xref: /openbmc/linux/drivers/spi/spi-fsl-qspi.c (revision c4f461a1)
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_FLSHCR			0x0c
67 #define QUADSPI_FLSHCR_TCSS_MASK	GENMASK(3, 0)
68 #define QUADSPI_FLSHCR_TCSH_MASK	GENMASK(11, 8)
69 #define QUADSPI_FLSHCR_TDH_MASK		GENMASK(17, 16)
70 
71 #define QUADSPI_BUF0CR                  0x10
72 #define QUADSPI_BUF1CR                  0x14
73 #define QUADSPI_BUF2CR                  0x18
74 #define QUADSPI_BUFXCR_INVALID_MSTRID   0xe
75 
76 #define QUADSPI_BUF3CR			0x1c
77 #define QUADSPI_BUF3CR_ALLMST_MASK	BIT(31)
78 #define QUADSPI_BUF3CR_ADATSZ(x)	((x) << 8)
79 #define QUADSPI_BUF3CR_ADATSZ_MASK	GENMASK(15, 8)
80 
81 #define QUADSPI_BFGENCR			0x20
82 #define QUADSPI_BFGENCR_SEQID(x)	((x) << 12)
83 
84 #define QUADSPI_BUF0IND			0x30
85 #define QUADSPI_BUF1IND			0x34
86 #define QUADSPI_BUF2IND			0x38
87 #define QUADSPI_SFAR			0x100
88 
89 #define QUADSPI_SMPR			0x108
90 #define QUADSPI_SMPR_DDRSMP_MASK	GENMASK(18, 16)
91 #define QUADSPI_SMPR_FSDLY_MASK		BIT(6)
92 #define QUADSPI_SMPR_FSPHS_MASK		BIT(5)
93 #define QUADSPI_SMPR_HSENA_MASK		BIT(0)
94 
95 #define QUADSPI_RBCT			0x110
96 #define QUADSPI_RBCT_WMRK_MASK		GENMASK(4, 0)
97 #define QUADSPI_RBCT_RXBRD_USEIPS	BIT(8)
98 
99 #define QUADSPI_TBDR			0x154
100 
101 #define QUADSPI_SR			0x15c
102 #define QUADSPI_SR_IP_ACC_MASK		BIT(1)
103 #define QUADSPI_SR_AHB_ACC_MASK		BIT(2)
104 
105 #define QUADSPI_FR			0x160
106 #define QUADSPI_FR_TFF_MASK		BIT(0)
107 
108 #define QUADSPI_RSER			0x164
109 #define QUADSPI_RSER_TFIE		BIT(0)
110 
111 #define QUADSPI_SPTRCLR			0x16c
112 #define QUADSPI_SPTRCLR_IPPTRC		BIT(8)
113 #define QUADSPI_SPTRCLR_BFPTRC		BIT(0)
114 
115 #define QUADSPI_SFA1AD			0x180
116 #define QUADSPI_SFA2AD			0x184
117 #define QUADSPI_SFB1AD			0x188
118 #define QUADSPI_SFB2AD			0x18c
119 #define QUADSPI_RBDR(x)			(0x200 + ((x) * 4))
120 
121 #define QUADSPI_LUTKEY			0x300
122 #define QUADSPI_LUTKEY_VALUE		0x5AF05AF0
123 
124 #define QUADSPI_LCKCR			0x304
125 #define QUADSPI_LCKER_LOCK		BIT(0)
126 #define QUADSPI_LCKER_UNLOCK		BIT(1)
127 
128 #define QUADSPI_LUT_BASE		0x310
129 #define QUADSPI_LUT_OFFSET		(SEQID_LUT * 4 * 4)
130 #define QUADSPI_LUT_REG(idx) \
131 	(QUADSPI_LUT_BASE + QUADSPI_LUT_OFFSET + (idx) * 4)
132 
133 /* Instruction set for the LUT register */
134 #define LUT_STOP		0
135 #define LUT_CMD			1
136 #define LUT_ADDR		2
137 #define LUT_DUMMY		3
138 #define LUT_MODE		4
139 #define LUT_MODE2		5
140 #define LUT_MODE4		6
141 #define LUT_FSL_READ		7
142 #define LUT_FSL_WRITE		8
143 #define LUT_JMP_ON_CS		9
144 #define LUT_ADDR_DDR		10
145 #define LUT_MODE_DDR		11
146 #define LUT_MODE2_DDR		12
147 #define LUT_MODE4_DDR		13
148 #define LUT_FSL_READ_DDR	14
149 #define LUT_FSL_WRITE_DDR	15
150 #define LUT_DATA_LEARN		16
151 
152 /*
153  * The PAD definitions for LUT register.
154  *
155  * The pad stands for the number of IO lines [0:3].
156  * For example, the quad read needs four IO lines,
157  * so you should use LUT_PAD(4).
158  */
159 #define LUT_PAD(x) (fls(x) - 1)
160 
161 /*
162  * Macro for constructing the LUT entries with the following
163  * register layout:
164  *
165  *  ---------------------------------------------------
166  *  | INSTR1 | PAD1 | OPRND1 | INSTR0 | PAD0 | OPRND0 |
167  *  ---------------------------------------------------
168  */
169 #define LUT_DEF(idx, ins, pad, opr)					\
170 	((((ins) << 10) | ((pad) << 8) | (opr)) << (((idx) % 2) * 16))
171 
172 /* Controller needs driver to swap endianness */
173 #define QUADSPI_QUIRK_SWAP_ENDIAN	BIT(0)
174 
175 /* Controller needs 4x internal clock */
176 #define QUADSPI_QUIRK_4X_INT_CLK	BIT(1)
177 
178 /*
179  * TKT253890, the controller needs the driver to fill the txfifo with
180  * 16 bytes at least to trigger a data transfer, even though the extra
181  * data won't be transferred.
182  */
183 #define QUADSPI_QUIRK_TKT253890		BIT(2)
184 
185 /* TKT245618, the controller cannot wake up from wait mode */
186 #define QUADSPI_QUIRK_TKT245618		BIT(3)
187 
188 /*
189  * Controller adds QSPI_AMBA_BASE (base address of the mapped memory)
190  * internally. No need to add it when setting SFXXAD and SFAR registers
191  */
192 #define QUADSPI_QUIRK_BASE_INTERNAL	BIT(4)
193 
194 /*
195  * Controller uses TDH bits in register QUADSPI_FLSHCR.
196  * They need to be set in accordance with the DDR/SDR mode.
197  */
198 #define QUADSPI_QUIRK_USE_TDH_SETTING	BIT(5)
199 
200 struct fsl_qspi_devtype_data {
201 	unsigned int rxfifo;
202 	unsigned int txfifo;
203 	int invalid_mstrid;
204 	unsigned int ahb_buf_size;
205 	unsigned int quirks;
206 	bool little_endian;
207 };
208 
209 static const struct fsl_qspi_devtype_data vybrid_data = {
210 	.rxfifo = SZ_128,
211 	.txfifo = SZ_64,
212 	.invalid_mstrid = QUADSPI_BUFXCR_INVALID_MSTRID,
213 	.ahb_buf_size = SZ_1K,
214 	.quirks = QUADSPI_QUIRK_SWAP_ENDIAN,
215 	.little_endian = true,
216 };
217 
218 static const struct fsl_qspi_devtype_data imx6sx_data = {
219 	.rxfifo = SZ_128,
220 	.txfifo = SZ_512,
221 	.invalid_mstrid = QUADSPI_BUFXCR_INVALID_MSTRID,
222 	.ahb_buf_size = SZ_1K,
223 	.quirks = QUADSPI_QUIRK_4X_INT_CLK | QUADSPI_QUIRK_TKT245618,
224 	.little_endian = true,
225 };
226 
227 static const struct fsl_qspi_devtype_data imx7d_data = {
228 	.rxfifo = SZ_128,
229 	.txfifo = SZ_512,
230 	.invalid_mstrid = QUADSPI_BUFXCR_INVALID_MSTRID,
231 	.ahb_buf_size = SZ_1K,
232 	.quirks = QUADSPI_QUIRK_TKT253890 | QUADSPI_QUIRK_4X_INT_CLK |
233 		  QUADSPI_QUIRK_USE_TDH_SETTING,
234 	.little_endian = true,
235 };
236 
237 static const struct fsl_qspi_devtype_data imx6ul_data = {
238 	.rxfifo = SZ_128,
239 	.txfifo = SZ_512,
240 	.invalid_mstrid = QUADSPI_BUFXCR_INVALID_MSTRID,
241 	.ahb_buf_size = SZ_1K,
242 	.quirks = QUADSPI_QUIRK_TKT253890 | QUADSPI_QUIRK_4X_INT_CLK |
243 		  QUADSPI_QUIRK_USE_TDH_SETTING,
244 	.little_endian = true,
245 };
246 
247 static const struct fsl_qspi_devtype_data ls1021a_data = {
248 	.rxfifo = SZ_128,
249 	.txfifo = SZ_64,
250 	.invalid_mstrid = QUADSPI_BUFXCR_INVALID_MSTRID,
251 	.ahb_buf_size = SZ_1K,
252 	.quirks = 0,
253 	.little_endian = false,
254 };
255 
256 static const struct fsl_qspi_devtype_data ls2080a_data = {
257 	.rxfifo = SZ_128,
258 	.txfifo = SZ_64,
259 	.ahb_buf_size = SZ_1K,
260 	.invalid_mstrid = 0x0,
261 	.quirks = QUADSPI_QUIRK_TKT253890 | QUADSPI_QUIRK_BASE_INTERNAL,
262 	.little_endian = true,
263 };
264 
265 struct fsl_qspi {
266 	void __iomem *iobase;
267 	void __iomem *ahb_addr;
268 	u32 memmap_phy;
269 	struct clk *clk, *clk_en;
270 	struct device *dev;
271 	struct completion c;
272 	const struct fsl_qspi_devtype_data *devtype_data;
273 	struct mutex lock;
274 	struct pm_qos_request pm_qos_req;
275 	int selected;
276 };
277 
278 static inline int needs_swap_endian(struct fsl_qspi *q)
279 {
280 	return q->devtype_data->quirks & QUADSPI_QUIRK_SWAP_ENDIAN;
281 }
282 
283 static inline int needs_4x_clock(struct fsl_qspi *q)
284 {
285 	return q->devtype_data->quirks & QUADSPI_QUIRK_4X_INT_CLK;
286 }
287 
288 static inline int needs_fill_txfifo(struct fsl_qspi *q)
289 {
290 	return q->devtype_data->quirks & QUADSPI_QUIRK_TKT253890;
291 }
292 
293 static inline int needs_wakeup_wait_mode(struct fsl_qspi *q)
294 {
295 	return q->devtype_data->quirks & QUADSPI_QUIRK_TKT245618;
296 }
297 
298 static inline int needs_amba_base_offset(struct fsl_qspi *q)
299 {
300 	return !(q->devtype_data->quirks & QUADSPI_QUIRK_BASE_INTERNAL);
301 }
302 
303 static inline int needs_tdh_setting(struct fsl_qspi *q)
304 {
305 	return q->devtype_data->quirks & QUADSPI_QUIRK_USE_TDH_SETTING;
306 }
307 
308 /*
309  * An IC bug makes it necessary to rearrange the 32-bit data.
310  * Later chips, such as IMX6SLX, have fixed this bug.
311  */
312 static inline u32 fsl_qspi_endian_xchg(struct fsl_qspi *q, u32 a)
313 {
314 	return needs_swap_endian(q) ? __swab32(a) : a;
315 }
316 
317 /*
318  * R/W functions for big- or little-endian registers:
319  * The QSPI controller's endianness is independent of
320  * the CPU core's endianness. So far, although the CPU
321  * core is little-endian the QSPI controller can use
322  * big-endian or little-endian.
323  */
324 static void qspi_writel(struct fsl_qspi *q, u32 val, void __iomem *addr)
325 {
326 	if (q->devtype_data->little_endian)
327 		iowrite32(val, addr);
328 	else
329 		iowrite32be(val, addr);
330 }
331 
332 static u32 qspi_readl(struct fsl_qspi *q, void __iomem *addr)
333 {
334 	if (q->devtype_data->little_endian)
335 		return ioread32(addr);
336 
337 	return ioread32be(addr);
338 }
339 
340 static irqreturn_t fsl_qspi_irq_handler(int irq, void *dev_id)
341 {
342 	struct fsl_qspi *q = dev_id;
343 	u32 reg;
344 
345 	/* clear interrupt */
346 	reg = qspi_readl(q, q->iobase + QUADSPI_FR);
347 	qspi_writel(q, reg, q->iobase + QUADSPI_FR);
348 
349 	if (reg & QUADSPI_FR_TFF_MASK)
350 		complete(&q->c);
351 
352 	dev_dbg(q->dev, "QUADSPI_FR : 0x%.8x:0x%.8x\n", 0, reg);
353 	return IRQ_HANDLED;
354 }
355 
356 static int fsl_qspi_check_buswidth(struct fsl_qspi *q, u8 width)
357 {
358 	switch (width) {
359 	case 1:
360 	case 2:
361 	case 4:
362 		return 0;
363 	}
364 
365 	return -ENOTSUPP;
366 }
367 
368 static bool fsl_qspi_supports_op(struct spi_mem *mem,
369 				 const struct spi_mem_op *op)
370 {
371 	struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
372 	int ret;
373 
374 	ret = fsl_qspi_check_buswidth(q, op->cmd.buswidth);
375 
376 	if (op->addr.nbytes)
377 		ret |= fsl_qspi_check_buswidth(q, op->addr.buswidth);
378 
379 	if (op->dummy.nbytes)
380 		ret |= fsl_qspi_check_buswidth(q, op->dummy.buswidth);
381 
382 	if (op->data.nbytes)
383 		ret |= fsl_qspi_check_buswidth(q, op->data.buswidth);
384 
385 	if (ret)
386 		return false;
387 
388 	/*
389 	 * The number of instructions needed for the op, needs
390 	 * to fit into a single LUT entry.
391 	 */
392 	if (op->addr.nbytes +
393 	   (op->dummy.nbytes ? 1:0) +
394 	   (op->data.nbytes ? 1:0) > 6)
395 		return false;
396 
397 	/* Max 64 dummy clock cycles supported */
398 	if (op->dummy.nbytes &&
399 	    (op->dummy.nbytes * 8 / op->dummy.buswidth > 64))
400 		return false;
401 
402 	/* Max data length, check controller limits and alignment */
403 	if (op->data.dir == SPI_MEM_DATA_IN &&
404 	    (op->data.nbytes > q->devtype_data->ahb_buf_size ||
405 	     (op->data.nbytes > q->devtype_data->rxfifo - 4 &&
406 	      !IS_ALIGNED(op->data.nbytes, 8))))
407 		return false;
408 
409 	if (op->data.dir == SPI_MEM_DATA_OUT &&
410 	    op->data.nbytes > q->devtype_data->txfifo)
411 		return false;
412 
413 	return spi_mem_default_supports_op(mem, op);
414 }
415 
416 static void fsl_qspi_prepare_lut(struct fsl_qspi *q,
417 				 const struct spi_mem_op *op)
418 {
419 	void __iomem *base = q->iobase;
420 	u32 lutval[4] = {};
421 	int lutidx = 1, i;
422 
423 	lutval[0] |= LUT_DEF(0, LUT_CMD, LUT_PAD(op->cmd.buswidth),
424 			     op->cmd.opcode);
425 
426 	/*
427 	 * For some unknown reason, using LUT_ADDR doesn't work in some
428 	 * cases (at least with only one byte long addresses), so
429 	 * let's use LUT_MODE to write the address bytes one by one
430 	 */
431 	for (i = 0; i < op->addr.nbytes; i++) {
432 		u8 addrbyte = op->addr.val >> (8 * (op->addr.nbytes - i - 1));
433 
434 		lutval[lutidx / 2] |= LUT_DEF(lutidx, LUT_MODE,
435 					      LUT_PAD(op->addr.buswidth),
436 					      addrbyte);
437 		lutidx++;
438 	}
439 
440 	if (op->dummy.nbytes) {
441 		lutval[lutidx / 2] |= LUT_DEF(lutidx, LUT_DUMMY,
442 					      LUT_PAD(op->dummy.buswidth),
443 					      op->dummy.nbytes * 8 /
444 					      op->dummy.buswidth);
445 		lutidx++;
446 	}
447 
448 	if (op->data.nbytes) {
449 		lutval[lutidx / 2] |= LUT_DEF(lutidx,
450 					      op->data.dir == SPI_MEM_DATA_IN ?
451 					      LUT_FSL_READ : LUT_FSL_WRITE,
452 					      LUT_PAD(op->data.buswidth),
453 					      0);
454 		lutidx++;
455 	}
456 
457 	lutval[lutidx / 2] |= LUT_DEF(lutidx, LUT_STOP, 0, 0);
458 
459 	/* unlock LUT */
460 	qspi_writel(q, QUADSPI_LUTKEY_VALUE, q->iobase + QUADSPI_LUTKEY);
461 	qspi_writel(q, QUADSPI_LCKER_UNLOCK, q->iobase + QUADSPI_LCKCR);
462 
463 	/* fill LUT */
464 	for (i = 0; i < ARRAY_SIZE(lutval); i++)
465 		qspi_writel(q, lutval[i], base + QUADSPI_LUT_REG(i));
466 
467 	/* lock LUT */
468 	qspi_writel(q, QUADSPI_LUTKEY_VALUE, q->iobase + QUADSPI_LUTKEY);
469 	qspi_writel(q, QUADSPI_LCKER_LOCK, q->iobase + QUADSPI_LCKCR);
470 }
471 
472 static int fsl_qspi_clk_prep_enable(struct fsl_qspi *q)
473 {
474 	int ret;
475 
476 	ret = clk_prepare_enable(q->clk_en);
477 	if (ret)
478 		return ret;
479 
480 	ret = clk_prepare_enable(q->clk);
481 	if (ret) {
482 		clk_disable_unprepare(q->clk_en);
483 		return ret;
484 	}
485 
486 	if (needs_wakeup_wait_mode(q))
487 		cpu_latency_qos_add_request(&q->pm_qos_req, 0);
488 
489 	return 0;
490 }
491 
492 static void fsl_qspi_clk_disable_unprep(struct fsl_qspi *q)
493 {
494 	if (needs_wakeup_wait_mode(q))
495 		cpu_latency_qos_remove_request(&q->pm_qos_req);
496 
497 	clk_disable_unprepare(q->clk);
498 	clk_disable_unprepare(q->clk_en);
499 }
500 
501 /*
502  * If we have changed the content of the flash by writing or erasing, or if we
503  * read from flash with a different offset into the page buffer, we need to
504  * invalidate the AHB buffer. If we do not do so, we may read out the wrong
505  * data. The spec tells us reset the AHB domain and Serial Flash domain at
506  * the same time.
507  */
508 static void fsl_qspi_invalidate(struct fsl_qspi *q)
509 {
510 	u32 reg;
511 
512 	reg = qspi_readl(q, q->iobase + QUADSPI_MCR);
513 	reg |= QUADSPI_MCR_SWRSTHD_MASK | QUADSPI_MCR_SWRSTSD_MASK;
514 	qspi_writel(q, reg, q->iobase + QUADSPI_MCR);
515 
516 	/*
517 	 * The minimum delay : 1 AHB + 2 SFCK clocks.
518 	 * Delay 1 us is enough.
519 	 */
520 	udelay(1);
521 
522 	reg &= ~(QUADSPI_MCR_SWRSTHD_MASK | QUADSPI_MCR_SWRSTSD_MASK);
523 	qspi_writel(q, reg, q->iobase + QUADSPI_MCR);
524 }
525 
526 static void fsl_qspi_select_mem(struct fsl_qspi *q, struct spi_device *spi)
527 {
528 	unsigned long rate = spi->max_speed_hz;
529 	int ret;
530 
531 	if (q->selected == spi_get_chipselect(spi, 0))
532 		return;
533 
534 	if (needs_4x_clock(q))
535 		rate *= 4;
536 
537 	fsl_qspi_clk_disable_unprep(q);
538 
539 	ret = clk_set_rate(q->clk, rate);
540 	if (ret)
541 		return;
542 
543 	ret = fsl_qspi_clk_prep_enable(q);
544 	if (ret)
545 		return;
546 
547 	q->selected = spi_get_chipselect(spi, 0);
548 
549 	fsl_qspi_invalidate(q);
550 }
551 
552 static void fsl_qspi_read_ahb(struct fsl_qspi *q, const struct spi_mem_op *op)
553 {
554 	memcpy_fromio(op->data.buf.in,
555 		      q->ahb_addr + q->selected * q->devtype_data->ahb_buf_size,
556 		      op->data.nbytes);
557 }
558 
559 static void fsl_qspi_fill_txfifo(struct fsl_qspi *q,
560 				 const struct spi_mem_op *op)
561 {
562 	void __iomem *base = q->iobase;
563 	int i;
564 	u32 val;
565 
566 	for (i = 0; i < ALIGN_DOWN(op->data.nbytes, 4); i += 4) {
567 		memcpy(&val, op->data.buf.out + i, 4);
568 		val = fsl_qspi_endian_xchg(q, val);
569 		qspi_writel(q, val, base + QUADSPI_TBDR);
570 	}
571 
572 	if (i < op->data.nbytes) {
573 		memcpy(&val, op->data.buf.out + i, op->data.nbytes - i);
574 		val = fsl_qspi_endian_xchg(q, val);
575 		qspi_writel(q, val, base + QUADSPI_TBDR);
576 	}
577 
578 	if (needs_fill_txfifo(q)) {
579 		for (i = op->data.nbytes; i < 16; i += 4)
580 			qspi_writel(q, 0, base + QUADSPI_TBDR);
581 	}
582 }
583 
584 static void fsl_qspi_read_rxfifo(struct fsl_qspi *q,
585 			  const struct spi_mem_op *op)
586 {
587 	void __iomem *base = q->iobase;
588 	int i;
589 	u8 *buf = op->data.buf.in;
590 	u32 val;
591 
592 	for (i = 0; i < ALIGN_DOWN(op->data.nbytes, 4); i += 4) {
593 		val = qspi_readl(q, base + QUADSPI_RBDR(i / 4));
594 		val = fsl_qspi_endian_xchg(q, val);
595 		memcpy(buf + i, &val, 4);
596 	}
597 
598 	if (i < op->data.nbytes) {
599 		val = qspi_readl(q, base + QUADSPI_RBDR(i / 4));
600 		val = fsl_qspi_endian_xchg(q, val);
601 		memcpy(buf + i, &val, op->data.nbytes - i);
602 	}
603 }
604 
605 static int fsl_qspi_do_op(struct fsl_qspi *q, const struct spi_mem_op *op)
606 {
607 	void __iomem *base = q->iobase;
608 	int err = 0;
609 
610 	init_completion(&q->c);
611 
612 	/*
613 	 * Always start the sequence at the same index since we update
614 	 * the LUT at each exec_op() call. And also specify the DATA
615 	 * length, since it's has not been specified in the LUT.
616 	 */
617 	qspi_writel(q, op->data.nbytes | QUADSPI_IPCR_SEQID(SEQID_LUT),
618 		    base + QUADSPI_IPCR);
619 
620 	/* Wait for the interrupt. */
621 	if (!wait_for_completion_timeout(&q->c, msecs_to_jiffies(1000)))
622 		err = -ETIMEDOUT;
623 
624 	if (!err && op->data.nbytes && op->data.dir == SPI_MEM_DATA_IN)
625 		fsl_qspi_read_rxfifo(q, op);
626 
627 	return err;
628 }
629 
630 static int fsl_qspi_readl_poll_tout(struct fsl_qspi *q, void __iomem *base,
631 				    u32 mask, u32 delay_us, u32 timeout_us)
632 {
633 	u32 reg;
634 
635 	if (!q->devtype_data->little_endian)
636 		mask = (u32)cpu_to_be32(mask);
637 
638 	return readl_poll_timeout(base, reg, !(reg & mask), delay_us,
639 				  timeout_us);
640 }
641 
642 static int fsl_qspi_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
643 {
644 	struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
645 	void __iomem *base = q->iobase;
646 	u32 addr_offset = 0;
647 	int err = 0;
648 	int invalid_mstrid = q->devtype_data->invalid_mstrid;
649 
650 	mutex_lock(&q->lock);
651 
652 	/* wait for the controller being ready */
653 	fsl_qspi_readl_poll_tout(q, base + QUADSPI_SR, (QUADSPI_SR_IP_ACC_MASK |
654 				 QUADSPI_SR_AHB_ACC_MASK), 10, 1000);
655 
656 	fsl_qspi_select_mem(q, mem->spi);
657 
658 	if (needs_amba_base_offset(q))
659 		addr_offset = q->memmap_phy;
660 
661 	qspi_writel(q,
662 		    q->selected * q->devtype_data->ahb_buf_size + addr_offset,
663 		    base + QUADSPI_SFAR);
664 
665 	qspi_writel(q, qspi_readl(q, base + QUADSPI_MCR) |
666 		    QUADSPI_MCR_CLR_RXF_MASK | QUADSPI_MCR_CLR_TXF_MASK,
667 		    base + QUADSPI_MCR);
668 
669 	qspi_writel(q, QUADSPI_SPTRCLR_BFPTRC | QUADSPI_SPTRCLR_IPPTRC,
670 		    base + QUADSPI_SPTRCLR);
671 
672 	qspi_writel(q, invalid_mstrid, base + QUADSPI_BUF0CR);
673 	qspi_writel(q, invalid_mstrid, base + QUADSPI_BUF1CR);
674 	qspi_writel(q, invalid_mstrid, base + QUADSPI_BUF2CR);
675 
676 	fsl_qspi_prepare_lut(q, op);
677 
678 	/*
679 	 * If we have large chunks of data, we read them through the AHB bus
680 	 * by accessing the mapped memory. In all other cases we use
681 	 * IP commands to access the flash.
682 	 */
683 	if (op->data.nbytes > (q->devtype_data->rxfifo - 4) &&
684 	    op->data.dir == SPI_MEM_DATA_IN) {
685 		fsl_qspi_read_ahb(q, op);
686 	} else {
687 		qspi_writel(q, QUADSPI_RBCT_WMRK_MASK |
688 			    QUADSPI_RBCT_RXBRD_USEIPS, base + QUADSPI_RBCT);
689 
690 		if (op->data.nbytes && op->data.dir == SPI_MEM_DATA_OUT)
691 			fsl_qspi_fill_txfifo(q, op);
692 
693 		err = fsl_qspi_do_op(q, op);
694 	}
695 
696 	/* Invalidate the data in the AHB buffer. */
697 	fsl_qspi_invalidate(q);
698 
699 	mutex_unlock(&q->lock);
700 
701 	return err;
702 }
703 
704 static int fsl_qspi_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
705 {
706 	struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
707 
708 	if (op->data.dir == SPI_MEM_DATA_OUT) {
709 		if (op->data.nbytes > q->devtype_data->txfifo)
710 			op->data.nbytes = q->devtype_data->txfifo;
711 	} else {
712 		if (op->data.nbytes > q->devtype_data->ahb_buf_size)
713 			op->data.nbytes = q->devtype_data->ahb_buf_size;
714 		else if (op->data.nbytes > (q->devtype_data->rxfifo - 4))
715 			op->data.nbytes = ALIGN_DOWN(op->data.nbytes, 8);
716 	}
717 
718 	return 0;
719 }
720 
721 static int fsl_qspi_default_setup(struct fsl_qspi *q)
722 {
723 	void __iomem *base = q->iobase;
724 	u32 reg, addr_offset = 0;
725 	int ret;
726 
727 	/* disable and unprepare clock to avoid glitch pass to controller */
728 	fsl_qspi_clk_disable_unprep(q);
729 
730 	/* the default frequency, we will change it later if necessary. */
731 	ret = clk_set_rate(q->clk, 66000000);
732 	if (ret)
733 		return ret;
734 
735 	ret = fsl_qspi_clk_prep_enable(q);
736 	if (ret)
737 		return ret;
738 
739 	/* Reset the module */
740 	qspi_writel(q, QUADSPI_MCR_SWRSTSD_MASK | QUADSPI_MCR_SWRSTHD_MASK,
741 		    base + QUADSPI_MCR);
742 	udelay(1);
743 
744 	/* Disable the module */
745 	qspi_writel(q, QUADSPI_MCR_MDIS_MASK | QUADSPI_MCR_RESERVED_MASK,
746 		    base + QUADSPI_MCR);
747 
748 	/*
749 	 * Previous boot stages (BootROM, bootloader) might have used DDR
750 	 * mode and did not clear the TDH bits. As we currently use SDR mode
751 	 * only, clear the TDH bits if necessary.
752 	 */
753 	if (needs_tdh_setting(q))
754 		qspi_writel(q, qspi_readl(q, base + QUADSPI_FLSHCR) &
755 			    ~QUADSPI_FLSHCR_TDH_MASK,
756 			    base + QUADSPI_FLSHCR);
757 
758 	reg = qspi_readl(q, base + QUADSPI_SMPR);
759 	qspi_writel(q, reg & ~(QUADSPI_SMPR_FSDLY_MASK
760 			| QUADSPI_SMPR_FSPHS_MASK
761 			| QUADSPI_SMPR_HSENA_MASK
762 			| QUADSPI_SMPR_DDRSMP_MASK), base + QUADSPI_SMPR);
763 
764 	/* We only use the buffer3 for AHB read */
765 	qspi_writel(q, 0, base + QUADSPI_BUF0IND);
766 	qspi_writel(q, 0, base + QUADSPI_BUF1IND);
767 	qspi_writel(q, 0, base + QUADSPI_BUF2IND);
768 
769 	qspi_writel(q, QUADSPI_BFGENCR_SEQID(SEQID_LUT),
770 		    q->iobase + QUADSPI_BFGENCR);
771 	qspi_writel(q, QUADSPI_RBCT_WMRK_MASK, base + QUADSPI_RBCT);
772 	qspi_writel(q, QUADSPI_BUF3CR_ALLMST_MASK |
773 		    QUADSPI_BUF3CR_ADATSZ(q->devtype_data->ahb_buf_size / 8),
774 		    base + QUADSPI_BUF3CR);
775 
776 	if (needs_amba_base_offset(q))
777 		addr_offset = q->memmap_phy;
778 
779 	/*
780 	 * In HW there can be a maximum of four chips on two buses with
781 	 * two chip selects on each bus. We use four chip selects in SW
782 	 * to differentiate between the four chips.
783 	 * We use ahb_buf_size for each chip and set SFA1AD, SFA2AD, SFB1AD,
784 	 * SFB2AD accordingly.
785 	 */
786 	qspi_writel(q, q->devtype_data->ahb_buf_size + addr_offset,
787 		    base + QUADSPI_SFA1AD);
788 	qspi_writel(q, q->devtype_data->ahb_buf_size * 2 + addr_offset,
789 		    base + QUADSPI_SFA2AD);
790 	qspi_writel(q, q->devtype_data->ahb_buf_size * 3 + addr_offset,
791 		    base + QUADSPI_SFB1AD);
792 	qspi_writel(q, q->devtype_data->ahb_buf_size * 4 + addr_offset,
793 		    base + QUADSPI_SFB2AD);
794 
795 	q->selected = -1;
796 
797 	/* Enable the module */
798 	qspi_writel(q, QUADSPI_MCR_RESERVED_MASK | QUADSPI_MCR_END_CFG_MASK,
799 		    base + QUADSPI_MCR);
800 
801 	/* clear all interrupt status */
802 	qspi_writel(q, 0xffffffff, q->iobase + QUADSPI_FR);
803 
804 	/* enable the interrupt */
805 	qspi_writel(q, QUADSPI_RSER_TFIE, q->iobase + QUADSPI_RSER);
806 
807 	return 0;
808 }
809 
810 static const char *fsl_qspi_get_name(struct spi_mem *mem)
811 {
812 	struct fsl_qspi *q = spi_controller_get_devdata(mem->spi->master);
813 	struct device *dev = &mem->spi->dev;
814 	const char *name;
815 
816 	/*
817 	 * In order to keep mtdparts compatible with the old MTD driver at
818 	 * mtd/spi-nor/fsl-quadspi.c, we set a custom name derived from the
819 	 * platform_device of the controller.
820 	 */
821 	if (of_get_available_child_count(q->dev->of_node) == 1)
822 		return dev_name(q->dev);
823 
824 	name = devm_kasprintf(dev, GFP_KERNEL,
825 			      "%s-%d", dev_name(q->dev),
826 			      spi_get_chipselect(mem->spi, 0));
827 
828 	if (!name) {
829 		dev_err(dev, "failed to get memory for custom flash name\n");
830 		return ERR_PTR(-ENOMEM);
831 	}
832 
833 	return name;
834 }
835 
836 static const struct spi_controller_mem_ops fsl_qspi_mem_ops = {
837 	.adjust_op_size = fsl_qspi_adjust_op_size,
838 	.supports_op = fsl_qspi_supports_op,
839 	.exec_op = fsl_qspi_exec_op,
840 	.get_name = fsl_qspi_get_name,
841 };
842 
843 static int fsl_qspi_probe(struct platform_device *pdev)
844 {
845 	struct spi_controller *ctlr;
846 	struct device *dev = &pdev->dev;
847 	struct device_node *np = dev->of_node;
848 	struct resource *res;
849 	struct fsl_qspi *q;
850 	int ret;
851 
852 	ctlr = spi_alloc_master(&pdev->dev, sizeof(*q));
853 	if (!ctlr)
854 		return -ENOMEM;
855 
856 	ctlr->mode_bits = SPI_RX_DUAL | SPI_RX_QUAD |
857 			  SPI_TX_DUAL | SPI_TX_QUAD;
858 
859 	q = spi_controller_get_devdata(ctlr);
860 	q->dev = dev;
861 	q->devtype_data = of_device_get_match_data(dev);
862 	if (!q->devtype_data) {
863 		ret = -ENODEV;
864 		goto err_put_ctrl;
865 	}
866 
867 	platform_set_drvdata(pdev, q);
868 
869 	/* find the resources */
870 	q->iobase = devm_platform_ioremap_resource_byname(pdev, "QuadSPI");
871 	if (IS_ERR(q->iobase)) {
872 		ret = PTR_ERR(q->iobase);
873 		goto err_put_ctrl;
874 	}
875 
876 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
877 					"QuadSPI-memory");
878 	if (!res) {
879 		ret = -EINVAL;
880 		goto err_put_ctrl;
881 	}
882 	q->memmap_phy = res->start;
883 	/* Since there are 4 cs, map size required is 4 times ahb_buf_size */
884 	q->ahb_addr = devm_ioremap(dev, q->memmap_phy,
885 				   (q->devtype_data->ahb_buf_size * 4));
886 	if (!q->ahb_addr) {
887 		ret = -ENOMEM;
888 		goto err_put_ctrl;
889 	}
890 
891 	/* find the clocks */
892 	q->clk_en = devm_clk_get(dev, "qspi_en");
893 	if (IS_ERR(q->clk_en)) {
894 		ret = PTR_ERR(q->clk_en);
895 		goto err_put_ctrl;
896 	}
897 
898 	q->clk = devm_clk_get(dev, "qspi");
899 	if (IS_ERR(q->clk)) {
900 		ret = PTR_ERR(q->clk);
901 		goto err_put_ctrl;
902 	}
903 
904 	ret = fsl_qspi_clk_prep_enable(q);
905 	if (ret) {
906 		dev_err(dev, "can not enable the clock\n");
907 		goto err_put_ctrl;
908 	}
909 
910 	/* find the irq */
911 	ret = platform_get_irq(pdev, 0);
912 	if (ret < 0)
913 		goto err_disable_clk;
914 
915 	ret = devm_request_irq(dev, ret,
916 			fsl_qspi_irq_handler, 0, pdev->name, q);
917 	if (ret) {
918 		dev_err(dev, "failed to request irq: %d\n", ret);
919 		goto err_disable_clk;
920 	}
921 
922 	mutex_init(&q->lock);
923 
924 	ctlr->bus_num = -1;
925 	ctlr->num_chipselect = 4;
926 	ctlr->mem_ops = &fsl_qspi_mem_ops;
927 
928 	fsl_qspi_default_setup(q);
929 
930 	ctlr->dev.of_node = np;
931 
932 	ret = devm_spi_register_controller(dev, ctlr);
933 	if (ret)
934 		goto err_destroy_mutex;
935 
936 	return 0;
937 
938 err_destroy_mutex:
939 	mutex_destroy(&q->lock);
940 
941 err_disable_clk:
942 	fsl_qspi_clk_disable_unprep(q);
943 
944 err_put_ctrl:
945 	spi_controller_put(ctlr);
946 
947 	dev_err(dev, "Freescale QuadSPI probe failed\n");
948 	return ret;
949 }
950 
951 static void fsl_qspi_remove(struct platform_device *pdev)
952 {
953 	struct fsl_qspi *q = platform_get_drvdata(pdev);
954 
955 	/* disable the hardware */
956 	qspi_writel(q, QUADSPI_MCR_MDIS_MASK, q->iobase + QUADSPI_MCR);
957 	qspi_writel(q, 0x0, q->iobase + QUADSPI_RSER);
958 
959 	fsl_qspi_clk_disable_unprep(q);
960 
961 	mutex_destroy(&q->lock);
962 }
963 
964 static int fsl_qspi_suspend(struct device *dev)
965 {
966 	return 0;
967 }
968 
969 static int fsl_qspi_resume(struct device *dev)
970 {
971 	struct fsl_qspi *q = dev_get_drvdata(dev);
972 
973 	fsl_qspi_default_setup(q);
974 
975 	return 0;
976 }
977 
978 static const struct of_device_id fsl_qspi_dt_ids[] = {
979 	{ .compatible = "fsl,vf610-qspi", .data = &vybrid_data, },
980 	{ .compatible = "fsl,imx6sx-qspi", .data = &imx6sx_data, },
981 	{ .compatible = "fsl,imx7d-qspi", .data = &imx7d_data, },
982 	{ .compatible = "fsl,imx6ul-qspi", .data = &imx6ul_data, },
983 	{ .compatible = "fsl,ls1021a-qspi", .data = &ls1021a_data, },
984 	{ .compatible = "fsl,ls2080a-qspi", .data = &ls2080a_data, },
985 	{ /* sentinel */ }
986 };
987 MODULE_DEVICE_TABLE(of, fsl_qspi_dt_ids);
988 
989 static const struct dev_pm_ops fsl_qspi_pm_ops = {
990 	.suspend	= fsl_qspi_suspend,
991 	.resume		= fsl_qspi_resume,
992 };
993 
994 static struct platform_driver fsl_qspi_driver = {
995 	.driver = {
996 		.name	= "fsl-quadspi",
997 		.of_match_table = fsl_qspi_dt_ids,
998 		.pm =   &fsl_qspi_pm_ops,
999 	},
1000 	.probe          = fsl_qspi_probe,
1001 	.remove_new	= fsl_qspi_remove,
1002 };
1003 module_platform_driver(fsl_qspi_driver);
1004 
1005 MODULE_DESCRIPTION("Freescale QuadSPI Controller Driver");
1006 MODULE_AUTHOR("Freescale Semiconductor Inc.");
1007 MODULE_AUTHOR("Boris Brezillon <bbrezillon@kernel.org>");
1008 MODULE_AUTHOR("Frieder Schrempf <frieder.schrempf@kontron.de>");
1009 MODULE_AUTHOR("Yogesh Gaur <yogeshnarayan.gaur@nxp.com>");
1010 MODULE_AUTHOR("Suresh Gupta <suresh.gupta@nxp.com>");
1011 MODULE_LICENSE("GPL v2");
1012