xref: /openbmc/linux/drivers/spi/spi-mtk-nor.c (revision 27e45f2e)
1 // SPDX-License-Identifier: GPL-2.0
2 //
3 // Mediatek SPI NOR controller driver
4 //
5 // Copyright (C) 2020 Chuanhong Guo <gch981213@gmail.com>
6 
7 #include <linux/bits.h>
8 #include <linux/clk.h>
9 #include <linux/completion.h>
10 #include <linux/dma-mapping.h>
11 #include <linux/interrupt.h>
12 #include <linux/io.h>
13 #include <linux/iopoll.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/of_device.h>
17 #include <linux/pm_runtime.h>
18 #include <linux/spi/spi.h>
19 #include <linux/spi/spi-mem.h>
20 #include <linux/string.h>
21 
22 #define DRIVER_NAME "mtk-spi-nor"
23 
24 #define MTK_NOR_REG_CMD			0x00
25 #define MTK_NOR_CMD_WRITE		BIT(4)
26 #define MTK_NOR_CMD_PROGRAM		BIT(2)
27 #define MTK_NOR_CMD_READ		BIT(0)
28 #define MTK_NOR_CMD_MASK		GENMASK(5, 0)
29 
30 #define MTK_NOR_REG_PRG_CNT		0x04
31 #define MTK_NOR_PRG_CNT_MAX		56
32 #define MTK_NOR_REG_RDATA		0x0c
33 
34 #define MTK_NOR_REG_RADR0		0x10
35 #define MTK_NOR_REG_RADR(n)		(MTK_NOR_REG_RADR0 + 4 * (n))
36 #define MTK_NOR_REG_RADR3		0xc8
37 
38 #define MTK_NOR_REG_WDATA		0x1c
39 
40 #define MTK_NOR_REG_PRGDATA0		0x20
41 #define MTK_NOR_REG_PRGDATA(n)		(MTK_NOR_REG_PRGDATA0 + 4 * (n))
42 #define MTK_NOR_REG_PRGDATA_MAX		5
43 
44 #define MTK_NOR_REG_SHIFT0		0x38
45 #define MTK_NOR_REG_SHIFT(n)		(MTK_NOR_REG_SHIFT0 + 4 * (n))
46 #define MTK_NOR_REG_SHIFT_MAX		9
47 
48 #define MTK_NOR_REG_CFG1		0x60
49 #define MTK_NOR_FAST_READ		BIT(0)
50 
51 #define MTK_NOR_REG_CFG2		0x64
52 #define MTK_NOR_WR_CUSTOM_OP_EN		BIT(4)
53 #define MTK_NOR_WR_BUF_EN		BIT(0)
54 
55 #define MTK_NOR_REG_PP_DATA		0x98
56 
57 #define MTK_NOR_REG_IRQ_STAT		0xa8
58 #define MTK_NOR_REG_IRQ_EN		0xac
59 #define MTK_NOR_IRQ_DMA			BIT(7)
60 #define MTK_NOR_IRQ_MASK		GENMASK(7, 0)
61 
62 #define MTK_NOR_REG_CFG3		0xb4
63 #define MTK_NOR_DISABLE_WREN		BIT(7)
64 #define MTK_NOR_DISABLE_SR_POLL		BIT(5)
65 
66 #define MTK_NOR_REG_WP			0xc4
67 #define MTK_NOR_ENABLE_SF_CMD		0x30
68 
69 #define MTK_NOR_REG_BUSCFG		0xcc
70 #define MTK_NOR_4B_ADDR			BIT(4)
71 #define MTK_NOR_QUAD_ADDR		BIT(3)
72 #define MTK_NOR_QUAD_READ		BIT(2)
73 #define MTK_NOR_DUAL_ADDR		BIT(1)
74 #define MTK_NOR_DUAL_READ		BIT(0)
75 #define MTK_NOR_BUS_MODE_MASK		GENMASK(4, 0)
76 
77 #define MTK_NOR_REG_DMA_CTL		0x718
78 #define MTK_NOR_DMA_START		BIT(0)
79 
80 #define MTK_NOR_REG_DMA_FADR		0x71c
81 #define MTK_NOR_REG_DMA_DADR		0x720
82 #define MTK_NOR_REG_DMA_END_DADR	0x724
83 #define MTK_NOR_REG_CG_DIS		0x728
84 #define MTK_NOR_SFC_SW_RST		BIT(2)
85 
86 #define MTK_NOR_REG_DMA_DADR_HB		0x738
87 #define MTK_NOR_REG_DMA_END_DADR_HB	0x73c
88 
89 #define MTK_NOR_PRG_MAX_SIZE		6
90 // Reading DMA src/dst addresses have to be 16-byte aligned
91 #define MTK_NOR_DMA_ALIGN		16
92 #define MTK_NOR_DMA_ALIGN_MASK		(MTK_NOR_DMA_ALIGN - 1)
93 // and we allocate a bounce buffer if destination address isn't aligned.
94 #define MTK_NOR_BOUNCE_BUF_SIZE		PAGE_SIZE
95 
96 // Buffered page program can do one 128-byte transfer
97 #define MTK_NOR_PP_SIZE			128
98 
99 #define CLK_TO_US(sp, clkcnt)		DIV_ROUND_UP(clkcnt, sp->spi_freq / 1000000)
100 
101 struct mtk_nor_caps {
102 	u8 dma_bits;
103 
104 	/* extra_dummy_bit is adding for the IP of new SoCs.
105 	 * Some new SoCs modify the timing of fetching registers' values
106 	 * and IDs of nor flash, they need a extra_dummy_bit which can add
107 	 * more clock cycles for fetching data.
108 	 */
109 	u8 extra_dummy_bit;
110 };
111 
112 struct mtk_nor {
113 	struct spi_controller *ctlr;
114 	struct device *dev;
115 	void __iomem *base;
116 	u8 *buffer;
117 	dma_addr_t buffer_dma;
118 	struct clk *spi_clk;
119 	struct clk *ctlr_clk;
120 	struct clk *axi_clk;
121 	struct clk *axi_s_clk;
122 	unsigned int spi_freq;
123 	bool wbuf_en;
124 	bool has_irq;
125 	bool high_dma;
126 	struct completion op_done;
127 	const struct mtk_nor_caps *caps;
128 };
129 
130 static inline void mtk_nor_rmw(struct mtk_nor *sp, u32 reg, u32 set, u32 clr)
131 {
132 	u32 val = readl(sp->base + reg);
133 
134 	val &= ~clr;
135 	val |= set;
136 	writel(val, sp->base + reg);
137 }
138 
139 static inline int mtk_nor_cmd_exec(struct mtk_nor *sp, u32 cmd, ulong clk)
140 {
141 	ulong delay = CLK_TO_US(sp, clk);
142 	u32 reg;
143 	int ret;
144 
145 	writel(cmd, sp->base + MTK_NOR_REG_CMD);
146 	ret = readl_poll_timeout(sp->base + MTK_NOR_REG_CMD, reg, !(reg & cmd),
147 				 delay / 3, (delay + 1) * 200);
148 	if (ret < 0)
149 		dev_err(sp->dev, "command %u timeout.\n", cmd);
150 	return ret;
151 }
152 
153 static void mtk_nor_reset(struct mtk_nor *sp)
154 {
155 	mtk_nor_rmw(sp, MTK_NOR_REG_CG_DIS, 0, MTK_NOR_SFC_SW_RST);
156 	mb(); /* flush previous writes */
157 	mtk_nor_rmw(sp, MTK_NOR_REG_CG_DIS, MTK_NOR_SFC_SW_RST, 0);
158 	mb(); /* flush previous writes */
159 	writel(MTK_NOR_ENABLE_SF_CMD, sp->base + MTK_NOR_REG_WP);
160 }
161 
162 static void mtk_nor_set_addr(struct mtk_nor *sp, const struct spi_mem_op *op)
163 {
164 	u32 addr = op->addr.val;
165 	int i;
166 
167 	for (i = 0; i < 3; i++) {
168 		writeb(addr & 0xff, sp->base + MTK_NOR_REG_RADR(i));
169 		addr >>= 8;
170 	}
171 	if (op->addr.nbytes == 4) {
172 		writeb(addr & 0xff, sp->base + MTK_NOR_REG_RADR3);
173 		mtk_nor_rmw(sp, MTK_NOR_REG_BUSCFG, MTK_NOR_4B_ADDR, 0);
174 	} else {
175 		mtk_nor_rmw(sp, MTK_NOR_REG_BUSCFG, 0, MTK_NOR_4B_ADDR);
176 	}
177 }
178 
179 static bool need_bounce(struct mtk_nor *sp, const struct spi_mem_op *op)
180 {
181 	return ((uintptr_t)op->data.buf.in & MTK_NOR_DMA_ALIGN_MASK);
182 }
183 
184 static bool mtk_nor_match_read(const struct spi_mem_op *op)
185 {
186 	int dummy = 0;
187 
188 	if (op->dummy.nbytes)
189 		dummy = op->dummy.nbytes * BITS_PER_BYTE / op->dummy.buswidth;
190 
191 	if ((op->data.buswidth == 2) || (op->data.buswidth == 4)) {
192 		if (op->addr.buswidth == 1)
193 			return dummy == 8;
194 		else if (op->addr.buswidth == 2)
195 			return dummy == 4;
196 		else if (op->addr.buswidth == 4)
197 			return dummy == 6;
198 	} else if ((op->addr.buswidth == 1) && (op->data.buswidth == 1)) {
199 		if (op->cmd.opcode == 0x03)
200 			return dummy == 0;
201 		else if (op->cmd.opcode == 0x0b)
202 			return dummy == 8;
203 	}
204 	return false;
205 }
206 
207 static bool mtk_nor_match_prg(const struct spi_mem_op *op)
208 {
209 	int tx_len, rx_len, prg_len, prg_left;
210 
211 	// prg mode is spi-only.
212 	if ((op->cmd.buswidth > 1) || (op->addr.buswidth > 1) ||
213 	    (op->dummy.buswidth > 1) || (op->data.buswidth > 1))
214 		return false;
215 
216 	tx_len = op->cmd.nbytes + op->addr.nbytes;
217 
218 	if (op->data.dir == SPI_MEM_DATA_OUT) {
219 		// count dummy bytes only if we need to write data after it
220 		tx_len += op->dummy.nbytes;
221 
222 		// leave at least one byte for data
223 		if (tx_len > MTK_NOR_REG_PRGDATA_MAX)
224 			return false;
225 
226 		// if there's no addr, meaning adjust_op_size is impossible,
227 		// check data length as well.
228 		if ((!op->addr.nbytes) &&
229 		    (tx_len + op->data.nbytes > MTK_NOR_REG_PRGDATA_MAX + 1))
230 			return false;
231 	} else if (op->data.dir == SPI_MEM_DATA_IN) {
232 		if (tx_len > MTK_NOR_REG_PRGDATA_MAX + 1)
233 			return false;
234 
235 		rx_len = op->data.nbytes;
236 		prg_left = MTK_NOR_PRG_CNT_MAX / 8 - tx_len - op->dummy.nbytes;
237 		if (prg_left > MTK_NOR_REG_SHIFT_MAX + 1)
238 			prg_left = MTK_NOR_REG_SHIFT_MAX + 1;
239 		if (rx_len > prg_left) {
240 			if (!op->addr.nbytes)
241 				return false;
242 			rx_len = prg_left;
243 		}
244 
245 		prg_len = tx_len + op->dummy.nbytes + rx_len;
246 		if (prg_len > MTK_NOR_PRG_CNT_MAX / 8)
247 			return false;
248 	} else {
249 		prg_len = tx_len + op->dummy.nbytes;
250 		if (prg_len > MTK_NOR_PRG_CNT_MAX / 8)
251 			return false;
252 	}
253 	return true;
254 }
255 
256 static void mtk_nor_adj_prg_size(struct spi_mem_op *op)
257 {
258 	int tx_len, tx_left, prg_left;
259 
260 	tx_len = op->cmd.nbytes + op->addr.nbytes;
261 	if (op->data.dir == SPI_MEM_DATA_OUT) {
262 		tx_len += op->dummy.nbytes;
263 		tx_left = MTK_NOR_REG_PRGDATA_MAX + 1 - tx_len;
264 		if (op->data.nbytes > tx_left)
265 			op->data.nbytes = tx_left;
266 	} else if (op->data.dir == SPI_MEM_DATA_IN) {
267 		prg_left = MTK_NOR_PRG_CNT_MAX / 8 - tx_len - op->dummy.nbytes;
268 		if (prg_left > MTK_NOR_REG_SHIFT_MAX + 1)
269 			prg_left = MTK_NOR_REG_SHIFT_MAX + 1;
270 		if (op->data.nbytes > prg_left)
271 			op->data.nbytes = prg_left;
272 	}
273 }
274 
275 static int mtk_nor_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
276 {
277 	struct mtk_nor *sp = spi_controller_get_devdata(mem->spi->master);
278 
279 	if (!op->data.nbytes)
280 		return 0;
281 
282 	if ((op->addr.nbytes == 3) || (op->addr.nbytes == 4)) {
283 		if ((op->data.dir == SPI_MEM_DATA_IN) &&
284 		    mtk_nor_match_read(op)) {
285 			// limit size to prevent timeout calculation overflow
286 			if (op->data.nbytes > 0x400000)
287 				op->data.nbytes = 0x400000;
288 
289 			if ((op->addr.val & MTK_NOR_DMA_ALIGN_MASK) ||
290 			    (op->data.nbytes < MTK_NOR_DMA_ALIGN))
291 				op->data.nbytes = 1;
292 			else if (!need_bounce(sp, op))
293 				op->data.nbytes &= ~MTK_NOR_DMA_ALIGN_MASK;
294 			else if (op->data.nbytes > MTK_NOR_BOUNCE_BUF_SIZE)
295 				op->data.nbytes = MTK_NOR_BOUNCE_BUF_SIZE;
296 			return 0;
297 		} else if (op->data.dir == SPI_MEM_DATA_OUT) {
298 			if (op->data.nbytes >= MTK_NOR_PP_SIZE)
299 				op->data.nbytes = MTK_NOR_PP_SIZE;
300 			else
301 				op->data.nbytes = 1;
302 			return 0;
303 		}
304 	}
305 
306 	mtk_nor_adj_prg_size(op);
307 	return 0;
308 }
309 
310 static bool mtk_nor_supports_op(struct spi_mem *mem,
311 				const struct spi_mem_op *op)
312 {
313 	if (!spi_mem_default_supports_op(mem, op))
314 		return false;
315 
316 	if (op->cmd.buswidth != 1)
317 		return false;
318 
319 	if ((op->addr.nbytes == 3) || (op->addr.nbytes == 4)) {
320 		switch (op->data.dir) {
321 		case SPI_MEM_DATA_IN:
322 			if (mtk_nor_match_read(op))
323 				return true;
324 			break;
325 		case SPI_MEM_DATA_OUT:
326 			if ((op->addr.buswidth == 1) &&
327 			    (op->dummy.nbytes == 0) &&
328 			    (op->data.buswidth == 1))
329 				return true;
330 			break;
331 		default:
332 			break;
333 		}
334 	}
335 
336 	return mtk_nor_match_prg(op);
337 }
338 
339 static void mtk_nor_setup_bus(struct mtk_nor *sp, const struct spi_mem_op *op)
340 {
341 	u32 reg = 0;
342 
343 	if (op->addr.nbytes == 4)
344 		reg |= MTK_NOR_4B_ADDR;
345 
346 	if (op->data.buswidth == 4) {
347 		reg |= MTK_NOR_QUAD_READ;
348 		writeb(op->cmd.opcode, sp->base + MTK_NOR_REG_PRGDATA(4));
349 		if (op->addr.buswidth == 4)
350 			reg |= MTK_NOR_QUAD_ADDR;
351 	} else if (op->data.buswidth == 2) {
352 		reg |= MTK_NOR_DUAL_READ;
353 		writeb(op->cmd.opcode, sp->base + MTK_NOR_REG_PRGDATA(3));
354 		if (op->addr.buswidth == 2)
355 			reg |= MTK_NOR_DUAL_ADDR;
356 	} else {
357 		if (op->cmd.opcode == 0x0b)
358 			mtk_nor_rmw(sp, MTK_NOR_REG_CFG1, MTK_NOR_FAST_READ, 0);
359 		else
360 			mtk_nor_rmw(sp, MTK_NOR_REG_CFG1, 0, MTK_NOR_FAST_READ);
361 	}
362 	mtk_nor_rmw(sp, MTK_NOR_REG_BUSCFG, reg, MTK_NOR_BUS_MODE_MASK);
363 }
364 
365 static int mtk_nor_dma_exec(struct mtk_nor *sp, u32 from, unsigned int length,
366 			    dma_addr_t dma_addr)
367 {
368 	int ret = 0;
369 	u32 delay, timeout;
370 	u32 reg;
371 
372 	writel(from, sp->base + MTK_NOR_REG_DMA_FADR);
373 	writel(dma_addr, sp->base + MTK_NOR_REG_DMA_DADR);
374 	writel(dma_addr + length, sp->base + MTK_NOR_REG_DMA_END_DADR);
375 
376 	if (sp->high_dma) {
377 		writel(upper_32_bits(dma_addr),
378 		       sp->base + MTK_NOR_REG_DMA_DADR_HB);
379 		writel(upper_32_bits(dma_addr + length),
380 		       sp->base + MTK_NOR_REG_DMA_END_DADR_HB);
381 	}
382 
383 	if (sp->has_irq) {
384 		reinit_completion(&sp->op_done);
385 		mtk_nor_rmw(sp, MTK_NOR_REG_IRQ_EN, MTK_NOR_IRQ_DMA, 0);
386 	}
387 
388 	mtk_nor_rmw(sp, MTK_NOR_REG_DMA_CTL, MTK_NOR_DMA_START, 0);
389 
390 	delay = CLK_TO_US(sp, (length + 5) * BITS_PER_BYTE);
391 	timeout = (delay + 1) * 100;
392 
393 	if (sp->has_irq) {
394 		if (!wait_for_completion_timeout(&sp->op_done,
395 		    usecs_to_jiffies(max(timeout, 10000U))))
396 			ret = -ETIMEDOUT;
397 	} else {
398 		ret = readl_poll_timeout(sp->base + MTK_NOR_REG_DMA_CTL, reg,
399 					 !(reg & MTK_NOR_DMA_START), delay / 3,
400 					 timeout);
401 	}
402 
403 	if (ret < 0)
404 		dev_err(sp->dev, "dma read timeout.\n");
405 
406 	return ret;
407 }
408 
409 static int mtk_nor_read_bounce(struct mtk_nor *sp, const struct spi_mem_op *op)
410 {
411 	unsigned int rdlen;
412 	int ret;
413 
414 	if (op->data.nbytes & MTK_NOR_DMA_ALIGN_MASK)
415 		rdlen = (op->data.nbytes + MTK_NOR_DMA_ALIGN) & ~MTK_NOR_DMA_ALIGN_MASK;
416 	else
417 		rdlen = op->data.nbytes;
418 
419 	ret = mtk_nor_dma_exec(sp, op->addr.val, rdlen, sp->buffer_dma);
420 
421 	if (!ret)
422 		memcpy(op->data.buf.in, sp->buffer, op->data.nbytes);
423 
424 	return ret;
425 }
426 
427 static int mtk_nor_read_dma(struct mtk_nor *sp, const struct spi_mem_op *op)
428 {
429 	int ret;
430 	dma_addr_t dma_addr;
431 
432 	if (need_bounce(sp, op))
433 		return mtk_nor_read_bounce(sp, op);
434 
435 	dma_addr = dma_map_single(sp->dev, op->data.buf.in,
436 				  op->data.nbytes, DMA_FROM_DEVICE);
437 
438 	if (dma_mapping_error(sp->dev, dma_addr))
439 		return -EINVAL;
440 
441 	ret = mtk_nor_dma_exec(sp, op->addr.val, op->data.nbytes, dma_addr);
442 
443 	dma_unmap_single(sp->dev, dma_addr, op->data.nbytes, DMA_FROM_DEVICE);
444 
445 	return ret;
446 }
447 
448 static int mtk_nor_read_pio(struct mtk_nor *sp, const struct spi_mem_op *op)
449 {
450 	u8 *buf = op->data.buf.in;
451 	int ret;
452 
453 	ret = mtk_nor_cmd_exec(sp, MTK_NOR_CMD_READ, 6 * BITS_PER_BYTE);
454 	if (!ret)
455 		buf[0] = readb(sp->base + MTK_NOR_REG_RDATA);
456 	return ret;
457 }
458 
459 static int mtk_nor_setup_write_buffer(struct mtk_nor *sp, bool on)
460 {
461 	int ret;
462 	u32 val;
463 
464 	if (!(sp->wbuf_en ^ on))
465 		return 0;
466 
467 	val = readl(sp->base + MTK_NOR_REG_CFG2);
468 	if (on) {
469 		writel(val | MTK_NOR_WR_BUF_EN, sp->base + MTK_NOR_REG_CFG2);
470 		ret = readl_poll_timeout(sp->base + MTK_NOR_REG_CFG2, val,
471 					 val & MTK_NOR_WR_BUF_EN, 0, 10000);
472 	} else {
473 		writel(val & ~MTK_NOR_WR_BUF_EN, sp->base + MTK_NOR_REG_CFG2);
474 		ret = readl_poll_timeout(sp->base + MTK_NOR_REG_CFG2, val,
475 					 !(val & MTK_NOR_WR_BUF_EN), 0, 10000);
476 	}
477 
478 	if (!ret)
479 		sp->wbuf_en = on;
480 
481 	return ret;
482 }
483 
484 static int mtk_nor_pp_buffered(struct mtk_nor *sp, const struct spi_mem_op *op)
485 {
486 	const u8 *buf = op->data.buf.out;
487 	u32 val;
488 	int ret, i;
489 
490 	ret = mtk_nor_setup_write_buffer(sp, true);
491 	if (ret < 0)
492 		return ret;
493 
494 	for (i = 0; i < op->data.nbytes; i += 4) {
495 		val = buf[i + 3] << 24 | buf[i + 2] << 16 | buf[i + 1] << 8 |
496 		      buf[i];
497 		writel(val, sp->base + MTK_NOR_REG_PP_DATA);
498 	}
499 	return mtk_nor_cmd_exec(sp, MTK_NOR_CMD_WRITE,
500 				(op->data.nbytes + 5) * BITS_PER_BYTE);
501 }
502 
503 static int mtk_nor_pp_unbuffered(struct mtk_nor *sp,
504 				 const struct spi_mem_op *op)
505 {
506 	const u8 *buf = op->data.buf.out;
507 	int ret;
508 
509 	ret = mtk_nor_setup_write_buffer(sp, false);
510 	if (ret < 0)
511 		return ret;
512 	writeb(buf[0], sp->base + MTK_NOR_REG_WDATA);
513 	return mtk_nor_cmd_exec(sp, MTK_NOR_CMD_WRITE, 6 * BITS_PER_BYTE);
514 }
515 
516 static int mtk_nor_spi_mem_prg(struct mtk_nor *sp, const struct spi_mem_op *op)
517 {
518 	int rx_len = 0;
519 	int reg_offset = MTK_NOR_REG_PRGDATA_MAX;
520 	int tx_len, prg_len;
521 	int i, ret;
522 	void __iomem *reg;
523 	u8 bufbyte;
524 
525 	tx_len = op->cmd.nbytes + op->addr.nbytes;
526 
527 	// count dummy bytes only if we need to write data after it
528 	if (op->data.dir == SPI_MEM_DATA_OUT)
529 		tx_len += op->dummy.nbytes + op->data.nbytes;
530 	else if (op->data.dir == SPI_MEM_DATA_IN)
531 		rx_len = op->data.nbytes;
532 
533 	prg_len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes +
534 		  op->data.nbytes;
535 
536 	// an invalid op may reach here if the caller calls exec_op without
537 	// adjust_op_size. return -EINVAL instead of -ENOTSUPP so that
538 	// spi-mem won't try this op again with generic spi transfers.
539 	if ((tx_len > MTK_NOR_REG_PRGDATA_MAX + 1) ||
540 	    (rx_len > MTK_NOR_REG_SHIFT_MAX + 1) ||
541 	    (prg_len > MTK_NOR_PRG_CNT_MAX / 8))
542 		return -EINVAL;
543 
544 	// fill tx data
545 	for (i = op->cmd.nbytes; i > 0; i--, reg_offset--) {
546 		reg = sp->base + MTK_NOR_REG_PRGDATA(reg_offset);
547 		bufbyte = (op->cmd.opcode >> ((i - 1) * BITS_PER_BYTE)) & 0xff;
548 		writeb(bufbyte, reg);
549 	}
550 
551 	for (i = op->addr.nbytes; i > 0; i--, reg_offset--) {
552 		reg = sp->base + MTK_NOR_REG_PRGDATA(reg_offset);
553 		bufbyte = (op->addr.val >> ((i - 1) * BITS_PER_BYTE)) & 0xff;
554 		writeb(bufbyte, reg);
555 	}
556 
557 	if (op->data.dir == SPI_MEM_DATA_OUT) {
558 		for (i = 0; i < op->dummy.nbytes; i++, reg_offset--) {
559 			reg = sp->base + MTK_NOR_REG_PRGDATA(reg_offset);
560 			writeb(0, reg);
561 		}
562 
563 		for (i = 0; i < op->data.nbytes; i++, reg_offset--) {
564 			reg = sp->base + MTK_NOR_REG_PRGDATA(reg_offset);
565 			writeb(((const u8 *)(op->data.buf.out))[i], reg);
566 		}
567 	}
568 
569 	for (; reg_offset >= 0; reg_offset--) {
570 		reg = sp->base + MTK_NOR_REG_PRGDATA(reg_offset);
571 		writeb(0, reg);
572 	}
573 
574 	// trigger op
575 	if (rx_len)
576 		writel(prg_len * BITS_PER_BYTE + sp->caps->extra_dummy_bit,
577 		       sp->base + MTK_NOR_REG_PRG_CNT);
578 	else
579 		writel(prg_len * BITS_PER_BYTE, sp->base + MTK_NOR_REG_PRG_CNT);
580 
581 	ret = mtk_nor_cmd_exec(sp, MTK_NOR_CMD_PROGRAM,
582 			       prg_len * BITS_PER_BYTE);
583 	if (ret)
584 		return ret;
585 
586 	// fetch read data
587 	reg_offset = 0;
588 	if (op->data.dir == SPI_MEM_DATA_IN) {
589 		for (i = op->data.nbytes - 1; i >= 0; i--, reg_offset++) {
590 			reg = sp->base + MTK_NOR_REG_SHIFT(reg_offset);
591 			((u8 *)(op->data.buf.in))[i] = readb(reg);
592 		}
593 	}
594 
595 	return 0;
596 }
597 
598 static int mtk_nor_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
599 {
600 	struct mtk_nor *sp = spi_controller_get_devdata(mem->spi->master);
601 	int ret;
602 
603 	if ((op->data.nbytes == 0) ||
604 	    ((op->addr.nbytes != 3) && (op->addr.nbytes != 4)))
605 		return mtk_nor_spi_mem_prg(sp, op);
606 
607 	if (op->data.dir == SPI_MEM_DATA_OUT) {
608 		mtk_nor_set_addr(sp, op);
609 		writeb(op->cmd.opcode, sp->base + MTK_NOR_REG_PRGDATA0);
610 		if (op->data.nbytes == MTK_NOR_PP_SIZE)
611 			return mtk_nor_pp_buffered(sp, op);
612 		return mtk_nor_pp_unbuffered(sp, op);
613 	}
614 
615 	if ((op->data.dir == SPI_MEM_DATA_IN) && mtk_nor_match_read(op)) {
616 		ret = mtk_nor_setup_write_buffer(sp, false);
617 		if (ret < 0)
618 			return ret;
619 		mtk_nor_setup_bus(sp, op);
620 		if (op->data.nbytes == 1) {
621 			mtk_nor_set_addr(sp, op);
622 			return mtk_nor_read_pio(sp, op);
623 		} else {
624 			ret = mtk_nor_read_dma(sp, op);
625 			if (unlikely(ret)) {
626 				/* Handle rare bus glitch */
627 				mtk_nor_reset(sp);
628 				mtk_nor_setup_bus(sp, op);
629 				return mtk_nor_read_dma(sp, op);
630 			}
631 
632 			return ret;
633 		}
634 	}
635 
636 	return mtk_nor_spi_mem_prg(sp, op);
637 }
638 
639 static int mtk_nor_setup(struct spi_device *spi)
640 {
641 	struct mtk_nor *sp = spi_controller_get_devdata(spi->master);
642 
643 	if (spi->max_speed_hz && (spi->max_speed_hz < sp->spi_freq)) {
644 		dev_err(&spi->dev, "spi clock should be %u Hz.\n",
645 			sp->spi_freq);
646 		return -EINVAL;
647 	}
648 	spi->max_speed_hz = sp->spi_freq;
649 
650 	return 0;
651 }
652 
653 static int mtk_nor_transfer_one_message(struct spi_controller *master,
654 					struct spi_message *m)
655 {
656 	struct mtk_nor *sp = spi_controller_get_devdata(master);
657 	struct spi_transfer *t = NULL;
658 	unsigned long trx_len = 0;
659 	int stat = 0;
660 	int reg_offset = MTK_NOR_REG_PRGDATA_MAX;
661 	void __iomem *reg;
662 	const u8 *txbuf;
663 	u8 *rxbuf;
664 	int i;
665 
666 	list_for_each_entry(t, &m->transfers, transfer_list) {
667 		txbuf = t->tx_buf;
668 		for (i = 0; i < t->len; i++, reg_offset--) {
669 			reg = sp->base + MTK_NOR_REG_PRGDATA(reg_offset);
670 			if (txbuf)
671 				writeb(txbuf[i], reg);
672 			else
673 				writeb(0, reg);
674 		}
675 		trx_len += t->len;
676 	}
677 
678 	writel(trx_len * BITS_PER_BYTE, sp->base + MTK_NOR_REG_PRG_CNT);
679 
680 	stat = mtk_nor_cmd_exec(sp, MTK_NOR_CMD_PROGRAM,
681 				trx_len * BITS_PER_BYTE);
682 	if (stat < 0)
683 		goto msg_done;
684 
685 	reg_offset = trx_len - 1;
686 	list_for_each_entry(t, &m->transfers, transfer_list) {
687 		rxbuf = t->rx_buf;
688 		for (i = 0; i < t->len; i++, reg_offset--) {
689 			reg = sp->base + MTK_NOR_REG_SHIFT(reg_offset);
690 			if (rxbuf)
691 				rxbuf[i] = readb(reg);
692 		}
693 	}
694 
695 	m->actual_length = trx_len;
696 msg_done:
697 	m->status = stat;
698 	spi_finalize_current_message(master);
699 
700 	return 0;
701 }
702 
703 static void mtk_nor_disable_clk(struct mtk_nor *sp)
704 {
705 	clk_disable_unprepare(sp->spi_clk);
706 	clk_disable_unprepare(sp->ctlr_clk);
707 	clk_disable_unprepare(sp->axi_clk);
708 	clk_disable_unprepare(sp->axi_s_clk);
709 }
710 
711 static int mtk_nor_enable_clk(struct mtk_nor *sp)
712 {
713 	int ret;
714 
715 	ret = clk_prepare_enable(sp->spi_clk);
716 	if (ret)
717 		return ret;
718 
719 	ret = clk_prepare_enable(sp->ctlr_clk);
720 	if (ret) {
721 		clk_disable_unprepare(sp->spi_clk);
722 		return ret;
723 	}
724 
725 	ret = clk_prepare_enable(sp->axi_clk);
726 	if (ret) {
727 		clk_disable_unprepare(sp->spi_clk);
728 		clk_disable_unprepare(sp->ctlr_clk);
729 		return ret;
730 	}
731 
732 	ret = clk_prepare_enable(sp->axi_s_clk);
733 	if (ret) {
734 		clk_disable_unprepare(sp->spi_clk);
735 		clk_disable_unprepare(sp->ctlr_clk);
736 		clk_disable_unprepare(sp->axi_clk);
737 		return ret;
738 	}
739 
740 	return 0;
741 }
742 
743 static void mtk_nor_init(struct mtk_nor *sp)
744 {
745 	writel(0, sp->base + MTK_NOR_REG_IRQ_EN);
746 	writel(MTK_NOR_IRQ_MASK, sp->base + MTK_NOR_REG_IRQ_STAT);
747 
748 	writel(MTK_NOR_ENABLE_SF_CMD, sp->base + MTK_NOR_REG_WP);
749 	mtk_nor_rmw(sp, MTK_NOR_REG_CFG2, MTK_NOR_WR_CUSTOM_OP_EN, 0);
750 	mtk_nor_rmw(sp, MTK_NOR_REG_CFG3,
751 		    MTK_NOR_DISABLE_WREN | MTK_NOR_DISABLE_SR_POLL, 0);
752 }
753 
754 static irqreturn_t mtk_nor_irq_handler(int irq, void *data)
755 {
756 	struct mtk_nor *sp = data;
757 	u32 irq_status, irq_enabled;
758 
759 	irq_status = readl(sp->base + MTK_NOR_REG_IRQ_STAT);
760 	irq_enabled = readl(sp->base + MTK_NOR_REG_IRQ_EN);
761 	// write status back to clear interrupt
762 	writel(irq_status, sp->base + MTK_NOR_REG_IRQ_STAT);
763 
764 	if (!(irq_status & irq_enabled))
765 		return IRQ_NONE;
766 
767 	if (irq_status & MTK_NOR_IRQ_DMA) {
768 		complete(&sp->op_done);
769 		writel(0, sp->base + MTK_NOR_REG_IRQ_EN);
770 	}
771 
772 	return IRQ_HANDLED;
773 }
774 
775 static size_t mtk_max_msg_size(struct spi_device *spi)
776 {
777 	return MTK_NOR_PRG_MAX_SIZE;
778 }
779 
780 static const struct spi_controller_mem_ops mtk_nor_mem_ops = {
781 	.adjust_op_size = mtk_nor_adjust_op_size,
782 	.supports_op = mtk_nor_supports_op,
783 	.exec_op = mtk_nor_exec_op
784 };
785 
786 static const struct mtk_nor_caps mtk_nor_caps_mt8173 = {
787 	.dma_bits = 32,
788 	.extra_dummy_bit = 0,
789 };
790 
791 static const struct mtk_nor_caps mtk_nor_caps_mt8186 = {
792 	.dma_bits = 32,
793 	.extra_dummy_bit = 1,
794 };
795 
796 static const struct mtk_nor_caps mtk_nor_caps_mt8192 = {
797 	.dma_bits = 36,
798 	.extra_dummy_bit = 0,
799 };
800 
801 static const struct of_device_id mtk_nor_match[] = {
802 	{ .compatible = "mediatek,mt8173-nor", .data = &mtk_nor_caps_mt8173 },
803 	{ .compatible = "mediatek,mt8186-nor", .data = &mtk_nor_caps_mt8186 },
804 	{ .compatible = "mediatek,mt8192-nor", .data = &mtk_nor_caps_mt8192 },
805 	{ /* sentinel */ }
806 };
807 MODULE_DEVICE_TABLE(of, mtk_nor_match);
808 
809 static int mtk_nor_probe(struct platform_device *pdev)
810 {
811 	struct spi_controller *ctlr;
812 	struct mtk_nor *sp;
813 	struct mtk_nor_caps *caps;
814 	void __iomem *base;
815 	struct clk *spi_clk, *ctlr_clk, *axi_clk, *axi_s_clk;
816 	int ret, irq;
817 
818 	base = devm_platform_ioremap_resource(pdev, 0);
819 	if (IS_ERR(base))
820 		return PTR_ERR(base);
821 
822 	spi_clk = devm_clk_get(&pdev->dev, "spi");
823 	if (IS_ERR(spi_clk))
824 		return PTR_ERR(spi_clk);
825 
826 	ctlr_clk = devm_clk_get(&pdev->dev, "sf");
827 	if (IS_ERR(ctlr_clk))
828 		return PTR_ERR(ctlr_clk);
829 
830 	axi_clk = devm_clk_get_optional(&pdev->dev, "axi");
831 	if (IS_ERR(axi_clk))
832 		return PTR_ERR(axi_clk);
833 
834 	axi_s_clk = devm_clk_get_optional(&pdev->dev, "axi_s");
835 	if (IS_ERR(axi_s_clk))
836 		return PTR_ERR(axi_s_clk);
837 
838 	caps = (struct mtk_nor_caps *)of_device_get_match_data(&pdev->dev);
839 
840 	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(caps->dma_bits));
841 	if (ret) {
842 		dev_err(&pdev->dev, "failed to set dma mask(%u)\n", caps->dma_bits);
843 		return ret;
844 	}
845 
846 	ctlr = devm_spi_alloc_master(&pdev->dev, sizeof(*sp));
847 	if (!ctlr) {
848 		dev_err(&pdev->dev, "failed to allocate spi controller\n");
849 		return -ENOMEM;
850 	}
851 
852 	ctlr->bits_per_word_mask = SPI_BPW_MASK(8);
853 	ctlr->dev.of_node = pdev->dev.of_node;
854 	ctlr->max_message_size = mtk_max_msg_size;
855 	ctlr->mem_ops = &mtk_nor_mem_ops;
856 	ctlr->mode_bits = SPI_RX_DUAL | SPI_RX_QUAD | SPI_TX_DUAL | SPI_TX_QUAD;
857 	ctlr->num_chipselect = 1;
858 	ctlr->setup = mtk_nor_setup;
859 	ctlr->transfer_one_message = mtk_nor_transfer_one_message;
860 	ctlr->auto_runtime_pm = true;
861 
862 	dev_set_drvdata(&pdev->dev, ctlr);
863 
864 	sp = spi_controller_get_devdata(ctlr);
865 	sp->base = base;
866 	sp->has_irq = false;
867 	sp->wbuf_en = false;
868 	sp->ctlr = ctlr;
869 	sp->dev = &pdev->dev;
870 	sp->spi_clk = spi_clk;
871 	sp->ctlr_clk = ctlr_clk;
872 	sp->axi_clk = axi_clk;
873 	sp->axi_s_clk = axi_s_clk;
874 	sp->caps = caps;
875 	sp->high_dma = caps->dma_bits > 32;
876 	sp->buffer = dmam_alloc_coherent(&pdev->dev,
877 				MTK_NOR_BOUNCE_BUF_SIZE + MTK_NOR_DMA_ALIGN,
878 				&sp->buffer_dma, GFP_KERNEL);
879 	if (!sp->buffer)
880 		return -ENOMEM;
881 
882 	if ((uintptr_t)sp->buffer & MTK_NOR_DMA_ALIGN_MASK) {
883 		dev_err(sp->dev, "misaligned allocation of internal buffer.\n");
884 		return -ENOMEM;
885 	}
886 
887 	ret = mtk_nor_enable_clk(sp);
888 	if (ret < 0)
889 		return ret;
890 
891 	sp->spi_freq = clk_get_rate(sp->spi_clk);
892 
893 	mtk_nor_init(sp);
894 
895 	irq = platform_get_irq_optional(pdev, 0);
896 
897 	if (irq < 0) {
898 		dev_warn(sp->dev, "IRQ not available.");
899 	} else {
900 		ret = devm_request_irq(sp->dev, irq, mtk_nor_irq_handler, 0,
901 				       pdev->name, sp);
902 		if (ret < 0) {
903 			dev_warn(sp->dev, "failed to request IRQ.");
904 		} else {
905 			init_completion(&sp->op_done);
906 			sp->has_irq = true;
907 		}
908 	}
909 
910 	pm_runtime_set_autosuspend_delay(&pdev->dev, -1);
911 	pm_runtime_use_autosuspend(&pdev->dev);
912 	pm_runtime_set_active(&pdev->dev);
913 	pm_runtime_enable(&pdev->dev);
914 	pm_runtime_get_noresume(&pdev->dev);
915 
916 	ret = devm_spi_register_controller(&pdev->dev, ctlr);
917 	if (ret < 0)
918 		goto err_probe;
919 
920 	pm_runtime_mark_last_busy(&pdev->dev);
921 	pm_runtime_put_autosuspend(&pdev->dev);
922 
923 	dev_info(&pdev->dev, "spi frequency: %d Hz\n", sp->spi_freq);
924 
925 	return 0;
926 
927 err_probe:
928 	pm_runtime_disable(&pdev->dev);
929 	pm_runtime_set_suspended(&pdev->dev);
930 	pm_runtime_dont_use_autosuspend(&pdev->dev);
931 
932 	mtk_nor_disable_clk(sp);
933 
934 	return ret;
935 }
936 
937 static int mtk_nor_remove(struct platform_device *pdev)
938 {
939 	struct spi_controller *ctlr = dev_get_drvdata(&pdev->dev);
940 	struct mtk_nor *sp = spi_controller_get_devdata(ctlr);
941 
942 	pm_runtime_disable(&pdev->dev);
943 	pm_runtime_set_suspended(&pdev->dev);
944 	pm_runtime_dont_use_autosuspend(&pdev->dev);
945 
946 	mtk_nor_disable_clk(sp);
947 
948 	return 0;
949 }
950 
951 static int __maybe_unused mtk_nor_runtime_suspend(struct device *dev)
952 {
953 	struct spi_controller *ctlr = dev_get_drvdata(dev);
954 	struct mtk_nor *sp = spi_controller_get_devdata(ctlr);
955 
956 	mtk_nor_disable_clk(sp);
957 
958 	return 0;
959 }
960 
961 static int __maybe_unused mtk_nor_runtime_resume(struct device *dev)
962 {
963 	struct spi_controller *ctlr = dev_get_drvdata(dev);
964 	struct mtk_nor *sp = spi_controller_get_devdata(ctlr);
965 
966 	return mtk_nor_enable_clk(sp);
967 }
968 
969 static int __maybe_unused mtk_nor_suspend(struct device *dev)
970 {
971 	return pm_runtime_force_suspend(dev);
972 }
973 
974 static int __maybe_unused mtk_nor_resume(struct device *dev)
975 {
976 	struct spi_controller *ctlr = dev_get_drvdata(dev);
977 	struct mtk_nor *sp = spi_controller_get_devdata(ctlr);
978 	int ret;
979 
980 	ret = pm_runtime_force_resume(dev);
981 	if (ret)
982 		return ret;
983 
984 	mtk_nor_init(sp);
985 
986 	return 0;
987 }
988 
989 static const struct dev_pm_ops mtk_nor_pm_ops = {
990 	SET_RUNTIME_PM_OPS(mtk_nor_runtime_suspend,
991 			   mtk_nor_runtime_resume, NULL)
992 	SET_SYSTEM_SLEEP_PM_OPS(mtk_nor_suspend, mtk_nor_resume)
993 };
994 
995 static struct platform_driver mtk_nor_driver = {
996 	.driver = {
997 		.name = DRIVER_NAME,
998 		.of_match_table = mtk_nor_match,
999 		.pm = &mtk_nor_pm_ops,
1000 	},
1001 	.probe = mtk_nor_probe,
1002 	.remove = mtk_nor_remove,
1003 };
1004 
1005 module_platform_driver(mtk_nor_driver);
1006 
1007 MODULE_DESCRIPTION("Mediatek SPI NOR controller driver");
1008 MODULE_AUTHOR("Chuanhong Guo <gch981213@gmail.com>");
1009 MODULE_LICENSE("GPL v2");
1010 MODULE_ALIAS("platform:" DRIVER_NAME);
1011