xref: /linux/drivers/spi/spi-dw-core.c (revision 447e140e66fd226350b3ce86cffc965eaae4c856)
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/bitfield.h>
9 #include <linux/bitops.h>
10 #include <linux/dma-mapping.h>
11 #include <linux/interrupt.h>
12 #include <linux/module.h>
13 #include <linux/preempt.h>
14 #include <linux/highmem.h>
15 #include <linux/delay.h>
16 #include <linux/slab.h>
17 #include <linux/spi/spi.h>
18 #include <linux/spi/spi-mem.h>
19 #include <linux/string.h>
20 #include <linux/of.h>
21 
22 #include "spi-dw.h"
23 
24 #ifdef CONFIG_DEBUG_FS
25 #include <linux/debugfs.h>
26 #endif
27 
28 /* Slave spi_device related */
29 struct dw_spi_chip_data {
30 	u32 cr0;
31 	u32 rx_sample_dly;	/* RX sample delay */
32 };
33 
34 #ifdef CONFIG_DEBUG_FS
35 
36 #define DW_SPI_DBGFS_REG(_name, _off)	\
37 {					\
38 	.name = _name,			\
39 	.offset = _off,			\
40 }
41 
42 static const struct debugfs_reg32 dw_spi_dbgfs_regs[] = {
43 	DW_SPI_DBGFS_REG("CTRLR0", DW_SPI_CTRLR0),
44 	DW_SPI_DBGFS_REG("CTRLR1", DW_SPI_CTRLR1),
45 	DW_SPI_DBGFS_REG("SSIENR", DW_SPI_SSIENR),
46 	DW_SPI_DBGFS_REG("SER", DW_SPI_SER),
47 	DW_SPI_DBGFS_REG("BAUDR", DW_SPI_BAUDR),
48 	DW_SPI_DBGFS_REG("TXFTLR", DW_SPI_TXFTLR),
49 	DW_SPI_DBGFS_REG("RXFTLR", DW_SPI_RXFTLR),
50 	DW_SPI_DBGFS_REG("TXFLR", DW_SPI_TXFLR),
51 	DW_SPI_DBGFS_REG("RXFLR", DW_SPI_RXFLR),
52 	DW_SPI_DBGFS_REG("SR", DW_SPI_SR),
53 	DW_SPI_DBGFS_REG("IMR", DW_SPI_IMR),
54 	DW_SPI_DBGFS_REG("ISR", DW_SPI_ISR),
55 	DW_SPI_DBGFS_REG("DMACR", DW_SPI_DMACR),
56 	DW_SPI_DBGFS_REG("DMATDLR", DW_SPI_DMATDLR),
57 	DW_SPI_DBGFS_REG("DMARDLR", DW_SPI_DMARDLR),
58 	DW_SPI_DBGFS_REG("RX_SAMPLE_DLY", DW_SPI_RX_SAMPLE_DLY),
59 };
60 
61 static void dw_spi_debugfs_init(struct dw_spi *dws)
62 {
63 	char name[32];
64 
65 	snprintf(name, 32, "dw_spi%d", dws->host->bus_num);
66 	dws->debugfs = debugfs_create_dir(name, NULL);
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 
74 static void dw_spi_debugfs_remove(struct dw_spi *dws)
75 {
76 	debugfs_remove_recursive(dws->debugfs);
77 }
78 
79 #else
80 static inline void dw_spi_debugfs_init(struct dw_spi *dws)
81 {
82 }
83 
84 static inline void dw_spi_debugfs_remove(struct dw_spi *dws)
85 {
86 }
87 #endif /* CONFIG_DEBUG_FS */
88 
89 void dw_spi_set_cs(struct spi_device *spi, bool enable)
90 {
91 	struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
92 	bool cs_high = !!(spi->mode & SPI_CS_HIGH);
93 
94 	/*
95 	 * DW SPI controller demands any native CS being set in order to
96 	 * proceed with data transfer. So in order to activate the SPI
97 	 * communications we must set a corresponding bit in the Slave
98 	 * Enable register no matter whether the SPI core is configured to
99 	 * support active-high or active-low CS level.
100 	 */
101 	if (cs_high == enable)
102 		dw_writel(dws, DW_SPI_SER, BIT(spi_get_chipselect(spi, 0)));
103 	else
104 		dw_writel(dws, DW_SPI_SER, 0);
105 }
106 EXPORT_SYMBOL_NS_GPL(dw_spi_set_cs, SPI_DW_CORE);
107 
108 /* Return the max entries we can fill into tx fifo */
109 static inline u32 dw_spi_tx_max(struct dw_spi *dws)
110 {
111 	u32 tx_room, rxtx_gap;
112 
113 	tx_room = dws->fifo_len - dw_readl(dws, DW_SPI_TXFLR);
114 
115 	/*
116 	 * Another concern is about the tx/rx mismatch, we
117 	 * though to use (dws->fifo_len - rxflr - txflr) as
118 	 * one maximum value for tx, but it doesn't cover the
119 	 * data which is out of tx/rx fifo and inside the
120 	 * shift registers. So a control from sw point of
121 	 * view is taken.
122 	 */
123 	rxtx_gap = dws->fifo_len - (dws->rx_len - dws->tx_len);
124 
125 	return min3((u32)dws->tx_len, tx_room, rxtx_gap);
126 }
127 
128 /* Return the max entries we should read out of rx fifo */
129 static inline u32 dw_spi_rx_max(struct dw_spi *dws)
130 {
131 	return min_t(u32, dws->rx_len, dw_readl(dws, DW_SPI_RXFLR));
132 }
133 
134 static void dw_writer(struct dw_spi *dws)
135 {
136 	u32 max = dw_spi_tx_max(dws);
137 	u32 txw = 0;
138 
139 	while (max--) {
140 		if (dws->tx) {
141 			if (dws->n_bytes == 1)
142 				txw = *(u8 *)(dws->tx);
143 			else if (dws->n_bytes == 2)
144 				txw = *(u16 *)(dws->tx);
145 			else
146 				txw = *(u32 *)(dws->tx);
147 
148 			dws->tx += dws->n_bytes;
149 		}
150 		dw_write_io_reg(dws, DW_SPI_DR, txw);
151 		--dws->tx_len;
152 	}
153 }
154 
155 static void dw_reader(struct dw_spi *dws)
156 {
157 	u32 max = dw_spi_rx_max(dws);
158 	u32 rxw;
159 
160 	while (max--) {
161 		rxw = dw_read_io_reg(dws, DW_SPI_DR);
162 		if (dws->rx) {
163 			if (dws->n_bytes == 1)
164 				*(u8 *)(dws->rx) = rxw;
165 			else if (dws->n_bytes == 2)
166 				*(u16 *)(dws->rx) = rxw;
167 			else
168 				*(u32 *)(dws->rx) = rxw;
169 
170 			dws->rx += dws->n_bytes;
171 		}
172 		--dws->rx_len;
173 	}
174 }
175 
176 int dw_spi_check_status(struct dw_spi *dws, bool raw)
177 {
178 	u32 irq_status;
179 	int ret = 0;
180 
181 	if (raw)
182 		irq_status = dw_readl(dws, DW_SPI_RISR);
183 	else
184 		irq_status = dw_readl(dws, DW_SPI_ISR);
185 
186 	if (irq_status & DW_SPI_INT_RXOI) {
187 		dev_err(&dws->host->dev, "RX FIFO overflow detected\n");
188 		ret = -EIO;
189 	}
190 
191 	if (irq_status & DW_SPI_INT_RXUI) {
192 		dev_err(&dws->host->dev, "RX FIFO underflow detected\n");
193 		ret = -EIO;
194 	}
195 
196 	if (irq_status & DW_SPI_INT_TXOI) {
197 		dev_err(&dws->host->dev, "TX FIFO overflow detected\n");
198 		ret = -EIO;
199 	}
200 
201 	/* Generically handle the erroneous situation */
202 	if (ret) {
203 		dw_spi_reset_chip(dws);
204 		if (dws->host->cur_msg)
205 			dws->host->cur_msg->status = ret;
206 	}
207 
208 	return ret;
209 }
210 EXPORT_SYMBOL_NS_GPL(dw_spi_check_status, SPI_DW_CORE);
211 
212 static irqreturn_t dw_spi_transfer_handler(struct dw_spi *dws)
213 {
214 	u16 irq_status = dw_readl(dws, DW_SPI_ISR);
215 
216 	if (dw_spi_check_status(dws, false)) {
217 		spi_finalize_current_transfer(dws->host);
218 		return IRQ_HANDLED;
219 	}
220 
221 	/*
222 	 * Read data from the Rx FIFO every time we've got a chance executing
223 	 * this method. If there is nothing left to receive, terminate the
224 	 * procedure. Otherwise adjust the Rx FIFO Threshold level if it's a
225 	 * final stage of the transfer. By doing so we'll get the next IRQ
226 	 * right when the leftover incoming data is received.
227 	 */
228 	dw_reader(dws);
229 	if (!dws->rx_len) {
230 		dw_spi_mask_intr(dws, 0xff);
231 		spi_finalize_current_transfer(dws->host);
232 	} else if (dws->rx_len <= dw_readl(dws, DW_SPI_RXFTLR)) {
233 		dw_writel(dws, DW_SPI_RXFTLR, dws->rx_len - 1);
234 	}
235 
236 	/*
237 	 * Send data out if Tx FIFO Empty IRQ is received. The IRQ will be
238 	 * disabled after the data transmission is finished so not to
239 	 * have the TXE IRQ flood at the final stage of the transfer.
240 	 */
241 	if (irq_status & DW_SPI_INT_TXEI) {
242 		dw_writer(dws);
243 		if (!dws->tx_len)
244 			dw_spi_mask_intr(dws, DW_SPI_INT_TXEI);
245 	}
246 
247 	return IRQ_HANDLED;
248 }
249 
250 static irqreturn_t dw_spi_irq(int irq, void *dev_id)
251 {
252 	struct spi_controller *host = dev_id;
253 	struct dw_spi *dws = spi_controller_get_devdata(host);
254 	u16 irq_status = dw_readl(dws, DW_SPI_ISR) & DW_SPI_INT_MASK;
255 
256 	if (!irq_status)
257 		return IRQ_NONE;
258 
259 	if (!host->cur_msg) {
260 		dw_spi_mask_intr(dws, 0xff);
261 		return IRQ_HANDLED;
262 	}
263 
264 	return dws->transfer_handler(dws);
265 }
266 
267 static u32 dw_spi_prepare_cr0(struct dw_spi *dws, struct spi_device *spi)
268 {
269 	u32 cr0 = 0;
270 
271 	if (dw_spi_ip_is(dws, PSSI)) {
272 		/* CTRLR0[ 5: 4] Frame Format */
273 		cr0 |= FIELD_PREP(DW_PSSI_CTRLR0_FRF_MASK, DW_SPI_CTRLR0_FRF_MOTO_SPI);
274 
275 		/*
276 		 * SPI mode (SCPOL|SCPH)
277 		 * CTRLR0[ 6] Serial Clock Phase
278 		 * CTRLR0[ 7] Serial Clock Polarity
279 		 */
280 		if (spi->mode & SPI_CPOL)
281 			cr0 |= DW_PSSI_CTRLR0_SCPOL;
282 		if (spi->mode & SPI_CPHA)
283 			cr0 |= DW_PSSI_CTRLR0_SCPHA;
284 
285 		/* CTRLR0[11] Shift Register Loop */
286 		if (spi->mode & SPI_LOOP)
287 			cr0 |= DW_PSSI_CTRLR0_SRL;
288 	} else {
289 		/* CTRLR0[ 7: 6] Frame Format */
290 		cr0 |= FIELD_PREP(DW_HSSI_CTRLR0_FRF_MASK, DW_SPI_CTRLR0_FRF_MOTO_SPI);
291 
292 		/*
293 		 * SPI mode (SCPOL|SCPH)
294 		 * CTRLR0[ 8] Serial Clock Phase
295 		 * CTRLR0[ 9] Serial Clock Polarity
296 		 */
297 		if (spi->mode & SPI_CPOL)
298 			cr0 |= DW_HSSI_CTRLR0_SCPOL;
299 		if (spi->mode & SPI_CPHA)
300 			cr0 |= DW_HSSI_CTRLR0_SCPHA;
301 
302 		/* CTRLR0[13] Shift Register Loop */
303 		if (spi->mode & SPI_LOOP)
304 			cr0 |= DW_HSSI_CTRLR0_SRL;
305 
306 		/* CTRLR0[31] MST */
307 		if (dw_spi_ver_is_ge(dws, HSSI, 102A))
308 			cr0 |= DW_HSSI_CTRLR0_MST;
309 	}
310 
311 	return cr0;
312 }
313 
314 void dw_spi_update_config(struct dw_spi *dws, struct spi_device *spi,
315 			  struct dw_spi_cfg *cfg)
316 {
317 	struct dw_spi_chip_data *chip = spi_get_ctldata(spi);
318 	u32 cr0 = chip->cr0;
319 	u32 speed_hz;
320 	u16 clk_div;
321 
322 	/* CTRLR0[ 4/3: 0] or CTRLR0[ 20: 16] Data Frame Size */
323 	cr0 |= (cfg->dfs - 1) << dws->dfs_offset;
324 
325 	if (dw_spi_ip_is(dws, PSSI))
326 		/* CTRLR0[ 9:8] Transfer Mode */
327 		cr0 |= FIELD_PREP(DW_PSSI_CTRLR0_TMOD_MASK, cfg->tmode);
328 	else
329 		/* CTRLR0[11:10] Transfer Mode */
330 		cr0 |= FIELD_PREP(DW_HSSI_CTRLR0_TMOD_MASK, cfg->tmode);
331 
332 	dw_writel(dws, DW_SPI_CTRLR0, cr0);
333 
334 	if (cfg->tmode == DW_SPI_CTRLR0_TMOD_EPROMREAD ||
335 	    cfg->tmode == DW_SPI_CTRLR0_TMOD_RO)
336 		dw_writel(dws, DW_SPI_CTRLR1, cfg->ndf ? cfg->ndf - 1 : 0);
337 
338 	/* Note DW APB SSI clock divider doesn't support odd numbers */
339 	clk_div = (DIV_ROUND_UP(dws->max_freq, cfg->freq) + 1) & 0xfffe;
340 	speed_hz = dws->max_freq / clk_div;
341 
342 	if (dws->current_freq != speed_hz) {
343 		dw_spi_set_clk(dws, clk_div);
344 		dws->current_freq = speed_hz;
345 	}
346 
347 	/* Update RX sample delay if required */
348 	if (dws->cur_rx_sample_dly != chip->rx_sample_dly) {
349 		dw_writel(dws, DW_SPI_RX_SAMPLE_DLY, chip->rx_sample_dly);
350 		dws->cur_rx_sample_dly = chip->rx_sample_dly;
351 	}
352 }
353 EXPORT_SYMBOL_NS_GPL(dw_spi_update_config, SPI_DW_CORE);
354 
355 static void dw_spi_irq_setup(struct dw_spi *dws)
356 {
357 	u16 level;
358 	u8 imask;
359 
360 	/*
361 	 * Originally Tx and Rx data lengths match. Rx FIFO Threshold level
362 	 * will be adjusted at the final stage of the IRQ-based SPI transfer
363 	 * execution so not to lose the leftover of the incoming data.
364 	 */
365 	level = min_t(unsigned int, dws->fifo_len / 2, dws->tx_len);
366 	dw_writel(dws, DW_SPI_TXFTLR, level);
367 	dw_writel(dws, DW_SPI_RXFTLR, level - 1);
368 
369 	dws->transfer_handler = dw_spi_transfer_handler;
370 
371 	imask = DW_SPI_INT_TXEI | DW_SPI_INT_TXOI |
372 		DW_SPI_INT_RXUI | DW_SPI_INT_RXOI | DW_SPI_INT_RXFI;
373 	dw_spi_umask_intr(dws, imask);
374 }
375 
376 /*
377  * The iterative procedure of the poll-based transfer is simple: write as much
378  * as possible to the Tx FIFO, wait until the pending to receive data is ready
379  * to be read, read it from the Rx FIFO and check whether the performed
380  * procedure has been successful.
381  *
382  * Note this method the same way as the IRQ-based transfer won't work well for
383  * the SPI devices connected to the controller with native CS due to the
384  * automatic CS assertion/de-assertion.
385  */
386 static int dw_spi_poll_transfer(struct dw_spi *dws,
387 				struct spi_transfer *transfer)
388 {
389 	struct spi_delay delay;
390 	u16 nbits;
391 	int ret;
392 
393 	delay.unit = SPI_DELAY_UNIT_SCK;
394 	nbits = dws->n_bytes * BITS_PER_BYTE;
395 
396 	do {
397 		dw_writer(dws);
398 
399 		delay.value = nbits * (dws->rx_len - dws->tx_len);
400 		spi_delay_exec(&delay, transfer);
401 
402 		dw_reader(dws);
403 
404 		ret = dw_spi_check_status(dws, true);
405 		if (ret)
406 			return ret;
407 	} while (dws->rx_len);
408 
409 	return 0;
410 }
411 
412 static int dw_spi_transfer_one(struct spi_controller *host,
413 			       struct spi_device *spi,
414 			       struct spi_transfer *transfer)
415 {
416 	struct dw_spi *dws = spi_controller_get_devdata(host);
417 	struct dw_spi_cfg cfg = {
418 		.tmode = DW_SPI_CTRLR0_TMOD_TR,
419 		.dfs = transfer->bits_per_word,
420 		.freq = transfer->speed_hz,
421 	};
422 	int ret;
423 
424 	dws->dma_mapped = 0;
425 	dws->n_bytes = roundup_pow_of_two(BITS_TO_BYTES(transfer->bits_per_word));
426 	dws->tx = (void *)transfer->tx_buf;
427 	dws->tx_len = transfer->len / dws->n_bytes;
428 	dws->rx = transfer->rx_buf;
429 	dws->rx_len = dws->tx_len;
430 
431 	/* Ensure the data above is visible for all CPUs */
432 	smp_mb();
433 
434 	dw_spi_enable_chip(dws, 0);
435 
436 	dw_spi_update_config(dws, spi, &cfg);
437 
438 	transfer->effective_speed_hz = dws->current_freq;
439 
440 	/* Check if current transfer is a DMA transaction */
441 	if (host->can_dma && host->can_dma(host, spi, transfer))
442 		dws->dma_mapped = host->cur_msg_mapped;
443 
444 	/* For poll mode just disable all interrupts */
445 	dw_spi_mask_intr(dws, 0xff);
446 
447 	if (dws->dma_mapped) {
448 		ret = dws->dma_ops->dma_setup(dws, transfer);
449 		if (ret)
450 			return ret;
451 	}
452 
453 	dw_spi_enable_chip(dws, 1);
454 
455 	if (dws->dma_mapped)
456 		return dws->dma_ops->dma_transfer(dws, transfer);
457 	else if (dws->irq == IRQ_NOTCONNECTED)
458 		return dw_spi_poll_transfer(dws, transfer);
459 
460 	dw_spi_irq_setup(dws);
461 
462 	return 1;
463 }
464 
465 static void dw_spi_handle_err(struct spi_controller *host,
466 			      struct spi_message *msg)
467 {
468 	struct dw_spi *dws = spi_controller_get_devdata(host);
469 
470 	if (dws->dma_mapped)
471 		dws->dma_ops->dma_stop(dws);
472 
473 	dw_spi_reset_chip(dws);
474 }
475 
476 static int dw_spi_adjust_mem_op_size(struct spi_mem *mem, struct spi_mem_op *op)
477 {
478 	if (op->data.dir == SPI_MEM_DATA_IN)
479 		op->data.nbytes = clamp_val(op->data.nbytes, 0, DW_SPI_NDF_MASK + 1);
480 
481 	return 0;
482 }
483 
484 static bool dw_spi_supports_mem_op(struct spi_mem *mem,
485 				   const struct spi_mem_op *op)
486 {
487 	if (op->data.buswidth > 1 || op->addr.buswidth > 1 ||
488 	    op->dummy.buswidth > 1 || op->cmd.buswidth > 1)
489 		return false;
490 
491 	return spi_mem_default_supports_op(mem, op);
492 }
493 
494 static int dw_spi_init_mem_buf(struct dw_spi *dws, const struct spi_mem_op *op)
495 {
496 	unsigned int i, j, len;
497 	u8 *out;
498 
499 	/*
500 	 * Calculate the total length of the EEPROM command transfer and
501 	 * either use the pre-allocated buffer or create a temporary one.
502 	 */
503 	len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
504 	if (op->data.dir == SPI_MEM_DATA_OUT)
505 		len += op->data.nbytes;
506 
507 	if (len <= DW_SPI_BUF_SIZE) {
508 		out = dws->buf;
509 	} else {
510 		out = kzalloc(len, GFP_KERNEL);
511 		if (!out)
512 			return -ENOMEM;
513 	}
514 
515 	/*
516 	 * Collect the operation code, address and dummy bytes into the single
517 	 * buffer. If it's a transfer with data to be sent, also copy it into the
518 	 * single buffer in order to speed the data transmission up.
519 	 */
520 	for (i = 0; i < op->cmd.nbytes; ++i)
521 		out[i] = DW_SPI_GET_BYTE(op->cmd.opcode, op->cmd.nbytes - i - 1);
522 	for (j = 0; j < op->addr.nbytes; ++i, ++j)
523 		out[i] = DW_SPI_GET_BYTE(op->addr.val, op->addr.nbytes - j - 1);
524 	for (j = 0; j < op->dummy.nbytes; ++i, ++j)
525 		out[i] = 0x0;
526 
527 	if (op->data.dir == SPI_MEM_DATA_OUT)
528 		memcpy(&out[i], op->data.buf.out, op->data.nbytes);
529 
530 	dws->n_bytes = 1;
531 	dws->tx = out;
532 	dws->tx_len = len;
533 	if (op->data.dir == SPI_MEM_DATA_IN) {
534 		dws->rx = op->data.buf.in;
535 		dws->rx_len = op->data.nbytes;
536 	} else {
537 		dws->rx = NULL;
538 		dws->rx_len = 0;
539 	}
540 
541 	return 0;
542 }
543 
544 static void dw_spi_free_mem_buf(struct dw_spi *dws)
545 {
546 	if (dws->tx != dws->buf)
547 		kfree(dws->tx);
548 }
549 
550 static int dw_spi_write_then_read(struct dw_spi *dws, struct spi_device *spi)
551 {
552 	u32 room, entries, sts;
553 	unsigned int len;
554 	u8 *buf;
555 
556 	/*
557 	 * At initial stage we just pre-fill the Tx FIFO in with no rush,
558 	 * since native CS hasn't been enabled yet and the automatic data
559 	 * transmission won't start til we do that.
560 	 */
561 	len = min(dws->fifo_len, dws->tx_len);
562 	buf = dws->tx;
563 	while (len--)
564 		dw_write_io_reg(dws, DW_SPI_DR, *buf++);
565 
566 	/*
567 	 * After setting any bit in the SER register the transmission will
568 	 * start automatically. We have to keep up with that procedure
569 	 * otherwise the CS de-assertion will happen whereupon the memory
570 	 * operation will be pre-terminated.
571 	 */
572 	len = dws->tx_len - ((void *)buf - dws->tx);
573 	dw_spi_set_cs(spi, false);
574 	while (len) {
575 		entries = readl_relaxed(dws->regs + DW_SPI_TXFLR);
576 		if (!entries) {
577 			dev_err(&dws->host->dev, "CS de-assertion on Tx\n");
578 			return -EIO;
579 		}
580 		room = min(dws->fifo_len - entries, len);
581 		for (; room; --room, --len)
582 			dw_write_io_reg(dws, DW_SPI_DR, *buf++);
583 	}
584 
585 	/*
586 	 * Data fetching will start automatically if the EEPROM-read mode is
587 	 * activated. We have to keep up with the incoming data pace to
588 	 * prevent the Rx FIFO overflow causing the inbound data loss.
589 	 */
590 	len = dws->rx_len;
591 	buf = dws->rx;
592 	while (len) {
593 		entries = readl_relaxed(dws->regs + DW_SPI_RXFLR);
594 		if (!entries) {
595 			sts = readl_relaxed(dws->regs + DW_SPI_RISR);
596 			if (sts & DW_SPI_INT_RXOI) {
597 				dev_err(&dws->host->dev, "FIFO overflow on Rx\n");
598 				return -EIO;
599 			}
600 			continue;
601 		}
602 		entries = min(entries, len);
603 		for (; entries; --entries, --len)
604 			*buf++ = dw_read_io_reg(dws, DW_SPI_DR);
605 	}
606 
607 	return 0;
608 }
609 
610 static inline bool dw_spi_ctlr_busy(struct dw_spi *dws)
611 {
612 	return dw_readl(dws, DW_SPI_SR) & DW_SPI_SR_BUSY;
613 }
614 
615 static int dw_spi_wait_mem_op_done(struct dw_spi *dws)
616 {
617 	int retry = DW_SPI_WAIT_RETRIES;
618 	struct spi_delay delay;
619 	unsigned long ns, us;
620 	u32 nents;
621 
622 	nents = dw_readl(dws, DW_SPI_TXFLR);
623 	ns = NSEC_PER_SEC / dws->current_freq * nents;
624 	ns *= dws->n_bytes * BITS_PER_BYTE;
625 	if (ns <= NSEC_PER_USEC) {
626 		delay.unit = SPI_DELAY_UNIT_NSECS;
627 		delay.value = ns;
628 	} else {
629 		us = DIV_ROUND_UP(ns, NSEC_PER_USEC);
630 		delay.unit = SPI_DELAY_UNIT_USECS;
631 		delay.value = clamp_val(us, 0, USHRT_MAX);
632 	}
633 
634 	while (dw_spi_ctlr_busy(dws) && retry--)
635 		spi_delay_exec(&delay, NULL);
636 
637 	if (retry < 0) {
638 		dev_err(&dws->host->dev, "Mem op hanged up\n");
639 		return -EIO;
640 	}
641 
642 	return 0;
643 }
644 
645 static void dw_spi_stop_mem_op(struct dw_spi *dws, struct spi_device *spi)
646 {
647 	dw_spi_enable_chip(dws, 0);
648 	dw_spi_set_cs(spi, true);
649 	dw_spi_enable_chip(dws, 1);
650 }
651 
652 /*
653  * The SPI memory operation implementation below is the best choice for the
654  * devices, which are selected by the native chip-select lane. It's
655  * specifically developed to workaround the problem with automatic chip-select
656  * lane toggle when there is no data in the Tx FIFO buffer. Luckily the current
657  * SPI-mem core calls exec_op() callback only if the GPIO-based CS is
658  * unavailable.
659  */
660 static int dw_spi_exec_mem_op(struct spi_mem *mem, const struct spi_mem_op *op)
661 {
662 	struct dw_spi *dws = spi_controller_get_devdata(mem->spi->controller);
663 	struct dw_spi_cfg cfg;
664 	unsigned long flags;
665 	int ret;
666 
667 	/*
668 	 * Collect the outbound data into a single buffer to speed the
669 	 * transmission up at least on the initial stage.
670 	 */
671 	ret = dw_spi_init_mem_buf(dws, op);
672 	if (ret)
673 		return ret;
674 
675 	/*
676 	 * DW SPI EEPROM-read mode is required only for the SPI memory Data-IN
677 	 * operation. Transmit-only mode is suitable for the rest of them.
678 	 */
679 	cfg.dfs = 8;
680 	cfg.freq = clamp(mem->spi->max_speed_hz, 0U, dws->max_mem_freq);
681 	if (op->data.dir == SPI_MEM_DATA_IN) {
682 		cfg.tmode = DW_SPI_CTRLR0_TMOD_EPROMREAD;
683 		cfg.ndf = op->data.nbytes;
684 	} else {
685 		cfg.tmode = DW_SPI_CTRLR0_TMOD_TO;
686 	}
687 
688 	dw_spi_enable_chip(dws, 0);
689 
690 	dw_spi_update_config(dws, mem->spi, &cfg);
691 
692 	dw_spi_mask_intr(dws, 0xff);
693 
694 	dw_spi_enable_chip(dws, 1);
695 
696 	/*
697 	 * DW APB SSI controller has very nasty peculiarities. First originally
698 	 * (without any vendor-specific modifications) it doesn't provide a
699 	 * direct way to set and clear the native chip-select signal. Instead
700 	 * the controller asserts the CS lane if Tx FIFO isn't empty and a
701 	 * transmission is going on, and automatically de-asserts it back to
702 	 * the high level if the Tx FIFO doesn't have anything to be pushed
703 	 * out. Due to that a multi-tasking or heavy IRQs activity might be
704 	 * fatal, since the transfer procedure preemption may cause the Tx FIFO
705 	 * getting empty and sudden CS de-assertion, which in the middle of the
706 	 * transfer will most likely cause the data loss. Secondly the
707 	 * EEPROM-read or Read-only DW SPI transfer modes imply the incoming
708 	 * data being automatically pulled in into the Rx FIFO. So if the
709 	 * driver software is late in fetching the data from the FIFO before
710 	 * it's overflown, new incoming data will be lost. In order to make
711 	 * sure the executed memory operations are CS-atomic and to prevent the
712 	 * Rx FIFO overflow we have to disable the local interrupts so to block
713 	 * any preemption during the subsequent IO operations.
714 	 *
715 	 * Note. At some circumstances disabling IRQs may not help to prevent
716 	 * the problems described above. The CS de-assertion and Rx FIFO
717 	 * overflow may still happen due to the relatively slow system bus or
718 	 * CPU not working fast enough, so the write-then-read algo implemented
719 	 * here just won't keep up with the SPI bus data transfer. Such
720 	 * situation is highly platform specific and is supposed to be fixed by
721 	 * manually restricting the SPI bus frequency using the
722 	 * dws->max_mem_freq parameter.
723 	 */
724 	local_irq_save(flags);
725 	preempt_disable();
726 
727 	ret = dw_spi_write_then_read(dws, mem->spi);
728 
729 	local_irq_restore(flags);
730 	preempt_enable();
731 
732 	/*
733 	 * Wait for the operation being finished and check the controller
734 	 * status only if there hasn't been any run-time error detected. In the
735 	 * former case it's just pointless. In the later one to prevent an
736 	 * additional error message printing since any hw error flag being set
737 	 * would be due to an error detected on the data transfer.
738 	 */
739 	if (!ret) {
740 		ret = dw_spi_wait_mem_op_done(dws);
741 		if (!ret)
742 			ret = dw_spi_check_status(dws, true);
743 	}
744 
745 	dw_spi_stop_mem_op(dws, mem->spi);
746 
747 	dw_spi_free_mem_buf(dws);
748 
749 	return ret;
750 }
751 
752 /*
753  * Initialize the default memory operations if a glue layer hasn't specified
754  * custom ones. Direct mapping operations will be preserved anyway since DW SPI
755  * controller doesn't have an embedded dirmap interface. Note the memory
756  * operations implemented in this driver is the best choice only for the DW APB
757  * SSI controller with standard native CS functionality. If a hardware vendor
758  * has fixed the automatic CS assertion/de-assertion peculiarity, then it will
759  * be safer to use the normal SPI-messages-based transfers implementation.
760  */
761 static void dw_spi_init_mem_ops(struct dw_spi *dws)
762 {
763 	if (!dws->mem_ops.exec_op && !(dws->caps & DW_SPI_CAP_CS_OVERRIDE) &&
764 	    !dws->set_cs) {
765 		dws->mem_ops.adjust_op_size = dw_spi_adjust_mem_op_size;
766 		dws->mem_ops.supports_op = dw_spi_supports_mem_op;
767 		dws->mem_ops.exec_op = dw_spi_exec_mem_op;
768 		if (!dws->max_mem_freq)
769 			dws->max_mem_freq = dws->max_freq;
770 	}
771 }
772 
773 /* This may be called twice for each spi dev */
774 static int dw_spi_setup(struct spi_device *spi)
775 {
776 	struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
777 	struct dw_spi_chip_data *chip;
778 
779 	/* Only alloc on first setup */
780 	chip = spi_get_ctldata(spi);
781 	if (!chip) {
782 		struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
783 		u32 rx_sample_dly_ns;
784 
785 		chip = kzalloc(sizeof(*chip), GFP_KERNEL);
786 		if (!chip)
787 			return -ENOMEM;
788 		spi_set_ctldata(spi, chip);
789 		/* Get specific / default rx-sample-delay */
790 		if (device_property_read_u32(&spi->dev,
791 					     "rx-sample-delay-ns",
792 					     &rx_sample_dly_ns) != 0)
793 			/* Use default controller value */
794 			rx_sample_dly_ns = dws->def_rx_sample_dly_ns;
795 		chip->rx_sample_dly = DIV_ROUND_CLOSEST(rx_sample_dly_ns,
796 							NSEC_PER_SEC /
797 							dws->max_freq);
798 	}
799 
800 	/*
801 	 * Update CR0 data each time the setup callback is invoked since
802 	 * the device parameters could have been changed, for instance, by
803 	 * the MMC SPI driver or something else.
804 	 */
805 	chip->cr0 = dw_spi_prepare_cr0(dws, spi);
806 
807 	return 0;
808 }
809 
810 static void dw_spi_cleanup(struct spi_device *spi)
811 {
812 	struct dw_spi_chip_data *chip = spi_get_ctldata(spi);
813 
814 	kfree(chip);
815 	spi_set_ctldata(spi, NULL);
816 }
817 
818 /* Restart the controller, disable all interrupts, clean rx fifo */
819 static void dw_spi_hw_init(struct device *dev, struct dw_spi *dws)
820 {
821 	dw_spi_reset_chip(dws);
822 
823 	/*
824 	 * Retrieve the Synopsys component version if it hasn't been specified
825 	 * by the platform. CoreKit version ID is encoded as a 3-chars ASCII
826 	 * code enclosed with '*' (typical for the most of Synopsys IP-cores).
827 	 */
828 	if (!dws->ver) {
829 		dws->ver = dw_readl(dws, DW_SPI_VERSION);
830 
831 		dev_dbg(dev, "Synopsys DWC%sSSI v%c.%c%c\n",
832 			dw_spi_ip_is(dws, PSSI) ? " APB " : " ",
833 			DW_SPI_GET_BYTE(dws->ver, 3), DW_SPI_GET_BYTE(dws->ver, 2),
834 			DW_SPI_GET_BYTE(dws->ver, 1));
835 	}
836 
837 	/*
838 	 * Try to detect the number of native chip-selects if the platform
839 	 * driver didn't set it up. There can be up to 16 lines configured.
840 	 */
841 	if (!dws->num_cs) {
842 		u32 ser;
843 
844 		dw_writel(dws, DW_SPI_SER, 0xffff);
845 		ser = dw_readl(dws, DW_SPI_SER);
846 		dw_writel(dws, DW_SPI_SER, 0);
847 
848 		dws->num_cs = hweight16(ser);
849 	}
850 
851 	/*
852 	 * Try to detect the FIFO depth if not set by interface driver,
853 	 * the depth could be from 2 to 256 from HW spec
854 	 */
855 	if (!dws->fifo_len) {
856 		u32 fifo;
857 
858 		for (fifo = 1; fifo < 256; fifo++) {
859 			dw_writel(dws, DW_SPI_TXFTLR, fifo);
860 			if (fifo != dw_readl(dws, DW_SPI_TXFTLR))
861 				break;
862 		}
863 		dw_writel(dws, DW_SPI_TXFTLR, 0);
864 
865 		dws->fifo_len = (fifo == 1) ? 0 : fifo;
866 		dev_dbg(dev, "Detected FIFO size: %u bytes\n", dws->fifo_len);
867 	}
868 
869 	/*
870 	 * Detect CTRLR0.DFS field size and offset by testing the lowest bits
871 	 * writability. Note DWC SSI controller also has the extended DFS, but
872 	 * with zero offset.
873 	 */
874 	if (dw_spi_ip_is(dws, PSSI)) {
875 		u32 cr0, tmp = dw_readl(dws, DW_SPI_CTRLR0);
876 
877 		dw_spi_enable_chip(dws, 0);
878 		dw_writel(dws, DW_SPI_CTRLR0, 0xffffffff);
879 		cr0 = dw_readl(dws, DW_SPI_CTRLR0);
880 		dw_writel(dws, DW_SPI_CTRLR0, tmp);
881 		dw_spi_enable_chip(dws, 1);
882 
883 		if (!(cr0 & DW_PSSI_CTRLR0_DFS_MASK)) {
884 			dws->caps |= DW_SPI_CAP_DFS32;
885 			dws->dfs_offset = __bf_shf(DW_PSSI_CTRLR0_DFS32_MASK);
886 			dev_dbg(dev, "Detected 32-bits max data frame size\n");
887 		}
888 	} else {
889 		dws->caps |= DW_SPI_CAP_DFS32;
890 	}
891 
892 	/* enable HW fixup for explicit CS deselect for Amazon's alpine chip */
893 	if (dws->caps & DW_SPI_CAP_CS_OVERRIDE)
894 		dw_writel(dws, DW_SPI_CS_OVERRIDE, 0xF);
895 }
896 
897 int dw_spi_add_host(struct device *dev, struct dw_spi *dws)
898 {
899 	struct spi_controller *host;
900 	int ret;
901 
902 	if (!dws)
903 		return -EINVAL;
904 
905 	host = spi_alloc_host(dev, 0);
906 	if (!host)
907 		return -ENOMEM;
908 
909 	device_set_node(&host->dev, dev_fwnode(dev));
910 
911 	dws->host = host;
912 	dws->dma_addr = (dma_addr_t)(dws->paddr + DW_SPI_DR);
913 
914 	spi_controller_set_devdata(host, dws);
915 
916 	/* Basic HW init */
917 	dw_spi_hw_init(dev, dws);
918 
919 	ret = request_irq(dws->irq, dw_spi_irq, IRQF_SHARED, dev_name(dev),
920 			  host);
921 	if (ret < 0 && ret != -ENOTCONN) {
922 		dev_err(dev, "can not get IRQ\n");
923 		goto err_free_host;
924 	}
925 
926 	dw_spi_init_mem_ops(dws);
927 
928 	host->use_gpio_descriptors = true;
929 	host->mode_bits = SPI_CPOL | SPI_CPHA | SPI_LOOP;
930 	if (dws->caps & DW_SPI_CAP_DFS32)
931 		host->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
932 	else
933 		host->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 16);
934 	host->bus_num = dws->bus_num;
935 	host->num_chipselect = dws->num_cs;
936 	host->setup = dw_spi_setup;
937 	host->cleanup = dw_spi_cleanup;
938 	if (dws->set_cs)
939 		host->set_cs = dws->set_cs;
940 	else
941 		host->set_cs = dw_spi_set_cs;
942 	host->transfer_one = dw_spi_transfer_one;
943 	host->handle_err = dw_spi_handle_err;
944 	if (dws->mem_ops.exec_op)
945 		host->mem_ops = &dws->mem_ops;
946 	host->max_speed_hz = dws->max_freq;
947 	host->flags = SPI_CONTROLLER_GPIO_SS;
948 	host->auto_runtime_pm = true;
949 
950 	/* Get default rx sample delay */
951 	device_property_read_u32(dev, "rx-sample-delay-ns",
952 				 &dws->def_rx_sample_dly_ns);
953 
954 	if (dws->dma_ops && dws->dma_ops->dma_init) {
955 		ret = dws->dma_ops->dma_init(dev, dws);
956 		if (ret == -EPROBE_DEFER) {
957 			goto err_free_irq;
958 		} else if (ret) {
959 			dev_warn(dev, "DMA init failed\n");
960 		} else {
961 			host->can_dma = dws->dma_ops->can_dma;
962 			host->flags |= SPI_CONTROLLER_MUST_TX;
963 		}
964 	}
965 
966 	ret = spi_register_controller(host);
967 	if (ret) {
968 		dev_err_probe(dev, ret, "problem registering spi host\n");
969 		goto err_dma_exit;
970 	}
971 
972 	dw_spi_debugfs_init(dws);
973 	return 0;
974 
975 err_dma_exit:
976 	if (dws->dma_ops && dws->dma_ops->dma_exit)
977 		dws->dma_ops->dma_exit(dws);
978 	dw_spi_enable_chip(dws, 0);
979 err_free_irq:
980 	free_irq(dws->irq, host);
981 err_free_host:
982 	spi_controller_put(host);
983 	return ret;
984 }
985 EXPORT_SYMBOL_NS_GPL(dw_spi_add_host, SPI_DW_CORE);
986 
987 void dw_spi_remove_host(struct dw_spi *dws)
988 {
989 	dw_spi_debugfs_remove(dws);
990 
991 	spi_unregister_controller(dws->host);
992 
993 	if (dws->dma_ops && dws->dma_ops->dma_exit)
994 		dws->dma_ops->dma_exit(dws);
995 
996 	dw_spi_shutdown_chip(dws);
997 
998 	free_irq(dws->irq, dws->host);
999 }
1000 EXPORT_SYMBOL_NS_GPL(dw_spi_remove_host, SPI_DW_CORE);
1001 
1002 int dw_spi_suspend_host(struct dw_spi *dws)
1003 {
1004 	int ret;
1005 
1006 	ret = spi_controller_suspend(dws->host);
1007 	if (ret)
1008 		return ret;
1009 
1010 	dw_spi_shutdown_chip(dws);
1011 	return 0;
1012 }
1013 EXPORT_SYMBOL_NS_GPL(dw_spi_suspend_host, SPI_DW_CORE);
1014 
1015 int dw_spi_resume_host(struct dw_spi *dws)
1016 {
1017 	dw_spi_hw_init(&dws->host->dev, dws);
1018 	return spi_controller_resume(dws->host);
1019 }
1020 EXPORT_SYMBOL_NS_GPL(dw_spi_resume_host, SPI_DW_CORE);
1021 
1022 MODULE_AUTHOR("Feng Tang <feng.tang@intel.com>");
1023 MODULE_DESCRIPTION("Driver for DesignWare SPI controller core");
1024 MODULE_LICENSE("GPL v2");
1025