xref: /openbmc/linux/drivers/spi/spi-dw-core.c (revision 48cc39c3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Designware SPI core controller driver (refer pxa2xx_spi.c)
4  *
5  * Copyright (c) 2009, Intel Corporation.
6  */
7 
8 #include <linux/dma-mapping.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/preempt.h>
12 #include <linux/highmem.h>
13 #include <linux/delay.h>
14 #include <linux/slab.h>
15 #include <linux/spi/spi.h>
16 #include <linux/spi/spi-mem.h>
17 #include <linux/string.h>
18 #include <linux/of.h>
19 
20 #include "spi-dw.h"
21 
22 #ifdef CONFIG_DEBUG_FS
23 #include <linux/debugfs.h>
24 #endif
25 
26 /* Slave spi_device related */
27 struct chip_data {
28 	u32 cr0;
29 	u32 rx_sample_dly;	/* RX sample delay */
30 };
31 
32 #ifdef CONFIG_DEBUG_FS
33 
34 #define DW_SPI_DBGFS_REG(_name, _off)	\
35 {					\
36 	.name = _name,			\
37 	.offset = _off,			\
38 }
39 
40 static const struct debugfs_reg32 dw_spi_dbgfs_regs[] = {
41 	DW_SPI_DBGFS_REG("CTRLR0", DW_SPI_CTRLR0),
42 	DW_SPI_DBGFS_REG("CTRLR1", DW_SPI_CTRLR1),
43 	DW_SPI_DBGFS_REG("SSIENR", DW_SPI_SSIENR),
44 	DW_SPI_DBGFS_REG("SER", DW_SPI_SER),
45 	DW_SPI_DBGFS_REG("BAUDR", DW_SPI_BAUDR),
46 	DW_SPI_DBGFS_REG("TXFTLR", DW_SPI_TXFTLR),
47 	DW_SPI_DBGFS_REG("RXFTLR", DW_SPI_RXFTLR),
48 	DW_SPI_DBGFS_REG("TXFLR", DW_SPI_TXFLR),
49 	DW_SPI_DBGFS_REG("RXFLR", DW_SPI_RXFLR),
50 	DW_SPI_DBGFS_REG("SR", DW_SPI_SR),
51 	DW_SPI_DBGFS_REG("IMR", DW_SPI_IMR),
52 	DW_SPI_DBGFS_REG("ISR", DW_SPI_ISR),
53 	DW_SPI_DBGFS_REG("DMACR", DW_SPI_DMACR),
54 	DW_SPI_DBGFS_REG("DMATDLR", DW_SPI_DMATDLR),
55 	DW_SPI_DBGFS_REG("DMARDLR", DW_SPI_DMARDLR),
56 	DW_SPI_DBGFS_REG("RX_SAMPLE_DLY", DW_SPI_RX_SAMPLE_DLY),
57 };
58 
59 static int dw_spi_debugfs_init(struct dw_spi *dws)
60 {
61 	char name[32];
62 
63 	snprintf(name, 32, "dw_spi%d", dws->master->bus_num);
64 	dws->debugfs = debugfs_create_dir(name, NULL);
65 	if (!dws->debugfs)
66 		return -ENOMEM;
67 
68 	dws->regset.regs = dw_spi_dbgfs_regs;
69 	dws->regset.nregs = ARRAY_SIZE(dw_spi_dbgfs_regs);
70 	dws->regset.base = dws->regs;
71 	debugfs_create_regset32("registers", 0400, dws->debugfs, &dws->regset);
72 
73 	return 0;
74 }
75 
76 static void dw_spi_debugfs_remove(struct dw_spi *dws)
77 {
78 	debugfs_remove_recursive(dws->debugfs);
79 }
80 
81 #else
82 static inline int dw_spi_debugfs_init(struct dw_spi *dws)
83 {
84 	return 0;
85 }
86 
87 static inline void dw_spi_debugfs_remove(struct dw_spi *dws)
88 {
89 }
90 #endif /* CONFIG_DEBUG_FS */
91 
92 void dw_spi_set_cs(struct spi_device *spi, bool enable)
93 {
94 	struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
95 	bool cs_high = !!(spi->mode & SPI_CS_HIGH);
96 
97 	/*
98 	 * DW SPI controller demands any native CS being set in order to
99 	 * proceed with data transfer. So in order to activate the SPI
100 	 * communications we must set a corresponding bit in the Slave
101 	 * Enable register no matter whether the SPI core is configured to
102 	 * support active-high or active-low CS level.
103 	 */
104 	if (cs_high == enable)
105 		dw_writel(dws, DW_SPI_SER, BIT(spi->chip_select));
106 	else
107 		dw_writel(dws, DW_SPI_SER, 0);
108 }
109 EXPORT_SYMBOL_GPL(dw_spi_set_cs);
110 
111 /* Return the max entries we can fill into tx fifo */
112 static inline u32 tx_max(struct dw_spi *dws)
113 {
114 	u32 tx_room, rxtx_gap;
115 
116 	tx_room = dws->fifo_len - dw_readl(dws, DW_SPI_TXFLR);
117 
118 	/*
119 	 * Another concern is about the tx/rx mismatch, we
120 	 * though to use (dws->fifo_len - rxflr - txflr) as
121 	 * one maximum value for tx, but it doesn't cover the
122 	 * data which is out of tx/rx fifo and inside the
123 	 * shift registers. So a control from sw point of
124 	 * view is taken.
125 	 */
126 	rxtx_gap = dws->fifo_len - (dws->rx_len - dws->tx_len);
127 
128 	return min3((u32)dws->tx_len, tx_room, rxtx_gap);
129 }
130 
131 /* Return the max entries we should read out of rx fifo */
132 static inline u32 rx_max(struct dw_spi *dws)
133 {
134 	return min_t(u32, dws->rx_len, dw_readl(dws, DW_SPI_RXFLR));
135 }
136 
137 static void dw_writer(struct dw_spi *dws)
138 {
139 	u32 max = tx_max(dws);
140 	u32 txw = 0;
141 
142 	while (max--) {
143 		if (dws->tx) {
144 			if (dws->n_bytes == 1)
145 				txw = *(u8 *)(dws->tx);
146 			else if (dws->n_bytes == 2)
147 				txw = *(u16 *)(dws->tx);
148 			else
149 				txw = *(u32 *)(dws->tx);
150 
151 			dws->tx += dws->n_bytes;
152 		}
153 		dw_write_io_reg(dws, DW_SPI_DR, txw);
154 		--dws->tx_len;
155 	}
156 }
157 
158 static void dw_reader(struct dw_spi *dws)
159 {
160 	u32 max = rx_max(dws);
161 	u32 rxw;
162 
163 	while (max--) {
164 		rxw = dw_read_io_reg(dws, DW_SPI_DR);
165 		if (dws->rx) {
166 			if (dws->n_bytes == 1)
167 				*(u8 *)(dws->rx) = rxw;
168 			else if (dws->n_bytes == 2)
169 				*(u16 *)(dws->rx) = rxw;
170 			else
171 				*(u32 *)(dws->rx) = rxw;
172 
173 			dws->rx += dws->n_bytes;
174 		}
175 		--dws->rx_len;
176 	}
177 }
178 
179 int dw_spi_check_status(struct dw_spi *dws, bool raw)
180 {
181 	u32 irq_status;
182 	int ret = 0;
183 
184 	if (raw)
185 		irq_status = dw_readl(dws, DW_SPI_RISR);
186 	else
187 		irq_status = dw_readl(dws, DW_SPI_ISR);
188 
189 	if (irq_status & SPI_INT_RXOI) {
190 		dev_err(&dws->master->dev, "RX FIFO overflow detected\n");
191 		ret = -EIO;
192 	}
193 
194 	if (irq_status & SPI_INT_RXUI) {
195 		dev_err(&dws->master->dev, "RX FIFO underflow detected\n");
196 		ret = -EIO;
197 	}
198 
199 	if (irq_status & SPI_INT_TXOI) {
200 		dev_err(&dws->master->dev, "TX FIFO overflow detected\n");
201 		ret = -EIO;
202 	}
203 
204 	/* Generically handle the erroneous situation */
205 	if (ret) {
206 		spi_reset_chip(dws);
207 		if (dws->master->cur_msg)
208 			dws->master->cur_msg->status = ret;
209 	}
210 
211 	return ret;
212 }
213 EXPORT_SYMBOL_GPL(dw_spi_check_status);
214 
215 static irqreturn_t dw_spi_transfer_handler(struct dw_spi *dws)
216 {
217 	u16 irq_status = dw_readl(dws, DW_SPI_ISR);
218 
219 	if (dw_spi_check_status(dws, false)) {
220 		spi_finalize_current_transfer(dws->master);
221 		return IRQ_HANDLED;
222 	}
223 
224 	/*
225 	 * Read data from the Rx FIFO every time we've got a chance executing
226 	 * this method. If there is nothing left to receive, terminate the
227 	 * procedure. Otherwise adjust the Rx FIFO Threshold level if it's a
228 	 * final stage of the transfer. By doing so we'll get the next IRQ
229 	 * right when the leftover incoming data is received.
230 	 */
231 	dw_reader(dws);
232 	if (!dws->rx_len) {
233 		spi_mask_intr(dws, 0xff);
234 		spi_finalize_current_transfer(dws->master);
235 	} else if (dws->rx_len <= dw_readl(dws, DW_SPI_RXFTLR)) {
236 		dw_writel(dws, DW_SPI_RXFTLR, dws->rx_len - 1);
237 	}
238 
239 	/*
240 	 * Send data out if Tx FIFO Empty IRQ is received. The IRQ will be
241 	 * disabled after the data transmission is finished so not to
242 	 * have the TXE IRQ flood at the final stage of the transfer.
243 	 */
244 	if (irq_status & SPI_INT_TXEI) {
245 		dw_writer(dws);
246 		if (!dws->tx_len)
247 			spi_mask_intr(dws, SPI_INT_TXEI);
248 	}
249 
250 	return IRQ_HANDLED;
251 }
252 
253 static irqreturn_t dw_spi_irq(int irq, void *dev_id)
254 {
255 	struct spi_controller *master = dev_id;
256 	struct dw_spi *dws = spi_controller_get_devdata(master);
257 	u16 irq_status = dw_readl(dws, DW_SPI_ISR) & 0x3f;
258 
259 	if (!irq_status)
260 		return IRQ_NONE;
261 
262 	if (!master->cur_msg) {
263 		spi_mask_intr(dws, 0xff);
264 		return IRQ_HANDLED;
265 	}
266 
267 	return dws->transfer_handler(dws);
268 }
269 
270 static u32 dw_spi_prepare_cr0(struct dw_spi *dws, struct spi_device *spi)
271 {
272 	u32 cr0 = 0;
273 
274 	if (!(dws->caps & DW_SPI_CAP_DWC_SSI)) {
275 		/* CTRLR0[ 5: 4] Frame Format */
276 		cr0 |= SSI_MOTO_SPI << SPI_FRF_OFFSET;
277 
278 		/*
279 		 * SPI mode (SCPOL|SCPH)
280 		 * CTRLR0[ 6] Serial Clock Phase
281 		 * CTRLR0[ 7] Serial Clock Polarity
282 		 */
283 		cr0 |= ((spi->mode & SPI_CPOL) ? 1 : 0) << SPI_SCOL_OFFSET;
284 		cr0 |= ((spi->mode & SPI_CPHA) ? 1 : 0) << SPI_SCPH_OFFSET;
285 
286 		/* CTRLR0[11] Shift Register Loop */
287 		cr0 |= ((spi->mode & SPI_LOOP) ? 1 : 0) << SPI_SRL_OFFSET;
288 	} else {
289 		/* CTRLR0[ 7: 6] Frame Format */
290 		cr0 |= SSI_MOTO_SPI << DWC_SSI_CTRLR0_FRF_OFFSET;
291 
292 		/*
293 		 * SPI mode (SCPOL|SCPH)
294 		 * CTRLR0[ 8] Serial Clock Phase
295 		 * CTRLR0[ 9] Serial Clock Polarity
296 		 */
297 		cr0 |= ((spi->mode & SPI_CPOL) ? 1 : 0) << DWC_SSI_CTRLR0_SCPOL_OFFSET;
298 		cr0 |= ((spi->mode & SPI_CPHA) ? 1 : 0) << DWC_SSI_CTRLR0_SCPH_OFFSET;
299 
300 		/* CTRLR0[13] Shift Register Loop */
301 		cr0 |= ((spi->mode & SPI_LOOP) ? 1 : 0) << DWC_SSI_CTRLR0_SRL_OFFSET;
302 
303 		if (dws->caps & DW_SPI_CAP_KEEMBAY_MST)
304 			cr0 |= DWC_SSI_CTRLR0_KEEMBAY_MST;
305 	}
306 
307 	return cr0;
308 }
309 
310 void dw_spi_update_config(struct dw_spi *dws, struct spi_device *spi,
311 			  struct dw_spi_cfg *cfg)
312 {
313 	struct chip_data *chip = spi_get_ctldata(spi);
314 	u32 cr0 = chip->cr0;
315 	u32 speed_hz;
316 	u16 clk_div;
317 
318 	/* CTRLR0[ 4/3: 0] or CTRLR0[ 20: 16] Data Frame Size */
319 	cr0 |= (cfg->dfs - 1) << dws->dfs_offset;
320 
321 	if (!(dws->caps & DW_SPI_CAP_DWC_SSI))
322 		/* CTRLR0[ 9:8] Transfer Mode */
323 		cr0 |= cfg->tmode << SPI_TMOD_OFFSET;
324 	else
325 		/* CTRLR0[11:10] Transfer Mode */
326 		cr0 |= cfg->tmode << DWC_SSI_CTRLR0_TMOD_OFFSET;
327 
328 	dw_writel(dws, DW_SPI_CTRLR0, cr0);
329 
330 	if (cfg->tmode == SPI_TMOD_EPROMREAD || cfg->tmode == SPI_TMOD_RO)
331 		dw_writel(dws, DW_SPI_CTRLR1, cfg->ndf ? cfg->ndf - 1 : 0);
332 
333 	/* Note DW APB SSI clock divider doesn't support odd numbers */
334 	clk_div = (DIV_ROUND_UP(dws->max_freq, cfg->freq) + 1) & 0xfffe;
335 	speed_hz = dws->max_freq / clk_div;
336 
337 	if (dws->current_freq != speed_hz) {
338 		spi_set_clk(dws, clk_div);
339 		dws->current_freq = speed_hz;
340 	}
341 
342 	/* Update RX sample delay if required */
343 	if (dws->cur_rx_sample_dly != chip->rx_sample_dly) {
344 		dw_writel(dws, DW_SPI_RX_SAMPLE_DLY, chip->rx_sample_dly);
345 		dws->cur_rx_sample_dly = chip->rx_sample_dly;
346 	}
347 }
348 EXPORT_SYMBOL_GPL(dw_spi_update_config);
349 
350 static void dw_spi_irq_setup(struct dw_spi *dws)
351 {
352 	u16 level;
353 	u8 imask;
354 
355 	/*
356 	 * Originally Tx and Rx data lengths match. Rx FIFO Threshold level
357 	 * will be adjusted at the final stage of the IRQ-based SPI transfer
358 	 * execution so not to lose the leftover of the incoming data.
359 	 */
360 	level = min_t(u16, dws->fifo_len / 2, dws->tx_len);
361 	dw_writel(dws, DW_SPI_TXFTLR, level);
362 	dw_writel(dws, DW_SPI_RXFTLR, level - 1);
363 
364 	dws->transfer_handler = dw_spi_transfer_handler;
365 
366 	imask = SPI_INT_TXEI | SPI_INT_TXOI | SPI_INT_RXUI | SPI_INT_RXOI |
367 		SPI_INT_RXFI;
368 	spi_umask_intr(dws, imask);
369 }
370 
371 /*
372  * The iterative procedure of the poll-based transfer is simple: write as much
373  * as possible to the Tx FIFO, wait until the pending to receive data is ready
374  * to be read, read it from the Rx FIFO and check whether the performed
375  * procedure has been successful.
376  *
377  * Note this method the same way as the IRQ-based transfer won't work well for
378  * the SPI devices connected to the controller with native CS due to the
379  * automatic CS assertion/de-assertion.
380  */
381 static int dw_spi_poll_transfer(struct dw_spi *dws,
382 				struct spi_transfer *transfer)
383 {
384 	struct spi_delay delay;
385 	u16 nbits;
386 	int ret;
387 
388 	delay.unit = SPI_DELAY_UNIT_SCK;
389 	nbits = dws->n_bytes * BITS_PER_BYTE;
390 
391 	do {
392 		dw_writer(dws);
393 
394 		delay.value = nbits * (dws->rx_len - dws->tx_len);
395 		spi_delay_exec(&delay, transfer);
396 
397 		dw_reader(dws);
398 
399 		ret = dw_spi_check_status(dws, true);
400 		if (ret)
401 			return ret;
402 	} while (dws->rx_len);
403 
404 	return 0;
405 }
406 
407 static int dw_spi_transfer_one(struct spi_controller *master,
408 		struct spi_device *spi, struct spi_transfer *transfer)
409 {
410 	struct dw_spi *dws = spi_controller_get_devdata(master);
411 	struct dw_spi_cfg cfg = {
412 		.tmode = SPI_TMOD_TR,
413 		.dfs = transfer->bits_per_word,
414 		.freq = transfer->speed_hz,
415 	};
416 	int ret;
417 
418 	dws->dma_mapped = 0;
419 	dws->n_bytes = DIV_ROUND_UP(transfer->bits_per_word, BITS_PER_BYTE);
420 	dws->tx = (void *)transfer->tx_buf;
421 	dws->tx_len = transfer->len / dws->n_bytes;
422 	dws->rx = transfer->rx_buf;
423 	dws->rx_len = dws->tx_len;
424 
425 	/* Ensure the data above is visible for all CPUs */
426 	smp_mb();
427 
428 	spi_enable_chip(dws, 0);
429 
430 	dw_spi_update_config(dws, spi, &cfg);
431 
432 	transfer->effective_speed_hz = dws->current_freq;
433 
434 	/* Check if current transfer is a DMA transaction */
435 	if (master->can_dma && master->can_dma(master, spi, transfer))
436 		dws->dma_mapped = master->cur_msg_mapped;
437 
438 	/* For poll mode just disable all interrupts */
439 	spi_mask_intr(dws, 0xff);
440 
441 	if (dws->dma_mapped) {
442 		ret = dws->dma_ops->dma_setup(dws, transfer);
443 		if (ret)
444 			return ret;
445 	}
446 
447 	spi_enable_chip(dws, 1);
448 
449 	if (dws->dma_mapped)
450 		return dws->dma_ops->dma_transfer(dws, transfer);
451 	else if (dws->irq == IRQ_NOTCONNECTED)
452 		return dw_spi_poll_transfer(dws, transfer);
453 
454 	dw_spi_irq_setup(dws);
455 
456 	return 1;
457 }
458 
459 static void dw_spi_handle_err(struct spi_controller *master,
460 		struct spi_message *msg)
461 {
462 	struct dw_spi *dws = spi_controller_get_devdata(master);
463 
464 	if (dws->dma_mapped)
465 		dws->dma_ops->dma_stop(dws);
466 
467 	spi_reset_chip(dws);
468 }
469 
470 static int dw_spi_adjust_mem_op_size(struct spi_mem *mem, struct spi_mem_op *op)
471 {
472 	if (op->data.dir == SPI_MEM_DATA_IN)
473 		op->data.nbytes = clamp_val(op->data.nbytes, 0, SPI_NDF_MASK + 1);
474 
475 	return 0;
476 }
477 
478 static bool dw_spi_supports_mem_op(struct spi_mem *mem,
479 				   const struct spi_mem_op *op)
480 {
481 	if (op->data.buswidth > 1 || op->addr.buswidth > 1 ||
482 	    op->dummy.buswidth > 1 || op->cmd.buswidth > 1)
483 		return false;
484 
485 	return spi_mem_default_supports_op(mem, op);
486 }
487 
488 static int dw_spi_init_mem_buf(struct dw_spi *dws, const struct spi_mem_op *op)
489 {
490 	unsigned int i, j, len;
491 	u8 *out;
492 
493 	/*
494 	 * Calculate the total length of the EEPROM command transfer and
495 	 * either use the pre-allocated buffer or create a temporary one.
496 	 */
497 	len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
498 	if (op->data.dir == SPI_MEM_DATA_OUT)
499 		len += op->data.nbytes;
500 
501 	if (len <= SPI_BUF_SIZE) {
502 		out = dws->buf;
503 	} else {
504 		out = kzalloc(len, GFP_KERNEL);
505 		if (!out)
506 			return -ENOMEM;
507 	}
508 
509 	/*
510 	 * Collect the operation code, address and dummy bytes into the single
511 	 * buffer. If it's a transfer with data to be sent, also copy it into the
512 	 * single buffer in order to speed the data transmission up.
513 	 */
514 	for (i = 0; i < op->cmd.nbytes; ++i)
515 		out[i] = SPI_GET_BYTE(op->cmd.opcode, op->cmd.nbytes - i - 1);
516 	for (j = 0; j < op->addr.nbytes; ++i, ++j)
517 		out[i] = SPI_GET_BYTE(op->addr.val, op->addr.nbytes - j - 1);
518 	for (j = 0; j < op->dummy.nbytes; ++i, ++j)
519 		out[i] = 0x0;
520 
521 	if (op->data.dir == SPI_MEM_DATA_OUT)
522 		memcpy(&out[i], op->data.buf.out, op->data.nbytes);
523 
524 	dws->n_bytes = 1;
525 	dws->tx = out;
526 	dws->tx_len = len;
527 	if (op->data.dir == SPI_MEM_DATA_IN) {
528 		dws->rx = op->data.buf.in;
529 		dws->rx_len = op->data.nbytes;
530 	} else {
531 		dws->rx = NULL;
532 		dws->rx_len = 0;
533 	}
534 
535 	return 0;
536 }
537 
538 static void dw_spi_free_mem_buf(struct dw_spi *dws)
539 {
540 	if (dws->tx != dws->buf)
541 		kfree(dws->tx);
542 }
543 
544 static int dw_spi_write_then_read(struct dw_spi *dws, struct spi_device *spi)
545 {
546 	u32 room, entries, sts;
547 	unsigned int len;
548 	u8 *buf;
549 
550 	/*
551 	 * At initial stage we just pre-fill the Tx FIFO in with no rush,
552 	 * since native CS hasn't been enabled yet and the automatic data
553 	 * transmission won't start til we do that.
554 	 */
555 	len = min(dws->fifo_len, dws->tx_len);
556 	buf = dws->tx;
557 	while (len--)
558 		dw_write_io_reg(dws, DW_SPI_DR, *buf++);
559 
560 	/*
561 	 * After setting any bit in the SER register the transmission will
562 	 * start automatically. We have to keep up with that procedure
563 	 * otherwise the CS de-assertion will happen whereupon the memory
564 	 * operation will be pre-terminated.
565 	 */
566 	len = dws->tx_len - ((void *)buf - dws->tx);
567 	dw_spi_set_cs(spi, false);
568 	while (len) {
569 		entries = readl_relaxed(dws->regs + DW_SPI_TXFLR);
570 		if (!entries) {
571 			dev_err(&dws->master->dev, "CS de-assertion on Tx\n");
572 			return -EIO;
573 		}
574 		room = min(dws->fifo_len - entries, len);
575 		for (; room; --room, --len)
576 			dw_write_io_reg(dws, DW_SPI_DR, *buf++);
577 	}
578 
579 	/*
580 	 * Data fetching will start automatically if the EEPROM-read mode is
581 	 * activated. We have to keep up with the incoming data pace to
582 	 * prevent the Rx FIFO overflow causing the inbound data loss.
583 	 */
584 	len = dws->rx_len;
585 	buf = dws->rx;
586 	while (len) {
587 		entries = readl_relaxed(dws->regs + DW_SPI_RXFLR);
588 		if (!entries) {
589 			sts = readl_relaxed(dws->regs + DW_SPI_RISR);
590 			if (sts & SPI_INT_RXOI) {
591 				dev_err(&dws->master->dev, "FIFO overflow on Rx\n");
592 				return -EIO;
593 			}
594 			continue;
595 		}
596 		entries = min(entries, len);
597 		for (; entries; --entries, --len)
598 			*buf++ = dw_read_io_reg(dws, DW_SPI_DR);
599 	}
600 
601 	return 0;
602 }
603 
604 static inline bool dw_spi_ctlr_busy(struct dw_spi *dws)
605 {
606 	return dw_readl(dws, DW_SPI_SR) & SR_BUSY;
607 }
608 
609 static int dw_spi_wait_mem_op_done(struct dw_spi *dws)
610 {
611 	int retry = SPI_WAIT_RETRIES;
612 	struct spi_delay delay;
613 	unsigned long ns, us;
614 	u32 nents;
615 
616 	nents = dw_readl(dws, DW_SPI_TXFLR);
617 	ns = NSEC_PER_SEC / dws->current_freq * nents;
618 	ns *= dws->n_bytes * BITS_PER_BYTE;
619 	if (ns <= NSEC_PER_USEC) {
620 		delay.unit = SPI_DELAY_UNIT_NSECS;
621 		delay.value = ns;
622 	} else {
623 		us = DIV_ROUND_UP(ns, NSEC_PER_USEC);
624 		delay.unit = SPI_DELAY_UNIT_USECS;
625 		delay.value = clamp_val(us, 0, USHRT_MAX);
626 	}
627 
628 	while (dw_spi_ctlr_busy(dws) && retry--)
629 		spi_delay_exec(&delay, NULL);
630 
631 	if (retry < 0) {
632 		dev_err(&dws->master->dev, "Mem op hanged up\n");
633 		return -EIO;
634 	}
635 
636 	return 0;
637 }
638 
639 static void dw_spi_stop_mem_op(struct dw_spi *dws, struct spi_device *spi)
640 {
641 	spi_enable_chip(dws, 0);
642 	dw_spi_set_cs(spi, true);
643 	spi_enable_chip(dws, 1);
644 }
645 
646 /*
647  * The SPI memory operation implementation below is the best choice for the
648  * devices, which are selected by the native chip-select lane. It's
649  * specifically developed to workaround the problem with automatic chip-select
650  * lane toggle when there is no data in the Tx FIFO buffer. Luckily the current
651  * SPI-mem core calls exec_op() callback only if the GPIO-based CS is
652  * unavailable.
653  */
654 static int dw_spi_exec_mem_op(struct spi_mem *mem, const struct spi_mem_op *op)
655 {
656 	struct dw_spi *dws = spi_controller_get_devdata(mem->spi->controller);
657 	struct dw_spi_cfg cfg;
658 	unsigned long flags;
659 	int ret;
660 
661 	/*
662 	 * Collect the outbound data into a single buffer to speed the
663 	 * transmission up at least on the initial stage.
664 	 */
665 	ret = dw_spi_init_mem_buf(dws, op);
666 	if (ret)
667 		return ret;
668 
669 	/*
670 	 * DW SPI EEPROM-read mode is required only for the SPI memory Data-IN
671 	 * operation. Transmit-only mode is suitable for the rest of them.
672 	 */
673 	cfg.dfs = 8;
674 	cfg.freq = clamp(mem->spi->max_speed_hz, 0U, dws->max_mem_freq);
675 	if (op->data.dir == SPI_MEM_DATA_IN) {
676 		cfg.tmode = SPI_TMOD_EPROMREAD;
677 		cfg.ndf = op->data.nbytes;
678 	} else {
679 		cfg.tmode = SPI_TMOD_TO;
680 	}
681 
682 	spi_enable_chip(dws, 0);
683 
684 	dw_spi_update_config(dws, mem->spi, &cfg);
685 
686 	spi_mask_intr(dws, 0xff);
687 
688 	spi_enable_chip(dws, 1);
689 
690 	/*
691 	 * DW APB SSI controller has very nasty peculiarities. First originally
692 	 * (without any vendor-specific modifications) it doesn't provide a
693 	 * direct way to set and clear the native chip-select signal. Instead
694 	 * the controller asserts the CS lane if Tx FIFO isn't empty and a
695 	 * transmission is going on, and automatically de-asserts it back to
696 	 * the high level if the Tx FIFO doesn't have anything to be pushed
697 	 * out. Due to that a multi-tasking or heavy IRQs activity might be
698 	 * fatal, since the transfer procedure preemption may cause the Tx FIFO
699 	 * getting empty and sudden CS de-assertion, which in the middle of the
700 	 * transfer will most likely cause the data loss. Secondly the
701 	 * EEPROM-read or Read-only DW SPI transfer modes imply the incoming
702 	 * data being automatically pulled in into the Rx FIFO. So if the
703 	 * driver software is late in fetching the data from the FIFO before
704 	 * it's overflown, new incoming data will be lost. In order to make
705 	 * sure the executed memory operations are CS-atomic and to prevent the
706 	 * Rx FIFO overflow we have to disable the local interrupts so to block
707 	 * any preemption during the subsequent IO operations.
708 	 *
709 	 * Note. At some circumstances disabling IRQs may not help to prevent
710 	 * the problems described above. The CS de-assertion and Rx FIFO
711 	 * overflow may still happen due to the relatively slow system bus or
712 	 * CPU not working fast enough, so the write-then-read algo implemented
713 	 * here just won't keep up with the SPI bus data transfer. Such
714 	 * situation is highly platform specific and is supposed to be fixed by
715 	 * manually restricting the SPI bus frequency using the
716 	 * dws->max_mem_freq parameter.
717 	 */
718 	local_irq_save(flags);
719 	preempt_disable();
720 
721 	ret = dw_spi_write_then_read(dws, mem->spi);
722 
723 	local_irq_restore(flags);
724 	preempt_enable();
725 
726 	/*
727 	 * Wait for the operation being finished and check the controller
728 	 * status only if there hasn't been any run-time error detected. In the
729 	 * former case it's just pointless. In the later one to prevent an
730 	 * additional error message printing since any hw error flag being set
731 	 * would be due to an error detected on the data transfer.
732 	 */
733 	if (!ret) {
734 		ret = dw_spi_wait_mem_op_done(dws);
735 		if (!ret)
736 			ret = dw_spi_check_status(dws, true);
737 	}
738 
739 	dw_spi_stop_mem_op(dws, mem->spi);
740 
741 	dw_spi_free_mem_buf(dws);
742 
743 	return ret;
744 }
745 
746 /*
747  * Initialize the default memory operations if a glue layer hasn't specified
748  * custom ones. Direct mapping operations will be preserved anyway since DW SPI
749  * controller doesn't have an embedded dirmap interface. Note the memory
750  * operations implemented in this driver is the best choice only for the DW APB
751  * SSI controller with standard native CS functionality. If a hardware vendor
752  * has fixed the automatic CS assertion/de-assertion peculiarity, then it will
753  * be safer to use the normal SPI-messages-based transfers implementation.
754  */
755 static void dw_spi_init_mem_ops(struct dw_spi *dws)
756 {
757 	if (!dws->mem_ops.exec_op && !(dws->caps & DW_SPI_CAP_CS_OVERRIDE) &&
758 	    !dws->set_cs) {
759 		dws->mem_ops.adjust_op_size = dw_spi_adjust_mem_op_size;
760 		dws->mem_ops.supports_op = dw_spi_supports_mem_op;
761 		dws->mem_ops.exec_op = dw_spi_exec_mem_op;
762 		if (!dws->max_mem_freq)
763 			dws->max_mem_freq = dws->max_freq;
764 	}
765 }
766 
767 /* This may be called twice for each spi dev */
768 static int dw_spi_setup(struct spi_device *spi)
769 {
770 	struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
771 	struct chip_data *chip;
772 
773 	/* Only alloc on first setup */
774 	chip = spi_get_ctldata(spi);
775 	if (!chip) {
776 		struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
777 		u32 rx_sample_dly_ns;
778 
779 		chip = kzalloc(sizeof(struct chip_data), GFP_KERNEL);
780 		if (!chip)
781 			return -ENOMEM;
782 		spi_set_ctldata(spi, chip);
783 		/* Get specific / default rx-sample-delay */
784 		if (device_property_read_u32(&spi->dev,
785 					     "rx-sample-delay-ns",
786 					     &rx_sample_dly_ns) != 0)
787 			/* Use default controller value */
788 			rx_sample_dly_ns = dws->def_rx_sample_dly_ns;
789 		chip->rx_sample_dly = DIV_ROUND_CLOSEST(rx_sample_dly_ns,
790 							NSEC_PER_SEC /
791 							dws->max_freq);
792 	}
793 
794 	/*
795 	 * Update CR0 data each time the setup callback is invoked since
796 	 * the device parameters could have been changed, for instance, by
797 	 * the MMC SPI driver or something else.
798 	 */
799 	chip->cr0 = dw_spi_prepare_cr0(dws, spi);
800 
801 	return 0;
802 }
803 
804 static void dw_spi_cleanup(struct spi_device *spi)
805 {
806 	struct chip_data *chip = spi_get_ctldata(spi);
807 
808 	kfree(chip);
809 	spi_set_ctldata(spi, NULL);
810 }
811 
812 /* Restart the controller, disable all interrupts, clean rx fifo */
813 static void spi_hw_init(struct device *dev, struct dw_spi *dws)
814 {
815 	spi_reset_chip(dws);
816 
817 	/*
818 	 * Try to detect the FIFO depth if not set by interface driver,
819 	 * the depth could be from 2 to 256 from HW spec
820 	 */
821 	if (!dws->fifo_len) {
822 		u32 fifo;
823 
824 		for (fifo = 1; fifo < 256; fifo++) {
825 			dw_writel(dws, DW_SPI_TXFTLR, fifo);
826 			if (fifo != dw_readl(dws, DW_SPI_TXFTLR))
827 				break;
828 		}
829 		dw_writel(dws, DW_SPI_TXFTLR, 0);
830 
831 		dws->fifo_len = (fifo == 1) ? 0 : fifo;
832 		dev_dbg(dev, "Detected FIFO size: %u bytes\n", dws->fifo_len);
833 	}
834 
835 	/*
836 	 * Detect CTRLR0.DFS field size and offset by testing the lowest bits
837 	 * writability. Note DWC SSI controller also has the extended DFS, but
838 	 * with zero offset.
839 	 */
840 	if (!(dws->caps & DW_SPI_CAP_DWC_SSI)) {
841 		u32 cr0, tmp = dw_readl(dws, DW_SPI_CTRLR0);
842 
843 		spi_enable_chip(dws, 0);
844 		dw_writel(dws, DW_SPI_CTRLR0, 0xffffffff);
845 		cr0 = dw_readl(dws, DW_SPI_CTRLR0);
846 		dw_writel(dws, DW_SPI_CTRLR0, tmp);
847 		spi_enable_chip(dws, 1);
848 
849 		if (!(cr0 & SPI_DFS_MASK)) {
850 			dws->caps |= DW_SPI_CAP_DFS32;
851 			dws->dfs_offset = SPI_DFS32_OFFSET;
852 			dev_dbg(dev, "Detected 32-bits max data frame size\n");
853 		}
854 	} else {
855 		dws->caps |= DW_SPI_CAP_DFS32;
856 	}
857 
858 	/* enable HW fixup for explicit CS deselect for Amazon's alpine chip */
859 	if (dws->caps & DW_SPI_CAP_CS_OVERRIDE)
860 		dw_writel(dws, DW_SPI_CS_OVERRIDE, 0xF);
861 }
862 
863 int dw_spi_add_host(struct device *dev, struct dw_spi *dws)
864 {
865 	struct spi_controller *master;
866 	int ret;
867 
868 	if (!dws)
869 		return -EINVAL;
870 
871 	master = spi_alloc_master(dev, 0);
872 	if (!master)
873 		return -ENOMEM;
874 
875 	dws->master = master;
876 	dws->dma_addr = (dma_addr_t)(dws->paddr + DW_SPI_DR);
877 
878 	spi_controller_set_devdata(master, dws);
879 
880 	/* Basic HW init */
881 	spi_hw_init(dev, dws);
882 
883 	ret = request_irq(dws->irq, dw_spi_irq, IRQF_SHARED, dev_name(dev),
884 			  master);
885 	if (ret < 0 && ret != -ENOTCONN) {
886 		dev_err(dev, "can not get IRQ\n");
887 		goto err_free_master;
888 	}
889 
890 	dw_spi_init_mem_ops(dws);
891 
892 	master->use_gpio_descriptors = true;
893 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_LOOP;
894 	if (dws->caps & DW_SPI_CAP_DFS32)
895 		master->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
896 	else
897 		master->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 16);
898 	master->bus_num = dws->bus_num;
899 	master->num_chipselect = dws->num_cs;
900 	master->setup = dw_spi_setup;
901 	master->cleanup = dw_spi_cleanup;
902 	if (dws->set_cs)
903 		master->set_cs = dws->set_cs;
904 	else
905 		master->set_cs = dw_spi_set_cs;
906 	master->transfer_one = dw_spi_transfer_one;
907 	master->handle_err = dw_spi_handle_err;
908 	if (dws->mem_ops.exec_op)
909 		master->mem_ops = &dws->mem_ops;
910 	master->max_speed_hz = dws->max_freq;
911 	master->dev.of_node = dev->of_node;
912 	master->dev.fwnode = dev->fwnode;
913 	master->flags = SPI_MASTER_GPIO_SS;
914 	master->auto_runtime_pm = true;
915 
916 	/* Get default rx sample delay */
917 	device_property_read_u32(dev, "rx-sample-delay-ns",
918 				 &dws->def_rx_sample_dly_ns);
919 
920 	if (dws->dma_ops && dws->dma_ops->dma_init) {
921 		ret = dws->dma_ops->dma_init(dev, dws);
922 		if (ret) {
923 			dev_warn(dev, "DMA init failed\n");
924 		} else {
925 			master->can_dma = dws->dma_ops->can_dma;
926 			master->flags |= SPI_CONTROLLER_MUST_TX;
927 		}
928 	}
929 
930 	ret = spi_register_controller(master);
931 	if (ret) {
932 		dev_err(&master->dev, "problem registering spi master\n");
933 		goto err_dma_exit;
934 	}
935 
936 	dw_spi_debugfs_init(dws);
937 	return 0;
938 
939 err_dma_exit:
940 	if (dws->dma_ops && dws->dma_ops->dma_exit)
941 		dws->dma_ops->dma_exit(dws);
942 	spi_enable_chip(dws, 0);
943 	free_irq(dws->irq, master);
944 err_free_master:
945 	spi_controller_put(master);
946 	return ret;
947 }
948 EXPORT_SYMBOL_GPL(dw_spi_add_host);
949 
950 void dw_spi_remove_host(struct dw_spi *dws)
951 {
952 	dw_spi_debugfs_remove(dws);
953 
954 	spi_unregister_controller(dws->master);
955 
956 	if (dws->dma_ops && dws->dma_ops->dma_exit)
957 		dws->dma_ops->dma_exit(dws);
958 
959 	spi_shutdown_chip(dws);
960 
961 	free_irq(dws->irq, dws->master);
962 }
963 EXPORT_SYMBOL_GPL(dw_spi_remove_host);
964 
965 int dw_spi_suspend_host(struct dw_spi *dws)
966 {
967 	int ret;
968 
969 	ret = spi_controller_suspend(dws->master);
970 	if (ret)
971 		return ret;
972 
973 	spi_shutdown_chip(dws);
974 	return 0;
975 }
976 EXPORT_SYMBOL_GPL(dw_spi_suspend_host);
977 
978 int dw_spi_resume_host(struct dw_spi *dws)
979 {
980 	spi_hw_init(&dws->master->dev, dws);
981 	return spi_controller_resume(dws->master);
982 }
983 EXPORT_SYMBOL_GPL(dw_spi_resume_host);
984 
985 MODULE_AUTHOR("Feng Tang <feng.tang@intel.com>");
986 MODULE_DESCRIPTION("Driver for DesignWare SPI controller core");
987 MODULE_LICENSE("GPL v2");
988