xref: /openbmc/linux/drivers/spi/spi-atmel.c (revision 32981ea5)
1 /*
2  * Driver for Atmel AT32 and AT91 SPI Controllers
3  *
4  * Copyright (C) 2006 Atmel Corporation
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  */
10 
11 #include <linux/kernel.h>
12 #include <linux/clk.h>
13 #include <linux/module.h>
14 #include <linux/platform_device.h>
15 #include <linux/delay.h>
16 #include <linux/dma-mapping.h>
17 #include <linux/dmaengine.h>
18 #include <linux/err.h>
19 #include <linux/interrupt.h>
20 #include <linux/spi/spi.h>
21 #include <linux/slab.h>
22 #include <linux/platform_data/dma-atmel.h>
23 #include <linux/of.h>
24 
25 #include <linux/io.h>
26 #include <linux/gpio.h>
27 #include <linux/pinctrl/consumer.h>
28 #include <linux/pm_runtime.h>
29 
30 /* SPI register offsets */
31 #define SPI_CR					0x0000
32 #define SPI_MR					0x0004
33 #define SPI_RDR					0x0008
34 #define SPI_TDR					0x000c
35 #define SPI_SR					0x0010
36 #define SPI_IER					0x0014
37 #define SPI_IDR					0x0018
38 #define SPI_IMR					0x001c
39 #define SPI_CSR0				0x0030
40 #define SPI_CSR1				0x0034
41 #define SPI_CSR2				0x0038
42 #define SPI_CSR3				0x003c
43 #define SPI_FMR					0x0040
44 #define SPI_FLR					0x0044
45 #define SPI_VERSION				0x00fc
46 #define SPI_RPR					0x0100
47 #define SPI_RCR					0x0104
48 #define SPI_TPR					0x0108
49 #define SPI_TCR					0x010c
50 #define SPI_RNPR				0x0110
51 #define SPI_RNCR				0x0114
52 #define SPI_TNPR				0x0118
53 #define SPI_TNCR				0x011c
54 #define SPI_PTCR				0x0120
55 #define SPI_PTSR				0x0124
56 
57 /* Bitfields in CR */
58 #define SPI_SPIEN_OFFSET			0
59 #define SPI_SPIEN_SIZE				1
60 #define SPI_SPIDIS_OFFSET			1
61 #define SPI_SPIDIS_SIZE				1
62 #define SPI_SWRST_OFFSET			7
63 #define SPI_SWRST_SIZE				1
64 #define SPI_LASTXFER_OFFSET			24
65 #define SPI_LASTXFER_SIZE			1
66 #define SPI_TXFCLR_OFFSET			16
67 #define SPI_TXFCLR_SIZE				1
68 #define SPI_RXFCLR_OFFSET			17
69 #define SPI_RXFCLR_SIZE				1
70 #define SPI_FIFOEN_OFFSET			30
71 #define SPI_FIFOEN_SIZE				1
72 #define SPI_FIFODIS_OFFSET			31
73 #define SPI_FIFODIS_SIZE			1
74 
75 /* Bitfields in MR */
76 #define SPI_MSTR_OFFSET				0
77 #define SPI_MSTR_SIZE				1
78 #define SPI_PS_OFFSET				1
79 #define SPI_PS_SIZE				1
80 #define SPI_PCSDEC_OFFSET			2
81 #define SPI_PCSDEC_SIZE				1
82 #define SPI_FDIV_OFFSET				3
83 #define SPI_FDIV_SIZE				1
84 #define SPI_MODFDIS_OFFSET			4
85 #define SPI_MODFDIS_SIZE			1
86 #define SPI_WDRBT_OFFSET			5
87 #define SPI_WDRBT_SIZE				1
88 #define SPI_LLB_OFFSET				7
89 #define SPI_LLB_SIZE				1
90 #define SPI_PCS_OFFSET				16
91 #define SPI_PCS_SIZE				4
92 #define SPI_DLYBCS_OFFSET			24
93 #define SPI_DLYBCS_SIZE				8
94 
95 /* Bitfields in RDR */
96 #define SPI_RD_OFFSET				0
97 #define SPI_RD_SIZE				16
98 
99 /* Bitfields in TDR */
100 #define SPI_TD_OFFSET				0
101 #define SPI_TD_SIZE				16
102 
103 /* Bitfields in SR */
104 #define SPI_RDRF_OFFSET				0
105 #define SPI_RDRF_SIZE				1
106 #define SPI_TDRE_OFFSET				1
107 #define SPI_TDRE_SIZE				1
108 #define SPI_MODF_OFFSET				2
109 #define SPI_MODF_SIZE				1
110 #define SPI_OVRES_OFFSET			3
111 #define SPI_OVRES_SIZE				1
112 #define SPI_ENDRX_OFFSET			4
113 #define SPI_ENDRX_SIZE				1
114 #define SPI_ENDTX_OFFSET			5
115 #define SPI_ENDTX_SIZE				1
116 #define SPI_RXBUFF_OFFSET			6
117 #define SPI_RXBUFF_SIZE				1
118 #define SPI_TXBUFE_OFFSET			7
119 #define SPI_TXBUFE_SIZE				1
120 #define SPI_NSSR_OFFSET				8
121 #define SPI_NSSR_SIZE				1
122 #define SPI_TXEMPTY_OFFSET			9
123 #define SPI_TXEMPTY_SIZE			1
124 #define SPI_SPIENS_OFFSET			16
125 #define SPI_SPIENS_SIZE				1
126 #define SPI_TXFEF_OFFSET			24
127 #define SPI_TXFEF_SIZE				1
128 #define SPI_TXFFF_OFFSET			25
129 #define SPI_TXFFF_SIZE				1
130 #define SPI_TXFTHF_OFFSET			26
131 #define SPI_TXFTHF_SIZE				1
132 #define SPI_RXFEF_OFFSET			27
133 #define SPI_RXFEF_SIZE				1
134 #define SPI_RXFFF_OFFSET			28
135 #define SPI_RXFFF_SIZE				1
136 #define SPI_RXFTHF_OFFSET			29
137 #define SPI_RXFTHF_SIZE				1
138 #define SPI_TXFPTEF_OFFSET			30
139 #define SPI_TXFPTEF_SIZE			1
140 #define SPI_RXFPTEF_OFFSET			31
141 #define SPI_RXFPTEF_SIZE			1
142 
143 /* Bitfields in CSR0 */
144 #define SPI_CPOL_OFFSET				0
145 #define SPI_CPOL_SIZE				1
146 #define SPI_NCPHA_OFFSET			1
147 #define SPI_NCPHA_SIZE				1
148 #define SPI_CSAAT_OFFSET			3
149 #define SPI_CSAAT_SIZE				1
150 #define SPI_BITS_OFFSET				4
151 #define SPI_BITS_SIZE				4
152 #define SPI_SCBR_OFFSET				8
153 #define SPI_SCBR_SIZE				8
154 #define SPI_DLYBS_OFFSET			16
155 #define SPI_DLYBS_SIZE				8
156 #define SPI_DLYBCT_OFFSET			24
157 #define SPI_DLYBCT_SIZE				8
158 
159 /* Bitfields in RCR */
160 #define SPI_RXCTR_OFFSET			0
161 #define SPI_RXCTR_SIZE				16
162 
163 /* Bitfields in TCR */
164 #define SPI_TXCTR_OFFSET			0
165 #define SPI_TXCTR_SIZE				16
166 
167 /* Bitfields in RNCR */
168 #define SPI_RXNCR_OFFSET			0
169 #define SPI_RXNCR_SIZE				16
170 
171 /* Bitfields in TNCR */
172 #define SPI_TXNCR_OFFSET			0
173 #define SPI_TXNCR_SIZE				16
174 
175 /* Bitfields in PTCR */
176 #define SPI_RXTEN_OFFSET			0
177 #define SPI_RXTEN_SIZE				1
178 #define SPI_RXTDIS_OFFSET			1
179 #define SPI_RXTDIS_SIZE				1
180 #define SPI_TXTEN_OFFSET			8
181 #define SPI_TXTEN_SIZE				1
182 #define SPI_TXTDIS_OFFSET			9
183 #define SPI_TXTDIS_SIZE				1
184 
185 /* Bitfields in FMR */
186 #define SPI_TXRDYM_OFFSET			0
187 #define SPI_TXRDYM_SIZE				2
188 #define SPI_RXRDYM_OFFSET			4
189 #define SPI_RXRDYM_SIZE				2
190 #define SPI_TXFTHRES_OFFSET			16
191 #define SPI_TXFTHRES_SIZE			6
192 #define SPI_RXFTHRES_OFFSET			24
193 #define SPI_RXFTHRES_SIZE			6
194 
195 /* Bitfields in FLR */
196 #define SPI_TXFL_OFFSET				0
197 #define SPI_TXFL_SIZE				6
198 #define SPI_RXFL_OFFSET				16
199 #define SPI_RXFL_SIZE				6
200 
201 /* Constants for BITS */
202 #define SPI_BITS_8_BPT				0
203 #define SPI_BITS_9_BPT				1
204 #define SPI_BITS_10_BPT				2
205 #define SPI_BITS_11_BPT				3
206 #define SPI_BITS_12_BPT				4
207 #define SPI_BITS_13_BPT				5
208 #define SPI_BITS_14_BPT				6
209 #define SPI_BITS_15_BPT				7
210 #define SPI_BITS_16_BPT				8
211 #define SPI_ONE_DATA				0
212 #define SPI_TWO_DATA				1
213 #define SPI_FOUR_DATA				2
214 
215 /* Bit manipulation macros */
216 #define SPI_BIT(name) \
217 	(1 << SPI_##name##_OFFSET)
218 #define SPI_BF(name, value) \
219 	(((value) & ((1 << SPI_##name##_SIZE) - 1)) << SPI_##name##_OFFSET)
220 #define SPI_BFEXT(name, value) \
221 	(((value) >> SPI_##name##_OFFSET) & ((1 << SPI_##name##_SIZE) - 1))
222 #define SPI_BFINS(name, value, old) \
223 	(((old) & ~(((1 << SPI_##name##_SIZE) - 1) << SPI_##name##_OFFSET)) \
224 	  | SPI_BF(name, value))
225 
226 /* Register access macros */
227 #ifdef CONFIG_AVR32
228 #define spi_readl(port, reg) \
229 	__raw_readl((port)->regs + SPI_##reg)
230 #define spi_writel(port, reg, value) \
231 	__raw_writel((value), (port)->regs + SPI_##reg)
232 
233 #define spi_readw(port, reg) \
234 	__raw_readw((port)->regs + SPI_##reg)
235 #define spi_writew(port, reg, value) \
236 	__raw_writew((value), (port)->regs + SPI_##reg)
237 
238 #define spi_readb(port, reg) \
239 	__raw_readb((port)->regs + SPI_##reg)
240 #define spi_writeb(port, reg, value) \
241 	__raw_writeb((value), (port)->regs + SPI_##reg)
242 #else
243 #define spi_readl(port, reg) \
244 	readl_relaxed((port)->regs + SPI_##reg)
245 #define spi_writel(port, reg, value) \
246 	writel_relaxed((value), (port)->regs + SPI_##reg)
247 
248 #define spi_readw(port, reg) \
249 	readw_relaxed((port)->regs + SPI_##reg)
250 #define spi_writew(port, reg, value) \
251 	writew_relaxed((value), (port)->regs + SPI_##reg)
252 
253 #define spi_readb(port, reg) \
254 	readb_relaxed((port)->regs + SPI_##reg)
255 #define spi_writeb(port, reg, value) \
256 	writeb_relaxed((value), (port)->regs + SPI_##reg)
257 #endif
258 /* use PIO for small transfers, avoiding DMA setup/teardown overhead and
259  * cache operations; better heuristics consider wordsize and bitrate.
260  */
261 #define DMA_MIN_BYTES	16
262 
263 #define SPI_DMA_TIMEOUT		(msecs_to_jiffies(1000))
264 
265 #define AUTOSUSPEND_TIMEOUT	2000
266 
267 struct atmel_spi_dma {
268 	struct dma_chan			*chan_rx;
269 	struct dma_chan			*chan_tx;
270 	struct scatterlist		sgrx;
271 	struct scatterlist		sgtx;
272 	struct dma_async_tx_descriptor	*data_desc_rx;
273 	struct dma_async_tx_descriptor	*data_desc_tx;
274 
275 	struct at_dma_slave	dma_slave;
276 };
277 
278 struct atmel_spi_caps {
279 	bool	is_spi2;
280 	bool	has_wdrbt;
281 	bool	has_dma_support;
282 };
283 
284 /*
285  * The core SPI transfer engine just talks to a register bank to set up
286  * DMA transfers; transfer queue progress is driven by IRQs.  The clock
287  * framework provides the base clock, subdivided for each spi_device.
288  */
289 struct atmel_spi {
290 	spinlock_t		lock;
291 	unsigned long		flags;
292 
293 	phys_addr_t		phybase;
294 	void __iomem		*regs;
295 	int			irq;
296 	struct clk		*clk;
297 	struct platform_device	*pdev;
298 
299 	struct spi_transfer	*current_transfer;
300 	int			current_remaining_bytes;
301 	int			done_status;
302 
303 	struct completion	xfer_completion;
304 
305 	/* scratch buffer */
306 	void			*buffer;
307 	dma_addr_t		buffer_dma;
308 
309 	struct atmel_spi_caps	caps;
310 
311 	bool			use_dma;
312 	bool			use_pdc;
313 	bool			use_cs_gpios;
314 	/* dmaengine data */
315 	struct atmel_spi_dma	dma;
316 
317 	bool			keep_cs;
318 	bool			cs_active;
319 
320 	u32			fifo_size;
321 };
322 
323 /* Controller-specific per-slave state */
324 struct atmel_spi_device {
325 	unsigned int		npcs_pin;
326 	u32			csr;
327 };
328 
329 #define BUFFER_SIZE		PAGE_SIZE
330 #define INVALID_DMA_ADDRESS	0xffffffff
331 
332 /*
333  * Version 2 of the SPI controller has
334  *  - CR.LASTXFER
335  *  - SPI_MR.DIV32 may become FDIV or must-be-zero (here: always zero)
336  *  - SPI_SR.TXEMPTY, SPI_SR.NSSR (and corresponding irqs)
337  *  - SPI_CSRx.CSAAT
338  *  - SPI_CSRx.SBCR allows faster clocking
339  */
340 static bool atmel_spi_is_v2(struct atmel_spi *as)
341 {
342 	return as->caps.is_spi2;
343 }
344 
345 /*
346  * Earlier SPI controllers (e.g. on at91rm9200) have a design bug whereby
347  * they assume that spi slave device state will not change on deselect, so
348  * that automagic deselection is OK.  ("NPCSx rises if no data is to be
349  * transmitted")  Not so!  Workaround uses nCSx pins as GPIOs; or newer
350  * controllers have CSAAT and friends.
351  *
352  * Since the CSAAT functionality is a bit weird on newer controllers as
353  * well, we use GPIO to control nCSx pins on all controllers, updating
354  * MR.PCS to avoid confusing the controller.  Using GPIOs also lets us
355  * support active-high chipselects despite the controller's belief that
356  * only active-low devices/systems exists.
357  *
358  * However, at91rm9200 has a second erratum whereby nCS0 doesn't work
359  * right when driven with GPIO.  ("Mode Fault does not allow more than one
360  * Master on Chip Select 0.")  No workaround exists for that ... so for
361  * nCS0 on that chip, we (a) don't use the GPIO, (b) can't support CS_HIGH,
362  * and (c) will trigger that first erratum in some cases.
363  */
364 
365 static void cs_activate(struct atmel_spi *as, struct spi_device *spi)
366 {
367 	struct atmel_spi_device *asd = spi->controller_state;
368 	unsigned active = spi->mode & SPI_CS_HIGH;
369 	u32 mr;
370 
371 	if (atmel_spi_is_v2(as)) {
372 		spi_writel(as, CSR0 + 4 * spi->chip_select, asd->csr);
373 		/* For the low SPI version, there is a issue that PDC transfer
374 		 * on CS1,2,3 needs SPI_CSR0.BITS config as SPI_CSR1,2,3.BITS
375 		 */
376 		spi_writel(as, CSR0, asd->csr);
377 		if (as->caps.has_wdrbt) {
378 			spi_writel(as, MR,
379 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
380 					| SPI_BIT(WDRBT)
381 					| SPI_BIT(MODFDIS)
382 					| SPI_BIT(MSTR));
383 		} else {
384 			spi_writel(as, MR,
385 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
386 					| SPI_BIT(MODFDIS)
387 					| SPI_BIT(MSTR));
388 		}
389 
390 		mr = spi_readl(as, MR);
391 		if (as->use_cs_gpios)
392 			gpio_set_value(asd->npcs_pin, active);
393 	} else {
394 		u32 cpol = (spi->mode & SPI_CPOL) ? SPI_BIT(CPOL) : 0;
395 		int i;
396 		u32 csr;
397 
398 		/* Make sure clock polarity is correct */
399 		for (i = 0; i < spi->master->num_chipselect; i++) {
400 			csr = spi_readl(as, CSR0 + 4 * i);
401 			if ((csr ^ cpol) & SPI_BIT(CPOL))
402 				spi_writel(as, CSR0 + 4 * i,
403 						csr ^ SPI_BIT(CPOL));
404 		}
405 
406 		mr = spi_readl(as, MR);
407 		mr = SPI_BFINS(PCS, ~(1 << spi->chip_select), mr);
408 		if (as->use_cs_gpios && spi->chip_select != 0)
409 			gpio_set_value(asd->npcs_pin, active);
410 		spi_writel(as, MR, mr);
411 	}
412 
413 	dev_dbg(&spi->dev, "activate %u%s, mr %08x\n",
414 			asd->npcs_pin, active ? " (high)" : "",
415 			mr);
416 }
417 
418 static void cs_deactivate(struct atmel_spi *as, struct spi_device *spi)
419 {
420 	struct atmel_spi_device *asd = spi->controller_state;
421 	unsigned active = spi->mode & SPI_CS_HIGH;
422 	u32 mr;
423 
424 	/* only deactivate *this* device; sometimes transfers to
425 	 * another device may be active when this routine is called.
426 	 */
427 	mr = spi_readl(as, MR);
428 	if (~SPI_BFEXT(PCS, mr) & (1 << spi->chip_select)) {
429 		mr = SPI_BFINS(PCS, 0xf, mr);
430 		spi_writel(as, MR, mr);
431 	}
432 
433 	dev_dbg(&spi->dev, "DEactivate %u%s, mr %08x\n",
434 			asd->npcs_pin, active ? " (low)" : "",
435 			mr);
436 
437 	if (!as->use_cs_gpios)
438 		spi_writel(as, CR, SPI_BIT(LASTXFER));
439 	else if (atmel_spi_is_v2(as) || spi->chip_select != 0)
440 		gpio_set_value(asd->npcs_pin, !active);
441 }
442 
443 static void atmel_spi_lock(struct atmel_spi *as) __acquires(&as->lock)
444 {
445 	spin_lock_irqsave(&as->lock, as->flags);
446 }
447 
448 static void atmel_spi_unlock(struct atmel_spi *as) __releases(&as->lock)
449 {
450 	spin_unlock_irqrestore(&as->lock, as->flags);
451 }
452 
453 static inline bool atmel_spi_use_dma(struct atmel_spi *as,
454 				struct spi_transfer *xfer)
455 {
456 	return as->use_dma && xfer->len >= DMA_MIN_BYTES;
457 }
458 
459 static int atmel_spi_dma_slave_config(struct atmel_spi *as,
460 				struct dma_slave_config *slave_config,
461 				u8 bits_per_word)
462 {
463 	int err = 0;
464 
465 	if (bits_per_word > 8) {
466 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
467 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
468 	} else {
469 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
470 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
471 	}
472 
473 	slave_config->dst_addr = (dma_addr_t)as->phybase + SPI_TDR;
474 	slave_config->src_addr = (dma_addr_t)as->phybase + SPI_RDR;
475 	slave_config->src_maxburst = 1;
476 	slave_config->dst_maxburst = 1;
477 	slave_config->device_fc = false;
478 
479 	/*
480 	 * This driver uses fixed peripheral select mode (PS bit set to '0' in
481 	 * the Mode Register).
482 	 * So according to the datasheet, when FIFOs are available (and
483 	 * enabled), the Transmit FIFO operates in Multiple Data Mode.
484 	 * In this mode, up to 2 data, not 4, can be written into the Transmit
485 	 * Data Register in a single access.
486 	 * However, the first data has to be written into the lowest 16 bits and
487 	 * the second data into the highest 16 bits of the Transmit
488 	 * Data Register. For 8bit data (the most frequent case), it would
489 	 * require to rework tx_buf so each data would actualy fit 16 bits.
490 	 * So we'd rather write only one data at the time. Hence the transmit
491 	 * path works the same whether FIFOs are available (and enabled) or not.
492 	 */
493 	slave_config->direction = DMA_MEM_TO_DEV;
494 	if (dmaengine_slave_config(as->dma.chan_tx, slave_config)) {
495 		dev_err(&as->pdev->dev,
496 			"failed to configure tx dma channel\n");
497 		err = -EINVAL;
498 	}
499 
500 	/*
501 	 * This driver configures the spi controller for master mode (MSTR bit
502 	 * set to '1' in the Mode Register).
503 	 * So according to the datasheet, when FIFOs are available (and
504 	 * enabled), the Receive FIFO operates in Single Data Mode.
505 	 * So the receive path works the same whether FIFOs are available (and
506 	 * enabled) or not.
507 	 */
508 	slave_config->direction = DMA_DEV_TO_MEM;
509 	if (dmaengine_slave_config(as->dma.chan_rx, slave_config)) {
510 		dev_err(&as->pdev->dev,
511 			"failed to configure rx dma channel\n");
512 		err = -EINVAL;
513 	}
514 
515 	return err;
516 }
517 
518 static int atmel_spi_configure_dma(struct atmel_spi *as)
519 {
520 	struct dma_slave_config	slave_config;
521 	struct device *dev = &as->pdev->dev;
522 	int err;
523 
524 	dma_cap_mask_t mask;
525 	dma_cap_zero(mask);
526 	dma_cap_set(DMA_SLAVE, mask);
527 
528 	as->dma.chan_tx = dma_request_slave_channel_reason(dev, "tx");
529 	if (IS_ERR(as->dma.chan_tx)) {
530 		err = PTR_ERR(as->dma.chan_tx);
531 		if (err == -EPROBE_DEFER) {
532 			dev_warn(dev, "no DMA channel available at the moment\n");
533 			return err;
534 		}
535 		dev_err(dev,
536 			"DMA TX channel not available, SPI unable to use DMA\n");
537 		err = -EBUSY;
538 		goto error;
539 	}
540 
541 	/*
542 	 * No reason to check EPROBE_DEFER here since we have already requested
543 	 * tx channel. If it fails here, it's for another reason.
544 	 */
545 	as->dma.chan_rx = dma_request_slave_channel(dev, "rx");
546 
547 	if (!as->dma.chan_rx) {
548 		dev_err(dev,
549 			"DMA RX channel not available, SPI unable to use DMA\n");
550 		err = -EBUSY;
551 		goto error;
552 	}
553 
554 	err = atmel_spi_dma_slave_config(as, &slave_config, 8);
555 	if (err)
556 		goto error;
557 
558 	dev_info(&as->pdev->dev,
559 			"Using %s (tx) and %s (rx) for DMA transfers\n",
560 			dma_chan_name(as->dma.chan_tx),
561 			dma_chan_name(as->dma.chan_rx));
562 	return 0;
563 error:
564 	if (as->dma.chan_rx)
565 		dma_release_channel(as->dma.chan_rx);
566 	if (!IS_ERR(as->dma.chan_tx))
567 		dma_release_channel(as->dma.chan_tx);
568 	return err;
569 }
570 
571 static void atmel_spi_stop_dma(struct atmel_spi *as)
572 {
573 	if (as->dma.chan_rx)
574 		dmaengine_terminate_all(as->dma.chan_rx);
575 	if (as->dma.chan_tx)
576 		dmaengine_terminate_all(as->dma.chan_tx);
577 }
578 
579 static void atmel_spi_release_dma(struct atmel_spi *as)
580 {
581 	if (as->dma.chan_rx)
582 		dma_release_channel(as->dma.chan_rx);
583 	if (as->dma.chan_tx)
584 		dma_release_channel(as->dma.chan_tx);
585 }
586 
587 /* This function is called by the DMA driver from tasklet context */
588 static void dma_callback(void *data)
589 {
590 	struct spi_master	*master = data;
591 	struct atmel_spi	*as = spi_master_get_devdata(master);
592 
593 	complete(&as->xfer_completion);
594 }
595 
596 /*
597  * Next transfer using PIO without FIFO.
598  */
599 static void atmel_spi_next_xfer_single(struct spi_master *master,
600 				       struct spi_transfer *xfer)
601 {
602 	struct atmel_spi	*as = spi_master_get_devdata(master);
603 	unsigned long xfer_pos = xfer->len - as->current_remaining_bytes;
604 
605 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_pio\n");
606 
607 	/* Make sure data is not remaining in RDR */
608 	spi_readl(as, RDR);
609 	while (spi_readl(as, SR) & SPI_BIT(RDRF)) {
610 		spi_readl(as, RDR);
611 		cpu_relax();
612 	}
613 
614 	if (xfer->tx_buf) {
615 		if (xfer->bits_per_word > 8)
616 			spi_writel(as, TDR, *(u16 *)(xfer->tx_buf + xfer_pos));
617 		else
618 			spi_writel(as, TDR, *(u8 *)(xfer->tx_buf + xfer_pos));
619 	} else {
620 		spi_writel(as, TDR, 0);
621 	}
622 
623 	dev_dbg(master->dev.parent,
624 		"  start pio xfer %p: len %u tx %p rx %p bitpw %d\n",
625 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
626 		xfer->bits_per_word);
627 
628 	/* Enable relevant interrupts */
629 	spi_writel(as, IER, SPI_BIT(RDRF) | SPI_BIT(OVRES));
630 }
631 
632 /*
633  * Next transfer using PIO with FIFO.
634  */
635 static void atmel_spi_next_xfer_fifo(struct spi_master *master,
636 				     struct spi_transfer *xfer)
637 {
638 	struct atmel_spi *as = spi_master_get_devdata(master);
639 	u32 current_remaining_data, num_data;
640 	u32 offset = xfer->len - as->current_remaining_bytes;
641 	const u16 *words = (const u16 *)((u8 *)xfer->tx_buf + offset);
642 	const u8  *bytes = (const u8  *)((u8 *)xfer->tx_buf + offset);
643 	u16 td0, td1;
644 	u32 fifomr;
645 
646 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_fifo\n");
647 
648 	/* Compute the number of data to transfer in the current iteration */
649 	current_remaining_data = ((xfer->bits_per_word > 8) ?
650 				  ((u32)as->current_remaining_bytes >> 1) :
651 				  (u32)as->current_remaining_bytes);
652 	num_data = min(current_remaining_data, as->fifo_size);
653 
654 	/* Flush RX and TX FIFOs */
655 	spi_writel(as, CR, SPI_BIT(RXFCLR) | SPI_BIT(TXFCLR));
656 	while (spi_readl(as, FLR))
657 		cpu_relax();
658 
659 	/* Set RX FIFO Threshold to the number of data to transfer */
660 	fifomr = spi_readl(as, FMR);
661 	spi_writel(as, FMR, SPI_BFINS(RXFTHRES, num_data, fifomr));
662 
663 	/* Clear FIFO flags in the Status Register, especially RXFTHF */
664 	(void)spi_readl(as, SR);
665 
666 	/* Fill TX FIFO */
667 	while (num_data >= 2) {
668 		if (xfer->tx_buf) {
669 			if (xfer->bits_per_word > 8) {
670 				td0 = *words++;
671 				td1 = *words++;
672 			} else {
673 				td0 = *bytes++;
674 				td1 = *bytes++;
675 			}
676 		} else {
677 			td0 = 0;
678 			td1 = 0;
679 		}
680 
681 		spi_writel(as, TDR, (td1 << 16) | td0);
682 		num_data -= 2;
683 	}
684 
685 	if (num_data) {
686 		if (xfer->tx_buf) {
687 			if (xfer->bits_per_word > 8)
688 				td0 = *words++;
689 			else
690 				td0 = *bytes++;
691 		} else {
692 			td0 = 0;
693 		}
694 
695 		spi_writew(as, TDR, td0);
696 		num_data--;
697 	}
698 
699 	dev_dbg(master->dev.parent,
700 		"  start fifo xfer %p: len %u tx %p rx %p bitpw %d\n",
701 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
702 		xfer->bits_per_word);
703 
704 	/*
705 	 * Enable RX FIFO Threshold Flag interrupt to be notified about
706 	 * transfer completion.
707 	 */
708 	spi_writel(as, IER, SPI_BIT(RXFTHF) | SPI_BIT(OVRES));
709 }
710 
711 /*
712  * Next transfer using PIO.
713  */
714 static void atmel_spi_next_xfer_pio(struct spi_master *master,
715 				    struct spi_transfer *xfer)
716 {
717 	struct atmel_spi *as = spi_master_get_devdata(master);
718 
719 	if (as->fifo_size)
720 		atmel_spi_next_xfer_fifo(master, xfer);
721 	else
722 		atmel_spi_next_xfer_single(master, xfer);
723 }
724 
725 /*
726  * Submit next transfer for DMA.
727  */
728 static int atmel_spi_next_xfer_dma_submit(struct spi_master *master,
729 				struct spi_transfer *xfer,
730 				u32 *plen)
731 {
732 	struct atmel_spi	*as = spi_master_get_devdata(master);
733 	struct dma_chan		*rxchan = as->dma.chan_rx;
734 	struct dma_chan		*txchan = as->dma.chan_tx;
735 	struct dma_async_tx_descriptor *rxdesc;
736 	struct dma_async_tx_descriptor *txdesc;
737 	struct dma_slave_config	slave_config;
738 	dma_cookie_t		cookie;
739 	u32	len = *plen;
740 
741 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_dma_submit\n");
742 
743 	/* Check that the channels are available */
744 	if (!rxchan || !txchan)
745 		return -ENODEV;
746 
747 	/* release lock for DMA operations */
748 	atmel_spi_unlock(as);
749 
750 	/* prepare the RX dma transfer */
751 	sg_init_table(&as->dma.sgrx, 1);
752 	if (xfer->rx_buf) {
753 		as->dma.sgrx.dma_address = xfer->rx_dma + xfer->len - *plen;
754 	} else {
755 		as->dma.sgrx.dma_address = as->buffer_dma;
756 		if (len > BUFFER_SIZE)
757 			len = BUFFER_SIZE;
758 	}
759 
760 	/* prepare the TX dma transfer */
761 	sg_init_table(&as->dma.sgtx, 1);
762 	if (xfer->tx_buf) {
763 		as->dma.sgtx.dma_address = xfer->tx_dma + xfer->len - *plen;
764 	} else {
765 		as->dma.sgtx.dma_address = as->buffer_dma;
766 		if (len > BUFFER_SIZE)
767 			len = BUFFER_SIZE;
768 		memset(as->buffer, 0, len);
769 	}
770 
771 	sg_dma_len(&as->dma.sgtx) = len;
772 	sg_dma_len(&as->dma.sgrx) = len;
773 
774 	*plen = len;
775 
776 	if (atmel_spi_dma_slave_config(as, &slave_config,
777 				       xfer->bits_per_word))
778 		goto err_exit;
779 
780 	/* Send both scatterlists */
781 	rxdesc = dmaengine_prep_slave_sg(rxchan, &as->dma.sgrx, 1,
782 					 DMA_FROM_DEVICE,
783 					 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
784 	if (!rxdesc)
785 		goto err_dma;
786 
787 	txdesc = dmaengine_prep_slave_sg(txchan, &as->dma.sgtx, 1,
788 					 DMA_TO_DEVICE,
789 					 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
790 	if (!txdesc)
791 		goto err_dma;
792 
793 	dev_dbg(master->dev.parent,
794 		"  start dma xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
795 		xfer, xfer->len, xfer->tx_buf, (unsigned long long)xfer->tx_dma,
796 		xfer->rx_buf, (unsigned long long)xfer->rx_dma);
797 
798 	/* Enable relevant interrupts */
799 	spi_writel(as, IER, SPI_BIT(OVRES));
800 
801 	/* Put the callback on the RX transfer only, that should finish last */
802 	rxdesc->callback = dma_callback;
803 	rxdesc->callback_param = master;
804 
805 	/* Submit and fire RX and TX with TX last so we're ready to read! */
806 	cookie = rxdesc->tx_submit(rxdesc);
807 	if (dma_submit_error(cookie))
808 		goto err_dma;
809 	cookie = txdesc->tx_submit(txdesc);
810 	if (dma_submit_error(cookie))
811 		goto err_dma;
812 	rxchan->device->device_issue_pending(rxchan);
813 	txchan->device->device_issue_pending(txchan);
814 
815 	/* take back lock */
816 	atmel_spi_lock(as);
817 	return 0;
818 
819 err_dma:
820 	spi_writel(as, IDR, SPI_BIT(OVRES));
821 	atmel_spi_stop_dma(as);
822 err_exit:
823 	atmel_spi_lock(as);
824 	return -ENOMEM;
825 }
826 
827 static void atmel_spi_next_xfer_data(struct spi_master *master,
828 				struct spi_transfer *xfer,
829 				dma_addr_t *tx_dma,
830 				dma_addr_t *rx_dma,
831 				u32 *plen)
832 {
833 	struct atmel_spi	*as = spi_master_get_devdata(master);
834 	u32			len = *plen;
835 
836 	/* use scratch buffer only when rx or tx data is unspecified */
837 	if (xfer->rx_buf)
838 		*rx_dma = xfer->rx_dma + xfer->len - *plen;
839 	else {
840 		*rx_dma = as->buffer_dma;
841 		if (len > BUFFER_SIZE)
842 			len = BUFFER_SIZE;
843 	}
844 
845 	if (xfer->tx_buf)
846 		*tx_dma = xfer->tx_dma + xfer->len - *plen;
847 	else {
848 		*tx_dma = as->buffer_dma;
849 		if (len > BUFFER_SIZE)
850 			len = BUFFER_SIZE;
851 		memset(as->buffer, 0, len);
852 		dma_sync_single_for_device(&as->pdev->dev,
853 				as->buffer_dma, len, DMA_TO_DEVICE);
854 	}
855 
856 	*plen = len;
857 }
858 
859 static int atmel_spi_set_xfer_speed(struct atmel_spi *as,
860 				    struct spi_device *spi,
861 				    struct spi_transfer *xfer)
862 {
863 	u32			scbr, csr;
864 	unsigned long		bus_hz;
865 
866 	/* v1 chips start out at half the peripheral bus speed. */
867 	bus_hz = clk_get_rate(as->clk);
868 	if (!atmel_spi_is_v2(as))
869 		bus_hz /= 2;
870 
871 	/*
872 	 * Calculate the lowest divider that satisfies the
873 	 * constraint, assuming div32/fdiv/mbz == 0.
874 	 */
875 	scbr = DIV_ROUND_UP(bus_hz, xfer->speed_hz);
876 
877 	/*
878 	 * If the resulting divider doesn't fit into the
879 	 * register bitfield, we can't satisfy the constraint.
880 	 */
881 	if (scbr >= (1 << SPI_SCBR_SIZE)) {
882 		dev_err(&spi->dev,
883 			"setup: %d Hz too slow, scbr %u; min %ld Hz\n",
884 			xfer->speed_hz, scbr, bus_hz/255);
885 		return -EINVAL;
886 	}
887 	if (scbr == 0) {
888 		dev_err(&spi->dev,
889 			"setup: %d Hz too high, scbr %u; max %ld Hz\n",
890 			xfer->speed_hz, scbr, bus_hz);
891 		return -EINVAL;
892 	}
893 	csr = spi_readl(as, CSR0 + 4 * spi->chip_select);
894 	csr = SPI_BFINS(SCBR, scbr, csr);
895 	spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
896 
897 	return 0;
898 }
899 
900 /*
901  * Submit next transfer for PDC.
902  * lock is held, spi irq is blocked
903  */
904 static void atmel_spi_pdc_next_xfer(struct spi_master *master,
905 					struct spi_message *msg,
906 					struct spi_transfer *xfer)
907 {
908 	struct atmel_spi	*as = spi_master_get_devdata(master);
909 	u32			len;
910 	dma_addr_t		tx_dma, rx_dma;
911 
912 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
913 
914 	len = as->current_remaining_bytes;
915 	atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
916 	as->current_remaining_bytes -= len;
917 
918 	spi_writel(as, RPR, rx_dma);
919 	spi_writel(as, TPR, tx_dma);
920 
921 	if (msg->spi->bits_per_word > 8)
922 		len >>= 1;
923 	spi_writel(as, RCR, len);
924 	spi_writel(as, TCR, len);
925 
926 	dev_dbg(&msg->spi->dev,
927 		"  start xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
928 		xfer, xfer->len, xfer->tx_buf,
929 		(unsigned long long)xfer->tx_dma, xfer->rx_buf,
930 		(unsigned long long)xfer->rx_dma);
931 
932 	if (as->current_remaining_bytes) {
933 		len = as->current_remaining_bytes;
934 		atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
935 		as->current_remaining_bytes -= len;
936 
937 		spi_writel(as, RNPR, rx_dma);
938 		spi_writel(as, TNPR, tx_dma);
939 
940 		if (msg->spi->bits_per_word > 8)
941 			len >>= 1;
942 		spi_writel(as, RNCR, len);
943 		spi_writel(as, TNCR, len);
944 
945 		dev_dbg(&msg->spi->dev,
946 			"  next xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
947 			xfer, xfer->len, xfer->tx_buf,
948 			(unsigned long long)xfer->tx_dma, xfer->rx_buf,
949 			(unsigned long long)xfer->rx_dma);
950 	}
951 
952 	/* REVISIT: We're waiting for RXBUFF before we start the next
953 	 * transfer because we need to handle some difficult timing
954 	 * issues otherwise. If we wait for TXBUFE in one transfer and
955 	 * then starts waiting for RXBUFF in the next, it's difficult
956 	 * to tell the difference between the RXBUFF interrupt we're
957 	 * actually waiting for and the RXBUFF interrupt of the
958 	 * previous transfer.
959 	 *
960 	 * It should be doable, though. Just not now...
961 	 */
962 	spi_writel(as, IER, SPI_BIT(RXBUFF) | SPI_BIT(OVRES));
963 	spi_writel(as, PTCR, SPI_BIT(TXTEN) | SPI_BIT(RXTEN));
964 }
965 
966 /*
967  * For DMA, tx_buf/tx_dma have the same relationship as rx_buf/rx_dma:
968  *  - The buffer is either valid for CPU access, else NULL
969  *  - If the buffer is valid, so is its DMA address
970  *
971  * This driver manages the dma address unless message->is_dma_mapped.
972  */
973 static int
974 atmel_spi_dma_map_xfer(struct atmel_spi *as, struct spi_transfer *xfer)
975 {
976 	struct device	*dev = &as->pdev->dev;
977 
978 	xfer->tx_dma = xfer->rx_dma = INVALID_DMA_ADDRESS;
979 	if (xfer->tx_buf) {
980 		/* tx_buf is a const void* where we need a void * for the dma
981 		 * mapping */
982 		void *nonconst_tx = (void *)xfer->tx_buf;
983 
984 		xfer->tx_dma = dma_map_single(dev,
985 				nonconst_tx, xfer->len,
986 				DMA_TO_DEVICE);
987 		if (dma_mapping_error(dev, xfer->tx_dma))
988 			return -ENOMEM;
989 	}
990 	if (xfer->rx_buf) {
991 		xfer->rx_dma = dma_map_single(dev,
992 				xfer->rx_buf, xfer->len,
993 				DMA_FROM_DEVICE);
994 		if (dma_mapping_error(dev, xfer->rx_dma)) {
995 			if (xfer->tx_buf)
996 				dma_unmap_single(dev,
997 						xfer->tx_dma, xfer->len,
998 						DMA_TO_DEVICE);
999 			return -ENOMEM;
1000 		}
1001 	}
1002 	return 0;
1003 }
1004 
1005 static void atmel_spi_dma_unmap_xfer(struct spi_master *master,
1006 				     struct spi_transfer *xfer)
1007 {
1008 	if (xfer->tx_dma != INVALID_DMA_ADDRESS)
1009 		dma_unmap_single(master->dev.parent, xfer->tx_dma,
1010 				 xfer->len, DMA_TO_DEVICE);
1011 	if (xfer->rx_dma != INVALID_DMA_ADDRESS)
1012 		dma_unmap_single(master->dev.parent, xfer->rx_dma,
1013 				 xfer->len, DMA_FROM_DEVICE);
1014 }
1015 
1016 static void atmel_spi_disable_pdc_transfer(struct atmel_spi *as)
1017 {
1018 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1019 }
1020 
1021 static void
1022 atmel_spi_pump_single_data(struct atmel_spi *as, struct spi_transfer *xfer)
1023 {
1024 	u8		*rxp;
1025 	u16		*rxp16;
1026 	unsigned long	xfer_pos = xfer->len - as->current_remaining_bytes;
1027 
1028 	if (xfer->rx_buf) {
1029 		if (xfer->bits_per_word > 8) {
1030 			rxp16 = (u16 *)(((u8 *)xfer->rx_buf) + xfer_pos);
1031 			*rxp16 = spi_readl(as, RDR);
1032 		} else {
1033 			rxp = ((u8 *)xfer->rx_buf) + xfer_pos;
1034 			*rxp = spi_readl(as, RDR);
1035 		}
1036 	} else {
1037 		spi_readl(as, RDR);
1038 	}
1039 	if (xfer->bits_per_word > 8) {
1040 		if (as->current_remaining_bytes > 2)
1041 			as->current_remaining_bytes -= 2;
1042 		else
1043 			as->current_remaining_bytes = 0;
1044 	} else {
1045 		as->current_remaining_bytes--;
1046 	}
1047 }
1048 
1049 static void
1050 atmel_spi_pump_fifo_data(struct atmel_spi *as, struct spi_transfer *xfer)
1051 {
1052 	u32 fifolr = spi_readl(as, FLR);
1053 	u32 num_bytes, num_data = SPI_BFEXT(RXFL, fifolr);
1054 	u32 offset = xfer->len - as->current_remaining_bytes;
1055 	u16 *words = (u16 *)((u8 *)xfer->rx_buf + offset);
1056 	u8  *bytes = (u8  *)((u8 *)xfer->rx_buf + offset);
1057 	u16 rd; /* RD field is the lowest 16 bits of RDR */
1058 
1059 	/* Update the number of remaining bytes to transfer */
1060 	num_bytes = ((xfer->bits_per_word > 8) ?
1061 		     (num_data << 1) :
1062 		     num_data);
1063 
1064 	if (as->current_remaining_bytes > num_bytes)
1065 		as->current_remaining_bytes -= num_bytes;
1066 	else
1067 		as->current_remaining_bytes = 0;
1068 
1069 	/* Handle odd number of bytes when data are more than 8bit width */
1070 	if (xfer->bits_per_word > 8)
1071 		as->current_remaining_bytes &= ~0x1;
1072 
1073 	/* Read data */
1074 	while (num_data) {
1075 		rd = spi_readl(as, RDR);
1076 		if (xfer->rx_buf) {
1077 			if (xfer->bits_per_word > 8)
1078 				*words++ = rd;
1079 			else
1080 				*bytes++ = rd;
1081 		}
1082 		num_data--;
1083 	}
1084 }
1085 
1086 /* Called from IRQ
1087  *
1088  * Must update "current_remaining_bytes" to keep track of data
1089  * to transfer.
1090  */
1091 static void
1092 atmel_spi_pump_pio_data(struct atmel_spi *as, struct spi_transfer *xfer)
1093 {
1094 	if (as->fifo_size)
1095 		atmel_spi_pump_fifo_data(as, xfer);
1096 	else
1097 		atmel_spi_pump_single_data(as, xfer);
1098 }
1099 
1100 /* Interrupt
1101  *
1102  * No need for locking in this Interrupt handler: done_status is the
1103  * only information modified.
1104  */
1105 static irqreturn_t
1106 atmel_spi_pio_interrupt(int irq, void *dev_id)
1107 {
1108 	struct spi_master	*master = dev_id;
1109 	struct atmel_spi	*as = spi_master_get_devdata(master);
1110 	u32			status, pending, imr;
1111 	struct spi_transfer	*xfer;
1112 	int			ret = IRQ_NONE;
1113 
1114 	imr = spi_readl(as, IMR);
1115 	status = spi_readl(as, SR);
1116 	pending = status & imr;
1117 
1118 	if (pending & SPI_BIT(OVRES)) {
1119 		ret = IRQ_HANDLED;
1120 		spi_writel(as, IDR, SPI_BIT(OVRES));
1121 		dev_warn(master->dev.parent, "overrun\n");
1122 
1123 		/*
1124 		 * When we get an overrun, we disregard the current
1125 		 * transfer. Data will not be copied back from any
1126 		 * bounce buffer and msg->actual_len will not be
1127 		 * updated with the last xfer.
1128 		 *
1129 		 * We will also not process any remaning transfers in
1130 		 * the message.
1131 		 */
1132 		as->done_status = -EIO;
1133 		smp_wmb();
1134 
1135 		/* Clear any overrun happening while cleaning up */
1136 		spi_readl(as, SR);
1137 
1138 		complete(&as->xfer_completion);
1139 
1140 	} else if (pending & (SPI_BIT(RDRF) | SPI_BIT(RXFTHF))) {
1141 		atmel_spi_lock(as);
1142 
1143 		if (as->current_remaining_bytes) {
1144 			ret = IRQ_HANDLED;
1145 			xfer = as->current_transfer;
1146 			atmel_spi_pump_pio_data(as, xfer);
1147 			if (!as->current_remaining_bytes)
1148 				spi_writel(as, IDR, pending);
1149 
1150 			complete(&as->xfer_completion);
1151 		}
1152 
1153 		atmel_spi_unlock(as);
1154 	} else {
1155 		WARN_ONCE(pending, "IRQ not handled, pending = %x\n", pending);
1156 		ret = IRQ_HANDLED;
1157 		spi_writel(as, IDR, pending);
1158 	}
1159 
1160 	return ret;
1161 }
1162 
1163 static irqreturn_t
1164 atmel_spi_pdc_interrupt(int irq, void *dev_id)
1165 {
1166 	struct spi_master	*master = dev_id;
1167 	struct atmel_spi	*as = spi_master_get_devdata(master);
1168 	u32			status, pending, imr;
1169 	int			ret = IRQ_NONE;
1170 
1171 	imr = spi_readl(as, IMR);
1172 	status = spi_readl(as, SR);
1173 	pending = status & imr;
1174 
1175 	if (pending & SPI_BIT(OVRES)) {
1176 
1177 		ret = IRQ_HANDLED;
1178 
1179 		spi_writel(as, IDR, (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX)
1180 				     | SPI_BIT(OVRES)));
1181 
1182 		/* Clear any overrun happening while cleaning up */
1183 		spi_readl(as, SR);
1184 
1185 		as->done_status = -EIO;
1186 
1187 		complete(&as->xfer_completion);
1188 
1189 	} else if (pending & (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX))) {
1190 		ret = IRQ_HANDLED;
1191 
1192 		spi_writel(as, IDR, pending);
1193 
1194 		complete(&as->xfer_completion);
1195 	}
1196 
1197 	return ret;
1198 }
1199 
1200 static int atmel_spi_setup(struct spi_device *spi)
1201 {
1202 	struct atmel_spi	*as;
1203 	struct atmel_spi_device	*asd;
1204 	u32			csr;
1205 	unsigned int		bits = spi->bits_per_word;
1206 	unsigned int		npcs_pin;
1207 	int			ret;
1208 
1209 	as = spi_master_get_devdata(spi->master);
1210 
1211 	/* see notes above re chipselect */
1212 	if (!atmel_spi_is_v2(as)
1213 			&& spi->chip_select == 0
1214 			&& (spi->mode & SPI_CS_HIGH)) {
1215 		dev_dbg(&spi->dev, "setup: can't be active-high\n");
1216 		return -EINVAL;
1217 	}
1218 
1219 	csr = SPI_BF(BITS, bits - 8);
1220 	if (spi->mode & SPI_CPOL)
1221 		csr |= SPI_BIT(CPOL);
1222 	if (!(spi->mode & SPI_CPHA))
1223 		csr |= SPI_BIT(NCPHA);
1224 	if (!as->use_cs_gpios)
1225 		csr |= SPI_BIT(CSAAT);
1226 
1227 	/* DLYBS is mostly irrelevant since we manage chipselect using GPIOs.
1228 	 *
1229 	 * DLYBCT would add delays between words, slowing down transfers.
1230 	 * It could potentially be useful to cope with DMA bottlenecks, but
1231 	 * in those cases it's probably best to just use a lower bitrate.
1232 	 */
1233 	csr |= SPI_BF(DLYBS, 0);
1234 	csr |= SPI_BF(DLYBCT, 0);
1235 
1236 	/* chipselect must have been muxed as GPIO (e.g. in board setup) */
1237 	npcs_pin = (unsigned long)spi->controller_data;
1238 
1239 	if (!as->use_cs_gpios)
1240 		npcs_pin = spi->chip_select;
1241 	else if (gpio_is_valid(spi->cs_gpio))
1242 		npcs_pin = spi->cs_gpio;
1243 
1244 	asd = spi->controller_state;
1245 	if (!asd) {
1246 		asd = kzalloc(sizeof(struct atmel_spi_device), GFP_KERNEL);
1247 		if (!asd)
1248 			return -ENOMEM;
1249 
1250 		if (as->use_cs_gpios) {
1251 			ret = gpio_request(npcs_pin, dev_name(&spi->dev));
1252 			if (ret) {
1253 				kfree(asd);
1254 				return ret;
1255 			}
1256 
1257 			gpio_direction_output(npcs_pin,
1258 					      !(spi->mode & SPI_CS_HIGH));
1259 		}
1260 
1261 		asd->npcs_pin = npcs_pin;
1262 		spi->controller_state = asd;
1263 	}
1264 
1265 	asd->csr = csr;
1266 
1267 	dev_dbg(&spi->dev,
1268 		"setup: bpw %u mode 0x%x -> csr%d %08x\n",
1269 		bits, spi->mode, spi->chip_select, csr);
1270 
1271 	if (!atmel_spi_is_v2(as))
1272 		spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
1273 
1274 	return 0;
1275 }
1276 
1277 static int atmel_spi_one_transfer(struct spi_master *master,
1278 					struct spi_message *msg,
1279 					struct spi_transfer *xfer)
1280 {
1281 	struct atmel_spi	*as;
1282 	struct spi_device	*spi = msg->spi;
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 	if (!(xfer->tx_buf || xfer->rx_buf) && xfer->len) {
1293 		dev_dbg(&spi->dev, "missing rx or tx buf\n");
1294 		return -EINVAL;
1295 	}
1296 
1297 	asd = spi->controller_state;
1298 	bits = (asd->csr >> 4) & 0xf;
1299 	if (bits != xfer->bits_per_word - 8) {
1300 		dev_dbg(&spi->dev,
1301 			"you can't yet change bits_per_word in transfers\n");
1302 		return -ENOPROTOOPT;
1303 	}
1304 
1305 	/*
1306 	 * DMA map early, for performance (empties dcache ASAP) and
1307 	 * better fault reporting.
1308 	 */
1309 	if ((!msg->is_dma_mapped)
1310 		&& (atmel_spi_use_dma(as, xfer)	|| as->use_pdc)) {
1311 		if (atmel_spi_dma_map_xfer(as, xfer) < 0)
1312 			return -ENOMEM;
1313 	}
1314 
1315 	atmel_spi_set_xfer_speed(as, msg->spi, xfer);
1316 
1317 	as->done_status = 0;
1318 	as->current_transfer = xfer;
1319 	as->current_remaining_bytes = xfer->len;
1320 	while (as->current_remaining_bytes) {
1321 		reinit_completion(&as->xfer_completion);
1322 
1323 		if (as->use_pdc) {
1324 			atmel_spi_pdc_next_xfer(master, msg, xfer);
1325 		} else if (atmel_spi_use_dma(as, xfer)) {
1326 			len = as->current_remaining_bytes;
1327 			ret = atmel_spi_next_xfer_dma_submit(master,
1328 								xfer, &len);
1329 			if (ret) {
1330 				dev_err(&spi->dev,
1331 					"unable to use DMA, fallback to PIO\n");
1332 				atmel_spi_next_xfer_pio(master, xfer);
1333 			} else {
1334 				as->current_remaining_bytes -= len;
1335 				if (as->current_remaining_bytes < 0)
1336 					as->current_remaining_bytes = 0;
1337 			}
1338 		} else {
1339 			atmel_spi_next_xfer_pio(master, xfer);
1340 		}
1341 
1342 		/* interrupts are disabled, so free the lock for schedule */
1343 		atmel_spi_unlock(as);
1344 		dma_timeout = wait_for_completion_timeout(&as->xfer_completion,
1345 							  SPI_DMA_TIMEOUT);
1346 		atmel_spi_lock(as);
1347 		if (WARN_ON(dma_timeout == 0)) {
1348 			dev_err(&spi->dev, "spi transfer timeout\n");
1349 			as->done_status = -EIO;
1350 		}
1351 
1352 		if (as->done_status)
1353 			break;
1354 	}
1355 
1356 	if (as->done_status) {
1357 		if (as->use_pdc) {
1358 			dev_warn(master->dev.parent,
1359 				"overrun (%u/%u remaining)\n",
1360 				spi_readl(as, TCR), spi_readl(as, RCR));
1361 
1362 			/*
1363 			 * Clean up DMA registers and make sure the data
1364 			 * registers are empty.
1365 			 */
1366 			spi_writel(as, RNCR, 0);
1367 			spi_writel(as, TNCR, 0);
1368 			spi_writel(as, RCR, 0);
1369 			spi_writel(as, TCR, 0);
1370 			for (timeout = 1000; timeout; timeout--)
1371 				if (spi_readl(as, SR) & SPI_BIT(TXEMPTY))
1372 					break;
1373 			if (!timeout)
1374 				dev_warn(master->dev.parent,
1375 					 "timeout waiting for TXEMPTY");
1376 			while (spi_readl(as, SR) & SPI_BIT(RDRF))
1377 				spi_readl(as, RDR);
1378 
1379 			/* Clear any overrun happening while cleaning up */
1380 			spi_readl(as, SR);
1381 
1382 		} else if (atmel_spi_use_dma(as, xfer)) {
1383 			atmel_spi_stop_dma(as);
1384 		}
1385 
1386 		if (!msg->is_dma_mapped
1387 			&& (atmel_spi_use_dma(as, xfer) || as->use_pdc))
1388 			atmel_spi_dma_unmap_xfer(master, xfer);
1389 
1390 		return 0;
1391 
1392 	} else {
1393 		/* only update length if no error */
1394 		msg->actual_length += xfer->len;
1395 	}
1396 
1397 	if (!msg->is_dma_mapped
1398 		&& (atmel_spi_use_dma(as, xfer) || as->use_pdc))
1399 		atmel_spi_dma_unmap_xfer(master, xfer);
1400 
1401 	if (xfer->delay_usecs)
1402 		udelay(xfer->delay_usecs);
1403 
1404 	if (xfer->cs_change) {
1405 		if (list_is_last(&xfer->transfer_list,
1406 				 &msg->transfers)) {
1407 			as->keep_cs = true;
1408 		} else {
1409 			as->cs_active = !as->cs_active;
1410 			if (as->cs_active)
1411 				cs_activate(as, msg->spi);
1412 			else
1413 				cs_deactivate(as, msg->spi);
1414 		}
1415 	}
1416 
1417 	return 0;
1418 }
1419 
1420 static int atmel_spi_transfer_one_message(struct spi_master *master,
1421 						struct spi_message *msg)
1422 {
1423 	struct atmel_spi *as;
1424 	struct spi_transfer *xfer;
1425 	struct spi_device *spi = msg->spi;
1426 	int ret = 0;
1427 
1428 	as = spi_master_get_devdata(master);
1429 
1430 	dev_dbg(&spi->dev, "new message %p submitted for %s\n",
1431 					msg, dev_name(&spi->dev));
1432 
1433 	atmel_spi_lock(as);
1434 	cs_activate(as, spi);
1435 
1436 	as->cs_active = true;
1437 	as->keep_cs = false;
1438 
1439 	msg->status = 0;
1440 	msg->actual_length = 0;
1441 
1442 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1443 		ret = atmel_spi_one_transfer(master, msg, xfer);
1444 		if (ret)
1445 			goto msg_done;
1446 	}
1447 
1448 	if (as->use_pdc)
1449 		atmel_spi_disable_pdc_transfer(as);
1450 
1451 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1452 		dev_dbg(&spi->dev,
1453 			"  xfer %p: len %u tx %p/%pad rx %p/%pad\n",
1454 			xfer, xfer->len,
1455 			xfer->tx_buf, &xfer->tx_dma,
1456 			xfer->rx_buf, &xfer->rx_dma);
1457 	}
1458 
1459 msg_done:
1460 	if (!as->keep_cs)
1461 		cs_deactivate(as, msg->spi);
1462 
1463 	atmel_spi_unlock(as);
1464 
1465 	msg->status = as->done_status;
1466 	spi_finalize_current_message(spi->master);
1467 
1468 	return ret;
1469 }
1470 
1471 static void atmel_spi_cleanup(struct spi_device *spi)
1472 {
1473 	struct atmel_spi_device	*asd = spi->controller_state;
1474 	unsigned		gpio = (unsigned long) spi->controller_data;
1475 
1476 	if (!asd)
1477 		return;
1478 
1479 	spi->controller_state = NULL;
1480 	gpio_free(gpio);
1481 	kfree(asd);
1482 }
1483 
1484 static inline unsigned int atmel_get_version(struct atmel_spi *as)
1485 {
1486 	return spi_readl(as, VERSION) & 0x00000fff;
1487 }
1488 
1489 static void atmel_get_caps(struct atmel_spi *as)
1490 {
1491 	unsigned int version;
1492 
1493 	version = atmel_get_version(as);
1494 	dev_info(&as->pdev->dev, "version: 0x%x\n", version);
1495 
1496 	as->caps.is_spi2 = version > 0x121;
1497 	as->caps.has_wdrbt = version >= 0x210;
1498 	as->caps.has_dma_support = version >= 0x212;
1499 }
1500 
1501 /*-------------------------------------------------------------------------*/
1502 
1503 static int atmel_spi_probe(struct platform_device *pdev)
1504 {
1505 	struct resource		*regs;
1506 	int			irq;
1507 	struct clk		*clk;
1508 	int			ret;
1509 	struct spi_master	*master;
1510 	struct atmel_spi	*as;
1511 
1512 	/* Select default pin state */
1513 	pinctrl_pm_select_default_state(&pdev->dev);
1514 
1515 	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1516 	if (!regs)
1517 		return -ENXIO;
1518 
1519 	irq = platform_get_irq(pdev, 0);
1520 	if (irq < 0)
1521 		return irq;
1522 
1523 	clk = devm_clk_get(&pdev->dev, "spi_clk");
1524 	if (IS_ERR(clk))
1525 		return PTR_ERR(clk);
1526 
1527 	/* setup spi core then atmel-specific driver state */
1528 	ret = -ENOMEM;
1529 	master = spi_alloc_master(&pdev->dev, sizeof(*as));
1530 	if (!master)
1531 		goto out_free;
1532 
1533 	/* the spi->mode bits understood by this driver: */
1534 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
1535 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(8, 16);
1536 	master->dev.of_node = pdev->dev.of_node;
1537 	master->bus_num = pdev->id;
1538 	master->num_chipselect = master->dev.of_node ? 0 : 4;
1539 	master->setup = atmel_spi_setup;
1540 	master->transfer_one_message = atmel_spi_transfer_one_message;
1541 	master->cleanup = atmel_spi_cleanup;
1542 	master->auto_runtime_pm = true;
1543 	platform_set_drvdata(pdev, master);
1544 
1545 	as = spi_master_get_devdata(master);
1546 
1547 	/*
1548 	 * Scratch buffer is used for throwaway rx and tx data.
1549 	 * It's coherent to minimize dcache pollution.
1550 	 */
1551 	as->buffer = dma_alloc_coherent(&pdev->dev, BUFFER_SIZE,
1552 					&as->buffer_dma, GFP_KERNEL);
1553 	if (!as->buffer)
1554 		goto out_free;
1555 
1556 	spin_lock_init(&as->lock);
1557 
1558 	as->pdev = pdev;
1559 	as->regs = devm_ioremap_resource(&pdev->dev, regs);
1560 	if (IS_ERR(as->regs)) {
1561 		ret = PTR_ERR(as->regs);
1562 		goto out_free_buffer;
1563 	}
1564 	as->phybase = regs->start;
1565 	as->irq = irq;
1566 	as->clk = clk;
1567 
1568 	init_completion(&as->xfer_completion);
1569 
1570 	atmel_get_caps(as);
1571 
1572 	as->use_cs_gpios = true;
1573 	if (atmel_spi_is_v2(as) &&
1574 	    pdev->dev.of_node &&
1575 	    !of_get_property(pdev->dev.of_node, "cs-gpios", NULL)) {
1576 		as->use_cs_gpios = false;
1577 		master->num_chipselect = 4;
1578 	}
1579 
1580 	as->use_dma = false;
1581 	as->use_pdc = false;
1582 	if (as->caps.has_dma_support) {
1583 		ret = atmel_spi_configure_dma(as);
1584 		if (ret == 0)
1585 			as->use_dma = true;
1586 		else if (ret == -EPROBE_DEFER)
1587 			return ret;
1588 	} else {
1589 		as->use_pdc = true;
1590 	}
1591 
1592 	if (as->caps.has_dma_support && !as->use_dma)
1593 		dev_info(&pdev->dev, "Atmel SPI Controller using PIO only\n");
1594 
1595 	if (as->use_pdc) {
1596 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pdc_interrupt,
1597 					0, dev_name(&pdev->dev), master);
1598 	} else {
1599 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pio_interrupt,
1600 					0, dev_name(&pdev->dev), master);
1601 	}
1602 	if (ret)
1603 		goto out_unmap_regs;
1604 
1605 	/* Initialize the hardware */
1606 	ret = clk_prepare_enable(clk);
1607 	if (ret)
1608 		goto out_free_irq;
1609 	spi_writel(as, CR, SPI_BIT(SWRST));
1610 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1611 	if (as->caps.has_wdrbt) {
1612 		spi_writel(as, MR, SPI_BIT(WDRBT) | SPI_BIT(MODFDIS)
1613 				| SPI_BIT(MSTR));
1614 	} else {
1615 		spi_writel(as, MR, SPI_BIT(MSTR) | SPI_BIT(MODFDIS));
1616 	}
1617 
1618 	if (as->use_pdc)
1619 		spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1620 	spi_writel(as, CR, SPI_BIT(SPIEN));
1621 
1622 	as->fifo_size = 0;
1623 	if (!of_property_read_u32(pdev->dev.of_node, "atmel,fifo-size",
1624 				  &as->fifo_size)) {
1625 		dev_info(&pdev->dev, "Using FIFO (%u data)\n", as->fifo_size);
1626 		spi_writel(as, CR, SPI_BIT(FIFOEN));
1627 	}
1628 
1629 	/* go! */
1630 	dev_info(&pdev->dev, "Atmel SPI Controller at 0x%08lx (irq %d)\n",
1631 			(unsigned long)regs->start, irq);
1632 
1633 	pm_runtime_set_autosuspend_delay(&pdev->dev, AUTOSUSPEND_TIMEOUT);
1634 	pm_runtime_use_autosuspend(&pdev->dev);
1635 	pm_runtime_set_active(&pdev->dev);
1636 	pm_runtime_enable(&pdev->dev);
1637 
1638 	ret = devm_spi_register_master(&pdev->dev, master);
1639 	if (ret)
1640 		goto out_free_dma;
1641 
1642 	return 0;
1643 
1644 out_free_dma:
1645 	pm_runtime_disable(&pdev->dev);
1646 	pm_runtime_set_suspended(&pdev->dev);
1647 
1648 	if (as->use_dma)
1649 		atmel_spi_release_dma(as);
1650 
1651 	spi_writel(as, CR, SPI_BIT(SWRST));
1652 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1653 	clk_disable_unprepare(clk);
1654 out_free_irq:
1655 out_unmap_regs:
1656 out_free_buffer:
1657 	dma_free_coherent(&pdev->dev, BUFFER_SIZE, as->buffer,
1658 			as->buffer_dma);
1659 out_free:
1660 	spi_master_put(master);
1661 	return ret;
1662 }
1663 
1664 static int atmel_spi_remove(struct platform_device *pdev)
1665 {
1666 	struct spi_master	*master = platform_get_drvdata(pdev);
1667 	struct atmel_spi	*as = spi_master_get_devdata(master);
1668 
1669 	pm_runtime_get_sync(&pdev->dev);
1670 
1671 	/* reset the hardware and block queue progress */
1672 	spin_lock_irq(&as->lock);
1673 	if (as->use_dma) {
1674 		atmel_spi_stop_dma(as);
1675 		atmel_spi_release_dma(as);
1676 	}
1677 
1678 	spi_writel(as, CR, SPI_BIT(SWRST));
1679 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1680 	spi_readl(as, SR);
1681 	spin_unlock_irq(&as->lock);
1682 
1683 	dma_free_coherent(&pdev->dev, BUFFER_SIZE, as->buffer,
1684 			as->buffer_dma);
1685 
1686 	clk_disable_unprepare(as->clk);
1687 
1688 	pm_runtime_put_noidle(&pdev->dev);
1689 	pm_runtime_disable(&pdev->dev);
1690 
1691 	return 0;
1692 }
1693 
1694 #ifdef CONFIG_PM
1695 static int atmel_spi_runtime_suspend(struct device *dev)
1696 {
1697 	struct spi_master *master = dev_get_drvdata(dev);
1698 	struct atmel_spi *as = spi_master_get_devdata(master);
1699 
1700 	clk_disable_unprepare(as->clk);
1701 	pinctrl_pm_select_sleep_state(dev);
1702 
1703 	return 0;
1704 }
1705 
1706 static int atmel_spi_runtime_resume(struct device *dev)
1707 {
1708 	struct spi_master *master = dev_get_drvdata(dev);
1709 	struct atmel_spi *as = spi_master_get_devdata(master);
1710 
1711 	pinctrl_pm_select_default_state(dev);
1712 
1713 	return clk_prepare_enable(as->clk);
1714 }
1715 
1716 #ifdef CONFIG_PM_SLEEP
1717 static int atmel_spi_suspend(struct device *dev)
1718 {
1719 	struct spi_master *master = dev_get_drvdata(dev);
1720 	int ret;
1721 
1722 	/* Stop the queue running */
1723 	ret = spi_master_suspend(master);
1724 	if (ret) {
1725 		dev_warn(dev, "cannot suspend master\n");
1726 		return ret;
1727 	}
1728 
1729 	if (!pm_runtime_suspended(dev))
1730 		atmel_spi_runtime_suspend(dev);
1731 
1732 	return 0;
1733 }
1734 
1735 static int atmel_spi_resume(struct device *dev)
1736 {
1737 	struct spi_master *master = dev_get_drvdata(dev);
1738 	int ret;
1739 
1740 	if (!pm_runtime_suspended(dev)) {
1741 		ret = atmel_spi_runtime_resume(dev);
1742 		if (ret)
1743 			return ret;
1744 	}
1745 
1746 	/* Start the queue running */
1747 	ret = spi_master_resume(master);
1748 	if (ret)
1749 		dev_err(dev, "problem starting queue (%d)\n", ret);
1750 
1751 	return ret;
1752 }
1753 #endif
1754 
1755 static const struct dev_pm_ops atmel_spi_pm_ops = {
1756 	SET_SYSTEM_SLEEP_PM_OPS(atmel_spi_suspend, atmel_spi_resume)
1757 	SET_RUNTIME_PM_OPS(atmel_spi_runtime_suspend,
1758 			   atmel_spi_runtime_resume, NULL)
1759 };
1760 #define ATMEL_SPI_PM_OPS	(&atmel_spi_pm_ops)
1761 #else
1762 #define ATMEL_SPI_PM_OPS	NULL
1763 #endif
1764 
1765 #if defined(CONFIG_OF)
1766 static const struct of_device_id atmel_spi_dt_ids[] = {
1767 	{ .compatible = "atmel,at91rm9200-spi" },
1768 	{ /* sentinel */ }
1769 };
1770 
1771 MODULE_DEVICE_TABLE(of, atmel_spi_dt_ids);
1772 #endif
1773 
1774 static struct platform_driver atmel_spi_driver = {
1775 	.driver		= {
1776 		.name	= "atmel_spi",
1777 		.pm	= ATMEL_SPI_PM_OPS,
1778 		.of_match_table	= of_match_ptr(atmel_spi_dt_ids),
1779 	},
1780 	.probe		= atmel_spi_probe,
1781 	.remove		= atmel_spi_remove,
1782 };
1783 module_platform_driver(atmel_spi_driver);
1784 
1785 MODULE_DESCRIPTION("Atmel AT32/AT91 SPI Controller driver");
1786 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1787 MODULE_LICENSE("GPL");
1788 MODULE_ALIAS("platform:atmel_spi");
1789