start to print out the BCH info */
205 	dev_err(this->dev, "Show BCH registers :\n");
206 	for (i = 0; i <= HW_BCH_VERSION / 0x10 + 1; i++) {
207 		reg = readl(r->bch_regs + i * 0x10);
208 		dev_err(this->dev, "offset 0x%.3x : 0x%.8x\n", i * 0x10, reg);
209 	}
210 	dev_err(this->dev, "BCH Geometry :\n"
211 		"GF length              : %u\n"
212 		"ECC Strength           : %u\n"
213 		"Page Size in Bytes     : %u\n"
214 		"Metadata Size in Bytes : %u\n"
215 		"ECC Chunk Size in Bytes: %u\n"
216 		"ECC Chunk Count        : %u\n"
217 		"Payload Size in Bytes  : %u\n"
218 		"Auxiliary Size in Bytes: %u\n"
219 		"Auxiliary Status Offset: %u\n"
220 		"Block Mark Byte Offset : %u\n"
221 		"Block Mark Bit Offset  : %u\n",
222 		geo->gf_len,
223 		geo->ecc_strength,
224 		geo->page_size,
225 		geo->metadata_size,
226 		geo->ecc_chunk_size,
227 		geo->ecc_chunk_count,
228 		geo->payload_size,
229 		geo->auxiliary_size,
230 		geo->auxiliary_status_offset,
231 		geo->block_mark_byte_offset,
232 		geo->block_mark_bit_offset);
233 }
234 
235 /* Configures the geometry for BCH.  */
bch_set_geometry(struct gpmi_nand_data * this)236 int bch_set_geometry(struct gpmi_nand_data *this)
237 {
238 	struct resources *r = &this->resources;
239 	struct bch_geometry *bch_geo = &this->bch_geometry;
240 	unsigned int block_count;
241 	unsigned int block_size;
242 	unsigned int metadata_size;
243 	unsigned int ecc_strength;
244 	unsigned int page_size;
245 	unsigned int gf_len;
246 	int ret;
247 
248 	ret = common_nfc_set_geometry(this);
249 	if (ret)
250 		return ret;
251 
252 	block_count   = bch_geo->ecc_chunk_count - 1;
253 	block_size    = bch_geo->ecc_chunk_size;
254 	metadata_size = bch_geo->metadata_size;
255 	ecc_strength  = bch_geo->ecc_strength >> 1;
256 	page_size     = bch_geo->page_size;
257 	gf_len        = bch_geo->gf_len;
258 
259 	ret = gpmi_enable_clk(this);
260 	if (ret)
261 		return ret;
262 
263 	/*
264 	* Due to erratum #2847 of the MX23, the BCH cannot be soft reset on this
265 	* chip, otherwise it will lock up. So we skip resetting BCH on the MX23.
266 	* On the other hand, the MX28 needs the reset, because one case has been
267 	* seen where the BCH produced ECC errors constantly after 10000
268 	* consecutive reboots. The latter case has not been seen on the MX23
269 	* yet, still we don't know if it could happen there as well.
270 	*/
271 	ret = gpmi_reset_block(r->bch_regs, GPMI_IS_MX23(this));
272 	if (ret)
273 		goto err_out;
274 
275 	/* Configure layout 0. */
276 	writel(BF_BCH_FLASH0LAYOUT0_NBLOCKS(block_count)
277 			| BF_BCH_FLASH0LAYOUT0_META_SIZE(metadata_size)
278 			| BF_BCH_FLASH0LAYOUT0_ECC0(ecc_strength, this)
279 			| BF_BCH_FLASH0LAYOUT0_GF(gf_len, this)
280 			| BF_BCH_FLASH0LAYOUT0_DATA0_SIZE(block_size, this),
281 			r->bch_regs + HW_BCH_FLASH0LAYOUT0);
282 
283 	writel(BF_BCH_FLASH0LAYOUT1_PAGE_SIZE(page_size)
284 			| BF_BCH_FLASH0LAYOUT1_ECCN(ecc_strength, this)
285 			| BF_BCH_FLASH0LAYOUT1_GF(gf_len, this)
286 			| BF_BCH_FLASH0LAYOUT1_DATAN_SIZE(block_size, this),
287 			r->bch_regs + HW_BCH_FLASH0LAYOUT1);
288 
289 	/* Set *all* chip selects to use layout 0. */
290 	writel(0, r->bch_regs + HW_BCH_LAYOUTSELECT);
291 
292 	/* Enable interrupts. */
293 	writel(BM_BCH_CTRL_COMPLETE_IRQ_EN,
294 				r->bch_regs + HW_BCH_CTRL_SET);
295 
296 	gpmi_disable_clk(this);
297 	return 0;
298 err_out:
299 	gpmi_disable_clk(this);
300 	return ret;
301 }
302 
303 /*
304  * <1> Firstly, we should know what's the GPMI-clock means.
305  *     The GPMI-clock is the internal clock in the gpmi nand controller.
306  *     If you set 100MHz to gpmi nand controller, the GPMI-clock's period
307  *     is 10ns. Mark the GPMI-clock's period as GPMI-clock-period.
308  *
309  * <2> Secondly, we should know what's the frequency on the nand chip pins.
310  *     The frequency on the nand chip pins is derived from the GPMI-clock.
311  *     We can get it from the following equation:
312  *
313  *         F = G / (DS + DH)
314  *
315  *         F  : the frequency on the nand chip pins.
316  *         G  : the GPMI clock, such as 100MHz.
317  *         DS : GPMI_HW_GPMI_TIMING0:DATA_SETUP
318  *         DH : GPMI_HW_GPMI_TIMING0:DATA_HOLD
319  *
320  * <3> Thirdly, when the frequency on the nand chip pins is above 33MHz,
321  *     the nand EDO(extended Data Out) timing could be applied.
322  *     The GPMI implements a feedback read strobe to sample the read data.
323  *     The feedback read strobe can be delayed to support the nand EDO timing
324  *     where the read strobe may deasserts before the read data is valid, and
325  *     read data is valid for some time after read strobe.
326  *
327  *     The following figure illustrates some aspects of a NAND Flash read:
328  *
329  *                   |<---tREA---->|
330  *                   |             |
331  *                   |         |   |
332  *                   |<--tRP-->|   |
333  *                   |         |   |
334  *                  __          ___|__________________________________
335  *     RDN            \________/   |
336  *                                 |
337  *                                 /---------\
338  *     Read Data    --------------<           >---------
339  *                                 \---------/
340  *                                |     |
341  *                                |<-D->|
342  *     FeedbackRDN  ________             ____________
343  *                          \___________/
344  *
345  *          D stands for delay, set in the HW_GPMI_CTRL1:RDN_DELAY.
346  *
347  *
348  * <4> Now, we begin to describe how to compute the right RDN_DELAY.
349  *
350  *  4.1) From the aspect of the nand chip pins:
351  *        Delay = (tREA + C - tRP)               {1}
352  *
353  *        tREA : the maximum read access time.
354  *        C    : a constant to adjust the delay. default is 4000ps.
355  *        tRP  : the read pulse width, which is exactly:
356  *                   tRP = (GPMI-clock-period) * DATA_SETUP
357  *
358  *  4.2) From the aspect of the GPMI nand controller:
359  *         Delay = RDN_DELAY * 0.125 * RP        {2}
360  *
361  *         RP   : the DLL reference period.
362  *            if (GPMI-clock-period > DLL_THRETHOLD)
363  *                   RP = GPMI-clock-period / 2;
364  *            else
365  *                   RP = GPMI-clock-period;
366  *
367  *            Set the HW_GPMI_CTRL1:HALF_PERIOD if GPMI-clock-period
368  *            is greater DLL_THRETHOLD. In other SOCs, the DLL_THRETHOLD
369  *            is 16000ps, but in mx6q, we use 12000ps.
370  *
371  *  4.3) since {1} equals {2}, we get:
372  *
373  *                     (tREA + 4000 - tRP) * 8
374  *         RDN_DELAY = -----------------------     {3}
375  *                           RP
376  */
gpmi_nfc_compute_timings(struct gpmi_nand_data * this,const struct nand_sdr_timings * sdr)377 static void gpmi_nfc_compute_timings(struct gpmi_nand_data *this,
378 				     const struct nand_sdr_timings *sdr)
379 {
380 	struct gpmi_nfc_hardware_timing *hw = &this->hw;
381 	unsigned int dll_threshold_ps = this->devdata->max_chain_delay;
382 	unsigned int period_ps, reference_period_ps;
383 	unsigned int data_setup_cycles, data_hold_cycles, addr_setup_cycles;
384 	unsigned int tRP_ps;
385 	bool use_half_period;
386 	int sample_delay_ps, sample_delay_factor;
387 	u16 busy_timeout_cycles;
388 	u8 wrn_dly_sel;
389 
390 	if (sdr->tRC_min >= 30000) {
391 		/* ONFI non-EDO modes [0-3] */
392 		hw->clk_rate = 22000000;
393 		wrn_dly_sel = BV_GPMI_CTRL1_WRN_DLY_SEL_4_TO_8NS;
394 	} else if (sdr->tRC_min >= 25000) {
395 		/* ONFI EDO mode 4 */
396 		hw->clk_rate = 80000000;
397 		wrn_dly_sel = BV_GPMI_CTRL1_WRN_DLY_SEL_NO_DELAY;
398 	} else {
399 		/* ONFI EDO mode 5 */
400 		hw->clk_rate = 100000000;
401 		wrn_dly_sel = BV_GPMI_CTRL1_WRN_DLY_SEL_NO_DELAY;
402 	}
403 
404 	/* SDR core timings are given in picoseconds */
405 	period_ps = div_u64((u64)NSEC_PER_SEC * 1000, hw->clk_rate);
406 
407 	addr_setup_cycles = TO_CYCLES(sdr->tALS_min, period_ps);
408 	data_setup_cycles = TO_CYCLES(sdr->tDS_min, period_ps);
409 	data_hold_cycles = TO_CYCLES(sdr->tDH_min, period_ps);
410 	busy_timeout_cycles = TO_CYCLES(sdr->tWB_max + sdr->tR_max, period_ps);
411 
412 	hw->timing0 = BF_GPMI_TIMING0_ADDRESS_SETUP(addr_setup_cycles) |
413 		      BF_GPMI_TIMING0_DATA_HOLD(data_hold_cycles) |
414 		      BF_GPMI_TIMING0_DATA_SETUP(data_setup_cycles);
415 	hw->timing1 = BF_GPMI_TIMING1_BUSY_TIMEOUT(busy_timeout_cycles * 4096);
416 
417 	/*
418 	 * Derive NFC ideal delay from {3}:
419 	 *
420 	 *                     (tREA + 4000 - tRP) * 8
421 	 *         RDN_DELAY = -----------------------
422 	 *                                RP
423 	 */
424 	if (period_ps > dll_threshold_ps) {
425 		use_half_period = true;
426 		reference_period_ps = period_ps / 2;
427 	} else {
428 		use_half_period = false;
429 		reference_period_ps = period_ps;
430 	}
431 
432 	tRP_ps = data_setup_cycles * period_ps;
433 	sample_delay_ps = (sdr->tREA_max + 4000 - tRP_ps) * 8;
434 	if (sample_delay_ps > 0)
435 		sample_delay_factor = sample_delay_ps / reference_period_ps;
436 	else
437 		sample_delay_factor = 0;
438 
439 	hw->ctrl1n = BF_GPMI_CTRL1_WRN_DLY_SEL(wrn_dly_sel);
440 	if (sample_delay_factor)
441 		hw->ctrl1n |= BF_GPMI_CTRL1_RDN_DELAY(sample_delay_factor) |
442 			      BM_GPMI_CTRL1_DLL_ENABLE |
443 			      (use_half_period ? BM_GPMI_CTRL1_HALF_PERIOD : 0);
444 }
445 
gpmi_nfc_apply_timings(struct gpmi_nand_data * this)446 void gpmi_nfc_apply_timings(struct gpmi_nand_data *this)
447 {
448 	struct gpmi_nfc_hardware_timing *hw = &this->hw;
449 	struct resources *r = &this->resources;
450 	void __iomem *gpmi_regs = r->gpmi_regs;
451 	unsigned int dll_wait_time_us;
452 
453 	clk_set_rate(r->clock[0], hw->clk_rate);
454 
455 	writel(hw->timing0, gpmi_regs + HW_GPMI_TIMING0);
456 	writel(hw->timing1, gpmi_regs + HW_GPMI_TIMING1);
457 
458 	/*
459 	 * Clear several CTRL1 fields, DLL must be disabled when setting
460 	 * RDN_DELAY or HALF_PERIOD.
461 	 */
462 	writel(BM_GPMI_CTRL1_CLEAR_MASK, gpmi_regs + HW_GPMI_CTRL1_CLR);
463 	writel(hw->ctrl1n, gpmi_regs + HW_GPMI_CTRL1_SET);
464 
465 	/* Wait 64 clock cycles before using the GPMI after enabling the DLL */
466 	dll_wait_time_us = USEC_PER_SEC / hw->clk_rate * 64;
467 	if (!dll_wait_time_us)
468 		dll_wait_time_us = 1;
469 
470 	/* Wait for the DLL to settle. */
471 	udelay(dll_wait_time_us);
472 }
473 
gpmi_setup_data_interface(struct mtd_info * mtd,int chipnr,const struct nand_data_interface * conf)474 int gpmi_setup_data_interface(struct mtd_info *mtd, int chipnr,
475 			      const struct nand_data_interface *conf)
476 {
477 	struct nand_chip *chip = mtd_to_nand(mtd);
478 	struct gpmi_nand_data *this = nand_get_controller_data(chip);
479 	const struct nand_sdr_timings *sdr;
480 
481 	/* Retrieve required NAND timings */
482 	sdr = nand_get_sdr_timings(conf);
483 	if (IS_ERR(sdr))
484 		return PTR_ERR(sdr);
485 
486 	/* Only MX6 GPMI controller can reach EDO timings */
487 	if (sdr->tRC_min <= 25000 && !GPMI_IS_MX6(this))
488 		return -ENOTSUPP;
489 
490 	/* Stop here if this call was just a check */
491 	if (chipnr < 0)
492 		return 0;
493 
494 	/* Do the actual derivation of the controller timings */
495 	gpmi_nfc_compute_timings(this, sdr);
496 
497 	this->hw.must_apply_timings = true;
498 
499 	return 0;
500 }
501 
502 /* Clears a BCH interrupt. */
gpmi_clear_bch(struct gpmi_nand_data * this)503 void gpmi_clear_bch(struct gpmi_nand_data *this)
504 {
505 	struct resources *r = &this->resources;
506 	writel(BM_BCH_CTRL_COMPLETE_IRQ, r->bch_regs + HW_BCH_CTRL_CLR);
507 }
508 
509 /* Returns the Ready/Busy status of the given chip. */
gpmi_is_ready(struct gpmi_nand_data * this,unsigned chip)510 int gpmi_is_ready(struct gpmi_nand_data *this, unsigned chip)
511 {
512 	struct resources *r = &this->resources;
513 	uint32_t mask = 0;
514 	uint32_t reg = 0;
515 
516 	if (GPMI_IS_MX23(this)) {
517 		mask = MX23_BM_GPMI_DEBUG_READY0 << chip;
518 		reg = readl(r->gpmi_regs + HW_GPMI_DEBUG);
519 	} else if (GPMI_IS_MX28(this) || GPMI_IS_MX6(this)) {
520 		/*
521 		 * In the imx6, all the ready/busy pins are bound
522 		 * together. So we only need to check chip 0.
523 		 */
524 		if (GPMI_IS_MX6(this))
525 			chip = 0;
526 
527 		/* MX28 shares the same R/B register as MX6Q. */
528 		mask = MX28_BF_GPMI_STAT_READY_BUSY(1 << chip);
529 		reg = readl(r->gpmi_regs + HW_GPMI_STAT);
530 	} else
531 		dev_err(this->dev, "unknown arch.\n");
532 	return reg & mask;
533 }
534 
gpmi_send_command(struct gpmi_nand_data * this)535 int gpmi_send_command(struct gpmi_nand_data *this)
536 {
537 	struct dma_chan *channel = get_dma_chan(this);
538 	struct dma_async_tx_descriptor *desc;
539 	struct scatterlist *sgl;
540 	int chip = this->current_chip;
541 	int ret;
542 	u32 pio[3];
543 
544 	/* [1] send out the PIO words */
545 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(BV_GPMI_CTRL0_COMMAND_MODE__WRITE)
546 		| BM_GPMI_CTRL0_WORD_LENGTH
547 		| BF_GPMI_CTRL0_CS(chip, this)
548 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
549 		| BF_GPMI_CTRL0_ADDRESS(BV_GPMI_CTRL0_ADDRESS__NAND_CLE)
550 		| BM_GPMI_CTRL0_ADDRESS_INCREMENT
551 		| BF_GPMI_CTRL0_XFER_COUNT(this->command_length);
552 	pio[1] = pio[2] = 0;
553 	desc = dmaengine_prep_slave_sg(channel,
554 					(struct scatterlist *)pio,
555 					ARRAY_SIZE(pio), DMA_TRANS_NONE, 0);
556 	if (!desc)
557 		return -EINVAL;
558 
559 	/* [2] send out the COMMAND + ADDRESS string stored in @buffer */
560 	sgl = &this->cmd_sgl;
561 
562 	sg_init_one(sgl, this->cmd_buffer, this->command_length);
563 	dma_map_sg(this->dev, sgl, 1, DMA_TO_DEVICE);
564 	desc = dmaengine_prep_slave_sg(channel,
565 				sgl, 1, DMA_MEM_TO_DEV,
566 				DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
567 	if (!desc)
568 		return -EINVAL;
569 
570 	/* [3] submit the DMA */
571 	ret = start_dma_without_bch_irq(this, desc);
572 
573 	dma_unmap_sg(this->dev, sgl, 1, DMA_TO_DEVICE);
574 
575 	return ret;
576 }
577 
gpmi_send_data(struct gpmi_nand_data * this,const void * buf,int len)578 int gpmi_send_data(struct gpmi_nand_data *this, const void *buf, int len)
579 {
580 	struct dma_async_tx_descriptor *desc;
581 	struct dma_chan *channel = get_dma_chan(this);
582 	int chip = this->current_chip;
583 	int ret;
584 	uint32_t command_mode;
585 	uint32_t address;
586 	u32 pio[2];
587 
588 	/* [1] PIO */
589 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__WRITE;
590 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
591 
592 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
593 		| BM_GPMI_CTRL0_WORD_LENGTH
594 		| BF_GPMI_CTRL0_CS(chip, this)
595 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
596 		| BF_GPMI_CTRL0_ADDRESS(address)
597 		| BF_GPMI_CTRL0_XFER_COUNT(len);
598 	pio[1] = 0;
599 	desc = dmaengine_prep_slave_sg(channel, (struct scatterlist *)pio,
600 					ARRAY_SIZE(pio), DMA_TRANS_NONE, 0);
601 	if (!desc)
602 		return -EINVAL;
603 
604 	/* [2] send DMA request */
605 	prepare_data_dma(this, buf, len, DMA_TO_DEVICE);
606 	desc = dmaengine_prep_slave_sg(channel, &this->data_sgl,
607 					1, DMA_MEM_TO_DEV,
608 					DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
609 	if (!desc)
610 		return -EINVAL;
611 
612 	/* [3] submit the DMA */
613 	ret = start_dma_without_bch_irq(this, desc);
614 
615 	dma_unmap_sg(this->dev, &this->data_sgl, 1, DMA_TO_DEVICE);
616 
617 	return ret;
618 }
619 
gpmi_read_data(struct gpmi_nand_data * this,void * buf,int len)620 int gpmi_read_data(struct gpmi_nand_data *this, void *buf, int len)
621 {
622 	struct dma_async_tx_descriptor *desc;
623 	struct dma_chan *channel = get_dma_chan(this);
624 	int chip = this->current_chip;
625 	int ret;
626 	u32 pio[2];
627 	bool direct;
628 
629 	/* [1] : send PIO */
630 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(BV_GPMI_CTRL0_COMMAND_MODE__READ)
631 		| BM_GPMI_CTRL0_WORD_LENGTH
632 		| BF_GPMI_CTRL0_CS(chip, this)
633 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
634 		| BF_GPMI_CTRL0_ADDRESS(BV_GPMI_CTRL0_ADDRESS__NAND_DATA)
635 		| BF_GPMI_CTRL0_XFER_COUNT(len);
636 	pio[1] = 0;
637 	desc = dmaengine_prep_slave_sg(channel,
638 					(struct scatterlist *)pio,
639 					ARRAY_SIZE(pio), DMA_TRANS_NONE, 0);
640 	if (!desc)
641 		return -EINVAL;
642 
643 	/* [2] : send DMA request */
644 	direct = prepare_data_dma(this, buf, len, DMA_FROM_DEVICE);
645 	desc = dmaengine_prep_slave_sg(channel, &this->data_sgl,
646 					1, DMA_DEV_TO_MEM,
647 					DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
648 	if (!desc)
649 		return -EINVAL;
650 
651 	/* [3] : submit the DMA */
652 
653 	ret = start_dma_without_bch_irq(this, desc);
654 
655 	dma_unmap_sg(this->dev, &this->data_sgl, 1, DMA_FROM_DEVICE);
656 	if (!direct)
657 		memcpy(buf, this->data_buffer_dma, len);
658 
659 	return ret;
660 }
661 
gpmi_send_page(struct gpmi_nand_data * this,dma_addr_t payload,dma_addr_t auxiliary)662 int gpmi_send_page(struct gpmi_nand_data *this,
663 			dma_addr_t payload, dma_addr_t auxiliary)
664 {
665 	struct bch_geometry *geo = &this->bch_geometry;
666 	uint32_t command_mode;
667 	uint32_t address;
668 	uint32_t ecc_command;
669 	uint32_t buffer_mask;
670 	struct dma_async_tx_descriptor *desc;
671 	struct dma_chan *channel = get_dma_chan(this);
672 	int chip = this->current_chip;
673 	u32 pio[6];
674 
675 	/* A DMA descriptor that does an ECC page read. */
676 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__WRITE;
677 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
678 	ecc_command  = BV_GPMI_ECCCTRL_ECC_CMD__BCH_ENCODE;
679 	buffer_mask  = BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_PAGE |
680 				BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_AUXONLY;
681 
682 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
683 		| BM_GPMI_CTRL0_WORD_LENGTH
684 		| BF_GPMI_CTRL0_CS(chip, this)
685 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
686 		| BF_GPMI_CTRL0_ADDRESS(address)
687 		| BF_GPMI_CTRL0_XFER_COUNT(0);
688 	pio[1] = 0;
689 	pio[2] = BM_GPMI_ECCCTRL_ENABLE_ECC
690 		| BF_GPMI_ECCCTRL_ECC_CMD(ecc_command)
691 		| BF_GPMI_ECCCTRL_BUFFER_MASK(buffer_mask);
692 	pio[3] = geo->page_size;
693 	pio[4] = payload;
694 	pio[5] = auxiliary;
695 
696 	desc = dmaengine_prep_slave_sg(channel,
697 					(struct scatterlist *)pio,
698 					ARRAY_SIZE(pio), DMA_TRANS_NONE,
699 					DMA_CTRL_ACK);
700 	if (!desc)
701 		return -EINVAL;
702 
703 	return start_dma_with_bch_irq(this, desc);
704 }
705 
gpmi_read_page(struct gpmi_nand_data * this,dma_addr_t payload,dma_addr_t auxiliary)706 int gpmi_read_page(struct gpmi_nand_data *this,
707 				dma_addr_t payload, dma_addr_t auxiliary)
708 {
709 	struct bch_geometry *geo = &this->bch_geometry;
710 	uint32_t command_mode;
711 	uint32_t address;
712 	uint32_t ecc_command;
713 	uint32_t buffer_mask;
714 	struct dma_async_tx_descriptor *desc;
715 	struct dma_chan *channel = get_dma_chan(this);
716 	int chip = this->current_chip;
717 	u32 pio[6];
718 
719 	/* [1] Wait for the chip to report ready. */
720 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__WAIT_FOR_READY;
721 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
722 
723 	pio[0] =  BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
724 		| BM_GPMI_CTRL0_WORD_LENGTH
725 		| BF_GPMI_CTRL0_CS(chip, this)
726 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
727 		| BF_GPMI_CTRL0_ADDRESS(address)
728 		| BF_GPMI_CTRL0_XFER_COUNT(0);
729 	pio[1] = 0;
730 	desc = dmaengine_prep_slave_sg(channel,
731 				(struct scatterlist *)pio, 2,
732 				DMA_TRANS_NONE, 0);
733 	if (!desc)
734 		return -EINVAL;
735 
736 	/* [2] Enable the BCH block and read. */
737 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__READ;
738 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
739 	ecc_command  = BV_GPMI_ECCCTRL_ECC_CMD__BCH_DECODE;
740 	buffer_mask  = BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_PAGE
741 			| BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_AUXONLY;
742 
743 	pio[0] =  BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
744 		| BM_GPMI_CTRL0_WORD_LENGTH
745 		| BF_GPMI_CTRL0_CS(chip, this)
746 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
747 		| BF_GPMI_CTRL0_ADDRESS(address)
748 		| BF_GPMI_CTRL0_XFER_COUNT(geo->page_size);
749 
750 	pio[1] = 0;
751 	pio[2] =  BM_GPMI_ECCCTRL_ENABLE_ECC
752 		| BF_GPMI_ECCCTRL_ECC_CMD(ecc_command)
753 		| BF_GPMI_ECCCTRL_BUFFER_MASK(buffer_mask);
754 	pio[3] = geo->page_size;
755 	pio[4] = payload;
756 	pio[5] = auxiliary;
757 	desc = dmaengine_prep_slave_sg(channel,
758 					(struct scatterlist *)pio,
759 					ARRAY_SIZE(pio), DMA_TRANS_NONE,
760 					DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
761 	if (!desc)
762 		return -EINVAL;
763 
764 	/* [3] Disable the BCH block */
765 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__WAIT_FOR_READY;
766 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
767 
768 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
769 		| BM_GPMI_CTRL0_WORD_LENGTH
770 		| BF_GPMI_CTRL0_CS(chip, this)
771 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
772 		| BF_GPMI_CTRL0_ADDRESS(address)
773 		| BF_GPMI_CTRL0_XFER_COUNT(geo->page_size);
774 	pio[1] = 0;
775 	pio[2] = 0; /* clear GPMI_HW_GPMI_ECCCTRL, disable the BCH. */
776 	desc = dmaengine_prep_slave_sg(channel,
777 				(struct scatterlist *)pio, 3,
778 				DMA_TRANS_NONE,
779 				DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
780 	if (!desc)
781 		return -EINVAL;
782 
783 	/* [4] submit the DMA */
784 	return start_dma_with_bch_irq(this, desc);
785 }
786 
787 /**
788  * gpmi_copy_bits - copy bits from one memory region to another
789  * @dst: destination buffer
790  * @dst_bit_off: bit offset we're starting to write at
791  * @src: source buffer
792  * @src_bit_off: bit offset we're starting to read from
793  * @nbits: number of bits to copy
794  *
795  * This functions copies bits from one memory region to another, and is used by
796  * the GPMI driver to copy ECC sections which are not guaranteed to be byte
797  * aligned.
798  *
799  * src and dst should not overlap.
800  *
801  */
gpmi_copy_bits(u8 * dst,size_t dst_bit_off,const u8 * src,size_t src_bit_off,size_t nbits)802 void gpmi_copy_bits(u8 *dst, size_t dst_bit_off,
803 		    const u8 *src, size_t src_bit_off,
804 		    size_t nbits)
805 {
806 	size_t i;
807 	size_t nbytes;
808 	u32 src_buffer = 0;
809 	size_t bits_in_src_buffer = 0;
810 
811 	if (!nbits)
812 		return;
813 
814 	/*
815 	 * Move src and dst pointers to the closest byte pointer and store bit
816 	 * offsets within a byte.
817 	 */
818 	src += src_bit_off / 8;
819 	src_bit_off %= 8;
820 
821 	dst += dst_bit_off / 8;
822 	dst_bit_off %= 8;
823 
824 	/*
825 	 * Initialize the src_buffer value with bits available in the first
826 	 * byte of data so that we end up with a byte aligned src pointer.
827 	 */
828 	if (src_bit_off) {
829 		src_buffer = src[0] >> src_bit_off;
830 		if (nbits >= (8 - src_bit_off)) {
831 			bits_in_src_buffer += 8 - src_bit_off;
832 		} else {
833 			src_buffer &= GENMASK(nbits - 1, 0);
834 			bits_in_src_buffer += nbits;
835 		}
836 		nbits -= bits_in_src_buffer;
837 		src++;
838 	}
839 
840 	/* Calculate the number of bytes that can be copied from src to dst. */
841 	nbytes = nbits / 8;
842 
843 	/* Try to align dst to a byte boundary. */
844 	if (dst_bit_off) {
845 		if (bits_in_src_buffer < (8 - dst_bit_off) && nbytes) {
846 			src_buffer |= src[0] << bits_in_src_buffer;
847 			bits_in_src_buffer += 8;
848 			src++;
849 			nbytes--;
850 		}
851 
852 		if (bits_in_src_buffer >= (8 - dst_bit_off)) {
853 			dst[0] &= GENMASK(dst_bit_off - 1, 0);
854 			dst[0] |= src_buffer << dst_bit_off;
855 			src_buffer >>= (8 - dst_bit_off);
856 			bits_in_src_buffer -= (8 - dst_bit_off);
857 			dst_bit_off = 0;
858 			dst++;
859 			if (bits_in_src_buffer > 7) {
860 				bits_in_src_buffer -= 8;
861 				dst[0] = src_buffer;
862 				dst++;
863 				src_buffer >>= 8;
864 			}
865 		}
866 	}
867 
868 	if (!bits_in_src_buffer && !dst_bit_off) {
869 		/*
870 		 * Both src and dst pointers are byte aligned, thus we can
871 		 * just use the optimized memcpy function.
872 		 */
873 		if (nbytes)
874 			memcpy(dst, src, nbytes);
875 	} else {
876 		/*
877 		 * src buffer is not byte aligned, hence we have to copy each
878 		 * src byte to the src_buffer variable before extracting a byte
879 		 * to store in dst.
880 		 */
881 		for (i = 0; i < nbytes; i++) {
882 			src_buffer |= src[i] << bits_in_src_buffer;
883 			dst[i] = src_buffer;
884 			src_buffer >>= 8;
885 		}
886 	}
887 	/* Update dst and src pointers */
888 	dst += nbytes;
889 	src += nbytes;
890 
891 	/*
892 	 * nbits is the number of remaining bits. It should not exceed 8 as
893 	 * we've already copied as much bytes as possible.
894 	 */
895 	nbits %= 8;
896 
897 	/*
898 	 * If there's no more bits to copy to the destination and src buffer
899 	 * was already byte aligned, then we're done.
900 	 */
901 	if (!nbits && !bits_in_src_buffer)
902 		return;
903 
904 	/* Copy the remaining bits to src_buffer */
905 	if (nbits)
906 		src_buffer |= (*src & GENMASK(nbits - 1, 0)) <<
907 			      bits_in_src_buffer;
908 	bits_in_src_buffer += nbits;
909 
910 	/*
911 	 * In case there were not enough bits to get a byte aligned dst buffer
912 	 * prepare the src_buffer variable to match the dst organization (shift
913 	 * src_buffer by dst_bit_off and retrieve the least significant bits
914 	 * from dst).
915 	 */
916 	if (dst_bit_off)
917 		src_buffer = (src_buffer << dst_bit_off) |
918 			     (*dst & GENMASK(dst_bit_off - 1, 0));
919 	bits_in_src_buffer += dst_bit_off;
920 
921 	/*
922 	 * Keep most significant bits from dst if we end up with an unaligned
923 	 * number of bits.
924 	 */
925 	nbytes = bits_in_src_buffer / 8;
926 	if (bits_in_src_buffer % 8) {
927 		src_buffer |= (dst[nbytes] &
928 			       GENMASK(7, bits_in_src_buffer % 8)) <<
929 			      (nbytes * 8);
930 		nbytes++;
931 	}
932 
933 	/* Copy the remaining bytes to dst */
934 	for (i = 0; i < nbytes; i++) {
935 		dst[i] = src_buffer;
936 		src_buffer >>= 8;
937 	}
938 }
939