xref: /linux/drivers/spi/spi-atmel.c (revision 0c7c237b1c35011ef0b8d30c1d5c20bc6ae7b69b)
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_MIN_TIMEOUT	(msecs_to_jiffies(1000))
237 #define SPI_DMA_TIMEOUT_PER_10K	(msecs_to_jiffies(4))
238 
239 #define AUTOSUSPEND_TIMEOUT	2000
240 
241 struct atmel_spi_caps {
242 	bool	is_spi2;
243 	bool	has_wdrbt;
244 	bool	has_dma_support;
245 	bool	has_pdc_support;
246 };
247 
248 /*
249  * The core SPI transfer engine just talks to a register bank to set up
250  * DMA transfers; transfer queue progress is driven by IRQs.  The clock
251  * framework provides the base clock, subdivided for each spi_device.
252  */
253 struct atmel_spi {
254 	spinlock_t		lock;
255 	unsigned long		flags;
256 
257 	phys_addr_t		phybase;
258 	void __iomem		*regs;
259 	int			irq;
260 	struct clk		*clk;
261 	struct platform_device	*pdev;
262 	unsigned long		spi_clk;
263 
264 	struct spi_transfer	*current_transfer;
265 	int			current_remaining_bytes;
266 	int			done_status;
267 	dma_addr_t		dma_addr_rx_bbuf;
268 	dma_addr_t		dma_addr_tx_bbuf;
269 	void			*addr_rx_bbuf;
270 	void			*addr_tx_bbuf;
271 
272 	struct completion	xfer_completion;
273 
274 	struct atmel_spi_caps	caps;
275 
276 	bool			use_dma;
277 	bool			use_pdc;
278 
279 	bool			keep_cs;
280 
281 	u32			fifo_size;
282 	u8			native_cs_free;
283 	u8			native_cs_for_gpio;
284 };
285 
286 /* Controller-specific per-slave state */
287 struct atmel_spi_device {
288 	u32			csr;
289 };
290 
291 #define SPI_MAX_DMA_XFER	65535 /* true for both PDC and DMA */
292 #define INVALID_DMA_ADDRESS	0xffffffff
293 
294 /*
295  * Version 2 of the SPI controller has
296  *  - CR.LASTXFER
297  *  - SPI_MR.DIV32 may become FDIV or must-be-zero (here: always zero)
298  *  - SPI_SR.TXEMPTY, SPI_SR.NSSR (and corresponding irqs)
299  *  - SPI_CSRx.CSAAT
300  *  - SPI_CSRx.SBCR allows faster clocking
301  */
302 static bool atmel_spi_is_v2(struct atmel_spi *as)
303 {
304 	return as->caps.is_spi2;
305 }
306 
307 /*
308  * Earlier SPI controllers (e.g. on at91rm9200) have a design bug whereby
309  * they assume that spi slave device state will not change on deselect, so
310  * that automagic deselection is OK.  ("NPCSx rises if no data is to be
311  * transmitted")  Not so!  Workaround uses nCSx pins as GPIOs; or newer
312  * controllers have CSAAT and friends.
313  *
314  * Even controller newer than ar91rm9200, using GPIOs can make sens as
315  * it lets us support active-high chipselects despite the controller's
316  * belief that only active-low devices/systems exists.
317  *
318  * However, at91rm9200 has a second erratum whereby nCS0 doesn't work
319  * right when driven with GPIO.  ("Mode Fault does not allow more than one
320  * Master on Chip Select 0.")  No workaround exists for that ... so for
321  * nCS0 on that chip, we (a) don't use the GPIO, (b) can't support CS_HIGH,
322  * and (c) will trigger that first erratum in some cases.
323  */
324 
325 static void cs_activate(struct atmel_spi *as, struct spi_device *spi)
326 {
327 	struct atmel_spi_device *asd = spi->controller_state;
328 	int chip_select;
329 	u32 mr;
330 
331 	if (spi_get_csgpiod(spi, 0))
332 		chip_select = as->native_cs_for_gpio;
333 	else
334 		chip_select = spi_get_chipselect(spi, 0);
335 
336 	if (atmel_spi_is_v2(as)) {
337 		spi_writel(as, CSR0 + 4 * chip_select, asd->csr);
338 		/* For the low SPI version, there is a issue that PDC transfer
339 		 * on CS1,2,3 needs SPI_CSR0.BITS config as SPI_CSR1,2,3.BITS
340 		 */
341 		spi_writel(as, CSR0, asd->csr);
342 		if (as->caps.has_wdrbt) {
343 			spi_writel(as, MR,
344 					SPI_BF(PCS, ~(0x01 << chip_select))
345 					| SPI_BIT(WDRBT)
346 					| SPI_BIT(MODFDIS)
347 					| SPI_BIT(MSTR));
348 		} else {
349 			spi_writel(as, MR,
350 					SPI_BF(PCS, ~(0x01 << chip_select))
351 					| SPI_BIT(MODFDIS)
352 					| SPI_BIT(MSTR));
353 		}
354 
355 		mr = spi_readl(as, MR);
356 	} else {
357 		u32 cpol = (spi->mode & SPI_CPOL) ? SPI_BIT(CPOL) : 0;
358 		int i;
359 		u32 csr;
360 
361 		/* Make sure clock polarity is correct */
362 		for (i = 0; i < spi->controller->num_chipselect; i++) {
363 			csr = spi_readl(as, CSR0 + 4 * i);
364 			if ((csr ^ cpol) & SPI_BIT(CPOL))
365 				spi_writel(as, CSR0 + 4 * i,
366 						csr ^ SPI_BIT(CPOL));
367 		}
368 
369 		mr = spi_readl(as, MR);
370 		mr = SPI_BFINS(PCS, ~(1 << chip_select), mr);
371 		spi_writel(as, MR, mr);
372 	}
373 
374 	dev_dbg(&spi->dev, "activate NPCS, mr %08x\n", mr);
375 }
376 
377 static void cs_deactivate(struct atmel_spi *as, struct spi_device *spi)
378 {
379 	int chip_select;
380 	u32 mr;
381 
382 	if (spi_get_csgpiod(spi, 0))
383 		chip_select = as->native_cs_for_gpio;
384 	else
385 		chip_select = spi_get_chipselect(spi, 0);
386 
387 	/* only deactivate *this* device; sometimes transfers to
388 	 * another device may be active when this routine is called.
389 	 */
390 	mr = spi_readl(as, MR);
391 	if (~SPI_BFEXT(PCS, mr) & (1 << chip_select)) {
392 		mr = SPI_BFINS(PCS, 0xf, mr);
393 		spi_writel(as, MR, mr);
394 	}
395 
396 	dev_dbg(&spi->dev, "DEactivate NPCS, mr %08x\n", mr);
397 
398 	if (!spi_get_csgpiod(spi, 0))
399 		spi_writel(as, CR, SPI_BIT(LASTXFER));
400 }
401 
402 static void atmel_spi_lock(struct atmel_spi *as) __acquires(&as->lock)
403 {
404 	spin_lock_irqsave(&as->lock, as->flags);
405 }
406 
407 static void atmel_spi_unlock(struct atmel_spi *as) __releases(&as->lock)
408 {
409 	spin_unlock_irqrestore(&as->lock, as->flags);
410 }
411 
412 static inline bool atmel_spi_is_vmalloc_xfer(struct spi_transfer *xfer)
413 {
414 	return is_vmalloc_addr(xfer->tx_buf) || is_vmalloc_addr(xfer->rx_buf);
415 }
416 
417 static inline bool atmel_spi_use_dma(struct atmel_spi *as,
418 				struct spi_transfer *xfer)
419 {
420 	return as->use_dma && xfer->len >= DMA_MIN_BYTES;
421 }
422 
423 static bool atmel_spi_can_dma(struct spi_controller *host,
424 			      struct spi_device *spi,
425 			      struct spi_transfer *xfer)
426 {
427 	struct atmel_spi *as = spi_controller_get_devdata(host);
428 
429 	if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5))
430 		return atmel_spi_use_dma(as, xfer) &&
431 			!atmel_spi_is_vmalloc_xfer(xfer);
432 	else
433 		return atmel_spi_use_dma(as, xfer);
434 
435 }
436 
437 static int atmel_spi_dma_slave_config(struct atmel_spi *as, u8 bits_per_word)
438 {
439 	struct spi_controller *host = platform_get_drvdata(as->pdev);
440 	struct dma_slave_config	slave_config;
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 actually 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 	if (dmaengine_slave_config(host->dma_tx, &slave_config)) {
472 		dev_err(&as->pdev->dev,
473 			"failed to configure tx dma channel\n");
474 		err = -EINVAL;
475 	}
476 
477 	/*
478 	 * This driver configures the spi controller for host mode (MSTR bit
479 	 * set to '1' in the Mode Register).
480 	 * So according to the datasheet, when FIFOs are available (and
481 	 * enabled), the Receive FIFO operates in Single Data Mode.
482 	 * So the receive path works the same whether FIFOs are available (and
483 	 * enabled) or not.
484 	 */
485 	if (dmaengine_slave_config(host->dma_rx, &slave_config)) {
486 		dev_err(&as->pdev->dev,
487 			"failed to configure rx dma channel\n");
488 		err = -EINVAL;
489 	}
490 
491 	return err;
492 }
493 
494 static int atmel_spi_configure_dma(struct spi_controller *host,
495 				   struct atmel_spi *as)
496 {
497 	struct device *dev = &as->pdev->dev;
498 	int err;
499 
500 	host->dma_tx = dma_request_chan(dev, "tx");
501 	if (IS_ERR(host->dma_tx)) {
502 		err = PTR_ERR(host->dma_tx);
503 		dev_dbg(dev, "No TX DMA channel, DMA is disabled\n");
504 		goto error_clear;
505 	}
506 
507 	host->dma_rx = dma_request_chan(dev, "rx");
508 	if (IS_ERR(host->dma_rx)) {
509 		err = PTR_ERR(host->dma_rx);
510 		/*
511 		 * No reason to check EPROBE_DEFER here since we have already
512 		 * requested tx channel.
513 		 */
514 		dev_dbg(dev, "No RX DMA channel, DMA is disabled\n");
515 		goto error;
516 	}
517 
518 	err = atmel_spi_dma_slave_config(as, 8);
519 	if (err)
520 		goto error;
521 
522 	dev_info(&as->pdev->dev,
523 			"Using %s (tx) and %s (rx) for DMA transfers\n",
524 			dma_chan_name(host->dma_tx),
525 			dma_chan_name(host->dma_rx));
526 
527 	return 0;
528 error:
529 	if (!IS_ERR(host->dma_rx))
530 		dma_release_channel(host->dma_rx);
531 	if (!IS_ERR(host->dma_tx))
532 		dma_release_channel(host->dma_tx);
533 error_clear:
534 	host->dma_tx = host->dma_rx = NULL;
535 	return err;
536 }
537 
538 static void atmel_spi_stop_dma(struct spi_controller *host)
539 {
540 	if (host->dma_rx)
541 		dmaengine_terminate_all(host->dma_rx);
542 	if (host->dma_tx)
543 		dmaengine_terminate_all(host->dma_tx);
544 }
545 
546 static void atmel_spi_release_dma(struct spi_controller *host)
547 {
548 	if (host->dma_rx) {
549 		dma_release_channel(host->dma_rx);
550 		host->dma_rx = NULL;
551 	}
552 	if (host->dma_tx) {
553 		dma_release_channel(host->dma_tx);
554 		host->dma_tx = NULL;
555 	}
556 }
557 
558 /* This function is called by the DMA driver from tasklet context */
559 static void dma_callback(void *data)
560 {
561 	struct spi_controller	*host = data;
562 	struct atmel_spi	*as = spi_controller_get_devdata(host);
563 
564 	if (is_vmalloc_addr(as->current_transfer->rx_buf) &&
565 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
566 		memcpy(as->current_transfer->rx_buf, as->addr_rx_bbuf,
567 		       as->current_transfer->len);
568 	}
569 	complete(&as->xfer_completion);
570 }
571 
572 /*
573  * Next transfer using PIO without FIFO.
574  */
575 static void atmel_spi_next_xfer_single(struct spi_controller *host,
576 				       struct spi_transfer *xfer)
577 {
578 	struct atmel_spi	*as = spi_controller_get_devdata(host);
579 	unsigned long xfer_pos = xfer->len - as->current_remaining_bytes;
580 
581 	dev_vdbg(host->dev.parent, "atmel_spi_next_xfer_pio\n");
582 
583 	/* Make sure data is not remaining in RDR */
584 	spi_readl(as, RDR);
585 	while (spi_readl(as, SR) & SPI_BIT(RDRF)) {
586 		spi_readl(as, RDR);
587 		cpu_relax();
588 	}
589 
590 	if (xfer->bits_per_word > 8)
591 		spi_writel(as, TDR, *(u16 *)(xfer->tx_buf + xfer_pos));
592 	else
593 		spi_writel(as, TDR, *(u8 *)(xfer->tx_buf + xfer_pos));
594 
595 	dev_dbg(host->dev.parent,
596 		"  start pio xfer %p: len %u tx %p rx %p bitpw %d\n",
597 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
598 		xfer->bits_per_word);
599 
600 	/* Enable relevant interrupts */
601 	spi_writel(as, IER, SPI_BIT(RDRF) | SPI_BIT(OVRES));
602 }
603 
604 /*
605  * Next transfer using PIO with FIFO.
606  */
607 static void atmel_spi_next_xfer_fifo(struct spi_controller *host,
608 				     struct spi_transfer *xfer)
609 {
610 	struct atmel_spi *as = spi_controller_get_devdata(host);
611 	u32 current_remaining_data, num_data;
612 	u32 offset = xfer->len - as->current_remaining_bytes;
613 	const u16 *words = (const u16 *)((u8 *)xfer->tx_buf + offset);
614 	const u8  *bytes = (const u8  *)((u8 *)xfer->tx_buf + offset);
615 	u16 td0, td1;
616 	u32 fifomr;
617 
618 	dev_vdbg(host->dev.parent, "atmel_spi_next_xfer_fifo\n");
619 
620 	/* Compute the number of data to transfer in the current iteration */
621 	current_remaining_data = ((xfer->bits_per_word > 8) ?
622 				  ((u32)as->current_remaining_bytes >> 1) :
623 				  (u32)as->current_remaining_bytes);
624 	num_data = min(current_remaining_data, as->fifo_size);
625 
626 	/* Flush RX and TX FIFOs */
627 	spi_writel(as, CR, SPI_BIT(RXFCLR) | SPI_BIT(TXFCLR));
628 	while (spi_readl(as, FLR))
629 		cpu_relax();
630 
631 	/* Set RX FIFO Threshold to the number of data to transfer */
632 	fifomr = spi_readl(as, FMR);
633 	spi_writel(as, FMR, SPI_BFINS(RXFTHRES, num_data, fifomr));
634 
635 	/* Clear FIFO flags in the Status Register, especially RXFTHF */
636 	(void)spi_readl(as, SR);
637 
638 	/* Fill TX FIFO */
639 	while (num_data >= 2) {
640 		if (xfer->bits_per_word > 8) {
641 			td0 = *words++;
642 			td1 = *words++;
643 		} else {
644 			td0 = *bytes++;
645 			td1 = *bytes++;
646 		}
647 
648 		spi_writel(as, TDR, (td1 << 16) | td0);
649 		num_data -= 2;
650 	}
651 
652 	if (num_data) {
653 		if (xfer->bits_per_word > 8)
654 			td0 = *words++;
655 		else
656 			td0 = *bytes++;
657 
658 		spi_writew(as, TDR, td0);
659 		num_data--;
660 	}
661 
662 	dev_dbg(host->dev.parent,
663 		"  start fifo xfer %p: len %u tx %p rx %p bitpw %d\n",
664 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
665 		xfer->bits_per_word);
666 
667 	/*
668 	 * Enable RX FIFO Threshold Flag interrupt to be notified about
669 	 * transfer completion.
670 	 */
671 	spi_writel(as, IER, SPI_BIT(RXFTHF) | SPI_BIT(OVRES));
672 }
673 
674 /*
675  * Next transfer using PIO.
676  */
677 static void atmel_spi_next_xfer_pio(struct spi_controller *host,
678 				    struct spi_transfer *xfer)
679 {
680 	struct atmel_spi *as = spi_controller_get_devdata(host);
681 
682 	if (as->fifo_size)
683 		atmel_spi_next_xfer_fifo(host, xfer);
684 	else
685 		atmel_spi_next_xfer_single(host, xfer);
686 }
687 
688 /*
689  * Submit next transfer for DMA.
690  */
691 static int atmel_spi_next_xfer_dma_submit(struct spi_controller *host,
692 				struct spi_transfer *xfer,
693 				u32 *plen)
694 {
695 	struct atmel_spi	*as = spi_controller_get_devdata(host);
696 	struct dma_chan		*rxchan = host->dma_rx;
697 	struct dma_chan		*txchan = host->dma_tx;
698 	struct dma_async_tx_descriptor *rxdesc;
699 	struct dma_async_tx_descriptor *txdesc;
700 	dma_cookie_t		cookie;
701 
702 	dev_vdbg(host->dev.parent, "atmel_spi_next_xfer_dma_submit\n");
703 
704 	/* Check that the channels are available */
705 	if (!rxchan || !txchan)
706 		return -ENODEV;
707 
708 
709 	*plen = xfer->len;
710 
711 	if (atmel_spi_dma_slave_config(as, xfer->bits_per_word))
712 		goto err_exit;
713 
714 	/* Send both scatterlists */
715 	if (atmel_spi_is_vmalloc_xfer(xfer) &&
716 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
717 		rxdesc = dmaengine_prep_slave_single(rxchan,
718 						     as->dma_addr_rx_bbuf,
719 						     xfer->len,
720 						     DMA_DEV_TO_MEM,
721 						     DMA_PREP_INTERRUPT |
722 						     DMA_CTRL_ACK);
723 	} else {
724 		rxdesc = dmaengine_prep_slave_sg(rxchan,
725 						 xfer->rx_sg.sgl,
726 						 xfer->rx_sg.nents,
727 						 DMA_DEV_TO_MEM,
728 						 DMA_PREP_INTERRUPT |
729 						 DMA_CTRL_ACK);
730 	}
731 	if (!rxdesc)
732 		goto err_dma;
733 
734 	if (atmel_spi_is_vmalloc_xfer(xfer) &&
735 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
736 		memcpy(as->addr_tx_bbuf, xfer->tx_buf, xfer->len);
737 		txdesc = dmaengine_prep_slave_single(txchan,
738 						     as->dma_addr_tx_bbuf,
739 						     xfer->len, DMA_MEM_TO_DEV,
740 						     DMA_PREP_INTERRUPT |
741 						     DMA_CTRL_ACK);
742 	} else {
743 		txdesc = dmaengine_prep_slave_sg(txchan,
744 						 xfer->tx_sg.sgl,
745 						 xfer->tx_sg.nents,
746 						 DMA_MEM_TO_DEV,
747 						 DMA_PREP_INTERRUPT |
748 						 DMA_CTRL_ACK);
749 	}
750 	if (!txdesc)
751 		goto err_dma;
752 
753 	dev_dbg(host->dev.parent,
754 		"  start dma xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
755 		xfer, xfer->len, xfer->tx_buf, (unsigned long long)xfer->tx_dma,
756 		xfer->rx_buf, (unsigned long long)xfer->rx_dma);
757 
758 	/* Enable relevant interrupts */
759 	spi_writel(as, IER, SPI_BIT(OVRES));
760 
761 	/* Put the callback on the RX transfer only, that should finish last */
762 	rxdesc->callback = dma_callback;
763 	rxdesc->callback_param = host;
764 
765 	/* Submit and fire RX and TX with TX last so we're ready to read! */
766 	cookie = rxdesc->tx_submit(rxdesc);
767 	if (dma_submit_error(cookie))
768 		goto err_dma;
769 	cookie = txdesc->tx_submit(txdesc);
770 	if (dma_submit_error(cookie))
771 		goto err_dma;
772 	rxchan->device->device_issue_pending(rxchan);
773 	txchan->device->device_issue_pending(txchan);
774 
775 	return 0;
776 
777 err_dma:
778 	spi_writel(as, IDR, SPI_BIT(OVRES));
779 	atmel_spi_stop_dma(host);
780 err_exit:
781 	return -ENOMEM;
782 }
783 
784 static void atmel_spi_next_xfer_data(struct spi_controller *host,
785 				struct spi_transfer *xfer,
786 				dma_addr_t *tx_dma,
787 				dma_addr_t *rx_dma,
788 				u32 *plen)
789 {
790 	*rx_dma = xfer->rx_dma + xfer->len - *plen;
791 	*tx_dma = xfer->tx_dma + xfer->len - *plen;
792 	if (*plen > host->max_dma_len)
793 		*plen = host->max_dma_len;
794 }
795 
796 static int atmel_spi_set_xfer_speed(struct atmel_spi *as,
797 				    struct spi_device *spi,
798 				    struct spi_transfer *xfer)
799 {
800 	u32			scbr, csr;
801 	unsigned long		bus_hz;
802 	int chip_select;
803 
804 	if (spi_get_csgpiod(spi, 0))
805 		chip_select = as->native_cs_for_gpio;
806 	else
807 		chip_select = spi_get_chipselect(spi, 0);
808 
809 	/* v1 chips start out at half the peripheral bus speed. */
810 	bus_hz = as->spi_clk;
811 	if (!atmel_spi_is_v2(as))
812 		bus_hz /= 2;
813 
814 	/*
815 	 * Calculate the lowest divider that satisfies the
816 	 * constraint, assuming div32/fdiv/mbz == 0.
817 	 */
818 	scbr = DIV_ROUND_UP(bus_hz, xfer->speed_hz);
819 
820 	/*
821 	 * If the resulting divider doesn't fit into the
822 	 * register bitfield, we can't satisfy the constraint.
823 	 */
824 	if (scbr >= (1 << SPI_SCBR_SIZE)) {
825 		dev_err(&spi->dev,
826 			"setup: %d Hz too slow, scbr %u; min %ld Hz\n",
827 			xfer->speed_hz, scbr, bus_hz/255);
828 		return -EINVAL;
829 	}
830 	if (scbr == 0) {
831 		dev_err(&spi->dev,
832 			"setup: %d Hz too high, scbr %u; max %ld Hz\n",
833 			xfer->speed_hz, scbr, bus_hz);
834 		return -EINVAL;
835 	}
836 	csr = spi_readl(as, CSR0 + 4 * chip_select);
837 	csr = SPI_BFINS(SCBR, scbr, csr);
838 	spi_writel(as, CSR0 + 4 * chip_select, csr);
839 	xfer->effective_speed_hz = bus_hz / scbr;
840 
841 	return 0;
842 }
843 
844 /*
845  * Submit next transfer for PDC.
846  * lock is held, spi irq is blocked
847  */
848 static void atmel_spi_pdc_next_xfer(struct spi_controller *host,
849 					struct spi_transfer *xfer)
850 {
851 	struct atmel_spi	*as = spi_controller_get_devdata(host);
852 	u32			len;
853 	dma_addr_t		tx_dma, rx_dma;
854 
855 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
856 
857 	len = as->current_remaining_bytes;
858 	atmel_spi_next_xfer_data(host, xfer, &tx_dma, &rx_dma, &len);
859 	as->current_remaining_bytes -= len;
860 
861 	spi_writel(as, RPR, rx_dma);
862 	spi_writel(as, TPR, tx_dma);
863 
864 	if (xfer->bits_per_word > 8)
865 		len >>= 1;
866 	spi_writel(as, RCR, len);
867 	spi_writel(as, TCR, len);
868 
869 	dev_dbg(&host->dev,
870 		"  start xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
871 		xfer, xfer->len, xfer->tx_buf,
872 		(unsigned long long)xfer->tx_dma, xfer->rx_buf,
873 		(unsigned long long)xfer->rx_dma);
874 
875 	if (as->current_remaining_bytes) {
876 		len = as->current_remaining_bytes;
877 		atmel_spi_next_xfer_data(host, xfer, &tx_dma, &rx_dma, &len);
878 		as->current_remaining_bytes -= len;
879 
880 		spi_writel(as, RNPR, rx_dma);
881 		spi_writel(as, TNPR, tx_dma);
882 
883 		if (xfer->bits_per_word > 8)
884 			len >>= 1;
885 		spi_writel(as, RNCR, len);
886 		spi_writel(as, TNCR, len);
887 
888 		dev_dbg(&host->dev,
889 			"  next xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
890 			xfer, xfer->len, xfer->tx_buf,
891 			(unsigned long long)xfer->tx_dma, xfer->rx_buf,
892 			(unsigned long long)xfer->rx_dma);
893 	}
894 
895 	/* REVISIT: We're waiting for RXBUFF before we start the next
896 	 * transfer because we need to handle some difficult timing
897 	 * issues otherwise. If we wait for TXBUFE in one transfer and
898 	 * then starts waiting for RXBUFF in the next, it's difficult
899 	 * to tell the difference between the RXBUFF interrupt we're
900 	 * actually waiting for and the RXBUFF interrupt of the
901 	 * previous transfer.
902 	 *
903 	 * It should be doable, though. Just not now...
904 	 */
905 	spi_writel(as, IER, SPI_BIT(RXBUFF) | SPI_BIT(OVRES));
906 	spi_writel(as, PTCR, SPI_BIT(TXTEN) | SPI_BIT(RXTEN));
907 }
908 
909 /*
910  * For DMA, tx_buf/tx_dma have the same relationship as rx_buf/rx_dma:
911  *  - The buffer is either valid for CPU access, else NULL
912  *  - If the buffer is valid, so is its DMA address
913  *
914  * This driver manages the dma address unless message->is_dma_mapped.
915  */
916 static int
917 atmel_spi_dma_map_xfer(struct atmel_spi *as, struct spi_transfer *xfer)
918 {
919 	struct device	*dev = &as->pdev->dev;
920 
921 	xfer->tx_dma = xfer->rx_dma = INVALID_DMA_ADDRESS;
922 	if (xfer->tx_buf) {
923 		/* tx_buf is a const void* where we need a void * for the dma
924 		 * mapping */
925 		void *nonconst_tx = (void *)xfer->tx_buf;
926 
927 		xfer->tx_dma = dma_map_single(dev,
928 				nonconst_tx, xfer->len,
929 				DMA_TO_DEVICE);
930 		if (dma_mapping_error(dev, xfer->tx_dma))
931 			return -ENOMEM;
932 	}
933 	if (xfer->rx_buf) {
934 		xfer->rx_dma = dma_map_single(dev,
935 				xfer->rx_buf, xfer->len,
936 				DMA_FROM_DEVICE);
937 		if (dma_mapping_error(dev, xfer->rx_dma)) {
938 			if (xfer->tx_buf)
939 				dma_unmap_single(dev,
940 						xfer->tx_dma, xfer->len,
941 						DMA_TO_DEVICE);
942 			return -ENOMEM;
943 		}
944 	}
945 	return 0;
946 }
947 
948 static void atmel_spi_dma_unmap_xfer(struct spi_controller *host,
949 				     struct spi_transfer *xfer)
950 {
951 	if (xfer->tx_dma != INVALID_DMA_ADDRESS)
952 		dma_unmap_single(host->dev.parent, xfer->tx_dma,
953 				 xfer->len, DMA_TO_DEVICE);
954 	if (xfer->rx_dma != INVALID_DMA_ADDRESS)
955 		dma_unmap_single(host->dev.parent, xfer->rx_dma,
956 				 xfer->len, DMA_FROM_DEVICE);
957 }
958 
959 static void atmel_spi_disable_pdc_transfer(struct atmel_spi *as)
960 {
961 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
962 }
963 
964 static void
965 atmel_spi_pump_single_data(struct atmel_spi *as, struct spi_transfer *xfer)
966 {
967 	u8		*rxp;
968 	u16		*rxp16;
969 	unsigned long	xfer_pos = xfer->len - as->current_remaining_bytes;
970 
971 	if (xfer->bits_per_word > 8) {
972 		rxp16 = (u16 *)(((u8 *)xfer->rx_buf) + xfer_pos);
973 		*rxp16 = spi_readl(as, RDR);
974 	} else {
975 		rxp = ((u8 *)xfer->rx_buf) + xfer_pos;
976 		*rxp = spi_readl(as, RDR);
977 	}
978 	if (xfer->bits_per_word > 8) {
979 		if (as->current_remaining_bytes > 2)
980 			as->current_remaining_bytes -= 2;
981 		else
982 			as->current_remaining_bytes = 0;
983 	} else {
984 		as->current_remaining_bytes--;
985 	}
986 }
987 
988 static void
989 atmel_spi_pump_fifo_data(struct atmel_spi *as, struct spi_transfer *xfer)
990 {
991 	u32 fifolr = spi_readl(as, FLR);
992 	u32 num_bytes, num_data = SPI_BFEXT(RXFL, fifolr);
993 	u32 offset = xfer->len - as->current_remaining_bytes;
994 	u16 *words = (u16 *)((u8 *)xfer->rx_buf + offset);
995 	u8  *bytes = (u8  *)((u8 *)xfer->rx_buf + offset);
996 	u16 rd; /* RD field is the lowest 16 bits of RDR */
997 
998 	/* Update the number of remaining bytes to transfer */
999 	num_bytes = ((xfer->bits_per_word > 8) ?
1000 		     (num_data << 1) :
1001 		     num_data);
1002 
1003 	if (as->current_remaining_bytes > num_bytes)
1004 		as->current_remaining_bytes -= num_bytes;
1005 	else
1006 		as->current_remaining_bytes = 0;
1007 
1008 	/* Handle odd number of bytes when data are more than 8bit width */
1009 	if (xfer->bits_per_word > 8)
1010 		as->current_remaining_bytes &= ~0x1;
1011 
1012 	/* Read data */
1013 	while (num_data) {
1014 		rd = spi_readl(as, RDR);
1015 		if (xfer->bits_per_word > 8)
1016 			*words++ = rd;
1017 		else
1018 			*bytes++ = rd;
1019 		num_data--;
1020 	}
1021 }
1022 
1023 /* Called from IRQ
1024  *
1025  * Must update "current_remaining_bytes" to keep track of data
1026  * to transfer.
1027  */
1028 static void
1029 atmel_spi_pump_pio_data(struct atmel_spi *as, struct spi_transfer *xfer)
1030 {
1031 	if (as->fifo_size)
1032 		atmel_spi_pump_fifo_data(as, xfer);
1033 	else
1034 		atmel_spi_pump_single_data(as, xfer);
1035 }
1036 
1037 /* Interrupt
1038  *
1039  */
1040 static irqreturn_t
1041 atmel_spi_pio_interrupt(int irq, void *dev_id)
1042 {
1043 	struct spi_controller	*host = dev_id;
1044 	struct atmel_spi	*as = spi_controller_get_devdata(host);
1045 	u32			status, pending, imr;
1046 	struct spi_transfer	*xfer;
1047 	int			ret = IRQ_NONE;
1048 
1049 	imr = spi_readl(as, IMR);
1050 	status = spi_readl(as, SR);
1051 	pending = status & imr;
1052 
1053 	if (pending & SPI_BIT(OVRES)) {
1054 		ret = IRQ_HANDLED;
1055 		spi_writel(as, IDR, SPI_BIT(OVRES));
1056 		dev_warn(host->dev.parent, "overrun\n");
1057 
1058 		/*
1059 		 * When we get an overrun, we disregard the current
1060 		 * transfer. Data will not be copied back from any
1061 		 * bounce buffer and msg->actual_len will not be
1062 		 * updated with the last xfer.
1063 		 *
1064 		 * We will also not process any remaning transfers in
1065 		 * the message.
1066 		 */
1067 		as->done_status = -EIO;
1068 		smp_wmb();
1069 
1070 		/* Clear any overrun happening while cleaning up */
1071 		spi_readl(as, SR);
1072 
1073 		complete(&as->xfer_completion);
1074 
1075 	} else if (pending & (SPI_BIT(RDRF) | SPI_BIT(RXFTHF))) {
1076 		atmel_spi_lock(as);
1077 
1078 		if (as->current_remaining_bytes) {
1079 			ret = IRQ_HANDLED;
1080 			xfer = as->current_transfer;
1081 			atmel_spi_pump_pio_data(as, xfer);
1082 			if (!as->current_remaining_bytes)
1083 				spi_writel(as, IDR, pending);
1084 
1085 			complete(&as->xfer_completion);
1086 		}
1087 
1088 		atmel_spi_unlock(as);
1089 	} else {
1090 		WARN_ONCE(pending, "IRQ not handled, pending = %x\n", pending);
1091 		ret = IRQ_HANDLED;
1092 		spi_writel(as, IDR, pending);
1093 	}
1094 
1095 	return ret;
1096 }
1097 
1098 static irqreturn_t
1099 atmel_spi_pdc_interrupt(int irq, void *dev_id)
1100 {
1101 	struct spi_controller	*host = dev_id;
1102 	struct atmel_spi	*as = spi_controller_get_devdata(host);
1103 	u32			status, pending, imr;
1104 	int			ret = IRQ_NONE;
1105 
1106 	imr = spi_readl(as, IMR);
1107 	status = spi_readl(as, SR);
1108 	pending = status & imr;
1109 
1110 	if (pending & SPI_BIT(OVRES)) {
1111 
1112 		ret = IRQ_HANDLED;
1113 
1114 		spi_writel(as, IDR, (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX)
1115 				     | SPI_BIT(OVRES)));
1116 
1117 		/* Clear any overrun happening while cleaning up */
1118 		spi_readl(as, SR);
1119 
1120 		as->done_status = -EIO;
1121 
1122 		complete(&as->xfer_completion);
1123 
1124 	} else if (pending & (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX))) {
1125 		ret = IRQ_HANDLED;
1126 
1127 		spi_writel(as, IDR, pending);
1128 
1129 		complete(&as->xfer_completion);
1130 	}
1131 
1132 	return ret;
1133 }
1134 
1135 static int atmel_word_delay_csr(struct spi_device *spi, struct atmel_spi *as)
1136 {
1137 	struct spi_delay *delay = &spi->word_delay;
1138 	u32 value = delay->value;
1139 
1140 	switch (delay->unit) {
1141 	case SPI_DELAY_UNIT_NSECS:
1142 		value /= 1000;
1143 		break;
1144 	case SPI_DELAY_UNIT_USECS:
1145 		break;
1146 	default:
1147 		return -EINVAL;
1148 	}
1149 
1150 	return (as->spi_clk / 1000000 * value) >> 5;
1151 }
1152 
1153 static void initialize_native_cs_for_gpio(struct atmel_spi *as)
1154 {
1155 	int i;
1156 	struct spi_controller *host = platform_get_drvdata(as->pdev);
1157 
1158 	if (!as->native_cs_free)
1159 		return; /* already initialized */
1160 
1161 	if (!host->cs_gpiods)
1162 		return; /* No CS GPIO */
1163 
1164 	/*
1165 	 * On the first version of the controller (AT91RM9200), CS0
1166 	 * can't be used associated with GPIO
1167 	 */
1168 	if (atmel_spi_is_v2(as))
1169 		i = 0;
1170 	else
1171 		i = 1;
1172 
1173 	for (; i < 4; i++)
1174 		if (host->cs_gpiods[i])
1175 			as->native_cs_free |= BIT(i);
1176 
1177 	if (as->native_cs_free)
1178 		as->native_cs_for_gpio = ffs(as->native_cs_free);
1179 }
1180 
1181 static int atmel_spi_setup(struct spi_device *spi)
1182 {
1183 	struct atmel_spi	*as;
1184 	struct atmel_spi_device	*asd;
1185 	u32			csr;
1186 	unsigned int		bits = spi->bits_per_word;
1187 	int chip_select;
1188 	int			word_delay_csr;
1189 
1190 	as = spi_controller_get_devdata(spi->controller);
1191 
1192 	/* see notes above re chipselect */
1193 	if (!spi_get_csgpiod(spi, 0) && (spi->mode & SPI_CS_HIGH)) {
1194 		dev_warn(&spi->dev, "setup: non GPIO CS can't be active-high\n");
1195 		return -EINVAL;
1196 	}
1197 
1198 	/* Setup() is called during spi_register_controller(aka
1199 	 * spi_register_master) but after all membmers of the cs_gpiod
1200 	 * array have been filled, so we can looked for which native
1201 	 * CS will be free for using with GPIO
1202 	 */
1203 	initialize_native_cs_for_gpio(as);
1204 
1205 	if (spi_get_csgpiod(spi, 0) && as->native_cs_free) {
1206 		dev_err(&spi->dev,
1207 			"No native CS available to support this GPIO CS\n");
1208 		return -EBUSY;
1209 	}
1210 
1211 	if (spi_get_csgpiod(spi, 0))
1212 		chip_select = as->native_cs_for_gpio;
1213 	else
1214 		chip_select = spi_get_chipselect(spi, 0);
1215 
1216 	csr = SPI_BF(BITS, bits - 8);
1217 	if (spi->mode & SPI_CPOL)
1218 		csr |= SPI_BIT(CPOL);
1219 	if (!(spi->mode & SPI_CPHA))
1220 		csr |= SPI_BIT(NCPHA);
1221 
1222 	if (!spi_get_csgpiod(spi, 0))
1223 		csr |= SPI_BIT(CSAAT);
1224 	csr |= SPI_BF(DLYBS, 0);
1225 
1226 	word_delay_csr = atmel_word_delay_csr(spi, as);
1227 	if (word_delay_csr < 0)
1228 		return word_delay_csr;
1229 
1230 	/* DLYBCT adds delays between words.  This is useful for slow devices
1231 	 * that need a bit of time to setup the next transfer.
1232 	 */
1233 	csr |= SPI_BF(DLYBCT, word_delay_csr);
1234 
1235 	asd = spi->controller_state;
1236 	if (!asd) {
1237 		asd = kzalloc(sizeof(struct atmel_spi_device), GFP_KERNEL);
1238 		if (!asd)
1239 			return -ENOMEM;
1240 
1241 		spi->controller_state = asd;
1242 	}
1243 
1244 	asd->csr = csr;
1245 
1246 	dev_dbg(&spi->dev,
1247 		"setup: bpw %u mode 0x%x -> csr%d %08x\n",
1248 		bits, spi->mode, spi_get_chipselect(spi, 0), csr);
1249 
1250 	if (!atmel_spi_is_v2(as))
1251 		spi_writel(as, CSR0 + 4 * chip_select, csr);
1252 
1253 	return 0;
1254 }
1255 
1256 static void atmel_spi_set_cs(struct spi_device *spi, bool enable)
1257 {
1258 	struct atmel_spi *as = spi_controller_get_devdata(spi->controller);
1259 	/* the core doesn't really pass us enable/disable, but CS HIGH vs CS LOW
1260 	 * since we already have routines for activate/deactivate translate
1261 	 * high/low to active/inactive
1262 	 */
1263 	enable = (!!(spi->mode & SPI_CS_HIGH) == enable);
1264 
1265 	if (enable) {
1266 		cs_activate(as, spi);
1267 	} else {
1268 		cs_deactivate(as, spi);
1269 	}
1270 
1271 }
1272 
1273 static int atmel_spi_one_transfer(struct spi_controller *host,
1274 					struct spi_device *spi,
1275 					struct spi_transfer *xfer)
1276 {
1277 	struct atmel_spi	*as;
1278 	u8			bits;
1279 	u32			len;
1280 	struct atmel_spi_device	*asd;
1281 	int			timeout;
1282 	int			ret;
1283 	unsigned int		dma_timeout;
1284 	long			ret_timeout;
1285 
1286 	as = spi_controller_get_devdata(host);
1287 
1288 	asd = spi->controller_state;
1289 	bits = (asd->csr >> 4) & 0xf;
1290 	if (bits != xfer->bits_per_word - 8) {
1291 		dev_dbg(&spi->dev,
1292 			"you can't yet change bits_per_word in transfers\n");
1293 		return -ENOPROTOOPT;
1294 	}
1295 
1296 	/*
1297 	 * DMA map early, for performance (empties dcache ASAP) and
1298 	 * better fault reporting.
1299 	 */
1300 	if ((!host->cur_msg->is_dma_mapped)
1301 		&& as->use_pdc) {
1302 		if (atmel_spi_dma_map_xfer(as, xfer) < 0)
1303 			return -ENOMEM;
1304 	}
1305 
1306 	atmel_spi_set_xfer_speed(as, spi, xfer);
1307 
1308 	as->done_status = 0;
1309 	as->current_transfer = xfer;
1310 	as->current_remaining_bytes = xfer->len;
1311 	while (as->current_remaining_bytes) {
1312 		reinit_completion(&as->xfer_completion);
1313 
1314 		if (as->use_pdc) {
1315 			atmel_spi_lock(as);
1316 			atmel_spi_pdc_next_xfer(host, xfer);
1317 			atmel_spi_unlock(as);
1318 		} else if (atmel_spi_use_dma(as, xfer)) {
1319 			len = as->current_remaining_bytes;
1320 			ret = atmel_spi_next_xfer_dma_submit(host,
1321 								xfer, &len);
1322 			if (ret) {
1323 				dev_err(&spi->dev,
1324 					"unable to use DMA, fallback to PIO\n");
1325 				as->done_status = ret;
1326 				break;
1327 			} else {
1328 				as->current_remaining_bytes -= len;
1329 				if (as->current_remaining_bytes < 0)
1330 					as->current_remaining_bytes = 0;
1331 			}
1332 		} else {
1333 			atmel_spi_lock(as);
1334 			atmel_spi_next_xfer_pio(host, xfer);
1335 			atmel_spi_unlock(as);
1336 		}
1337 
1338 		dma_timeout = msecs_to_jiffies(spi_controller_xfer_timeout(host, xfer));
1339 		ret_timeout = wait_for_completion_interruptible_timeout(&as->xfer_completion,
1340 									dma_timeout);
1341 		if (ret_timeout <= 0) {
1342 			dev_err(&spi->dev, "spi transfer %s\n",
1343 				!ret_timeout ? "timeout" : "canceled");
1344 			as->done_status = ret_timeout < 0 ? ret_timeout : -EIO;
1345 		}
1346 
1347 		if (as->done_status)
1348 			break;
1349 	}
1350 
1351 	if (as->done_status) {
1352 		if (as->use_pdc) {
1353 			dev_warn(host->dev.parent,
1354 				"overrun (%u/%u remaining)\n",
1355 				spi_readl(as, TCR), spi_readl(as, RCR));
1356 
1357 			/*
1358 			 * Clean up DMA registers and make sure the data
1359 			 * registers are empty.
1360 			 */
1361 			spi_writel(as, RNCR, 0);
1362 			spi_writel(as, TNCR, 0);
1363 			spi_writel(as, RCR, 0);
1364 			spi_writel(as, TCR, 0);
1365 			for (timeout = 1000; timeout; timeout--)
1366 				if (spi_readl(as, SR) & SPI_BIT(TXEMPTY))
1367 					break;
1368 			if (!timeout)
1369 				dev_warn(host->dev.parent,
1370 					 "timeout waiting for TXEMPTY");
1371 			while (spi_readl(as, SR) & SPI_BIT(RDRF))
1372 				spi_readl(as, RDR);
1373 
1374 			/* Clear any overrun happening while cleaning up */
1375 			spi_readl(as, SR);
1376 
1377 		} else if (atmel_spi_use_dma(as, xfer)) {
1378 			atmel_spi_stop_dma(host);
1379 		}
1380 	}
1381 
1382 	if (!host->cur_msg->is_dma_mapped
1383 		&& as->use_pdc)
1384 		atmel_spi_dma_unmap_xfer(host, xfer);
1385 
1386 	if (as->use_pdc)
1387 		atmel_spi_disable_pdc_transfer(as);
1388 
1389 	return as->done_status;
1390 }
1391 
1392 static void atmel_spi_cleanup(struct spi_device *spi)
1393 {
1394 	struct atmel_spi_device	*asd = spi->controller_state;
1395 
1396 	if (!asd)
1397 		return;
1398 
1399 	spi->controller_state = NULL;
1400 	kfree(asd);
1401 }
1402 
1403 static inline unsigned int atmel_get_version(struct atmel_spi *as)
1404 {
1405 	return spi_readl(as, VERSION) & 0x00000fff;
1406 }
1407 
1408 static void atmel_get_caps(struct atmel_spi *as)
1409 {
1410 	unsigned int version;
1411 
1412 	version = atmel_get_version(as);
1413 
1414 	as->caps.is_spi2 = version > 0x121;
1415 	as->caps.has_wdrbt = version >= 0x210;
1416 	as->caps.has_dma_support = version >= 0x212;
1417 	as->caps.has_pdc_support = version < 0x212;
1418 }
1419 
1420 static void atmel_spi_init(struct atmel_spi *as)
1421 {
1422 	spi_writel(as, CR, SPI_BIT(SWRST));
1423 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1424 
1425 	/* It is recommended to enable FIFOs first thing after reset */
1426 	if (as->fifo_size)
1427 		spi_writel(as, CR, SPI_BIT(FIFOEN));
1428 
1429 	if (as->caps.has_wdrbt) {
1430 		spi_writel(as, MR, SPI_BIT(WDRBT) | SPI_BIT(MODFDIS)
1431 				| SPI_BIT(MSTR));
1432 	} else {
1433 		spi_writel(as, MR, SPI_BIT(MSTR) | SPI_BIT(MODFDIS));
1434 	}
1435 
1436 	if (as->use_pdc)
1437 		spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1438 	spi_writel(as, CR, SPI_BIT(SPIEN));
1439 }
1440 
1441 static int atmel_spi_probe(struct platform_device *pdev)
1442 {
1443 	struct resource		*regs;
1444 	int			irq;
1445 	struct clk		*clk;
1446 	int			ret;
1447 	struct spi_controller	*host;
1448 	struct atmel_spi	*as;
1449 
1450 	/* Select default pin state */
1451 	pinctrl_pm_select_default_state(&pdev->dev);
1452 
1453 	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1454 	if (!regs)
1455 		return -ENXIO;
1456 
1457 	irq = platform_get_irq(pdev, 0);
1458 	if (irq < 0)
1459 		return irq;
1460 
1461 	clk = devm_clk_get(&pdev->dev, "spi_clk");
1462 	if (IS_ERR(clk))
1463 		return PTR_ERR(clk);
1464 
1465 	/* setup spi core then atmel-specific driver state */
1466 	host = spi_alloc_host(&pdev->dev, sizeof(*as));
1467 	if (!host)
1468 		return -ENOMEM;
1469 
1470 	/* the spi->mode bits understood by this driver: */
1471 	host->use_gpio_descriptors = true;
1472 	host->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
1473 	host->bits_per_word_mask = SPI_BPW_RANGE_MASK(8, 16);
1474 	host->dev.of_node = pdev->dev.of_node;
1475 	host->bus_num = pdev->id;
1476 	host->num_chipselect = 4;
1477 	host->setup = atmel_spi_setup;
1478 	host->flags = (SPI_MASTER_MUST_RX | SPI_MASTER_MUST_TX |
1479 			SPI_MASTER_GPIO_SS);
1480 	host->transfer_one = atmel_spi_one_transfer;
1481 	host->set_cs = atmel_spi_set_cs;
1482 	host->cleanup = atmel_spi_cleanup;
1483 	host->auto_runtime_pm = true;
1484 	host->max_dma_len = SPI_MAX_DMA_XFER;
1485 	host->can_dma = atmel_spi_can_dma;
1486 	platform_set_drvdata(pdev, host);
1487 
1488 	as = spi_controller_get_devdata(host);
1489 
1490 	spin_lock_init(&as->lock);
1491 
1492 	as->pdev = pdev;
1493 	as->regs = devm_ioremap_resource(&pdev->dev, regs);
1494 	if (IS_ERR(as->regs)) {
1495 		ret = PTR_ERR(as->regs);
1496 		goto out_unmap_regs;
1497 	}
1498 	as->phybase = regs->start;
1499 	as->irq = irq;
1500 	as->clk = clk;
1501 
1502 	init_completion(&as->xfer_completion);
1503 
1504 	atmel_get_caps(as);
1505 
1506 	as->use_dma = false;
1507 	as->use_pdc = false;
1508 	if (as->caps.has_dma_support) {
1509 		ret = atmel_spi_configure_dma(host, as);
1510 		if (ret == 0) {
1511 			as->use_dma = true;
1512 		} else if (ret == -EPROBE_DEFER) {
1513 			goto out_unmap_regs;
1514 		}
1515 	} else if (as->caps.has_pdc_support) {
1516 		as->use_pdc = true;
1517 	}
1518 
1519 	if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
1520 		as->addr_rx_bbuf = dma_alloc_coherent(&pdev->dev,
1521 						      SPI_MAX_DMA_XFER,
1522 						      &as->dma_addr_rx_bbuf,
1523 						      GFP_KERNEL | GFP_DMA);
1524 		if (!as->addr_rx_bbuf) {
1525 			as->use_dma = false;
1526 		} else {
1527 			as->addr_tx_bbuf = dma_alloc_coherent(&pdev->dev,
1528 					SPI_MAX_DMA_XFER,
1529 					&as->dma_addr_tx_bbuf,
1530 					GFP_KERNEL | GFP_DMA);
1531 			if (!as->addr_tx_bbuf) {
1532 				as->use_dma = false;
1533 				dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1534 						  as->addr_rx_bbuf,
1535 						  as->dma_addr_rx_bbuf);
1536 			}
1537 		}
1538 		if (!as->use_dma)
1539 			dev_info(host->dev.parent,
1540 				 "  can not allocate dma coherent memory\n");
1541 	}
1542 
1543 	if (as->caps.has_dma_support && !as->use_dma)
1544 		dev_info(&pdev->dev, "Atmel SPI Controller using PIO only\n");
1545 
1546 	if (as->use_pdc) {
1547 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pdc_interrupt,
1548 					0, dev_name(&pdev->dev), host);
1549 	} else {
1550 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pio_interrupt,
1551 					0, dev_name(&pdev->dev), host);
1552 	}
1553 	if (ret)
1554 		goto out_unmap_regs;
1555 
1556 	/* Initialize the hardware */
1557 	ret = clk_prepare_enable(clk);
1558 	if (ret)
1559 		goto out_free_irq;
1560 
1561 	as->spi_clk = clk_get_rate(clk);
1562 
1563 	as->fifo_size = 0;
1564 	if (!of_property_read_u32(pdev->dev.of_node, "atmel,fifo-size",
1565 				  &as->fifo_size)) {
1566 		dev_info(&pdev->dev, "Using FIFO (%u data)\n", as->fifo_size);
1567 	}
1568 
1569 	atmel_spi_init(as);
1570 
1571 	pm_runtime_set_autosuspend_delay(&pdev->dev, AUTOSUSPEND_TIMEOUT);
1572 	pm_runtime_use_autosuspend(&pdev->dev);
1573 	pm_runtime_set_active(&pdev->dev);
1574 	pm_runtime_enable(&pdev->dev);
1575 
1576 	ret = devm_spi_register_controller(&pdev->dev, host);
1577 	if (ret)
1578 		goto out_free_dma;
1579 
1580 	/* go! */
1581 	dev_info(&pdev->dev, "Atmel SPI Controller version 0x%x at 0x%08lx (irq %d)\n",
1582 			atmel_get_version(as), (unsigned long)regs->start,
1583 			irq);
1584 
1585 	return 0;
1586 
1587 out_free_dma:
1588 	pm_runtime_disable(&pdev->dev);
1589 	pm_runtime_set_suspended(&pdev->dev);
1590 
1591 	if (as->use_dma)
1592 		atmel_spi_release_dma(host);
1593 
1594 	spi_writel(as, CR, SPI_BIT(SWRST));
1595 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1596 	clk_disable_unprepare(clk);
1597 out_free_irq:
1598 out_unmap_regs:
1599 	spi_controller_put(host);
1600 	return ret;
1601 }
1602 
1603 static void atmel_spi_remove(struct platform_device *pdev)
1604 {
1605 	struct spi_controller	*host = platform_get_drvdata(pdev);
1606 	struct atmel_spi	*as = spi_controller_get_devdata(host);
1607 
1608 	pm_runtime_get_sync(&pdev->dev);
1609 
1610 	/* reset the hardware and block queue progress */
1611 	if (as->use_dma) {
1612 		atmel_spi_stop_dma(host);
1613 		atmel_spi_release_dma(host);
1614 		if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
1615 			dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1616 					  as->addr_tx_bbuf,
1617 					  as->dma_addr_tx_bbuf);
1618 			dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1619 					  as->addr_rx_bbuf,
1620 					  as->dma_addr_rx_bbuf);
1621 		}
1622 	}
1623 
1624 	spin_lock_irq(&as->lock);
1625 	spi_writel(as, CR, SPI_BIT(SWRST));
1626 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1627 	spi_readl(as, SR);
1628 	spin_unlock_irq(&as->lock);
1629 
1630 	clk_disable_unprepare(as->clk);
1631 
1632 	pm_runtime_put_noidle(&pdev->dev);
1633 	pm_runtime_disable(&pdev->dev);
1634 }
1635 
1636 static int atmel_spi_runtime_suspend(struct device *dev)
1637 {
1638 	struct spi_controller *host = dev_get_drvdata(dev);
1639 	struct atmel_spi *as = spi_controller_get_devdata(host);
1640 
1641 	clk_disable_unprepare(as->clk);
1642 	pinctrl_pm_select_sleep_state(dev);
1643 
1644 	return 0;
1645 }
1646 
1647 static int atmel_spi_runtime_resume(struct device *dev)
1648 {
1649 	struct spi_controller *host = dev_get_drvdata(dev);
1650 	struct atmel_spi *as = spi_controller_get_devdata(host);
1651 
1652 	pinctrl_pm_select_default_state(dev);
1653 
1654 	return clk_prepare_enable(as->clk);
1655 }
1656 
1657 static int atmel_spi_suspend(struct device *dev)
1658 {
1659 	struct spi_controller *host = dev_get_drvdata(dev);
1660 	int ret;
1661 
1662 	/* Stop the queue running */
1663 	ret = spi_controller_suspend(host);
1664 	if (ret)
1665 		return ret;
1666 
1667 	if (!pm_runtime_suspended(dev))
1668 		atmel_spi_runtime_suspend(dev);
1669 
1670 	return 0;
1671 }
1672 
1673 static int atmel_spi_resume(struct device *dev)
1674 {
1675 	struct spi_controller *host = dev_get_drvdata(dev);
1676 	struct atmel_spi *as = spi_controller_get_devdata(host);
1677 	int ret;
1678 
1679 	ret = clk_prepare_enable(as->clk);
1680 	if (ret)
1681 		return ret;
1682 
1683 	atmel_spi_init(as);
1684 
1685 	clk_disable_unprepare(as->clk);
1686 
1687 	if (!pm_runtime_suspended(dev)) {
1688 		ret = atmel_spi_runtime_resume(dev);
1689 		if (ret)
1690 			return ret;
1691 	}
1692 
1693 	/* Start the queue running */
1694 	return spi_controller_resume(host);
1695 }
1696 
1697 static const struct dev_pm_ops atmel_spi_pm_ops = {
1698 	SYSTEM_SLEEP_PM_OPS(atmel_spi_suspend, atmel_spi_resume)
1699 	RUNTIME_PM_OPS(atmel_spi_runtime_suspend,
1700 		       atmel_spi_runtime_resume, NULL)
1701 };
1702 
1703 static const struct of_device_id atmel_spi_dt_ids[] = {
1704 	{ .compatible = "atmel,at91rm9200-spi" },
1705 	{ /* sentinel */ }
1706 };
1707 
1708 MODULE_DEVICE_TABLE(of, atmel_spi_dt_ids);
1709 
1710 static struct platform_driver atmel_spi_driver = {
1711 	.driver		= {
1712 		.name	= "atmel_spi",
1713 		.pm	= pm_ptr(&atmel_spi_pm_ops),
1714 		.of_match_table	= atmel_spi_dt_ids,
1715 	},
1716 	.probe		= atmel_spi_probe,
1717 	.remove_new	= atmel_spi_remove,
1718 };
1719 module_platform_driver(atmel_spi_driver);
1720 
1721 MODULE_DESCRIPTION("Atmel AT32/AT91 SPI Controller driver");
1722 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1723 MODULE_LICENSE("GPL");
1724 MODULE_ALIAS("platform:atmel_spi");
1725