xref: /linux/drivers/spi/spi-fsl-qspi.c (revision 17cfcb68af3bc7d5e8ae08779b1853310a2949f3)
1 // SPDX-License-Identifier: GPL-2.0+
2 
3 /*
4  * Freescale QuadSPI driver.
5  *
6  * Copyright (C) 2013 Freescale Semiconductor, Inc.
7  * Copyright (C) 2018 Bootlin
8  * Copyright (C) 2018 exceet electronics GmbH
9  * Copyright (C) 2018 Kontron Electronics GmbH
10  *
11  * Transition to SPI MEM interface:
12  * Authors:
13  *     Boris Brezillon <bbrezillon@kernel.org>
14  *     Frieder Schrempf <frieder.schrempf@kontron.de>
15  *     Yogesh Gaur <yogeshnarayan.gaur@nxp.com>
16  *     Suresh Gupta <suresh.gupta@nxp.com>
17  *
18  * Based on the original fsl-quadspi.c spi-nor driver:
19  * Author: Freescale Semiconductor, Inc.
20  *
21  */
22 
23 #include <linux/bitops.h>
24 #include <linux/clk.h>
25 #include <linux/completion.h>
26 #include <linux/delay.h>
27 #include <linux/err.h>
28 #include <linux/errno.h>
29 #include <linux/interrupt.h>
30 #include <linux/io.h>
31 #include <linux/iopoll.h>
32 #include <linux/jiffies.h>
33 #include <linux/kernel.h>
34 #include <linux/module.h>
35 #include <linux/mutex.h>
36 #include <linux/of.h>
37 #include <linux/of_device.h>
38 #include <linux/platform_device.h>
39 #include <linux/pm_qos.h>
40 #include <linux/sizes.h>
41 
42 #include <linux/spi/spi.h>
43 #include <linux/spi/spi-mem.h>
44 
45 /*
46  * The driver only uses one single LUT entry, that is updated on
47  * each call of exec_op(). Index 0 is preset at boot with a basic
48  * read operation, so let's use the last entry (15).
49  */
50 #define	SEQID_LUT			15
51 
52 /* Registers used by the driver */
53 #define QUADSPI_MCR			0x00
54 #define QUADSPI_MCR_RESERVED_MASK	GENMASK(19, 16)
55 #define QUADSPI_MCR_MDIS_MASK		BIT(14)
56 #define QUADSPI_MCR_CLR_TXF_MASK	BIT(11)
57 #define QUADSPI_MCR_CLR_RXF_MASK	BIT(10)
58 #define QUADSPI_MCR_DDR_EN_MASK		BIT(7)
59 #define QUADSPI_MCR_END_CFG_MASK	GENMASK(3, 2)
60 #define QUADSPI_MCR_SWRSTHD_MASK	BIT(1)
61 #define QUADSPI_MCR_SWRSTSD_MASK	BIT(0)
62 
63 #define QUADSPI_IPCR			0x08
64 #define QUADSPI_IPCR_SEQID(x)		((x) << 24)
65 
66 #define QUADSPI_BUF3CR			0x1c
67 #define QUADSPI_BUF3CR_ALLMST_MASK	BIT(31)
68 #define QUADSPI_BUF3CR_ADATSZ(x)	((x) << 8)
69 #define QUADSPI_BUF3CR_ADATSZ_MASK	GENMASK(15, 8)
70 
71 #define QUADSPI_BFGENCR			0x20
72 #define QUADSPI_BFGENCR_SEQID(x)	((x) << 12)
73 
74 #define QUADSPI_BUF0IND			0x30
75 #define QUADSPI_BUF1IND			0x34
76 #define QUADSPI_BUF2IND			0x38
77 #define QUADSPI_SFAR			0x100
78 
79 #define QUADSPI_SMPR			0x108
80 #define QUADSPI_SMPR_DDRSMP_MASK	GENMASK(18, 16)
81 #define QUADSPI_SMPR_FSDLY_MASK		BIT(6)
82 #define QUADSPI_SMPR_FSPHS_MASK		BIT(5)
83 #define QUADSPI_SMPR_HSENA_MASK		BIT(0)
84 
85 #define QUADSPI_RBCT			0x110
86 #define QUADSPI_RBCT_WMRK_MASK		GENMASK(4, 0)
87 #define QUADSPI_RBCT_RXBRD_USEIPS	BIT(8)
88 
89 #define QUADSPI_TBDR			0x154
90 
91 #define QUADSPI_SR			0x15c
92 #define QUADSPI_SR_IP_ACC_MASK		BIT(1)
93 #define QUADSPI_SR_AHB_ACC_MASK		BIT(2)
94 
95 #define QUADSPI_FR			0x160
96 #define QUADSPI_FR_TFF_MASK		BIT(0)
97 
98 #define QUADSPI_SPTRCLR			0x16c
99 #define QUADSPI_SPTRCLR_IPPTRC		BIT(8)
100 #define QUADSPI_SPTRCLR_BFPTRC		BIT(0)
101 
102 #define QUADSPI_SFA1AD			0x180
103 #define QUADSPI_SFA2AD			0x184
104 #define QUADSPI_SFB1AD			0x188
105 #define QUADSPI_SFB2AD			0x18c
106 #define QUADSPI_RBDR(x)			(0x200 + ((x) * 4))
107 
108 #define QUADSPI_LUTKEY			0x300
109 #define QUADSPI_LUTKEY_VALUE		0x5AF05AF0
110 
111 #define QUADSPI_LCKCR			0x304
112 #define QUADSPI_LCKER_LOCK		BIT(0)
113 #define QUADSPI_LCKER_UNLOCK		BIT(1)
114 
115 #define QUADSPI_RSER			0x164
116 #define QUADSPI_RSER_TFIE		BIT(0)
117 
118 #define QUADSPI_LUT_BASE		0x310
119 #define QUADSPI_LUT_OFFSET		(SEQID_LUT * 4 * 4)
120 #define QUADSPI_LUT_REG(idx) \
121 	(QUADSPI_LUT_BASE + QUADSPI_LUT_OFFSET + (idx) * 4)
122 
123 /* Instruction set for the LUT register */
124 #define LUT_STOP		0
125 #define LUT_CMD			1
126 #define LUT_ADDR		2
127 #define LUT_DUMMY		3
128 #define LUT_MODE		4
129 #define LUT_MODE2		5
130 #define LUT_MODE4		6
131 #define LUT_FSL_READ		7
132 #define LUT_FSL_WRITE		8
133 #define LUT_JMP_ON_CS		9
134 #define LUT_ADDR_DDR		10
135 #define LUT_MODE_DDR		11
136 #define LUT_MODE2_DDR		12
137 #define LUT_MODE4_DDR		13
138 #define LUT_FSL_READ_DDR	14
139 #define LUT_FSL_WRITE_DDR	15
140 #define LUT_DATA_LEARN		16
141 
142 /*
143  * The PAD definitions for LUT register.
144  *
145  * The pad stands for the number of IO lines [0:3].
146  * For example, the quad read needs four IO lines,
147  * so you should use LUT_PAD(4).
148  */
149 #define LUT_PAD(x) (fls(x) - 1)
150 
151 /*
152  * Macro for constructing the LUT entries with the following
153  * register layout:
154  *
155  *  ---------------------------------------------------
156  *  | INSTR1 | PAD1 | OPRND1 | INSTR0 | PAD0 | OPRND0 |
157  *  ---------------------------------------------------
158  */
159 #define LUT_DEF(idx, ins, pad, opr)					\
160 	((((ins) << 10) | ((pad) << 8) | (opr)) << (((idx) % 2) * 16))
161 
162 /* Controller needs driver to swap endianness */
163 #define QUADSPI_QUIRK_SWAP_ENDIAN	BIT(0)
164 
165 /* Controller needs 4x internal clock */
166 #define QUADSPI_QUIRK_4X_INT_CLK	BIT(1)
167 
168 /*
169  * TKT253890, the controller needs the driver to fill the txfifo with
170  * 16 bytes at least to trigger a data transfer, even though the extra
171  * data won't be transferred.
172  */
173 #define QUADSPI_QUIRK_TKT253890		BIT(2)
174 
175 /* TKT245618, the controller cannot wake up from wait mode */
176 #define QUADSPI_QUIRK_TKT245618		BIT(3)
177 
178 /*
179  * Controller adds QSPI_AMBA_BASE (base address of the mapped memory)
180  * internally. No need to add it when setting SFXXAD and SFAR registers
181  */
182 #define QUADSPI_QUIRK_BASE_INTERNAL	BIT(4)
183 
184 struct fsl_qspi_devtype_data {
185 	unsigned int rxfifo;
186 	unsigned int txfifo;
187 	unsigned int ahb_buf_size;
188 	unsigned int quirks;
189 	bool little_endian;
190 };
191 
192 static const struct fsl_qspi_devtype_data vybrid_data = {
193 	.rxfifo = SZ_128,
194 	.txfifo = SZ_64,
195 	.ahb_buf_size = SZ_1K,
196 	.quirks = QUADSPI_QUIRK_SWAP_ENDIAN,
197 	.little_endian = true,
198 };
199 
200 static const struct fsl_qspi_devtype_data imx6sx_data = {
201 	.rxfifo = SZ_128,
202 	.txfifo = SZ_512,
203 	.ahb_buf_size = SZ_1K,
204 	.quirks = QUADSPI_QUIRK_4X_INT_CLK | QUADSPI_QUIRK_TKT245618,
205 	.little_endian = true,
206 };
207 
208 static const struct fsl_qspi_devtype_data imx7d_data = {
209 	.rxfifo = SZ_128,
210 	.txfifo = SZ_512,
211 	.ahb_buf_size = SZ_1K,
212 	.quirks = QUADSPI_QUIRK_TKT253890 | QUADSPI_QUIRK_4X_INT_CLK,
213 	.little_endian = true,
214 };
215 
216 static const struct fsl_qspi_devtype_data imx6ul_data = {
217 	.rxfifo = SZ_128,
218 	.txfifo = SZ_512,
219 	.ahb_buf_size = SZ_1K,
220 	.quirks = QUADSPI_QUIRK_TKT253890 | QUADSPI_QUIRK_4X_INT_CLK,
221 	.little_endian = true,
222 };
223 
224 static const struct fsl_qspi_devtype_data ls1021a_data = {
225 	.rxfifo = SZ_128,
226 	.txfifo = SZ_64,
227 	.ahb_buf_size = SZ_1K,
228 	.quirks = 0,
229 	.little_endian = false,
230 };
231 
232 static const struct fsl_qspi_devtype_data ls2080a_data = {
233 	.rxfifo = SZ_128,
234 	.txfifo = SZ_64,
235 	.ahb_buf_size = SZ_1K,
236 	.quirks = QUADSPI_QUIRK_TKT253890 | QUADSPI_QUIRK_BASE_INTERNAL,
237 	.little_endian = true,
238 };
239 
240 struct fsl_qspi {
241 	void __iomem *iobase;
242 	void __iomem *ahb_addr;
243 	u32 memmap_phy;
244 	struct clk *clk, *clk_en;
245 	struct device *dev;
246 	struct completion c;
247 	const struct fsl_qspi_devtype_data *devtype_data;
248 	struct mutex lock;
249 	struct pm_qos_request pm_qos_req;
250 	int selected;
251 };
252 
253 static inline int needs_swap_endian(struct fsl_qspi *q)
254 {
255 	return q->devtype_data->quirks & QUADSPI_QUIRK_SWAP_ENDIAN;
256 }
257 
258 static inline int needs_4x_clock(struct fsl_qspi *q)
259 {
260 	return q->devtype_data->quirks & QUADSPI_QUIRK_4X_INT_CLK;
261 }
262 
263 static inline int needs_fill_txfifo(struct fsl_qspi *q)
264 {
265 	return q->devtype_data->quirks & QUADSPI_QUIRK_TKT253890;
266 }
267 
268 static inline int needs_wakeup_wait_mode(struct fsl_qspi *q)
269 {
270 	return q->devtype_data->quirks & QUADSPI_QUIRK_TKT245618;
271 }
272 
273 static inline int needs_amba_base_offset(struct fsl_qspi *q)
274 {
275 	return !(q->devtype_data->quirks & QUADSPI_QUIRK_BASE_INTERNAL);
276 }
277 
278 /*
279  * An IC bug makes it necessary to rearrange the 32-bit data.
280  * Later chips, such as IMX6SLX, have fixed this bug.
281  */
282 static inline u32 fsl_qspi_endian_xchg(struct fsl_qspi *q, u32 a)
283 {
284 	return needs_swap_endian(q) ? __swab32(a) : a;
285 }
286 
287 /*
288  * R/W functions for big- or little-endian registers:
289  * The QSPI controller's endianness is independent of
290  * the CPU core's endianness. So far, although the CPU
291  * core is little-endian the QSPI controller can use
292  * big-endian or little-endian.
293  */
294 static void qspi_writel(struct fsl_qspi *q, u32 val, void __iomem *addr)
295 {
296 	if (q->devtype_data->little_endian)
297 		iowrite32(val, addr);
298 	else
299 		iowrite32be(val, addr);
300 }
301 
302 static u32 qspi_readl(struct fsl_qspi *q, void __iomem *addr)
303 {
304 	if (q->devtype_data->little_endian)
305 		return ioread32(addr);
306 
307 	return ioread32be(addr);
308 }
309 
310 static irqreturn_t fsl_qspi_irq_handler(int irq, void *dev_id)
311 {
312 	struct fsl_qspi *q = dev_id;
313 	u32 reg;
314 
315 	/* clear interrupt */
316 	reg = qspi_readl(q, q->iobase + QUADSPI_FR);
317 	qspi_writel(q, reg, q->iobase + QUADSPI_FR);
318 
319 	if (reg & QUADSPI_FR_TFF_MASK)
320 		complete(&q->c);
321 
322 	dev_dbg(q->dev, "QUADSPI_FR : 0x%.8x:0x%.8x\n", 0, reg);
323 	return IRQ_HANDLED;
324 }
325 
326 static int fsl_qspi_check_buswidth(struct fsl_qspi *q, u8 width)
327 {
328 	switch (width) {
329 	case 1:
330 	case 2:
331 	case 4:
332 		return 0;
333 	}
334 
335 	return -ENOTSUPP;
336 }
337 
338 static bool fsl_qspi_supports_op(struct spi_mem *mem,
339 				 const struct spi_mem_op *op)
340 {
341 	struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
342 	int ret;
343 
344 	ret = fsl_qspi_check_buswidth(q, op->cmd.buswidth);
345 
346 	if (op->addr.nbytes)
347 		ret |= fsl_qspi_check_buswidth(q, op->addr.buswidth);
348 
349 	if (op->dummy.nbytes)
350 		ret |= fsl_qspi_check_buswidth(q, op->dummy.buswidth);
351 
352 	if (op->data.nbytes)
353 		ret |= fsl_qspi_check_buswidth(q, op->data.buswidth);
354 
355 	if (ret)
356 		return false;
357 
358 	/*
359 	 * The number of instructions needed for the op, needs
360 	 * to fit into a single LUT entry.
361 	 */
362 	if (op->addr.nbytes +
363 	   (op->dummy.nbytes ? 1:0) +
364 	   (op->data.nbytes ? 1:0) > 6)
365 		return false;
366 
367 	/* Max 64 dummy clock cycles supported */
368 	if (op->dummy.nbytes &&
369 	    (op->dummy.nbytes * 8 / op->dummy.buswidth > 64))
370 		return false;
371 
372 	/* Max data length, check controller limits and alignment */
373 	if (op->data.dir == SPI_MEM_DATA_IN &&
374 	    (op->data.nbytes > q->devtype_data->ahb_buf_size ||
375 	     (op->data.nbytes > q->devtype_data->rxfifo - 4 &&
376 	      !IS_ALIGNED(op->data.nbytes, 8))))
377 		return false;
378 
379 	if (op->data.dir == SPI_MEM_DATA_OUT &&
380 	    op->data.nbytes > q->devtype_data->txfifo)
381 		return false;
382 
383 	return true;
384 }
385 
386 static void fsl_qspi_prepare_lut(struct fsl_qspi *q,
387 				 const struct spi_mem_op *op)
388 {
389 	void __iomem *base = q->iobase;
390 	u32 lutval[4] = {};
391 	int lutidx = 1, i;
392 
393 	lutval[0] |= LUT_DEF(0, LUT_CMD, LUT_PAD(op->cmd.buswidth),
394 			     op->cmd.opcode);
395 
396 	/*
397 	 * For some unknown reason, using LUT_ADDR doesn't work in some
398 	 * cases (at least with only one byte long addresses), so
399 	 * let's use LUT_MODE to write the address bytes one by one
400 	 */
401 	for (i = 0; i < op->addr.nbytes; i++) {
402 		u8 addrbyte = op->addr.val >> (8 * (op->addr.nbytes - i - 1));
403 
404 		lutval[lutidx / 2] |= LUT_DEF(lutidx, LUT_MODE,
405 					      LUT_PAD(op->addr.buswidth),
406 					      addrbyte);
407 		lutidx++;
408 	}
409 
410 	if (op->dummy.nbytes) {
411 		lutval[lutidx / 2] |= LUT_DEF(lutidx, LUT_DUMMY,
412 					      LUT_PAD(op->dummy.buswidth),
413 					      op->dummy.nbytes * 8 /
414 					      op->dummy.buswidth);
415 		lutidx++;
416 	}
417 
418 	if (op->data.nbytes) {
419 		lutval[lutidx / 2] |= LUT_DEF(lutidx,
420 					      op->data.dir == SPI_MEM_DATA_IN ?
421 					      LUT_FSL_READ : LUT_FSL_WRITE,
422 					      LUT_PAD(op->data.buswidth),
423 					      0);
424 		lutidx++;
425 	}
426 
427 	lutval[lutidx / 2] |= LUT_DEF(lutidx, LUT_STOP, 0, 0);
428 
429 	/* unlock LUT */
430 	qspi_writel(q, QUADSPI_LUTKEY_VALUE, q->iobase + QUADSPI_LUTKEY);
431 	qspi_writel(q, QUADSPI_LCKER_UNLOCK, q->iobase + QUADSPI_LCKCR);
432 
433 	/* fill LUT */
434 	for (i = 0; i < ARRAY_SIZE(lutval); i++)
435 		qspi_writel(q, lutval[i], base + QUADSPI_LUT_REG(i));
436 
437 	/* lock LUT */
438 	qspi_writel(q, QUADSPI_LUTKEY_VALUE, q->iobase + QUADSPI_LUTKEY);
439 	qspi_writel(q, QUADSPI_LCKER_LOCK, q->iobase + QUADSPI_LCKCR);
440 }
441 
442 static int fsl_qspi_clk_prep_enable(struct fsl_qspi *q)
443 {
444 	int ret;
445 
446 	ret = clk_prepare_enable(q->clk_en);
447 	if (ret)
448 		return ret;
449 
450 	ret = clk_prepare_enable(q->clk);
451 	if (ret) {
452 		clk_disable_unprepare(q->clk_en);
453 		return ret;
454 	}
455 
456 	if (needs_wakeup_wait_mode(q))
457 		pm_qos_add_request(&q->pm_qos_req, PM_QOS_CPU_DMA_LATENCY, 0);
458 
459 	return 0;
460 }
461 
462 static void fsl_qspi_clk_disable_unprep(struct fsl_qspi *q)
463 {
464 	if (needs_wakeup_wait_mode(q))
465 		pm_qos_remove_request(&q->pm_qos_req);
466 
467 	clk_disable_unprepare(q->clk);
468 	clk_disable_unprepare(q->clk_en);
469 }
470 
471 /*
472  * If we have changed the content of the flash by writing or erasing, or if we
473  * read from flash with a different offset into the page buffer, we need to
474  * invalidate the AHB buffer. If we do not do so, we may read out the wrong
475  * data. The spec tells us reset the AHB domain and Serial Flash domain at
476  * the same time.
477  */
478 static void fsl_qspi_invalidate(struct fsl_qspi *q)
479 {
480 	u32 reg;
481 
482 	reg = qspi_readl(q, q->iobase + QUADSPI_MCR);
483 	reg |= QUADSPI_MCR_SWRSTHD_MASK | QUADSPI_MCR_SWRSTSD_MASK;
484 	qspi_writel(q, reg, q->iobase + QUADSPI_MCR);
485 
486 	/*
487 	 * The minimum delay : 1 AHB + 2 SFCK clocks.
488 	 * Delay 1 us is enough.
489 	 */
490 	udelay(1);
491 
492 	reg &= ~(QUADSPI_MCR_SWRSTHD_MASK | QUADSPI_MCR_SWRSTSD_MASK);
493 	qspi_writel(q, reg, q->iobase + QUADSPI_MCR);
494 }
495 
496 static void fsl_qspi_select_mem(struct fsl_qspi *q, struct spi_device *spi)
497 {
498 	unsigned long rate = spi->max_speed_hz;
499 	int ret;
500 
501 	if (q->selected == spi->chip_select)
502 		return;
503 
504 	if (needs_4x_clock(q))
505 		rate *= 4;
506 
507 	fsl_qspi_clk_disable_unprep(q);
508 
509 	ret = clk_set_rate(q->clk, rate);
510 	if (ret)
511 		return;
512 
513 	ret = fsl_qspi_clk_prep_enable(q);
514 	if (ret)
515 		return;
516 
517 	q->selected = spi->chip_select;
518 
519 	fsl_qspi_invalidate(q);
520 }
521 
522 static void fsl_qspi_read_ahb(struct fsl_qspi *q, const struct spi_mem_op *op)
523 {
524 	memcpy_fromio(op->data.buf.in,
525 		      q->ahb_addr + q->selected * q->devtype_data->ahb_buf_size,
526 		      op->data.nbytes);
527 }
528 
529 static void fsl_qspi_fill_txfifo(struct fsl_qspi *q,
530 				 const struct spi_mem_op *op)
531 {
532 	void __iomem *base = q->iobase;
533 	int i;
534 	u32 val;
535 
536 	for (i = 0; i < ALIGN_DOWN(op->data.nbytes, 4); i += 4) {
537 		memcpy(&val, op->data.buf.out + i, 4);
538 		val = fsl_qspi_endian_xchg(q, val);
539 		qspi_writel(q, val, base + QUADSPI_TBDR);
540 	}
541 
542 	if (i < op->data.nbytes) {
543 		memcpy(&val, op->data.buf.out + i, op->data.nbytes - i);
544 		val = fsl_qspi_endian_xchg(q, val);
545 		qspi_writel(q, val, base + QUADSPI_TBDR);
546 	}
547 
548 	if (needs_fill_txfifo(q)) {
549 		for (i = op->data.nbytes; i < 16; i += 4)
550 			qspi_writel(q, 0, base + QUADSPI_TBDR);
551 	}
552 }
553 
554 static void fsl_qspi_read_rxfifo(struct fsl_qspi *q,
555 			  const struct spi_mem_op *op)
556 {
557 	void __iomem *base = q->iobase;
558 	int i;
559 	u8 *buf = op->data.buf.in;
560 	u32 val;
561 
562 	for (i = 0; i < ALIGN_DOWN(op->data.nbytes, 4); i += 4) {
563 		val = qspi_readl(q, base + QUADSPI_RBDR(i / 4));
564 		val = fsl_qspi_endian_xchg(q, val);
565 		memcpy(buf + i, &val, 4);
566 	}
567 
568 	if (i < op->data.nbytes) {
569 		val = qspi_readl(q, base + QUADSPI_RBDR(i / 4));
570 		val = fsl_qspi_endian_xchg(q, val);
571 		memcpy(buf + i, &val, op->data.nbytes - i);
572 	}
573 }
574 
575 static int fsl_qspi_do_op(struct fsl_qspi *q, const struct spi_mem_op *op)
576 {
577 	void __iomem *base = q->iobase;
578 	int err = 0;
579 
580 	init_completion(&q->c);
581 
582 	/*
583 	 * Always start the sequence at the same index since we update
584 	 * the LUT at each exec_op() call. And also specify the DATA
585 	 * length, since it's has not been specified in the LUT.
586 	 */
587 	qspi_writel(q, op->data.nbytes | QUADSPI_IPCR_SEQID(SEQID_LUT),
588 		    base + QUADSPI_IPCR);
589 
590 	/* Wait for the interrupt. */
591 	if (!wait_for_completion_timeout(&q->c, msecs_to_jiffies(1000)))
592 		err = -ETIMEDOUT;
593 
594 	if (!err && op->data.nbytes && op->data.dir == SPI_MEM_DATA_IN)
595 		fsl_qspi_read_rxfifo(q, op);
596 
597 	return err;
598 }
599 
600 static int fsl_qspi_readl_poll_tout(struct fsl_qspi *q, void __iomem *base,
601 				    u32 mask, u32 delay_us, u32 timeout_us)
602 {
603 	u32 reg;
604 
605 	if (!q->devtype_data->little_endian)
606 		mask = (u32)cpu_to_be32(mask);
607 
608 	return readl_poll_timeout(base, reg, !(reg & mask), delay_us,
609 				  timeout_us);
610 }
611 
612 static int fsl_qspi_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
613 {
614 	struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
615 	void __iomem *base = q->iobase;
616 	u32 addr_offset = 0;
617 	int err = 0;
618 
619 	mutex_lock(&q->lock);
620 
621 	/* wait for the controller being ready */
622 	fsl_qspi_readl_poll_tout(q, base + QUADSPI_SR, (QUADSPI_SR_IP_ACC_MASK |
623 				 QUADSPI_SR_AHB_ACC_MASK), 10, 1000);
624 
625 	fsl_qspi_select_mem(q, mem->spi);
626 
627 	if (needs_amba_base_offset(q))
628 		addr_offset = q->memmap_phy;
629 
630 	qspi_writel(q,
631 		    q->selected * q->devtype_data->ahb_buf_size + addr_offset,
632 		    base + QUADSPI_SFAR);
633 
634 	qspi_writel(q, qspi_readl(q, base + QUADSPI_MCR) |
635 		    QUADSPI_MCR_CLR_RXF_MASK | QUADSPI_MCR_CLR_TXF_MASK,
636 		    base + QUADSPI_MCR);
637 
638 	qspi_writel(q, QUADSPI_SPTRCLR_BFPTRC | QUADSPI_SPTRCLR_IPPTRC,
639 		    base + QUADSPI_SPTRCLR);
640 
641 	fsl_qspi_prepare_lut(q, op);
642 
643 	/*
644 	 * If we have large chunks of data, we read them through the AHB bus
645 	 * by accessing the mapped memory. In all other cases we use
646 	 * IP commands to access the flash.
647 	 */
648 	if (op->data.nbytes > (q->devtype_data->rxfifo - 4) &&
649 	    op->data.dir == SPI_MEM_DATA_IN) {
650 		fsl_qspi_read_ahb(q, op);
651 	} else {
652 		qspi_writel(q, QUADSPI_RBCT_WMRK_MASK |
653 			    QUADSPI_RBCT_RXBRD_USEIPS, base + QUADSPI_RBCT);
654 
655 		if (op->data.nbytes && op->data.dir == SPI_MEM_DATA_OUT)
656 			fsl_qspi_fill_txfifo(q, op);
657 
658 		err = fsl_qspi_do_op(q, op);
659 	}
660 
661 	/* Invalidate the data in the AHB buffer. */
662 	fsl_qspi_invalidate(q);
663 
664 	mutex_unlock(&q->lock);
665 
666 	return err;
667 }
668 
669 static int fsl_qspi_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
670 {
671 	struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
672 
673 	if (op->data.dir == SPI_MEM_DATA_OUT) {
674 		if (op->data.nbytes > q->devtype_data->txfifo)
675 			op->data.nbytes = q->devtype_data->txfifo;
676 	} else {
677 		if (op->data.nbytes > q->devtype_data->ahb_buf_size)
678 			op->data.nbytes = q->devtype_data->ahb_buf_size;
679 		else if (op->data.nbytes > (q->devtype_data->rxfifo - 4))
680 			op->data.nbytes = ALIGN_DOWN(op->data.nbytes, 8);
681 	}
682 
683 	return 0;
684 }
685 
686 static int fsl_qspi_default_setup(struct fsl_qspi *q)
687 {
688 	void __iomem *base = q->iobase;
689 	u32 reg, addr_offset = 0;
690 	int ret;
691 
692 	/* disable and unprepare clock to avoid glitch pass to controller */
693 	fsl_qspi_clk_disable_unprep(q);
694 
695 	/* the default frequency, we will change it later if necessary. */
696 	ret = clk_set_rate(q->clk, 66000000);
697 	if (ret)
698 		return ret;
699 
700 	ret = fsl_qspi_clk_prep_enable(q);
701 	if (ret)
702 		return ret;
703 
704 	/* Reset the module */
705 	qspi_writel(q, QUADSPI_MCR_SWRSTSD_MASK | QUADSPI_MCR_SWRSTHD_MASK,
706 		    base + QUADSPI_MCR);
707 	udelay(1);
708 
709 	/* Disable the module */
710 	qspi_writel(q, QUADSPI_MCR_MDIS_MASK | QUADSPI_MCR_RESERVED_MASK,
711 		    base + QUADSPI_MCR);
712 
713 	reg = qspi_readl(q, base + QUADSPI_SMPR);
714 	qspi_writel(q, reg & ~(QUADSPI_SMPR_FSDLY_MASK
715 			| QUADSPI_SMPR_FSPHS_MASK
716 			| QUADSPI_SMPR_HSENA_MASK
717 			| QUADSPI_SMPR_DDRSMP_MASK), base + QUADSPI_SMPR);
718 
719 	/* We only use the buffer3 for AHB read */
720 	qspi_writel(q, 0, base + QUADSPI_BUF0IND);
721 	qspi_writel(q, 0, base + QUADSPI_BUF1IND);
722 	qspi_writel(q, 0, base + QUADSPI_BUF2IND);
723 
724 	qspi_writel(q, QUADSPI_BFGENCR_SEQID(SEQID_LUT),
725 		    q->iobase + QUADSPI_BFGENCR);
726 	qspi_writel(q, QUADSPI_RBCT_WMRK_MASK, base + QUADSPI_RBCT);
727 	qspi_writel(q, QUADSPI_BUF3CR_ALLMST_MASK |
728 		    QUADSPI_BUF3CR_ADATSZ(q->devtype_data->ahb_buf_size / 8),
729 		    base + QUADSPI_BUF3CR);
730 
731 	if (needs_amba_base_offset(q))
732 		addr_offset = q->memmap_phy;
733 
734 	/*
735 	 * In HW there can be a maximum of four chips on two buses with
736 	 * two chip selects on each bus. We use four chip selects in SW
737 	 * to differentiate between the four chips.
738 	 * We use ahb_buf_size for each chip and set SFA1AD, SFA2AD, SFB1AD,
739 	 * SFB2AD accordingly.
740 	 */
741 	qspi_writel(q, q->devtype_data->ahb_buf_size + addr_offset,
742 		    base + QUADSPI_SFA1AD);
743 	qspi_writel(q, q->devtype_data->ahb_buf_size * 2 + addr_offset,
744 		    base + QUADSPI_SFA2AD);
745 	qspi_writel(q, q->devtype_data->ahb_buf_size * 3 + addr_offset,
746 		    base + QUADSPI_SFB1AD);
747 	qspi_writel(q, q->devtype_data->ahb_buf_size * 4 + addr_offset,
748 		    base + QUADSPI_SFB2AD);
749 
750 	q->selected = -1;
751 
752 	/* Enable the module */
753 	qspi_writel(q, QUADSPI_MCR_RESERVED_MASK | QUADSPI_MCR_END_CFG_MASK,
754 		    base + QUADSPI_MCR);
755 
756 	/* clear all interrupt status */
757 	qspi_writel(q, 0xffffffff, q->iobase + QUADSPI_FR);
758 
759 	/* enable the interrupt */
760 	qspi_writel(q, QUADSPI_RSER_TFIE, q->iobase + QUADSPI_RSER);
761 
762 	return 0;
763 }
764 
765 static const char *fsl_qspi_get_name(struct spi_mem *mem)
766 {
767 	struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
768 	struct device *dev = &mem->spi->dev;
769 	const char *name;
770 
771 	/*
772 	 * In order to keep mtdparts compatible with the old MTD driver at
773 	 * mtd/spi-nor/fsl-quadspi.c, we set a custom name derived from the
774 	 * platform_device of the controller.
775 	 */
776 	if (of_get_available_child_count(q->dev->of_node) == 1)
777 		return dev_name(q->dev);
778 
779 	name = devm_kasprintf(dev, GFP_KERNEL,
780 			      "%s-%d", dev_name(q->dev),
781 			      mem->spi->chip_select);
782 
783 	if (!name) {
784 		dev_err(dev, "failed to get memory for custom flash name\n");
785 		return ERR_PTR(-ENOMEM);
786 	}
787 
788 	return name;
789 }
790 
791 static const struct spi_controller_mem_ops fsl_qspi_mem_ops = {
792 	.adjust_op_size = fsl_qspi_adjust_op_size,
793 	.supports_op = fsl_qspi_supports_op,
794 	.exec_op = fsl_qspi_exec_op,
795 	.get_name = fsl_qspi_get_name,
796 };
797 
798 static int fsl_qspi_probe(struct platform_device *pdev)
799 {
800 	struct spi_controller *ctlr;
801 	struct device *dev = &pdev->dev;
802 	struct device_node *np = dev->of_node;
803 	struct resource *res;
804 	struct fsl_qspi *q;
805 	int ret;
806 
807 	ctlr = spi_alloc_master(&pdev->dev, sizeof(*q));
808 	if (!ctlr)
809 		return -ENOMEM;
810 
811 	ctlr->mode_bits = SPI_RX_DUAL | SPI_RX_QUAD |
812 			  SPI_TX_DUAL | SPI_TX_QUAD;
813 
814 	q = spi_controller_get_devdata(ctlr);
815 	q->dev = dev;
816 	q->devtype_data = of_device_get_match_data(dev);
817 	if (!q->devtype_data) {
818 		ret = -ENODEV;
819 		goto err_put_ctrl;
820 	}
821 
822 	platform_set_drvdata(pdev, q);
823 
824 	/* find the resources */
825 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "QuadSPI");
826 	q->iobase = devm_ioremap_resource(dev, res);
827 	if (IS_ERR(q->iobase)) {
828 		ret = PTR_ERR(q->iobase);
829 		goto err_put_ctrl;
830 	}
831 
832 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
833 					"QuadSPI-memory");
834 	q->ahb_addr = devm_ioremap_resource(dev, res);
835 	if (IS_ERR(q->ahb_addr)) {
836 		ret = PTR_ERR(q->ahb_addr);
837 		goto err_put_ctrl;
838 	}
839 
840 	q->memmap_phy = res->start;
841 
842 	/* find the clocks */
843 	q->clk_en = devm_clk_get(dev, "qspi_en");
844 	if (IS_ERR(q->clk_en)) {
845 		ret = PTR_ERR(q->clk_en);
846 		goto err_put_ctrl;
847 	}
848 
849 	q->clk = devm_clk_get(dev, "qspi");
850 	if (IS_ERR(q->clk)) {
851 		ret = PTR_ERR(q->clk);
852 		goto err_put_ctrl;
853 	}
854 
855 	ret = fsl_qspi_clk_prep_enable(q);
856 	if (ret) {
857 		dev_err(dev, "can not enable the clock\n");
858 		goto err_put_ctrl;
859 	}
860 
861 	/* find the irq */
862 	ret = platform_get_irq(pdev, 0);
863 	if (ret < 0)
864 		goto err_disable_clk;
865 
866 	ret = devm_request_irq(dev, ret,
867 			fsl_qspi_irq_handler, 0, pdev->name, q);
868 	if (ret) {
869 		dev_err(dev, "failed to request irq: %d\n", ret);
870 		goto err_disable_clk;
871 	}
872 
873 	mutex_init(&q->lock);
874 
875 	ctlr->bus_num = -1;
876 	ctlr->num_chipselect = 4;
877 	ctlr->mem_ops = &fsl_qspi_mem_ops;
878 
879 	fsl_qspi_default_setup(q);
880 
881 	ctlr->dev.of_node = np;
882 
883 	ret = devm_spi_register_controller(dev, ctlr);
884 	if (ret)
885 		goto err_destroy_mutex;
886 
887 	return 0;
888 
889 err_destroy_mutex:
890 	mutex_destroy(&q->lock);
891 
892 err_disable_clk:
893 	fsl_qspi_clk_disable_unprep(q);
894 
895 err_put_ctrl:
896 	spi_controller_put(ctlr);
897 
898 	dev_err(dev, "Freescale QuadSPI probe failed\n");
899 	return ret;
900 }
901 
902 static int fsl_qspi_remove(struct platform_device *pdev)
903 {
904 	struct fsl_qspi *q = platform_get_drvdata(pdev);
905 
906 	/* disable the hardware */
907 	qspi_writel(q, QUADSPI_MCR_MDIS_MASK, q->iobase + QUADSPI_MCR);
908 	qspi_writel(q, 0x0, q->iobase + QUADSPI_RSER);
909 
910 	fsl_qspi_clk_disable_unprep(q);
911 
912 	mutex_destroy(&q->lock);
913 
914 	return 0;
915 }
916 
917 static int fsl_qspi_suspend(struct device *dev)
918 {
919 	return 0;
920 }
921 
922 static int fsl_qspi_resume(struct device *dev)
923 {
924 	struct fsl_qspi *q = dev_get_drvdata(dev);
925 
926 	fsl_qspi_default_setup(q);
927 
928 	return 0;
929 }
930 
931 static const struct of_device_id fsl_qspi_dt_ids[] = {
932 	{ .compatible = "fsl,vf610-qspi", .data = &vybrid_data, },
933 	{ .compatible = "fsl,imx6sx-qspi", .data = &imx6sx_data, },
934 	{ .compatible = "fsl,imx7d-qspi", .data = &imx7d_data, },
935 	{ .compatible = "fsl,imx6ul-qspi", .data = &imx6ul_data, },
936 	{ .compatible = "fsl,ls1021a-qspi", .data = &ls1021a_data, },
937 	{ .compatible = "fsl,ls2080a-qspi", .data = &ls2080a_data, },
938 	{ /* sentinel */ }
939 };
940 MODULE_DEVICE_TABLE(of, fsl_qspi_dt_ids);
941 
942 static const struct dev_pm_ops fsl_qspi_pm_ops = {
943 	.suspend	= fsl_qspi_suspend,
944 	.resume		= fsl_qspi_resume,
945 };
946 
947 static struct platform_driver fsl_qspi_driver = {
948 	.driver = {
949 		.name	= "fsl-quadspi",
950 		.of_match_table = fsl_qspi_dt_ids,
951 		.pm =   &fsl_qspi_pm_ops,
952 	},
953 	.probe          = fsl_qspi_probe,
954 	.remove		= fsl_qspi_remove,
955 };
956 module_platform_driver(fsl_qspi_driver);
957 
958 MODULE_DESCRIPTION("Freescale QuadSPI Controller Driver");
959 MODULE_AUTHOR("Freescale Semiconductor Inc.");
960 MODULE_AUTHOR("Boris Brezillon <bbrezillon@kernel.org>");
961 MODULE_AUTHOR("Frieder Schrempf <frieder.schrempf@kontron.de>");
962 MODULE_AUTHOR("Yogesh Gaur <yogeshnarayan.gaur@nxp.com>");
963 MODULE_AUTHOR("Suresh Gupta <suresh.gupta@nxp.com>");
964 MODULE_LICENSE("GPL v2");
965