xref: /linux/drivers/spi/spi-atmel.c (revision 83bce9c2baa51e439480a713119a73d3c8b61083)
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/of_gpio.h>
28 #include <linux/pinctrl/consumer.h>
29 #include <linux/pm_runtime.h>
30 
31 /* SPI register offsets */
32 #define SPI_CR					0x0000
33 #define SPI_MR					0x0004
34 #define SPI_RDR					0x0008
35 #define SPI_TDR					0x000c
36 #define SPI_SR					0x0010
37 #define SPI_IER					0x0014
38 #define SPI_IDR					0x0018
39 #define SPI_IMR					0x001c
40 #define SPI_CSR0				0x0030
41 #define SPI_CSR1				0x0034
42 #define SPI_CSR2				0x0038
43 #define SPI_CSR3				0x003c
44 #define SPI_FMR					0x0040
45 #define SPI_FLR					0x0044
46 #define SPI_VERSION				0x00fc
47 #define SPI_RPR					0x0100
48 #define SPI_RCR					0x0104
49 #define SPI_TPR					0x0108
50 #define SPI_TCR					0x010c
51 #define SPI_RNPR				0x0110
52 #define SPI_RNCR				0x0114
53 #define SPI_TNPR				0x0118
54 #define SPI_TNCR				0x011c
55 #define SPI_PTCR				0x0120
56 #define SPI_PTSR				0x0124
57 
58 /* Bitfields in CR */
59 #define SPI_SPIEN_OFFSET			0
60 #define SPI_SPIEN_SIZE				1
61 #define SPI_SPIDIS_OFFSET			1
62 #define SPI_SPIDIS_SIZE				1
63 #define SPI_SWRST_OFFSET			7
64 #define SPI_SWRST_SIZE				1
65 #define SPI_LASTXFER_OFFSET			24
66 #define SPI_LASTXFER_SIZE			1
67 #define SPI_TXFCLR_OFFSET			16
68 #define SPI_TXFCLR_SIZE				1
69 #define SPI_RXFCLR_OFFSET			17
70 #define SPI_RXFCLR_SIZE				1
71 #define SPI_FIFOEN_OFFSET			30
72 #define SPI_FIFOEN_SIZE				1
73 #define SPI_FIFODIS_OFFSET			31
74 #define SPI_FIFODIS_SIZE			1
75 
76 /* Bitfields in MR */
77 #define SPI_MSTR_OFFSET				0
78 #define SPI_MSTR_SIZE				1
79 #define SPI_PS_OFFSET				1
80 #define SPI_PS_SIZE				1
81 #define SPI_PCSDEC_OFFSET			2
82 #define SPI_PCSDEC_SIZE				1
83 #define SPI_FDIV_OFFSET				3
84 #define SPI_FDIV_SIZE				1
85 #define SPI_MODFDIS_OFFSET			4
86 #define SPI_MODFDIS_SIZE			1
87 #define SPI_WDRBT_OFFSET			5
88 #define SPI_WDRBT_SIZE				1
89 #define SPI_LLB_OFFSET				7
90 #define SPI_LLB_SIZE				1
91 #define SPI_PCS_OFFSET				16
92 #define SPI_PCS_SIZE				4
93 #define SPI_DLYBCS_OFFSET			24
94 #define SPI_DLYBCS_SIZE				8
95 
96 /* Bitfields in RDR */
97 #define SPI_RD_OFFSET				0
98 #define SPI_RD_SIZE				16
99 
100 /* Bitfields in TDR */
101 #define SPI_TD_OFFSET				0
102 #define SPI_TD_SIZE				16
103 
104 /* Bitfields in SR */
105 #define SPI_RDRF_OFFSET				0
106 #define SPI_RDRF_SIZE				1
107 #define SPI_TDRE_OFFSET				1
108 #define SPI_TDRE_SIZE				1
109 #define SPI_MODF_OFFSET				2
110 #define SPI_MODF_SIZE				1
111 #define SPI_OVRES_OFFSET			3
112 #define SPI_OVRES_SIZE				1
113 #define SPI_ENDRX_OFFSET			4
114 #define SPI_ENDRX_SIZE				1
115 #define SPI_ENDTX_OFFSET			5
116 #define SPI_ENDTX_SIZE				1
117 #define SPI_RXBUFF_OFFSET			6
118 #define SPI_RXBUFF_SIZE				1
119 #define SPI_TXBUFE_OFFSET			7
120 #define SPI_TXBUFE_SIZE				1
121 #define SPI_NSSR_OFFSET				8
122 #define SPI_NSSR_SIZE				1
123 #define SPI_TXEMPTY_OFFSET			9
124 #define SPI_TXEMPTY_SIZE			1
125 #define SPI_SPIENS_OFFSET			16
126 #define SPI_SPIENS_SIZE				1
127 #define SPI_TXFEF_OFFSET			24
128 #define SPI_TXFEF_SIZE				1
129 #define SPI_TXFFF_OFFSET			25
130 #define SPI_TXFFF_SIZE				1
131 #define SPI_TXFTHF_OFFSET			26
132 #define SPI_TXFTHF_SIZE				1
133 #define SPI_RXFEF_OFFSET			27
134 #define SPI_RXFEF_SIZE				1
135 #define SPI_RXFFF_OFFSET			28
136 #define SPI_RXFFF_SIZE				1
137 #define SPI_RXFTHF_OFFSET			29
138 #define SPI_RXFTHF_SIZE				1
139 #define SPI_TXFPTEF_OFFSET			30
140 #define SPI_TXFPTEF_SIZE			1
141 #define SPI_RXFPTEF_OFFSET			31
142 #define SPI_RXFPTEF_SIZE			1
143 
144 /* Bitfields in CSR0 */
145 #define SPI_CPOL_OFFSET				0
146 #define SPI_CPOL_SIZE				1
147 #define SPI_NCPHA_OFFSET			1
148 #define SPI_NCPHA_SIZE				1
149 #define SPI_CSAAT_OFFSET			3
150 #define SPI_CSAAT_SIZE				1
151 #define SPI_BITS_OFFSET				4
152 #define SPI_BITS_SIZE				4
153 #define SPI_SCBR_OFFSET				8
154 #define SPI_SCBR_SIZE				8
155 #define SPI_DLYBS_OFFSET			16
156 #define SPI_DLYBS_SIZE				8
157 #define SPI_DLYBCT_OFFSET			24
158 #define SPI_DLYBCT_SIZE				8
159 
160 /* Bitfields in RCR */
161 #define SPI_RXCTR_OFFSET			0
162 #define SPI_RXCTR_SIZE				16
163 
164 /* Bitfields in TCR */
165 #define SPI_TXCTR_OFFSET			0
166 #define SPI_TXCTR_SIZE				16
167 
168 /* Bitfields in RNCR */
169 #define SPI_RXNCR_OFFSET			0
170 #define SPI_RXNCR_SIZE				16
171 
172 /* Bitfields in TNCR */
173 #define SPI_TXNCR_OFFSET			0
174 #define SPI_TXNCR_SIZE				16
175 
176 /* Bitfields in PTCR */
177 #define SPI_RXTEN_OFFSET			0
178 #define SPI_RXTEN_SIZE				1
179 #define SPI_RXTDIS_OFFSET			1
180 #define SPI_RXTDIS_SIZE				1
181 #define SPI_TXTEN_OFFSET			8
182 #define SPI_TXTEN_SIZE				1
183 #define SPI_TXTDIS_OFFSET			9
184 #define SPI_TXTDIS_SIZE				1
185 
186 /* Bitfields in FMR */
187 #define SPI_TXRDYM_OFFSET			0
188 #define SPI_TXRDYM_SIZE				2
189 #define SPI_RXRDYM_OFFSET			4
190 #define SPI_RXRDYM_SIZE				2
191 #define SPI_TXFTHRES_OFFSET			16
192 #define SPI_TXFTHRES_SIZE			6
193 #define SPI_RXFTHRES_OFFSET			24
194 #define SPI_RXFTHRES_SIZE			6
195 
196 /* Bitfields in FLR */
197 #define SPI_TXFL_OFFSET				0
198 #define SPI_TXFL_SIZE				6
199 #define SPI_RXFL_OFFSET				16
200 #define SPI_RXFL_SIZE				6
201 
202 /* Constants for BITS */
203 #define SPI_BITS_8_BPT				0
204 #define SPI_BITS_9_BPT				1
205 #define SPI_BITS_10_BPT				2
206 #define SPI_BITS_11_BPT				3
207 #define SPI_BITS_12_BPT				4
208 #define SPI_BITS_13_BPT				5
209 #define SPI_BITS_14_BPT				6
210 #define SPI_BITS_15_BPT				7
211 #define SPI_BITS_16_BPT				8
212 #define SPI_ONE_DATA				0
213 #define SPI_TWO_DATA				1
214 #define SPI_FOUR_DATA				2
215 
216 /* Bit manipulation macros */
217 #define SPI_BIT(name) \
218 	(1 << SPI_##name##_OFFSET)
219 #define SPI_BF(name, value) \
220 	(((value) & ((1 << SPI_##name##_SIZE) - 1)) << SPI_##name##_OFFSET)
221 #define SPI_BFEXT(name, value) \
222 	(((value) >> SPI_##name##_OFFSET) & ((1 << SPI_##name##_SIZE) - 1))
223 #define SPI_BFINS(name, value, old) \
224 	(((old) & ~(((1 << SPI_##name##_SIZE) - 1) << SPI_##name##_OFFSET)) \
225 	  | SPI_BF(name, value))
226 
227 /* Register access macros */
228 #ifdef CONFIG_AVR32
229 #define spi_readl(port, reg) \
230 	__raw_readl((port)->regs + SPI_##reg)
231 #define spi_writel(port, reg, value) \
232 	__raw_writel((value), (port)->regs + SPI_##reg)
233 
234 #define spi_readw(port, reg) \
235 	__raw_readw((port)->regs + SPI_##reg)
236 #define spi_writew(port, reg, value) \
237 	__raw_writew((value), (port)->regs + SPI_##reg)
238 
239 #define spi_readb(port, reg) \
240 	__raw_readb((port)->regs + SPI_##reg)
241 #define spi_writeb(port, reg, value) \
242 	__raw_writeb((value), (port)->regs + SPI_##reg)
243 #else
244 #define spi_readl(port, reg) \
245 	readl_relaxed((port)->regs + SPI_##reg)
246 #define spi_writel(port, reg, value) \
247 	writel_relaxed((value), (port)->regs + SPI_##reg)
248 
249 #define spi_readw(port, reg) \
250 	readw_relaxed((port)->regs + SPI_##reg)
251 #define spi_writew(port, reg, value) \
252 	writew_relaxed((value), (port)->regs + SPI_##reg)
253 
254 #define spi_readb(port, reg) \
255 	readb_relaxed((port)->regs + SPI_##reg)
256 #define spi_writeb(port, reg, value) \
257 	writeb_relaxed((value), (port)->regs + SPI_##reg)
258 #endif
259 /* use PIO for small transfers, avoiding DMA setup/teardown overhead and
260  * cache operations; better heuristics consider wordsize and bitrate.
261  */
262 #define DMA_MIN_BYTES	16
263 
264 #define SPI_DMA_TIMEOUT		(msecs_to_jiffies(1000))
265 
266 #define AUTOSUSPEND_TIMEOUT	2000
267 
268 struct atmel_spi_caps {
269 	bool	is_spi2;
270 	bool	has_wdrbt;
271 	bool	has_dma_support;
272 };
273 
274 /*
275  * The core SPI transfer engine just talks to a register bank to set up
276  * DMA transfers; transfer queue progress is driven by IRQs.  The clock
277  * framework provides the base clock, subdivided for each spi_device.
278  */
279 struct atmel_spi {
280 	spinlock_t		lock;
281 	unsigned long		flags;
282 
283 	phys_addr_t		phybase;
284 	void __iomem		*regs;
285 	int			irq;
286 	struct clk		*clk;
287 	struct platform_device	*pdev;
288 	unsigned long		spi_clk;
289 
290 	struct spi_transfer	*current_transfer;
291 	int			current_remaining_bytes;
292 	int			done_status;
293 
294 	struct completion	xfer_completion;
295 
296 	struct atmel_spi_caps	caps;
297 
298 	bool			use_dma;
299 	bool			use_pdc;
300 	bool			use_cs_gpios;
301 
302 	bool			keep_cs;
303 	bool			cs_active;
304 
305 	u32			fifo_size;
306 };
307 
308 /* Controller-specific per-slave state */
309 struct atmel_spi_device {
310 	unsigned int		npcs_pin;
311 	u32			csr;
312 };
313 
314 #define SPI_MAX_DMA_XFER	65535 /* true for both PDC and DMA */
315 #define INVALID_DMA_ADDRESS	0xffffffff
316 
317 /*
318  * Version 2 of the SPI controller has
319  *  - CR.LASTXFER
320  *  - SPI_MR.DIV32 may become FDIV or must-be-zero (here: always zero)
321  *  - SPI_SR.TXEMPTY, SPI_SR.NSSR (and corresponding irqs)
322  *  - SPI_CSRx.CSAAT
323  *  - SPI_CSRx.SBCR allows faster clocking
324  */
325 static bool atmel_spi_is_v2(struct atmel_spi *as)
326 {
327 	return as->caps.is_spi2;
328 }
329 
330 /*
331  * Earlier SPI controllers (e.g. on at91rm9200) have a design bug whereby
332  * they assume that spi slave device state will not change on deselect, so
333  * that automagic deselection is OK.  ("NPCSx rises if no data is to be
334  * transmitted")  Not so!  Workaround uses nCSx pins as GPIOs; or newer
335  * controllers have CSAAT and friends.
336  *
337  * Since the CSAAT functionality is a bit weird on newer controllers as
338  * well, we use GPIO to control nCSx pins on all controllers, updating
339  * MR.PCS to avoid confusing the controller.  Using GPIOs also lets us
340  * support active-high chipselects despite the controller's belief that
341  * only active-low devices/systems exists.
342  *
343  * However, at91rm9200 has a second erratum whereby nCS0 doesn't work
344  * right when driven with GPIO.  ("Mode Fault does not allow more than one
345  * Master on Chip Select 0.")  No workaround exists for that ... so for
346  * nCS0 on that chip, we (a) don't use the GPIO, (b) can't support CS_HIGH,
347  * and (c) will trigger that first erratum in some cases.
348  */
349 
350 static void cs_activate(struct atmel_spi *as, struct spi_device *spi)
351 {
352 	struct atmel_spi_device *asd = spi->controller_state;
353 	unsigned active = spi->mode & SPI_CS_HIGH;
354 	u32 mr;
355 
356 	if (atmel_spi_is_v2(as)) {
357 		spi_writel(as, CSR0 + 4 * spi->chip_select, asd->csr);
358 		/* For the low SPI version, there is a issue that PDC transfer
359 		 * on CS1,2,3 needs SPI_CSR0.BITS config as SPI_CSR1,2,3.BITS
360 		 */
361 		spi_writel(as, CSR0, asd->csr);
362 		if (as->caps.has_wdrbt) {
363 			spi_writel(as, MR,
364 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
365 					| SPI_BIT(WDRBT)
366 					| SPI_BIT(MODFDIS)
367 					| SPI_BIT(MSTR));
368 		} else {
369 			spi_writel(as, MR,
370 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
371 					| SPI_BIT(MODFDIS)
372 					| SPI_BIT(MSTR));
373 		}
374 
375 		mr = spi_readl(as, MR);
376 		if (as->use_cs_gpios)
377 			gpio_set_value(asd->npcs_pin, active);
378 	} else {
379 		u32 cpol = (spi->mode & SPI_CPOL) ? SPI_BIT(CPOL) : 0;
380 		int i;
381 		u32 csr;
382 
383 		/* Make sure clock polarity is correct */
384 		for (i = 0; i < spi->master->num_chipselect; i++) {
385 			csr = spi_readl(as, CSR0 + 4 * i);
386 			if ((csr ^ cpol) & SPI_BIT(CPOL))
387 				spi_writel(as, CSR0 + 4 * i,
388 						csr ^ SPI_BIT(CPOL));
389 		}
390 
391 		mr = spi_readl(as, MR);
392 		mr = SPI_BFINS(PCS, ~(1 << spi->chip_select), mr);
393 		if (as->use_cs_gpios && spi->chip_select != 0)
394 			gpio_set_value(asd->npcs_pin, active);
395 		spi_writel(as, MR, mr);
396 	}
397 
398 	dev_dbg(&spi->dev, "activate %u%s, mr %08x\n",
399 			asd->npcs_pin, active ? " (high)" : "",
400 			mr);
401 }
402 
403 static void cs_deactivate(struct atmel_spi *as, struct spi_device *spi)
404 {
405 	struct atmel_spi_device *asd = spi->controller_state;
406 	unsigned active = spi->mode & SPI_CS_HIGH;
407 	u32 mr;
408 
409 	/* only deactivate *this* device; sometimes transfers to
410 	 * another device may be active when this routine is called.
411 	 */
412 	mr = spi_readl(as, MR);
413 	if (~SPI_BFEXT(PCS, mr) & (1 << spi->chip_select)) {
414 		mr = SPI_BFINS(PCS, 0xf, mr);
415 		spi_writel(as, MR, mr);
416 	}
417 
418 	dev_dbg(&spi->dev, "DEactivate %u%s, mr %08x\n",
419 			asd->npcs_pin, active ? " (low)" : "",
420 			mr);
421 
422 	if (!as->use_cs_gpios)
423 		spi_writel(as, CR, SPI_BIT(LASTXFER));
424 	else if (atmel_spi_is_v2(as) || spi->chip_select != 0)
425 		gpio_set_value(asd->npcs_pin, !active);
426 }
427 
428 static void atmel_spi_lock(struct atmel_spi *as) __acquires(&as->lock)
429 {
430 	spin_lock_irqsave(&as->lock, as->flags);
431 }
432 
433 static void atmel_spi_unlock(struct atmel_spi *as) __releases(&as->lock)
434 {
435 	spin_unlock_irqrestore(&as->lock, as->flags);
436 }
437 
438 static inline bool atmel_spi_use_dma(struct atmel_spi *as,
439 				struct spi_transfer *xfer)
440 {
441 	return as->use_dma && xfer->len >= DMA_MIN_BYTES;
442 }
443 
444 static bool atmel_spi_can_dma(struct spi_master *master,
445 			      struct spi_device *spi,
446 			      struct spi_transfer *xfer)
447 {
448 	struct atmel_spi *as = spi_master_get_devdata(master);
449 
450 	return atmel_spi_use_dma(as, xfer);
451 }
452 
453 static int atmel_spi_dma_slave_config(struct atmel_spi *as,
454 				struct dma_slave_config *slave_config,
455 				u8 bits_per_word)
456 {
457 	struct spi_master *master = platform_get_drvdata(as->pdev);
458 	int err = 0;
459 
460 	if (bits_per_word > 8) {
461 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
462 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
463 	} else {
464 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
465 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
466 	}
467 
468 	slave_config->dst_addr = (dma_addr_t)as->phybase + SPI_TDR;
469 	slave_config->src_addr = (dma_addr_t)as->phybase + SPI_RDR;
470 	slave_config->src_maxburst = 1;
471 	slave_config->dst_maxburst = 1;
472 	slave_config->device_fc = false;
473 
474 	/*
475 	 * This driver uses fixed peripheral select mode (PS bit set to '0' in
476 	 * the Mode Register).
477 	 * So according to the datasheet, when FIFOs are available (and
478 	 * enabled), the Transmit FIFO operates in Multiple Data Mode.
479 	 * In this mode, up to 2 data, not 4, can be written into the Transmit
480 	 * Data Register in a single access.
481 	 * However, the first data has to be written into the lowest 16 bits and
482 	 * the second data into the highest 16 bits of the Transmit
483 	 * Data Register. For 8bit data (the most frequent case), it would
484 	 * require to rework tx_buf so each data would actualy fit 16 bits.
485 	 * So we'd rather write only one data at the time. Hence the transmit
486 	 * path works the same whether FIFOs are available (and enabled) or not.
487 	 */
488 	slave_config->direction = DMA_MEM_TO_DEV;
489 	if (dmaengine_slave_config(master->dma_tx, slave_config)) {
490 		dev_err(&as->pdev->dev,
491 			"failed to configure tx dma channel\n");
492 		err = -EINVAL;
493 	}
494 
495 	/*
496 	 * This driver configures the spi controller for master mode (MSTR bit
497 	 * set to '1' in the Mode Register).
498 	 * So according to the datasheet, when FIFOs are available (and
499 	 * enabled), the Receive FIFO operates in Single Data Mode.
500 	 * So the receive path works the same whether FIFOs are available (and
501 	 * enabled) or not.
502 	 */
503 	slave_config->direction = DMA_DEV_TO_MEM;
504 	if (dmaengine_slave_config(master->dma_rx, slave_config)) {
505 		dev_err(&as->pdev->dev,
506 			"failed to configure rx dma channel\n");
507 		err = -EINVAL;
508 	}
509 
510 	return err;
511 }
512 
513 static int atmel_spi_configure_dma(struct spi_master *master,
514 				   struct atmel_spi *as)
515 {
516 	struct dma_slave_config	slave_config;
517 	struct device *dev = &as->pdev->dev;
518 	int err;
519 
520 	dma_cap_mask_t mask;
521 	dma_cap_zero(mask);
522 	dma_cap_set(DMA_SLAVE, mask);
523 
524 	master->dma_tx = dma_request_slave_channel_reason(dev, "tx");
525 	if (IS_ERR(master->dma_tx)) {
526 		err = PTR_ERR(master->dma_tx);
527 		if (err == -EPROBE_DEFER) {
528 			dev_warn(dev, "no DMA channel available at the moment\n");
529 			goto error_clear;
530 		}
531 		dev_err(dev,
532 			"DMA TX channel not available, SPI unable to use DMA\n");
533 		err = -EBUSY;
534 		goto error_clear;
535 	}
536 
537 	/*
538 	 * No reason to check EPROBE_DEFER here since we have already requested
539 	 * tx channel. If it fails here, it's for another reason.
540 	 */
541 	master->dma_rx = dma_request_slave_channel(dev, "rx");
542 
543 	if (!master->dma_rx) {
544 		dev_err(dev,
545 			"DMA RX channel not available, SPI unable to use DMA\n");
546 		err = -EBUSY;
547 		goto error;
548 	}
549 
550 	err = atmel_spi_dma_slave_config(as, &slave_config, 8);
551 	if (err)
552 		goto error;
553 
554 	dev_info(&as->pdev->dev,
555 			"Using %s (tx) and %s (rx) for DMA transfers\n",
556 			dma_chan_name(master->dma_tx),
557 			dma_chan_name(master->dma_rx));
558 
559 	return 0;
560 error:
561 	if (master->dma_rx)
562 		dma_release_channel(master->dma_rx);
563 	if (!IS_ERR(master->dma_tx))
564 		dma_release_channel(master->dma_tx);
565 error_clear:
566 	master->dma_tx = master->dma_rx = NULL;
567 	return err;
568 }
569 
570 static void atmel_spi_stop_dma(struct spi_master *master)
571 {
572 	if (master->dma_rx)
573 		dmaengine_terminate_all(master->dma_rx);
574 	if (master->dma_tx)
575 		dmaengine_terminate_all(master->dma_tx);
576 }
577 
578 static void atmel_spi_release_dma(struct spi_master *master)
579 {
580 	if (master->dma_rx) {
581 		dma_release_channel(master->dma_rx);
582 		master->dma_rx = NULL;
583 	}
584 	if (master->dma_tx) {
585 		dma_release_channel(master->dma_tx);
586 		master->dma_tx = NULL;
587 	}
588 }
589 
590 /* This function is called by the DMA driver from tasklet context */
591 static void dma_callback(void *data)
592 {
593 	struct spi_master	*master = data;
594 	struct atmel_spi	*as = spi_master_get_devdata(master);
595 
596 	complete(&as->xfer_completion);
597 }
598 
599 /*
600  * Next transfer using PIO without FIFO.
601  */
602 static void atmel_spi_next_xfer_single(struct spi_master *master,
603 				       struct spi_transfer *xfer)
604 {
605 	struct atmel_spi	*as = spi_master_get_devdata(master);
606 	unsigned long xfer_pos = xfer->len - as->current_remaining_bytes;
607 
608 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_pio\n");
609 
610 	/* Make sure data is not remaining in RDR */
611 	spi_readl(as, RDR);
612 	while (spi_readl(as, SR) & SPI_BIT(RDRF)) {
613 		spi_readl(as, RDR);
614 		cpu_relax();
615 	}
616 
617 	if (xfer->bits_per_word > 8)
618 		spi_writel(as, TDR, *(u16 *)(xfer->tx_buf + xfer_pos));
619 	else
620 		spi_writel(as, TDR, *(u8 *)(xfer->tx_buf + xfer_pos));
621 
622 	dev_dbg(master->dev.parent,
623 		"  start pio xfer %p: len %u tx %p rx %p bitpw %d\n",
624 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
625 		xfer->bits_per_word);
626 
627 	/* Enable relevant interrupts */
628 	spi_writel(as, IER, SPI_BIT(RDRF) | SPI_BIT(OVRES));
629 }
630 
631 /*
632  * Next transfer using PIO with FIFO.
633  */
634 static void atmel_spi_next_xfer_fifo(struct spi_master *master,
635 				     struct spi_transfer *xfer)
636 {
637 	struct atmel_spi *as = spi_master_get_devdata(master);
638 	u32 current_remaining_data, num_data;
639 	u32 offset = xfer->len - as->current_remaining_bytes;
640 	const u16 *words = (const u16 *)((u8 *)xfer->tx_buf + offset);
641 	const u8  *bytes = (const u8  *)((u8 *)xfer->tx_buf + offset);
642 	u16 td0, td1;
643 	u32 fifomr;
644 
645 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_fifo\n");
646 
647 	/* Compute the number of data to transfer in the current iteration */
648 	current_remaining_data = ((xfer->bits_per_word > 8) ?
649 				  ((u32)as->current_remaining_bytes >> 1) :
650 				  (u32)as->current_remaining_bytes);
651 	num_data = min(current_remaining_data, as->fifo_size);
652 
653 	/* Flush RX and TX FIFOs */
654 	spi_writel(as, CR, SPI_BIT(RXFCLR) | SPI_BIT(TXFCLR));
655 	while (spi_readl(as, FLR))
656 		cpu_relax();
657 
658 	/* Set RX FIFO Threshold to the number of data to transfer */
659 	fifomr = spi_readl(as, FMR);
660 	spi_writel(as, FMR, SPI_BFINS(RXFTHRES, num_data, fifomr));
661 
662 	/* Clear FIFO flags in the Status Register, especially RXFTHF */
663 	(void)spi_readl(as, SR);
664 
665 	/* Fill TX FIFO */
666 	while (num_data >= 2) {
667 		if (xfer->bits_per_word > 8) {
668 			td0 = *words++;
669 			td1 = *words++;
670 		} else {
671 			td0 = *bytes++;
672 			td1 = *bytes++;
673 		}
674 
675 		spi_writel(as, TDR, (td1 << 16) | td0);
676 		num_data -= 2;
677 	}
678 
679 	if (num_data) {
680 		if (xfer->bits_per_word > 8)
681 			td0 = *words++;
682 		else
683 			td0 = *bytes++;
684 
685 		spi_writew(as, TDR, td0);
686 		num_data--;
687 	}
688 
689 	dev_dbg(master->dev.parent,
690 		"  start fifo xfer %p: len %u tx %p rx %p bitpw %d\n",
691 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
692 		xfer->bits_per_word);
693 
694 	/*
695 	 * Enable RX FIFO Threshold Flag interrupt to be notified about
696 	 * transfer completion.
697 	 */
698 	spi_writel(as, IER, SPI_BIT(RXFTHF) | SPI_BIT(OVRES));
699 }
700 
701 /*
702  * Next transfer using PIO.
703  */
704 static void atmel_spi_next_xfer_pio(struct spi_master *master,
705 				    struct spi_transfer *xfer)
706 {
707 	struct atmel_spi *as = spi_master_get_devdata(master);
708 
709 	if (as->fifo_size)
710 		atmel_spi_next_xfer_fifo(master, xfer);
711 	else
712 		atmel_spi_next_xfer_single(master, xfer);
713 }
714 
715 /*
716  * Submit next transfer for DMA.
717  */
718 static int atmel_spi_next_xfer_dma_submit(struct spi_master *master,
719 				struct spi_transfer *xfer,
720 				u32 *plen)
721 {
722 	struct atmel_spi	*as = spi_master_get_devdata(master);
723 	struct dma_chan		*rxchan = master->dma_rx;
724 	struct dma_chan		*txchan = master->dma_tx;
725 	struct dma_async_tx_descriptor *rxdesc;
726 	struct dma_async_tx_descriptor *txdesc;
727 	struct dma_slave_config	slave_config;
728 	dma_cookie_t		cookie;
729 
730 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_dma_submit\n");
731 
732 	/* Check that the channels are available */
733 	if (!rxchan || !txchan)
734 		return -ENODEV;
735 
736 	/* release lock for DMA operations */
737 	atmel_spi_unlock(as);
738 
739 	*plen = xfer->len;
740 
741 	if (atmel_spi_dma_slave_config(as, &slave_config,
742 				       xfer->bits_per_word))
743 		goto err_exit;
744 
745 	/* Send both scatterlists */
746 	rxdesc = dmaengine_prep_slave_sg(rxchan,
747 					 xfer->rx_sg.sgl, xfer->rx_sg.nents,
748 					 DMA_FROM_DEVICE,
749 					 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
750 	if (!rxdesc)
751 		goto err_dma;
752 
753 	txdesc = dmaengine_prep_slave_sg(txchan,
754 					 xfer->tx_sg.sgl, xfer->tx_sg.nents,
755 					 DMA_TO_DEVICE,
756 					 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
757 	if (!txdesc)
758 		goto err_dma;
759 
760 	dev_dbg(master->dev.parent,
761 		"  start dma xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
762 		xfer, xfer->len, xfer->tx_buf, (unsigned long long)xfer->tx_dma,
763 		xfer->rx_buf, (unsigned long long)xfer->rx_dma);
764 
765 	/* Enable relevant interrupts */
766 	spi_writel(as, IER, SPI_BIT(OVRES));
767 
768 	/* Put the callback on the RX transfer only, that should finish last */
769 	rxdesc->callback = dma_callback;
770 	rxdesc->callback_param = master;
771 
772 	/* Submit and fire RX and TX with TX last so we're ready to read! */
773 	cookie = rxdesc->tx_submit(rxdesc);
774 	if (dma_submit_error(cookie))
775 		goto err_dma;
776 	cookie = txdesc->tx_submit(txdesc);
777 	if (dma_submit_error(cookie))
778 		goto err_dma;
779 	rxchan->device->device_issue_pending(rxchan);
780 	txchan->device->device_issue_pending(txchan);
781 
782 	/* take back lock */
783 	atmel_spi_lock(as);
784 	return 0;
785 
786 err_dma:
787 	spi_writel(as, IDR, SPI_BIT(OVRES));
788 	atmel_spi_stop_dma(master);
789 err_exit:
790 	atmel_spi_lock(as);
791 	return -ENOMEM;
792 }
793 
794 static void atmel_spi_next_xfer_data(struct spi_master *master,
795 				struct spi_transfer *xfer,
796 				dma_addr_t *tx_dma,
797 				dma_addr_t *rx_dma,
798 				u32 *plen)
799 {
800 	*rx_dma = xfer->rx_dma + xfer->len - *plen;
801 	*tx_dma = xfer->tx_dma + xfer->len - *plen;
802 	if (*plen > master->max_dma_len)
803 		*plen = master->max_dma_len;
804 }
805 
806 static int atmel_spi_set_xfer_speed(struct atmel_spi *as,
807 				    struct spi_device *spi,
808 				    struct spi_transfer *xfer)
809 {
810 	u32			scbr, csr;
811 	unsigned long		bus_hz;
812 
813 	/* v1 chips start out at half the peripheral bus speed. */
814 	bus_hz = as->spi_clk;
815 	if (!atmel_spi_is_v2(as))
816 		bus_hz /= 2;
817 
818 	/*
819 	 * Calculate the lowest divider that satisfies the
820 	 * constraint, assuming div32/fdiv/mbz == 0.
821 	 */
822 	scbr = DIV_ROUND_UP(bus_hz, xfer->speed_hz);
823 
824 	/*
825 	 * If the resulting divider doesn't fit into the
826 	 * register bitfield, we can't satisfy the constraint.
827 	 */
828 	if (scbr >= (1 << SPI_SCBR_SIZE)) {
829 		dev_err(&spi->dev,
830 			"setup: %d Hz too slow, scbr %u; min %ld Hz\n",
831 			xfer->speed_hz, scbr, bus_hz/255);
832 		return -EINVAL;
833 	}
834 	if (scbr == 0) {
835 		dev_err(&spi->dev,
836 			"setup: %d Hz too high, scbr %u; max %ld Hz\n",
837 			xfer->speed_hz, scbr, bus_hz);
838 		return -EINVAL;
839 	}
840 	csr = spi_readl(as, CSR0 + 4 * spi->chip_select);
841 	csr = SPI_BFINS(SCBR, scbr, csr);
842 	spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
843 
844 	return 0;
845 }
846 
847 /*
848  * Submit next transfer for PDC.
849  * lock is held, spi irq is blocked
850  */
851 static void atmel_spi_pdc_next_xfer(struct spi_master *master,
852 					struct spi_message *msg,
853 					struct spi_transfer *xfer)
854 {
855 	struct atmel_spi	*as = spi_master_get_devdata(master);
856 	u32			len;
857 	dma_addr_t		tx_dma, rx_dma;
858 
859 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
860 
861 	len = as->current_remaining_bytes;
862 	atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
863 	as->current_remaining_bytes -= len;
864 
865 	spi_writel(as, RPR, rx_dma);
866 	spi_writel(as, TPR, tx_dma);
867 
868 	if (msg->spi->bits_per_word > 8)
869 		len >>= 1;
870 	spi_writel(as, RCR, len);
871 	spi_writel(as, TCR, len);
872 
873 	dev_dbg(&msg->spi->dev,
874 		"  start xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
875 		xfer, xfer->len, xfer->tx_buf,
876 		(unsigned long long)xfer->tx_dma, xfer->rx_buf,
877 		(unsigned long long)xfer->rx_dma);
878 
879 	if (as->current_remaining_bytes) {
880 		len = as->current_remaining_bytes;
881 		atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
882 		as->current_remaining_bytes -= len;
883 
884 		spi_writel(as, RNPR, rx_dma);
885 		spi_writel(as, TNPR, tx_dma);
886 
887 		if (msg->spi->bits_per_word > 8)
888 			len >>= 1;
889 		spi_writel(as, RNCR, len);
890 		spi_writel(as, TNCR, len);
891 
892 		dev_dbg(&msg->spi->dev,
893 			"  next xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
894 			xfer, xfer->len, xfer->tx_buf,
895 			(unsigned long long)xfer->tx_dma, xfer->rx_buf,
896 			(unsigned long long)xfer->rx_dma);
897 	}
898 
899 	/* REVISIT: We're waiting for RXBUFF before we start the next
900 	 * transfer because we need to handle some difficult timing
901 	 * issues otherwise. If we wait for TXBUFE in one transfer and
902 	 * then starts waiting for RXBUFF in the next, it's difficult
903 	 * to tell the difference between the RXBUFF interrupt we're
904 	 * actually waiting for and the RXBUFF interrupt of the
905 	 * previous transfer.
906 	 *
907 	 * It should be doable, though. Just not now...
908 	 */
909 	spi_writel(as, IER, SPI_BIT(RXBUFF) | SPI_BIT(OVRES));
910 	spi_writel(as, PTCR, SPI_BIT(TXTEN) | SPI_BIT(RXTEN));
911 }
912 
913 /*
914  * For DMA, tx_buf/tx_dma have the same relationship as rx_buf/rx_dma:
915  *  - The buffer is either valid for CPU access, else NULL
916  *  - If the buffer is valid, so is its DMA address
917  *
918  * This driver manages the dma address unless message->is_dma_mapped.
919  */
920 static int
921 atmel_spi_dma_map_xfer(struct atmel_spi *as, struct spi_transfer *xfer)
922 {
923 	struct device	*dev = &as->pdev->dev;
924 
925 	xfer->tx_dma = xfer->rx_dma = INVALID_DMA_ADDRESS;
926 	if (xfer->tx_buf) {
927 		/* tx_buf is a const void* where we need a void * for the dma
928 		 * mapping */
929 		void *nonconst_tx = (void *)xfer->tx_buf;
930 
931 		xfer->tx_dma = dma_map_single(dev,
932 				nonconst_tx, xfer->len,
933 				DMA_TO_DEVICE);
934 		if (dma_mapping_error(dev, xfer->tx_dma))
935 			return -ENOMEM;
936 	}
937 	if (xfer->rx_buf) {
938 		xfer->rx_dma = dma_map_single(dev,
939 				xfer->rx_buf, xfer->len,
940 				DMA_FROM_DEVICE);
941 		if (dma_mapping_error(dev, xfer->rx_dma)) {
942 			if (xfer->tx_buf)
943 				dma_unmap_single(dev,
944 						xfer->tx_dma, xfer->len,
945 						DMA_TO_DEVICE);
946 			return -ENOMEM;
947 		}
948 	}
949 	return 0;
950 }
951 
952 static void atmel_spi_dma_unmap_xfer(struct spi_master *master,
953 				     struct spi_transfer *xfer)
954 {
955 	if (xfer->tx_dma != INVALID_DMA_ADDRESS)
956 		dma_unmap_single(master->dev.parent, xfer->tx_dma,
957 				 xfer->len, DMA_TO_DEVICE);
958 	if (xfer->rx_dma != INVALID_DMA_ADDRESS)
959 		dma_unmap_single(master->dev.parent, xfer->rx_dma,
960 				 xfer->len, DMA_FROM_DEVICE);
961 }
962 
963 static void atmel_spi_disable_pdc_transfer(struct atmel_spi *as)
964 {
965 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
966 }
967 
968 static void
969 atmel_spi_pump_single_data(struct atmel_spi *as, struct spi_transfer *xfer)
970 {
971 	u8		*rxp;
972 	u16		*rxp16;
973 	unsigned long	xfer_pos = xfer->len - as->current_remaining_bytes;
974 
975 	if (xfer->bits_per_word > 8) {
976 		rxp16 = (u16 *)(((u8 *)xfer->rx_buf) + xfer_pos);
977 		*rxp16 = spi_readl(as, RDR);
978 	} else {
979 		rxp = ((u8 *)xfer->rx_buf) + xfer_pos;
980 		*rxp = spi_readl(as, RDR);
981 	}
982 	if (xfer->bits_per_word > 8) {
983 		if (as->current_remaining_bytes > 2)
984 			as->current_remaining_bytes -= 2;
985 		else
986 			as->current_remaining_bytes = 0;
987 	} else {
988 		as->current_remaining_bytes--;
989 	}
990 }
991 
992 static void
993 atmel_spi_pump_fifo_data(struct atmel_spi *as, struct spi_transfer *xfer)
994 {
995 	u32 fifolr = spi_readl(as, FLR);
996 	u32 num_bytes, num_data = SPI_BFEXT(RXFL, fifolr);
997 	u32 offset = xfer->len - as->current_remaining_bytes;
998 	u16 *words = (u16 *)((u8 *)xfer->rx_buf + offset);
999 	u8  *bytes = (u8  *)((u8 *)xfer->rx_buf + offset);
1000 	u16 rd; /* RD field is the lowest 16 bits of RDR */
1001 
1002 	/* Update the number of remaining bytes to transfer */
1003 	num_bytes = ((xfer->bits_per_word > 8) ?
1004 		     (num_data << 1) :
1005 		     num_data);
1006 
1007 	if (as->current_remaining_bytes > num_bytes)
1008 		as->current_remaining_bytes -= num_bytes;
1009 	else
1010 		as->current_remaining_bytes = 0;
1011 
1012 	/* Handle odd number of bytes when data are more than 8bit width */
1013 	if (xfer->bits_per_word > 8)
1014 		as->current_remaining_bytes &= ~0x1;
1015 
1016 	/* Read data */
1017 	while (num_data) {
1018 		rd = spi_readl(as, RDR);
1019 		if (xfer->bits_per_word > 8)
1020 			*words++ = rd;
1021 		else
1022 			*bytes++ = rd;
1023 		num_data--;
1024 	}
1025 }
1026 
1027 /* Called from IRQ
1028  *
1029  * Must update "current_remaining_bytes" to keep track of data
1030  * to transfer.
1031  */
1032 static void
1033 atmel_spi_pump_pio_data(struct atmel_spi *as, struct spi_transfer *xfer)
1034 {
1035 	if (as->fifo_size)
1036 		atmel_spi_pump_fifo_data(as, xfer);
1037 	else
1038 		atmel_spi_pump_single_data(as, xfer);
1039 }
1040 
1041 /* Interrupt
1042  *
1043  * No need for locking in this Interrupt handler: done_status is the
1044  * only information modified.
1045  */
1046 static irqreturn_t
1047 atmel_spi_pio_interrupt(int irq, void *dev_id)
1048 {
1049 	struct spi_master	*master = dev_id;
1050 	struct atmel_spi	*as = spi_master_get_devdata(master);
1051 	u32			status, pending, imr;
1052 	struct spi_transfer	*xfer;
1053 	int			ret = IRQ_NONE;
1054 
1055 	imr = spi_readl(as, IMR);
1056 	status = spi_readl(as, SR);
1057 	pending = status & imr;
1058 
1059 	if (pending & SPI_BIT(OVRES)) {
1060 		ret = IRQ_HANDLED;
1061 		spi_writel(as, IDR, SPI_BIT(OVRES));
1062 		dev_warn(master->dev.parent, "overrun\n");
1063 
1064 		/*
1065 		 * When we get an overrun, we disregard the current
1066 		 * transfer. Data will not be copied back from any
1067 		 * bounce buffer and msg->actual_len will not be
1068 		 * updated with the last xfer.
1069 		 *
1070 		 * We will also not process any remaning transfers in
1071 		 * the message.
1072 		 */
1073 		as->done_status = -EIO;
1074 		smp_wmb();
1075 
1076 		/* Clear any overrun happening while cleaning up */
1077 		spi_readl(as, SR);
1078 
1079 		complete(&as->xfer_completion);
1080 
1081 	} else if (pending & (SPI_BIT(RDRF) | SPI_BIT(RXFTHF))) {
1082 		atmel_spi_lock(as);
1083 
1084 		if (as->current_remaining_bytes) {
1085 			ret = IRQ_HANDLED;
1086 			xfer = as->current_transfer;
1087 			atmel_spi_pump_pio_data(as, xfer);
1088 			if (!as->current_remaining_bytes)
1089 				spi_writel(as, IDR, pending);
1090 
1091 			complete(&as->xfer_completion);
1092 		}
1093 
1094 		atmel_spi_unlock(as);
1095 	} else {
1096 		WARN_ONCE(pending, "IRQ not handled, pending = %x\n", pending);
1097 		ret = IRQ_HANDLED;
1098 		spi_writel(as, IDR, pending);
1099 	}
1100 
1101 	return ret;
1102 }
1103 
1104 static irqreturn_t
1105 atmel_spi_pdc_interrupt(int irq, void *dev_id)
1106 {
1107 	struct spi_master	*master = dev_id;
1108 	struct atmel_spi	*as = spi_master_get_devdata(master);
1109 	u32			status, pending, imr;
1110 	int			ret = IRQ_NONE;
1111 
1112 	imr = spi_readl(as, IMR);
1113 	status = spi_readl(as, SR);
1114 	pending = status & imr;
1115 
1116 	if (pending & SPI_BIT(OVRES)) {
1117 
1118 		ret = IRQ_HANDLED;
1119 
1120 		spi_writel(as, IDR, (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX)
1121 				     | SPI_BIT(OVRES)));
1122 
1123 		/* Clear any overrun happening while cleaning up */
1124 		spi_readl(as, SR);
1125 
1126 		as->done_status = -EIO;
1127 
1128 		complete(&as->xfer_completion);
1129 
1130 	} else if (pending & (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX))) {
1131 		ret = IRQ_HANDLED;
1132 
1133 		spi_writel(as, IDR, pending);
1134 
1135 		complete(&as->xfer_completion);
1136 	}
1137 
1138 	return ret;
1139 }
1140 
1141 static int atmel_spi_setup(struct spi_device *spi)
1142 {
1143 	struct atmel_spi	*as;
1144 	struct atmel_spi_device	*asd;
1145 	u32			csr;
1146 	unsigned int		bits = spi->bits_per_word;
1147 	unsigned int		npcs_pin;
1148 
1149 	as = spi_master_get_devdata(spi->master);
1150 
1151 	/* see notes above re chipselect */
1152 	if (!atmel_spi_is_v2(as)
1153 			&& spi->chip_select == 0
1154 			&& (spi->mode & SPI_CS_HIGH)) {
1155 		dev_dbg(&spi->dev, "setup: can't be active-high\n");
1156 		return -EINVAL;
1157 	}
1158 
1159 	csr = SPI_BF(BITS, bits - 8);
1160 	if (spi->mode & SPI_CPOL)
1161 		csr |= SPI_BIT(CPOL);
1162 	if (!(spi->mode & SPI_CPHA))
1163 		csr |= SPI_BIT(NCPHA);
1164 	if (!as->use_cs_gpios)
1165 		csr |= SPI_BIT(CSAAT);
1166 
1167 	/* DLYBS is mostly irrelevant since we manage chipselect using GPIOs.
1168 	 *
1169 	 * DLYBCT would add delays between words, slowing down transfers.
1170 	 * It could potentially be useful to cope with DMA bottlenecks, but
1171 	 * in those cases it's probably best to just use a lower bitrate.
1172 	 */
1173 	csr |= SPI_BF(DLYBS, 0);
1174 	csr |= SPI_BF(DLYBCT, 0);
1175 
1176 	/* chipselect must have been muxed as GPIO (e.g. in board setup) */
1177 	npcs_pin = (unsigned long)spi->controller_data;
1178 
1179 	if (!as->use_cs_gpios)
1180 		npcs_pin = spi->chip_select;
1181 	else if (gpio_is_valid(spi->cs_gpio))
1182 		npcs_pin = spi->cs_gpio;
1183 
1184 	asd = spi->controller_state;
1185 	if (!asd) {
1186 		asd = kzalloc(sizeof(struct atmel_spi_device), GFP_KERNEL);
1187 		if (!asd)
1188 			return -ENOMEM;
1189 
1190 		if (as->use_cs_gpios)
1191 			gpio_direction_output(npcs_pin,
1192 					      !(spi->mode & SPI_CS_HIGH));
1193 
1194 		asd->npcs_pin = npcs_pin;
1195 		spi->controller_state = asd;
1196 	}
1197 
1198 	asd->csr = csr;
1199 
1200 	dev_dbg(&spi->dev,
1201 		"setup: bpw %u mode 0x%x -> csr%d %08x\n",
1202 		bits, spi->mode, spi->chip_select, csr);
1203 
1204 	if (!atmel_spi_is_v2(as))
1205 		spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
1206 
1207 	return 0;
1208 }
1209 
1210 static int atmel_spi_one_transfer(struct spi_master *master,
1211 					struct spi_message *msg,
1212 					struct spi_transfer *xfer)
1213 {
1214 	struct atmel_spi	*as;
1215 	struct spi_device	*spi = msg->spi;
1216 	u8			bits;
1217 	u32			len;
1218 	struct atmel_spi_device	*asd;
1219 	int			timeout;
1220 	int			ret;
1221 	unsigned long		dma_timeout;
1222 
1223 	as = spi_master_get_devdata(master);
1224 
1225 	if (!(xfer->tx_buf || xfer->rx_buf) && xfer->len) {
1226 		dev_dbg(&spi->dev, "missing rx or tx buf\n");
1227 		return -EINVAL;
1228 	}
1229 
1230 	asd = spi->controller_state;
1231 	bits = (asd->csr >> 4) & 0xf;
1232 	if (bits != xfer->bits_per_word - 8) {
1233 		dev_dbg(&spi->dev,
1234 			"you can't yet change bits_per_word in transfers\n");
1235 		return -ENOPROTOOPT;
1236 	}
1237 
1238 	/*
1239 	 * DMA map early, for performance (empties dcache ASAP) and
1240 	 * better fault reporting.
1241 	 */
1242 	if ((!msg->is_dma_mapped)
1243 		&& as->use_pdc) {
1244 		if (atmel_spi_dma_map_xfer(as, xfer) < 0)
1245 			return -ENOMEM;
1246 	}
1247 
1248 	atmel_spi_set_xfer_speed(as, msg->spi, xfer);
1249 
1250 	as->done_status = 0;
1251 	as->current_transfer = xfer;
1252 	as->current_remaining_bytes = xfer->len;
1253 	while (as->current_remaining_bytes) {
1254 		reinit_completion(&as->xfer_completion);
1255 
1256 		if (as->use_pdc) {
1257 			atmel_spi_pdc_next_xfer(master, msg, xfer);
1258 		} else if (atmel_spi_use_dma(as, xfer)) {
1259 			len = as->current_remaining_bytes;
1260 			ret = atmel_spi_next_xfer_dma_submit(master,
1261 								xfer, &len);
1262 			if (ret) {
1263 				dev_err(&spi->dev,
1264 					"unable to use DMA, fallback to PIO\n");
1265 				atmel_spi_next_xfer_pio(master, xfer);
1266 			} else {
1267 				as->current_remaining_bytes -= len;
1268 				if (as->current_remaining_bytes < 0)
1269 					as->current_remaining_bytes = 0;
1270 			}
1271 		} else {
1272 			atmel_spi_next_xfer_pio(master, xfer);
1273 		}
1274 
1275 		/* interrupts are disabled, so free the lock for schedule */
1276 		atmel_spi_unlock(as);
1277 		dma_timeout = wait_for_completion_timeout(&as->xfer_completion,
1278 							  SPI_DMA_TIMEOUT);
1279 		atmel_spi_lock(as);
1280 		if (WARN_ON(dma_timeout == 0)) {
1281 			dev_err(&spi->dev, "spi transfer timeout\n");
1282 			as->done_status = -EIO;
1283 		}
1284 
1285 		if (as->done_status)
1286 			break;
1287 	}
1288 
1289 	if (as->done_status) {
1290 		if (as->use_pdc) {
1291 			dev_warn(master->dev.parent,
1292 				"overrun (%u/%u remaining)\n",
1293 				spi_readl(as, TCR), spi_readl(as, RCR));
1294 
1295 			/*
1296 			 * Clean up DMA registers and make sure the data
1297 			 * registers are empty.
1298 			 */
1299 			spi_writel(as, RNCR, 0);
1300 			spi_writel(as, TNCR, 0);
1301 			spi_writel(as, RCR, 0);
1302 			spi_writel(as, TCR, 0);
1303 			for (timeout = 1000; timeout; timeout--)
1304 				if (spi_readl(as, SR) & SPI_BIT(TXEMPTY))
1305 					break;
1306 			if (!timeout)
1307 				dev_warn(master->dev.parent,
1308 					 "timeout waiting for TXEMPTY");
1309 			while (spi_readl(as, SR) & SPI_BIT(RDRF))
1310 				spi_readl(as, RDR);
1311 
1312 			/* Clear any overrun happening while cleaning up */
1313 			spi_readl(as, SR);
1314 
1315 		} else if (atmel_spi_use_dma(as, xfer)) {
1316 			atmel_spi_stop_dma(master);
1317 		}
1318 
1319 		if (!msg->is_dma_mapped
1320 			&& as->use_pdc)
1321 			atmel_spi_dma_unmap_xfer(master, xfer);
1322 
1323 		return 0;
1324 
1325 	} else {
1326 		/* only update length if no error */
1327 		msg->actual_length += xfer->len;
1328 	}
1329 
1330 	if (!msg->is_dma_mapped
1331 		&& as->use_pdc)
1332 		atmel_spi_dma_unmap_xfer(master, xfer);
1333 
1334 	if (xfer->delay_usecs)
1335 		udelay(xfer->delay_usecs);
1336 
1337 	if (xfer->cs_change) {
1338 		if (list_is_last(&xfer->transfer_list,
1339 				 &msg->transfers)) {
1340 			as->keep_cs = true;
1341 		} else {
1342 			as->cs_active = !as->cs_active;
1343 			if (as->cs_active)
1344 				cs_activate(as, msg->spi);
1345 			else
1346 				cs_deactivate(as, msg->spi);
1347 		}
1348 	}
1349 
1350 	return 0;
1351 }
1352 
1353 static int atmel_spi_transfer_one_message(struct spi_master *master,
1354 						struct spi_message *msg)
1355 {
1356 	struct atmel_spi *as;
1357 	struct spi_transfer *xfer;
1358 	struct spi_device *spi = msg->spi;
1359 	int ret = 0;
1360 
1361 	as = spi_master_get_devdata(master);
1362 
1363 	dev_dbg(&spi->dev, "new message %p submitted for %s\n",
1364 					msg, dev_name(&spi->dev));
1365 
1366 	atmel_spi_lock(as);
1367 	cs_activate(as, spi);
1368 
1369 	as->cs_active = true;
1370 	as->keep_cs = false;
1371 
1372 	msg->status = 0;
1373 	msg->actual_length = 0;
1374 
1375 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1376 		ret = atmel_spi_one_transfer(master, msg, xfer);
1377 		if (ret)
1378 			goto msg_done;
1379 	}
1380 
1381 	if (as->use_pdc)
1382 		atmel_spi_disable_pdc_transfer(as);
1383 
1384 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1385 		dev_dbg(&spi->dev,
1386 			"  xfer %p: len %u tx %p/%pad rx %p/%pad\n",
1387 			xfer, xfer->len,
1388 			xfer->tx_buf, &xfer->tx_dma,
1389 			xfer->rx_buf, &xfer->rx_dma);
1390 	}
1391 
1392 msg_done:
1393 	if (!as->keep_cs)
1394 		cs_deactivate(as, msg->spi);
1395 
1396 	atmel_spi_unlock(as);
1397 
1398 	msg->status = as->done_status;
1399 	spi_finalize_current_message(spi->master);
1400 
1401 	return ret;
1402 }
1403 
1404 static void atmel_spi_cleanup(struct spi_device *spi)
1405 {
1406 	struct atmel_spi_device	*asd = spi->controller_state;
1407 
1408 	if (!asd)
1409 		return;
1410 
1411 	spi->controller_state = NULL;
1412 	kfree(asd);
1413 }
1414 
1415 static inline unsigned int atmel_get_version(struct atmel_spi *as)
1416 {
1417 	return spi_readl(as, VERSION) & 0x00000fff;
1418 }
1419 
1420 static void atmel_get_caps(struct atmel_spi *as)
1421 {
1422 	unsigned int version;
1423 
1424 	version = atmel_get_version(as);
1425 	dev_info(&as->pdev->dev, "version: 0x%x\n", version);
1426 
1427 	as->caps.is_spi2 = version > 0x121;
1428 	as->caps.has_wdrbt = version >= 0x210;
1429 	as->caps.has_dma_support = version >= 0x212;
1430 }
1431 
1432 /*-------------------------------------------------------------------------*/
1433 static int atmel_spi_gpio_cs(struct platform_device *pdev)
1434 {
1435 	struct spi_master	*master = platform_get_drvdata(pdev);
1436 	struct atmel_spi	*as = spi_master_get_devdata(master);
1437 	struct device_node	*np = master->dev.of_node;
1438 	int			i;
1439 	int			ret = 0;
1440 	int			nb = 0;
1441 
1442 	if (!as->use_cs_gpios)
1443 		return 0;
1444 
1445 	if (!np)
1446 		return 0;
1447 
1448 	nb = of_gpio_named_count(np, "cs-gpios");
1449 	for (i = 0; i < nb; i++) {
1450 		int cs_gpio = of_get_named_gpio(pdev->dev.of_node,
1451 						"cs-gpios", i);
1452 
1453 		if (cs_gpio == -EPROBE_DEFER)
1454 			return cs_gpio;
1455 
1456 		if (gpio_is_valid(cs_gpio)) {
1457 			ret = devm_gpio_request(&pdev->dev, cs_gpio,
1458 						dev_name(&pdev->dev));
1459 			if (ret)
1460 				return ret;
1461 		}
1462 	}
1463 
1464 	return 0;
1465 }
1466 
1467 static int atmel_spi_probe(struct platform_device *pdev)
1468 {
1469 	struct resource		*regs;
1470 	int			irq;
1471 	struct clk		*clk;
1472 	int			ret;
1473 	struct spi_master	*master;
1474 	struct atmel_spi	*as;
1475 
1476 	/* Select default pin state */
1477 	pinctrl_pm_select_default_state(&pdev->dev);
1478 
1479 	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1480 	if (!regs)
1481 		return -ENXIO;
1482 
1483 	irq = platform_get_irq(pdev, 0);
1484 	if (irq < 0)
1485 		return irq;
1486 
1487 	clk = devm_clk_get(&pdev->dev, "spi_clk");
1488 	if (IS_ERR(clk))
1489 		return PTR_ERR(clk);
1490 
1491 	/* setup spi core then atmel-specific driver state */
1492 	ret = -ENOMEM;
1493 	master = spi_alloc_master(&pdev->dev, sizeof(*as));
1494 	if (!master)
1495 		goto out_free;
1496 
1497 	/* the spi->mode bits understood by this driver: */
1498 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
1499 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(8, 16);
1500 	master->dev.of_node = pdev->dev.of_node;
1501 	master->bus_num = pdev->id;
1502 	master->num_chipselect = master->dev.of_node ? 0 : 4;
1503 	master->setup = atmel_spi_setup;
1504 	master->flags = (SPI_MASTER_MUST_RX | SPI_MASTER_MUST_TX);
1505 	master->transfer_one_message = atmel_spi_transfer_one_message;
1506 	master->cleanup = atmel_spi_cleanup;
1507 	master->auto_runtime_pm = true;
1508 	master->max_dma_len = SPI_MAX_DMA_XFER;
1509 	master->can_dma = atmel_spi_can_dma;
1510 	platform_set_drvdata(pdev, master);
1511 
1512 	as = spi_master_get_devdata(master);
1513 
1514 	spin_lock_init(&as->lock);
1515 
1516 	as->pdev = pdev;
1517 	as->regs = devm_ioremap_resource(&pdev->dev, regs);
1518 	if (IS_ERR(as->regs)) {
1519 		ret = PTR_ERR(as->regs);
1520 		goto out_unmap_regs;
1521 	}
1522 	as->phybase = regs->start;
1523 	as->irq = irq;
1524 	as->clk = clk;
1525 
1526 	init_completion(&as->xfer_completion);
1527 
1528 	atmel_get_caps(as);
1529 
1530 	as->use_cs_gpios = true;
1531 	if (atmel_spi_is_v2(as) &&
1532 	    pdev->dev.of_node &&
1533 	    !of_get_property(pdev->dev.of_node, "cs-gpios", NULL)) {
1534 		as->use_cs_gpios = false;
1535 		master->num_chipselect = 4;
1536 	}
1537 
1538 	ret = atmel_spi_gpio_cs(pdev);
1539 	if (ret)
1540 		goto out_unmap_regs;
1541 
1542 	as->use_dma = false;
1543 	as->use_pdc = false;
1544 	if (as->caps.has_dma_support) {
1545 		ret = atmel_spi_configure_dma(master, as);
1546 		if (ret == 0) {
1547 			as->use_dma = true;
1548 		} else if (ret == -EPROBE_DEFER) {
1549 			return ret;
1550 		}
1551 	} else {
1552 		as->use_pdc = true;
1553 	}
1554 
1555 	if (as->caps.has_dma_support && !as->use_dma)
1556 		dev_info(&pdev->dev, "Atmel SPI Controller using PIO only\n");
1557 
1558 	if (as->use_pdc) {
1559 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pdc_interrupt,
1560 					0, dev_name(&pdev->dev), master);
1561 	} else {
1562 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pio_interrupt,
1563 					0, dev_name(&pdev->dev), master);
1564 	}
1565 	if (ret)
1566 		goto out_unmap_regs;
1567 
1568 	/* Initialize the hardware */
1569 	ret = clk_prepare_enable(clk);
1570 	if (ret)
1571 		goto out_free_irq;
1572 
1573 	as->spi_clk = clk_get_rate(clk);
1574 
1575 	spi_writel(as, CR, SPI_BIT(SWRST));
1576 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1577 	if (as->caps.has_wdrbt) {
1578 		spi_writel(as, MR, SPI_BIT(WDRBT) | SPI_BIT(MODFDIS)
1579 				| SPI_BIT(MSTR));
1580 	} else {
1581 		spi_writel(as, MR, SPI_BIT(MSTR) | SPI_BIT(MODFDIS));
1582 	}
1583 
1584 	if (as->use_pdc)
1585 		spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1586 	spi_writel(as, CR, SPI_BIT(SPIEN));
1587 
1588 	as->fifo_size = 0;
1589 	if (!of_property_read_u32(pdev->dev.of_node, "atmel,fifo-size",
1590 				  &as->fifo_size)) {
1591 		dev_info(&pdev->dev, "Using FIFO (%u data)\n", as->fifo_size);
1592 		spi_writel(as, CR, SPI_BIT(FIFOEN));
1593 	}
1594 
1595 	pm_runtime_set_autosuspend_delay(&pdev->dev, AUTOSUSPEND_TIMEOUT);
1596 	pm_runtime_use_autosuspend(&pdev->dev);
1597 	pm_runtime_set_active(&pdev->dev);
1598 	pm_runtime_enable(&pdev->dev);
1599 
1600 	ret = devm_spi_register_master(&pdev->dev, master);
1601 	if (ret)
1602 		goto out_free_dma;
1603 
1604 	/* go! */
1605 	dev_info(&pdev->dev, "Atmel SPI Controller at 0x%08lx (irq %d)\n",
1606 			(unsigned long)regs->start, irq);
1607 
1608 	return 0;
1609 
1610 out_free_dma:
1611 	pm_runtime_disable(&pdev->dev);
1612 	pm_runtime_set_suspended(&pdev->dev);
1613 
1614 	if (as->use_dma)
1615 		atmel_spi_release_dma(master);
1616 
1617 	spi_writel(as, CR, SPI_BIT(SWRST));
1618 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1619 	clk_disable_unprepare(clk);
1620 out_free_irq:
1621 out_unmap_regs:
1622 out_free:
1623 	spi_master_put(master);
1624 	return ret;
1625 }
1626 
1627 static int atmel_spi_remove(struct platform_device *pdev)
1628 {
1629 	struct spi_master	*master = platform_get_drvdata(pdev);
1630 	struct atmel_spi	*as = spi_master_get_devdata(master);
1631 
1632 	pm_runtime_get_sync(&pdev->dev);
1633 
1634 	/* reset the hardware and block queue progress */
1635 	spin_lock_irq(&as->lock);
1636 	if (as->use_dma) {
1637 		atmel_spi_stop_dma(master);
1638 		atmel_spi_release_dma(master);
1639 	}
1640 
1641 	spi_writel(as, CR, SPI_BIT(SWRST));
1642 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1643 	spi_readl(as, SR);
1644 	spin_unlock_irq(&as->lock);
1645 
1646 	clk_disable_unprepare(as->clk);
1647 
1648 	pm_runtime_put_noidle(&pdev->dev);
1649 	pm_runtime_disable(&pdev->dev);
1650 
1651 	return 0;
1652 }
1653 
1654 #ifdef CONFIG_PM
1655 static int atmel_spi_runtime_suspend(struct device *dev)
1656 {
1657 	struct spi_master *master = dev_get_drvdata(dev);
1658 	struct atmel_spi *as = spi_master_get_devdata(master);
1659 
1660 	clk_disable_unprepare(as->clk);
1661 	pinctrl_pm_select_sleep_state(dev);
1662 
1663 	return 0;
1664 }
1665 
1666 static int atmel_spi_runtime_resume(struct device *dev)
1667 {
1668 	struct spi_master *master = dev_get_drvdata(dev);
1669 	struct atmel_spi *as = spi_master_get_devdata(master);
1670 
1671 	pinctrl_pm_select_default_state(dev);
1672 
1673 	return clk_prepare_enable(as->clk);
1674 }
1675 
1676 #ifdef CONFIG_PM_SLEEP
1677 static int atmel_spi_suspend(struct device *dev)
1678 {
1679 	struct spi_master *master = dev_get_drvdata(dev);
1680 	int ret;
1681 
1682 	/* Stop the queue running */
1683 	ret = spi_master_suspend(master);
1684 	if (ret) {
1685 		dev_warn(dev, "cannot suspend master\n");
1686 		return ret;
1687 	}
1688 
1689 	if (!pm_runtime_suspended(dev))
1690 		atmel_spi_runtime_suspend(dev);
1691 
1692 	return 0;
1693 }
1694 
1695 static int atmel_spi_resume(struct device *dev)
1696 {
1697 	struct spi_master *master = dev_get_drvdata(dev);
1698 	int ret;
1699 
1700 	if (!pm_runtime_suspended(dev)) {
1701 		ret = atmel_spi_runtime_resume(dev);
1702 		if (ret)
1703 			return ret;
1704 	}
1705 
1706 	/* Start the queue running */
1707 	ret = spi_master_resume(master);
1708 	if (ret)
1709 		dev_err(dev, "problem starting queue (%d)\n", ret);
1710 
1711 	return ret;
1712 }
1713 #endif
1714 
1715 static const struct dev_pm_ops atmel_spi_pm_ops = {
1716 	SET_SYSTEM_SLEEP_PM_OPS(atmel_spi_suspend, atmel_spi_resume)
1717 	SET_RUNTIME_PM_OPS(atmel_spi_runtime_suspend,
1718 			   atmel_spi_runtime_resume, NULL)
1719 };
1720 #define ATMEL_SPI_PM_OPS	(&atmel_spi_pm_ops)
1721 #else
1722 #define ATMEL_SPI_PM_OPS	NULL
1723 #endif
1724 
1725 #if defined(CONFIG_OF)
1726 static const struct of_device_id atmel_spi_dt_ids[] = {
1727 	{ .compatible = "atmel,at91rm9200-spi" },
1728 	{ /* sentinel */ }
1729 };
1730 
1731 MODULE_DEVICE_TABLE(of, atmel_spi_dt_ids);
1732 #endif
1733 
1734 static struct platform_driver atmel_spi_driver = {
1735 	.driver		= {
1736 		.name	= "atmel_spi",
1737 		.pm	= ATMEL_SPI_PM_OPS,
1738 		.of_match_table	= of_match_ptr(atmel_spi_dt_ids),
1739 	},
1740 	.probe		= atmel_spi_probe,
1741 	.remove		= atmel_spi_remove,
1742 };
1743 module_platform_driver(atmel_spi_driver);
1744 
1745 MODULE_DESCRIPTION("Atmel AT32/AT91 SPI Controller driver");
1746 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1747 MODULE_LICENSE("GPL");
1748 MODULE_ALIAS("platform:atmel_spi");
1749