xref: /openbmc/linux/drivers/spi/spi-atmel.c (revision b38269ec)
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,
437 				struct dma_slave_config *slave_config,
438 				u8 bits_per_word)
439 {
440 	struct spi_master *master = platform_get_drvdata(as->pdev);
441 	int err = 0;
442 
443 	if (bits_per_word > 8) {
444 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
445 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
446 	} else {
447 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
448 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
449 	}
450 
451 	slave_config->dst_addr = (dma_addr_t)as->phybase + SPI_TDR;
452 	slave_config->src_addr = (dma_addr_t)as->phybase + SPI_RDR;
453 	slave_config->src_maxburst = 1;
454 	slave_config->dst_maxburst = 1;
455 	slave_config->device_fc = false;
456 
457 	/*
458 	 * This driver uses fixed peripheral select mode (PS bit set to '0' in
459 	 * the Mode Register).
460 	 * So according to the datasheet, when FIFOs are available (and
461 	 * enabled), the Transmit FIFO operates in Multiple Data Mode.
462 	 * In this mode, up to 2 data, not 4, can be written into the Transmit
463 	 * Data Register in a single access.
464 	 * However, the first data has to be written into the lowest 16 bits and
465 	 * the second data into the highest 16 bits of the Transmit
466 	 * Data Register. For 8bit data (the most frequent case), it would
467 	 * require to rework tx_buf so each data would actualy fit 16 bits.
468 	 * So we'd rather write only one data at the time. Hence the transmit
469 	 * path works the same whether FIFOs are available (and enabled) or not.
470 	 */
471 	slave_config->direction = DMA_MEM_TO_DEV;
472 	if (dmaengine_slave_config(master->dma_tx, slave_config)) {
473 		dev_err(&as->pdev->dev,
474 			"failed to configure tx dma channel\n");
475 		err = -EINVAL;
476 	}
477 
478 	/*
479 	 * This driver configures the spi controller for master mode (MSTR bit
480 	 * set to '1' in the Mode Register).
481 	 * So according to the datasheet, when FIFOs are available (and
482 	 * enabled), the Receive FIFO operates in Single Data Mode.
483 	 * So the receive path works the same whether FIFOs are available (and
484 	 * enabled) or not.
485 	 */
486 	slave_config->direction = DMA_DEV_TO_MEM;
487 	if (dmaengine_slave_config(master->dma_rx, slave_config)) {
488 		dev_err(&as->pdev->dev,
489 			"failed to configure rx dma channel\n");
490 		err = -EINVAL;
491 	}
492 
493 	return err;
494 }
495 
496 static int atmel_spi_configure_dma(struct spi_master *master,
497 				   struct atmel_spi *as)
498 {
499 	struct dma_slave_config	slave_config;
500 	struct device *dev = &as->pdev->dev;
501 	int err;
502 
503 	master->dma_tx = dma_request_chan(dev, "tx");
504 	if (IS_ERR(master->dma_tx)) {
505 		err = PTR_ERR(master->dma_tx);
506 		dev_dbg(dev, "No TX DMA channel, DMA is disabled\n");
507 		goto error_clear;
508 	}
509 
510 	master->dma_rx = dma_request_chan(dev, "rx");
511 	if (IS_ERR(master->dma_rx)) {
512 		err = PTR_ERR(master->dma_rx);
513 		/*
514 		 * No reason to check EPROBE_DEFER here since we have already
515 		 * requested tx channel.
516 		 */
517 		dev_dbg(dev, "No RX DMA channel, DMA is disabled\n");
518 		goto error;
519 	}
520 
521 	err = atmel_spi_dma_slave_config(as, &slave_config, 8);
522 	if (err)
523 		goto error;
524 
525 	dev_info(&as->pdev->dev,
526 			"Using %s (tx) and %s (rx) for DMA transfers\n",
527 			dma_chan_name(master->dma_tx),
528 			dma_chan_name(master->dma_rx));
529 
530 	return 0;
531 error:
532 	if (!IS_ERR(master->dma_rx))
533 		dma_release_channel(master->dma_rx);
534 	if (!IS_ERR(master->dma_tx))
535 		dma_release_channel(master->dma_tx);
536 error_clear:
537 	master->dma_tx = master->dma_rx = NULL;
538 	return err;
539 }
540 
541 static void atmel_spi_stop_dma(struct spi_master *master)
542 {
543 	if (master->dma_rx)
544 		dmaengine_terminate_all(master->dma_rx);
545 	if (master->dma_tx)
546 		dmaengine_terminate_all(master->dma_tx);
547 }
548 
549 static void atmel_spi_release_dma(struct spi_master *master)
550 {
551 	if (master->dma_rx) {
552 		dma_release_channel(master->dma_rx);
553 		master->dma_rx = NULL;
554 	}
555 	if (master->dma_tx) {
556 		dma_release_channel(master->dma_tx);
557 		master->dma_tx = NULL;
558 	}
559 }
560 
561 /* This function is called by the DMA driver from tasklet context */
562 static void dma_callback(void *data)
563 {
564 	struct spi_master	*master = data;
565 	struct atmel_spi	*as = spi_master_get_devdata(master);
566 
567 	if (is_vmalloc_addr(as->current_transfer->rx_buf) &&
568 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
569 		memcpy(as->current_transfer->rx_buf, as->addr_rx_bbuf,
570 		       as->current_transfer->len);
571 	}
572 	complete(&as->xfer_completion);
573 }
574 
575 /*
576  * Next transfer using PIO without FIFO.
577  */
578 static void atmel_spi_next_xfer_single(struct spi_master *master,
579 				       struct spi_transfer *xfer)
580 {
581 	struct atmel_spi	*as = spi_master_get_devdata(master);
582 	unsigned long xfer_pos = xfer->len - as->current_remaining_bytes;
583 
584 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_pio\n");
585 
586 	/* Make sure data is not remaining in RDR */
587 	spi_readl(as, RDR);
588 	while (spi_readl(as, SR) & SPI_BIT(RDRF)) {
589 		spi_readl(as, RDR);
590 		cpu_relax();
591 	}
592 
593 	if (xfer->bits_per_word > 8)
594 		spi_writel(as, TDR, *(u16 *)(xfer->tx_buf + xfer_pos));
595 	else
596 		spi_writel(as, TDR, *(u8 *)(xfer->tx_buf + xfer_pos));
597 
598 	dev_dbg(master->dev.parent,
599 		"  start pio xfer %p: len %u tx %p rx %p bitpw %d\n",
600 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
601 		xfer->bits_per_word);
602 
603 	/* Enable relevant interrupts */
604 	spi_writel(as, IER, SPI_BIT(RDRF) | SPI_BIT(OVRES));
605 }
606 
607 /*
608  * Next transfer using PIO with FIFO.
609  */
610 static void atmel_spi_next_xfer_fifo(struct spi_master *master,
611 				     struct spi_transfer *xfer)
612 {
613 	struct atmel_spi *as = spi_master_get_devdata(master);
614 	u32 current_remaining_data, num_data;
615 	u32 offset = xfer->len - as->current_remaining_bytes;
616 	const u16 *words = (const u16 *)((u8 *)xfer->tx_buf + offset);
617 	const u8  *bytes = (const u8  *)((u8 *)xfer->tx_buf + offset);
618 	u16 td0, td1;
619 	u32 fifomr;
620 
621 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_fifo\n");
622 
623 	/* Compute the number of data to transfer in the current iteration */
624 	current_remaining_data = ((xfer->bits_per_word > 8) ?
625 				  ((u32)as->current_remaining_bytes >> 1) :
626 				  (u32)as->current_remaining_bytes);
627 	num_data = min(current_remaining_data, as->fifo_size);
628 
629 	/* Flush RX and TX FIFOs */
630 	spi_writel(as, CR, SPI_BIT(RXFCLR) | SPI_BIT(TXFCLR));
631 	while (spi_readl(as, FLR))
632 		cpu_relax();
633 
634 	/* Set RX FIFO Threshold to the number of data to transfer */
635 	fifomr = spi_readl(as, FMR);
636 	spi_writel(as, FMR, SPI_BFINS(RXFTHRES, num_data, fifomr));
637 
638 	/* Clear FIFO flags in the Status Register, especially RXFTHF */
639 	(void)spi_readl(as, SR);
640 
641 	/* Fill TX FIFO */
642 	while (num_data >= 2) {
643 		if (xfer->bits_per_word > 8) {
644 			td0 = *words++;
645 			td1 = *words++;
646 		} else {
647 			td0 = *bytes++;
648 			td1 = *bytes++;
649 		}
650 
651 		spi_writel(as, TDR, (td1 << 16) | td0);
652 		num_data -= 2;
653 	}
654 
655 	if (num_data) {
656 		if (xfer->bits_per_word > 8)
657 			td0 = *words++;
658 		else
659 			td0 = *bytes++;
660 
661 		spi_writew(as, TDR, td0);
662 		num_data--;
663 	}
664 
665 	dev_dbg(master->dev.parent,
666 		"  start fifo xfer %p: len %u tx %p rx %p bitpw %d\n",
667 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
668 		xfer->bits_per_word);
669 
670 	/*
671 	 * Enable RX FIFO Threshold Flag interrupt to be notified about
672 	 * transfer completion.
673 	 */
674 	spi_writel(as, IER, SPI_BIT(RXFTHF) | SPI_BIT(OVRES));
675 }
676 
677 /*
678  * Next transfer using PIO.
679  */
680 static void atmel_spi_next_xfer_pio(struct spi_master *master,
681 				    struct spi_transfer *xfer)
682 {
683 	struct atmel_spi *as = spi_master_get_devdata(master);
684 
685 	if (as->fifo_size)
686 		atmel_spi_next_xfer_fifo(master, xfer);
687 	else
688 		atmel_spi_next_xfer_single(master, xfer);
689 }
690 
691 /*
692  * Submit next transfer for DMA.
693  */
694 static int atmel_spi_next_xfer_dma_submit(struct spi_master *master,
695 				struct spi_transfer *xfer,
696 				u32 *plen)
697 {
698 	struct atmel_spi	*as = spi_master_get_devdata(master);
699 	struct dma_chan		*rxchan = master->dma_rx;
700 	struct dma_chan		*txchan = master->dma_tx;
701 	struct dma_async_tx_descriptor *rxdesc;
702 	struct dma_async_tx_descriptor *txdesc;
703 	struct dma_slave_config	slave_config;
704 	dma_cookie_t		cookie;
705 
706 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_dma_submit\n");
707 
708 	/* Check that the channels are available */
709 	if (!rxchan || !txchan)
710 		return -ENODEV;
711 
712 
713 	*plen = xfer->len;
714 
715 	if (atmel_spi_dma_slave_config(as, &slave_config,
716 				       xfer->bits_per_word))
717 		goto err_exit;
718 
719 	/* Send both scatterlists */
720 	if (atmel_spi_is_vmalloc_xfer(xfer) &&
721 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
722 		rxdesc = dmaengine_prep_slave_single(rxchan,
723 						     as->dma_addr_rx_bbuf,
724 						     xfer->len,
725 						     DMA_DEV_TO_MEM,
726 						     DMA_PREP_INTERRUPT |
727 						     DMA_CTRL_ACK);
728 	} else {
729 		rxdesc = dmaengine_prep_slave_sg(rxchan,
730 						 xfer->rx_sg.sgl,
731 						 xfer->rx_sg.nents,
732 						 DMA_DEV_TO_MEM,
733 						 DMA_PREP_INTERRUPT |
734 						 DMA_CTRL_ACK);
735 	}
736 	if (!rxdesc)
737 		goto err_dma;
738 
739 	if (atmel_spi_is_vmalloc_xfer(xfer) &&
740 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
741 		memcpy(as->addr_tx_bbuf, xfer->tx_buf, xfer->len);
742 		txdesc = dmaengine_prep_slave_single(txchan,
743 						     as->dma_addr_tx_bbuf,
744 						     xfer->len, DMA_MEM_TO_DEV,
745 						     DMA_PREP_INTERRUPT |
746 						     DMA_CTRL_ACK);
747 	} else {
748 		txdesc = dmaengine_prep_slave_sg(txchan,
749 						 xfer->tx_sg.sgl,
750 						 xfer->tx_sg.nents,
751 						 DMA_MEM_TO_DEV,
752 						 DMA_PREP_INTERRUPT |
753 						 DMA_CTRL_ACK);
754 	}
755 	if (!txdesc)
756 		goto err_dma;
757 
758 	dev_dbg(master->dev.parent,
759 		"  start dma xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
760 		xfer, xfer->len, xfer->tx_buf, (unsigned long long)xfer->tx_dma,
761 		xfer->rx_buf, (unsigned long long)xfer->rx_dma);
762 
763 	/* Enable relevant interrupts */
764 	spi_writel(as, IER, SPI_BIT(OVRES));
765 
766 	/* Put the callback on the RX transfer only, that should finish last */
767 	rxdesc->callback = dma_callback;
768 	rxdesc->callback_param = master;
769 
770 	/* Submit and fire RX and TX with TX last so we're ready to read! */
771 	cookie = rxdesc->tx_submit(rxdesc);
772 	if (dma_submit_error(cookie))
773 		goto err_dma;
774 	cookie = txdesc->tx_submit(txdesc);
775 	if (dma_submit_error(cookie))
776 		goto err_dma;
777 	rxchan->device->device_issue_pending(rxchan);
778 	txchan->device->device_issue_pending(txchan);
779 
780 	return 0;
781 
782 err_dma:
783 	spi_writel(as, IDR, SPI_BIT(OVRES));
784 	atmel_spi_stop_dma(master);
785 err_exit:
786 	return -ENOMEM;
787 }
788 
789 static void atmel_spi_next_xfer_data(struct spi_master *master,
790 				struct spi_transfer *xfer,
791 				dma_addr_t *tx_dma,
792 				dma_addr_t *rx_dma,
793 				u32 *plen)
794 {
795 	*rx_dma = xfer->rx_dma + xfer->len - *plen;
796 	*tx_dma = xfer->tx_dma + xfer->len - *plen;
797 	if (*plen > master->max_dma_len)
798 		*plen = master->max_dma_len;
799 }
800 
801 static int atmel_spi_set_xfer_speed(struct atmel_spi *as,
802 				    struct spi_device *spi,
803 				    struct spi_transfer *xfer)
804 {
805 	u32			scbr, csr;
806 	unsigned long		bus_hz;
807 	int chip_select;
808 
809 	if (spi->cs_gpiod)
810 		chip_select = as->native_cs_for_gpio;
811 	else
812 		chip_select = spi->chip_select;
813 
814 	/* v1 chips start out at half the peripheral bus speed. */
815 	bus_hz = as->spi_clk;
816 	if (!atmel_spi_is_v2(as))
817 		bus_hz /= 2;
818 
819 	/*
820 	 * Calculate the lowest divider that satisfies the
821 	 * constraint, assuming div32/fdiv/mbz == 0.
822 	 */
823 	scbr = DIV_ROUND_UP(bus_hz, xfer->speed_hz);
824 
825 	/*
826 	 * If the resulting divider doesn't fit into the
827 	 * register bitfield, we can't satisfy the constraint.
828 	 */
829 	if (scbr >= (1 << SPI_SCBR_SIZE)) {
830 		dev_err(&spi->dev,
831 			"setup: %d Hz too slow, scbr %u; min %ld Hz\n",
832 			xfer->speed_hz, scbr, bus_hz/255);
833 		return -EINVAL;
834 	}
835 	if (scbr == 0) {
836 		dev_err(&spi->dev,
837 			"setup: %d Hz too high, scbr %u; max %ld Hz\n",
838 			xfer->speed_hz, scbr, bus_hz);
839 		return -EINVAL;
840 	}
841 	csr = spi_readl(as, CSR0 + 4 * chip_select);
842 	csr = SPI_BFINS(SCBR, scbr, csr);
843 	spi_writel(as, CSR0 + 4 * chip_select, csr);
844 	xfer->effective_speed_hz = bus_hz / scbr;
845 
846 	return 0;
847 }
848 
849 /*
850  * Submit next transfer for PDC.
851  * lock is held, spi irq is blocked
852  */
853 static void atmel_spi_pdc_next_xfer(struct spi_master *master,
854 					struct spi_transfer *xfer)
855 {
856 	struct atmel_spi	*as = spi_master_get_devdata(master);
857 	u32			len;
858 	dma_addr_t		tx_dma, rx_dma;
859 
860 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
861 
862 	len = as->current_remaining_bytes;
863 	atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
864 	as->current_remaining_bytes -= len;
865 
866 	spi_writel(as, RPR, rx_dma);
867 	spi_writel(as, TPR, tx_dma);
868 
869 	if (xfer->bits_per_word > 8)
870 		len >>= 1;
871 	spi_writel(as, RCR, len);
872 	spi_writel(as, TCR, len);
873 
874 	dev_dbg(&master->dev,
875 		"  start xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
876 		xfer, xfer->len, xfer->tx_buf,
877 		(unsigned long long)xfer->tx_dma, xfer->rx_buf,
878 		(unsigned long long)xfer->rx_dma);
879 
880 	if (as->current_remaining_bytes) {
881 		len = as->current_remaining_bytes;
882 		atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
883 		as->current_remaining_bytes -= len;
884 
885 		spi_writel(as, RNPR, rx_dma);
886 		spi_writel(as, TNPR, tx_dma);
887 
888 		if (xfer->bits_per_word > 8)
889 			len >>= 1;
890 		spi_writel(as, RNCR, len);
891 		spi_writel(as, TNCR, len);
892 
893 		dev_dbg(&master->dev,
894 			"  next xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
895 			xfer, xfer->len, xfer->tx_buf,
896 			(unsigned long long)xfer->tx_dma, xfer->rx_buf,
897 			(unsigned long long)xfer->rx_dma);
898 	}
899 
900 	/* REVISIT: We're waiting for RXBUFF before we start the next
901 	 * transfer because we need to handle some difficult timing
902 	 * issues otherwise. If we wait for TXBUFE in one transfer and
903 	 * then starts waiting for RXBUFF in the next, it's difficult
904 	 * to tell the difference between the RXBUFF interrupt we're
905 	 * actually waiting for and the RXBUFF interrupt of the
906 	 * previous transfer.
907 	 *
908 	 * It should be doable, though. Just not now...
909 	 */
910 	spi_writel(as, IER, SPI_BIT(RXBUFF) | SPI_BIT(OVRES));
911 	spi_writel(as, PTCR, SPI_BIT(TXTEN) | SPI_BIT(RXTEN));
912 }
913 
914 /*
915  * For DMA, tx_buf/tx_dma have the same relationship as rx_buf/rx_dma:
916  *  - The buffer is either valid for CPU access, else NULL
917  *  - If the buffer is valid, so is its DMA address
918  *
919  * This driver manages the dma address unless message->is_dma_mapped.
920  */
921 static int
922 atmel_spi_dma_map_xfer(struct atmel_spi *as, struct spi_transfer *xfer)
923 {
924 	struct device	*dev = &as->pdev->dev;
925 
926 	xfer->tx_dma = xfer->rx_dma = INVALID_DMA_ADDRESS;
927 	if (xfer->tx_buf) {
928 		/* tx_buf is a const void* where we need a void * for the dma
929 		 * mapping */
930 		void *nonconst_tx = (void *)xfer->tx_buf;
931 
932 		xfer->tx_dma = dma_map_single(dev,
933 				nonconst_tx, xfer->len,
934 				DMA_TO_DEVICE);
935 		if (dma_mapping_error(dev, xfer->tx_dma))
936 			return -ENOMEM;
937 	}
938 	if (xfer->rx_buf) {
939 		xfer->rx_dma = dma_map_single(dev,
940 				xfer->rx_buf, xfer->len,
941 				DMA_FROM_DEVICE);
942 		if (dma_mapping_error(dev, xfer->rx_dma)) {
943 			if (xfer->tx_buf)
944 				dma_unmap_single(dev,
945 						xfer->tx_dma, xfer->len,
946 						DMA_TO_DEVICE);
947 			return -ENOMEM;
948 		}
949 	}
950 	return 0;
951 }
952 
953 static void atmel_spi_dma_unmap_xfer(struct spi_master *master,
954 				     struct spi_transfer *xfer)
955 {
956 	if (xfer->tx_dma != INVALID_DMA_ADDRESS)
957 		dma_unmap_single(master->dev.parent, xfer->tx_dma,
958 				 xfer->len, DMA_TO_DEVICE);
959 	if (xfer->rx_dma != INVALID_DMA_ADDRESS)
960 		dma_unmap_single(master->dev.parent, xfer->rx_dma,
961 				 xfer->len, DMA_FROM_DEVICE);
962 }
963 
964 static void atmel_spi_disable_pdc_transfer(struct atmel_spi *as)
965 {
966 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
967 }
968 
969 static void
970 atmel_spi_pump_single_data(struct atmel_spi *as, struct spi_transfer *xfer)
971 {
972 	u8		*rxp;
973 	u16		*rxp16;
974 	unsigned long	xfer_pos = xfer->len - as->current_remaining_bytes;
975 
976 	if (xfer->bits_per_word > 8) {
977 		rxp16 = (u16 *)(((u8 *)xfer->rx_buf) + xfer_pos);
978 		*rxp16 = spi_readl(as, RDR);
979 	} else {
980 		rxp = ((u8 *)xfer->rx_buf) + xfer_pos;
981 		*rxp = spi_readl(as, RDR);
982 	}
983 	if (xfer->bits_per_word > 8) {
984 		if (as->current_remaining_bytes > 2)
985 			as->current_remaining_bytes -= 2;
986 		else
987 			as->current_remaining_bytes = 0;
988 	} else {
989 		as->current_remaining_bytes--;
990 	}
991 }
992 
993 static void
994 atmel_spi_pump_fifo_data(struct atmel_spi *as, struct spi_transfer *xfer)
995 {
996 	u32 fifolr = spi_readl(as, FLR);
997 	u32 num_bytes, num_data = SPI_BFEXT(RXFL, fifolr);
998 	u32 offset = xfer->len - as->current_remaining_bytes;
999 	u16 *words = (u16 *)((u8 *)xfer->rx_buf + offset);
1000 	u8  *bytes = (u8  *)((u8 *)xfer->rx_buf + offset);
1001 	u16 rd; /* RD field is the lowest 16 bits of RDR */
1002 
1003 	/* Update the number of remaining bytes to transfer */
1004 	num_bytes = ((xfer->bits_per_word > 8) ?
1005 		     (num_data << 1) :
1006 		     num_data);
1007 
1008 	if (as->current_remaining_bytes > num_bytes)
1009 		as->current_remaining_bytes -= num_bytes;
1010 	else
1011 		as->current_remaining_bytes = 0;
1012 
1013 	/* Handle odd number of bytes when data are more than 8bit width */
1014 	if (xfer->bits_per_word > 8)
1015 		as->current_remaining_bytes &= ~0x1;
1016 
1017 	/* Read data */
1018 	while (num_data) {
1019 		rd = spi_readl(as, RDR);
1020 		if (xfer->bits_per_word > 8)
1021 			*words++ = rd;
1022 		else
1023 			*bytes++ = rd;
1024 		num_data--;
1025 	}
1026 }
1027 
1028 /* Called from IRQ
1029  *
1030  * Must update "current_remaining_bytes" to keep track of data
1031  * to transfer.
1032  */
1033 static void
1034 atmel_spi_pump_pio_data(struct atmel_spi *as, struct spi_transfer *xfer)
1035 {
1036 	if (as->fifo_size)
1037 		atmel_spi_pump_fifo_data(as, xfer);
1038 	else
1039 		atmel_spi_pump_single_data(as, xfer);
1040 }
1041 
1042 /* Interrupt
1043  *
1044  */
1045 static irqreturn_t
1046 atmel_spi_pio_interrupt(int irq, void *dev_id)
1047 {
1048 	struct spi_master	*master = dev_id;
1049 	struct atmel_spi	*as = spi_master_get_devdata(master);
1050 	u32			status, pending, imr;
1051 	struct spi_transfer	*xfer;
1052 	int			ret = IRQ_NONE;
1053 
1054 	imr = spi_readl(as, IMR);
1055 	status = spi_readl(as, SR);
1056 	pending = status & imr;
1057 
1058 	if (pending & SPI_BIT(OVRES)) {
1059 		ret = IRQ_HANDLED;
1060 		spi_writel(as, IDR, SPI_BIT(OVRES));
1061 		dev_warn(master->dev.parent, "overrun\n");
1062 
1063 		/*
1064 		 * When we get an overrun, we disregard the current
1065 		 * transfer. Data will not be copied back from any
1066 		 * bounce buffer and msg->actual_len will not be
1067 		 * updated with the last xfer.
1068 		 *
1069 		 * We will also not process any remaning transfers in
1070 		 * the message.
1071 		 */
1072 		as->done_status = -EIO;
1073 		smp_wmb();
1074 
1075 		/* Clear any overrun happening while cleaning up */
1076 		spi_readl(as, SR);
1077 
1078 		complete(&as->xfer_completion);
1079 
1080 	} else if (pending & (SPI_BIT(RDRF) | SPI_BIT(RXFTHF))) {
1081 		atmel_spi_lock(as);
1082 
1083 		if (as->current_remaining_bytes) {
1084 			ret = IRQ_HANDLED;
1085 			xfer = as->current_transfer;
1086 			atmel_spi_pump_pio_data(as, xfer);
1087 			if (!as->current_remaining_bytes)
1088 				spi_writel(as, IDR, pending);
1089 
1090 			complete(&as->xfer_completion);
1091 		}
1092 
1093 		atmel_spi_unlock(as);
1094 	} else {
1095 		WARN_ONCE(pending, "IRQ not handled, pending = %x\n", pending);
1096 		ret = IRQ_HANDLED;
1097 		spi_writel(as, IDR, pending);
1098 	}
1099 
1100 	return ret;
1101 }
1102 
1103 static irqreturn_t
1104 atmel_spi_pdc_interrupt(int irq, void *dev_id)
1105 {
1106 	struct spi_master	*master = dev_id;
1107 	struct atmel_spi	*as = spi_master_get_devdata(master);
1108 	u32			status, pending, imr;
1109 	int			ret = IRQ_NONE;
1110 
1111 	imr = spi_readl(as, IMR);
1112 	status = spi_readl(as, SR);
1113 	pending = status & imr;
1114 
1115 	if (pending & SPI_BIT(OVRES)) {
1116 
1117 		ret = IRQ_HANDLED;
1118 
1119 		spi_writel(as, IDR, (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX)
1120 				     | SPI_BIT(OVRES)));
1121 
1122 		/* Clear any overrun happening while cleaning up */
1123 		spi_readl(as, SR);
1124 
1125 		as->done_status = -EIO;
1126 
1127 		complete(&as->xfer_completion);
1128 
1129 	} else if (pending & (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX))) {
1130 		ret = IRQ_HANDLED;
1131 
1132 		spi_writel(as, IDR, pending);
1133 
1134 		complete(&as->xfer_completion);
1135 	}
1136 
1137 	return ret;
1138 }
1139 
1140 static int atmel_word_delay_csr(struct spi_device *spi, struct atmel_spi *as)
1141 {
1142 	struct spi_delay *delay = &spi->word_delay;
1143 	u32 value = delay->value;
1144 
1145 	switch (delay->unit) {
1146 	case SPI_DELAY_UNIT_NSECS:
1147 		value /= 1000;
1148 		break;
1149 	case SPI_DELAY_UNIT_USECS:
1150 		break;
1151 	default:
1152 		return -EINVAL;
1153 	}
1154 
1155 	return (as->spi_clk / 1000000 * value) >> 5;
1156 }
1157 
1158 static void initialize_native_cs_for_gpio(struct atmel_spi *as)
1159 {
1160 	int i;
1161 	struct spi_master *master = platform_get_drvdata(as->pdev);
1162 
1163 	if (!as->native_cs_free)
1164 		return; /* already initialized */
1165 
1166 	if (!master->cs_gpiods)
1167 		return; /* No CS GPIO */
1168 
1169 	/*
1170 	 * On the first version of the controller (AT91RM9200), CS0
1171 	 * can't be used associated with GPIO
1172 	 */
1173 	if (atmel_spi_is_v2(as))
1174 		i = 0;
1175 	else
1176 		i = 1;
1177 
1178 	for (; i < 4; i++)
1179 		if (master->cs_gpiods[i])
1180 			as->native_cs_free |= BIT(i);
1181 
1182 	if (as->native_cs_free)
1183 		as->native_cs_for_gpio = ffs(as->native_cs_free);
1184 }
1185 
1186 static int atmel_spi_setup(struct spi_device *spi)
1187 {
1188 	struct atmel_spi	*as;
1189 	struct atmel_spi_device	*asd;
1190 	u32			csr;
1191 	unsigned int		bits = spi->bits_per_word;
1192 	int chip_select;
1193 	int			word_delay_csr;
1194 
1195 	as = spi_master_get_devdata(spi->master);
1196 
1197 	/* see notes above re chipselect */
1198 	if (!spi->cs_gpiod && (spi->mode & SPI_CS_HIGH)) {
1199 		dev_warn(&spi->dev, "setup: non GPIO CS can't be active-high\n");
1200 		return -EINVAL;
1201 	}
1202 
1203 	/* Setup() is called during spi_register_controller(aka
1204 	 * spi_register_master) but after all membmers of the cs_gpiod
1205 	 * array have been filled, so we can looked for which native
1206 	 * CS will be free for using with GPIO
1207 	 */
1208 	initialize_native_cs_for_gpio(as);
1209 
1210 	if (spi->cs_gpiod && as->native_cs_free) {
1211 		dev_err(&spi->dev,
1212 			"No native CS available to support this GPIO CS\n");
1213 		return -EBUSY;
1214 	}
1215 
1216 	if (spi->cs_gpiod)
1217 		chip_select = as->native_cs_for_gpio;
1218 	else
1219 		chip_select = spi->chip_select;
1220 
1221 	csr = SPI_BF(BITS, bits - 8);
1222 	if (spi->mode & SPI_CPOL)
1223 		csr |= SPI_BIT(CPOL);
1224 	if (!(spi->mode & SPI_CPHA))
1225 		csr |= SPI_BIT(NCPHA);
1226 
1227 	if (!spi->cs_gpiod)
1228 		csr |= SPI_BIT(CSAAT);
1229 	csr |= SPI_BF(DLYBS, 0);
1230 
1231 	word_delay_csr = atmel_word_delay_csr(spi, as);
1232 	if (word_delay_csr < 0)
1233 		return word_delay_csr;
1234 
1235 	/* DLYBCT adds delays between words.  This is useful for slow devices
1236 	 * that need a bit of time to setup the next transfer.
1237 	 */
1238 	csr |= SPI_BF(DLYBCT, word_delay_csr);
1239 
1240 	asd = spi->controller_state;
1241 	if (!asd) {
1242 		asd = kzalloc(sizeof(struct atmel_spi_device), GFP_KERNEL);
1243 		if (!asd)
1244 			return -ENOMEM;
1245 
1246 		spi->controller_state = asd;
1247 	}
1248 
1249 	asd->csr = csr;
1250 
1251 	dev_dbg(&spi->dev,
1252 		"setup: bpw %u mode 0x%x -> csr%d %08x\n",
1253 		bits, spi->mode, spi->chip_select, csr);
1254 
1255 	if (!atmel_spi_is_v2(as))
1256 		spi_writel(as, CSR0 + 4 * chip_select, csr);
1257 
1258 	return 0;
1259 }
1260 
1261 static void atmel_spi_set_cs(struct spi_device *spi, bool enable)
1262 {
1263 	struct atmel_spi *as = spi_master_get_devdata(spi->master);
1264 	/* the core doesn't really pass us enable/disable, but CS HIGH vs CS LOW
1265 	 * since we already have routines for activate/deactivate translate
1266 	 * high/low to active/inactive
1267 	 */
1268 	enable = (!!(spi->mode & SPI_CS_HIGH) == enable);
1269 
1270 	if (enable) {
1271 		cs_activate(as, spi);
1272 	} else {
1273 		cs_deactivate(as, spi);
1274 	}
1275 
1276 }
1277 
1278 static int atmel_spi_one_transfer(struct spi_master *master,
1279 					struct spi_device *spi,
1280 					struct spi_transfer *xfer)
1281 {
1282 	struct atmel_spi	*as;
1283 	u8			bits;
1284 	u32			len;
1285 	struct atmel_spi_device	*asd;
1286 	int			timeout;
1287 	int			ret;
1288 	unsigned long		dma_timeout;
1289 
1290 	as = spi_master_get_devdata(master);
1291 
1292 	asd = spi->controller_state;
1293 	bits = (asd->csr >> 4) & 0xf;
1294 	if (bits != xfer->bits_per_word - 8) {
1295 		dev_dbg(&spi->dev,
1296 			"you can't yet change bits_per_word in transfers\n");
1297 		return -ENOPROTOOPT;
1298 	}
1299 
1300 	/*
1301 	 * DMA map early, for performance (empties dcache ASAP) and
1302 	 * better fault reporting.
1303 	 */
1304 	if ((!master->cur_msg_mapped)
1305 		&& as->use_pdc) {
1306 		if (atmel_spi_dma_map_xfer(as, xfer) < 0)
1307 			return -ENOMEM;
1308 	}
1309 
1310 	atmel_spi_set_xfer_speed(as, spi, xfer);
1311 
1312 	as->done_status = 0;
1313 	as->current_transfer = xfer;
1314 	as->current_remaining_bytes = xfer->len;
1315 	while (as->current_remaining_bytes) {
1316 		reinit_completion(&as->xfer_completion);
1317 
1318 		if (as->use_pdc) {
1319 			atmel_spi_lock(as);
1320 			atmel_spi_pdc_next_xfer(master, xfer);
1321 			atmel_spi_unlock(as);
1322 		} else if (atmel_spi_use_dma(as, xfer)) {
1323 			len = as->current_remaining_bytes;
1324 			ret = atmel_spi_next_xfer_dma_submit(master,
1325 								xfer, &len);
1326 			if (ret) {
1327 				dev_err(&spi->dev,
1328 					"unable to use DMA, fallback to PIO\n");
1329 				as->done_status = ret;
1330 				break;
1331 			} else {
1332 				as->current_remaining_bytes -= len;
1333 				if (as->current_remaining_bytes < 0)
1334 					as->current_remaining_bytes = 0;
1335 			}
1336 		} else {
1337 			atmel_spi_lock(as);
1338 			atmel_spi_next_xfer_pio(master, xfer);
1339 			atmel_spi_unlock(as);
1340 		}
1341 
1342 		dma_timeout = wait_for_completion_timeout(&as->xfer_completion,
1343 							  SPI_DMA_TIMEOUT);
1344 		if (WARN_ON(dma_timeout == 0)) {
1345 			dev_err(&spi->dev, "spi transfer timeout\n");
1346 			as->done_status = -EIO;
1347 		}
1348 
1349 		if (as->done_status)
1350 			break;
1351 	}
1352 
1353 	if (as->done_status) {
1354 		if (as->use_pdc) {
1355 			dev_warn(master->dev.parent,
1356 				"overrun (%u/%u remaining)\n",
1357 				spi_readl(as, TCR), spi_readl(as, RCR));
1358 
1359 			/*
1360 			 * Clean up DMA registers and make sure the data
1361 			 * registers are empty.
1362 			 */
1363 			spi_writel(as, RNCR, 0);
1364 			spi_writel(as, TNCR, 0);
1365 			spi_writel(as, RCR, 0);
1366 			spi_writel(as, TCR, 0);
1367 			for (timeout = 1000; timeout; timeout--)
1368 				if (spi_readl(as, SR) & SPI_BIT(TXEMPTY))
1369 					break;
1370 			if (!timeout)
1371 				dev_warn(master->dev.parent,
1372 					 "timeout waiting for TXEMPTY");
1373 			while (spi_readl(as, SR) & SPI_BIT(RDRF))
1374 				spi_readl(as, RDR);
1375 
1376 			/* Clear any overrun happening while cleaning up */
1377 			spi_readl(as, SR);
1378 
1379 		} else if (atmel_spi_use_dma(as, xfer)) {
1380 			atmel_spi_stop_dma(master);
1381 		}
1382 	}
1383 
1384 	if (!master->cur_msg_mapped
1385 		&& as->use_pdc)
1386 		atmel_spi_dma_unmap_xfer(master, xfer);
1387 
1388 	if (as->use_pdc)
1389 		atmel_spi_disable_pdc_transfer(as);
1390 
1391 	return as->done_status;
1392 }
1393 
1394 static void atmel_spi_cleanup(struct spi_device *spi)
1395 {
1396 	struct atmel_spi_device	*asd = spi->controller_state;
1397 
1398 	if (!asd)
1399 		return;
1400 
1401 	spi->controller_state = NULL;
1402 	kfree(asd);
1403 }
1404 
1405 static inline unsigned int atmel_get_version(struct atmel_spi *as)
1406 {
1407 	return spi_readl(as, VERSION) & 0x00000fff;
1408 }
1409 
1410 static void atmel_get_caps(struct atmel_spi *as)
1411 {
1412 	unsigned int version;
1413 
1414 	version = atmel_get_version(as);
1415 
1416 	as->caps.is_spi2 = version > 0x121;
1417 	as->caps.has_wdrbt = version >= 0x210;
1418 	as->caps.has_dma_support = version >= 0x212;
1419 	as->caps.has_pdc_support = version < 0x212;
1420 }
1421 
1422 static void atmel_spi_init(struct atmel_spi *as)
1423 {
1424 	spi_writel(as, CR, SPI_BIT(SWRST));
1425 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1426 
1427 	/* It is recommended to enable FIFOs first thing after reset */
1428 	if (as->fifo_size)
1429 		spi_writel(as, CR, SPI_BIT(FIFOEN));
1430 
1431 	if (as->caps.has_wdrbt) {
1432 		spi_writel(as, MR, SPI_BIT(WDRBT) | SPI_BIT(MODFDIS)
1433 				| SPI_BIT(MSTR));
1434 	} else {
1435 		spi_writel(as, MR, SPI_BIT(MSTR) | SPI_BIT(MODFDIS));
1436 	}
1437 
1438 	if (as->use_pdc)
1439 		spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1440 	spi_writel(as, CR, SPI_BIT(SPIEN));
1441 }
1442 
1443 static int atmel_spi_probe(struct platform_device *pdev)
1444 {
1445 	struct resource		*regs;
1446 	int			irq;
1447 	struct clk		*clk;
1448 	int			ret;
1449 	struct spi_master	*master;
1450 	struct atmel_spi	*as;
1451 
1452 	/* Select default pin state */
1453 	pinctrl_pm_select_default_state(&pdev->dev);
1454 
1455 	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1456 	if (!regs)
1457 		return -ENXIO;
1458 
1459 	irq = platform_get_irq(pdev, 0);
1460 	if (irq < 0)
1461 		return irq;
1462 
1463 	clk = devm_clk_get(&pdev->dev, "spi_clk");
1464 	if (IS_ERR(clk))
1465 		return PTR_ERR(clk);
1466 
1467 	/* setup spi core then atmel-specific driver state */
1468 	master = spi_alloc_master(&pdev->dev, sizeof(*as));
1469 	if (!master)
1470 		return -ENOMEM;
1471 
1472 	/* the spi->mode bits understood by this driver: */
1473 	master->use_gpio_descriptors = true;
1474 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
1475 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(8, 16);
1476 	master->dev.of_node = pdev->dev.of_node;
1477 	master->bus_num = pdev->id;
1478 	master->num_chipselect = 4;
1479 	master->setup = atmel_spi_setup;
1480 	master->flags = (SPI_MASTER_MUST_RX | SPI_MASTER_MUST_TX |
1481 			SPI_MASTER_GPIO_SS);
1482 	master->transfer_one = atmel_spi_one_transfer;
1483 	master->set_cs = atmel_spi_set_cs;
1484 	master->cleanup = atmel_spi_cleanup;
1485 	master->auto_runtime_pm = true;
1486 	master->max_dma_len = SPI_MAX_DMA_XFER;
1487 	master->can_dma = atmel_spi_can_dma;
1488 	platform_set_drvdata(pdev, master);
1489 
1490 	as = spi_master_get_devdata(master);
1491 
1492 	spin_lock_init(&as->lock);
1493 
1494 	as->pdev = pdev;
1495 	as->regs = devm_ioremap_resource(&pdev->dev, regs);
1496 	if (IS_ERR(as->regs)) {
1497 		ret = PTR_ERR(as->regs);
1498 		goto out_unmap_regs;
1499 	}
1500 	as->phybase = regs->start;
1501 	as->irq = irq;
1502 	as->clk = clk;
1503 
1504 	init_completion(&as->xfer_completion);
1505 
1506 	atmel_get_caps(as);
1507 
1508 	as->use_dma = false;
1509 	as->use_pdc = false;
1510 	if (as->caps.has_dma_support) {
1511 		ret = atmel_spi_configure_dma(master, as);
1512 		if (ret == 0) {
1513 			as->use_dma = true;
1514 		} else if (ret == -EPROBE_DEFER) {
1515 			goto out_unmap_regs;
1516 		}
1517 	} else if (as->caps.has_pdc_support) {
1518 		as->use_pdc = true;
1519 	}
1520 
1521 	if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
1522 		as->addr_rx_bbuf = dma_alloc_coherent(&pdev->dev,
1523 						      SPI_MAX_DMA_XFER,
1524 						      &as->dma_addr_rx_bbuf,
1525 						      GFP_KERNEL | GFP_DMA);
1526 		if (!as->addr_rx_bbuf) {
1527 			as->use_dma = false;
1528 		} else {
1529 			as->addr_tx_bbuf = dma_alloc_coherent(&pdev->dev,
1530 					SPI_MAX_DMA_XFER,
1531 					&as->dma_addr_tx_bbuf,
1532 					GFP_KERNEL | GFP_DMA);
1533 			if (!as->addr_tx_bbuf) {
1534 				as->use_dma = false;
1535 				dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1536 						  as->addr_rx_bbuf,
1537 						  as->dma_addr_rx_bbuf);
1538 			}
1539 		}
1540 		if (!as->use_dma)
1541 			dev_info(master->dev.parent,
1542 				 "  can not allocate dma coherent memory\n");
1543 	}
1544 
1545 	if (as->caps.has_dma_support && !as->use_dma)
1546 		dev_info(&pdev->dev, "Atmel SPI Controller using PIO only\n");
1547 
1548 	if (as->use_pdc) {
1549 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pdc_interrupt,
1550 					0, dev_name(&pdev->dev), master);
1551 	} else {
1552 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pio_interrupt,
1553 					0, dev_name(&pdev->dev), master);
1554 	}
1555 	if (ret)
1556 		goto out_unmap_regs;
1557 
1558 	/* Initialize the hardware */
1559 	ret = clk_prepare_enable(clk);
1560 	if (ret)
1561 		goto out_free_irq;
1562 
1563 	as->spi_clk = clk_get_rate(clk);
1564 
1565 	as->fifo_size = 0;
1566 	if (!of_property_read_u32(pdev->dev.of_node, "atmel,fifo-size",
1567 				  &as->fifo_size)) {
1568 		dev_info(&pdev->dev, "Using FIFO (%u data)\n", as->fifo_size);
1569 	}
1570 
1571 	atmel_spi_init(as);
1572 
1573 	pm_runtime_set_autosuspend_delay(&pdev->dev, AUTOSUSPEND_TIMEOUT);
1574 	pm_runtime_use_autosuspend(&pdev->dev);
1575 	pm_runtime_set_active(&pdev->dev);
1576 	pm_runtime_enable(&pdev->dev);
1577 
1578 	ret = devm_spi_register_master(&pdev->dev, master);
1579 	if (ret)
1580 		goto out_free_dma;
1581 
1582 	/* go! */
1583 	dev_info(&pdev->dev, "Atmel SPI Controller version 0x%x at 0x%08lx (irq %d)\n",
1584 			atmel_get_version(as), (unsigned long)regs->start,
1585 			irq);
1586 
1587 	return 0;
1588 
1589 out_free_dma:
1590 	pm_runtime_disable(&pdev->dev);
1591 	pm_runtime_set_suspended(&pdev->dev);
1592 
1593 	if (as->use_dma)
1594 		atmel_spi_release_dma(master);
1595 
1596 	spi_writel(as, CR, SPI_BIT(SWRST));
1597 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1598 	clk_disable_unprepare(clk);
1599 out_free_irq:
1600 out_unmap_regs:
1601 	spi_master_put(master);
1602 	return ret;
1603 }
1604 
1605 static int atmel_spi_remove(struct platform_device *pdev)
1606 {
1607 	struct spi_master	*master = platform_get_drvdata(pdev);
1608 	struct atmel_spi	*as = spi_master_get_devdata(master);
1609 
1610 	pm_runtime_get_sync(&pdev->dev);
1611 
1612 	/* reset the hardware and block queue progress */
1613 	if (as->use_dma) {
1614 		atmel_spi_stop_dma(master);
1615 		atmel_spi_release_dma(master);
1616 		if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
1617 			dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1618 					  as->addr_tx_bbuf,
1619 					  as->dma_addr_tx_bbuf);
1620 			dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1621 					  as->addr_rx_bbuf,
1622 					  as->dma_addr_rx_bbuf);
1623 		}
1624 	}
1625 
1626 	spin_lock_irq(&as->lock);
1627 	spi_writel(as, CR, SPI_BIT(SWRST));
1628 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1629 	spi_readl(as, SR);
1630 	spin_unlock_irq(&as->lock);
1631 
1632 	clk_disable_unprepare(as->clk);
1633 
1634 	pm_runtime_put_noidle(&pdev->dev);
1635 	pm_runtime_disable(&pdev->dev);
1636 
1637 	return 0;
1638 }
1639 
1640 #ifdef CONFIG_PM
1641 static int atmel_spi_runtime_suspend(struct device *dev)
1642 {
1643 	struct spi_master *master = dev_get_drvdata(dev);
1644 	struct atmel_spi *as = spi_master_get_devdata(master);
1645 
1646 	clk_disable_unprepare(as->clk);
1647 	pinctrl_pm_select_sleep_state(dev);
1648 
1649 	return 0;
1650 }
1651 
1652 static int atmel_spi_runtime_resume(struct device *dev)
1653 {
1654 	struct spi_master *master = dev_get_drvdata(dev);
1655 	struct atmel_spi *as = spi_master_get_devdata(master);
1656 
1657 	pinctrl_pm_select_default_state(dev);
1658 
1659 	return clk_prepare_enable(as->clk);
1660 }
1661 
1662 #ifdef CONFIG_PM_SLEEP
1663 static int atmel_spi_suspend(struct device *dev)
1664 {
1665 	struct spi_master *master = dev_get_drvdata(dev);
1666 	int ret;
1667 
1668 	/* Stop the queue running */
1669 	ret = spi_master_suspend(master);
1670 	if (ret)
1671 		return ret;
1672 
1673 	if (!pm_runtime_suspended(dev))
1674 		atmel_spi_runtime_suspend(dev);
1675 
1676 	return 0;
1677 }
1678 
1679 static int atmel_spi_resume(struct device *dev)
1680 {
1681 	struct spi_master *master = dev_get_drvdata(dev);
1682 	struct atmel_spi *as = spi_master_get_devdata(master);
1683 	int ret;
1684 
1685 	ret = clk_prepare_enable(as->clk);
1686 	if (ret)
1687 		return ret;
1688 
1689 	atmel_spi_init(as);
1690 
1691 	clk_disable_unprepare(as->clk);
1692 
1693 	if (!pm_runtime_suspended(dev)) {
1694 		ret = atmel_spi_runtime_resume(dev);
1695 		if (ret)
1696 			return ret;
1697 	}
1698 
1699 	/* Start the queue running */
1700 	return spi_master_resume(master);
1701 }
1702 #endif
1703 
1704 static const struct dev_pm_ops atmel_spi_pm_ops = {
1705 	SET_SYSTEM_SLEEP_PM_OPS(atmel_spi_suspend, atmel_spi_resume)
1706 	SET_RUNTIME_PM_OPS(atmel_spi_runtime_suspend,
1707 			   atmel_spi_runtime_resume, NULL)
1708 };
1709 #define ATMEL_SPI_PM_OPS	(&atmel_spi_pm_ops)
1710 #else
1711 #define ATMEL_SPI_PM_OPS	NULL
1712 #endif
1713 
1714 static const struct of_device_id atmel_spi_dt_ids[] = {
1715 	{ .compatible = "atmel,at91rm9200-spi" },
1716 	{ /* sentinel */ }
1717 };
1718 
1719 MODULE_DEVICE_TABLE(of, atmel_spi_dt_ids);
1720 
1721 static struct platform_driver atmel_spi_driver = {
1722 	.driver		= {
1723 		.name	= "atmel_spi",
1724 		.pm	= ATMEL_SPI_PM_OPS,
1725 		.of_match_table	= atmel_spi_dt_ids,
1726 	},
1727 	.probe		= atmel_spi_probe,
1728 	.remove		= atmel_spi_remove,
1729 };
1730 module_platform_driver(atmel_spi_driver);
1731 
1732 MODULE_DESCRIPTION("Atmel AT32/AT91 SPI Controller driver");
1733 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1734 MODULE_LICENSE("GPL");
1735 MODULE_ALIAS("platform:atmel_spi");
1736