xref: /linux/drivers/spi/spi-atmel.c (revision d060296cc0300ae8ed08004ebd3994bf325fa257)
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 	bool	has_pdc_support;
273 };
274 
275 /*
276  * The core SPI transfer engine just talks to a register bank to set up
277  * DMA transfers; transfer queue progress is driven by IRQs.  The clock
278  * framework provides the base clock, subdivided for each spi_device.
279  */
280 struct atmel_spi {
281 	spinlock_t		lock;
282 	unsigned long		flags;
283 
284 	phys_addr_t		phybase;
285 	void __iomem		*regs;
286 	int			irq;
287 	struct clk		*clk;
288 	struct platform_device	*pdev;
289 	unsigned long		spi_clk;
290 
291 	struct spi_transfer	*current_transfer;
292 	int			current_remaining_bytes;
293 	int			done_status;
294 
295 	struct completion	xfer_completion;
296 
297 	struct atmel_spi_caps	caps;
298 
299 	bool			use_dma;
300 	bool			use_pdc;
301 	bool			use_cs_gpios;
302 
303 	bool			keep_cs;
304 	bool			cs_active;
305 
306 	u32			fifo_size;
307 };
308 
309 /* Controller-specific per-slave state */
310 struct atmel_spi_device {
311 	unsigned int		npcs_pin;
312 	u32			csr;
313 };
314 
315 #define SPI_MAX_DMA_XFER	65535 /* true for both PDC and DMA */
316 #define INVALID_DMA_ADDRESS	0xffffffff
317 
318 /*
319  * Version 2 of the SPI controller has
320  *  - CR.LASTXFER
321  *  - SPI_MR.DIV32 may become FDIV or must-be-zero (here: always zero)
322  *  - SPI_SR.TXEMPTY, SPI_SR.NSSR (and corresponding irqs)
323  *  - SPI_CSRx.CSAAT
324  *  - SPI_CSRx.SBCR allows faster clocking
325  */
326 static bool atmel_spi_is_v2(struct atmel_spi *as)
327 {
328 	return as->caps.is_spi2;
329 }
330 
331 /*
332  * Earlier SPI controllers (e.g. on at91rm9200) have a design bug whereby
333  * they assume that spi slave device state will not change on deselect, so
334  * that automagic deselection is OK.  ("NPCSx rises if no data is to be
335  * transmitted")  Not so!  Workaround uses nCSx pins as GPIOs; or newer
336  * controllers have CSAAT and friends.
337  *
338  * Since the CSAAT functionality is a bit weird on newer controllers as
339  * well, we use GPIO to control nCSx pins on all controllers, updating
340  * MR.PCS to avoid confusing the controller.  Using GPIOs also lets us
341  * support active-high chipselects despite the controller's belief that
342  * only active-low devices/systems exists.
343  *
344  * However, at91rm9200 has a second erratum whereby nCS0 doesn't work
345  * right when driven with GPIO.  ("Mode Fault does not allow more than one
346  * Master on Chip Select 0.")  No workaround exists for that ... so for
347  * nCS0 on that chip, we (a) don't use the GPIO, (b) can't support CS_HIGH,
348  * and (c) will trigger that first erratum in some cases.
349  */
350 
351 static void cs_activate(struct atmel_spi *as, struct spi_device *spi)
352 {
353 	struct atmel_spi_device *asd = spi->controller_state;
354 	unsigned active = spi->mode & SPI_CS_HIGH;
355 	u32 mr;
356 
357 	if (atmel_spi_is_v2(as)) {
358 		spi_writel(as, CSR0 + 4 * spi->chip_select, asd->csr);
359 		/* For the low SPI version, there is a issue that PDC transfer
360 		 * on CS1,2,3 needs SPI_CSR0.BITS config as SPI_CSR1,2,3.BITS
361 		 */
362 		spi_writel(as, CSR0, asd->csr);
363 		if (as->caps.has_wdrbt) {
364 			spi_writel(as, MR,
365 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
366 					| SPI_BIT(WDRBT)
367 					| SPI_BIT(MODFDIS)
368 					| SPI_BIT(MSTR));
369 		} else {
370 			spi_writel(as, MR,
371 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
372 					| SPI_BIT(MODFDIS)
373 					| SPI_BIT(MSTR));
374 		}
375 
376 		mr = spi_readl(as, MR);
377 		if (as->use_cs_gpios)
378 			gpio_set_value(asd->npcs_pin, active);
379 	} else {
380 		u32 cpol = (spi->mode & SPI_CPOL) ? SPI_BIT(CPOL) : 0;
381 		int i;
382 		u32 csr;
383 
384 		/* Make sure clock polarity is correct */
385 		for (i = 0; i < spi->master->num_chipselect; i++) {
386 			csr = spi_readl(as, CSR0 + 4 * i);
387 			if ((csr ^ cpol) & SPI_BIT(CPOL))
388 				spi_writel(as, CSR0 + 4 * i,
389 						csr ^ SPI_BIT(CPOL));
390 		}
391 
392 		mr = spi_readl(as, MR);
393 		mr = SPI_BFINS(PCS, ~(1 << spi->chip_select), mr);
394 		if (as->use_cs_gpios && spi->chip_select != 0)
395 			gpio_set_value(asd->npcs_pin, active);
396 		spi_writel(as, MR, mr);
397 	}
398 
399 	dev_dbg(&spi->dev, "activate %u%s, mr %08x\n",
400 			asd->npcs_pin, active ? " (high)" : "",
401 			mr);
402 }
403 
404 static void cs_deactivate(struct atmel_spi *as, struct spi_device *spi)
405 {
406 	struct atmel_spi_device *asd = spi->controller_state;
407 	unsigned active = spi->mode & SPI_CS_HIGH;
408 	u32 mr;
409 
410 	/* only deactivate *this* device; sometimes transfers to
411 	 * another device may be active when this routine is called.
412 	 */
413 	mr = spi_readl(as, MR);
414 	if (~SPI_BFEXT(PCS, mr) & (1 << spi->chip_select)) {
415 		mr = SPI_BFINS(PCS, 0xf, mr);
416 		spi_writel(as, MR, mr);
417 	}
418 
419 	dev_dbg(&spi->dev, "DEactivate %u%s, mr %08x\n",
420 			asd->npcs_pin, active ? " (low)" : "",
421 			mr);
422 
423 	if (!as->use_cs_gpios)
424 		spi_writel(as, CR, SPI_BIT(LASTXFER));
425 	else if (atmel_spi_is_v2(as) || spi->chip_select != 0)
426 		gpio_set_value(asd->npcs_pin, !active);
427 }
428 
429 static void atmel_spi_lock(struct atmel_spi *as) __acquires(&as->lock)
430 {
431 	spin_lock_irqsave(&as->lock, as->flags);
432 }
433 
434 static void atmel_spi_unlock(struct atmel_spi *as) __releases(&as->lock)
435 {
436 	spin_unlock_irqrestore(&as->lock, as->flags);
437 }
438 
439 static inline bool atmel_spi_use_dma(struct atmel_spi *as,
440 				struct spi_transfer *xfer)
441 {
442 	return as->use_dma && xfer->len >= DMA_MIN_BYTES;
443 }
444 
445 static bool atmel_spi_can_dma(struct spi_master *master,
446 			      struct spi_device *spi,
447 			      struct spi_transfer *xfer)
448 {
449 	struct atmel_spi *as = spi_master_get_devdata(master);
450 
451 	return atmel_spi_use_dma(as, xfer);
452 }
453 
454 static int atmel_spi_dma_slave_config(struct atmel_spi *as,
455 				struct dma_slave_config *slave_config,
456 				u8 bits_per_word)
457 {
458 	struct spi_master *master = platform_get_drvdata(as->pdev);
459 	int err = 0;
460 
461 	if (bits_per_word > 8) {
462 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
463 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
464 	} else {
465 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
466 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
467 	}
468 
469 	slave_config->dst_addr = (dma_addr_t)as->phybase + SPI_TDR;
470 	slave_config->src_addr = (dma_addr_t)as->phybase + SPI_RDR;
471 	slave_config->src_maxburst = 1;
472 	slave_config->dst_maxburst = 1;
473 	slave_config->device_fc = false;
474 
475 	/*
476 	 * This driver uses fixed peripheral select mode (PS bit set to '0' in
477 	 * the Mode Register).
478 	 * So according to the datasheet, when FIFOs are available (and
479 	 * enabled), the Transmit FIFO operates in Multiple Data Mode.
480 	 * In this mode, up to 2 data, not 4, can be written into the Transmit
481 	 * Data Register in a single access.
482 	 * However, the first data has to be written into the lowest 16 bits and
483 	 * the second data into the highest 16 bits of the Transmit
484 	 * Data Register. For 8bit data (the most frequent case), it would
485 	 * require to rework tx_buf so each data would actualy fit 16 bits.
486 	 * So we'd rather write only one data at the time. Hence the transmit
487 	 * path works the same whether FIFOs are available (and enabled) or not.
488 	 */
489 	slave_config->direction = DMA_MEM_TO_DEV;
490 	if (dmaengine_slave_config(master->dma_tx, slave_config)) {
491 		dev_err(&as->pdev->dev,
492 			"failed to configure tx dma channel\n");
493 		err = -EINVAL;
494 	}
495 
496 	/*
497 	 * This driver configures the spi controller for master mode (MSTR bit
498 	 * set to '1' in the Mode Register).
499 	 * So according to the datasheet, when FIFOs are available (and
500 	 * enabled), the Receive FIFO operates in Single Data Mode.
501 	 * So the receive path works the same whether FIFOs are available (and
502 	 * enabled) or not.
503 	 */
504 	slave_config->direction = DMA_DEV_TO_MEM;
505 	if (dmaengine_slave_config(master->dma_rx, slave_config)) {
506 		dev_err(&as->pdev->dev,
507 			"failed to configure rx dma channel\n");
508 		err = -EINVAL;
509 	}
510 
511 	return err;
512 }
513 
514 static int atmel_spi_configure_dma(struct spi_master *master,
515 				   struct atmel_spi *as)
516 {
517 	struct dma_slave_config	slave_config;
518 	struct device *dev = &as->pdev->dev;
519 	int err;
520 
521 	dma_cap_mask_t mask;
522 	dma_cap_zero(mask);
523 	dma_cap_set(DMA_SLAVE, mask);
524 
525 	master->dma_tx = dma_request_slave_channel_reason(dev, "tx");
526 	if (IS_ERR(master->dma_tx)) {
527 		err = PTR_ERR(master->dma_tx);
528 		if (err == -EPROBE_DEFER) {
529 			dev_warn(dev, "no DMA channel available at the moment\n");
530 			goto error_clear;
531 		}
532 		dev_err(dev,
533 			"DMA TX channel not available, SPI unable to use DMA\n");
534 		err = -EBUSY;
535 		goto error_clear;
536 	}
537 
538 	/*
539 	 * No reason to check EPROBE_DEFER here since we have already requested
540 	 * tx channel. If it fails here, it's for another reason.
541 	 */
542 	master->dma_rx = dma_request_slave_channel(dev, "rx");
543 
544 	if (!master->dma_rx) {
545 		dev_err(dev,
546 			"DMA RX channel not available, SPI unable to use DMA\n");
547 		err = -EBUSY;
548 		goto error;
549 	}
550 
551 	err = atmel_spi_dma_slave_config(as, &slave_config, 8);
552 	if (err)
553 		goto error;
554 
555 	dev_info(&as->pdev->dev,
556 			"Using %s (tx) and %s (rx) for DMA transfers\n",
557 			dma_chan_name(master->dma_tx),
558 			dma_chan_name(master->dma_rx));
559 
560 	return 0;
561 error:
562 	if (master->dma_rx)
563 		dma_release_channel(master->dma_rx);
564 	if (!IS_ERR(master->dma_tx))
565 		dma_release_channel(master->dma_tx);
566 error_clear:
567 	master->dma_tx = master->dma_rx = NULL;
568 	return err;
569 }
570 
571 static void atmel_spi_stop_dma(struct spi_master *master)
572 {
573 	if (master->dma_rx)
574 		dmaengine_terminate_all(master->dma_rx);
575 	if (master->dma_tx)
576 		dmaengine_terminate_all(master->dma_tx);
577 }
578 
579 static void atmel_spi_release_dma(struct spi_master *master)
580 {
581 	if (master->dma_rx) {
582 		dma_release_channel(master->dma_rx);
583 		master->dma_rx = NULL;
584 	}
585 	if (master->dma_tx) {
586 		dma_release_channel(master->dma_tx);
587 		master->dma_tx = NULL;
588 	}
589 }
590 
591 /* This function is called by the DMA driver from tasklet context */
592 static void dma_callback(void *data)
593 {
594 	struct spi_master	*master = data;
595 	struct atmel_spi	*as = spi_master_get_devdata(master);
596 
597 	complete(&as->xfer_completion);
598 }
599 
600 /*
601  * Next transfer using PIO without FIFO.
602  */
603 static void atmel_spi_next_xfer_single(struct spi_master *master,
604 				       struct spi_transfer *xfer)
605 {
606 	struct atmel_spi	*as = spi_master_get_devdata(master);
607 	unsigned long xfer_pos = xfer->len - as->current_remaining_bytes;
608 
609 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_pio\n");
610 
611 	/* Make sure data is not remaining in RDR */
612 	spi_readl(as, RDR);
613 	while (spi_readl(as, SR) & SPI_BIT(RDRF)) {
614 		spi_readl(as, RDR);
615 		cpu_relax();
616 	}
617 
618 	if (xfer->bits_per_word > 8)
619 		spi_writel(as, TDR, *(u16 *)(xfer->tx_buf + xfer_pos));
620 	else
621 		spi_writel(as, TDR, *(u8 *)(xfer->tx_buf + xfer_pos));
622 
623 	dev_dbg(master->dev.parent,
624 		"  start pio xfer %p: len %u tx %p rx %p bitpw %d\n",
625 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
626 		xfer->bits_per_word);
627 
628 	/* Enable relevant interrupts */
629 	spi_writel(as, IER, SPI_BIT(RDRF) | SPI_BIT(OVRES));
630 }
631 
632 /*
633  * Next transfer using PIO with FIFO.
634  */
635 static void atmel_spi_next_xfer_fifo(struct spi_master *master,
636 				     struct spi_transfer *xfer)
637 {
638 	struct atmel_spi *as = spi_master_get_devdata(master);
639 	u32 current_remaining_data, num_data;
640 	u32 offset = xfer->len - as->current_remaining_bytes;
641 	const u16 *words = (const u16 *)((u8 *)xfer->tx_buf + offset);
642 	const u8  *bytes = (const u8  *)((u8 *)xfer->tx_buf + offset);
643 	u16 td0, td1;
644 	u32 fifomr;
645 
646 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_fifo\n");
647 
648 	/* Compute the number of data to transfer in the current iteration */
649 	current_remaining_data = ((xfer->bits_per_word > 8) ?
650 				  ((u32)as->current_remaining_bytes >> 1) :
651 				  (u32)as->current_remaining_bytes);
652 	num_data = min(current_remaining_data, as->fifo_size);
653 
654 	/* Flush RX and TX FIFOs */
655 	spi_writel(as, CR, SPI_BIT(RXFCLR) | SPI_BIT(TXFCLR));
656 	while (spi_readl(as, FLR))
657 		cpu_relax();
658 
659 	/* Set RX FIFO Threshold to the number of data to transfer */
660 	fifomr = spi_readl(as, FMR);
661 	spi_writel(as, FMR, SPI_BFINS(RXFTHRES, num_data, fifomr));
662 
663 	/* Clear FIFO flags in the Status Register, especially RXFTHF */
664 	(void)spi_readl(as, SR);
665 
666 	/* Fill TX FIFO */
667 	while (num_data >= 2) {
668 		if (xfer->bits_per_word > 8) {
669 			td0 = *words++;
670 			td1 = *words++;
671 		} else {
672 			td0 = *bytes++;
673 			td1 = *bytes++;
674 		}
675 
676 		spi_writel(as, TDR, (td1 << 16) | td0);
677 		num_data -= 2;
678 	}
679 
680 	if (num_data) {
681 		if (xfer->bits_per_word > 8)
682 			td0 = *words++;
683 		else
684 			td0 = *bytes++;
685 
686 		spi_writew(as, TDR, td0);
687 		num_data--;
688 	}
689 
690 	dev_dbg(master->dev.parent,
691 		"  start fifo xfer %p: len %u tx %p rx %p bitpw %d\n",
692 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
693 		xfer->bits_per_word);
694 
695 	/*
696 	 * Enable RX FIFO Threshold Flag interrupt to be notified about
697 	 * transfer completion.
698 	 */
699 	spi_writel(as, IER, SPI_BIT(RXFTHF) | SPI_BIT(OVRES));
700 }
701 
702 /*
703  * Next transfer using PIO.
704  */
705 static void atmel_spi_next_xfer_pio(struct spi_master *master,
706 				    struct spi_transfer *xfer)
707 {
708 	struct atmel_spi *as = spi_master_get_devdata(master);
709 
710 	if (as->fifo_size)
711 		atmel_spi_next_xfer_fifo(master, xfer);
712 	else
713 		atmel_spi_next_xfer_single(master, xfer);
714 }
715 
716 /*
717  * Submit next transfer for DMA.
718  */
719 static int atmel_spi_next_xfer_dma_submit(struct spi_master *master,
720 				struct spi_transfer *xfer,
721 				u32 *plen)
722 {
723 	struct atmel_spi	*as = spi_master_get_devdata(master);
724 	struct dma_chan		*rxchan = master->dma_rx;
725 	struct dma_chan		*txchan = master->dma_tx;
726 	struct dma_async_tx_descriptor *rxdesc;
727 	struct dma_async_tx_descriptor *txdesc;
728 	struct dma_slave_config	slave_config;
729 	dma_cookie_t		cookie;
730 
731 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_dma_submit\n");
732 
733 	/* Check that the channels are available */
734 	if (!rxchan || !txchan)
735 		return -ENODEV;
736 
737 	/* release lock for DMA operations */
738 	atmel_spi_unlock(as);
739 
740 	*plen = xfer->len;
741 
742 	if (atmel_spi_dma_slave_config(as, &slave_config,
743 				       xfer->bits_per_word))
744 		goto err_exit;
745 
746 	/* Send both scatterlists */
747 	rxdesc = dmaengine_prep_slave_sg(rxchan,
748 					 xfer->rx_sg.sgl, xfer->rx_sg.nents,
749 					 DMA_FROM_DEVICE,
750 					 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
751 	if (!rxdesc)
752 		goto err_dma;
753 
754 	txdesc = dmaengine_prep_slave_sg(txchan,
755 					 xfer->tx_sg.sgl, xfer->tx_sg.nents,
756 					 DMA_TO_DEVICE,
757 					 DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
758 	if (!txdesc)
759 		goto err_dma;
760 
761 	dev_dbg(master->dev.parent,
762 		"  start dma xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
763 		xfer, xfer->len, xfer->tx_buf, (unsigned long long)xfer->tx_dma,
764 		xfer->rx_buf, (unsigned long long)xfer->rx_dma);
765 
766 	/* Enable relevant interrupts */
767 	spi_writel(as, IER, SPI_BIT(OVRES));
768 
769 	/* Put the callback on the RX transfer only, that should finish last */
770 	rxdesc->callback = dma_callback;
771 	rxdesc->callback_param = master;
772 
773 	/* Submit and fire RX and TX with TX last so we're ready to read! */
774 	cookie = rxdesc->tx_submit(rxdesc);
775 	if (dma_submit_error(cookie))
776 		goto err_dma;
777 	cookie = txdesc->tx_submit(txdesc);
778 	if (dma_submit_error(cookie))
779 		goto err_dma;
780 	rxchan->device->device_issue_pending(rxchan);
781 	txchan->device->device_issue_pending(txchan);
782 
783 	/* take back lock */
784 	atmel_spi_lock(as);
785 	return 0;
786 
787 err_dma:
788 	spi_writel(as, IDR, SPI_BIT(OVRES));
789 	atmel_spi_stop_dma(master);
790 err_exit:
791 	atmel_spi_lock(as);
792 	return -ENOMEM;
793 }
794 
795 static void atmel_spi_next_xfer_data(struct spi_master *master,
796 				struct spi_transfer *xfer,
797 				dma_addr_t *tx_dma,
798 				dma_addr_t *rx_dma,
799 				u32 *plen)
800 {
801 	*rx_dma = xfer->rx_dma + xfer->len - *plen;
802 	*tx_dma = xfer->tx_dma + xfer->len - *plen;
803 	if (*plen > master->max_dma_len)
804 		*plen = master->max_dma_len;
805 }
806 
807 static int atmel_spi_set_xfer_speed(struct atmel_spi *as,
808 				    struct spi_device *spi,
809 				    struct spi_transfer *xfer)
810 {
811 	u32			scbr, csr;
812 	unsigned long		bus_hz;
813 
814 	/* v1 chips start out at half the peripheral bus speed. */
815 	bus_hz = as->spi_clk;
816 	if (!atmel_spi_is_v2(as))
817 		bus_hz /= 2;
818 
819 	/*
820 	 * Calculate the lowest divider that satisfies the
821 	 * constraint, assuming div32/fdiv/mbz == 0.
822 	 */
823 	scbr = DIV_ROUND_UP(bus_hz, xfer->speed_hz);
824 
825 	/*
826 	 * If the resulting divider doesn't fit into the
827 	 * register bitfield, we can't satisfy the constraint.
828 	 */
829 	if (scbr >= (1 << SPI_SCBR_SIZE)) {
830 		dev_err(&spi->dev,
831 			"setup: %d Hz too slow, scbr %u; min %ld Hz\n",
832 			xfer->speed_hz, scbr, bus_hz/255);
833 		return -EINVAL;
834 	}
835 	if (scbr == 0) {
836 		dev_err(&spi->dev,
837 			"setup: %d Hz too high, scbr %u; max %ld Hz\n",
838 			xfer->speed_hz, scbr, bus_hz);
839 		return -EINVAL;
840 	}
841 	csr = spi_readl(as, CSR0 + 4 * spi->chip_select);
842 	csr = SPI_BFINS(SCBR, scbr, csr);
843 	spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
844 
845 	return 0;
846 }
847 
848 /*
849  * Submit next transfer for PDC.
850  * lock is held, spi irq is blocked
851  */
852 static void atmel_spi_pdc_next_xfer(struct spi_master *master,
853 					struct spi_message *msg,
854 					struct spi_transfer *xfer)
855 {
856 	struct atmel_spi	*as = spi_master_get_devdata(master);
857 	u32			len;
858 	dma_addr_t		tx_dma, rx_dma;
859 
860 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
861 
862 	len = as->current_remaining_bytes;
863 	atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
864 	as->current_remaining_bytes -= len;
865 
866 	spi_writel(as, RPR, rx_dma);
867 	spi_writel(as, TPR, tx_dma);
868 
869 	if (msg->spi->bits_per_word > 8)
870 		len >>= 1;
871 	spi_writel(as, RCR, len);
872 	spi_writel(as, TCR, len);
873 
874 	dev_dbg(&msg->spi->dev,
875 		"  start xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
876 		xfer, xfer->len, xfer->tx_buf,
877 		(unsigned long long)xfer->tx_dma, xfer->rx_buf,
878 		(unsigned long long)xfer->rx_dma);
879 
880 	if (as->current_remaining_bytes) {
881 		len = as->current_remaining_bytes;
882 		atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
883 		as->current_remaining_bytes -= len;
884 
885 		spi_writel(as, RNPR, rx_dma);
886 		spi_writel(as, TNPR, tx_dma);
887 
888 		if (msg->spi->bits_per_word > 8)
889 			len >>= 1;
890 		spi_writel(as, RNCR, len);
891 		spi_writel(as, TNCR, len);
892 
893 		dev_dbg(&msg->spi->dev,
894 			"  next xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
895 			xfer, xfer->len, xfer->tx_buf,
896 			(unsigned long long)xfer->tx_dma, xfer->rx_buf,
897 			(unsigned long long)xfer->rx_dma);
898 	}
899 
900 	/* REVISIT: We're waiting for RXBUFF before we start the next
901 	 * transfer because we need to handle some difficult timing
902 	 * issues otherwise. If we wait for TXBUFE in one transfer and
903 	 * then starts waiting for RXBUFF in the next, it's difficult
904 	 * to tell the difference between the RXBUFF interrupt we're
905 	 * actually waiting for and the RXBUFF interrupt of the
906 	 * previous transfer.
907 	 *
908 	 * It should be doable, though. Just not now...
909 	 */
910 	spi_writel(as, IER, SPI_BIT(RXBUFF) | SPI_BIT(OVRES));
911 	spi_writel(as, PTCR, SPI_BIT(TXTEN) | SPI_BIT(RXTEN));
912 }
913 
914 /*
915  * For DMA, tx_buf/tx_dma have the same relationship as rx_buf/rx_dma:
916  *  - The buffer is either valid for CPU access, else NULL
917  *  - If the buffer is valid, so is its DMA address
918  *
919  * This driver manages the dma address unless message->is_dma_mapped.
920  */
921 static int
922 atmel_spi_dma_map_xfer(struct atmel_spi *as, struct spi_transfer *xfer)
923 {
924 	struct device	*dev = &as->pdev->dev;
925 
926 	xfer->tx_dma = xfer->rx_dma = INVALID_DMA_ADDRESS;
927 	if (xfer->tx_buf) {
928 		/* tx_buf is a const void* where we need a void * for the dma
929 		 * mapping */
930 		void *nonconst_tx = (void *)xfer->tx_buf;
931 
932 		xfer->tx_dma = dma_map_single(dev,
933 				nonconst_tx, xfer->len,
934 				DMA_TO_DEVICE);
935 		if (dma_mapping_error(dev, xfer->tx_dma))
936 			return -ENOMEM;
937 	}
938 	if (xfer->rx_buf) {
939 		xfer->rx_dma = dma_map_single(dev,
940 				xfer->rx_buf, xfer->len,
941 				DMA_FROM_DEVICE);
942 		if (dma_mapping_error(dev, xfer->rx_dma)) {
943 			if (xfer->tx_buf)
944 				dma_unmap_single(dev,
945 						xfer->tx_dma, xfer->len,
946 						DMA_TO_DEVICE);
947 			return -ENOMEM;
948 		}
949 	}
950 	return 0;
951 }
952 
953 static void atmel_spi_dma_unmap_xfer(struct spi_master *master,
954 				     struct spi_transfer *xfer)
955 {
956 	if (xfer->tx_dma != INVALID_DMA_ADDRESS)
957 		dma_unmap_single(master->dev.parent, xfer->tx_dma,
958 				 xfer->len, DMA_TO_DEVICE);
959 	if (xfer->rx_dma != INVALID_DMA_ADDRESS)
960 		dma_unmap_single(master->dev.parent, xfer->rx_dma,
961 				 xfer->len, DMA_FROM_DEVICE);
962 }
963 
964 static void atmel_spi_disable_pdc_transfer(struct atmel_spi *as)
965 {
966 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
967 }
968 
969 static void
970 atmel_spi_pump_single_data(struct atmel_spi *as, struct spi_transfer *xfer)
971 {
972 	u8		*rxp;
973 	u16		*rxp16;
974 	unsigned long	xfer_pos = xfer->len - as->current_remaining_bytes;
975 
976 	if (xfer->bits_per_word > 8) {
977 		rxp16 = (u16 *)(((u8 *)xfer->rx_buf) + xfer_pos);
978 		*rxp16 = spi_readl(as, RDR);
979 	} else {
980 		rxp = ((u8 *)xfer->rx_buf) + xfer_pos;
981 		*rxp = spi_readl(as, RDR);
982 	}
983 	if (xfer->bits_per_word > 8) {
984 		if (as->current_remaining_bytes > 2)
985 			as->current_remaining_bytes -= 2;
986 		else
987 			as->current_remaining_bytes = 0;
988 	} else {
989 		as->current_remaining_bytes--;
990 	}
991 }
992 
993 static void
994 atmel_spi_pump_fifo_data(struct atmel_spi *as, struct spi_transfer *xfer)
995 {
996 	u32 fifolr = spi_readl(as, FLR);
997 	u32 num_bytes, num_data = SPI_BFEXT(RXFL, fifolr);
998 	u32 offset = xfer->len - as->current_remaining_bytes;
999 	u16 *words = (u16 *)((u8 *)xfer->rx_buf + offset);
1000 	u8  *bytes = (u8  *)((u8 *)xfer->rx_buf + offset);
1001 	u16 rd; /* RD field is the lowest 16 bits of RDR */
1002 
1003 	/* Update the number of remaining bytes to transfer */
1004 	num_bytes = ((xfer->bits_per_word > 8) ?
1005 		     (num_data << 1) :
1006 		     num_data);
1007 
1008 	if (as->current_remaining_bytes > num_bytes)
1009 		as->current_remaining_bytes -= num_bytes;
1010 	else
1011 		as->current_remaining_bytes = 0;
1012 
1013 	/* Handle odd number of bytes when data are more than 8bit width */
1014 	if (xfer->bits_per_word > 8)
1015 		as->current_remaining_bytes &= ~0x1;
1016 
1017 	/* Read data */
1018 	while (num_data) {
1019 		rd = spi_readl(as, RDR);
1020 		if (xfer->bits_per_word > 8)
1021 			*words++ = rd;
1022 		else
1023 			*bytes++ = rd;
1024 		num_data--;
1025 	}
1026 }
1027 
1028 /* Called from IRQ
1029  *
1030  * Must update "current_remaining_bytes" to keep track of data
1031  * to transfer.
1032  */
1033 static void
1034 atmel_spi_pump_pio_data(struct atmel_spi *as, struct spi_transfer *xfer)
1035 {
1036 	if (as->fifo_size)
1037 		atmel_spi_pump_fifo_data(as, xfer);
1038 	else
1039 		atmel_spi_pump_single_data(as, xfer);
1040 }
1041 
1042 /* Interrupt
1043  *
1044  * No need for locking in this Interrupt handler: done_status is the
1045  * only information modified.
1046  */
1047 static irqreturn_t
1048 atmel_spi_pio_interrupt(int irq, void *dev_id)
1049 {
1050 	struct spi_master	*master = dev_id;
1051 	struct atmel_spi	*as = spi_master_get_devdata(master);
1052 	u32			status, pending, imr;
1053 	struct spi_transfer	*xfer;
1054 	int			ret = IRQ_NONE;
1055 
1056 	imr = spi_readl(as, IMR);
1057 	status = spi_readl(as, SR);
1058 	pending = status & imr;
1059 
1060 	if (pending & SPI_BIT(OVRES)) {
1061 		ret = IRQ_HANDLED;
1062 		spi_writel(as, IDR, SPI_BIT(OVRES));
1063 		dev_warn(master->dev.parent, "overrun\n");
1064 
1065 		/*
1066 		 * When we get an overrun, we disregard the current
1067 		 * transfer. Data will not be copied back from any
1068 		 * bounce buffer and msg->actual_len will not be
1069 		 * updated with the last xfer.
1070 		 *
1071 		 * We will also not process any remaning transfers in
1072 		 * the message.
1073 		 */
1074 		as->done_status = -EIO;
1075 		smp_wmb();
1076 
1077 		/* Clear any overrun happening while cleaning up */
1078 		spi_readl(as, SR);
1079 
1080 		complete(&as->xfer_completion);
1081 
1082 	} else if (pending & (SPI_BIT(RDRF) | SPI_BIT(RXFTHF))) {
1083 		atmel_spi_lock(as);
1084 
1085 		if (as->current_remaining_bytes) {
1086 			ret = IRQ_HANDLED;
1087 			xfer = as->current_transfer;
1088 			atmel_spi_pump_pio_data(as, xfer);
1089 			if (!as->current_remaining_bytes)
1090 				spi_writel(as, IDR, pending);
1091 
1092 			complete(&as->xfer_completion);
1093 		}
1094 
1095 		atmel_spi_unlock(as);
1096 	} else {
1097 		WARN_ONCE(pending, "IRQ not handled, pending = %x\n", pending);
1098 		ret = IRQ_HANDLED;
1099 		spi_writel(as, IDR, pending);
1100 	}
1101 
1102 	return ret;
1103 }
1104 
1105 static irqreturn_t
1106 atmel_spi_pdc_interrupt(int irq, void *dev_id)
1107 {
1108 	struct spi_master	*master = dev_id;
1109 	struct atmel_spi	*as = spi_master_get_devdata(master);
1110 	u32			status, pending, imr;
1111 	int			ret = IRQ_NONE;
1112 
1113 	imr = spi_readl(as, IMR);
1114 	status = spi_readl(as, SR);
1115 	pending = status & imr;
1116 
1117 	if (pending & SPI_BIT(OVRES)) {
1118 
1119 		ret = IRQ_HANDLED;
1120 
1121 		spi_writel(as, IDR, (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX)
1122 				     | SPI_BIT(OVRES)));
1123 
1124 		/* Clear any overrun happening while cleaning up */
1125 		spi_readl(as, SR);
1126 
1127 		as->done_status = -EIO;
1128 
1129 		complete(&as->xfer_completion);
1130 
1131 	} else if (pending & (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX))) {
1132 		ret = IRQ_HANDLED;
1133 
1134 		spi_writel(as, IDR, pending);
1135 
1136 		complete(&as->xfer_completion);
1137 	}
1138 
1139 	return ret;
1140 }
1141 
1142 static int atmel_spi_setup(struct spi_device *spi)
1143 {
1144 	struct atmel_spi	*as;
1145 	struct atmel_spi_device	*asd;
1146 	u32			csr;
1147 	unsigned int		bits = spi->bits_per_word;
1148 	unsigned int		npcs_pin;
1149 
1150 	as = spi_master_get_devdata(spi->master);
1151 
1152 	/* see notes above re chipselect */
1153 	if (!atmel_spi_is_v2(as)
1154 			&& spi->chip_select == 0
1155 			&& (spi->mode & SPI_CS_HIGH)) {
1156 		dev_dbg(&spi->dev, "setup: can't be active-high\n");
1157 		return -EINVAL;
1158 	}
1159 
1160 	csr = SPI_BF(BITS, bits - 8);
1161 	if (spi->mode & SPI_CPOL)
1162 		csr |= SPI_BIT(CPOL);
1163 	if (!(spi->mode & SPI_CPHA))
1164 		csr |= SPI_BIT(NCPHA);
1165 	if (!as->use_cs_gpios)
1166 		csr |= SPI_BIT(CSAAT);
1167 
1168 	/* DLYBS is mostly irrelevant since we manage chipselect using GPIOs.
1169 	 *
1170 	 * DLYBCT would add delays between words, slowing down transfers.
1171 	 * It could potentially be useful to cope with DMA bottlenecks, but
1172 	 * in those cases it's probably best to just use a lower bitrate.
1173 	 */
1174 	csr |= SPI_BF(DLYBS, 0);
1175 	csr |= SPI_BF(DLYBCT, 0);
1176 
1177 	/* chipselect must have been muxed as GPIO (e.g. in board setup) */
1178 	npcs_pin = (unsigned long)spi->controller_data;
1179 
1180 	if (!as->use_cs_gpios)
1181 		npcs_pin = spi->chip_select;
1182 	else if (gpio_is_valid(spi->cs_gpio))
1183 		npcs_pin = spi->cs_gpio;
1184 
1185 	asd = spi->controller_state;
1186 	if (!asd) {
1187 		asd = kzalloc(sizeof(struct atmel_spi_device), GFP_KERNEL);
1188 		if (!asd)
1189 			return -ENOMEM;
1190 
1191 		if (as->use_cs_gpios)
1192 			gpio_direction_output(npcs_pin,
1193 					      !(spi->mode & SPI_CS_HIGH));
1194 
1195 		asd->npcs_pin = npcs_pin;
1196 		spi->controller_state = asd;
1197 	}
1198 
1199 	asd->csr = csr;
1200 
1201 	dev_dbg(&spi->dev,
1202 		"setup: bpw %u mode 0x%x -> csr%d %08x\n",
1203 		bits, spi->mode, spi->chip_select, csr);
1204 
1205 	if (!atmel_spi_is_v2(as))
1206 		spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
1207 
1208 	return 0;
1209 }
1210 
1211 static int atmel_spi_one_transfer(struct spi_master *master,
1212 					struct spi_message *msg,
1213 					struct spi_transfer *xfer)
1214 {
1215 	struct atmel_spi	*as;
1216 	struct spi_device	*spi = msg->spi;
1217 	u8			bits;
1218 	u32			len;
1219 	struct atmel_spi_device	*asd;
1220 	int			timeout;
1221 	int			ret;
1222 	unsigned long		dma_timeout;
1223 
1224 	as = spi_master_get_devdata(master);
1225 
1226 	if (!(xfer->tx_buf || xfer->rx_buf) && xfer->len) {
1227 		dev_dbg(&spi->dev, "missing rx or tx buf\n");
1228 		return -EINVAL;
1229 	}
1230 
1231 	asd = spi->controller_state;
1232 	bits = (asd->csr >> 4) & 0xf;
1233 	if (bits != xfer->bits_per_word - 8) {
1234 		dev_dbg(&spi->dev,
1235 			"you can't yet change bits_per_word in transfers\n");
1236 		return -ENOPROTOOPT;
1237 	}
1238 
1239 	/*
1240 	 * DMA map early, for performance (empties dcache ASAP) and
1241 	 * better fault reporting.
1242 	 */
1243 	if ((!msg->is_dma_mapped)
1244 		&& as->use_pdc) {
1245 		if (atmel_spi_dma_map_xfer(as, xfer) < 0)
1246 			return -ENOMEM;
1247 	}
1248 
1249 	atmel_spi_set_xfer_speed(as, msg->spi, xfer);
1250 
1251 	as->done_status = 0;
1252 	as->current_transfer = xfer;
1253 	as->current_remaining_bytes = xfer->len;
1254 	while (as->current_remaining_bytes) {
1255 		reinit_completion(&as->xfer_completion);
1256 
1257 		if (as->use_pdc) {
1258 			atmel_spi_pdc_next_xfer(master, msg, xfer);
1259 		} else if (atmel_spi_use_dma(as, xfer)) {
1260 			len = as->current_remaining_bytes;
1261 			ret = atmel_spi_next_xfer_dma_submit(master,
1262 								xfer, &len);
1263 			if (ret) {
1264 				dev_err(&spi->dev,
1265 					"unable to use DMA, fallback to PIO\n");
1266 				atmel_spi_next_xfer_pio(master, xfer);
1267 			} else {
1268 				as->current_remaining_bytes -= len;
1269 				if (as->current_remaining_bytes < 0)
1270 					as->current_remaining_bytes = 0;
1271 			}
1272 		} else {
1273 			atmel_spi_next_xfer_pio(master, xfer);
1274 		}
1275 
1276 		/* interrupts are disabled, so free the lock for schedule */
1277 		atmel_spi_unlock(as);
1278 		dma_timeout = wait_for_completion_timeout(&as->xfer_completion,
1279 							  SPI_DMA_TIMEOUT);
1280 		atmel_spi_lock(as);
1281 		if (WARN_ON(dma_timeout == 0)) {
1282 			dev_err(&spi->dev, "spi transfer timeout\n");
1283 			as->done_status = -EIO;
1284 		}
1285 
1286 		if (as->done_status)
1287 			break;
1288 	}
1289 
1290 	if (as->done_status) {
1291 		if (as->use_pdc) {
1292 			dev_warn(master->dev.parent,
1293 				"overrun (%u/%u remaining)\n",
1294 				spi_readl(as, TCR), spi_readl(as, RCR));
1295 
1296 			/*
1297 			 * Clean up DMA registers and make sure the data
1298 			 * registers are empty.
1299 			 */
1300 			spi_writel(as, RNCR, 0);
1301 			spi_writel(as, TNCR, 0);
1302 			spi_writel(as, RCR, 0);
1303 			spi_writel(as, TCR, 0);
1304 			for (timeout = 1000; timeout; timeout--)
1305 				if (spi_readl(as, SR) & SPI_BIT(TXEMPTY))
1306 					break;
1307 			if (!timeout)
1308 				dev_warn(master->dev.parent,
1309 					 "timeout waiting for TXEMPTY");
1310 			while (spi_readl(as, SR) & SPI_BIT(RDRF))
1311 				spi_readl(as, RDR);
1312 
1313 			/* Clear any overrun happening while cleaning up */
1314 			spi_readl(as, SR);
1315 
1316 		} else if (atmel_spi_use_dma(as, xfer)) {
1317 			atmel_spi_stop_dma(master);
1318 		}
1319 
1320 		if (!msg->is_dma_mapped
1321 			&& as->use_pdc)
1322 			atmel_spi_dma_unmap_xfer(master, xfer);
1323 
1324 		return 0;
1325 
1326 	} else {
1327 		/* only update length if no error */
1328 		msg->actual_length += xfer->len;
1329 	}
1330 
1331 	if (!msg->is_dma_mapped
1332 		&& as->use_pdc)
1333 		atmel_spi_dma_unmap_xfer(master, xfer);
1334 
1335 	if (xfer->delay_usecs)
1336 		udelay(xfer->delay_usecs);
1337 
1338 	if (xfer->cs_change) {
1339 		if (list_is_last(&xfer->transfer_list,
1340 				 &msg->transfers)) {
1341 			as->keep_cs = true;
1342 		} else {
1343 			as->cs_active = !as->cs_active;
1344 			if (as->cs_active)
1345 				cs_activate(as, msg->spi);
1346 			else
1347 				cs_deactivate(as, msg->spi);
1348 		}
1349 	}
1350 
1351 	return 0;
1352 }
1353 
1354 static int atmel_spi_transfer_one_message(struct spi_master *master,
1355 						struct spi_message *msg)
1356 {
1357 	struct atmel_spi *as;
1358 	struct spi_transfer *xfer;
1359 	struct spi_device *spi = msg->spi;
1360 	int ret = 0;
1361 
1362 	as = spi_master_get_devdata(master);
1363 
1364 	dev_dbg(&spi->dev, "new message %p submitted for %s\n",
1365 					msg, dev_name(&spi->dev));
1366 
1367 	atmel_spi_lock(as);
1368 	cs_activate(as, spi);
1369 
1370 	as->cs_active = true;
1371 	as->keep_cs = false;
1372 
1373 	msg->status = 0;
1374 	msg->actual_length = 0;
1375 
1376 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1377 		ret = atmel_spi_one_transfer(master, msg, xfer);
1378 		if (ret)
1379 			goto msg_done;
1380 	}
1381 
1382 	if (as->use_pdc)
1383 		atmel_spi_disable_pdc_transfer(as);
1384 
1385 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1386 		dev_dbg(&spi->dev,
1387 			"  xfer %p: len %u tx %p/%pad rx %p/%pad\n",
1388 			xfer, xfer->len,
1389 			xfer->tx_buf, &xfer->tx_dma,
1390 			xfer->rx_buf, &xfer->rx_dma);
1391 	}
1392 
1393 msg_done:
1394 	if (!as->keep_cs)
1395 		cs_deactivate(as, msg->spi);
1396 
1397 	atmel_spi_unlock(as);
1398 
1399 	msg->status = as->done_status;
1400 	spi_finalize_current_message(spi->master);
1401 
1402 	return ret;
1403 }
1404 
1405 static void atmel_spi_cleanup(struct spi_device *spi)
1406 {
1407 	struct atmel_spi_device	*asd = spi->controller_state;
1408 
1409 	if (!asd)
1410 		return;
1411 
1412 	spi->controller_state = NULL;
1413 	kfree(asd);
1414 }
1415 
1416 static inline unsigned int atmel_get_version(struct atmel_spi *as)
1417 {
1418 	return spi_readl(as, VERSION) & 0x00000fff;
1419 }
1420 
1421 static void atmel_get_caps(struct atmel_spi *as)
1422 {
1423 	unsigned int version;
1424 
1425 	version = atmel_get_version(as);
1426 
1427 	as->caps.is_spi2 = version > 0x121;
1428 	as->caps.has_wdrbt = version >= 0x210;
1429 #ifdef CONFIG_SOC_SAM_V4_V5
1430 	/*
1431 	 * Atmel SoCs based on ARM9 (SAM9x) cores should not use spi_map_buf()
1432 	 * since this later function tries to map buffers with dma_map_sg()
1433 	 * even if they have not been allocated inside DMA-safe areas.
1434 	 * On SoCs based on Cortex A5 (SAMA5Dx), it works anyway because for
1435 	 * those ARM cores, the data cache follows the PIPT model.
1436 	 * Also the L2 cache controller of SAMA5D2 uses the PIPT model too.
1437 	 * In case of PIPT caches, there cannot be cache aliases.
1438 	 * However on ARM9 cores, the data cache follows the VIVT model, hence
1439 	 * the cache aliases issue can occur when buffers are allocated from
1440 	 * DMA-unsafe areas, by vmalloc() for instance, where cache coherency is
1441 	 * not taken into account or at least not handled completely (cache
1442 	 * lines of aliases are not invalidated).
1443 	 * This is not a theorical issue: it was reproduced when trying to mount
1444 	 * a UBI file-system on a at91sam9g35ek board.
1445 	 */
1446 	as->caps.has_dma_support = false;
1447 #else
1448 	as->caps.has_dma_support = version >= 0x212;
1449 #endif
1450 	as->caps.has_pdc_support = version < 0x212;
1451 }
1452 
1453 /*-------------------------------------------------------------------------*/
1454 static int atmel_spi_gpio_cs(struct platform_device *pdev)
1455 {
1456 	struct spi_master	*master = platform_get_drvdata(pdev);
1457 	struct atmel_spi	*as = spi_master_get_devdata(master);
1458 	struct device_node	*np = master->dev.of_node;
1459 	int			i;
1460 	int			ret = 0;
1461 	int			nb = 0;
1462 
1463 	if (!as->use_cs_gpios)
1464 		return 0;
1465 
1466 	if (!np)
1467 		return 0;
1468 
1469 	nb = of_gpio_named_count(np, "cs-gpios");
1470 	for (i = 0; i < nb; i++) {
1471 		int cs_gpio = of_get_named_gpio(pdev->dev.of_node,
1472 						"cs-gpios", i);
1473 
1474 		if (cs_gpio == -EPROBE_DEFER)
1475 			return cs_gpio;
1476 
1477 		if (gpio_is_valid(cs_gpio)) {
1478 			ret = devm_gpio_request(&pdev->dev, cs_gpio,
1479 						dev_name(&pdev->dev));
1480 			if (ret)
1481 				return ret;
1482 		}
1483 	}
1484 
1485 	return 0;
1486 }
1487 
1488 static void atmel_spi_init(struct atmel_spi *as)
1489 {
1490 	spi_writel(as, CR, SPI_BIT(SWRST));
1491 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1492 	if (as->caps.has_wdrbt) {
1493 		spi_writel(as, MR, SPI_BIT(WDRBT) | SPI_BIT(MODFDIS)
1494 				| SPI_BIT(MSTR));
1495 	} else {
1496 		spi_writel(as, MR, SPI_BIT(MSTR) | SPI_BIT(MODFDIS));
1497 	}
1498 
1499 	if (as->use_pdc)
1500 		spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1501 	spi_writel(as, CR, SPI_BIT(SPIEN));
1502 
1503 	if (as->fifo_size)
1504 		spi_writel(as, CR, SPI_BIT(FIFOEN));
1505 }
1506 
1507 static int atmel_spi_probe(struct platform_device *pdev)
1508 {
1509 	struct resource		*regs;
1510 	int			irq;
1511 	struct clk		*clk;
1512 	int			ret;
1513 	struct spi_master	*master;
1514 	struct atmel_spi	*as;
1515 
1516 	/* Select default pin state */
1517 	pinctrl_pm_select_default_state(&pdev->dev);
1518 
1519 	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1520 	if (!regs)
1521 		return -ENXIO;
1522 
1523 	irq = platform_get_irq(pdev, 0);
1524 	if (irq < 0)
1525 		return irq;
1526 
1527 	clk = devm_clk_get(&pdev->dev, "spi_clk");
1528 	if (IS_ERR(clk))
1529 		return PTR_ERR(clk);
1530 
1531 	/* setup spi core then atmel-specific driver state */
1532 	ret = -ENOMEM;
1533 	master = spi_alloc_master(&pdev->dev, sizeof(*as));
1534 	if (!master)
1535 		goto out_free;
1536 
1537 	/* the spi->mode bits understood by this driver: */
1538 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
1539 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(8, 16);
1540 	master->dev.of_node = pdev->dev.of_node;
1541 	master->bus_num = pdev->id;
1542 	master->num_chipselect = master->dev.of_node ? 0 : 4;
1543 	master->setup = atmel_spi_setup;
1544 	master->flags = (SPI_MASTER_MUST_RX | SPI_MASTER_MUST_TX);
1545 	master->transfer_one_message = atmel_spi_transfer_one_message;
1546 	master->cleanup = atmel_spi_cleanup;
1547 	master->auto_runtime_pm = true;
1548 	master->max_dma_len = SPI_MAX_DMA_XFER;
1549 	master->can_dma = atmel_spi_can_dma;
1550 	platform_set_drvdata(pdev, master);
1551 
1552 	as = spi_master_get_devdata(master);
1553 
1554 	spin_lock_init(&as->lock);
1555 
1556 	as->pdev = pdev;
1557 	as->regs = devm_ioremap_resource(&pdev->dev, regs);
1558 	if (IS_ERR(as->regs)) {
1559 		ret = PTR_ERR(as->regs);
1560 		goto out_unmap_regs;
1561 	}
1562 	as->phybase = regs->start;
1563 	as->irq = irq;
1564 	as->clk = clk;
1565 
1566 	init_completion(&as->xfer_completion);
1567 
1568 	atmel_get_caps(as);
1569 
1570 	as->use_cs_gpios = true;
1571 	if (atmel_spi_is_v2(as) &&
1572 	    pdev->dev.of_node &&
1573 	    !of_get_property(pdev->dev.of_node, "cs-gpios", NULL)) {
1574 		as->use_cs_gpios = false;
1575 		master->num_chipselect = 4;
1576 	}
1577 
1578 	ret = atmel_spi_gpio_cs(pdev);
1579 	if (ret)
1580 		goto out_unmap_regs;
1581 
1582 	as->use_dma = false;
1583 	as->use_pdc = false;
1584 	if (as->caps.has_dma_support) {
1585 		ret = atmel_spi_configure_dma(master, as);
1586 		if (ret == 0) {
1587 			as->use_dma = true;
1588 		} else if (ret == -EPROBE_DEFER) {
1589 			return ret;
1590 		}
1591 	} else if (as->caps.has_pdc_support) {
1592 		as->use_pdc = true;
1593 	}
1594 
1595 	if (as->caps.has_dma_support && !as->use_dma)
1596 		dev_info(&pdev->dev, "Atmel SPI Controller using PIO only\n");
1597 
1598 	if (as->use_pdc) {
1599 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pdc_interrupt,
1600 					0, dev_name(&pdev->dev), master);
1601 	} else {
1602 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pio_interrupt,
1603 					0, dev_name(&pdev->dev), master);
1604 	}
1605 	if (ret)
1606 		goto out_unmap_regs;
1607 
1608 	/* Initialize the hardware */
1609 	ret = clk_prepare_enable(clk);
1610 	if (ret)
1611 		goto out_free_irq;
1612 
1613 	as->spi_clk = clk_get_rate(clk);
1614 
1615 	as->fifo_size = 0;
1616 	if (!of_property_read_u32(pdev->dev.of_node, "atmel,fifo-size",
1617 				  &as->fifo_size)) {
1618 		dev_info(&pdev->dev, "Using FIFO (%u data)\n", as->fifo_size);
1619 	}
1620 
1621 	atmel_spi_init(as);
1622 
1623 	pm_runtime_set_autosuspend_delay(&pdev->dev, AUTOSUSPEND_TIMEOUT);
1624 	pm_runtime_use_autosuspend(&pdev->dev);
1625 	pm_runtime_set_active(&pdev->dev);
1626 	pm_runtime_enable(&pdev->dev);
1627 
1628 	ret = devm_spi_register_master(&pdev->dev, master);
1629 	if (ret)
1630 		goto out_free_dma;
1631 
1632 	/* go! */
1633 	dev_info(&pdev->dev, "Atmel SPI Controller version 0x%x at 0x%08lx (irq %d)\n",
1634 			atmel_get_version(as), (unsigned long)regs->start,
1635 			irq);
1636 
1637 	return 0;
1638 
1639 out_free_dma:
1640 	pm_runtime_disable(&pdev->dev);
1641 	pm_runtime_set_suspended(&pdev->dev);
1642 
1643 	if (as->use_dma)
1644 		atmel_spi_release_dma(master);
1645 
1646 	spi_writel(as, CR, SPI_BIT(SWRST));
1647 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1648 	clk_disable_unprepare(clk);
1649 out_free_irq:
1650 out_unmap_regs:
1651 out_free:
1652 	spi_master_put(master);
1653 	return ret;
1654 }
1655 
1656 static int atmel_spi_remove(struct platform_device *pdev)
1657 {
1658 	struct spi_master	*master = platform_get_drvdata(pdev);
1659 	struct atmel_spi	*as = spi_master_get_devdata(master);
1660 
1661 	pm_runtime_get_sync(&pdev->dev);
1662 
1663 	/* reset the hardware and block queue progress */
1664 	spin_lock_irq(&as->lock);
1665 	if (as->use_dma) {
1666 		atmel_spi_stop_dma(master);
1667 		atmel_spi_release_dma(master);
1668 	}
1669 
1670 	spi_writel(as, CR, SPI_BIT(SWRST));
1671 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1672 	spi_readl(as, SR);
1673 	spin_unlock_irq(&as->lock);
1674 
1675 	clk_disable_unprepare(as->clk);
1676 
1677 	pm_runtime_put_noidle(&pdev->dev);
1678 	pm_runtime_disable(&pdev->dev);
1679 
1680 	return 0;
1681 }
1682 
1683 #ifdef CONFIG_PM
1684 static int atmel_spi_runtime_suspend(struct device *dev)
1685 {
1686 	struct spi_master *master = dev_get_drvdata(dev);
1687 	struct atmel_spi *as = spi_master_get_devdata(master);
1688 
1689 	clk_disable_unprepare(as->clk);
1690 	pinctrl_pm_select_sleep_state(dev);
1691 
1692 	return 0;
1693 }
1694 
1695 static int atmel_spi_runtime_resume(struct device *dev)
1696 {
1697 	struct spi_master *master = dev_get_drvdata(dev);
1698 	struct atmel_spi *as = spi_master_get_devdata(master);
1699 
1700 	pinctrl_pm_select_default_state(dev);
1701 
1702 	return clk_prepare_enable(as->clk);
1703 }
1704 
1705 #ifdef CONFIG_PM_SLEEP
1706 static int atmel_spi_suspend(struct device *dev)
1707 {
1708 	struct spi_master *master = dev_get_drvdata(dev);
1709 	int ret;
1710 
1711 	/* Stop the queue running */
1712 	ret = spi_master_suspend(master);
1713 	if (ret) {
1714 		dev_warn(dev, "cannot suspend master\n");
1715 		return ret;
1716 	}
1717 
1718 	if (!pm_runtime_suspended(dev))
1719 		atmel_spi_runtime_suspend(dev);
1720 
1721 	return 0;
1722 }
1723 
1724 static int atmel_spi_resume(struct device *dev)
1725 {
1726 	struct spi_master *master = dev_get_drvdata(dev);
1727 	struct atmel_spi *as = spi_master_get_devdata(master);
1728 	int ret;
1729 
1730 	ret = clk_prepare_enable(as->clk);
1731 	if (ret)
1732 		return ret;
1733 
1734 	atmel_spi_init(as);
1735 
1736 	clk_disable_unprepare(as->clk);
1737 
1738 	if (!pm_runtime_suspended(dev)) {
1739 		ret = atmel_spi_runtime_resume(dev);
1740 		if (ret)
1741 			return ret;
1742 	}
1743 
1744 	/* Start the queue running */
1745 	ret = spi_master_resume(master);
1746 	if (ret)
1747 		dev_err(dev, "problem starting queue (%d)\n", ret);
1748 
1749 	return ret;
1750 }
1751 #endif
1752 
1753 static const struct dev_pm_ops atmel_spi_pm_ops = {
1754 	SET_SYSTEM_SLEEP_PM_OPS(atmel_spi_suspend, atmel_spi_resume)
1755 	SET_RUNTIME_PM_OPS(atmel_spi_runtime_suspend,
1756 			   atmel_spi_runtime_resume, NULL)
1757 };
1758 #define ATMEL_SPI_PM_OPS	(&atmel_spi_pm_ops)
1759 #else
1760 #define ATMEL_SPI_PM_OPS	NULL
1761 #endif
1762 
1763 #if defined(CONFIG_OF)
1764 static const struct of_device_id atmel_spi_dt_ids[] = {
1765 	{ .compatible = "atmel,at91rm9200-spi" },
1766 	{ /* sentinel */ }
1767 };
1768 
1769 MODULE_DEVICE_TABLE(of, atmel_spi_dt_ids);
1770 #endif
1771 
1772 static struct platform_driver atmel_spi_driver = {
1773 	.driver		= {
1774 		.name	= "atmel_spi",
1775 		.pm	= ATMEL_SPI_PM_OPS,
1776 		.of_match_table	= of_match_ptr(atmel_spi_dt_ids),
1777 	},
1778 	.probe		= atmel_spi_probe,
1779 	.remove		= atmel_spi_remove,
1780 };
1781 module_platform_driver(atmel_spi_driver);
1782 
1783 MODULE_DESCRIPTION("Atmel AT32/AT91 SPI Controller driver");
1784 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1785 MODULE_LICENSE("GPL");
1786 MODULE_ALIAS("platform:atmel_spi");
1787