xref: /linux/drivers/spi/spi-atmel.c (revision 288440de9e5fdb4a3ff73864850f080c1250fc81)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Driver for Atmel AT32 and AT91 SPI Controllers
4  *
5  * Copyright (C) 2006 Atmel Corporation
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/clk.h>
10 #include <linux/module.h>
11 #include <linux/platform_device.h>
12 #include <linux/delay.h>
13 #include <linux/dma-mapping.h>
14 #include <linux/dmaengine.h>
15 #include <linux/err.h>
16 #include <linux/interrupt.h>
17 #include <linux/spi/spi.h>
18 #include <linux/slab.h>
19 #include <linux/of.h>
20 
21 #include <linux/io.h>
22 #include <linux/gpio/consumer.h>
23 #include <linux/pinctrl/consumer.h>
24 #include <linux/pm_runtime.h>
25 #include <trace/events/spi.h>
26 
27 /* SPI register offsets */
28 #define SPI_CR					0x0000
29 #define SPI_MR					0x0004
30 #define SPI_RDR					0x0008
31 #define SPI_TDR					0x000c
32 #define SPI_SR					0x0010
33 #define SPI_IER					0x0014
34 #define SPI_IDR					0x0018
35 #define SPI_IMR					0x001c
36 #define SPI_CSR0				0x0030
37 #define SPI_CSR1				0x0034
38 #define SPI_CSR2				0x0038
39 #define SPI_CSR3				0x003c
40 #define SPI_FMR					0x0040
41 #define SPI_FLR					0x0044
42 #define SPI_VERSION				0x00fc
43 #define SPI_RPR					0x0100
44 #define SPI_RCR					0x0104
45 #define SPI_TPR					0x0108
46 #define SPI_TCR					0x010c
47 #define SPI_RNPR				0x0110
48 #define SPI_RNCR				0x0114
49 #define SPI_TNPR				0x0118
50 #define SPI_TNCR				0x011c
51 #define SPI_PTCR				0x0120
52 #define SPI_PTSR				0x0124
53 
54 /* Bitfields in CR */
55 #define SPI_SPIEN_OFFSET			0
56 #define SPI_SPIEN_SIZE				1
57 #define SPI_SPIDIS_OFFSET			1
58 #define SPI_SPIDIS_SIZE				1
59 #define SPI_SWRST_OFFSET			7
60 #define SPI_SWRST_SIZE				1
61 #define SPI_LASTXFER_OFFSET			24
62 #define SPI_LASTXFER_SIZE			1
63 #define SPI_TXFCLR_OFFSET			16
64 #define SPI_TXFCLR_SIZE				1
65 #define SPI_RXFCLR_OFFSET			17
66 #define SPI_RXFCLR_SIZE				1
67 #define SPI_FIFOEN_OFFSET			30
68 #define SPI_FIFOEN_SIZE				1
69 #define SPI_FIFODIS_OFFSET			31
70 #define SPI_FIFODIS_SIZE			1
71 
72 /* Bitfields in MR */
73 #define SPI_MSTR_OFFSET				0
74 #define SPI_MSTR_SIZE				1
75 #define SPI_PS_OFFSET				1
76 #define SPI_PS_SIZE				1
77 #define SPI_PCSDEC_OFFSET			2
78 #define SPI_PCSDEC_SIZE				1
79 #define SPI_FDIV_OFFSET				3
80 #define SPI_FDIV_SIZE				1
81 #define SPI_MODFDIS_OFFSET			4
82 #define SPI_MODFDIS_SIZE			1
83 #define SPI_WDRBT_OFFSET			5
84 #define SPI_WDRBT_SIZE				1
85 #define SPI_LLB_OFFSET				7
86 #define SPI_LLB_SIZE				1
87 #define SPI_PCS_OFFSET				16
88 #define SPI_PCS_SIZE				4
89 #define SPI_DLYBCS_OFFSET			24
90 #define SPI_DLYBCS_SIZE				8
91 
92 /* Bitfields in RDR */
93 #define SPI_RD_OFFSET				0
94 #define SPI_RD_SIZE				16
95 
96 /* Bitfields in TDR */
97 #define SPI_TD_OFFSET				0
98 #define SPI_TD_SIZE				16
99 
100 /* Bitfields in SR */
101 #define SPI_RDRF_OFFSET				0
102 #define SPI_RDRF_SIZE				1
103 #define SPI_TDRE_OFFSET				1
104 #define SPI_TDRE_SIZE				1
105 #define SPI_MODF_OFFSET				2
106 #define SPI_MODF_SIZE				1
107 #define SPI_OVRES_OFFSET			3
108 #define SPI_OVRES_SIZE				1
109 #define SPI_ENDRX_OFFSET			4
110 #define SPI_ENDRX_SIZE				1
111 #define SPI_ENDTX_OFFSET			5
112 #define SPI_ENDTX_SIZE				1
113 #define SPI_RXBUFF_OFFSET			6
114 #define SPI_RXBUFF_SIZE				1
115 #define SPI_TXBUFE_OFFSET			7
116 #define SPI_TXBUFE_SIZE				1
117 #define SPI_NSSR_OFFSET				8
118 #define SPI_NSSR_SIZE				1
119 #define SPI_TXEMPTY_OFFSET			9
120 #define SPI_TXEMPTY_SIZE			1
121 #define SPI_SPIENS_OFFSET			16
122 #define SPI_SPIENS_SIZE				1
123 #define SPI_TXFEF_OFFSET			24
124 #define SPI_TXFEF_SIZE				1
125 #define SPI_TXFFF_OFFSET			25
126 #define SPI_TXFFF_SIZE				1
127 #define SPI_TXFTHF_OFFSET			26
128 #define SPI_TXFTHF_SIZE				1
129 #define SPI_RXFEF_OFFSET			27
130 #define SPI_RXFEF_SIZE				1
131 #define SPI_RXFFF_OFFSET			28
132 #define SPI_RXFFF_SIZE				1
133 #define SPI_RXFTHF_OFFSET			29
134 #define SPI_RXFTHF_SIZE				1
135 #define SPI_TXFPTEF_OFFSET			30
136 #define SPI_TXFPTEF_SIZE			1
137 #define SPI_RXFPTEF_OFFSET			31
138 #define SPI_RXFPTEF_SIZE			1
139 
140 /* Bitfields in CSR0 */
141 #define SPI_CPOL_OFFSET				0
142 #define SPI_CPOL_SIZE				1
143 #define SPI_NCPHA_OFFSET			1
144 #define SPI_NCPHA_SIZE				1
145 #define SPI_CSAAT_OFFSET			3
146 #define SPI_CSAAT_SIZE				1
147 #define SPI_BITS_OFFSET				4
148 #define SPI_BITS_SIZE				4
149 #define SPI_SCBR_OFFSET				8
150 #define SPI_SCBR_SIZE				8
151 #define SPI_DLYBS_OFFSET			16
152 #define SPI_DLYBS_SIZE				8
153 #define SPI_DLYBCT_OFFSET			24
154 #define SPI_DLYBCT_SIZE				8
155 
156 /* Bitfields in RCR */
157 #define SPI_RXCTR_OFFSET			0
158 #define SPI_RXCTR_SIZE				16
159 
160 /* Bitfields in TCR */
161 #define SPI_TXCTR_OFFSET			0
162 #define SPI_TXCTR_SIZE				16
163 
164 /* Bitfields in RNCR */
165 #define SPI_RXNCR_OFFSET			0
166 #define SPI_RXNCR_SIZE				16
167 
168 /* Bitfields in TNCR */
169 #define SPI_TXNCR_OFFSET			0
170 #define SPI_TXNCR_SIZE				16
171 
172 /* Bitfields in PTCR */
173 #define SPI_RXTEN_OFFSET			0
174 #define SPI_RXTEN_SIZE				1
175 #define SPI_RXTDIS_OFFSET			1
176 #define SPI_RXTDIS_SIZE				1
177 #define SPI_TXTEN_OFFSET			8
178 #define SPI_TXTEN_SIZE				1
179 #define SPI_TXTDIS_OFFSET			9
180 #define SPI_TXTDIS_SIZE				1
181 
182 /* Bitfields in FMR */
183 #define SPI_TXRDYM_OFFSET			0
184 #define SPI_TXRDYM_SIZE				2
185 #define SPI_RXRDYM_OFFSET			4
186 #define SPI_RXRDYM_SIZE				2
187 #define SPI_TXFTHRES_OFFSET			16
188 #define SPI_TXFTHRES_SIZE			6
189 #define SPI_RXFTHRES_OFFSET			24
190 #define SPI_RXFTHRES_SIZE			6
191 
192 /* Bitfields in FLR */
193 #define SPI_TXFL_OFFSET				0
194 #define SPI_TXFL_SIZE				6
195 #define SPI_RXFL_OFFSET				16
196 #define SPI_RXFL_SIZE				6
197 
198 /* Constants for BITS */
199 #define SPI_BITS_8_BPT				0
200 #define SPI_BITS_9_BPT				1
201 #define SPI_BITS_10_BPT				2
202 #define SPI_BITS_11_BPT				3
203 #define SPI_BITS_12_BPT				4
204 #define SPI_BITS_13_BPT				5
205 #define SPI_BITS_14_BPT				6
206 #define SPI_BITS_15_BPT				7
207 #define SPI_BITS_16_BPT				8
208 #define SPI_ONE_DATA				0
209 #define SPI_TWO_DATA				1
210 #define SPI_FOUR_DATA				2
211 
212 /* Bit manipulation macros */
213 #define SPI_BIT(name) \
214 	(1 << SPI_##name##_OFFSET)
215 #define SPI_BF(name, value) \
216 	(((value) & ((1 << SPI_##name##_SIZE) - 1)) << SPI_##name##_OFFSET)
217 #define SPI_BFEXT(name, value) \
218 	(((value) >> SPI_##name##_OFFSET) & ((1 << SPI_##name##_SIZE) - 1))
219 #define SPI_BFINS(name, value, old) \
220 	(((old) & ~(((1 << SPI_##name##_SIZE) - 1) << SPI_##name##_OFFSET)) \
221 	  | SPI_BF(name, value))
222 
223 /* Register access macros */
224 #define spi_readl(port, reg) \
225 	readl_relaxed((port)->regs + SPI_##reg)
226 #define spi_writel(port, reg, value) \
227 	writel_relaxed((value), (port)->regs + SPI_##reg)
228 #define spi_writew(port, reg, value) \
229 	writew_relaxed((value), (port)->regs + SPI_##reg)
230 
231 /* use PIO for small transfers, avoiding DMA setup/teardown overhead and
232  * cache operations; better heuristics consider wordsize and bitrate.
233  */
234 #define DMA_MIN_BYTES	16
235 
236 #define SPI_DMA_TIMEOUT		(msecs_to_jiffies(1000))
237 
238 #define AUTOSUSPEND_TIMEOUT	2000
239 
240 struct atmel_spi_caps {
241 	bool	is_spi2;
242 	bool	has_wdrbt;
243 	bool	has_dma_support;
244 	bool	has_pdc_support;
245 };
246 
247 /*
248  * The core SPI transfer engine just talks to a register bank to set up
249  * DMA transfers; transfer queue progress is driven by IRQs.  The clock
250  * framework provides the base clock, subdivided for each spi_device.
251  */
252 struct atmel_spi {
253 	spinlock_t		lock;
254 	unsigned long		flags;
255 
256 	phys_addr_t		phybase;
257 	void __iomem		*regs;
258 	int			irq;
259 	struct clk		*clk;
260 	struct platform_device	*pdev;
261 	unsigned long		spi_clk;
262 
263 	struct spi_transfer	*current_transfer;
264 	int			current_remaining_bytes;
265 	int			done_status;
266 	dma_addr_t		dma_addr_rx_bbuf;
267 	dma_addr_t		dma_addr_tx_bbuf;
268 	void			*addr_rx_bbuf;
269 	void			*addr_tx_bbuf;
270 
271 	struct completion	xfer_completion;
272 
273 	struct atmel_spi_caps	caps;
274 
275 	bool			use_dma;
276 	bool			use_pdc;
277 
278 	bool			keep_cs;
279 
280 	u32			fifo_size;
281 	u8			native_cs_free;
282 	u8			native_cs_for_gpio;
283 };
284 
285 /* Controller-specific per-slave state */
286 struct atmel_spi_device {
287 	u32			csr;
288 };
289 
290 #define SPI_MAX_DMA_XFER	65535 /* true for both PDC and DMA */
291 #define INVALID_DMA_ADDRESS	0xffffffff
292 
293 /*
294  * Version 2 of the SPI controller has
295  *  - CR.LASTXFER
296  *  - SPI_MR.DIV32 may become FDIV or must-be-zero (here: always zero)
297  *  - SPI_SR.TXEMPTY, SPI_SR.NSSR (and corresponding irqs)
298  *  - SPI_CSRx.CSAAT
299  *  - SPI_CSRx.SBCR allows faster clocking
300  */
301 static bool atmel_spi_is_v2(struct atmel_spi *as)
302 {
303 	return as->caps.is_spi2;
304 }
305 
306 /*
307  * Earlier SPI controllers (e.g. on at91rm9200) have a design bug whereby
308  * they assume that spi slave device state will not change on deselect, so
309  * that automagic deselection is OK.  ("NPCSx rises if no data is to be
310  * transmitted")  Not so!  Workaround uses nCSx pins as GPIOs; or newer
311  * controllers have CSAAT and friends.
312  *
313  * Even controller newer than ar91rm9200, using GPIOs can make sens as
314  * it lets us support active-high chipselects despite the controller's
315  * belief that only active-low devices/systems exists.
316  *
317  * However, at91rm9200 has a second erratum whereby nCS0 doesn't work
318  * right when driven with GPIO.  ("Mode Fault does not allow more than one
319  * Master on Chip Select 0.")  No workaround exists for that ... so for
320  * nCS0 on that chip, we (a) don't use the GPIO, (b) can't support CS_HIGH,
321  * and (c) will trigger that first erratum in some cases.
322  */
323 
324 static void cs_activate(struct atmel_spi *as, struct spi_device *spi)
325 {
326 	struct atmel_spi_device *asd = spi->controller_state;
327 	int chip_select;
328 	u32 mr;
329 
330 	if (spi->cs_gpiod)
331 		chip_select = as->native_cs_for_gpio;
332 	else
333 		chip_select = spi->chip_select;
334 
335 	if (atmel_spi_is_v2(as)) {
336 		spi_writel(as, CSR0 + 4 * chip_select, asd->csr);
337 		/* For the low SPI version, there is a issue that PDC transfer
338 		 * on CS1,2,3 needs SPI_CSR0.BITS config as SPI_CSR1,2,3.BITS
339 		 */
340 		spi_writel(as, CSR0, asd->csr);
341 		if (as->caps.has_wdrbt) {
342 			spi_writel(as, MR,
343 					SPI_BF(PCS, ~(0x01 << chip_select))
344 					| SPI_BIT(WDRBT)
345 					| SPI_BIT(MODFDIS)
346 					| SPI_BIT(MSTR));
347 		} else {
348 			spi_writel(as, MR,
349 					SPI_BF(PCS, ~(0x01 << chip_select))
350 					| SPI_BIT(MODFDIS)
351 					| SPI_BIT(MSTR));
352 		}
353 
354 		mr = spi_readl(as, MR);
355 	} else {
356 		u32 cpol = (spi->mode & SPI_CPOL) ? SPI_BIT(CPOL) : 0;
357 		int i;
358 		u32 csr;
359 
360 		/* Make sure clock polarity is correct */
361 		for (i = 0; i < spi->master->num_chipselect; i++) {
362 			csr = spi_readl(as, CSR0 + 4 * i);
363 			if ((csr ^ cpol) & SPI_BIT(CPOL))
364 				spi_writel(as, CSR0 + 4 * i,
365 						csr ^ SPI_BIT(CPOL));
366 		}
367 
368 		mr = spi_readl(as, MR);
369 		mr = SPI_BFINS(PCS, ~(1 << chip_select), mr);
370 		spi_writel(as, MR, mr);
371 	}
372 
373 	dev_dbg(&spi->dev, "activate NPCS, mr %08x\n", mr);
374 }
375 
376 static void cs_deactivate(struct atmel_spi *as, struct spi_device *spi)
377 {
378 	int chip_select;
379 	u32 mr;
380 
381 	if (spi->cs_gpiod)
382 		chip_select = as->native_cs_for_gpio;
383 	else
384 		chip_select = spi->chip_select;
385 
386 	/* only deactivate *this* device; sometimes transfers to
387 	 * another device may be active when this routine is called.
388 	 */
389 	mr = spi_readl(as, MR);
390 	if (~SPI_BFEXT(PCS, mr) & (1 << chip_select)) {
391 		mr = SPI_BFINS(PCS, 0xf, mr);
392 		spi_writel(as, MR, mr);
393 	}
394 
395 	dev_dbg(&spi->dev, "DEactivate NPCS, mr %08x\n", mr);
396 
397 	if (!spi->cs_gpiod)
398 		spi_writel(as, CR, SPI_BIT(LASTXFER));
399 }
400 
401 static void atmel_spi_lock(struct atmel_spi *as) __acquires(&as->lock)
402 {
403 	spin_lock_irqsave(&as->lock, as->flags);
404 }
405 
406 static void atmel_spi_unlock(struct atmel_spi *as) __releases(&as->lock)
407 {
408 	spin_unlock_irqrestore(&as->lock, as->flags);
409 }
410 
411 static inline bool atmel_spi_is_vmalloc_xfer(struct spi_transfer *xfer)
412 {
413 	return is_vmalloc_addr(xfer->tx_buf) || is_vmalloc_addr(xfer->rx_buf);
414 }
415 
416 static inline bool atmel_spi_use_dma(struct atmel_spi *as,
417 				struct spi_transfer *xfer)
418 {
419 	return as->use_dma && xfer->len >= DMA_MIN_BYTES;
420 }
421 
422 static bool atmel_spi_can_dma(struct spi_master *master,
423 			      struct spi_device *spi,
424 			      struct spi_transfer *xfer)
425 {
426 	struct atmel_spi *as = spi_master_get_devdata(master);
427 
428 	if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5))
429 		return atmel_spi_use_dma(as, xfer) &&
430 			!atmel_spi_is_vmalloc_xfer(xfer);
431 	else
432 		return atmel_spi_use_dma(as, xfer);
433 
434 }
435 
436 static int atmel_spi_dma_slave_config(struct atmel_spi *as, u8 bits_per_word)
437 {
438 	struct spi_master *master = platform_get_drvdata(as->pdev);
439 	struct dma_slave_config	slave_config;
440 	int err = 0;
441 
442 	if (bits_per_word > 8) {
443 		slave_config.dst_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
444 		slave_config.src_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
445 	} else {
446 		slave_config.dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
447 		slave_config.src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
448 	}
449 
450 	slave_config.dst_addr = (dma_addr_t)as->phybase + SPI_TDR;
451 	slave_config.src_addr = (dma_addr_t)as->phybase + SPI_RDR;
452 	slave_config.src_maxburst = 1;
453 	slave_config.dst_maxburst = 1;
454 	slave_config.device_fc = false;
455 
456 	/*
457 	 * This driver uses fixed peripheral select mode (PS bit set to '0' in
458 	 * the Mode Register).
459 	 * So according to the datasheet, when FIFOs are available (and
460 	 * enabled), the Transmit FIFO operates in Multiple Data Mode.
461 	 * In this mode, up to 2 data, not 4, can be written into the Transmit
462 	 * Data Register in a single access.
463 	 * However, the first data has to be written into the lowest 16 bits and
464 	 * the second data into the highest 16 bits of the Transmit
465 	 * Data Register. For 8bit data (the most frequent case), it would
466 	 * require to rework tx_buf so each data would actually fit 16 bits.
467 	 * So we'd rather write only one data at the time. Hence the transmit
468 	 * path works the same whether FIFOs are available (and enabled) or not.
469 	 */
470 	if (dmaengine_slave_config(master->dma_tx, &slave_config)) {
471 		dev_err(&as->pdev->dev,
472 			"failed to configure tx dma channel\n");
473 		err = -EINVAL;
474 	}
475 
476 	/*
477 	 * This driver configures the spi controller for master mode (MSTR bit
478 	 * set to '1' in the Mode Register).
479 	 * So according to the datasheet, when FIFOs are available (and
480 	 * enabled), the Receive FIFO operates in Single Data Mode.
481 	 * So the receive path works the same whether FIFOs are available (and
482 	 * enabled) or not.
483 	 */
484 	if (dmaengine_slave_config(master->dma_rx, &slave_config)) {
485 		dev_err(&as->pdev->dev,
486 			"failed to configure rx dma channel\n");
487 		err = -EINVAL;
488 	}
489 
490 	return err;
491 }
492 
493 static int atmel_spi_configure_dma(struct spi_master *master,
494 				   struct atmel_spi *as)
495 {
496 	struct device *dev = &as->pdev->dev;
497 	int err;
498 
499 	master->dma_tx = dma_request_chan(dev, "tx");
500 	if (IS_ERR(master->dma_tx)) {
501 		err = PTR_ERR(master->dma_tx);
502 		dev_dbg(dev, "No TX DMA channel, DMA is disabled\n");
503 		goto error_clear;
504 	}
505 
506 	master->dma_rx = dma_request_chan(dev, "rx");
507 	if (IS_ERR(master->dma_rx)) {
508 		err = PTR_ERR(master->dma_rx);
509 		/*
510 		 * No reason to check EPROBE_DEFER here since we have already
511 		 * requested tx channel.
512 		 */
513 		dev_dbg(dev, "No RX DMA channel, DMA is disabled\n");
514 		goto error;
515 	}
516 
517 	err = atmel_spi_dma_slave_config(as, 8);
518 	if (err)
519 		goto error;
520 
521 	dev_info(&as->pdev->dev,
522 			"Using %s (tx) and %s (rx) for DMA transfers\n",
523 			dma_chan_name(master->dma_tx),
524 			dma_chan_name(master->dma_rx));
525 
526 	return 0;
527 error:
528 	if (!IS_ERR(master->dma_rx))
529 		dma_release_channel(master->dma_rx);
530 	if (!IS_ERR(master->dma_tx))
531 		dma_release_channel(master->dma_tx);
532 error_clear:
533 	master->dma_tx = master->dma_rx = NULL;
534 	return err;
535 }
536 
537 static void atmel_spi_stop_dma(struct spi_master *master)
538 {
539 	if (master->dma_rx)
540 		dmaengine_terminate_all(master->dma_rx);
541 	if (master->dma_tx)
542 		dmaengine_terminate_all(master->dma_tx);
543 }
544 
545 static void atmel_spi_release_dma(struct spi_master *master)
546 {
547 	if (master->dma_rx) {
548 		dma_release_channel(master->dma_rx);
549 		master->dma_rx = NULL;
550 	}
551 	if (master->dma_tx) {
552 		dma_release_channel(master->dma_tx);
553 		master->dma_tx = NULL;
554 	}
555 }
556 
557 /* This function is called by the DMA driver from tasklet context */
558 static void dma_callback(void *data)
559 {
560 	struct spi_master	*master = data;
561 	struct atmel_spi	*as = spi_master_get_devdata(master);
562 
563 	if (is_vmalloc_addr(as->current_transfer->rx_buf) &&
564 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
565 		memcpy(as->current_transfer->rx_buf, as->addr_rx_bbuf,
566 		       as->current_transfer->len);
567 	}
568 	complete(&as->xfer_completion);
569 }
570 
571 /*
572  * Next transfer using PIO without FIFO.
573  */
574 static void atmel_spi_next_xfer_single(struct spi_master *master,
575 				       struct spi_transfer *xfer)
576 {
577 	struct atmel_spi	*as = spi_master_get_devdata(master);
578 	unsigned long xfer_pos = xfer->len - as->current_remaining_bytes;
579 
580 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_pio\n");
581 
582 	/* Make sure data is not remaining in RDR */
583 	spi_readl(as, RDR);
584 	while (spi_readl(as, SR) & SPI_BIT(RDRF)) {
585 		spi_readl(as, RDR);
586 		cpu_relax();
587 	}
588 
589 	if (xfer->bits_per_word > 8)
590 		spi_writel(as, TDR, *(u16 *)(xfer->tx_buf + xfer_pos));
591 	else
592 		spi_writel(as, TDR, *(u8 *)(xfer->tx_buf + xfer_pos));
593 
594 	dev_dbg(master->dev.parent,
595 		"  start pio xfer %p: len %u tx %p rx %p bitpw %d\n",
596 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
597 		xfer->bits_per_word);
598 
599 	/* Enable relevant interrupts */
600 	spi_writel(as, IER, SPI_BIT(RDRF) | SPI_BIT(OVRES));
601 }
602 
603 /*
604  * Next transfer using PIO with FIFO.
605  */
606 static void atmel_spi_next_xfer_fifo(struct spi_master *master,
607 				     struct spi_transfer *xfer)
608 {
609 	struct atmel_spi *as = spi_master_get_devdata(master);
610 	u32 current_remaining_data, num_data;
611 	u32 offset = xfer->len - as->current_remaining_bytes;
612 	const u16 *words = (const u16 *)((u8 *)xfer->tx_buf + offset);
613 	const u8  *bytes = (const u8  *)((u8 *)xfer->tx_buf + offset);
614 	u16 td0, td1;
615 	u32 fifomr;
616 
617 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_fifo\n");
618 
619 	/* Compute the number of data to transfer in the current iteration */
620 	current_remaining_data = ((xfer->bits_per_word > 8) ?
621 				  ((u32)as->current_remaining_bytes >> 1) :
622 				  (u32)as->current_remaining_bytes);
623 	num_data = min(current_remaining_data, as->fifo_size);
624 
625 	/* Flush RX and TX FIFOs */
626 	spi_writel(as, CR, SPI_BIT(RXFCLR) | SPI_BIT(TXFCLR));
627 	while (spi_readl(as, FLR))
628 		cpu_relax();
629 
630 	/* Set RX FIFO Threshold to the number of data to transfer */
631 	fifomr = spi_readl(as, FMR);
632 	spi_writel(as, FMR, SPI_BFINS(RXFTHRES, num_data, fifomr));
633 
634 	/* Clear FIFO flags in the Status Register, especially RXFTHF */
635 	(void)spi_readl(as, SR);
636 
637 	/* Fill TX FIFO */
638 	while (num_data >= 2) {
639 		if (xfer->bits_per_word > 8) {
640 			td0 = *words++;
641 			td1 = *words++;
642 		} else {
643 			td0 = *bytes++;
644 			td1 = *bytes++;
645 		}
646 
647 		spi_writel(as, TDR, (td1 << 16) | td0);
648 		num_data -= 2;
649 	}
650 
651 	if (num_data) {
652 		if (xfer->bits_per_word > 8)
653 			td0 = *words++;
654 		else
655 			td0 = *bytes++;
656 
657 		spi_writew(as, TDR, td0);
658 		num_data--;
659 	}
660 
661 	dev_dbg(master->dev.parent,
662 		"  start fifo xfer %p: len %u tx %p rx %p bitpw %d\n",
663 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
664 		xfer->bits_per_word);
665 
666 	/*
667 	 * Enable RX FIFO Threshold Flag interrupt to be notified about
668 	 * transfer completion.
669 	 */
670 	spi_writel(as, IER, SPI_BIT(RXFTHF) | SPI_BIT(OVRES));
671 }
672 
673 /*
674  * Next transfer using PIO.
675  */
676 static void atmel_spi_next_xfer_pio(struct spi_master *master,
677 				    struct spi_transfer *xfer)
678 {
679 	struct atmel_spi *as = spi_master_get_devdata(master);
680 
681 	if (as->fifo_size)
682 		atmel_spi_next_xfer_fifo(master, xfer);
683 	else
684 		atmel_spi_next_xfer_single(master, xfer);
685 }
686 
687 /*
688  * Submit next transfer for DMA.
689  */
690 static int atmel_spi_next_xfer_dma_submit(struct spi_master *master,
691 				struct spi_transfer *xfer,
692 				u32 *plen)
693 {
694 	struct atmel_spi	*as = spi_master_get_devdata(master);
695 	struct dma_chan		*rxchan = master->dma_rx;
696 	struct dma_chan		*txchan = master->dma_tx;
697 	struct dma_async_tx_descriptor *rxdesc;
698 	struct dma_async_tx_descriptor *txdesc;
699 	dma_cookie_t		cookie;
700 
701 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_dma_submit\n");
702 
703 	/* Check that the channels are available */
704 	if (!rxchan || !txchan)
705 		return -ENODEV;
706 
707 
708 	*plen = xfer->len;
709 
710 	if (atmel_spi_dma_slave_config(as, xfer->bits_per_word))
711 		goto err_exit;
712 
713 	/* Send both scatterlists */
714 	if (atmel_spi_is_vmalloc_xfer(xfer) &&
715 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
716 		rxdesc = dmaengine_prep_slave_single(rxchan,
717 						     as->dma_addr_rx_bbuf,
718 						     xfer->len,
719 						     DMA_DEV_TO_MEM,
720 						     DMA_PREP_INTERRUPT |
721 						     DMA_CTRL_ACK);
722 	} else {
723 		rxdesc = dmaengine_prep_slave_sg(rxchan,
724 						 xfer->rx_sg.sgl,
725 						 xfer->rx_sg.nents,
726 						 DMA_DEV_TO_MEM,
727 						 DMA_PREP_INTERRUPT |
728 						 DMA_CTRL_ACK);
729 	}
730 	if (!rxdesc)
731 		goto err_dma;
732 
733 	if (atmel_spi_is_vmalloc_xfer(xfer) &&
734 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
735 		memcpy(as->addr_tx_bbuf, xfer->tx_buf, xfer->len);
736 		txdesc = dmaengine_prep_slave_single(txchan,
737 						     as->dma_addr_tx_bbuf,
738 						     xfer->len, DMA_MEM_TO_DEV,
739 						     DMA_PREP_INTERRUPT |
740 						     DMA_CTRL_ACK);
741 	} else {
742 		txdesc = dmaengine_prep_slave_sg(txchan,
743 						 xfer->tx_sg.sgl,
744 						 xfer->tx_sg.nents,
745 						 DMA_MEM_TO_DEV,
746 						 DMA_PREP_INTERRUPT |
747 						 DMA_CTRL_ACK);
748 	}
749 	if (!txdesc)
750 		goto err_dma;
751 
752 	dev_dbg(master->dev.parent,
753 		"  start dma xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
754 		xfer, xfer->len, xfer->tx_buf, (unsigned long long)xfer->tx_dma,
755 		xfer->rx_buf, (unsigned long long)xfer->rx_dma);
756 
757 	/* Enable relevant interrupts */
758 	spi_writel(as, IER, SPI_BIT(OVRES));
759 
760 	/* Put the callback on the RX transfer only, that should finish last */
761 	rxdesc->callback = dma_callback;
762 	rxdesc->callback_param = master;
763 
764 	/* Submit and fire RX and TX with TX last so we're ready to read! */
765 	cookie = rxdesc->tx_submit(rxdesc);
766 	if (dma_submit_error(cookie))
767 		goto err_dma;
768 	cookie = txdesc->tx_submit(txdesc);
769 	if (dma_submit_error(cookie))
770 		goto err_dma;
771 	rxchan->device->device_issue_pending(rxchan);
772 	txchan->device->device_issue_pending(txchan);
773 
774 	return 0;
775 
776 err_dma:
777 	spi_writel(as, IDR, SPI_BIT(OVRES));
778 	atmel_spi_stop_dma(master);
779 err_exit:
780 	return -ENOMEM;
781 }
782 
783 static void atmel_spi_next_xfer_data(struct spi_master *master,
784 				struct spi_transfer *xfer,
785 				dma_addr_t *tx_dma,
786 				dma_addr_t *rx_dma,
787 				u32 *plen)
788 {
789 	*rx_dma = xfer->rx_dma + xfer->len - *plen;
790 	*tx_dma = xfer->tx_dma + xfer->len - *plen;
791 	if (*plen > master->max_dma_len)
792 		*plen = master->max_dma_len;
793 }
794 
795 static int atmel_spi_set_xfer_speed(struct atmel_spi *as,
796 				    struct spi_device *spi,
797 				    struct spi_transfer *xfer)
798 {
799 	u32			scbr, csr;
800 	unsigned long		bus_hz;
801 	int chip_select;
802 
803 	if (spi->cs_gpiod)
804 		chip_select = as->native_cs_for_gpio;
805 	else
806 		chip_select = spi->chip_select;
807 
808 	/* v1 chips start out at half the peripheral bus speed. */
809 	bus_hz = as->spi_clk;
810 	if (!atmel_spi_is_v2(as))
811 		bus_hz /= 2;
812 
813 	/*
814 	 * Calculate the lowest divider that satisfies the
815 	 * constraint, assuming div32/fdiv/mbz == 0.
816 	 */
817 	scbr = DIV_ROUND_UP(bus_hz, xfer->speed_hz);
818 
819 	/*
820 	 * If the resulting divider doesn't fit into the
821 	 * register bitfield, we can't satisfy the constraint.
822 	 */
823 	if (scbr >= (1 << SPI_SCBR_SIZE)) {
824 		dev_err(&spi->dev,
825 			"setup: %d Hz too slow, scbr %u; min %ld Hz\n",
826 			xfer->speed_hz, scbr, bus_hz/255);
827 		return -EINVAL;
828 	}
829 	if (scbr == 0) {
830 		dev_err(&spi->dev,
831 			"setup: %d Hz too high, scbr %u; max %ld Hz\n",
832 			xfer->speed_hz, scbr, bus_hz);
833 		return -EINVAL;
834 	}
835 	csr = spi_readl(as, CSR0 + 4 * chip_select);
836 	csr = SPI_BFINS(SCBR, scbr, csr);
837 	spi_writel(as, CSR0 + 4 * chip_select, csr);
838 	xfer->effective_speed_hz = bus_hz / scbr;
839 
840 	return 0;
841 }
842 
843 /*
844  * Submit next transfer for PDC.
845  * lock is held, spi irq is blocked
846  */
847 static void atmel_spi_pdc_next_xfer(struct spi_master *master,
848 					struct spi_transfer *xfer)
849 {
850 	struct atmel_spi	*as = spi_master_get_devdata(master);
851 	u32			len;
852 	dma_addr_t		tx_dma, rx_dma;
853 
854 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
855 
856 	len = as->current_remaining_bytes;
857 	atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
858 	as->current_remaining_bytes -= len;
859 
860 	spi_writel(as, RPR, rx_dma);
861 	spi_writel(as, TPR, tx_dma);
862 
863 	if (xfer->bits_per_word > 8)
864 		len >>= 1;
865 	spi_writel(as, RCR, len);
866 	spi_writel(as, TCR, len);
867 
868 	dev_dbg(&master->dev,
869 		"  start xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
870 		xfer, xfer->len, xfer->tx_buf,
871 		(unsigned long long)xfer->tx_dma, xfer->rx_buf,
872 		(unsigned long long)xfer->rx_dma);
873 
874 	if (as->current_remaining_bytes) {
875 		len = as->current_remaining_bytes;
876 		atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
877 		as->current_remaining_bytes -= len;
878 
879 		spi_writel(as, RNPR, rx_dma);
880 		spi_writel(as, TNPR, tx_dma);
881 
882 		if (xfer->bits_per_word > 8)
883 			len >>= 1;
884 		spi_writel(as, RNCR, len);
885 		spi_writel(as, TNCR, len);
886 
887 		dev_dbg(&master->dev,
888 			"  next xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
889 			xfer, xfer->len, xfer->tx_buf,
890 			(unsigned long long)xfer->tx_dma, xfer->rx_buf,
891 			(unsigned long long)xfer->rx_dma);
892 	}
893 
894 	/* REVISIT: We're waiting for RXBUFF before we start the next
895 	 * transfer because we need to handle some difficult timing
896 	 * issues otherwise. If we wait for TXBUFE in one transfer and
897 	 * then starts waiting for RXBUFF in the next, it's difficult
898 	 * to tell the difference between the RXBUFF interrupt we're
899 	 * actually waiting for and the RXBUFF interrupt of the
900 	 * previous transfer.
901 	 *
902 	 * It should be doable, though. Just not now...
903 	 */
904 	spi_writel(as, IER, SPI_BIT(RXBUFF) | SPI_BIT(OVRES));
905 	spi_writel(as, PTCR, SPI_BIT(TXTEN) | SPI_BIT(RXTEN));
906 }
907 
908 /*
909  * For DMA, tx_buf/tx_dma have the same relationship as rx_buf/rx_dma:
910  *  - The buffer is either valid for CPU access, else NULL
911  *  - If the buffer is valid, so is its DMA address
912  *
913  * This driver manages the dma address unless message->is_dma_mapped.
914  */
915 static int
916 atmel_spi_dma_map_xfer(struct atmel_spi *as, struct spi_transfer *xfer)
917 {
918 	struct device	*dev = &as->pdev->dev;
919 
920 	xfer->tx_dma = xfer->rx_dma = INVALID_DMA_ADDRESS;
921 	if (xfer->tx_buf) {
922 		/* tx_buf is a const void* where we need a void * for the dma
923 		 * mapping */
924 		void *nonconst_tx = (void *)xfer->tx_buf;
925 
926 		xfer->tx_dma = dma_map_single(dev,
927 				nonconst_tx, xfer->len,
928 				DMA_TO_DEVICE);
929 		if (dma_mapping_error(dev, xfer->tx_dma))
930 			return -ENOMEM;
931 	}
932 	if (xfer->rx_buf) {
933 		xfer->rx_dma = dma_map_single(dev,
934 				xfer->rx_buf, xfer->len,
935 				DMA_FROM_DEVICE);
936 		if (dma_mapping_error(dev, xfer->rx_dma)) {
937 			if (xfer->tx_buf)
938 				dma_unmap_single(dev,
939 						xfer->tx_dma, xfer->len,
940 						DMA_TO_DEVICE);
941 			return -ENOMEM;
942 		}
943 	}
944 	return 0;
945 }
946 
947 static void atmel_spi_dma_unmap_xfer(struct spi_master *master,
948 				     struct spi_transfer *xfer)
949 {
950 	if (xfer->tx_dma != INVALID_DMA_ADDRESS)
951 		dma_unmap_single(master->dev.parent, xfer->tx_dma,
952 				 xfer->len, DMA_TO_DEVICE);
953 	if (xfer->rx_dma != INVALID_DMA_ADDRESS)
954 		dma_unmap_single(master->dev.parent, xfer->rx_dma,
955 				 xfer->len, DMA_FROM_DEVICE);
956 }
957 
958 static void atmel_spi_disable_pdc_transfer(struct atmel_spi *as)
959 {
960 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
961 }
962 
963 static void
964 atmel_spi_pump_single_data(struct atmel_spi *as, struct spi_transfer *xfer)
965 {
966 	u8		*rxp;
967 	u16		*rxp16;
968 	unsigned long	xfer_pos = xfer->len - as->current_remaining_bytes;
969 
970 	if (xfer->bits_per_word > 8) {
971 		rxp16 = (u16 *)(((u8 *)xfer->rx_buf) + xfer_pos);
972 		*rxp16 = spi_readl(as, RDR);
973 	} else {
974 		rxp = ((u8 *)xfer->rx_buf) + xfer_pos;
975 		*rxp = spi_readl(as, RDR);
976 	}
977 	if (xfer->bits_per_word > 8) {
978 		if (as->current_remaining_bytes > 2)
979 			as->current_remaining_bytes -= 2;
980 		else
981 			as->current_remaining_bytes = 0;
982 	} else {
983 		as->current_remaining_bytes--;
984 	}
985 }
986 
987 static void
988 atmel_spi_pump_fifo_data(struct atmel_spi *as, struct spi_transfer *xfer)
989 {
990 	u32 fifolr = spi_readl(as, FLR);
991 	u32 num_bytes, num_data = SPI_BFEXT(RXFL, fifolr);
992 	u32 offset = xfer->len - as->current_remaining_bytes;
993 	u16 *words = (u16 *)((u8 *)xfer->rx_buf + offset);
994 	u8  *bytes = (u8  *)((u8 *)xfer->rx_buf + offset);
995 	u16 rd; /* RD field is the lowest 16 bits of RDR */
996 
997 	/* Update the number of remaining bytes to transfer */
998 	num_bytes = ((xfer->bits_per_word > 8) ?
999 		     (num_data << 1) :
1000 		     num_data);
1001 
1002 	if (as->current_remaining_bytes > num_bytes)
1003 		as->current_remaining_bytes -= num_bytes;
1004 	else
1005 		as->current_remaining_bytes = 0;
1006 
1007 	/* Handle odd number of bytes when data are more than 8bit width */
1008 	if (xfer->bits_per_word > 8)
1009 		as->current_remaining_bytes &= ~0x1;
1010 
1011 	/* Read data */
1012 	while (num_data) {
1013 		rd = spi_readl(as, RDR);
1014 		if (xfer->bits_per_word > 8)
1015 			*words++ = rd;
1016 		else
1017 			*bytes++ = rd;
1018 		num_data--;
1019 	}
1020 }
1021 
1022 /* Called from IRQ
1023  *
1024  * Must update "current_remaining_bytes" to keep track of data
1025  * to transfer.
1026  */
1027 static void
1028 atmel_spi_pump_pio_data(struct atmel_spi *as, struct spi_transfer *xfer)
1029 {
1030 	if (as->fifo_size)
1031 		atmel_spi_pump_fifo_data(as, xfer);
1032 	else
1033 		atmel_spi_pump_single_data(as, xfer);
1034 }
1035 
1036 /* Interrupt
1037  *
1038  */
1039 static irqreturn_t
1040 atmel_spi_pio_interrupt(int irq, void *dev_id)
1041 {
1042 	struct spi_master	*master = dev_id;
1043 	struct atmel_spi	*as = spi_master_get_devdata(master);
1044 	u32			status, pending, imr;
1045 	struct spi_transfer	*xfer;
1046 	int			ret = IRQ_NONE;
1047 
1048 	imr = spi_readl(as, IMR);
1049 	status = spi_readl(as, SR);
1050 	pending = status & imr;
1051 
1052 	if (pending & SPI_BIT(OVRES)) {
1053 		ret = IRQ_HANDLED;
1054 		spi_writel(as, IDR, SPI_BIT(OVRES));
1055 		dev_warn(master->dev.parent, "overrun\n");
1056 
1057 		/*
1058 		 * When we get an overrun, we disregard the current
1059 		 * transfer. Data will not be copied back from any
1060 		 * bounce buffer and msg->actual_len will not be
1061 		 * updated with the last xfer.
1062 		 *
1063 		 * We will also not process any remaning transfers in
1064 		 * the message.
1065 		 */
1066 		as->done_status = -EIO;
1067 		smp_wmb();
1068 
1069 		/* Clear any overrun happening while cleaning up */
1070 		spi_readl(as, SR);
1071 
1072 		complete(&as->xfer_completion);
1073 
1074 	} else if (pending & (SPI_BIT(RDRF) | SPI_BIT(RXFTHF))) {
1075 		atmel_spi_lock(as);
1076 
1077 		if (as->current_remaining_bytes) {
1078 			ret = IRQ_HANDLED;
1079 			xfer = as->current_transfer;
1080 			atmel_spi_pump_pio_data(as, xfer);
1081 			if (!as->current_remaining_bytes)
1082 				spi_writel(as, IDR, pending);
1083 
1084 			complete(&as->xfer_completion);
1085 		}
1086 
1087 		atmel_spi_unlock(as);
1088 	} else {
1089 		WARN_ONCE(pending, "IRQ not handled, pending = %x\n", pending);
1090 		ret = IRQ_HANDLED;
1091 		spi_writel(as, IDR, pending);
1092 	}
1093 
1094 	return ret;
1095 }
1096 
1097 static irqreturn_t
1098 atmel_spi_pdc_interrupt(int irq, void *dev_id)
1099 {
1100 	struct spi_master	*master = dev_id;
1101 	struct atmel_spi	*as = spi_master_get_devdata(master);
1102 	u32			status, pending, imr;
1103 	int			ret = IRQ_NONE;
1104 
1105 	imr = spi_readl(as, IMR);
1106 	status = spi_readl(as, SR);
1107 	pending = status & imr;
1108 
1109 	if (pending & SPI_BIT(OVRES)) {
1110 
1111 		ret = IRQ_HANDLED;
1112 
1113 		spi_writel(as, IDR, (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX)
1114 				     | SPI_BIT(OVRES)));
1115 
1116 		/* Clear any overrun happening while cleaning up */
1117 		spi_readl(as, SR);
1118 
1119 		as->done_status = -EIO;
1120 
1121 		complete(&as->xfer_completion);
1122 
1123 	} else if (pending & (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX))) {
1124 		ret = IRQ_HANDLED;
1125 
1126 		spi_writel(as, IDR, pending);
1127 
1128 		complete(&as->xfer_completion);
1129 	}
1130 
1131 	return ret;
1132 }
1133 
1134 static int atmel_word_delay_csr(struct spi_device *spi, struct atmel_spi *as)
1135 {
1136 	struct spi_delay *delay = &spi->word_delay;
1137 	u32 value = delay->value;
1138 
1139 	switch (delay->unit) {
1140 	case SPI_DELAY_UNIT_NSECS:
1141 		value /= 1000;
1142 		break;
1143 	case SPI_DELAY_UNIT_USECS:
1144 		break;
1145 	default:
1146 		return -EINVAL;
1147 	}
1148 
1149 	return (as->spi_clk / 1000000 * value) >> 5;
1150 }
1151 
1152 static void initialize_native_cs_for_gpio(struct atmel_spi *as)
1153 {
1154 	int i;
1155 	struct spi_master *master = platform_get_drvdata(as->pdev);
1156 
1157 	if (!as->native_cs_free)
1158 		return; /* already initialized */
1159 
1160 	if (!master->cs_gpiods)
1161 		return; /* No CS GPIO */
1162 
1163 	/*
1164 	 * On the first version of the controller (AT91RM9200), CS0
1165 	 * can't be used associated with GPIO
1166 	 */
1167 	if (atmel_spi_is_v2(as))
1168 		i = 0;
1169 	else
1170 		i = 1;
1171 
1172 	for (; i < 4; i++)
1173 		if (master->cs_gpiods[i])
1174 			as->native_cs_free |= BIT(i);
1175 
1176 	if (as->native_cs_free)
1177 		as->native_cs_for_gpio = ffs(as->native_cs_free);
1178 }
1179 
1180 static int atmel_spi_setup(struct spi_device *spi)
1181 {
1182 	struct atmel_spi	*as;
1183 	struct atmel_spi_device	*asd;
1184 	u32			csr;
1185 	unsigned int		bits = spi->bits_per_word;
1186 	int chip_select;
1187 	int			word_delay_csr;
1188 
1189 	as = spi_master_get_devdata(spi->master);
1190 
1191 	/* see notes above re chipselect */
1192 	if (!spi->cs_gpiod && (spi->mode & SPI_CS_HIGH)) {
1193 		dev_warn(&spi->dev, "setup: non GPIO CS can't be active-high\n");
1194 		return -EINVAL;
1195 	}
1196 
1197 	/* Setup() is called during spi_register_controller(aka
1198 	 * spi_register_master) but after all membmers of the cs_gpiod
1199 	 * array have been filled, so we can looked for which native
1200 	 * CS will be free for using with GPIO
1201 	 */
1202 	initialize_native_cs_for_gpio(as);
1203 
1204 	if (spi->cs_gpiod && as->native_cs_free) {
1205 		dev_err(&spi->dev,
1206 			"No native CS available to support this GPIO CS\n");
1207 		return -EBUSY;
1208 	}
1209 
1210 	if (spi->cs_gpiod)
1211 		chip_select = as->native_cs_for_gpio;
1212 	else
1213 		chip_select = spi->chip_select;
1214 
1215 	csr = SPI_BF(BITS, bits - 8);
1216 	if (spi->mode & SPI_CPOL)
1217 		csr |= SPI_BIT(CPOL);
1218 	if (!(spi->mode & SPI_CPHA))
1219 		csr |= SPI_BIT(NCPHA);
1220 
1221 	if (!spi->cs_gpiod)
1222 		csr |= SPI_BIT(CSAAT);
1223 	csr |= SPI_BF(DLYBS, 0);
1224 
1225 	word_delay_csr = atmel_word_delay_csr(spi, as);
1226 	if (word_delay_csr < 0)
1227 		return word_delay_csr;
1228 
1229 	/* DLYBCT adds delays between words.  This is useful for slow devices
1230 	 * that need a bit of time to setup the next transfer.
1231 	 */
1232 	csr |= SPI_BF(DLYBCT, word_delay_csr);
1233 
1234 	asd = spi->controller_state;
1235 	if (!asd) {
1236 		asd = kzalloc(sizeof(struct atmel_spi_device), GFP_KERNEL);
1237 		if (!asd)
1238 			return -ENOMEM;
1239 
1240 		spi->controller_state = asd;
1241 	}
1242 
1243 	asd->csr = csr;
1244 
1245 	dev_dbg(&spi->dev,
1246 		"setup: bpw %u mode 0x%x -> csr%d %08x\n",
1247 		bits, spi->mode, spi->chip_select, csr);
1248 
1249 	if (!atmel_spi_is_v2(as))
1250 		spi_writel(as, CSR0 + 4 * chip_select, csr);
1251 
1252 	return 0;
1253 }
1254 
1255 static void atmel_spi_set_cs(struct spi_device *spi, bool enable)
1256 {
1257 	struct atmel_spi *as = spi_master_get_devdata(spi->master);
1258 	/* the core doesn't really pass us enable/disable, but CS HIGH vs CS LOW
1259 	 * since we already have routines for activate/deactivate translate
1260 	 * high/low to active/inactive
1261 	 */
1262 	enable = (!!(spi->mode & SPI_CS_HIGH) == enable);
1263 
1264 	if (enable) {
1265 		cs_activate(as, spi);
1266 	} else {
1267 		cs_deactivate(as, spi);
1268 	}
1269 
1270 }
1271 
1272 static int atmel_spi_one_transfer(struct spi_master *master,
1273 					struct spi_device *spi,
1274 					struct spi_transfer *xfer)
1275 {
1276 	struct atmel_spi	*as;
1277 	u8			bits;
1278 	u32			len;
1279 	struct atmel_spi_device	*asd;
1280 	int			timeout;
1281 	int			ret;
1282 	unsigned long		dma_timeout;
1283 
1284 	as = spi_master_get_devdata(master);
1285 
1286 	asd = spi->controller_state;
1287 	bits = (asd->csr >> 4) & 0xf;
1288 	if (bits != xfer->bits_per_word - 8) {
1289 		dev_dbg(&spi->dev,
1290 			"you can't yet change bits_per_word in transfers\n");
1291 		return -ENOPROTOOPT;
1292 	}
1293 
1294 	/*
1295 	 * DMA map early, for performance (empties dcache ASAP) and
1296 	 * better fault reporting.
1297 	 */
1298 	if ((!master->cur_msg->is_dma_mapped)
1299 		&& as->use_pdc) {
1300 		if (atmel_spi_dma_map_xfer(as, xfer) < 0)
1301 			return -ENOMEM;
1302 	}
1303 
1304 	atmel_spi_set_xfer_speed(as, spi, xfer);
1305 
1306 	as->done_status = 0;
1307 	as->current_transfer = xfer;
1308 	as->current_remaining_bytes = xfer->len;
1309 	while (as->current_remaining_bytes) {
1310 		reinit_completion(&as->xfer_completion);
1311 
1312 		if (as->use_pdc) {
1313 			atmel_spi_lock(as);
1314 			atmel_spi_pdc_next_xfer(master, xfer);
1315 			atmel_spi_unlock(as);
1316 		} else if (atmel_spi_use_dma(as, xfer)) {
1317 			len = as->current_remaining_bytes;
1318 			ret = atmel_spi_next_xfer_dma_submit(master,
1319 								xfer, &len);
1320 			if (ret) {
1321 				dev_err(&spi->dev,
1322 					"unable to use DMA, fallback to PIO\n");
1323 				as->done_status = ret;
1324 				break;
1325 			} else {
1326 				as->current_remaining_bytes -= len;
1327 				if (as->current_remaining_bytes < 0)
1328 					as->current_remaining_bytes = 0;
1329 			}
1330 		} else {
1331 			atmel_spi_lock(as);
1332 			atmel_spi_next_xfer_pio(master, xfer);
1333 			atmel_spi_unlock(as);
1334 		}
1335 
1336 		dma_timeout = wait_for_completion_timeout(&as->xfer_completion,
1337 							  SPI_DMA_TIMEOUT);
1338 		if (WARN_ON(dma_timeout == 0)) {
1339 			dev_err(&spi->dev, "spi transfer timeout\n");
1340 			as->done_status = -EIO;
1341 		}
1342 
1343 		if (as->done_status)
1344 			break;
1345 	}
1346 
1347 	if (as->done_status) {
1348 		if (as->use_pdc) {
1349 			dev_warn(master->dev.parent,
1350 				"overrun (%u/%u remaining)\n",
1351 				spi_readl(as, TCR), spi_readl(as, RCR));
1352 
1353 			/*
1354 			 * Clean up DMA registers and make sure the data
1355 			 * registers are empty.
1356 			 */
1357 			spi_writel(as, RNCR, 0);
1358 			spi_writel(as, TNCR, 0);
1359 			spi_writel(as, RCR, 0);
1360 			spi_writel(as, TCR, 0);
1361 			for (timeout = 1000; timeout; timeout--)
1362 				if (spi_readl(as, SR) & SPI_BIT(TXEMPTY))
1363 					break;
1364 			if (!timeout)
1365 				dev_warn(master->dev.parent,
1366 					 "timeout waiting for TXEMPTY");
1367 			while (spi_readl(as, SR) & SPI_BIT(RDRF))
1368 				spi_readl(as, RDR);
1369 
1370 			/* Clear any overrun happening while cleaning up */
1371 			spi_readl(as, SR);
1372 
1373 		} else if (atmel_spi_use_dma(as, xfer)) {
1374 			atmel_spi_stop_dma(master);
1375 		}
1376 	}
1377 
1378 	if (!master->cur_msg->is_dma_mapped
1379 		&& as->use_pdc)
1380 		atmel_spi_dma_unmap_xfer(master, xfer);
1381 
1382 	if (as->use_pdc)
1383 		atmel_spi_disable_pdc_transfer(as);
1384 
1385 	return as->done_status;
1386 }
1387 
1388 static void atmel_spi_cleanup(struct spi_device *spi)
1389 {
1390 	struct atmel_spi_device	*asd = spi->controller_state;
1391 
1392 	if (!asd)
1393 		return;
1394 
1395 	spi->controller_state = NULL;
1396 	kfree(asd);
1397 }
1398 
1399 static inline unsigned int atmel_get_version(struct atmel_spi *as)
1400 {
1401 	return spi_readl(as, VERSION) & 0x00000fff;
1402 }
1403 
1404 static void atmel_get_caps(struct atmel_spi *as)
1405 {
1406 	unsigned int version;
1407 
1408 	version = atmel_get_version(as);
1409 
1410 	as->caps.is_spi2 = version > 0x121;
1411 	as->caps.has_wdrbt = version >= 0x210;
1412 	as->caps.has_dma_support = version >= 0x212;
1413 	as->caps.has_pdc_support = version < 0x212;
1414 }
1415 
1416 static void atmel_spi_init(struct atmel_spi *as)
1417 {
1418 	spi_writel(as, CR, SPI_BIT(SWRST));
1419 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1420 
1421 	/* It is recommended to enable FIFOs first thing after reset */
1422 	if (as->fifo_size)
1423 		spi_writel(as, CR, SPI_BIT(FIFOEN));
1424 
1425 	if (as->caps.has_wdrbt) {
1426 		spi_writel(as, MR, SPI_BIT(WDRBT) | SPI_BIT(MODFDIS)
1427 				| SPI_BIT(MSTR));
1428 	} else {
1429 		spi_writel(as, MR, SPI_BIT(MSTR) | SPI_BIT(MODFDIS));
1430 	}
1431 
1432 	if (as->use_pdc)
1433 		spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1434 	spi_writel(as, CR, SPI_BIT(SPIEN));
1435 }
1436 
1437 static int atmel_spi_probe(struct platform_device *pdev)
1438 {
1439 	struct resource		*regs;
1440 	int			irq;
1441 	struct clk		*clk;
1442 	int			ret;
1443 	struct spi_master	*master;
1444 	struct atmel_spi	*as;
1445 
1446 	/* Select default pin state */
1447 	pinctrl_pm_select_default_state(&pdev->dev);
1448 
1449 	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1450 	if (!regs)
1451 		return -ENXIO;
1452 
1453 	irq = platform_get_irq(pdev, 0);
1454 	if (irq < 0)
1455 		return irq;
1456 
1457 	clk = devm_clk_get(&pdev->dev, "spi_clk");
1458 	if (IS_ERR(clk))
1459 		return PTR_ERR(clk);
1460 
1461 	/* setup spi core then atmel-specific driver state */
1462 	master = spi_alloc_master(&pdev->dev, sizeof(*as));
1463 	if (!master)
1464 		return -ENOMEM;
1465 
1466 	/* the spi->mode bits understood by this driver: */
1467 	master->use_gpio_descriptors = true;
1468 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
1469 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(8, 16);
1470 	master->dev.of_node = pdev->dev.of_node;
1471 	master->bus_num = pdev->id;
1472 	master->num_chipselect = 4;
1473 	master->setup = atmel_spi_setup;
1474 	master->flags = (SPI_MASTER_MUST_RX | SPI_MASTER_MUST_TX |
1475 			SPI_MASTER_GPIO_SS);
1476 	master->transfer_one = atmel_spi_one_transfer;
1477 	master->set_cs = atmel_spi_set_cs;
1478 	master->cleanup = atmel_spi_cleanup;
1479 	master->auto_runtime_pm = true;
1480 	master->max_dma_len = SPI_MAX_DMA_XFER;
1481 	master->can_dma = atmel_spi_can_dma;
1482 	platform_set_drvdata(pdev, master);
1483 
1484 	as = spi_master_get_devdata(master);
1485 
1486 	spin_lock_init(&as->lock);
1487 
1488 	as->pdev = pdev;
1489 	as->regs = devm_ioremap_resource(&pdev->dev, regs);
1490 	if (IS_ERR(as->regs)) {
1491 		ret = PTR_ERR(as->regs);
1492 		goto out_unmap_regs;
1493 	}
1494 	as->phybase = regs->start;
1495 	as->irq = irq;
1496 	as->clk = clk;
1497 
1498 	init_completion(&as->xfer_completion);
1499 
1500 	atmel_get_caps(as);
1501 
1502 	as->use_dma = false;
1503 	as->use_pdc = false;
1504 	if (as->caps.has_dma_support) {
1505 		ret = atmel_spi_configure_dma(master, as);
1506 		if (ret == 0) {
1507 			as->use_dma = true;
1508 		} else if (ret == -EPROBE_DEFER) {
1509 			goto out_unmap_regs;
1510 		}
1511 	} else if (as->caps.has_pdc_support) {
1512 		as->use_pdc = true;
1513 	}
1514 
1515 	if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
1516 		as->addr_rx_bbuf = dma_alloc_coherent(&pdev->dev,
1517 						      SPI_MAX_DMA_XFER,
1518 						      &as->dma_addr_rx_bbuf,
1519 						      GFP_KERNEL | GFP_DMA);
1520 		if (!as->addr_rx_bbuf) {
1521 			as->use_dma = false;
1522 		} else {
1523 			as->addr_tx_bbuf = dma_alloc_coherent(&pdev->dev,
1524 					SPI_MAX_DMA_XFER,
1525 					&as->dma_addr_tx_bbuf,
1526 					GFP_KERNEL | GFP_DMA);
1527 			if (!as->addr_tx_bbuf) {
1528 				as->use_dma = false;
1529 				dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1530 						  as->addr_rx_bbuf,
1531 						  as->dma_addr_rx_bbuf);
1532 			}
1533 		}
1534 		if (!as->use_dma)
1535 			dev_info(master->dev.parent,
1536 				 "  can not allocate dma coherent memory\n");
1537 	}
1538 
1539 	if (as->caps.has_dma_support && !as->use_dma)
1540 		dev_info(&pdev->dev, "Atmel SPI Controller using PIO only\n");
1541 
1542 	if (as->use_pdc) {
1543 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pdc_interrupt,
1544 					0, dev_name(&pdev->dev), master);
1545 	} else {
1546 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pio_interrupt,
1547 					0, dev_name(&pdev->dev), master);
1548 	}
1549 	if (ret)
1550 		goto out_unmap_regs;
1551 
1552 	/* Initialize the hardware */
1553 	ret = clk_prepare_enable(clk);
1554 	if (ret)
1555 		goto out_free_irq;
1556 
1557 	as->spi_clk = clk_get_rate(clk);
1558 
1559 	as->fifo_size = 0;
1560 	if (!of_property_read_u32(pdev->dev.of_node, "atmel,fifo-size",
1561 				  &as->fifo_size)) {
1562 		dev_info(&pdev->dev, "Using FIFO (%u data)\n", as->fifo_size);
1563 	}
1564 
1565 	atmel_spi_init(as);
1566 
1567 	pm_runtime_set_autosuspend_delay(&pdev->dev, AUTOSUSPEND_TIMEOUT);
1568 	pm_runtime_use_autosuspend(&pdev->dev);
1569 	pm_runtime_set_active(&pdev->dev);
1570 	pm_runtime_enable(&pdev->dev);
1571 
1572 	ret = devm_spi_register_master(&pdev->dev, master);
1573 	if (ret)
1574 		goto out_free_dma;
1575 
1576 	/* go! */
1577 	dev_info(&pdev->dev, "Atmel SPI Controller version 0x%x at 0x%08lx (irq %d)\n",
1578 			atmel_get_version(as), (unsigned long)regs->start,
1579 			irq);
1580 
1581 	return 0;
1582 
1583 out_free_dma:
1584 	pm_runtime_disable(&pdev->dev);
1585 	pm_runtime_set_suspended(&pdev->dev);
1586 
1587 	if (as->use_dma)
1588 		atmel_spi_release_dma(master);
1589 
1590 	spi_writel(as, CR, SPI_BIT(SWRST));
1591 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1592 	clk_disable_unprepare(clk);
1593 out_free_irq:
1594 out_unmap_regs:
1595 	spi_master_put(master);
1596 	return ret;
1597 }
1598 
1599 static int atmel_spi_remove(struct platform_device *pdev)
1600 {
1601 	struct spi_master	*master = platform_get_drvdata(pdev);
1602 	struct atmel_spi	*as = spi_master_get_devdata(master);
1603 
1604 	pm_runtime_get_sync(&pdev->dev);
1605 
1606 	/* reset the hardware and block queue progress */
1607 	if (as->use_dma) {
1608 		atmel_spi_stop_dma(master);
1609 		atmel_spi_release_dma(master);
1610 		if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
1611 			dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1612 					  as->addr_tx_bbuf,
1613 					  as->dma_addr_tx_bbuf);
1614 			dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1615 					  as->addr_rx_bbuf,
1616 					  as->dma_addr_rx_bbuf);
1617 		}
1618 	}
1619 
1620 	spin_lock_irq(&as->lock);
1621 	spi_writel(as, CR, SPI_BIT(SWRST));
1622 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1623 	spi_readl(as, SR);
1624 	spin_unlock_irq(&as->lock);
1625 
1626 	clk_disable_unprepare(as->clk);
1627 
1628 	pm_runtime_put_noidle(&pdev->dev);
1629 	pm_runtime_disable(&pdev->dev);
1630 
1631 	return 0;
1632 }
1633 
1634 static int atmel_spi_runtime_suspend(struct device *dev)
1635 {
1636 	struct spi_master *master = dev_get_drvdata(dev);
1637 	struct atmel_spi *as = spi_master_get_devdata(master);
1638 
1639 	clk_disable_unprepare(as->clk);
1640 	pinctrl_pm_select_sleep_state(dev);
1641 
1642 	return 0;
1643 }
1644 
1645 static int atmel_spi_runtime_resume(struct device *dev)
1646 {
1647 	struct spi_master *master = dev_get_drvdata(dev);
1648 	struct atmel_spi *as = spi_master_get_devdata(master);
1649 
1650 	pinctrl_pm_select_default_state(dev);
1651 
1652 	return clk_prepare_enable(as->clk);
1653 }
1654 
1655 static int atmel_spi_suspend(struct device *dev)
1656 {
1657 	struct spi_master *master = dev_get_drvdata(dev);
1658 	int ret;
1659 
1660 	/* Stop the queue running */
1661 	ret = spi_master_suspend(master);
1662 	if (ret)
1663 		return ret;
1664 
1665 	if (!pm_runtime_suspended(dev))
1666 		atmel_spi_runtime_suspend(dev);
1667 
1668 	return 0;
1669 }
1670 
1671 static int atmel_spi_resume(struct device *dev)
1672 {
1673 	struct spi_master *master = dev_get_drvdata(dev);
1674 	struct atmel_spi *as = spi_master_get_devdata(master);
1675 	int ret;
1676 
1677 	ret = clk_prepare_enable(as->clk);
1678 	if (ret)
1679 		return ret;
1680 
1681 	atmel_spi_init(as);
1682 
1683 	clk_disable_unprepare(as->clk);
1684 
1685 	if (!pm_runtime_suspended(dev)) {
1686 		ret = atmel_spi_runtime_resume(dev);
1687 		if (ret)
1688 			return ret;
1689 	}
1690 
1691 	/* Start the queue running */
1692 	return spi_master_resume(master);
1693 }
1694 
1695 static const struct dev_pm_ops atmel_spi_pm_ops = {
1696 	SYSTEM_SLEEP_PM_OPS(atmel_spi_suspend, atmel_spi_resume)
1697 	RUNTIME_PM_OPS(atmel_spi_runtime_suspend,
1698 		       atmel_spi_runtime_resume, NULL)
1699 };
1700 
1701 static const struct of_device_id atmel_spi_dt_ids[] = {
1702 	{ .compatible = "atmel,at91rm9200-spi" },
1703 	{ /* sentinel */ }
1704 };
1705 
1706 MODULE_DEVICE_TABLE(of, atmel_spi_dt_ids);
1707 
1708 static struct platform_driver atmel_spi_driver = {
1709 	.driver		= {
1710 		.name	= "atmel_spi",
1711 		.pm	= pm_ptr(&atmel_spi_pm_ops),
1712 		.of_match_table	= atmel_spi_dt_ids,
1713 	},
1714 	.probe		= atmel_spi_probe,
1715 	.remove		= atmel_spi_remove,
1716 };
1717 module_platform_driver(atmel_spi_driver);
1718 
1719 MODULE_DESCRIPTION("Atmel AT32/AT91 SPI Controller driver");
1720 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1721 MODULE_LICENSE("GPL");
1722 MODULE_ALIAS("platform:atmel_spi");
1723