xref: /openbmc/linux/drivers/spi/spi-dw-core.c (revision e2942062)
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 void 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 
67 	dws->regset.regs = dw_spi_dbgfs_regs;
68 	dws->regset.nregs = ARRAY_SIZE(dw_spi_dbgfs_regs);
69 	dws->regset.base = dws->regs;
70 	debugfs_create_regset32("registers", 0400, dws->debugfs, &dws->regset);
71 }
72 
73 static void dw_spi_debugfs_remove(struct dw_spi *dws)
74 {
75 	debugfs_remove_recursive(dws->debugfs);
76 }
77 
78 #else
79 static inline void dw_spi_debugfs_init(struct dw_spi *dws)
80 {
81 }
82 
83 static inline void dw_spi_debugfs_remove(struct dw_spi *dws)
84 {
85 }
86 #endif /* CONFIG_DEBUG_FS */
87 
88 void dw_spi_set_cs(struct spi_device *spi, bool enable)
89 {
90 	struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
91 	bool cs_high = !!(spi->mode & SPI_CS_HIGH);
92 
93 	/*
94 	 * DW SPI controller demands any native CS being set in order to
95 	 * proceed with data transfer. So in order to activate the SPI
96 	 * communications we must set a corresponding bit in the Slave
97 	 * Enable register no matter whether the SPI core is configured to
98 	 * support active-high or active-low CS level.
99 	 */
100 	if (cs_high == enable)
101 		dw_writel(dws, DW_SPI_SER, BIT(spi_get_chipselect(spi, 0)));
102 	else
103 		dw_writel(dws, DW_SPI_SER, 0);
104 }
105 EXPORT_SYMBOL_NS_GPL(dw_spi_set_cs, SPI_DW_CORE);
106 
107 /* Return the max entries we can fill into tx fifo */
108 static inline u32 dw_spi_tx_max(struct dw_spi *dws)
109 {
110 	u32 tx_room, rxtx_gap;
111 
112 	tx_room = dws->fifo_len - dw_readl(dws, DW_SPI_TXFLR);
113 
114 	/*
115 	 * Another concern is about the tx/rx mismatch, we
116 	 * though to use (dws->fifo_len - rxflr - txflr) as
117 	 * one maximum value for tx, but it doesn't cover the
118 	 * data which is out of tx/rx fifo and inside the
119 	 * shift registers. So a control from sw point of
120 	 * view is taken.
121 	 */
122 	rxtx_gap = dws->fifo_len - (dws->rx_len - dws->tx_len);
123 
124 	return min3((u32)dws->tx_len, tx_room, rxtx_gap);
125 }
126 
127 /* Return the max entries we should read out of rx fifo */
128 static inline u32 dw_spi_rx_max(struct dw_spi *dws)
129 {
130 	return min_t(u32, dws->rx_len, dw_readl(dws, DW_SPI_RXFLR));
131 }
132 
133 static void dw_writer(struct dw_spi *dws)
134 {
135 	u32 max = dw_spi_tx_max(dws);
136 	u32 txw = 0;
137 
138 	while (max--) {
139 		if (dws->tx) {
140 			if (dws->n_bytes == 1)
141 				txw = *(u8 *)(dws->tx);
142 			else if (dws->n_bytes == 2)
143 				txw = *(u16 *)(dws->tx);
144 			else
145 				txw = *(u32 *)(dws->tx);
146 
147 			dws->tx += dws->n_bytes;
148 		}
149 		dw_write_io_reg(dws, DW_SPI_DR, txw);
150 		--dws->tx_len;
151 	}
152 }
153 
154 static void dw_reader(struct dw_spi *dws)
155 {
156 	u32 max = dw_spi_rx_max(dws);
157 	u32 rxw;
158 
159 	while (max--) {
160 		rxw = dw_read_io_reg(dws, DW_SPI_DR);
161 		if (dws->rx) {
162 			if (dws->n_bytes == 1)
163 				*(u8 *)(dws->rx) = rxw;
164 			else if (dws->n_bytes == 2)
165 				*(u16 *)(dws->rx) = rxw;
166 			else
167 				*(u32 *)(dws->rx) = rxw;
168 
169 			dws->rx += dws->n_bytes;
170 		}
171 		--dws->rx_len;
172 	}
173 }
174 
175 int dw_spi_check_status(struct dw_spi *dws, bool raw)
176 {
177 	u32 irq_status;
178 	int ret = 0;
179 
180 	if (raw)
181 		irq_status = dw_readl(dws, DW_SPI_RISR);
182 	else
183 		irq_status = dw_readl(dws, DW_SPI_ISR);
184 
185 	if (irq_status & DW_SPI_INT_RXOI) {
186 		dev_err(&dws->master->dev, "RX FIFO overflow detected\n");
187 		ret = -EIO;
188 	}
189 
190 	if (irq_status & DW_SPI_INT_RXUI) {
191 		dev_err(&dws->master->dev, "RX FIFO underflow detected\n");
192 		ret = -EIO;
193 	}
194 
195 	if (irq_status & DW_SPI_INT_TXOI) {
196 		dev_err(&dws->master->dev, "TX FIFO overflow detected\n");
197 		ret = -EIO;
198 	}
199 
200 	/* Generically handle the erroneous situation */
201 	if (ret) {
202 		dw_spi_reset_chip(dws);
203 		if (dws->master->cur_msg)
204 			dws->master->cur_msg->status = ret;
205 	}
206 
207 	return ret;
208 }
209 EXPORT_SYMBOL_NS_GPL(dw_spi_check_status, SPI_DW_CORE);
210 
211 static irqreturn_t dw_spi_transfer_handler(struct dw_spi *dws)
212 {
213 	u16 irq_status = dw_readl(dws, DW_SPI_ISR);
214 
215 	if (dw_spi_check_status(dws, false)) {
216 		spi_finalize_current_transfer(dws->master);
217 		return IRQ_HANDLED;
218 	}
219 
220 	/*
221 	 * Read data from the Rx FIFO every time we've got a chance executing
222 	 * this method. If there is nothing left to receive, terminate the
223 	 * procedure. Otherwise adjust the Rx FIFO Threshold level if it's a
224 	 * final stage of the transfer. By doing so we'll get the next IRQ
225 	 * right when the leftover incoming data is received.
226 	 */
227 	dw_reader(dws);
228 	if (!dws->rx_len) {
229 		dw_spi_mask_intr(dws, 0xff);
230 		spi_finalize_current_transfer(dws->master);
231 	} else if (dws->rx_len <= dw_readl(dws, DW_SPI_RXFTLR)) {
232 		dw_writel(dws, DW_SPI_RXFTLR, dws->rx_len - 1);
233 	}
234 
235 	/*
236 	 * Send data out if Tx FIFO Empty IRQ is received. The IRQ will be
237 	 * disabled after the data transmission is finished so not to
238 	 * have the TXE IRQ flood at the final stage of the transfer.
239 	 */
240 	if (irq_status & DW_SPI_INT_TXEI) {
241 		dw_writer(dws);
242 		if (!dws->tx_len)
243 			dw_spi_mask_intr(dws, DW_SPI_INT_TXEI);
244 	}
245 
246 	return IRQ_HANDLED;
247 }
248 
249 static irqreturn_t dw_spi_irq(int irq, void *dev_id)
250 {
251 	struct spi_controller *master = dev_id;
252 	struct dw_spi *dws = spi_controller_get_devdata(master);
253 	u16 irq_status = dw_readl(dws, DW_SPI_ISR) & DW_SPI_INT_MASK;
254 
255 	if (!irq_status)
256 		return IRQ_NONE;
257 
258 	if (!master->cur_msg) {
259 		dw_spi_mask_intr(dws, 0xff);
260 		return IRQ_HANDLED;
261 	}
262 
263 	return dws->transfer_handler(dws);
264 }
265 
266 static u32 dw_spi_prepare_cr0(struct dw_spi *dws, struct spi_device *spi)
267 {
268 	u32 cr0 = 0;
269 
270 	if (dw_spi_ip_is(dws, PSSI)) {
271 		/* CTRLR0[ 5: 4] Frame Format */
272 		cr0 |= FIELD_PREP(DW_PSSI_CTRLR0_FRF_MASK, DW_SPI_CTRLR0_FRF_MOTO_SPI);
273 
274 		/*
275 		 * SPI mode (SCPOL|SCPH)
276 		 * CTRLR0[ 6] Serial Clock Phase
277 		 * CTRLR0[ 7] Serial Clock Polarity
278 		 */
279 		if (spi->mode & SPI_CPOL)
280 			cr0 |= DW_PSSI_CTRLR0_SCPOL;
281 		if (spi->mode & SPI_CPHA)
282 			cr0 |= DW_PSSI_CTRLR0_SCPHA;
283 
284 		/* CTRLR0[11] Shift Register Loop */
285 		if (spi->mode & SPI_LOOP)
286 			cr0 |= DW_PSSI_CTRLR0_SRL;
287 	} else {
288 		/* CTRLR0[ 7: 6] Frame Format */
289 		cr0 |= FIELD_PREP(DW_HSSI_CTRLR0_FRF_MASK, DW_SPI_CTRLR0_FRF_MOTO_SPI);
290 
291 		/*
292 		 * SPI mode (SCPOL|SCPH)
293 		 * CTRLR0[ 8] Serial Clock Phase
294 		 * CTRLR0[ 9] Serial Clock Polarity
295 		 */
296 		if (spi->mode & SPI_CPOL)
297 			cr0 |= DW_HSSI_CTRLR0_SCPOL;
298 		if (spi->mode & SPI_CPHA)
299 			cr0 |= DW_HSSI_CTRLR0_SCPHA;
300 
301 		/* CTRLR0[13] Shift Register Loop */
302 		if (spi->mode & SPI_LOOP)
303 			cr0 |= DW_HSSI_CTRLR0_SRL;
304 
305 		/* CTRLR0[31] MST */
306 		if (dw_spi_ver_is_ge(dws, HSSI, 102A))
307 			cr0 |= DW_HSSI_CTRLR0_MST;
308 	}
309 
310 	return cr0;
311 }
312 
313 void dw_spi_update_config(struct dw_spi *dws, struct spi_device *spi,
314 			  struct dw_spi_cfg *cfg)
315 {
316 	struct dw_spi_chip_data *chip = spi_get_ctldata(spi);
317 	u32 cr0 = chip->cr0;
318 	u32 speed_hz;
319 	u16 clk_div;
320 
321 	/* CTRLR0[ 4/3: 0] or CTRLR0[ 20: 16] Data Frame Size */
322 	cr0 |= (cfg->dfs - 1) << dws->dfs_offset;
323 
324 	if (dw_spi_ip_is(dws, PSSI))
325 		/* CTRLR0[ 9:8] Transfer Mode */
326 		cr0 |= FIELD_PREP(DW_PSSI_CTRLR0_TMOD_MASK, cfg->tmode);
327 	else
328 		/* CTRLR0[11:10] Transfer Mode */
329 		cr0 |= FIELD_PREP(DW_HSSI_CTRLR0_TMOD_MASK, cfg->tmode);
330 
331 	dw_writel(dws, DW_SPI_CTRLR0, cr0);
332 
333 	if (cfg->tmode == DW_SPI_CTRLR0_TMOD_EPROMREAD ||
334 	    cfg->tmode == DW_SPI_CTRLR0_TMOD_RO)
335 		dw_writel(dws, DW_SPI_CTRLR1, cfg->ndf ? cfg->ndf - 1 : 0);
336 
337 	/* Note DW APB SSI clock divider doesn't support odd numbers */
338 	clk_div = (DIV_ROUND_UP(dws->max_freq, cfg->freq) + 1) & 0xfffe;
339 	speed_hz = dws->max_freq / clk_div;
340 
341 	if (dws->current_freq != speed_hz) {
342 		dw_spi_set_clk(dws, clk_div);
343 		dws->current_freq = speed_hz;
344 	}
345 
346 	/* Update RX sample delay if required */
347 	if (dws->cur_rx_sample_dly != chip->rx_sample_dly) {
348 		dw_writel(dws, DW_SPI_RX_SAMPLE_DLY, chip->rx_sample_dly);
349 		dws->cur_rx_sample_dly = chip->rx_sample_dly;
350 	}
351 }
352 EXPORT_SYMBOL_NS_GPL(dw_spi_update_config, SPI_DW_CORE);
353 
354 static void dw_spi_irq_setup(struct dw_spi *dws)
355 {
356 	u16 level;
357 	u8 imask;
358 
359 	/*
360 	 * Originally Tx and Rx data lengths match. Rx FIFO Threshold level
361 	 * will be adjusted at the final stage of the IRQ-based SPI transfer
362 	 * execution so not to lose the leftover of the incoming data.
363 	 */
364 	level = min_t(unsigned int, dws->fifo_len / 2, dws->tx_len);
365 	dw_writel(dws, DW_SPI_TXFTLR, level);
366 	dw_writel(dws, DW_SPI_RXFTLR, level - 1);
367 
368 	dws->transfer_handler = dw_spi_transfer_handler;
369 
370 	imask = DW_SPI_INT_TXEI | DW_SPI_INT_TXOI |
371 		DW_SPI_INT_RXUI | DW_SPI_INT_RXOI | DW_SPI_INT_RXFI;
372 	dw_spi_umask_intr(dws, imask);
373 }
374 
375 /*
376  * The iterative procedure of the poll-based transfer is simple: write as much
377  * as possible to the Tx FIFO, wait until the pending to receive data is ready
378  * to be read, read it from the Rx FIFO and check whether the performed
379  * procedure has been successful.
380  *
381  * Note this method the same way as the IRQ-based transfer won't work well for
382  * the SPI devices connected to the controller with native CS due to the
383  * automatic CS assertion/de-assertion.
384  */
385 static int dw_spi_poll_transfer(struct dw_spi *dws,
386 				struct spi_transfer *transfer)
387 {
388 	struct spi_delay delay;
389 	u16 nbits;
390 	int ret;
391 
392 	delay.unit = SPI_DELAY_UNIT_SCK;
393 	nbits = dws->n_bytes * BITS_PER_BYTE;
394 
395 	do {
396 		dw_writer(dws);
397 
398 		delay.value = nbits * (dws->rx_len - dws->tx_len);
399 		spi_delay_exec(&delay, transfer);
400 
401 		dw_reader(dws);
402 
403 		ret = dw_spi_check_status(dws, true);
404 		if (ret)
405 			return ret;
406 	} while (dws->rx_len);
407 
408 	return 0;
409 }
410 
411 static int dw_spi_transfer_one(struct spi_controller *master,
412 			       struct spi_device *spi,
413 			       struct spi_transfer *transfer)
414 {
415 	struct dw_spi *dws = spi_controller_get_devdata(master);
416 	struct dw_spi_cfg cfg = {
417 		.tmode = DW_SPI_CTRLR0_TMOD_TR,
418 		.dfs = transfer->bits_per_word,
419 		.freq = transfer->speed_hz,
420 	};
421 	int ret;
422 
423 	dws->dma_mapped = 0;
424 	dws->n_bytes =
425 		roundup_pow_of_two(DIV_ROUND_UP(transfer->bits_per_word,
426 						BITS_PER_BYTE));
427 
428 	dws->tx = (void *)transfer->tx_buf;
429 	dws->tx_len = transfer->len / dws->n_bytes;
430 	dws->rx = transfer->rx_buf;
431 	dws->rx_len = dws->tx_len;
432 
433 	/* Ensure the data above is visible for all CPUs */
434 	smp_mb();
435 
436 	dw_spi_enable_chip(dws, 0);
437 
438 	dw_spi_update_config(dws, spi, &cfg);
439 
440 	transfer->effective_speed_hz = dws->current_freq;
441 
442 	/* Check if current transfer is a DMA transaction */
443 	if (master->can_dma && master->can_dma(master, spi, transfer))
444 		dws->dma_mapped = master->cur_msg_mapped;
445 
446 	/* For poll mode just disable all interrupts */
447 	dw_spi_mask_intr(dws, 0xff);
448 
449 	if (dws->dma_mapped) {
450 		ret = dws->dma_ops->dma_setup(dws, transfer);
451 		if (ret)
452 			return ret;
453 	}
454 
455 	dw_spi_enable_chip(dws, 1);
456 
457 	if (dws->dma_mapped)
458 		return dws->dma_ops->dma_transfer(dws, transfer);
459 	else if (dws->irq == IRQ_NOTCONNECTED)
460 		return dw_spi_poll_transfer(dws, transfer);
461 
462 	dw_spi_irq_setup(dws);
463 
464 	return 1;
465 }
466 
467 static void dw_spi_handle_err(struct spi_controller *master,
468 			      struct spi_message *msg)
469 {
470 	struct dw_spi *dws = spi_controller_get_devdata(master);
471 
472 	if (dws->dma_mapped)
473 		dws->dma_ops->dma_stop(dws);
474 
475 	dw_spi_reset_chip(dws);
476 }
477 
478 static int dw_spi_adjust_mem_op_size(struct spi_mem *mem, struct spi_mem_op *op)
479 {
480 	if (op->data.dir == SPI_MEM_DATA_IN)
481 		op->data.nbytes = clamp_val(op->data.nbytes, 0, DW_SPI_NDF_MASK + 1);
482 
483 	return 0;
484 }
485 
486 static bool dw_spi_supports_mem_op(struct spi_mem *mem,
487 				   const struct spi_mem_op *op)
488 {
489 	if (op->data.buswidth > 1 || op->addr.buswidth > 1 ||
490 	    op->dummy.buswidth > 1 || op->cmd.buswidth > 1)
491 		return false;
492 
493 	return spi_mem_default_supports_op(mem, op);
494 }
495 
496 static int dw_spi_init_mem_buf(struct dw_spi *dws, const struct spi_mem_op *op)
497 {
498 	unsigned int i, j, len;
499 	u8 *out;
500 
501 	/*
502 	 * Calculate the total length of the EEPROM command transfer and
503 	 * either use the pre-allocated buffer or create a temporary one.
504 	 */
505 	len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
506 	if (op->data.dir == SPI_MEM_DATA_OUT)
507 		len += op->data.nbytes;
508 
509 	if (len <= DW_SPI_BUF_SIZE) {
510 		out = dws->buf;
511 	} else {
512 		out = kzalloc(len, GFP_KERNEL);
513 		if (!out)
514 			return -ENOMEM;
515 	}
516 
517 	/*
518 	 * Collect the operation code, address and dummy bytes into the single
519 	 * buffer. If it's a transfer with data to be sent, also copy it into the
520 	 * single buffer in order to speed the data transmission up.
521 	 */
522 	for (i = 0; i < op->cmd.nbytes; ++i)
523 		out[i] = DW_SPI_GET_BYTE(op->cmd.opcode, op->cmd.nbytes - i - 1);
524 	for (j = 0; j < op->addr.nbytes; ++i, ++j)
525 		out[i] = DW_SPI_GET_BYTE(op->addr.val, op->addr.nbytes - j - 1);
526 	for (j = 0; j < op->dummy.nbytes; ++i, ++j)
527 		out[i] = 0x0;
528 
529 	if (op->data.dir == SPI_MEM_DATA_OUT)
530 		memcpy(&out[i], op->data.buf.out, op->data.nbytes);
531 
532 	dws->n_bytes = 1;
533 	dws->tx = out;
534 	dws->tx_len = len;
535 	if (op->data.dir == SPI_MEM_DATA_IN) {
536 		dws->rx = op->data.buf.in;
537 		dws->rx_len = op->data.nbytes;
538 	} else {
539 		dws->rx = NULL;
540 		dws->rx_len = 0;
541 	}
542 
543 	return 0;
544 }
545 
546 static void dw_spi_free_mem_buf(struct dw_spi *dws)
547 {
548 	if (dws->tx != dws->buf)
549 		kfree(dws->tx);
550 }
551 
552 static int dw_spi_write_then_read(struct dw_spi *dws, struct spi_device *spi)
553 {
554 	u32 room, entries, sts;
555 	unsigned int len;
556 	u8 *buf;
557 
558 	/*
559 	 * At initial stage we just pre-fill the Tx FIFO in with no rush,
560 	 * since native CS hasn't been enabled yet and the automatic data
561 	 * transmission won't start til we do that.
562 	 */
563 	len = min(dws->fifo_len, dws->tx_len);
564 	buf = dws->tx;
565 	while (len--)
566 		dw_write_io_reg(dws, DW_SPI_DR, *buf++);
567 
568 	/*
569 	 * After setting any bit in the SER register the transmission will
570 	 * start automatically. We have to keep up with that procedure
571 	 * otherwise the CS de-assertion will happen whereupon the memory
572 	 * operation will be pre-terminated.
573 	 */
574 	len = dws->tx_len - ((void *)buf - dws->tx);
575 	dw_spi_set_cs(spi, false);
576 	while (len) {
577 		entries = readl_relaxed(dws->regs + DW_SPI_TXFLR);
578 		if (!entries) {
579 			dev_err(&dws->master->dev, "CS de-assertion on Tx\n");
580 			return -EIO;
581 		}
582 		room = min(dws->fifo_len - entries, len);
583 		for (; room; --room, --len)
584 			dw_write_io_reg(dws, DW_SPI_DR, *buf++);
585 	}
586 
587 	/*
588 	 * Data fetching will start automatically if the EEPROM-read mode is
589 	 * activated. We have to keep up with the incoming data pace to
590 	 * prevent the Rx FIFO overflow causing the inbound data loss.
591 	 */
592 	len = dws->rx_len;
593 	buf = dws->rx;
594 	while (len) {
595 		entries = readl_relaxed(dws->regs + DW_SPI_RXFLR);
596 		if (!entries) {
597 			sts = readl_relaxed(dws->regs + DW_SPI_RISR);
598 			if (sts & DW_SPI_INT_RXOI) {
599 				dev_err(&dws->master->dev, "FIFO overflow on Rx\n");
600 				return -EIO;
601 			}
602 			continue;
603 		}
604 		entries = min(entries, len);
605 		for (; entries; --entries, --len)
606 			*buf++ = dw_read_io_reg(dws, DW_SPI_DR);
607 	}
608 
609 	return 0;
610 }
611 
612 static inline bool dw_spi_ctlr_busy(struct dw_spi *dws)
613 {
614 	return dw_readl(dws, DW_SPI_SR) & DW_SPI_SR_BUSY;
615 }
616 
617 static int dw_spi_wait_mem_op_done(struct dw_spi *dws)
618 {
619 	int retry = DW_SPI_WAIT_RETRIES;
620 	struct spi_delay delay;
621 	unsigned long ns, us;
622 	u32 nents;
623 
624 	nents = dw_readl(dws, DW_SPI_TXFLR);
625 	ns = NSEC_PER_SEC / dws->current_freq * nents;
626 	ns *= dws->n_bytes * BITS_PER_BYTE;
627 	if (ns <= NSEC_PER_USEC) {
628 		delay.unit = SPI_DELAY_UNIT_NSECS;
629 		delay.value = ns;
630 	} else {
631 		us = DIV_ROUND_UP(ns, NSEC_PER_USEC);
632 		delay.unit = SPI_DELAY_UNIT_USECS;
633 		delay.value = clamp_val(us, 0, USHRT_MAX);
634 	}
635 
636 	while (dw_spi_ctlr_busy(dws) && retry--)
637 		spi_delay_exec(&delay, NULL);
638 
639 	if (retry < 0) {
640 		dev_err(&dws->master->dev, "Mem op hanged up\n");
641 		return -EIO;
642 	}
643 
644 	return 0;
645 }
646 
647 static void dw_spi_stop_mem_op(struct dw_spi *dws, struct spi_device *spi)
648 {
649 	dw_spi_enable_chip(dws, 0);
650 	dw_spi_set_cs(spi, true);
651 	dw_spi_enable_chip(dws, 1);
652 }
653 
654 /*
655  * The SPI memory operation implementation below is the best choice for the
656  * devices, which are selected by the native chip-select lane. It's
657  * specifically developed to workaround the problem with automatic chip-select
658  * lane toggle when there is no data in the Tx FIFO buffer. Luckily the current
659  * SPI-mem core calls exec_op() callback only if the GPIO-based CS is
660  * unavailable.
661  */
662 static int dw_spi_exec_mem_op(struct spi_mem *mem, const struct spi_mem_op *op)
663 {
664 	struct dw_spi *dws = spi_controller_get_devdata(mem->spi->controller);
665 	struct dw_spi_cfg cfg;
666 	unsigned long flags;
667 	int ret;
668 
669 	/*
670 	 * Collect the outbound data into a single buffer to speed the
671 	 * transmission up at least on the initial stage.
672 	 */
673 	ret = dw_spi_init_mem_buf(dws, op);
674 	if (ret)
675 		return ret;
676 
677 	/*
678 	 * DW SPI EEPROM-read mode is required only for the SPI memory Data-IN
679 	 * operation. Transmit-only mode is suitable for the rest of them.
680 	 */
681 	cfg.dfs = 8;
682 	cfg.freq = clamp(mem->spi->max_speed_hz, 0U, dws->max_mem_freq);
683 	if (op->data.dir == SPI_MEM_DATA_IN) {
684 		cfg.tmode = DW_SPI_CTRLR0_TMOD_EPROMREAD;
685 		cfg.ndf = op->data.nbytes;
686 	} else {
687 		cfg.tmode = DW_SPI_CTRLR0_TMOD_TO;
688 	}
689 
690 	dw_spi_enable_chip(dws, 0);
691 
692 	dw_spi_update_config(dws, mem->spi, &cfg);
693 
694 	dw_spi_mask_intr(dws, 0xff);
695 
696 	dw_spi_enable_chip(dws, 1);
697 
698 	/*
699 	 * DW APB SSI controller has very nasty peculiarities. First originally
700 	 * (without any vendor-specific modifications) it doesn't provide a
701 	 * direct way to set and clear the native chip-select signal. Instead
702 	 * the controller asserts the CS lane if Tx FIFO isn't empty and a
703 	 * transmission is going on, and automatically de-asserts it back to
704 	 * the high level if the Tx FIFO doesn't have anything to be pushed
705 	 * out. Due to that a multi-tasking or heavy IRQs activity might be
706 	 * fatal, since the transfer procedure preemption may cause the Tx FIFO
707 	 * getting empty and sudden CS de-assertion, which in the middle of the
708 	 * transfer will most likely cause the data loss. Secondly the
709 	 * EEPROM-read or Read-only DW SPI transfer modes imply the incoming
710 	 * data being automatically pulled in into the Rx FIFO. So if the
711 	 * driver software is late in fetching the data from the FIFO before
712 	 * it's overflown, new incoming data will be lost. In order to make
713 	 * sure the executed memory operations are CS-atomic and to prevent the
714 	 * Rx FIFO overflow we have to disable the local interrupts so to block
715 	 * any preemption during the subsequent IO operations.
716 	 *
717 	 * Note. At some circumstances disabling IRQs may not help to prevent
718 	 * the problems described above. The CS de-assertion and Rx FIFO
719 	 * overflow may still happen due to the relatively slow system bus or
720 	 * CPU not working fast enough, so the write-then-read algo implemented
721 	 * here just won't keep up with the SPI bus data transfer. Such
722 	 * situation is highly platform specific and is supposed to be fixed by
723 	 * manually restricting the SPI bus frequency using the
724 	 * dws->max_mem_freq parameter.
725 	 */
726 	local_irq_save(flags);
727 	preempt_disable();
728 
729 	ret = dw_spi_write_then_read(dws, mem->spi);
730 
731 	local_irq_restore(flags);
732 	preempt_enable();
733 
734 	/*
735 	 * Wait for the operation being finished and check the controller
736 	 * status only if there hasn't been any run-time error detected. In the
737 	 * former case it's just pointless. In the later one to prevent an
738 	 * additional error message printing since any hw error flag being set
739 	 * would be due to an error detected on the data transfer.
740 	 */
741 	if (!ret) {
742 		ret = dw_spi_wait_mem_op_done(dws);
743 		if (!ret)
744 			ret = dw_spi_check_status(dws, true);
745 	}
746 
747 	dw_spi_stop_mem_op(dws, mem->spi);
748 
749 	dw_spi_free_mem_buf(dws);
750 
751 	return ret;
752 }
753 
754 /*
755  * Initialize the default memory operations if a glue layer hasn't specified
756  * custom ones. Direct mapping operations will be preserved anyway since DW SPI
757  * controller doesn't have an embedded dirmap interface. Note the memory
758  * operations implemented in this driver is the best choice only for the DW APB
759  * SSI controller with standard native CS functionality. If a hardware vendor
760  * has fixed the automatic CS assertion/de-assertion peculiarity, then it will
761  * be safer to use the normal SPI-messages-based transfers implementation.
762  */
763 static void dw_spi_init_mem_ops(struct dw_spi *dws)
764 {
765 	if (!dws->mem_ops.exec_op && !(dws->caps & DW_SPI_CAP_CS_OVERRIDE) &&
766 	    !dws->set_cs) {
767 		dws->mem_ops.adjust_op_size = dw_spi_adjust_mem_op_size;
768 		dws->mem_ops.supports_op = dw_spi_supports_mem_op;
769 		dws->mem_ops.exec_op = dw_spi_exec_mem_op;
770 		if (!dws->max_mem_freq)
771 			dws->max_mem_freq = dws->max_freq;
772 	}
773 }
774 
775 /* This may be called twice for each spi dev */
776 static int dw_spi_setup(struct spi_device *spi)
777 {
778 	struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
779 	struct dw_spi_chip_data *chip;
780 
781 	/* Only alloc on first setup */
782 	chip = spi_get_ctldata(spi);
783 	if (!chip) {
784 		struct dw_spi *dws = spi_controller_get_devdata(spi->controller);
785 		u32 rx_sample_dly_ns;
786 
787 		chip = kzalloc(sizeof(*chip), GFP_KERNEL);
788 		if (!chip)
789 			return -ENOMEM;
790 		spi_set_ctldata(spi, chip);
791 		/* Get specific / default rx-sample-delay */
792 		if (device_property_read_u32(&spi->dev,
793 					     "rx-sample-delay-ns",
794 					     &rx_sample_dly_ns) != 0)
795 			/* Use default controller value */
796 			rx_sample_dly_ns = dws->def_rx_sample_dly_ns;
797 		chip->rx_sample_dly = DIV_ROUND_CLOSEST(rx_sample_dly_ns,
798 							NSEC_PER_SEC /
799 							dws->max_freq);
800 	}
801 
802 	/*
803 	 * Update CR0 data each time the setup callback is invoked since
804 	 * the device parameters could have been changed, for instance, by
805 	 * the MMC SPI driver or something else.
806 	 */
807 	chip->cr0 = dw_spi_prepare_cr0(dws, spi);
808 
809 	return 0;
810 }
811 
812 static void dw_spi_cleanup(struct spi_device *spi)
813 {
814 	struct dw_spi_chip_data *chip = spi_get_ctldata(spi);
815 
816 	kfree(chip);
817 	spi_set_ctldata(spi, NULL);
818 }
819 
820 /* Restart the controller, disable all interrupts, clean rx fifo */
821 static void dw_spi_hw_init(struct device *dev, struct dw_spi *dws)
822 {
823 	dw_spi_reset_chip(dws);
824 
825 	/*
826 	 * Retrieve the Synopsys component version if it hasn't been specified
827 	 * by the platform. CoreKit version ID is encoded as a 3-chars ASCII
828 	 * code enclosed with '*' (typical for the most of Synopsys IP-cores).
829 	 */
830 	if (!dws->ver) {
831 		dws->ver = dw_readl(dws, DW_SPI_VERSION);
832 
833 		dev_dbg(dev, "Synopsys DWC%sSSI v%c.%c%c\n",
834 			dw_spi_ip_is(dws, PSSI) ? " APB " : " ",
835 			DW_SPI_GET_BYTE(dws->ver, 3), DW_SPI_GET_BYTE(dws->ver, 2),
836 			DW_SPI_GET_BYTE(dws->ver, 1));
837 	}
838 
839 	/*
840 	 * Try to detect the FIFO depth if not set by interface driver,
841 	 * the depth could be from 2 to 256 from HW spec
842 	 */
843 	if (!dws->fifo_len) {
844 		u32 fifo;
845 
846 		for (fifo = 1; fifo < 256; fifo++) {
847 			dw_writel(dws, DW_SPI_TXFTLR, fifo);
848 			if (fifo != dw_readl(dws, DW_SPI_TXFTLR))
849 				break;
850 		}
851 		dw_writel(dws, DW_SPI_TXFTLR, 0);
852 
853 		dws->fifo_len = (fifo == 1) ? 0 : fifo;
854 		dev_dbg(dev, "Detected FIFO size: %u bytes\n", dws->fifo_len);
855 	}
856 
857 	/*
858 	 * Detect CTRLR0.DFS field size and offset by testing the lowest bits
859 	 * writability. Note DWC SSI controller also has the extended DFS, but
860 	 * with zero offset.
861 	 */
862 	if (dw_spi_ip_is(dws, PSSI)) {
863 		u32 cr0, tmp = dw_readl(dws, DW_SPI_CTRLR0);
864 
865 		dw_spi_enable_chip(dws, 0);
866 		dw_writel(dws, DW_SPI_CTRLR0, 0xffffffff);
867 		cr0 = dw_readl(dws, DW_SPI_CTRLR0);
868 		dw_writel(dws, DW_SPI_CTRLR0, tmp);
869 		dw_spi_enable_chip(dws, 1);
870 
871 		if (!(cr0 & DW_PSSI_CTRLR0_DFS_MASK)) {
872 			dws->caps |= DW_SPI_CAP_DFS32;
873 			dws->dfs_offset = __bf_shf(DW_PSSI_CTRLR0_DFS32_MASK);
874 			dev_dbg(dev, "Detected 32-bits max data frame size\n");
875 		}
876 	} else {
877 		dws->caps |= DW_SPI_CAP_DFS32;
878 	}
879 
880 	/* enable HW fixup for explicit CS deselect for Amazon's alpine chip */
881 	if (dws->caps & DW_SPI_CAP_CS_OVERRIDE)
882 		dw_writel(dws, DW_SPI_CS_OVERRIDE, 0xF);
883 }
884 
885 int dw_spi_add_host(struct device *dev, struct dw_spi *dws)
886 {
887 	struct spi_controller *master;
888 	int ret;
889 
890 	if (!dws)
891 		return -EINVAL;
892 
893 	master = spi_alloc_master(dev, 0);
894 	if (!master)
895 		return -ENOMEM;
896 
897 	device_set_node(&master->dev, dev_fwnode(dev));
898 
899 	dws->master = master;
900 	dws->dma_addr = (dma_addr_t)(dws->paddr + DW_SPI_DR);
901 
902 	spi_controller_set_devdata(master, dws);
903 
904 	/* Basic HW init */
905 	dw_spi_hw_init(dev, dws);
906 
907 	ret = request_irq(dws->irq, dw_spi_irq, IRQF_SHARED, dev_name(dev),
908 			  master);
909 	if (ret < 0 && ret != -ENOTCONN) {
910 		dev_err(dev, "can not get IRQ\n");
911 		goto err_free_master;
912 	}
913 
914 	dw_spi_init_mem_ops(dws);
915 
916 	master->use_gpio_descriptors = true;
917 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_LOOP;
918 	if (dws->caps & DW_SPI_CAP_DFS32)
919 		master->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
920 	else
921 		master->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 16);
922 	master->bus_num = dws->bus_num;
923 	master->num_chipselect = dws->num_cs;
924 	master->setup = dw_spi_setup;
925 	master->cleanup = dw_spi_cleanup;
926 	if (dws->set_cs)
927 		master->set_cs = dws->set_cs;
928 	else
929 		master->set_cs = dw_spi_set_cs;
930 	master->transfer_one = dw_spi_transfer_one;
931 	master->handle_err = dw_spi_handle_err;
932 	if (dws->mem_ops.exec_op)
933 		master->mem_ops = &dws->mem_ops;
934 	master->max_speed_hz = dws->max_freq;
935 	master->flags = SPI_MASTER_GPIO_SS;
936 	master->auto_runtime_pm = true;
937 
938 	/* Get default rx sample delay */
939 	device_property_read_u32(dev, "rx-sample-delay-ns",
940 				 &dws->def_rx_sample_dly_ns);
941 
942 	if (dws->dma_ops && dws->dma_ops->dma_init) {
943 		ret = dws->dma_ops->dma_init(dev, dws);
944 		if (ret == -EPROBE_DEFER) {
945 			goto err_free_irq;
946 		} else if (ret) {
947 			dev_warn(dev, "DMA init failed\n");
948 		} else {
949 			master->can_dma = dws->dma_ops->can_dma;
950 			master->flags |= SPI_CONTROLLER_MUST_TX;
951 		}
952 	}
953 
954 	ret = spi_register_controller(master);
955 	if (ret) {
956 		dev_err_probe(dev, ret, "problem registering spi master\n");
957 		goto err_dma_exit;
958 	}
959 
960 	dw_spi_debugfs_init(dws);
961 	return 0;
962 
963 err_dma_exit:
964 	if (dws->dma_ops && dws->dma_ops->dma_exit)
965 		dws->dma_ops->dma_exit(dws);
966 	dw_spi_enable_chip(dws, 0);
967 err_free_irq:
968 	free_irq(dws->irq, master);
969 err_free_master:
970 	spi_controller_put(master);
971 	return ret;
972 }
973 EXPORT_SYMBOL_NS_GPL(dw_spi_add_host, SPI_DW_CORE);
974 
975 void dw_spi_remove_host(struct dw_spi *dws)
976 {
977 	dw_spi_debugfs_remove(dws);
978 
979 	spi_unregister_controller(dws->master);
980 
981 	if (dws->dma_ops && dws->dma_ops->dma_exit)
982 		dws->dma_ops->dma_exit(dws);
983 
984 	dw_spi_shutdown_chip(dws);
985 
986 	free_irq(dws->irq, dws->master);
987 }
988 EXPORT_SYMBOL_NS_GPL(dw_spi_remove_host, SPI_DW_CORE);
989 
990 int dw_spi_suspend_host(struct dw_spi *dws)
991 {
992 	int ret;
993 
994 	ret = spi_controller_suspend(dws->master);
995 	if (ret)
996 		return ret;
997 
998 	dw_spi_shutdown_chip(dws);
999 	return 0;
1000 }
1001 EXPORT_SYMBOL_NS_GPL(dw_spi_suspend_host, SPI_DW_CORE);
1002 
1003 int dw_spi_resume_host(struct dw_spi *dws)
1004 {
1005 	dw_spi_hw_init(&dws->master->dev, dws);
1006 	return spi_controller_resume(dws->master);
1007 }
1008 EXPORT_SYMBOL_NS_GPL(dw_spi_resume_host, SPI_DW_CORE);
1009 
1010 MODULE_AUTHOR("Feng Tang <feng.tang@intel.com>");
1011 MODULE_DESCRIPTION("Driver for DesignWare SPI controller core");
1012 MODULE_LICENSE("GPL v2");
1013