xref: /openbmc/linux/drivers/spi/spi-dw-core.c (revision fbb6b31a)
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/dma-mapping.h>
10 #include <linux/interrupt.h>
11 #include <linux/module.h>
12 #include <linux/preempt.h>
13 #include <linux/highmem.h>
14 #include <linux/delay.h>
15 #include <linux/slab.h>
16 #include <linux/spi/spi.h>
17 #include <linux/spi/spi-mem.h>
18 #include <linux/string.h>
19 #include <linux/of.h>
20 
21 #include "spi-dw.h"
22 
23 #ifdef CONFIG_DEBUG_FS
24 #include <linux/debugfs.h>
25 #endif
26 
27 /* Slave spi_device related */
28 struct dw_spi_chip_data {
29 	u32 cr0;
30 	u32 rx_sample_dly;	/* RX sample delay */
31 };
32 
33 #ifdef CONFIG_DEBUG_FS
34 
35 #define DW_SPI_DBGFS_REG(_name, _off)	\
36 {					\
37 	.name = _name,			\
38 	.offset = _off,			\
39 }
40 
41 static const struct debugfs_reg32 dw_spi_dbgfs_regs[] = {
42 	DW_SPI_DBGFS_REG("CTRLR0", DW_SPI_CTRLR0),
43 	DW_SPI_DBGFS_REG("CTRLR1", DW_SPI_CTRLR1),
44 	DW_SPI_DBGFS_REG("SSIENR", DW_SPI_SSIENR),
45 	DW_SPI_DBGFS_REG("SER", DW_SPI_SER),
46 	DW_SPI_DBGFS_REG("BAUDR", DW_SPI_BAUDR),
47 	DW_SPI_DBGFS_REG("TXFTLR", DW_SPI_TXFTLR),
48 	DW_SPI_DBGFS_REG("RXFTLR", DW_SPI_RXFTLR),
49 	DW_SPI_DBGFS_REG("TXFLR", DW_SPI_TXFLR),
50 	DW_SPI_DBGFS_REG("RXFLR", DW_SPI_RXFLR),
51 	DW_SPI_DBGFS_REG("SR", DW_SPI_SR),
52 	DW_SPI_DBGFS_REG("IMR", DW_SPI_IMR),
53 	DW_SPI_DBGFS_REG("ISR", DW_SPI_ISR),
54 	DW_SPI_DBGFS_REG("DMACR", DW_SPI_DMACR),
55 	DW_SPI_DBGFS_REG("DMATDLR", DW_SPI_DMATDLR),
56 	DW_SPI_DBGFS_REG("DMARDLR", DW_SPI_DMARDLR),
57 	DW_SPI_DBGFS_REG("RX_SAMPLE_DLY", DW_SPI_RX_SAMPLE_DLY),
58 };
59 
60 static int dw_spi_debugfs_init(struct dw_spi *dws)
61 {
62 	char name[32];
63 
64 	snprintf(name, 32, "dw_spi%d", dws->master->bus_num);
65 	dws->debugfs = debugfs_create_dir(name, NULL);
66 	if (!dws->debugfs)
67 		return -ENOMEM;
68 
69 	dws->regset.regs = dw_spi_dbgfs_regs;
70 	dws->regset.nregs = ARRAY_SIZE(dw_spi_dbgfs_regs);
71 	dws->regset.base = dws->regs;
72 	debugfs_create_regset32("registers", 0400, dws->debugfs, &dws->regset);
73 
74 	return 0;
75 }
76 
77 static void dw_spi_debugfs_remove(struct dw_spi *dws)
78 {
79 	debugfs_remove_recursive(dws->debugfs);
80 }
81 
82 #else
83 static inline int dw_spi_debugfs_init(struct dw_spi *dws)
84 {
85 	return 0;
86 }
87 
88 static inline void dw_spi_debugfs_remove(struct dw_spi *dws)
89 {
90 }
91 #endif /* CONFIG_DEBUG_FS */
92 
93 void dw_spi_set_cs(struct spi_device *spi, bool enable)
94 {
95 	struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
96 	bool cs_high = !!(spi->mode & SPI_CS_HIGH);
97 
98 	/*
99 	 * DW SPI controller demands any native CS being set in order to
100 	 * proceed with data transfer. So in order to activate the SPI
101 	 * communications we must set a corresponding bit in the Slave
102 	 * Enable register no matter whether the SPI core is configured to
103 	 * support active-high or active-low CS level.
104 	 */
105 	if (cs_high == enable)
106 		dw_writel(dws, DW_SPI_SER, BIT(spi->chip_select));
107 	else
108 		dw_writel(dws, DW_SPI_SER, 0);
109 }
110 EXPORT_SYMBOL_NS_GPL(dw_spi_set_cs, SPI_DW_CORE);
111 
112 /* Return the max entries we can fill into tx fifo */
113 static inline u32 dw_spi_tx_max(struct dw_spi *dws)
114 {
115 	u32 tx_room, rxtx_gap;
116 
117 	tx_room = dws->fifo_len - dw_readl(dws, DW_SPI_TXFLR);
118 
119 	/*
120 	 * Another concern is about the tx/rx mismatch, we
121 	 * though to use (dws->fifo_len - rxflr - txflr) as
122 	 * one maximum value for tx, but it doesn't cover the
123 	 * data which is out of tx/rx fifo and inside the
124 	 * shift registers. So a control from sw point of
125 	 * view is taken.
126 	 */
127 	rxtx_gap = dws->fifo_len - (dws->rx_len - dws->tx_len);
128 
129 	return min3((u32)dws->tx_len, tx_room, rxtx_gap);
130 }
131 
132 /* Return the max entries we should read out of rx fifo */
133 static inline u32 dw_spi_rx_max(struct dw_spi *dws)
134 {
135 	return min_t(u32, dws->rx_len, dw_readl(dws, DW_SPI_RXFLR));
136 }
137 
138 static void dw_writer(struct dw_spi *dws)
139 {
140 	u32 max = dw_spi_tx_max(dws);
141 	u32 txw = 0;
142 
143 	while (max--) {
144 		if (dws->tx) {
145 			if (dws->n_bytes == 1)
146 				txw = *(u8 *)(dws->tx);
147 			else if (dws->n_bytes == 2)
148 				txw = *(u16 *)(dws->tx);
149 			else
150 				txw = *(u32 *)(dws->tx);
151 
152 			dws->tx += dws->n_bytes;
153 		}
154 		dw_write_io_reg(dws, DW_SPI_DR, txw);
155 		--dws->tx_len;
156 	}
157 }
158 
159 static void dw_reader(struct dw_spi *dws)
160 {
161 	u32 max = dw_spi_rx_max(dws);
162 	u32 rxw;
163 
164 	while (max--) {
165 		rxw = dw_read_io_reg(dws, DW_SPI_DR);
166 		if (dws->rx) {
167 			if (dws->n_bytes == 1)
168 				*(u8 *)(dws->rx) = rxw;
169 			else if (dws->n_bytes == 2)
170 				*(u16 *)(dws->rx) = rxw;
171 			else
172 				*(u32 *)(dws->rx) = rxw;
173 
174 			dws->rx += dws->n_bytes;
175 		}
176 		--dws->rx_len;
177 	}
178 }
179 
180 int dw_spi_check_status(struct dw_spi *dws, bool raw)
181 {
182 	u32 irq_status;
183 	int ret = 0;
184 
185 	if (raw)
186 		irq_status = dw_readl(dws, DW_SPI_RISR);
187 	else
188 		irq_status = dw_readl(dws, DW_SPI_ISR);
189 
190 	if (irq_status & DW_SPI_INT_RXOI) {
191 		dev_err(&dws->master->dev, "RX FIFO overflow detected\n");
192 		ret = -EIO;
193 	}
194 
195 	if (irq_status & DW_SPI_INT_RXUI) {
196 		dev_err(&dws->master->dev, "RX FIFO underflow detected\n");
197 		ret = -EIO;
198 	}
199 
200 	if (irq_status & DW_SPI_INT_TXOI) {
201 		dev_err(&dws->master->dev, "TX FIFO overflow detected\n");
202 		ret = -EIO;
203 	}
204 
205 	/* Generically handle the erroneous situation */
206 	if (ret) {
207 		dw_spi_reset_chip(dws);
208 		if (dws->master->cur_msg)
209 			dws->master->cur_msg->status = ret;
210 	}
211 
212 	return ret;
213 }
214 EXPORT_SYMBOL_NS_GPL(dw_spi_check_status, SPI_DW_CORE);
215 
216 static irqreturn_t dw_spi_transfer_handler(struct dw_spi *dws)
217 {
218 	u16 irq_status = dw_readl(dws, DW_SPI_ISR);
219 
220 	if (dw_spi_check_status(dws, false)) {
221 		spi_finalize_current_transfer(dws->master);
222 		return IRQ_HANDLED;
223 	}
224 
225 	/*
226 	 * Read data from the Rx FIFO every time we've got a chance executing
227 	 * this method. If there is nothing left to receive, terminate the
228 	 * procedure. Otherwise adjust the Rx FIFO Threshold level if it's a
229 	 * final stage of the transfer. By doing so we'll get the next IRQ
230 	 * right when the leftover incoming data is received.
231 	 */
232 	dw_reader(dws);
233 	if (!dws->rx_len) {
234 		dw_spi_mask_intr(dws, 0xff);
235 		spi_finalize_current_transfer(dws->master);
236 	} else if (dws->rx_len <= dw_readl(dws, DW_SPI_RXFTLR)) {
237 		dw_writel(dws, DW_SPI_RXFTLR, dws->rx_len - 1);
238 	}
239 
240 	/*
241 	 * Send data out if Tx FIFO Empty IRQ is received. The IRQ will be
242 	 * disabled after the data transmission is finished so not to
243 	 * have the TXE IRQ flood at the final stage of the transfer.
244 	 */
245 	if (irq_status & DW_SPI_INT_TXEI) {
246 		dw_writer(dws);
247 		if (!dws->tx_len)
248 			dw_spi_mask_intr(dws, DW_SPI_INT_TXEI);
249 	}
250 
251 	return IRQ_HANDLED;
252 }
253 
254 static irqreturn_t dw_spi_irq(int irq, void *dev_id)
255 {
256 	struct spi_controller *master = dev_id;
257 	struct dw_spi *dws = spi_controller_get_devdata(master);
258 	u16 irq_status = dw_readl(dws, DW_SPI_ISR) & DW_SPI_INT_MASK;
259 
260 	if (!irq_status)
261 		return IRQ_NONE;
262 
263 	if (!master->cur_msg) {
264 		dw_spi_mask_intr(dws, 0xff);
265 		return IRQ_HANDLED;
266 	}
267 
268 	return dws->transfer_handler(dws);
269 }
270 
271 static u32 dw_spi_prepare_cr0(struct dw_spi *dws, struct spi_device *spi)
272 {
273 	u32 cr0 = 0;
274 
275 	if (dw_spi_ip_is(dws, PSSI)) {
276 		/* CTRLR0[ 5: 4] Frame Format */
277 		cr0 |= FIELD_PREP(DW_PSSI_CTRLR0_FRF_MASK, DW_SPI_CTRLR0_FRF_MOTO_SPI);
278 
279 		/*
280 		 * SPI mode (SCPOL|SCPH)
281 		 * CTRLR0[ 6] Serial Clock Phase
282 		 * CTRLR0[ 7] Serial Clock Polarity
283 		 */
284 		if (spi->mode & SPI_CPOL)
285 			cr0 |= DW_PSSI_CTRLR0_SCPOL;
286 		if (spi->mode & SPI_CPHA)
287 			cr0 |= DW_PSSI_CTRLR0_SCPHA;
288 
289 		/* CTRLR0[11] Shift Register Loop */
290 		if (spi->mode & SPI_LOOP)
291 			cr0 |= DW_PSSI_CTRLR0_SRL;
292 	} else {
293 		/* CTRLR0[ 7: 6] Frame Format */
294 		cr0 |= FIELD_PREP(DW_HSSI_CTRLR0_FRF_MASK, DW_SPI_CTRLR0_FRF_MOTO_SPI);
295 
296 		/*
297 		 * SPI mode (SCPOL|SCPH)
298 		 * CTRLR0[ 8] Serial Clock Phase
299 		 * CTRLR0[ 9] Serial Clock Polarity
300 		 */
301 		if (spi->mode & SPI_CPOL)
302 			cr0 |= DW_HSSI_CTRLR0_SCPOL;
303 		if (spi->mode & SPI_CPHA)
304 			cr0 |= DW_HSSI_CTRLR0_SCPHA;
305 
306 		/* CTRLR0[13] Shift Register Loop */
307 		if (spi->mode & SPI_LOOP)
308 			cr0 |= DW_HSSI_CTRLR0_SRL;
309 
310 		if (dws->caps & DW_SPI_CAP_KEEMBAY_MST)
311 			cr0 |= DW_HSSI_CTRLR0_KEEMBAY_MST;
312 	}
313 
314 	return cr0;
315 }
316 
317 void dw_spi_update_config(struct dw_spi *dws, struct spi_device *spi,
318 			  struct dw_spi_cfg *cfg)
319 {
320 	struct dw_spi_chip_data *chip = spi_get_ctldata(spi);
321 	u32 cr0 = chip->cr0;
322 	u32 speed_hz;
323 	u16 clk_div;
324 
325 	/* CTRLR0[ 4/3: 0] or CTRLR0[ 20: 16] Data Frame Size */
326 	cr0 |= (cfg->dfs - 1) << dws->dfs_offset;
327 
328 	if (dw_spi_ip_is(dws, PSSI))
329 		/* CTRLR0[ 9:8] Transfer Mode */
330 		cr0 |= FIELD_PREP(DW_PSSI_CTRLR0_TMOD_MASK, cfg->tmode);
331 	else
332 		/* CTRLR0[11:10] Transfer Mode */
333 		cr0 |= FIELD_PREP(DW_HSSI_CTRLR0_TMOD_MASK, cfg->tmode);
334 
335 	dw_writel(dws, DW_SPI_CTRLR0, cr0);
336 
337 	if (cfg->tmode == DW_SPI_CTRLR0_TMOD_EPROMREAD ||
338 	    cfg->tmode == DW_SPI_CTRLR0_TMOD_RO)
339 		dw_writel(dws, DW_SPI_CTRLR1, cfg->ndf ? cfg->ndf - 1 : 0);
340 
341 	/* Note DW APB SSI clock divider doesn't support odd numbers */
342 	clk_div = (DIV_ROUND_UP(dws->max_freq, cfg->freq) + 1) & 0xfffe;
343 	speed_hz = dws->max_freq / clk_div;
344 
345 	if (dws->current_freq != speed_hz) {
346 		dw_spi_set_clk(dws, clk_div);
347 		dws->current_freq = speed_hz;
348 	}
349 
350 	/* Update RX sample delay if required */
351 	if (dws->cur_rx_sample_dly != chip->rx_sample_dly) {
352 		dw_writel(dws, DW_SPI_RX_SAMPLE_DLY, chip->rx_sample_dly);
353 		dws->cur_rx_sample_dly = chip->rx_sample_dly;
354 	}
355 }
356 EXPORT_SYMBOL_NS_GPL(dw_spi_update_config, SPI_DW_CORE);
357 
358 static void dw_spi_irq_setup(struct dw_spi *dws)
359 {
360 	u16 level;
361 	u8 imask;
362 
363 	/*
364 	 * Originally Tx and Rx data lengths match. Rx FIFO Threshold level
365 	 * will be adjusted at the final stage of the IRQ-based SPI transfer
366 	 * execution so not to lose the leftover of the incoming data.
367 	 */
368 	level = min_t(u16, dws->fifo_len / 2, dws->tx_len);
369 	dw_writel(dws, DW_SPI_TXFTLR, level);
370 	dw_writel(dws, DW_SPI_RXFTLR, level - 1);
371 
372 	dws->transfer_handler = dw_spi_transfer_handler;
373 
374 	imask = DW_SPI_INT_TXEI | DW_SPI_INT_TXOI |
375 		DW_SPI_INT_RXUI | DW_SPI_INT_RXOI | DW_SPI_INT_RXFI;
376 	dw_spi_umask_intr(dws, imask);
377 }
378 
379 /*
380  * The iterative procedure of the poll-based transfer is simple: write as much
381  * as possible to the Tx FIFO, wait until the pending to receive data is ready
382  * to be read, read it from the Rx FIFO and check whether the performed
383  * procedure has been successful.
384  *
385  * Note this method the same way as the IRQ-based transfer won't work well for
386  * the SPI devices connected to the controller with native CS due to the
387  * automatic CS assertion/de-assertion.
388  */
389 static int dw_spi_poll_transfer(struct dw_spi *dws,
390 				struct spi_transfer *transfer)
391 {
392 	struct spi_delay delay;
393 	u16 nbits;
394 	int ret;
395 
396 	delay.unit = SPI_DELAY_UNIT_SCK;
397 	nbits = dws->n_bytes * BITS_PER_BYTE;
398 
399 	do {
400 		dw_writer(dws);
401 
402 		delay.value = nbits * (dws->rx_len - dws->tx_len);
403 		spi_delay_exec(&delay, transfer);
404 
405 		dw_reader(dws);
406 
407 		ret = dw_spi_check_status(dws, true);
408 		if (ret)
409 			return ret;
410 	} while (dws->rx_len);
411 
412 	return 0;
413 }
414 
415 static int dw_spi_transfer_one(struct spi_controller *master,
416 			       struct spi_device *spi,
417 			       struct spi_transfer *transfer)
418 {
419 	struct dw_spi *dws = spi_controller_get_devdata(master);
420 	struct dw_spi_cfg cfg = {
421 		.tmode = DW_SPI_CTRLR0_TMOD_TR,
422 		.dfs = transfer->bits_per_word,
423 		.freq = transfer->speed_hz,
424 	};
425 	int ret;
426 
427 	dws->dma_mapped = 0;
428 	dws->n_bytes = DIV_ROUND_UP(transfer->bits_per_word, BITS_PER_BYTE);
429 	dws->tx = (void *)transfer->tx_buf;
430 	dws->tx_len = transfer->len / dws->n_bytes;
431 	dws->rx = transfer->rx_buf;
432 	dws->rx_len = dws->tx_len;
433 
434 	/* Ensure the data above is visible for all CPUs */
435 	smp_mb();
436 
437 	dw_spi_enable_chip(dws, 0);
438 
439 	dw_spi_update_config(dws, spi, &cfg);
440 
441 	transfer->effective_speed_hz = dws->current_freq;
442 
443 	/* Check if current transfer is a DMA transaction */
444 	if (master->can_dma && master->can_dma(master, spi, transfer))
445 		dws->dma_mapped = master->cur_msg_mapped;
446 
447 	/* For poll mode just disable all interrupts */
448 	dw_spi_mask_intr(dws, 0xff);
449 
450 	if (dws->dma_mapped) {
451 		ret = dws->dma_ops->dma_setup(dws, transfer);
452 		if (ret)
453 			return ret;
454 	}
455 
456 	dw_spi_enable_chip(dws, 1);
457 
458 	if (dws->dma_mapped)
459 		return dws->dma_ops->dma_transfer(dws, transfer);
460 	else if (dws->irq == IRQ_NOTCONNECTED)
461 		return dw_spi_poll_transfer(dws, transfer);
462 
463 	dw_spi_irq_setup(dws);
464 
465 	return 1;
466 }
467 
468 static void dw_spi_handle_err(struct spi_controller *master,
469 			      struct spi_message *msg)
470 {
471 	struct dw_spi *dws = spi_controller_get_devdata(master);
472 
473 	if (dws->dma_mapped)
474 		dws->dma_ops->dma_stop(dws);
475 
476 	dw_spi_reset_chip(dws);
477 }
478 
479 static int dw_spi_adjust_mem_op_size(struct spi_mem *mem, struct spi_mem_op *op)
480 {
481 	if (op->data.dir == SPI_MEM_DATA_IN)
482 		op->data.nbytes = clamp_val(op->data.nbytes, 0, DW_SPI_NDF_MASK + 1);
483 
484 	return 0;
485 }
486 
487 static bool dw_spi_supports_mem_op(struct spi_mem *mem,
488 				   const struct spi_mem_op *op)
489 {
490 	if (op->data.buswidth > 1 || op->addr.buswidth > 1 ||
491 	    op->dummy.buswidth > 1 || op->cmd.buswidth > 1)
492 		return false;
493 
494 	return spi_mem_default_supports_op(mem, op);
495 }
496 
497 static int dw_spi_init_mem_buf(struct dw_spi *dws, const struct spi_mem_op *op)
498 {
499 	unsigned int i, j, len;
500 	u8 *out;
501 
502 	/*
503 	 * Calculate the total length of the EEPROM command transfer and
504 	 * either use the pre-allocated buffer or create a temporary one.
505 	 */
506 	len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
507 	if (op->data.dir == SPI_MEM_DATA_OUT)
508 		len += op->data.nbytes;
509 
510 	if (len <= DW_SPI_BUF_SIZE) {
511 		out = dws->buf;
512 	} else {
513 		out = kzalloc(len, GFP_KERNEL);
514 		if (!out)
515 			return -ENOMEM;
516 	}
517 
518 	/*
519 	 * Collect the operation code, address and dummy bytes into the single
520 	 * buffer. If it's a transfer with data to be sent, also copy it into the
521 	 * single buffer in order to speed the data transmission up.
522 	 */
523 	for (i = 0; i < op->cmd.nbytes; ++i)
524 		out[i] = DW_SPI_GET_BYTE(op->cmd.opcode, op->cmd.nbytes - i - 1);
525 	for (j = 0; j < op->addr.nbytes; ++i, ++j)
526 		out[i] = DW_SPI_GET_BYTE(op->addr.val, op->addr.nbytes - j - 1);
527 	for (j = 0; j < op->dummy.nbytes; ++i, ++j)
528 		out[i] = 0x0;
529 
530 	if (op->data.dir == SPI_MEM_DATA_OUT)
531 		memcpy(&out[i], op->data.buf.out, op->data.nbytes);
532 
533 	dws->n_bytes = 1;
534 	dws->tx = out;
535 	dws->tx_len = len;
536 	if (op->data.dir == SPI_MEM_DATA_IN) {
537 		dws->rx = op->data.buf.in;
538 		dws->rx_len = op->data.nbytes;
539 	} else {
540 		dws->rx = NULL;
541 		dws->rx_len = 0;
542 	}
543 
544 	return 0;
545 }
546 
547 static void dw_spi_free_mem_buf(struct dw_spi *dws)
548 {
549 	if (dws->tx != dws->buf)
550 		kfree(dws->tx);
551 }
552 
553 static int dw_spi_write_then_read(struct dw_spi *dws, struct spi_device *spi)
554 {
555 	u32 room, entries, sts;
556 	unsigned int len;
557 	u8 *buf;
558 
559 	/*
560 	 * At initial stage we just pre-fill the Tx FIFO in with no rush,
561 	 * since native CS hasn't been enabled yet and the automatic data
562 	 * transmission won't start til we do that.
563 	 */
564 	len = min(dws->fifo_len, dws->tx_len);
565 	buf = dws->tx;
566 	while (len--)
567 		dw_write_io_reg(dws, DW_SPI_DR, *buf++);
568 
569 	/*
570 	 * After setting any bit in the SER register the transmission will
571 	 * start automatically. We have to keep up with that procedure
572 	 * otherwise the CS de-assertion will happen whereupon the memory
573 	 * operation will be pre-terminated.
574 	 */
575 	len = dws->tx_len - ((void *)buf - dws->tx);
576 	dw_spi_set_cs(spi, false);
577 	while (len) {
578 		entries = readl_relaxed(dws->regs + DW_SPI_TXFLR);
579 		if (!entries) {
580 			dev_err(&dws->master->dev, "CS de-assertion on Tx\n");
581 			return -EIO;
582 		}
583 		room = min(dws->fifo_len - entries, len);
584 		for (; room; --room, --len)
585 			dw_write_io_reg(dws, DW_SPI_DR, *buf++);
586 	}
587 
588 	/*
589 	 * Data fetching will start automatically if the EEPROM-read mode is
590 	 * activated. We have to keep up with the incoming data pace to
591 	 * prevent the Rx FIFO overflow causing the inbound data loss.
592 	 */
593 	len = dws->rx_len;
594 	buf = dws->rx;
595 	while (len) {
596 		entries = readl_relaxed(dws->regs + DW_SPI_RXFLR);
597 		if (!entries) {
598 			sts = readl_relaxed(dws->regs + DW_SPI_RISR);
599 			if (sts & DW_SPI_INT_RXOI) {
600 				dev_err(&dws->master->dev, "FIFO overflow on Rx\n");
601 				return -EIO;
602 			}
603 			continue;
604 		}
605 		entries = min(entries, len);
606 		for (; entries; --entries, --len)
607 			*buf++ = dw_read_io_reg(dws, DW_SPI_DR);
608 	}
609 
610 	return 0;
611 }
612 
613 static inline bool dw_spi_ctlr_busy(struct dw_spi *dws)
614 {
615 	return dw_readl(dws, DW_SPI_SR) & DW_SPI_SR_BUSY;
616 }
617 
618 static int dw_spi_wait_mem_op_done(struct dw_spi *dws)
619 {
620 	int retry = DW_SPI_WAIT_RETRIES;
621 	struct spi_delay delay;
622 	unsigned long ns, us;
623 	u32 nents;
624 
625 	nents = dw_readl(dws, DW_SPI_TXFLR);
626 	ns = NSEC_PER_SEC / dws->current_freq * nents;
627 	ns *= dws->n_bytes * BITS_PER_BYTE;
628 	if (ns <= NSEC_PER_USEC) {
629 		delay.unit = SPI_DELAY_UNIT_NSECS;
630 		delay.value = ns;
631 	} else {
632 		us = DIV_ROUND_UP(ns, NSEC_PER_USEC);
633 		delay.unit = SPI_DELAY_UNIT_USECS;
634 		delay.value = clamp_val(us, 0, USHRT_MAX);
635 	}
636 
637 	while (dw_spi_ctlr_busy(dws) && retry--)
638 		spi_delay_exec(&delay, NULL);
639 
640 	if (retry < 0) {
641 		dev_err(&dws->master->dev, "Mem op hanged up\n");
642 		return -EIO;
643 	}
644 
645 	return 0;
646 }
647 
648 static void dw_spi_stop_mem_op(struct dw_spi *dws, struct spi_device *spi)
649 {
650 	dw_spi_enable_chip(dws, 0);
651 	dw_spi_set_cs(spi, true);
652 	dw_spi_enable_chip(dws, 1);
653 }
654 
655 /*
656  * The SPI memory operation implementation below is the best choice for the
657  * devices, which are selected by the native chip-select lane. It's
658  * specifically developed to workaround the problem with automatic chip-select
659  * lane toggle when there is no data in the Tx FIFO buffer. Luckily the current
660  * SPI-mem core calls exec_op() callback only if the GPIO-based CS is
661  * unavailable.
662  */
663 static int dw_spi_exec_mem_op(struct spi_mem *mem, const struct spi_mem_op *op)
664 {
665 	struct dw_spi *dws = spi_controller_get_devdata(mem->spi->controller);
666 	struct dw_spi_cfg cfg;
667 	unsigned long flags;
668 	int ret;
669 
670 	/*
671 	 * Collect the outbound data into a single buffer to speed the
672 	 * transmission up at least on the initial stage.
673 	 */
674 	ret = dw_spi_init_mem_buf(dws, op);
675 	if (ret)
676 		return ret;
677 
678 	/*
679 	 * DW SPI EEPROM-read mode is required only for the SPI memory Data-IN
680 	 * operation. Transmit-only mode is suitable for the rest of them.
681 	 */
682 	cfg.dfs = 8;
683 	cfg.freq = clamp(mem->spi->max_speed_hz, 0U, dws->max_mem_freq);
684 	if (op->data.dir == SPI_MEM_DATA_IN) {
685 		cfg.tmode = DW_SPI_CTRLR0_TMOD_EPROMREAD;
686 		cfg.ndf = op->data.nbytes;
687 	} else {
688 		cfg.tmode = DW_SPI_CTRLR0_TMOD_TO;
689 	}
690 
691 	dw_spi_enable_chip(dws, 0);
692 
693 	dw_spi_update_config(dws, mem->spi, &cfg);
694 
695 	dw_spi_mask_intr(dws, 0xff);
696 
697 	dw_spi_enable_chip(dws, 1);
698 
699 	/*
700 	 * DW APB SSI controller has very nasty peculiarities. First originally
701 	 * (without any vendor-specific modifications) it doesn't provide a
702 	 * direct way to set and clear the native chip-select signal. Instead
703 	 * the controller asserts the CS lane if Tx FIFO isn't empty and a
704 	 * transmission is going on, and automatically de-asserts it back to
705 	 * the high level if the Tx FIFO doesn't have anything to be pushed
706 	 * out. Due to that a multi-tasking or heavy IRQs activity might be
707 	 * fatal, since the transfer procedure preemption may cause the Tx FIFO
708 	 * getting empty and sudden CS de-assertion, which in the middle of the
709 	 * transfer will most likely cause the data loss. Secondly the
710 	 * EEPROM-read or Read-only DW SPI transfer modes imply the incoming
711 	 * data being automatically pulled in into the Rx FIFO. So if the
712 	 * driver software is late in fetching the data from the FIFO before
713 	 * it's overflown, new incoming data will be lost. In order to make
714 	 * sure the executed memory operations are CS-atomic and to prevent the
715 	 * Rx FIFO overflow we have to disable the local interrupts so to block
716 	 * any preemption during the subsequent IO operations.
717 	 *
718 	 * Note. At some circumstances disabling IRQs may not help to prevent
719 	 * the problems described above. The CS de-assertion and Rx FIFO
720 	 * overflow may still happen due to the relatively slow system bus or
721 	 * CPU not working fast enough, so the write-then-read algo implemented
722 	 * here just won't keep up with the SPI bus data transfer. Such
723 	 * situation is highly platform specific and is supposed to be fixed by
724 	 * manually restricting the SPI bus frequency using the
725 	 * dws->max_mem_freq parameter.
726 	 */
727 	local_irq_save(flags);
728 	preempt_disable();
729 
730 	ret = dw_spi_write_then_read(dws, mem->spi);
731 
732 	local_irq_restore(flags);
733 	preempt_enable();
734 
735 	/*
736 	 * Wait for the operation being finished and check the controller
737 	 * status only if there hasn't been any run-time error detected. In the
738 	 * former case it's just pointless. In the later one to prevent an
739 	 * additional error message printing since any hw error flag being set
740 	 * would be due to an error detected on the data transfer.
741 	 */
742 	if (!ret) {
743 		ret = dw_spi_wait_mem_op_done(dws);
744 		if (!ret)
745 			ret = dw_spi_check_status(dws, true);
746 	}
747 
748 	dw_spi_stop_mem_op(dws, mem->spi);
749 
750 	dw_spi_free_mem_buf(dws);
751 
752 	return ret;
753 }
754 
755 /*
756  * Initialize the default memory operations if a glue layer hasn't specified
757  * custom ones. Direct mapping operations will be preserved anyway since DW SPI
758  * controller doesn't have an embedded dirmap interface. Note the memory
759  * operations implemented in this driver is the best choice only for the DW APB
760  * SSI controller with standard native CS functionality. If a hardware vendor
761  * has fixed the automatic CS assertion/de-assertion peculiarity, then it will
762  * be safer to use the normal SPI-messages-based transfers implementation.
763  */
764 static void dw_spi_init_mem_ops(struct dw_spi *dws)
765 {
766 	if (!dws->mem_ops.exec_op && !(dws->caps & DW_SPI_CAP_CS_OVERRIDE) &&
767 	    !dws->set_cs) {
768 		dws->mem_ops.adjust_op_size = dw_spi_adjust_mem_op_size;
769 		dws->mem_ops.supports_op = dw_spi_supports_mem_op;
770 		dws->mem_ops.exec_op = dw_spi_exec_mem_op;
771 		if (!dws->max_mem_freq)
772 			dws->max_mem_freq = dws->max_freq;
773 	}
774 }
775 
776 /* This may be called twice for each spi dev */
777 static int dw_spi_setup(struct spi_device *spi)
778 {
779 	struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
780 	struct dw_spi_chip_data *chip;
781 
782 	/* Only alloc on first setup */
783 	chip = spi_get_ctldata(spi);
784 	if (!chip) {
785 		struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
786 		u32 rx_sample_dly_ns;
787 
788 		chip = kzalloc(sizeof(*chip), GFP_KERNEL);
789 		if (!chip)
790 			return -ENOMEM;
791 		spi_set_ctldata(spi, chip);
792 		/* Get specific / default rx-sample-delay */
793 		if (device_property_read_u32(&spi->dev,
794 					     "rx-sample-delay-ns",
795 					     &rx_sample_dly_ns) != 0)
796 			/* Use default controller value */
797 			rx_sample_dly_ns = dws->def_rx_sample_dly_ns;
798 		chip->rx_sample_dly = DIV_ROUND_CLOSEST(rx_sample_dly_ns,
799 							NSEC_PER_SEC /
800 							dws->max_freq);
801 	}
802 
803 	/*
804 	 * Update CR0 data each time the setup callback is invoked since
805 	 * the device parameters could have been changed, for instance, by
806 	 * the MMC SPI driver or something else.
807 	 */
808 	chip->cr0 = dw_spi_prepare_cr0(dws, spi);
809 
810 	return 0;
811 }
812 
813 static void dw_spi_cleanup(struct spi_device *spi)
814 {
815 	struct dw_spi_chip_data *chip = spi_get_ctldata(spi);
816 
817 	kfree(chip);
818 	spi_set_ctldata(spi, NULL);
819 }
820 
821 /* Restart the controller, disable all interrupts, clean rx fifo */
822 static void dw_spi_hw_init(struct device *dev, struct dw_spi *dws)
823 {
824 	dw_spi_reset_chip(dws);
825 
826 	/*
827 	 * Retrieve the Synopsys component version if it hasn't been specified
828 	 * by the platform. CoreKit version ID is encoded as a 3-chars ASCII
829 	 * code enclosed with '*' (typical for the most of Synopsys IP-cores).
830 	 */
831 	if (!dws->ver) {
832 		dws->ver = dw_readl(dws, DW_SPI_VERSION);
833 
834 		dev_dbg(dev, "Synopsys DWC%sSSI v%c.%c%c\n",
835 			dw_spi_ip_is(dws, PSSI) ? " APB " : " ",
836 			DW_SPI_GET_BYTE(dws->ver, 3), DW_SPI_GET_BYTE(dws->ver, 2),
837 			DW_SPI_GET_BYTE(dws->ver, 1));
838 	}
839 
840 	/*
841 	 * Try to detect the FIFO depth if not set by interface driver,
842 	 * the depth could be from 2 to 256 from HW spec
843 	 */
844 	if (!dws->fifo_len) {
845 		u32 fifo;
846 
847 		for (fifo = 1; fifo < 256; fifo++) {
848 			dw_writel(dws, DW_SPI_TXFTLR, fifo);
849 			if (fifo != dw_readl(dws, DW_SPI_TXFTLR))
850 				break;
851 		}
852 		dw_writel(dws, DW_SPI_TXFTLR, 0);
853 
854 		dws->fifo_len = (fifo == 1) ? 0 : fifo;
855 		dev_dbg(dev, "Detected FIFO size: %u bytes\n", dws->fifo_len);
856 	}
857 
858 	/*
859 	 * Detect CTRLR0.DFS field size and offset by testing the lowest bits
860 	 * writability. Note DWC SSI controller also has the extended DFS, but
861 	 * with zero offset.
862 	 */
863 	if (dw_spi_ip_is(dws, PSSI)) {
864 		u32 cr0, tmp = dw_readl(dws, DW_SPI_CTRLR0);
865 
866 		dw_spi_enable_chip(dws, 0);
867 		dw_writel(dws, DW_SPI_CTRLR0, 0xffffffff);
868 		cr0 = dw_readl(dws, DW_SPI_CTRLR0);
869 		dw_writel(dws, DW_SPI_CTRLR0, tmp);
870 		dw_spi_enable_chip(dws, 1);
871 
872 		if (!(cr0 & DW_PSSI_CTRLR0_DFS_MASK)) {
873 			dws->caps |= DW_SPI_CAP_DFS32;
874 			dws->dfs_offset = __bf_shf(DW_PSSI_CTRLR0_DFS32_MASK);
875 			dev_dbg(dev, "Detected 32-bits max data frame size\n");
876 		}
877 	} else {
878 		dws->caps |= DW_SPI_CAP_DFS32;
879 	}
880 
881 	/* enable HW fixup for explicit CS deselect for Amazon's alpine chip */
882 	if (dws->caps & DW_SPI_CAP_CS_OVERRIDE)
883 		dw_writel(dws, DW_SPI_CS_OVERRIDE, 0xF);
884 }
885 
886 int dw_spi_add_host(struct device *dev, struct dw_spi *dws)
887 {
888 	struct spi_controller *master;
889 	int ret;
890 
891 	if (!dws)
892 		return -EINVAL;
893 
894 	master = spi_alloc_master(dev, 0);
895 	if (!master)
896 		return -ENOMEM;
897 
898 	device_set_node(&master->dev, dev_fwnode(dev));
899 
900 	dws->master = master;
901 	dws->dma_addr = (dma_addr_t)(dws->paddr + DW_SPI_DR);
902 
903 	spi_controller_set_devdata(master, dws);
904 
905 	/* Basic HW init */
906 	dw_spi_hw_init(dev, dws);
907 
908 	ret = request_irq(dws->irq, dw_spi_irq, IRQF_SHARED, dev_name(dev),
909 			  master);
910 	if (ret < 0 && ret != -ENOTCONN) {
911 		dev_err(dev, "can not get IRQ\n");
912 		goto err_free_master;
913 	}
914 
915 	dw_spi_init_mem_ops(dws);
916 
917 	master->use_gpio_descriptors = true;
918 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_LOOP;
919 	if (dws->caps & DW_SPI_CAP_DFS32)
920 		master->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
921 	else
922 		master->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 16);
923 	master->bus_num = dws->bus_num;
924 	master->num_chipselect = dws->num_cs;
925 	master->setup = dw_spi_setup;
926 	master->cleanup = dw_spi_cleanup;
927 	if (dws->set_cs)
928 		master->set_cs = dws->set_cs;
929 	else
930 		master->set_cs = dw_spi_set_cs;
931 	master->transfer_one = dw_spi_transfer_one;
932 	master->handle_err = dw_spi_handle_err;
933 	if (dws->mem_ops.exec_op)
934 		master->mem_ops = &dws->mem_ops;
935 	master->max_speed_hz = dws->max_freq;
936 	master->flags = SPI_MASTER_GPIO_SS;
937 	master->auto_runtime_pm = true;
938 
939 	/* Get default rx sample delay */
940 	device_property_read_u32(dev, "rx-sample-delay-ns",
941 				 &dws->def_rx_sample_dly_ns);
942 
943 	if (dws->dma_ops && dws->dma_ops->dma_init) {
944 		ret = dws->dma_ops->dma_init(dev, dws);
945 		if (ret) {
946 			dev_warn(dev, "DMA init failed\n");
947 		} else {
948 			master->can_dma = dws->dma_ops->can_dma;
949 			master->flags |= SPI_CONTROLLER_MUST_TX;
950 		}
951 	}
952 
953 	ret = spi_register_controller(master);
954 	if (ret) {
955 		dev_err(&master->dev, "problem registering spi master\n");
956 		goto err_dma_exit;
957 	}
958 
959 	dw_spi_debugfs_init(dws);
960 	return 0;
961 
962 err_dma_exit:
963 	if (dws->dma_ops && dws->dma_ops->dma_exit)
964 		dws->dma_ops->dma_exit(dws);
965 	dw_spi_enable_chip(dws, 0);
966 	free_irq(dws->irq, master);
967 err_free_master:
968 	spi_controller_put(master);
969 	return ret;
970 }
971 EXPORT_SYMBOL_NS_GPL(dw_spi_add_host, SPI_DW_CORE);
972 
973 void dw_spi_remove_host(struct dw_spi *dws)
974 {
975 	dw_spi_debugfs_remove(dws);
976 
977 	spi_unregister_controller(dws->master);
978 
979 	if (dws->dma_ops && dws->dma_ops->dma_exit)
980 		dws->dma_ops->dma_exit(dws);
981 
982 	dw_spi_shutdown_chip(dws);
983 
984 	free_irq(dws->irq, dws->master);
985 }
986 EXPORT_SYMBOL_NS_GPL(dw_spi_remove_host, SPI_DW_CORE);
987 
988 int dw_spi_suspend_host(struct dw_spi *dws)
989 {
990 	int ret;
991 
992 	ret = spi_controller_suspend(dws->master);
993 	if (ret)
994 		return ret;
995 
996 	dw_spi_shutdown_chip(dws);
997 	return 0;
998 }
999 EXPORT_SYMBOL_NS_GPL(dw_spi_suspend_host, SPI_DW_CORE);
1000 
1001 int dw_spi_resume_host(struct dw_spi *dws)
1002 {
1003 	dw_spi_hw_init(&dws->master->dev, dws);
1004 	return spi_controller_resume(dws->master);
1005 }
1006 EXPORT_SYMBOL_NS_GPL(dw_spi_resume_host, SPI_DW_CORE);
1007 
1008 MODULE_AUTHOR("Feng Tang <feng.tang@intel.com>");
1009 MODULE_DESCRIPTION("Driver for DesignWare SPI controller core");
1010 MODULE_LICENSE("GPL v2");
1011