1 /*
2  * Freescale GPMI NAND Flash Driver
3  *
4  * Copyright (C) 2008-2011 Freescale Semiconductor, Inc.
5  * Copyright (C) 2008 Embedded Alley Solutions, Inc.
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation; either version 2 of the License, or
10  * (at your option) any later version.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License along
18  * with this program; if not, write to the Free Software Foundation, Inc.,
19  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
20  */
21 #include <linux/mtd/gpmi-nand.h>
22 #include <linux/delay.h>
23 #include <linux/clk.h>
24 #include <mach/mxs.h>
25 
26 #include "gpmi-nand.h"
27 #include "gpmi-regs.h"
28 #include "bch-regs.h"
29 
30 struct timing_threshod timing_default_threshold = {
31 	.max_data_setup_cycles       = (BM_GPMI_TIMING0_DATA_SETUP >>
32 						BP_GPMI_TIMING0_DATA_SETUP),
33 	.internal_data_setup_in_ns   = 0,
34 	.max_sample_delay_factor     = (BM_GPMI_CTRL1_RDN_DELAY >>
35 						BP_GPMI_CTRL1_RDN_DELAY),
36 	.max_dll_clock_period_in_ns  = 32,
37 	.max_dll_delay_in_ns         = 16,
38 };
39 
40 /*
41  * Clear the bit and poll it cleared.  This is usually called with
42  * a reset address and mask being either SFTRST(bit 31) or CLKGATE
43  * (bit 30).
44  */
clear_poll_bit(void __iomem * addr,u32 mask)45 static int clear_poll_bit(void __iomem *addr, u32 mask)
46 {
47 	int timeout = 0x400;
48 
49 	/* clear the bit */
50 	__mxs_clrl(mask, addr);
51 
52 	/*
53 	 * SFTRST needs 3 GPMI clocks to settle, the reference manual
54 	 * recommends to wait 1us.
55 	 */
56 	udelay(1);
57 
58 	/* poll the bit becoming clear */
59 	while ((readl(addr) & mask) && --timeout)
60 		/* nothing */;
61 
62 	return !timeout;
63 }
64 
65 #define MODULE_CLKGATE		(1 << 30)
66 #define MODULE_SFTRST		(1 << 31)
67 /*
68  * The current mxs_reset_block() will do two things:
69  *  [1] enable the module.
70  *  [2] reset the module.
71  *
72  * In most of the cases, it's ok.
73  * But in MX23, there is a hardware bug in the BCH block (see erratum #2847).
74  * If you try to soft reset the BCH block, it becomes unusable until
75  * the next hard reset. This case occurs in the NAND boot mode. When the board
76  * boots by NAND, the ROM of the chip will initialize the BCH blocks itself.
77  * So If the driver tries to reset the BCH again, the BCH will not work anymore.
78  * You will see a DMA timeout in this case. The bug has been fixed
79  * in the following chips, such as MX28.
80  *
81  * To avoid this bug, just add a new parameter `just_enable` for
82  * the mxs_reset_block(), and rewrite it here.
83  */
gpmi_reset_block(void __iomem * reset_addr,bool just_enable)84 static int gpmi_reset_block(void __iomem *reset_addr, bool just_enable)
85 {
86 	int ret;
87 	int timeout = 0x400;
88 
89 	/* clear and poll SFTRST */
90 	ret = clear_poll_bit(reset_addr, MODULE_SFTRST);
91 	if (unlikely(ret))
92 		goto error;
93 
94 	/* clear CLKGATE */
95 	__mxs_clrl(MODULE_CLKGATE, reset_addr);
96 
97 	if (!just_enable) {
98 		/* set SFTRST to reset the block */
99 		__mxs_setl(MODULE_SFTRST, reset_addr);
100 		udelay(1);
101 
102 		/* poll CLKGATE becoming set */
103 		while ((!(readl(reset_addr) & MODULE_CLKGATE)) && --timeout)
104 			/* nothing */;
105 		if (unlikely(!timeout))
106 			goto error;
107 	}
108 
109 	/* clear and poll SFTRST */
110 	ret = clear_poll_bit(reset_addr, MODULE_SFTRST);
111 	if (unlikely(ret))
112 		goto error;
113 
114 	/* clear and poll CLKGATE */
115 	ret = clear_poll_bit(reset_addr, MODULE_CLKGATE);
116 	if (unlikely(ret))
117 		goto error;
118 
119 	return 0;
120 
121 error:
122 	pr_err("%s(%p): module reset timeout\n", __func__, reset_addr);
123 	return -ETIMEDOUT;
124 }
125 
gpmi_init(struct gpmi_nand_data * this)126 int gpmi_init(struct gpmi_nand_data *this)
127 {
128 	struct resources *r = &this->resources;
129 	int ret;
130 
131 	ret = clk_prepare_enable(r->clock);
132 	if (ret)
133 		goto err_out;
134 	ret = gpmi_reset_block(r->gpmi_regs, false);
135 	if (ret)
136 		goto err_out;
137 
138 	/* Choose NAND mode. */
139 	writel(BM_GPMI_CTRL1_GPMI_MODE, r->gpmi_regs + HW_GPMI_CTRL1_CLR);
140 
141 	/* Set the IRQ polarity. */
142 	writel(BM_GPMI_CTRL1_ATA_IRQRDY_POLARITY,
143 				r->gpmi_regs + HW_GPMI_CTRL1_SET);
144 
145 	/* Disable Write-Protection. */
146 	writel(BM_GPMI_CTRL1_DEV_RESET, r->gpmi_regs + HW_GPMI_CTRL1_SET);
147 
148 	/* Select BCH ECC. */
149 	writel(BM_GPMI_CTRL1_BCH_MODE, r->gpmi_regs + HW_GPMI_CTRL1_SET);
150 
151 	clk_disable_unprepare(r->clock);
152 	return 0;
153 err_out:
154 	return ret;
155 }
156 
157 /* This function is very useful. It is called only when the bug occur. */
gpmi_dump_info(struct gpmi_nand_data * this)158 void gpmi_dump_info(struct gpmi_nand_data *this)
159 {
160 	struct resources *r = &this->resources;
161 	struct bch_geometry *geo = &this->bch_geometry;
162 	u32 reg;
163 	int i;
164 
165 	pr_err("Show GPMI registers :\n");
166 	for (i = 0; i <= HW_GPMI_DEBUG / 0x10 + 1; i++) {
167 		reg = readl(r->gpmi_regs + i * 0x10);
168 		pr_err("offset 0x%.3x : 0x%.8x\n", i * 0x10, reg);
169 	}
170 
171 	/* start to print out the BCH info */
172 	pr_err("BCH Geometry :\n");
173 	pr_err("GF length              : %u\n", geo->gf_len);
174 	pr_err("ECC Strength           : %u\n", geo->ecc_strength);
175 	pr_err("Page Size in Bytes     : %u\n", geo->page_size);
176 	pr_err("Metadata Size in Bytes : %u\n", geo->metadata_size);
177 	pr_err("ECC Chunk Size in Bytes: %u\n", geo->ecc_chunk_size);
178 	pr_err("ECC Chunk Count        : %u\n", geo->ecc_chunk_count);
179 	pr_err("Payload Size in Bytes  : %u\n", geo->payload_size);
180 	pr_err("Auxiliary Size in Bytes: %u\n", geo->auxiliary_size);
181 	pr_err("Auxiliary Status Offset: %u\n", geo->auxiliary_status_offset);
182 	pr_err("Block Mark Byte Offset : %u\n", geo->block_mark_byte_offset);
183 	pr_err("Block Mark Bit Offset  : %u\n", geo->block_mark_bit_offset);
184 }
185 
186 /* Configures the geometry for BCH.  */
bch_set_geometry(struct gpmi_nand_data * this)187 int bch_set_geometry(struct gpmi_nand_data *this)
188 {
189 	struct resources *r = &this->resources;
190 	struct bch_geometry *bch_geo = &this->bch_geometry;
191 	unsigned int block_count;
192 	unsigned int block_size;
193 	unsigned int metadata_size;
194 	unsigned int ecc_strength;
195 	unsigned int page_size;
196 	int ret;
197 
198 	if (common_nfc_set_geometry(this))
199 		return !0;
200 
201 	block_count   = bch_geo->ecc_chunk_count - 1;
202 	block_size    = bch_geo->ecc_chunk_size;
203 	metadata_size = bch_geo->metadata_size;
204 	ecc_strength  = bch_geo->ecc_strength >> 1;
205 	page_size     = bch_geo->page_size;
206 
207 	ret = clk_prepare_enable(r->clock);
208 	if (ret)
209 		goto err_out;
210 
211 	/*
212 	* Due to erratum #2847 of the MX23, the BCH cannot be soft reset on this
213 	* chip, otherwise it will lock up. So we skip resetting BCH on the MX23.
214 	* On the other hand, the MX28 needs the reset, because one case has been
215 	* seen where the BCH produced ECC errors constantly after 10000
216 	* consecutive reboots. The latter case has not been seen on the MX23 yet,
217 	* still we don't know if it could happen there as well.
218 	*/
219 	ret = gpmi_reset_block(r->bch_regs, GPMI_IS_MX23(this));
220 	if (ret)
221 		goto err_out;
222 
223 	/* Configure layout 0. */
224 	writel(BF_BCH_FLASH0LAYOUT0_NBLOCKS(block_count)
225 			| BF_BCH_FLASH0LAYOUT0_META_SIZE(metadata_size)
226 			| BF_BCH_FLASH0LAYOUT0_ECC0(ecc_strength)
227 			| BF_BCH_FLASH0LAYOUT0_DATA0_SIZE(block_size),
228 			r->bch_regs + HW_BCH_FLASH0LAYOUT0);
229 
230 	writel(BF_BCH_FLASH0LAYOUT1_PAGE_SIZE(page_size)
231 			| BF_BCH_FLASH0LAYOUT1_ECCN(ecc_strength)
232 			| BF_BCH_FLASH0LAYOUT1_DATAN_SIZE(block_size),
233 			r->bch_regs + HW_BCH_FLASH0LAYOUT1);
234 
235 	/* Set *all* chip selects to use layout 0. */
236 	writel(0, r->bch_regs + HW_BCH_LAYOUTSELECT);
237 
238 	/* Enable interrupts. */
239 	writel(BM_BCH_CTRL_COMPLETE_IRQ_EN,
240 				r->bch_regs + HW_BCH_CTRL_SET);
241 
242 	clk_disable_unprepare(r->clock);
243 	return 0;
244 err_out:
245 	return ret;
246 }
247 
248 /* Converts time in nanoseconds to cycles. */
ns_to_cycles(unsigned int time,unsigned int period,unsigned int min)249 static unsigned int ns_to_cycles(unsigned int time,
250 			unsigned int period, unsigned int min)
251 {
252 	unsigned int k;
253 
254 	k = (time + period - 1) / period;
255 	return max(k, min);
256 }
257 
258 /* Apply timing to current hardware conditions. */
gpmi_nfc_compute_hardware_timing(struct gpmi_nand_data * this,struct gpmi_nfc_hardware_timing * hw)259 static int gpmi_nfc_compute_hardware_timing(struct gpmi_nand_data *this,
260 					struct gpmi_nfc_hardware_timing *hw)
261 {
262 	struct gpmi_nand_platform_data *pdata = this->pdata;
263 	struct timing_threshod *nfc = &timing_default_threshold;
264 	struct nand_chip *nand = &this->nand;
265 	struct nand_timing target = this->timing;
266 	bool improved_timing_is_available;
267 	unsigned long clock_frequency_in_hz;
268 	unsigned int clock_period_in_ns;
269 	bool dll_use_half_periods;
270 	unsigned int dll_delay_shift;
271 	unsigned int max_sample_delay_in_ns;
272 	unsigned int address_setup_in_cycles;
273 	unsigned int data_setup_in_ns;
274 	unsigned int data_setup_in_cycles;
275 	unsigned int data_hold_in_cycles;
276 	int ideal_sample_delay_in_ns;
277 	unsigned int sample_delay_factor;
278 	int tEYE;
279 	unsigned int min_prop_delay_in_ns = pdata->min_prop_delay_in_ns;
280 	unsigned int max_prop_delay_in_ns = pdata->max_prop_delay_in_ns;
281 
282 	/*
283 	 * If there are multiple chips, we need to relax the timings to allow
284 	 * for signal distortion due to higher capacitance.
285 	 */
286 	if (nand->numchips > 2) {
287 		target.data_setup_in_ns    += 10;
288 		target.data_hold_in_ns     += 10;
289 		target.address_setup_in_ns += 10;
290 	} else if (nand->numchips > 1) {
291 		target.data_setup_in_ns    += 5;
292 		target.data_hold_in_ns     += 5;
293 		target.address_setup_in_ns += 5;
294 	}
295 
296 	/* Check if improved timing information is available. */
297 	improved_timing_is_available =
298 		(target.tREA_in_ns  >= 0) &&
299 		(target.tRLOH_in_ns >= 0) &&
300 		(target.tRHOH_in_ns >= 0) ;
301 
302 	/* Inspect the clock. */
303 	clock_frequency_in_hz = nfc->clock_frequency_in_hz;
304 	clock_period_in_ns    = 1000000000 / clock_frequency_in_hz;
305 
306 	/*
307 	 * The NFC quantizes setup and hold parameters in terms of clock cycles.
308 	 * Here, we quantize the setup and hold timing parameters to the
309 	 * next-highest clock period to make sure we apply at least the
310 	 * specified times.
311 	 *
312 	 * For data setup and data hold, the hardware interprets a value of zero
313 	 * as the largest possible delay. This is not what's intended by a zero
314 	 * in the input parameter, so we impose a minimum of one cycle.
315 	 */
316 	data_setup_in_cycles    = ns_to_cycles(target.data_setup_in_ns,
317 							clock_period_in_ns, 1);
318 	data_hold_in_cycles     = ns_to_cycles(target.data_hold_in_ns,
319 							clock_period_in_ns, 1);
320 	address_setup_in_cycles = ns_to_cycles(target.address_setup_in_ns,
321 							clock_period_in_ns, 0);
322 
323 	/*
324 	 * The clock's period affects the sample delay in a number of ways:
325 	 *
326 	 * (1) The NFC HAL tells us the maximum clock period the sample delay
327 	 *     DLL can tolerate. If the clock period is greater than half that
328 	 *     maximum, we must configure the DLL to be driven by half periods.
329 	 *
330 	 * (2) We need to convert from an ideal sample delay, in ns, to a
331 	 *     "sample delay factor," which the NFC uses. This factor depends on
332 	 *     whether we're driving the DLL with full or half periods.
333 	 *     Paraphrasing the reference manual:
334 	 *
335 	 *         AD = SDF x 0.125 x RP
336 	 *
337 	 * where:
338 	 *
339 	 *     AD   is the applied delay, in ns.
340 	 *     SDF  is the sample delay factor, which is dimensionless.
341 	 *     RP   is the reference period, in ns, which is a full clock period
342 	 *          if the DLL is being driven by full periods, or half that if
343 	 *          the DLL is being driven by half periods.
344 	 *
345 	 * Let's re-arrange this in a way that's more useful to us:
346 	 *
347 	 *                        8
348 	 *         SDF  =  AD x ----
349 	 *                       RP
350 	 *
351 	 * The reference period is either the clock period or half that, so this
352 	 * is:
353 	 *
354 	 *                        8       AD x DDF
355 	 *         SDF  =  AD x -----  =  --------
356 	 *                      f x P        P
357 	 *
358 	 * where:
359 	 *
360 	 *       f  is 1 or 1/2, depending on how we're driving the DLL.
361 	 *       P  is the clock period.
362 	 *     DDF  is the DLL Delay Factor, a dimensionless value that
363 	 *          incorporates all the constants in the conversion.
364 	 *
365 	 * DDF will be either 8 or 16, both of which are powers of two. We can
366 	 * reduce the cost of this conversion by using bit shifts instead of
367 	 * multiplication or division. Thus:
368 	 *
369 	 *                 AD << DDS
370 	 *         SDF  =  ---------
371 	 *                     P
372 	 *
373 	 *     or
374 	 *
375 	 *         AD  =  (SDF >> DDS) x P
376 	 *
377 	 * where:
378 	 *
379 	 *     DDS  is the DLL Delay Shift, the logarithm to base 2 of the DDF.
380 	 */
381 	if (clock_period_in_ns > (nfc->max_dll_clock_period_in_ns >> 1)) {
382 		dll_use_half_periods = true;
383 		dll_delay_shift      = 3 + 1;
384 	} else {
385 		dll_use_half_periods = false;
386 		dll_delay_shift      = 3;
387 	}
388 
389 	/*
390 	 * Compute the maximum sample delay the NFC allows, under current
391 	 * conditions. If the clock is running too slowly, no sample delay is
392 	 * possible.
393 	 */
394 	if (clock_period_in_ns > nfc->max_dll_clock_period_in_ns)
395 		max_sample_delay_in_ns = 0;
396 	else {
397 		/*
398 		 * Compute the delay implied by the largest sample delay factor
399 		 * the NFC allows.
400 		 */
401 		max_sample_delay_in_ns =
402 			(nfc->max_sample_delay_factor * clock_period_in_ns) >>
403 								dll_delay_shift;
404 
405 		/*
406 		 * Check if the implied sample delay larger than the NFC
407 		 * actually allows.
408 		 */
409 		if (max_sample_delay_in_ns > nfc->max_dll_delay_in_ns)
410 			max_sample_delay_in_ns = nfc->max_dll_delay_in_ns;
411 	}
412 
413 	/*
414 	 * Check if improved timing information is available. If not, we have to
415 	 * use a less-sophisticated algorithm.
416 	 */
417 	if (!improved_timing_is_available) {
418 		/*
419 		 * Fold the read setup time required by the NFC into the ideal
420 		 * sample delay.
421 		 */
422 		ideal_sample_delay_in_ns = target.gpmi_sample_delay_in_ns +
423 						nfc->internal_data_setup_in_ns;
424 
425 		/*
426 		 * The ideal sample delay may be greater than the maximum
427 		 * allowed by the NFC. If so, we can trade off sample delay time
428 		 * for more data setup time.
429 		 *
430 		 * In each iteration of the following loop, we add a cycle to
431 		 * the data setup time and subtract a corresponding amount from
432 		 * the sample delay until we've satisified the constraints or
433 		 * can't do any better.
434 		 */
435 		while ((ideal_sample_delay_in_ns > max_sample_delay_in_ns) &&
436 			(data_setup_in_cycles < nfc->max_data_setup_cycles)) {
437 
438 			data_setup_in_cycles++;
439 			ideal_sample_delay_in_ns -= clock_period_in_ns;
440 
441 			if (ideal_sample_delay_in_ns < 0)
442 				ideal_sample_delay_in_ns = 0;
443 
444 		}
445 
446 		/*
447 		 * Compute the sample delay factor that corresponds most closely
448 		 * to the ideal sample delay. If the result is too large for the
449 		 * NFC, use the maximum value.
450 		 *
451 		 * Notice that we use the ns_to_cycles function to compute the
452 		 * sample delay factor. We do this because the form of the
453 		 * computation is the same as that for calculating cycles.
454 		 */
455 		sample_delay_factor =
456 			ns_to_cycles(
457 				ideal_sample_delay_in_ns << dll_delay_shift,
458 							clock_period_in_ns, 0);
459 
460 		if (sample_delay_factor > nfc->max_sample_delay_factor)
461 			sample_delay_factor = nfc->max_sample_delay_factor;
462 
463 		/* Skip to the part where we return our results. */
464 		goto return_results;
465 	}
466 
467 	/*
468 	 * If control arrives here, we have more detailed timing information,
469 	 * so we can use a better algorithm.
470 	 */
471 
472 	/*
473 	 * Fold the read setup time required by the NFC into the maximum
474 	 * propagation delay.
475 	 */
476 	max_prop_delay_in_ns += nfc->internal_data_setup_in_ns;
477 
478 	/*
479 	 * Earlier, we computed the number of clock cycles required to satisfy
480 	 * the data setup time. Now, we need to know the actual nanoseconds.
481 	 */
482 	data_setup_in_ns = clock_period_in_ns * data_setup_in_cycles;
483 
484 	/*
485 	 * Compute tEYE, the width of the data eye when reading from the NAND
486 	 * Flash. The eye width is fundamentally determined by the data setup
487 	 * time, perturbed by propagation delays and some characteristics of the
488 	 * NAND Flash device.
489 	 *
490 	 * start of the eye = max_prop_delay + tREA
491 	 * end of the eye   = min_prop_delay + tRHOH + data_setup
492 	 */
493 	tEYE = (int)min_prop_delay_in_ns + (int)target.tRHOH_in_ns +
494 							(int)data_setup_in_ns;
495 
496 	tEYE -= (int)max_prop_delay_in_ns + (int)target.tREA_in_ns;
497 
498 	/*
499 	 * The eye must be open. If it's not, we can try to open it by
500 	 * increasing its main forcer, the data setup time.
501 	 *
502 	 * In each iteration of the following loop, we increase the data setup
503 	 * time by a single clock cycle. We do this until either the eye is
504 	 * open or we run into NFC limits.
505 	 */
506 	while ((tEYE <= 0) &&
507 			(data_setup_in_cycles < nfc->max_data_setup_cycles)) {
508 		/* Give a cycle to data setup. */
509 		data_setup_in_cycles++;
510 		/* Synchronize the data setup time with the cycles. */
511 		data_setup_in_ns += clock_period_in_ns;
512 		/* Adjust tEYE accordingly. */
513 		tEYE += clock_period_in_ns;
514 	}
515 
516 	/*
517 	 * When control arrives here, the eye is open. The ideal time to sample
518 	 * the data is in the center of the eye:
519 	 *
520 	 *     end of the eye + start of the eye
521 	 *     ---------------------------------  -  data_setup
522 	 *                    2
523 	 *
524 	 * After some algebra, this simplifies to the code immediately below.
525 	 */
526 	ideal_sample_delay_in_ns =
527 		((int)max_prop_delay_in_ns +
528 			(int)target.tREA_in_ns +
529 				(int)min_prop_delay_in_ns +
530 					(int)target.tRHOH_in_ns -
531 						(int)data_setup_in_ns) >> 1;
532 
533 	/*
534 	 * The following figure illustrates some aspects of a NAND Flash read:
535 	 *
536 	 *
537 	 *           __                   _____________________________________
538 	 * RDN         \_________________/
539 	 *
540 	 *                                         <---- tEYE ----->
541 	 *                                        /-----------------\
542 	 * Read Data ----------------------------<                   >---------
543 	 *                                        \-----------------/
544 	 *             ^                 ^                 ^              ^
545 	 *             |                 |                 |              |
546 	 *             |<--Data Setup -->|<--Delay Time -->|              |
547 	 *             |                 |                 |              |
548 	 *             |                 |                                |
549 	 *             |                 |<--   Quantized Delay Time   -->|
550 	 *             |                 |                                |
551 	 *
552 	 *
553 	 * We have some issues we must now address:
554 	 *
555 	 * (1) The *ideal* sample delay time must not be negative. If it is, we
556 	 *     jam it to zero.
557 	 *
558 	 * (2) The *ideal* sample delay time must not be greater than that
559 	 *     allowed by the NFC. If it is, we can increase the data setup
560 	 *     time, which will reduce the delay between the end of the data
561 	 *     setup and the center of the eye. It will also make the eye
562 	 *     larger, which might help with the next issue...
563 	 *
564 	 * (3) The *quantized* sample delay time must not fall either before the
565 	 *     eye opens or after it closes (the latter is the problem
566 	 *     illustrated in the above figure).
567 	 */
568 
569 	/* Jam a negative ideal sample delay to zero. */
570 	if (ideal_sample_delay_in_ns < 0)
571 		ideal_sample_delay_in_ns = 0;
572 
573 	/*
574 	 * Extend the data setup as needed to reduce the ideal sample delay
575 	 * below the maximum permitted by the NFC.
576 	 */
577 	while ((ideal_sample_delay_in_ns > max_sample_delay_in_ns) &&
578 			(data_setup_in_cycles < nfc->max_data_setup_cycles)) {
579 
580 		/* Give a cycle to data setup. */
581 		data_setup_in_cycles++;
582 		/* Synchronize the data setup time with the cycles. */
583 		data_setup_in_ns += clock_period_in_ns;
584 		/* Adjust tEYE accordingly. */
585 		tEYE += clock_period_in_ns;
586 
587 		/*
588 		 * Decrease the ideal sample delay by one half cycle, to keep it
589 		 * in the middle of the eye.
590 		 */
591 		ideal_sample_delay_in_ns -= (clock_period_in_ns >> 1);
592 
593 		/* Jam a negative ideal sample delay to zero. */
594 		if (ideal_sample_delay_in_ns < 0)
595 			ideal_sample_delay_in_ns = 0;
596 	}
597 
598 	/*
599 	 * Compute the sample delay factor that corresponds to the ideal sample
600 	 * delay. If the result is too large, then use the maximum allowed
601 	 * value.
602 	 *
603 	 * Notice that we use the ns_to_cycles function to compute the sample
604 	 * delay factor. We do this because the form of the computation is the
605 	 * same as that for calculating cycles.
606 	 */
607 	sample_delay_factor =
608 		ns_to_cycles(ideal_sample_delay_in_ns << dll_delay_shift,
609 							clock_period_in_ns, 0);
610 
611 	if (sample_delay_factor > nfc->max_sample_delay_factor)
612 		sample_delay_factor = nfc->max_sample_delay_factor;
613 
614 	/*
615 	 * These macros conveniently encapsulate a computation we'll use to
616 	 * continuously evaluate whether or not the data sample delay is inside
617 	 * the eye.
618 	 */
619 	#define IDEAL_DELAY  ((int) ideal_sample_delay_in_ns)
620 
621 	#define QUANTIZED_DELAY  \
622 		((int) ((sample_delay_factor * clock_period_in_ns) >> \
623 							dll_delay_shift))
624 
625 	#define DELAY_ERROR  (abs(QUANTIZED_DELAY - IDEAL_DELAY))
626 
627 	#define SAMPLE_IS_NOT_WITHIN_THE_EYE  (DELAY_ERROR > (tEYE >> 1))
628 
629 	/*
630 	 * While the quantized sample time falls outside the eye, reduce the
631 	 * sample delay or extend the data setup to move the sampling point back
632 	 * toward the eye. Do not allow the number of data setup cycles to
633 	 * exceed the maximum allowed by the NFC.
634 	 */
635 	while (SAMPLE_IS_NOT_WITHIN_THE_EYE &&
636 			(data_setup_in_cycles < nfc->max_data_setup_cycles)) {
637 		/*
638 		 * If control arrives here, the quantized sample delay falls
639 		 * outside the eye. Check if it's before the eye opens, or after
640 		 * the eye closes.
641 		 */
642 		if (QUANTIZED_DELAY > IDEAL_DELAY) {
643 			/*
644 			 * If control arrives here, the quantized sample delay
645 			 * falls after the eye closes. Decrease the quantized
646 			 * delay time and then go back to re-evaluate.
647 			 */
648 			if (sample_delay_factor != 0)
649 				sample_delay_factor--;
650 			continue;
651 		}
652 
653 		/*
654 		 * If control arrives here, the quantized sample delay falls
655 		 * before the eye opens. Shift the sample point by increasing
656 		 * data setup time. This will also make the eye larger.
657 		 */
658 
659 		/* Give a cycle to data setup. */
660 		data_setup_in_cycles++;
661 		/* Synchronize the data setup time with the cycles. */
662 		data_setup_in_ns += clock_period_in_ns;
663 		/* Adjust tEYE accordingly. */
664 		tEYE += clock_period_in_ns;
665 
666 		/*
667 		 * Decrease the ideal sample delay by one half cycle, to keep it
668 		 * in the middle of the eye.
669 		 */
670 		ideal_sample_delay_in_ns -= (clock_period_in_ns >> 1);
671 
672 		/* ...and one less period for the delay time. */
673 		ideal_sample_delay_in_ns -= clock_period_in_ns;
674 
675 		/* Jam a negative ideal sample delay to zero. */
676 		if (ideal_sample_delay_in_ns < 0)
677 			ideal_sample_delay_in_ns = 0;
678 
679 		/*
680 		 * We have a new ideal sample delay, so re-compute the quantized
681 		 * delay.
682 		 */
683 		sample_delay_factor =
684 			ns_to_cycles(
685 				ideal_sample_delay_in_ns << dll_delay_shift,
686 							clock_period_in_ns, 0);
687 
688 		if (sample_delay_factor > nfc->max_sample_delay_factor)
689 			sample_delay_factor = nfc->max_sample_delay_factor;
690 	}
691 
692 	/* Control arrives here when we're ready to return our results. */
693 return_results:
694 	hw->data_setup_in_cycles    = data_setup_in_cycles;
695 	hw->data_hold_in_cycles     = data_hold_in_cycles;
696 	hw->address_setup_in_cycles = address_setup_in_cycles;
697 	hw->use_half_periods        = dll_use_half_periods;
698 	hw->sample_delay_factor     = sample_delay_factor;
699 
700 	/* Return success. */
701 	return 0;
702 }
703 
704 /* Begin the I/O */
gpmi_begin(struct gpmi_nand_data * this)705 void gpmi_begin(struct gpmi_nand_data *this)
706 {
707 	struct resources *r = &this->resources;
708 	struct timing_threshod *nfc = &timing_default_threshold;
709 	unsigned char  *gpmi_regs = r->gpmi_regs;
710 	unsigned int   clock_period_in_ns;
711 	uint32_t       reg;
712 	unsigned int   dll_wait_time_in_us;
713 	struct gpmi_nfc_hardware_timing  hw;
714 	int ret;
715 
716 	/* Enable the clock. */
717 	ret = clk_prepare_enable(r->clock);
718 	if (ret) {
719 		pr_err("We failed in enable the clk\n");
720 		goto err_out;
721 	}
722 
723 	/* set ready/busy timeout */
724 	writel(0x500 << BP_GPMI_TIMING1_BUSY_TIMEOUT,
725 		gpmi_regs + HW_GPMI_TIMING1);
726 
727 	/* Get the timing information we need. */
728 	nfc->clock_frequency_in_hz = clk_get_rate(r->clock);
729 	clock_period_in_ns = 1000000000 / nfc->clock_frequency_in_hz;
730 
731 	gpmi_nfc_compute_hardware_timing(this, &hw);
732 
733 	/* Set up all the simple timing parameters. */
734 	reg = BF_GPMI_TIMING0_ADDRESS_SETUP(hw.address_setup_in_cycles) |
735 		BF_GPMI_TIMING0_DATA_HOLD(hw.data_hold_in_cycles)         |
736 		BF_GPMI_TIMING0_DATA_SETUP(hw.data_setup_in_cycles)       ;
737 
738 	writel(reg, gpmi_regs + HW_GPMI_TIMING0);
739 
740 	/*
741 	 * DLL_ENABLE must be set to 0 when setting RDN_DELAY or HALF_PERIOD.
742 	 */
743 	writel(BM_GPMI_CTRL1_DLL_ENABLE, gpmi_regs + HW_GPMI_CTRL1_CLR);
744 
745 	/* Clear out the DLL control fields. */
746 	writel(BM_GPMI_CTRL1_RDN_DELAY,   gpmi_regs + HW_GPMI_CTRL1_CLR);
747 	writel(BM_GPMI_CTRL1_HALF_PERIOD, gpmi_regs + HW_GPMI_CTRL1_CLR);
748 
749 	/* If no sample delay is called for, return immediately. */
750 	if (!hw.sample_delay_factor)
751 		return;
752 
753 	/* Configure the HALF_PERIOD flag. */
754 	if (hw.use_half_periods)
755 		writel(BM_GPMI_CTRL1_HALF_PERIOD,
756 						gpmi_regs + HW_GPMI_CTRL1_SET);
757 
758 	/* Set the delay factor. */
759 	writel(BF_GPMI_CTRL1_RDN_DELAY(hw.sample_delay_factor),
760 						gpmi_regs + HW_GPMI_CTRL1_SET);
761 
762 	/* Enable the DLL. */
763 	writel(BM_GPMI_CTRL1_DLL_ENABLE, gpmi_regs + HW_GPMI_CTRL1_SET);
764 
765 	/*
766 	 * After we enable the GPMI DLL, we have to wait 64 clock cycles before
767 	 * we can use the GPMI.
768 	 *
769 	 * Calculate the amount of time we need to wait, in microseconds.
770 	 */
771 	dll_wait_time_in_us = (clock_period_in_ns * 64) / 1000;
772 
773 	if (!dll_wait_time_in_us)
774 		dll_wait_time_in_us = 1;
775 
776 	/* Wait for the DLL to settle. */
777 	udelay(dll_wait_time_in_us);
778 
779 err_out:
780 	return;
781 }
782 
gpmi_end(struct gpmi_nand_data * this)783 void gpmi_end(struct gpmi_nand_data *this)
784 {
785 	struct resources *r = &this->resources;
786 	clk_disable_unprepare(r->clock);
787 }
788 
789 /* Clears a BCH interrupt. */
gpmi_clear_bch(struct gpmi_nand_data * this)790 void gpmi_clear_bch(struct gpmi_nand_data *this)
791 {
792 	struct resources *r = &this->resources;
793 	writel(BM_BCH_CTRL_COMPLETE_IRQ, r->bch_regs + HW_BCH_CTRL_CLR);
794 }
795 
796 /* Returns the Ready/Busy status of the given chip. */
gpmi_is_ready(struct gpmi_nand_data * this,unsigned chip)797 int gpmi_is_ready(struct gpmi_nand_data *this, unsigned chip)
798 {
799 	struct resources *r = &this->resources;
800 	uint32_t mask = 0;
801 	uint32_t reg = 0;
802 
803 	if (GPMI_IS_MX23(this)) {
804 		mask = MX23_BM_GPMI_DEBUG_READY0 << chip;
805 		reg = readl(r->gpmi_regs + HW_GPMI_DEBUG);
806 	} else if (GPMI_IS_MX28(this)) {
807 		mask = MX28_BF_GPMI_STAT_READY_BUSY(1 << chip);
808 		reg = readl(r->gpmi_regs + HW_GPMI_STAT);
809 	} else
810 		pr_err("unknow arch.\n");
811 	return reg & mask;
812 }
813 
set_dma_type(struct gpmi_nand_data * this,enum dma_ops_type type)814 static inline void set_dma_type(struct gpmi_nand_data *this,
815 					enum dma_ops_type type)
816 {
817 	this->last_dma_type = this->dma_type;
818 	this->dma_type = type;
819 }
820 
gpmi_send_command(struct gpmi_nand_data * this)821 int gpmi_send_command(struct gpmi_nand_data *this)
822 {
823 	struct dma_chan *channel = get_dma_chan(this);
824 	struct dma_async_tx_descriptor *desc;
825 	struct scatterlist *sgl;
826 	int chip = this->current_chip;
827 	u32 pio[3];
828 
829 	/* [1] send out the PIO words */
830 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(BV_GPMI_CTRL0_COMMAND_MODE__WRITE)
831 		| BM_GPMI_CTRL0_WORD_LENGTH
832 		| BF_GPMI_CTRL0_CS(chip, this)
833 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
834 		| BF_GPMI_CTRL0_ADDRESS(BV_GPMI_CTRL0_ADDRESS__NAND_CLE)
835 		| BM_GPMI_CTRL0_ADDRESS_INCREMENT
836 		| BF_GPMI_CTRL0_XFER_COUNT(this->command_length);
837 	pio[1] = pio[2] = 0;
838 	desc = channel->device->device_prep_slave_sg(channel,
839 					(struct scatterlist *)pio,
840 					ARRAY_SIZE(pio), DMA_TRANS_NONE, 0);
841 	if (!desc) {
842 		pr_err("step 1 error\n");
843 		return -1;
844 	}
845 
846 	/* [2] send out the COMMAND + ADDRESS string stored in @buffer */
847 	sgl = &this->cmd_sgl;
848 
849 	sg_init_one(sgl, this->cmd_buffer, this->command_length);
850 	dma_map_sg(this->dev, sgl, 1, DMA_TO_DEVICE);
851 	desc = channel->device->device_prep_slave_sg(channel,
852 					sgl, 1, DMA_MEM_TO_DEV, 1);
853 	if (!desc) {
854 		pr_err("step 2 error\n");
855 		return -1;
856 	}
857 
858 	/* [3] submit the DMA */
859 	set_dma_type(this, DMA_FOR_COMMAND);
860 	return start_dma_without_bch_irq(this, desc);
861 }
862 
gpmi_send_data(struct gpmi_nand_data * this)863 int gpmi_send_data(struct gpmi_nand_data *this)
864 {
865 	struct dma_async_tx_descriptor *desc;
866 	struct dma_chan *channel = get_dma_chan(this);
867 	int chip = this->current_chip;
868 	uint32_t command_mode;
869 	uint32_t address;
870 	u32 pio[2];
871 
872 	/* [1] PIO */
873 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__WRITE;
874 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
875 
876 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
877 		| BM_GPMI_CTRL0_WORD_LENGTH
878 		| BF_GPMI_CTRL0_CS(chip, this)
879 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
880 		| BF_GPMI_CTRL0_ADDRESS(address)
881 		| BF_GPMI_CTRL0_XFER_COUNT(this->upper_len);
882 	pio[1] = 0;
883 	desc = channel->device->device_prep_slave_sg(channel,
884 					(struct scatterlist *)pio,
885 					ARRAY_SIZE(pio), DMA_TRANS_NONE, 0);
886 	if (!desc) {
887 		pr_err("step 1 error\n");
888 		return -1;
889 	}
890 
891 	/* [2] send DMA request */
892 	prepare_data_dma(this, DMA_TO_DEVICE);
893 	desc = channel->device->device_prep_slave_sg(channel, &this->data_sgl,
894 						1, DMA_MEM_TO_DEV, 1);
895 	if (!desc) {
896 		pr_err("step 2 error\n");
897 		return -1;
898 	}
899 	/* [3] submit the DMA */
900 	set_dma_type(this, DMA_FOR_WRITE_DATA);
901 	return start_dma_without_bch_irq(this, desc);
902 }
903 
gpmi_read_data(struct gpmi_nand_data * this)904 int gpmi_read_data(struct gpmi_nand_data *this)
905 {
906 	struct dma_async_tx_descriptor *desc;
907 	struct dma_chan *channel = get_dma_chan(this);
908 	int chip = this->current_chip;
909 	u32 pio[2];
910 
911 	/* [1] : send PIO */
912 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(BV_GPMI_CTRL0_COMMAND_MODE__READ)
913 		| BM_GPMI_CTRL0_WORD_LENGTH
914 		| BF_GPMI_CTRL0_CS(chip, this)
915 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
916 		| BF_GPMI_CTRL0_ADDRESS(BV_GPMI_CTRL0_ADDRESS__NAND_DATA)
917 		| BF_GPMI_CTRL0_XFER_COUNT(this->upper_len);
918 	pio[1] = 0;
919 	desc = channel->device->device_prep_slave_sg(channel,
920 					(struct scatterlist *)pio,
921 					ARRAY_SIZE(pio), DMA_TRANS_NONE, 0);
922 	if (!desc) {
923 		pr_err("step 1 error\n");
924 		return -1;
925 	}
926 
927 	/* [2] : send DMA request */
928 	prepare_data_dma(this, DMA_FROM_DEVICE);
929 	desc = channel->device->device_prep_slave_sg(channel, &this->data_sgl,
930 						1, DMA_DEV_TO_MEM, 1);
931 	if (!desc) {
932 		pr_err("step 2 error\n");
933 		return -1;
934 	}
935 
936 	/* [3] : submit the DMA */
937 	set_dma_type(this, DMA_FOR_READ_DATA);
938 	return start_dma_without_bch_irq(this, desc);
939 }
940 
gpmi_send_page(struct gpmi_nand_data * this,dma_addr_t payload,dma_addr_t auxiliary)941 int gpmi_send_page(struct gpmi_nand_data *this,
942 			dma_addr_t payload, dma_addr_t auxiliary)
943 {
944 	struct bch_geometry *geo = &this->bch_geometry;
945 	uint32_t command_mode;
946 	uint32_t address;
947 	uint32_t ecc_command;
948 	uint32_t buffer_mask;
949 	struct dma_async_tx_descriptor *desc;
950 	struct dma_chan *channel = get_dma_chan(this);
951 	int chip = this->current_chip;
952 	u32 pio[6];
953 
954 	/* A DMA descriptor that does an ECC page read. */
955 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__WRITE;
956 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
957 	ecc_command  = BV_GPMI_ECCCTRL_ECC_CMD__BCH_ENCODE;
958 	buffer_mask  = BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_PAGE |
959 				BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_AUXONLY;
960 
961 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
962 		| BM_GPMI_CTRL0_WORD_LENGTH
963 		| BF_GPMI_CTRL0_CS(chip, this)
964 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
965 		| BF_GPMI_CTRL0_ADDRESS(address)
966 		| BF_GPMI_CTRL0_XFER_COUNT(0);
967 	pio[1] = 0;
968 	pio[2] = BM_GPMI_ECCCTRL_ENABLE_ECC
969 		| BF_GPMI_ECCCTRL_ECC_CMD(ecc_command)
970 		| BF_GPMI_ECCCTRL_BUFFER_MASK(buffer_mask);
971 	pio[3] = geo->page_size;
972 	pio[4] = payload;
973 	pio[5] = auxiliary;
974 
975 	desc = channel->device->device_prep_slave_sg(channel,
976 					(struct scatterlist *)pio,
977 					ARRAY_SIZE(pio), DMA_TRANS_NONE, 0);
978 	if (!desc) {
979 		pr_err("step 2 error\n");
980 		return -1;
981 	}
982 	set_dma_type(this, DMA_FOR_WRITE_ECC_PAGE);
983 	return start_dma_with_bch_irq(this, desc);
984 }
985 
gpmi_read_page(struct gpmi_nand_data * this,dma_addr_t payload,dma_addr_t auxiliary)986 int gpmi_read_page(struct gpmi_nand_data *this,
987 				dma_addr_t payload, dma_addr_t auxiliary)
988 {
989 	struct bch_geometry *geo = &this->bch_geometry;
990 	uint32_t command_mode;
991 	uint32_t address;
992 	uint32_t ecc_command;
993 	uint32_t buffer_mask;
994 	struct dma_async_tx_descriptor *desc;
995 	struct dma_chan *channel = get_dma_chan(this);
996 	int chip = this->current_chip;
997 	u32 pio[6];
998 
999 	/* [1] Wait for the chip to report ready. */
1000 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__WAIT_FOR_READY;
1001 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
1002 
1003 	pio[0] =  BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
1004 		| BM_GPMI_CTRL0_WORD_LENGTH
1005 		| BF_GPMI_CTRL0_CS(chip, this)
1006 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
1007 		| BF_GPMI_CTRL0_ADDRESS(address)
1008 		| BF_GPMI_CTRL0_XFER_COUNT(0);
1009 	pio[1] = 0;
1010 	desc = channel->device->device_prep_slave_sg(channel,
1011 				(struct scatterlist *)pio, 2,
1012 				DMA_TRANS_NONE, 0);
1013 	if (!desc) {
1014 		pr_err("step 1 error\n");
1015 		return -1;
1016 	}
1017 
1018 	/* [2] Enable the BCH block and read. */
1019 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__READ;
1020 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
1021 	ecc_command  = BV_GPMI_ECCCTRL_ECC_CMD__BCH_DECODE;
1022 	buffer_mask  = BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_PAGE
1023 			| BV_GPMI_ECCCTRL_BUFFER_MASK__BCH_AUXONLY;
1024 
1025 	pio[0] =  BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
1026 		| BM_GPMI_CTRL0_WORD_LENGTH
1027 		| BF_GPMI_CTRL0_CS(chip, this)
1028 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
1029 		| BF_GPMI_CTRL0_ADDRESS(address)
1030 		| BF_GPMI_CTRL0_XFER_COUNT(geo->page_size);
1031 
1032 	pio[1] = 0;
1033 	pio[2] =  BM_GPMI_ECCCTRL_ENABLE_ECC
1034 		| BF_GPMI_ECCCTRL_ECC_CMD(ecc_command)
1035 		| BF_GPMI_ECCCTRL_BUFFER_MASK(buffer_mask);
1036 	pio[3] = geo->page_size;
1037 	pio[4] = payload;
1038 	pio[5] = auxiliary;
1039 	desc = channel->device->device_prep_slave_sg(channel,
1040 					(struct scatterlist *)pio,
1041 					ARRAY_SIZE(pio), DMA_TRANS_NONE, 1);
1042 	if (!desc) {
1043 		pr_err("step 2 error\n");
1044 		return -1;
1045 	}
1046 
1047 	/* [3] Disable the BCH block */
1048 	command_mode = BV_GPMI_CTRL0_COMMAND_MODE__WAIT_FOR_READY;
1049 	address      = BV_GPMI_CTRL0_ADDRESS__NAND_DATA;
1050 
1051 	pio[0] = BF_GPMI_CTRL0_COMMAND_MODE(command_mode)
1052 		| BM_GPMI_CTRL0_WORD_LENGTH
1053 		| BF_GPMI_CTRL0_CS(chip, this)
1054 		| BF_GPMI_CTRL0_LOCK_CS(LOCK_CS_ENABLE, this)
1055 		| BF_GPMI_CTRL0_ADDRESS(address)
1056 		| BF_GPMI_CTRL0_XFER_COUNT(geo->page_size);
1057 	pio[1] = 0;
1058 	desc = channel->device->device_prep_slave_sg(channel,
1059 				(struct scatterlist *)pio, 2,
1060 				DMA_TRANS_NONE, 1);
1061 	if (!desc) {
1062 		pr_err("step 3 error\n");
1063 		return -1;
1064 	}
1065 
1066 	/* [4] submit the DMA */
1067 	set_dma_type(this, DMA_FOR_READ_ECC_PAGE);
1068 	return start_dma_with_bch_irq(this, desc);
1069 }
1070