xref: /linux/drivers/spi/spi-atmel.c (revision c0e297dc61f8d4453e07afbea1fa8d0e67cd4a34)
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/atmel.h>
23 #include <linux/platform_data/dma-atmel.h>
24 #include <linux/of.h>
25 
26 #include <linux/io.h>
27 #include <linux/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_dma {
269 	struct dma_chan			*chan_rx;
270 	struct dma_chan			*chan_tx;
271 	struct scatterlist		sgrx;
272 	struct scatterlist		sgtx;
273 	struct dma_async_tx_descriptor	*data_desc_rx;
274 	struct dma_async_tx_descriptor	*data_desc_tx;
275 
276 	struct at_dma_slave	dma_slave;
277 };
278 
279 struct atmel_spi_caps {
280 	bool	is_spi2;
281 	bool	has_wdrbt;
282 	bool	has_dma_support;
283 };
284 
285 /*
286  * The core SPI transfer engine just talks to a register bank to set up
287  * DMA transfers; transfer queue progress is driven by IRQs.  The clock
288  * framework provides the base clock, subdivided for each spi_device.
289  */
290 struct atmel_spi {
291 	spinlock_t		lock;
292 	unsigned long		flags;
293 
294 	phys_addr_t		phybase;
295 	void __iomem		*regs;
296 	int			irq;
297 	struct clk		*clk;
298 	struct platform_device	*pdev;
299 
300 	struct spi_transfer	*current_transfer;
301 	int			current_remaining_bytes;
302 	int			done_status;
303 
304 	struct completion	xfer_completion;
305 
306 	/* scratch buffer */
307 	void			*buffer;
308 	dma_addr_t		buffer_dma;
309 
310 	struct atmel_spi_caps	caps;
311 
312 	bool			use_dma;
313 	bool			use_pdc;
314 	bool			use_cs_gpios;
315 	/* dmaengine data */
316 	struct atmel_spi_dma	dma;
317 
318 	bool			keep_cs;
319 	bool			cs_active;
320 
321 	u32			fifo_size;
322 };
323 
324 /* Controller-specific per-slave state */
325 struct atmel_spi_device {
326 	unsigned int		npcs_pin;
327 	u32			csr;
328 };
329 
330 #define BUFFER_SIZE		PAGE_SIZE
331 #define INVALID_DMA_ADDRESS	0xffffffff
332 
333 /*
334  * Version 2 of the SPI controller has
335  *  - CR.LASTXFER
336  *  - SPI_MR.DIV32 may become FDIV or must-be-zero (here: always zero)
337  *  - SPI_SR.TXEMPTY, SPI_SR.NSSR (and corresponding irqs)
338  *  - SPI_CSRx.CSAAT
339  *  - SPI_CSRx.SBCR allows faster clocking
340  */
341 static bool atmel_spi_is_v2(struct atmel_spi *as)
342 {
343 	return as->caps.is_spi2;
344 }
345 
346 /*
347  * Earlier SPI controllers (e.g. on at91rm9200) have a design bug whereby
348  * they assume that spi slave device state will not change on deselect, so
349  * that automagic deselection is OK.  ("NPCSx rises if no data is to be
350  * transmitted")  Not so!  Workaround uses nCSx pins as GPIOs; or newer
351  * controllers have CSAAT and friends.
352  *
353  * Since the CSAAT functionality is a bit weird on newer controllers as
354  * well, we use GPIO to control nCSx pins on all controllers, updating
355  * MR.PCS to avoid confusing the controller.  Using GPIOs also lets us
356  * support active-high chipselects despite the controller's belief that
357  * only active-low devices/systems exists.
358  *
359  * However, at91rm9200 has a second erratum whereby nCS0 doesn't work
360  * right when driven with GPIO.  ("Mode Fault does not allow more than one
361  * Master on Chip Select 0.")  No workaround exists for that ... so for
362  * nCS0 on that chip, we (a) don't use the GPIO, (b) can't support CS_HIGH,
363  * and (c) will trigger that first erratum in some cases.
364  */
365 
366 static void cs_activate(struct atmel_spi *as, struct spi_device *spi)
367 {
368 	struct atmel_spi_device *asd = spi->controller_state;
369 	unsigned active = spi->mode & SPI_CS_HIGH;
370 	u32 mr;
371 
372 	if (atmel_spi_is_v2(as)) {
373 		spi_writel(as, CSR0 + 4 * spi->chip_select, asd->csr);
374 		/* For the low SPI version, there is a issue that PDC transfer
375 		 * on CS1,2,3 needs SPI_CSR0.BITS config as SPI_CSR1,2,3.BITS
376 		 */
377 		spi_writel(as, CSR0, asd->csr);
378 		if (as->caps.has_wdrbt) {
379 			spi_writel(as, MR,
380 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
381 					| SPI_BIT(WDRBT)
382 					| SPI_BIT(MODFDIS)
383 					| SPI_BIT(MSTR));
384 		} else {
385 			spi_writel(as, MR,
386 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
387 					| SPI_BIT(MODFDIS)
388 					| SPI_BIT(MSTR));
389 		}
390 
391 		mr = spi_readl(as, MR);
392 		if (as->use_cs_gpios)
393 			gpio_set_value(asd->npcs_pin, active);
394 	} else {
395 		u32 cpol = (spi->mode & SPI_CPOL) ? SPI_BIT(CPOL) : 0;
396 		int i;
397 		u32 csr;
398 
399 		/* Make sure clock polarity is correct */
400 		for (i = 0; i < spi->master->num_chipselect; i++) {
401 			csr = spi_readl(as, CSR0 + 4 * i);
402 			if ((csr ^ cpol) & SPI_BIT(CPOL))
403 				spi_writel(as, CSR0 + 4 * i,
404 						csr ^ SPI_BIT(CPOL));
405 		}
406 
407 		mr = spi_readl(as, MR);
408 		mr = SPI_BFINS(PCS, ~(1 << spi->chip_select), mr);
409 		if (as->use_cs_gpios && spi->chip_select != 0)
410 			gpio_set_value(asd->npcs_pin, active);
411 		spi_writel(as, MR, mr);
412 	}
413 
414 	dev_dbg(&spi->dev, "activate %u%s, mr %08x\n",
415 			asd->npcs_pin, active ? " (high)" : "",
416 			mr);
417 }
418 
419 static void cs_deactivate(struct atmel_spi *as, struct spi_device *spi)
420 {
421 	struct atmel_spi_device *asd = spi->controller_state;
422 	unsigned active = spi->mode & SPI_CS_HIGH;
423 	u32 mr;
424 
425 	/* only deactivate *this* device; sometimes transfers to
426 	 * another device may be active when this routine is called.
427 	 */
428 	mr = spi_readl(as, MR);
429 	if (~SPI_BFEXT(PCS, mr) & (1 << spi->chip_select)) {
430 		mr = SPI_BFINS(PCS, 0xf, mr);
431 		spi_writel(as, MR, mr);
432 	}
433 
434 	dev_dbg(&spi->dev, "DEactivate %u%s, mr %08x\n",
435 			asd->npcs_pin, active ? " (low)" : "",
436 			mr);
437 
438 	if (!as->use_cs_gpios)
439 		spi_writel(as, CR, SPI_BIT(LASTXFER));
440 	else if (atmel_spi_is_v2(as) || spi->chip_select != 0)
441 		gpio_set_value(asd->npcs_pin, !active);
442 }
443 
444 static void atmel_spi_lock(struct atmel_spi *as) __acquires(&as->lock)
445 {
446 	spin_lock_irqsave(&as->lock, as->flags);
447 }
448 
449 static void atmel_spi_unlock(struct atmel_spi *as) __releases(&as->lock)
450 {
451 	spin_unlock_irqrestore(&as->lock, as->flags);
452 }
453 
454 static inline bool atmel_spi_use_dma(struct atmel_spi *as,
455 				struct spi_transfer *xfer)
456 {
457 	return as->use_dma && xfer->len >= DMA_MIN_BYTES;
458 }
459 
460 static int atmel_spi_dma_slave_config(struct atmel_spi *as,
461 				struct dma_slave_config *slave_config,
462 				u8 bits_per_word)
463 {
464 	int err = 0;
465 
466 	if (bits_per_word > 8) {
467 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
468 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
469 	} else {
470 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
471 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
472 	}
473 
474 	slave_config->dst_addr = (dma_addr_t)as->phybase + SPI_TDR;
475 	slave_config->src_addr = (dma_addr_t)as->phybase + SPI_RDR;
476 	slave_config->src_maxburst = 1;
477 	slave_config->dst_maxburst = 1;
478 	slave_config->device_fc = false;
479 
480 	/*
481 	 * This driver uses fixed peripheral select mode (PS bit set to '0' in
482 	 * the Mode Register).
483 	 * So according to the datasheet, when FIFOs are available (and
484 	 * enabled), the Transmit FIFO operates in Multiple Data Mode.
485 	 * In this mode, up to 2 data, not 4, can be written into the Transmit
486 	 * Data Register in a single access.
487 	 * However, the first data has to be written into the lowest 16 bits and
488 	 * the second data into the highest 16 bits of the Transmit
489 	 * Data Register. For 8bit data (the most frequent case), it would
490 	 * require to rework tx_buf so each data would actualy fit 16 bits.
491 	 * So we'd rather write only one data at the time. Hence the transmit
492 	 * path works the same whether FIFOs are available (and enabled) or not.
493 	 */
494 	slave_config->direction = DMA_MEM_TO_DEV;
495 	if (dmaengine_slave_config(as->dma.chan_tx, slave_config)) {
496 		dev_err(&as->pdev->dev,
497 			"failed to configure tx dma channel\n");
498 		err = -EINVAL;
499 	}
500 
501 	/*
502 	 * This driver configures the spi controller for master mode (MSTR bit
503 	 * set to '1' in the Mode Register).
504 	 * So according to the datasheet, when FIFOs are available (and
505 	 * enabled), the Receive FIFO operates in Single Data Mode.
506 	 * So the receive path works the same whether FIFOs are available (and
507 	 * enabled) or not.
508 	 */
509 	slave_config->direction = DMA_DEV_TO_MEM;
510 	if (dmaengine_slave_config(as->dma.chan_rx, slave_config)) {
511 		dev_err(&as->pdev->dev,
512 			"failed to configure rx dma channel\n");
513 		err = -EINVAL;
514 	}
515 
516 	return err;
517 }
518 
519 static int atmel_spi_configure_dma(struct atmel_spi *as)
520 {
521 	struct dma_slave_config	slave_config;
522 	struct device *dev = &as->pdev->dev;
523 	int err;
524 
525 	dma_cap_mask_t mask;
526 	dma_cap_zero(mask);
527 	dma_cap_set(DMA_SLAVE, mask);
528 
529 	as->dma.chan_tx = dma_request_slave_channel_reason(dev, "tx");
530 	if (IS_ERR(as->dma.chan_tx)) {
531 		err = PTR_ERR(as->dma.chan_tx);
532 		if (err == -EPROBE_DEFER) {
533 			dev_warn(dev, "no DMA channel available at the moment\n");
534 			return err;
535 		}
536 		dev_err(dev,
537 			"DMA TX channel not available, SPI unable to use DMA\n");
538 		err = -EBUSY;
539 		goto error;
540 	}
541 
542 	/*
543 	 * No reason to check EPROBE_DEFER here since we have already requested
544 	 * tx channel. If it fails here, it's for another reason.
545 	 */
546 	as->dma.chan_rx = dma_request_slave_channel(dev, "rx");
547 
548 	if (!as->dma.chan_rx) {
549 		dev_err(dev,
550 			"DMA RX channel not available, SPI unable to use DMA\n");
551 		err = -EBUSY;
552 		goto error;
553 	}
554 
555 	err = atmel_spi_dma_slave_config(as, &slave_config, 8);
556 	if (err)
557 		goto error;
558 
559 	dev_info(&as->pdev->dev,
560 			"Using %s (tx) and %s (rx) for DMA transfers\n",
561 			dma_chan_name(as->dma.chan_tx),
562 			dma_chan_name(as->dma.chan_rx));
563 	return 0;
564 error:
565 	if (as->dma.chan_rx)
566 		dma_release_channel(as->dma.chan_rx);
567 	if (!IS_ERR(as->dma.chan_tx))
568 		dma_release_channel(as->dma.chan_tx);
569 	return err;
570 }
571 
572 static void atmel_spi_stop_dma(struct atmel_spi *as)
573 {
574 	if (as->dma.chan_rx)
575 		dmaengine_terminate_all(as->dma.chan_rx);
576 	if (as->dma.chan_tx)
577 		dmaengine_terminate_all(as->dma.chan_tx);
578 }
579 
580 static void atmel_spi_release_dma(struct atmel_spi *as)
581 {
582 	if (as->dma.chan_rx)
583 		dma_release_channel(as->dma.chan_rx);
584 	if (as->dma.chan_tx)
585 		dma_release_channel(as->dma.chan_tx);
586 }
587 
588 /* This function is called by the DMA driver from tasklet context */
589 static void dma_callback(void *data)
590 {
591 	struct spi_master	*master = data;
592 	struct atmel_spi	*as = spi_master_get_devdata(master);
593 
594 	complete(&as->xfer_completion);
595 }
596 
597 /*
598  * Next transfer using PIO without FIFO.
599  */
600 static void atmel_spi_next_xfer_single(struct spi_master *master,
601 				       struct spi_transfer *xfer)
602 {
603 	struct atmel_spi	*as = spi_master_get_devdata(master);
604 	unsigned long xfer_pos = xfer->len - as->current_remaining_bytes;
605 
606 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_pio\n");
607 
608 	/* Make sure data is not remaining in RDR */
609 	spi_readl(as, RDR);
610 	while (spi_readl(as, SR) & SPI_BIT(RDRF)) {
611 		spi_readl(as, RDR);
612 		cpu_relax();
613 	}
614 
615 	if (xfer->tx_buf) {
616 		if (xfer->bits_per_word > 8)
617 			spi_writel(as, TDR, *(u16 *)(xfer->tx_buf + xfer_pos));
618 		else
619 			spi_writel(as, TDR, *(u8 *)(xfer->tx_buf + xfer_pos));
620 	} else {
621 		spi_writel(as, TDR, 0);
622 	}
623 
624 	dev_dbg(master->dev.parent,
625 		"  start pio xfer %p: len %u tx %p rx %p bitpw %d\n",
626 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
627 		xfer->bits_per_word);
628 
629 	/* Enable relevant interrupts */
630 	spi_writel(as, IER, SPI_BIT(RDRF) | SPI_BIT(OVRES));
631 }
632 
633 /*
634  * Next transfer using PIO with FIFO.
635  */
636 static void atmel_spi_next_xfer_fifo(struct spi_master *master,
637 				     struct spi_transfer *xfer)
638 {
639 	struct atmel_spi *as = spi_master_get_devdata(master);
640 	u32 current_remaining_data, num_data;
641 	u32 offset = xfer->len - as->current_remaining_bytes;
642 	const u16 *words = (const u16 *)((u8 *)xfer->tx_buf + offset);
643 	const u8  *bytes = (const u8  *)((u8 *)xfer->tx_buf + offset);
644 	u16 td0, td1;
645 	u32 fifomr;
646 
647 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_fifo\n");
648 
649 	/* Compute the number of data to transfer in the current iteration */
650 	current_remaining_data = ((xfer->bits_per_word > 8) ?
651 				  ((u32)as->current_remaining_bytes >> 1) :
652 				  (u32)as->current_remaining_bytes);
653 	num_data = min(current_remaining_data, as->fifo_size);
654 
655 	/* Flush RX and TX FIFOs */
656 	spi_writel(as, CR, SPI_BIT(RXFCLR) | SPI_BIT(TXFCLR));
657 	while (spi_readl(as, FLR))
658 		cpu_relax();
659 
660 	/* Set RX FIFO Threshold to the number of data to transfer */
661 	fifomr = spi_readl(as, FMR);
662 	spi_writel(as, FMR, SPI_BFINS(RXFTHRES, num_data, fifomr));
663 
664 	/* Clear FIFO flags in the Status Register, especially RXFTHF */
665 	(void)spi_readl(as, SR);
666 
667 	/* Fill TX FIFO */
668 	while (num_data >= 2) {
669 		if (xfer->tx_buf) {
670 			if (xfer->bits_per_word > 8) {
671 				td0 = *words++;
672 				td1 = *words++;
673 			} else {
674 				td0 = *bytes++;
675 				td1 = *bytes++;
676 			}
677 		} else {
678 			td0 = 0;
679 			td1 = 0;
680 		}
681 
682 		spi_writel(as, TDR, (td1 << 16) | td0);
683 		num_data -= 2;
684 	}
685 
686 	if (num_data) {
687 		if (xfer->tx_buf) {
688 			if (xfer->bits_per_word > 8)
689 				td0 = *words++;
690 			else
691 				td0 = *bytes++;
692 		} else {
693 			td0 = 0;
694 		}
695 
696 		spi_writew(as, TDR, td0);
697 		num_data--;
698 	}
699 
700 	dev_dbg(master->dev.parent,
701 		"  start fifo xfer %p: len %u tx %p rx %p bitpw %d\n",
702 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
703 		xfer->bits_per_word);
704 
705 	/*
706 	 * Enable RX FIFO Threshold Flag interrupt to be notified about
707 	 * transfer completion.
708 	 */
709 	spi_writel(as, IER, SPI_BIT(RXFTHF) | SPI_BIT(OVRES));
710 }
711 
712 /*
713  * Next transfer using PIO.
714  */
715 static void atmel_spi_next_xfer_pio(struct spi_master *master,
716 				    struct spi_transfer *xfer)
717 {
718 	struct atmel_spi *as = spi_master_get_devdata(master);
719 
720 	if (as->fifo_size)
721 		atmel_spi_next_xfer_fifo(master, xfer);
722 	else
723 		atmel_spi_next_xfer_single(master, xfer);
724 }
725 
726 /*
727  * Submit next transfer for DMA.
728  */
729 static int atmel_spi_next_xfer_dma_submit(struct spi_master *master,
730 				struct spi_transfer *xfer,
731 				u32 *plen)
732 {
733 	struct atmel_spi	*as = spi_master_get_devdata(master);
734 	struct dma_chan		*rxchan = as->dma.chan_rx;
735 	struct dma_chan		*txchan = as->dma.chan_tx;
736 	struct dma_async_tx_descriptor *rxdesc;
737 	struct dma_async_tx_descriptor *txdesc;
738 	struct dma_slave_config	slave_config;
739 	dma_cookie_t		cookie;
740 	u32	len = *plen;
741 
742 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_dma_submit\n");
743 
744 	/* Check that the channels are available */
745 	if (!rxchan || !txchan)
746 		return -ENODEV;
747 
748 	/* release lock for DMA operations */
749 	atmel_spi_unlock(as);
750 
751 	/* prepare the RX dma transfer */
752 	sg_init_table(&as->dma.sgrx, 1);
753 	if (xfer->rx_buf) {
754 		as->dma.sgrx.dma_address = xfer->rx_dma + xfer->len - *plen;
755 	} else {
756 		as->dma.sgrx.dma_address = as->buffer_dma;
757 		if (len > BUFFER_SIZE)
758 			len = BUFFER_SIZE;
759 	}
760 
761 	/* prepare the TX dma transfer */
762 	sg_init_table(&as->dma.sgtx, 1);
763 	if (xfer->tx_buf) {
764 		as->dma.sgtx.dma_address = xfer->tx_dma + xfer->len - *plen;
765 	} else {
766 		as->dma.sgtx.dma_address = as->buffer_dma;
767 		if (len > BUFFER_SIZE)
768 			len = BUFFER_SIZE;
769 		memset(as->buffer, 0, len);
770 	}
771 
772 	sg_dma_len(&as->dma.sgtx) = len;
773 	sg_dma_len(&as->dma.sgrx) = len;
774 
775 	*plen = len;
776 
777 	if (atmel_spi_dma_slave_config(as, &slave_config, 8))
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 	if (xfer->speed_hz)
876 		scbr = DIV_ROUND_UP(bus_hz, xfer->speed_hz);
877 	else
878 		/*
879 		 * This can happend if max_speed is null.
880 		 * In this case, we set the lowest possible speed
881 		 */
882 		scbr = 0xff;
883 
884 	/*
885 	 * If the resulting divider doesn't fit into the
886 	 * register bitfield, we can't satisfy the constraint.
887 	 */
888 	if (scbr >= (1 << SPI_SCBR_SIZE)) {
889 		dev_err(&spi->dev,
890 			"setup: %d Hz too slow, scbr %u; min %ld Hz\n",
891 			xfer->speed_hz, scbr, bus_hz/255);
892 		return -EINVAL;
893 	}
894 	if (scbr == 0) {
895 		dev_err(&spi->dev,
896 			"setup: %d Hz too high, scbr %u; max %ld Hz\n",
897 			xfer->speed_hz, scbr, bus_hz);
898 		return -EINVAL;
899 	}
900 	csr = spi_readl(as, CSR0 + 4 * spi->chip_select);
901 	csr = SPI_BFINS(SCBR, scbr, csr);
902 	spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
903 
904 	return 0;
905 }
906 
907 /*
908  * Submit next transfer for PDC.
909  * lock is held, spi irq is blocked
910  */
911 static void atmel_spi_pdc_next_xfer(struct spi_master *master,
912 					struct spi_message *msg,
913 					struct spi_transfer *xfer)
914 {
915 	struct atmel_spi	*as = spi_master_get_devdata(master);
916 	u32			len;
917 	dma_addr_t		tx_dma, rx_dma;
918 
919 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
920 
921 	len = as->current_remaining_bytes;
922 	atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
923 	as->current_remaining_bytes -= len;
924 
925 	spi_writel(as, RPR, rx_dma);
926 	spi_writel(as, TPR, tx_dma);
927 
928 	if (msg->spi->bits_per_word > 8)
929 		len >>= 1;
930 	spi_writel(as, RCR, len);
931 	spi_writel(as, TCR, len);
932 
933 	dev_dbg(&msg->spi->dev,
934 		"  start xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
935 		xfer, xfer->len, xfer->tx_buf,
936 		(unsigned long long)xfer->tx_dma, xfer->rx_buf,
937 		(unsigned long long)xfer->rx_dma);
938 
939 	if (as->current_remaining_bytes) {
940 		len = as->current_remaining_bytes;
941 		atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
942 		as->current_remaining_bytes -= len;
943 
944 		spi_writel(as, RNPR, rx_dma);
945 		spi_writel(as, TNPR, tx_dma);
946 
947 		if (msg->spi->bits_per_word > 8)
948 			len >>= 1;
949 		spi_writel(as, RNCR, len);
950 		spi_writel(as, TNCR, len);
951 
952 		dev_dbg(&msg->spi->dev,
953 			"  next xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
954 			xfer, xfer->len, xfer->tx_buf,
955 			(unsigned long long)xfer->tx_dma, xfer->rx_buf,
956 			(unsigned long long)xfer->rx_dma);
957 	}
958 
959 	/* REVISIT: We're waiting for RXBUFF before we start the next
960 	 * transfer because we need to handle some difficult timing
961 	 * issues otherwise. If we wait for TXBUFE in one transfer and
962 	 * then starts waiting for RXBUFF in the next, it's difficult
963 	 * to tell the difference between the RXBUFF interrupt we're
964 	 * actually waiting for and the RXBUFF interrupt of the
965 	 * previous transfer.
966 	 *
967 	 * It should be doable, though. Just not now...
968 	 */
969 	spi_writel(as, IER, SPI_BIT(RXBUFF) | SPI_BIT(OVRES));
970 	spi_writel(as, PTCR, SPI_BIT(TXTEN) | SPI_BIT(RXTEN));
971 }
972 
973 /*
974  * For DMA, tx_buf/tx_dma have the same relationship as rx_buf/rx_dma:
975  *  - The buffer is either valid for CPU access, else NULL
976  *  - If the buffer is valid, so is its DMA address
977  *
978  * This driver manages the dma address unless message->is_dma_mapped.
979  */
980 static int
981 atmel_spi_dma_map_xfer(struct atmel_spi *as, struct spi_transfer *xfer)
982 {
983 	struct device	*dev = &as->pdev->dev;
984 
985 	xfer->tx_dma = xfer->rx_dma = INVALID_DMA_ADDRESS;
986 	if (xfer->tx_buf) {
987 		/* tx_buf is a const void* where we need a void * for the dma
988 		 * mapping */
989 		void *nonconst_tx = (void *)xfer->tx_buf;
990 
991 		xfer->tx_dma = dma_map_single(dev,
992 				nonconst_tx, xfer->len,
993 				DMA_TO_DEVICE);
994 		if (dma_mapping_error(dev, xfer->tx_dma))
995 			return -ENOMEM;
996 	}
997 	if (xfer->rx_buf) {
998 		xfer->rx_dma = dma_map_single(dev,
999 				xfer->rx_buf, xfer->len,
1000 				DMA_FROM_DEVICE);
1001 		if (dma_mapping_error(dev, xfer->rx_dma)) {
1002 			if (xfer->tx_buf)
1003 				dma_unmap_single(dev,
1004 						xfer->tx_dma, xfer->len,
1005 						DMA_TO_DEVICE);
1006 			return -ENOMEM;
1007 		}
1008 	}
1009 	return 0;
1010 }
1011 
1012 static void atmel_spi_dma_unmap_xfer(struct spi_master *master,
1013 				     struct spi_transfer *xfer)
1014 {
1015 	if (xfer->tx_dma != INVALID_DMA_ADDRESS)
1016 		dma_unmap_single(master->dev.parent, xfer->tx_dma,
1017 				 xfer->len, DMA_TO_DEVICE);
1018 	if (xfer->rx_dma != INVALID_DMA_ADDRESS)
1019 		dma_unmap_single(master->dev.parent, xfer->rx_dma,
1020 				 xfer->len, DMA_FROM_DEVICE);
1021 }
1022 
1023 static void atmel_spi_disable_pdc_transfer(struct atmel_spi *as)
1024 {
1025 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1026 }
1027 
1028 static void
1029 atmel_spi_pump_single_data(struct atmel_spi *as, struct spi_transfer *xfer)
1030 {
1031 	u8		*rxp;
1032 	u16		*rxp16;
1033 	unsigned long	xfer_pos = xfer->len - as->current_remaining_bytes;
1034 
1035 	if (xfer->rx_buf) {
1036 		if (xfer->bits_per_word > 8) {
1037 			rxp16 = (u16 *)(((u8 *)xfer->rx_buf) + xfer_pos);
1038 			*rxp16 = spi_readl(as, RDR);
1039 		} else {
1040 			rxp = ((u8 *)xfer->rx_buf) + xfer_pos;
1041 			*rxp = spi_readl(as, RDR);
1042 		}
1043 	} else {
1044 		spi_readl(as, RDR);
1045 	}
1046 	if (xfer->bits_per_word > 8) {
1047 		if (as->current_remaining_bytes > 2)
1048 			as->current_remaining_bytes -= 2;
1049 		else
1050 			as->current_remaining_bytes = 0;
1051 	} else {
1052 		as->current_remaining_bytes--;
1053 	}
1054 }
1055 
1056 static void
1057 atmel_spi_pump_fifo_data(struct atmel_spi *as, struct spi_transfer *xfer)
1058 {
1059 	u32 fifolr = spi_readl(as, FLR);
1060 	u32 num_bytes, num_data = SPI_BFEXT(RXFL, fifolr);
1061 	u32 offset = xfer->len - as->current_remaining_bytes;
1062 	u16 *words = (u16 *)((u8 *)xfer->rx_buf + offset);
1063 	u8  *bytes = (u8  *)((u8 *)xfer->rx_buf + offset);
1064 	u16 rd; /* RD field is the lowest 16 bits of RDR */
1065 
1066 	/* Update the number of remaining bytes to transfer */
1067 	num_bytes = ((xfer->bits_per_word > 8) ?
1068 		     (num_data << 1) :
1069 		     num_data);
1070 
1071 	if (as->current_remaining_bytes > num_bytes)
1072 		as->current_remaining_bytes -= num_bytes;
1073 	else
1074 		as->current_remaining_bytes = 0;
1075 
1076 	/* Handle odd number of bytes when data are more than 8bit width */
1077 	if (xfer->bits_per_word > 8)
1078 		as->current_remaining_bytes &= ~0x1;
1079 
1080 	/* Read data */
1081 	while (num_data) {
1082 		rd = spi_readl(as, RDR);
1083 		if (xfer->rx_buf) {
1084 			if (xfer->bits_per_word > 8)
1085 				*words++ = rd;
1086 			else
1087 				*bytes++ = rd;
1088 		}
1089 		num_data--;
1090 	}
1091 }
1092 
1093 /* Called from IRQ
1094  *
1095  * Must update "current_remaining_bytes" to keep track of data
1096  * to transfer.
1097  */
1098 static void
1099 atmel_spi_pump_pio_data(struct atmel_spi *as, struct spi_transfer *xfer)
1100 {
1101 	if (as->fifo_size)
1102 		atmel_spi_pump_fifo_data(as, xfer);
1103 	else
1104 		atmel_spi_pump_single_data(as, xfer);
1105 }
1106 
1107 /* Interrupt
1108  *
1109  * No need for locking in this Interrupt handler: done_status is the
1110  * only information modified.
1111  */
1112 static irqreturn_t
1113 atmel_spi_pio_interrupt(int irq, void *dev_id)
1114 {
1115 	struct spi_master	*master = dev_id;
1116 	struct atmel_spi	*as = spi_master_get_devdata(master);
1117 	u32			status, pending, imr;
1118 	struct spi_transfer	*xfer;
1119 	int			ret = IRQ_NONE;
1120 
1121 	imr = spi_readl(as, IMR);
1122 	status = spi_readl(as, SR);
1123 	pending = status & imr;
1124 
1125 	if (pending & SPI_BIT(OVRES)) {
1126 		ret = IRQ_HANDLED;
1127 		spi_writel(as, IDR, SPI_BIT(OVRES));
1128 		dev_warn(master->dev.parent, "overrun\n");
1129 
1130 		/*
1131 		 * When we get an overrun, we disregard the current
1132 		 * transfer. Data will not be copied back from any
1133 		 * bounce buffer and msg->actual_len will not be
1134 		 * updated with the last xfer.
1135 		 *
1136 		 * We will also not process any remaning transfers in
1137 		 * the message.
1138 		 */
1139 		as->done_status = -EIO;
1140 		smp_wmb();
1141 
1142 		/* Clear any overrun happening while cleaning up */
1143 		spi_readl(as, SR);
1144 
1145 		complete(&as->xfer_completion);
1146 
1147 	} else if (pending & (SPI_BIT(RDRF) | SPI_BIT(RXFTHF))) {
1148 		atmel_spi_lock(as);
1149 
1150 		if (as->current_remaining_bytes) {
1151 			ret = IRQ_HANDLED;
1152 			xfer = as->current_transfer;
1153 			atmel_spi_pump_pio_data(as, xfer);
1154 			if (!as->current_remaining_bytes)
1155 				spi_writel(as, IDR, pending);
1156 
1157 			complete(&as->xfer_completion);
1158 		}
1159 
1160 		atmel_spi_unlock(as);
1161 	} else {
1162 		WARN_ONCE(pending, "IRQ not handled, pending = %x\n", pending);
1163 		ret = IRQ_HANDLED;
1164 		spi_writel(as, IDR, pending);
1165 	}
1166 
1167 	return ret;
1168 }
1169 
1170 static irqreturn_t
1171 atmel_spi_pdc_interrupt(int irq, void *dev_id)
1172 {
1173 	struct spi_master	*master = dev_id;
1174 	struct atmel_spi	*as = spi_master_get_devdata(master);
1175 	u32			status, pending, imr;
1176 	int			ret = IRQ_NONE;
1177 
1178 	imr = spi_readl(as, IMR);
1179 	status = spi_readl(as, SR);
1180 	pending = status & imr;
1181 
1182 	if (pending & SPI_BIT(OVRES)) {
1183 
1184 		ret = IRQ_HANDLED;
1185 
1186 		spi_writel(as, IDR, (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX)
1187 				     | SPI_BIT(OVRES)));
1188 
1189 		/* Clear any overrun happening while cleaning up */
1190 		spi_readl(as, SR);
1191 
1192 		as->done_status = -EIO;
1193 
1194 		complete(&as->xfer_completion);
1195 
1196 	} else if (pending & (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX))) {
1197 		ret = IRQ_HANDLED;
1198 
1199 		spi_writel(as, IDR, pending);
1200 
1201 		complete(&as->xfer_completion);
1202 	}
1203 
1204 	return ret;
1205 }
1206 
1207 static int atmel_spi_setup(struct spi_device *spi)
1208 {
1209 	struct atmel_spi	*as;
1210 	struct atmel_spi_device	*asd;
1211 	u32			csr;
1212 	unsigned int		bits = spi->bits_per_word;
1213 	unsigned int		npcs_pin;
1214 	int			ret;
1215 
1216 	as = spi_master_get_devdata(spi->master);
1217 
1218 	/* see notes above re chipselect */
1219 	if (!atmel_spi_is_v2(as)
1220 			&& spi->chip_select == 0
1221 			&& (spi->mode & SPI_CS_HIGH)) {
1222 		dev_dbg(&spi->dev, "setup: can't be active-high\n");
1223 		return -EINVAL;
1224 	}
1225 
1226 	csr = SPI_BF(BITS, bits - 8);
1227 	if (spi->mode & SPI_CPOL)
1228 		csr |= SPI_BIT(CPOL);
1229 	if (!(spi->mode & SPI_CPHA))
1230 		csr |= SPI_BIT(NCPHA);
1231 	if (!as->use_cs_gpios)
1232 		csr |= SPI_BIT(CSAAT);
1233 
1234 	/* DLYBS is mostly irrelevant since we manage chipselect using GPIOs.
1235 	 *
1236 	 * DLYBCT would add delays between words, slowing down transfers.
1237 	 * It could potentially be useful to cope with DMA bottlenecks, but
1238 	 * in those cases it's probably best to just use a lower bitrate.
1239 	 */
1240 	csr |= SPI_BF(DLYBS, 0);
1241 	csr |= SPI_BF(DLYBCT, 0);
1242 
1243 	/* chipselect must have been muxed as GPIO (e.g. in board setup) */
1244 	npcs_pin = (unsigned long)spi->controller_data;
1245 
1246 	if (!as->use_cs_gpios)
1247 		npcs_pin = spi->chip_select;
1248 	else if (gpio_is_valid(spi->cs_gpio))
1249 		npcs_pin = spi->cs_gpio;
1250 
1251 	asd = spi->controller_state;
1252 	if (!asd) {
1253 		asd = kzalloc(sizeof(struct atmel_spi_device), GFP_KERNEL);
1254 		if (!asd)
1255 			return -ENOMEM;
1256 
1257 		if (as->use_cs_gpios) {
1258 			ret = gpio_request(npcs_pin, dev_name(&spi->dev));
1259 			if (ret) {
1260 				kfree(asd);
1261 				return ret;
1262 			}
1263 
1264 			gpio_direction_output(npcs_pin,
1265 					      !(spi->mode & SPI_CS_HIGH));
1266 		}
1267 
1268 		asd->npcs_pin = npcs_pin;
1269 		spi->controller_state = asd;
1270 	}
1271 
1272 	asd->csr = csr;
1273 
1274 	dev_dbg(&spi->dev,
1275 		"setup: bpw %u mode 0x%x -> csr%d %08x\n",
1276 		bits, spi->mode, spi->chip_select, csr);
1277 
1278 	if (!atmel_spi_is_v2(as))
1279 		spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
1280 
1281 	return 0;
1282 }
1283 
1284 static int atmel_spi_one_transfer(struct spi_master *master,
1285 					struct spi_message *msg,
1286 					struct spi_transfer *xfer)
1287 {
1288 	struct atmel_spi	*as;
1289 	struct spi_device	*spi = msg->spi;
1290 	u8			bits;
1291 	u32			len;
1292 	struct atmel_spi_device	*asd;
1293 	int			timeout;
1294 	int			ret;
1295 	unsigned long		dma_timeout;
1296 
1297 	as = spi_master_get_devdata(master);
1298 
1299 	if (!(xfer->tx_buf || xfer->rx_buf) && xfer->len) {
1300 		dev_dbg(&spi->dev, "missing rx or tx buf\n");
1301 		return -EINVAL;
1302 	}
1303 
1304 	if (xfer->bits_per_word) {
1305 		asd = spi->controller_state;
1306 		bits = (asd->csr >> 4) & 0xf;
1307 		if (bits != xfer->bits_per_word - 8) {
1308 			dev_dbg(&spi->dev,
1309 			"you can't yet change bits_per_word in transfers\n");
1310 			return -ENOPROTOOPT;
1311 		}
1312 	}
1313 
1314 	/*
1315 	 * DMA map early, for performance (empties dcache ASAP) and
1316 	 * better fault reporting.
1317 	 */
1318 	if ((!msg->is_dma_mapped)
1319 		&& (atmel_spi_use_dma(as, xfer)	|| as->use_pdc)) {
1320 		if (atmel_spi_dma_map_xfer(as, xfer) < 0)
1321 			return -ENOMEM;
1322 	}
1323 
1324 	atmel_spi_set_xfer_speed(as, msg->spi, xfer);
1325 
1326 	as->done_status = 0;
1327 	as->current_transfer = xfer;
1328 	as->current_remaining_bytes = xfer->len;
1329 	while (as->current_remaining_bytes) {
1330 		reinit_completion(&as->xfer_completion);
1331 
1332 		if (as->use_pdc) {
1333 			atmel_spi_pdc_next_xfer(master, msg, xfer);
1334 		} else if (atmel_spi_use_dma(as, xfer)) {
1335 			len = as->current_remaining_bytes;
1336 			ret = atmel_spi_next_xfer_dma_submit(master,
1337 								xfer, &len);
1338 			if (ret) {
1339 				dev_err(&spi->dev,
1340 					"unable to use DMA, fallback to PIO\n");
1341 				atmel_spi_next_xfer_pio(master, xfer);
1342 			} else {
1343 				as->current_remaining_bytes -= len;
1344 				if (as->current_remaining_bytes < 0)
1345 					as->current_remaining_bytes = 0;
1346 			}
1347 		} else {
1348 			atmel_spi_next_xfer_pio(master, xfer);
1349 		}
1350 
1351 		/* interrupts are disabled, so free the lock for schedule */
1352 		atmel_spi_unlock(as);
1353 		dma_timeout = wait_for_completion_timeout(&as->xfer_completion,
1354 							  SPI_DMA_TIMEOUT);
1355 		atmel_spi_lock(as);
1356 		if (WARN_ON(dma_timeout == 0)) {
1357 			dev_err(&spi->dev, "spi transfer timeout\n");
1358 			as->done_status = -EIO;
1359 		}
1360 
1361 		if (as->done_status)
1362 			break;
1363 	}
1364 
1365 	if (as->done_status) {
1366 		if (as->use_pdc) {
1367 			dev_warn(master->dev.parent,
1368 				"overrun (%u/%u remaining)\n",
1369 				spi_readl(as, TCR), spi_readl(as, RCR));
1370 
1371 			/*
1372 			 * Clean up DMA registers and make sure the data
1373 			 * registers are empty.
1374 			 */
1375 			spi_writel(as, RNCR, 0);
1376 			spi_writel(as, TNCR, 0);
1377 			spi_writel(as, RCR, 0);
1378 			spi_writel(as, TCR, 0);
1379 			for (timeout = 1000; timeout; timeout--)
1380 				if (spi_readl(as, SR) & SPI_BIT(TXEMPTY))
1381 					break;
1382 			if (!timeout)
1383 				dev_warn(master->dev.parent,
1384 					 "timeout waiting for TXEMPTY");
1385 			while (spi_readl(as, SR) & SPI_BIT(RDRF))
1386 				spi_readl(as, RDR);
1387 
1388 			/* Clear any overrun happening while cleaning up */
1389 			spi_readl(as, SR);
1390 
1391 		} else if (atmel_spi_use_dma(as, xfer)) {
1392 			atmel_spi_stop_dma(as);
1393 		}
1394 
1395 		if (!msg->is_dma_mapped
1396 			&& (atmel_spi_use_dma(as, xfer) || as->use_pdc))
1397 			atmel_spi_dma_unmap_xfer(master, xfer);
1398 
1399 		return 0;
1400 
1401 	} else {
1402 		/* only update length if no error */
1403 		msg->actual_length += xfer->len;
1404 	}
1405 
1406 	if (!msg->is_dma_mapped
1407 		&& (atmel_spi_use_dma(as, xfer) || as->use_pdc))
1408 		atmel_spi_dma_unmap_xfer(master, xfer);
1409 
1410 	if (xfer->delay_usecs)
1411 		udelay(xfer->delay_usecs);
1412 
1413 	if (xfer->cs_change) {
1414 		if (list_is_last(&xfer->transfer_list,
1415 				 &msg->transfers)) {
1416 			as->keep_cs = true;
1417 		} else {
1418 			as->cs_active = !as->cs_active;
1419 			if (as->cs_active)
1420 				cs_activate(as, msg->spi);
1421 			else
1422 				cs_deactivate(as, msg->spi);
1423 		}
1424 	}
1425 
1426 	return 0;
1427 }
1428 
1429 static int atmel_spi_transfer_one_message(struct spi_master *master,
1430 						struct spi_message *msg)
1431 {
1432 	struct atmel_spi *as;
1433 	struct spi_transfer *xfer;
1434 	struct spi_device *spi = msg->spi;
1435 	int ret = 0;
1436 
1437 	as = spi_master_get_devdata(master);
1438 
1439 	dev_dbg(&spi->dev, "new message %p submitted for %s\n",
1440 					msg, dev_name(&spi->dev));
1441 
1442 	atmel_spi_lock(as);
1443 	cs_activate(as, spi);
1444 
1445 	as->cs_active = true;
1446 	as->keep_cs = false;
1447 
1448 	msg->status = 0;
1449 	msg->actual_length = 0;
1450 
1451 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1452 		ret = atmel_spi_one_transfer(master, msg, xfer);
1453 		if (ret)
1454 			goto msg_done;
1455 	}
1456 
1457 	if (as->use_pdc)
1458 		atmel_spi_disable_pdc_transfer(as);
1459 
1460 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1461 		dev_dbg(&spi->dev,
1462 			"  xfer %p: len %u tx %p/%pad rx %p/%pad\n",
1463 			xfer, xfer->len,
1464 			xfer->tx_buf, &xfer->tx_dma,
1465 			xfer->rx_buf, &xfer->rx_dma);
1466 	}
1467 
1468 msg_done:
1469 	if (!as->keep_cs)
1470 		cs_deactivate(as, msg->spi);
1471 
1472 	atmel_spi_unlock(as);
1473 
1474 	msg->status = as->done_status;
1475 	spi_finalize_current_message(spi->master);
1476 
1477 	return ret;
1478 }
1479 
1480 static void atmel_spi_cleanup(struct spi_device *spi)
1481 {
1482 	struct atmel_spi_device	*asd = spi->controller_state;
1483 	unsigned		gpio = (unsigned long) spi->controller_data;
1484 
1485 	if (!asd)
1486 		return;
1487 
1488 	spi->controller_state = NULL;
1489 	gpio_free(gpio);
1490 	kfree(asd);
1491 }
1492 
1493 static inline unsigned int atmel_get_version(struct atmel_spi *as)
1494 {
1495 	return spi_readl(as, VERSION) & 0x00000fff;
1496 }
1497 
1498 static void atmel_get_caps(struct atmel_spi *as)
1499 {
1500 	unsigned int version;
1501 
1502 	version = atmel_get_version(as);
1503 	dev_info(&as->pdev->dev, "version: 0x%x\n", version);
1504 
1505 	as->caps.is_spi2 = version > 0x121;
1506 	as->caps.has_wdrbt = version >= 0x210;
1507 	as->caps.has_dma_support = version >= 0x212;
1508 }
1509 
1510 /*-------------------------------------------------------------------------*/
1511 
1512 static int atmel_spi_probe(struct platform_device *pdev)
1513 {
1514 	struct resource		*regs;
1515 	int			irq;
1516 	struct clk		*clk;
1517 	int			ret;
1518 	struct spi_master	*master;
1519 	struct atmel_spi	*as;
1520 
1521 	/* Select default pin state */
1522 	pinctrl_pm_select_default_state(&pdev->dev);
1523 
1524 	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1525 	if (!regs)
1526 		return -ENXIO;
1527 
1528 	irq = platform_get_irq(pdev, 0);
1529 	if (irq < 0)
1530 		return irq;
1531 
1532 	clk = devm_clk_get(&pdev->dev, "spi_clk");
1533 	if (IS_ERR(clk))
1534 		return PTR_ERR(clk);
1535 
1536 	/* setup spi core then atmel-specific driver state */
1537 	ret = -ENOMEM;
1538 	master = spi_alloc_master(&pdev->dev, sizeof(*as));
1539 	if (!master)
1540 		goto out_free;
1541 
1542 	/* the spi->mode bits understood by this driver: */
1543 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
1544 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(8, 16);
1545 	master->dev.of_node = pdev->dev.of_node;
1546 	master->bus_num = pdev->id;
1547 	master->num_chipselect = master->dev.of_node ? 0 : 4;
1548 	master->setup = atmel_spi_setup;
1549 	master->transfer_one_message = atmel_spi_transfer_one_message;
1550 	master->cleanup = atmel_spi_cleanup;
1551 	master->auto_runtime_pm = true;
1552 	platform_set_drvdata(pdev, master);
1553 
1554 	as = spi_master_get_devdata(master);
1555 
1556 	/*
1557 	 * Scratch buffer is used for throwaway rx and tx data.
1558 	 * It's coherent to minimize dcache pollution.
1559 	 */
1560 	as->buffer = dma_alloc_coherent(&pdev->dev, BUFFER_SIZE,
1561 					&as->buffer_dma, GFP_KERNEL);
1562 	if (!as->buffer)
1563 		goto out_free;
1564 
1565 	spin_lock_init(&as->lock);
1566 
1567 	as->pdev = pdev;
1568 	as->regs = devm_ioremap_resource(&pdev->dev, regs);
1569 	if (IS_ERR(as->regs)) {
1570 		ret = PTR_ERR(as->regs);
1571 		goto out_free_buffer;
1572 	}
1573 	as->phybase = regs->start;
1574 	as->irq = irq;
1575 	as->clk = clk;
1576 
1577 	init_completion(&as->xfer_completion);
1578 
1579 	atmel_get_caps(as);
1580 
1581 	as->use_cs_gpios = true;
1582 	if (atmel_spi_is_v2(as) &&
1583 	    !of_get_property(pdev->dev.of_node, "cs-gpios", NULL)) {
1584 		as->use_cs_gpios = false;
1585 		master->num_chipselect = 4;
1586 	}
1587 
1588 	as->use_dma = false;
1589 	as->use_pdc = false;
1590 	if (as->caps.has_dma_support) {
1591 		ret = atmel_spi_configure_dma(as);
1592 		if (ret == 0)
1593 			as->use_dma = true;
1594 		else if (ret == -EPROBE_DEFER)
1595 			return ret;
1596 	} else {
1597 		as->use_pdc = true;
1598 	}
1599 
1600 	if (as->caps.has_dma_support && !as->use_dma)
1601 		dev_info(&pdev->dev, "Atmel SPI Controller using PIO only\n");
1602 
1603 	if (as->use_pdc) {
1604 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pdc_interrupt,
1605 					0, dev_name(&pdev->dev), master);
1606 	} else {
1607 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pio_interrupt,
1608 					0, dev_name(&pdev->dev), master);
1609 	}
1610 	if (ret)
1611 		goto out_unmap_regs;
1612 
1613 	/* Initialize the hardware */
1614 	ret = clk_prepare_enable(clk);
1615 	if (ret)
1616 		goto out_free_irq;
1617 	spi_writel(as, CR, SPI_BIT(SWRST));
1618 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1619 	if (as->caps.has_wdrbt) {
1620 		spi_writel(as, MR, SPI_BIT(WDRBT) | SPI_BIT(MODFDIS)
1621 				| SPI_BIT(MSTR));
1622 	} else {
1623 		spi_writel(as, MR, SPI_BIT(MSTR) | SPI_BIT(MODFDIS));
1624 	}
1625 
1626 	if (as->use_pdc)
1627 		spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1628 	spi_writel(as, CR, SPI_BIT(SPIEN));
1629 
1630 	as->fifo_size = 0;
1631 	if (!of_property_read_u32(pdev->dev.of_node, "atmel,fifo-size",
1632 				  &as->fifo_size)) {
1633 		dev_info(&pdev->dev, "Using FIFO (%u data)\n", as->fifo_size);
1634 		spi_writel(as, CR, SPI_BIT(FIFOEN));
1635 	}
1636 
1637 	/* go! */
1638 	dev_info(&pdev->dev, "Atmel SPI Controller at 0x%08lx (irq %d)\n",
1639 			(unsigned long)regs->start, irq);
1640 
1641 	pm_runtime_set_autosuspend_delay(&pdev->dev, AUTOSUSPEND_TIMEOUT);
1642 	pm_runtime_use_autosuspend(&pdev->dev);
1643 	pm_runtime_set_active(&pdev->dev);
1644 	pm_runtime_enable(&pdev->dev);
1645 
1646 	ret = devm_spi_register_master(&pdev->dev, master);
1647 	if (ret)
1648 		goto out_free_dma;
1649 
1650 	return 0;
1651 
1652 out_free_dma:
1653 	pm_runtime_disable(&pdev->dev);
1654 	pm_runtime_set_suspended(&pdev->dev);
1655 
1656 	if (as->use_dma)
1657 		atmel_spi_release_dma(as);
1658 
1659 	spi_writel(as, CR, SPI_BIT(SWRST));
1660 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1661 	clk_disable_unprepare(clk);
1662 out_free_irq:
1663 out_unmap_regs:
1664 out_free_buffer:
1665 	dma_free_coherent(&pdev->dev, BUFFER_SIZE, as->buffer,
1666 			as->buffer_dma);
1667 out_free:
1668 	spi_master_put(master);
1669 	return ret;
1670 }
1671 
1672 static int atmel_spi_remove(struct platform_device *pdev)
1673 {
1674 	struct spi_master	*master = platform_get_drvdata(pdev);
1675 	struct atmel_spi	*as = spi_master_get_devdata(master);
1676 
1677 	pm_runtime_get_sync(&pdev->dev);
1678 
1679 	/* reset the hardware and block queue progress */
1680 	spin_lock_irq(&as->lock);
1681 	if (as->use_dma) {
1682 		atmel_spi_stop_dma(as);
1683 		atmel_spi_release_dma(as);
1684 	}
1685 
1686 	spi_writel(as, CR, SPI_BIT(SWRST));
1687 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1688 	spi_readl(as, SR);
1689 	spin_unlock_irq(&as->lock);
1690 
1691 	dma_free_coherent(&pdev->dev, BUFFER_SIZE, as->buffer,
1692 			as->buffer_dma);
1693 
1694 	clk_disable_unprepare(as->clk);
1695 
1696 	pm_runtime_put_noidle(&pdev->dev);
1697 	pm_runtime_disable(&pdev->dev);
1698 
1699 	return 0;
1700 }
1701 
1702 #ifdef CONFIG_PM
1703 static int atmel_spi_runtime_suspend(struct device *dev)
1704 {
1705 	struct spi_master *master = dev_get_drvdata(dev);
1706 	struct atmel_spi *as = spi_master_get_devdata(master);
1707 
1708 	clk_disable_unprepare(as->clk);
1709 	pinctrl_pm_select_sleep_state(dev);
1710 
1711 	return 0;
1712 }
1713 
1714 static int atmel_spi_runtime_resume(struct device *dev)
1715 {
1716 	struct spi_master *master = dev_get_drvdata(dev);
1717 	struct atmel_spi *as = spi_master_get_devdata(master);
1718 
1719 	pinctrl_pm_select_default_state(dev);
1720 
1721 	return clk_prepare_enable(as->clk);
1722 }
1723 
1724 static int atmel_spi_suspend(struct device *dev)
1725 {
1726 	struct spi_master *master = dev_get_drvdata(dev);
1727 	int ret;
1728 
1729 	/* Stop the queue running */
1730 	ret = spi_master_suspend(master);
1731 	if (ret) {
1732 		dev_warn(dev, "cannot suspend master\n");
1733 		return ret;
1734 	}
1735 
1736 	if (!pm_runtime_suspended(dev))
1737 		atmel_spi_runtime_suspend(dev);
1738 
1739 	return 0;
1740 }
1741 
1742 static int atmel_spi_resume(struct device *dev)
1743 {
1744 	struct spi_master *master = dev_get_drvdata(dev);
1745 	int ret;
1746 
1747 	if (!pm_runtime_suspended(dev)) {
1748 		ret = atmel_spi_runtime_resume(dev);
1749 		if (ret)
1750 			return ret;
1751 	}
1752 
1753 	/* Start the queue running */
1754 	ret = spi_master_resume(master);
1755 	if (ret)
1756 		dev_err(dev, "problem starting queue (%d)\n", ret);
1757 
1758 	return ret;
1759 }
1760 
1761 static const struct dev_pm_ops atmel_spi_pm_ops = {
1762 	SET_SYSTEM_SLEEP_PM_OPS(atmel_spi_suspend, atmel_spi_resume)
1763 	SET_RUNTIME_PM_OPS(atmel_spi_runtime_suspend,
1764 			   atmel_spi_runtime_resume, NULL)
1765 };
1766 #define ATMEL_SPI_PM_OPS	(&atmel_spi_pm_ops)
1767 #else
1768 #define ATMEL_SPI_PM_OPS	NULL
1769 #endif
1770 
1771 #if defined(CONFIG_OF)
1772 static const struct of_device_id atmel_spi_dt_ids[] = {
1773 	{ .compatible = "atmel,at91rm9200-spi" },
1774 	{ /* sentinel */ }
1775 };
1776 
1777 MODULE_DEVICE_TABLE(of, atmel_spi_dt_ids);
1778 #endif
1779 
1780 static struct platform_driver atmel_spi_driver = {
1781 	.driver		= {
1782 		.name	= "atmel_spi",
1783 		.pm	= ATMEL_SPI_PM_OPS,
1784 		.of_match_table	= of_match_ptr(atmel_spi_dt_ids),
1785 	},
1786 	.probe		= atmel_spi_probe,
1787 	.remove		= atmel_spi_remove,
1788 };
1789 module_platform_driver(atmel_spi_driver);
1790 
1791 MODULE_DESCRIPTION("Atmel AT32/AT91 SPI Controller driver");
1792 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1793 MODULE_LICENSE("GPL");
1794 MODULE_ALIAS("platform:atmel_spi");
1795