xref: /openbmc/linux/drivers/spi/spi-atmel.c (revision b830f94f)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Driver for Atmel AT32 and AT91 SPI Controllers
4  *
5  * Copyright (C) 2006 Atmel Corporation
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/clk.h>
10 #include <linux/module.h>
11 #include <linux/platform_device.h>
12 #include <linux/delay.h>
13 #include <linux/dma-mapping.h>
14 #include <linux/dmaengine.h>
15 #include <linux/err.h>
16 #include <linux/interrupt.h>
17 #include <linux/spi/spi.h>
18 #include <linux/slab.h>
19 #include <linux/platform_data/dma-atmel.h>
20 #include <linux/of.h>
21 
22 #include <linux/io.h>
23 #include <linux/gpio/consumer.h>
24 #include <linux/pinctrl/consumer.h>
25 #include <linux/pm_runtime.h>
26 
27 /* SPI register offsets */
28 #define SPI_CR					0x0000
29 #define SPI_MR					0x0004
30 #define SPI_RDR					0x0008
31 #define SPI_TDR					0x000c
32 #define SPI_SR					0x0010
33 #define SPI_IER					0x0014
34 #define SPI_IDR					0x0018
35 #define SPI_IMR					0x001c
36 #define SPI_CSR0				0x0030
37 #define SPI_CSR1				0x0034
38 #define SPI_CSR2				0x0038
39 #define SPI_CSR3				0x003c
40 #define SPI_FMR					0x0040
41 #define SPI_FLR					0x0044
42 #define SPI_VERSION				0x00fc
43 #define SPI_RPR					0x0100
44 #define SPI_RCR					0x0104
45 #define SPI_TPR					0x0108
46 #define SPI_TCR					0x010c
47 #define SPI_RNPR				0x0110
48 #define SPI_RNCR				0x0114
49 #define SPI_TNPR				0x0118
50 #define SPI_TNCR				0x011c
51 #define SPI_PTCR				0x0120
52 #define SPI_PTSR				0x0124
53 
54 /* Bitfields in CR */
55 #define SPI_SPIEN_OFFSET			0
56 #define SPI_SPIEN_SIZE				1
57 #define SPI_SPIDIS_OFFSET			1
58 #define SPI_SPIDIS_SIZE				1
59 #define SPI_SWRST_OFFSET			7
60 #define SPI_SWRST_SIZE				1
61 #define SPI_LASTXFER_OFFSET			24
62 #define SPI_LASTXFER_SIZE			1
63 #define SPI_TXFCLR_OFFSET			16
64 #define SPI_TXFCLR_SIZE				1
65 #define SPI_RXFCLR_OFFSET			17
66 #define SPI_RXFCLR_SIZE				1
67 #define SPI_FIFOEN_OFFSET			30
68 #define SPI_FIFOEN_SIZE				1
69 #define SPI_FIFODIS_OFFSET			31
70 #define SPI_FIFODIS_SIZE			1
71 
72 /* Bitfields in MR */
73 #define SPI_MSTR_OFFSET				0
74 #define SPI_MSTR_SIZE				1
75 #define SPI_PS_OFFSET				1
76 #define SPI_PS_SIZE				1
77 #define SPI_PCSDEC_OFFSET			2
78 #define SPI_PCSDEC_SIZE				1
79 #define SPI_FDIV_OFFSET				3
80 #define SPI_FDIV_SIZE				1
81 #define SPI_MODFDIS_OFFSET			4
82 #define SPI_MODFDIS_SIZE			1
83 #define SPI_WDRBT_OFFSET			5
84 #define SPI_WDRBT_SIZE				1
85 #define SPI_LLB_OFFSET				7
86 #define SPI_LLB_SIZE				1
87 #define SPI_PCS_OFFSET				16
88 #define SPI_PCS_SIZE				4
89 #define SPI_DLYBCS_OFFSET			24
90 #define SPI_DLYBCS_SIZE				8
91 
92 /* Bitfields in RDR */
93 #define SPI_RD_OFFSET				0
94 #define SPI_RD_SIZE				16
95 
96 /* Bitfields in TDR */
97 #define SPI_TD_OFFSET				0
98 #define SPI_TD_SIZE				16
99 
100 /* Bitfields in SR */
101 #define SPI_RDRF_OFFSET				0
102 #define SPI_RDRF_SIZE				1
103 #define SPI_TDRE_OFFSET				1
104 #define SPI_TDRE_SIZE				1
105 #define SPI_MODF_OFFSET				2
106 #define SPI_MODF_SIZE				1
107 #define SPI_OVRES_OFFSET			3
108 #define SPI_OVRES_SIZE				1
109 #define SPI_ENDRX_OFFSET			4
110 #define SPI_ENDRX_SIZE				1
111 #define SPI_ENDTX_OFFSET			5
112 #define SPI_ENDTX_SIZE				1
113 #define SPI_RXBUFF_OFFSET			6
114 #define SPI_RXBUFF_SIZE				1
115 #define SPI_TXBUFE_OFFSET			7
116 #define SPI_TXBUFE_SIZE				1
117 #define SPI_NSSR_OFFSET				8
118 #define SPI_NSSR_SIZE				1
119 #define SPI_TXEMPTY_OFFSET			9
120 #define SPI_TXEMPTY_SIZE			1
121 #define SPI_SPIENS_OFFSET			16
122 #define SPI_SPIENS_SIZE				1
123 #define SPI_TXFEF_OFFSET			24
124 #define SPI_TXFEF_SIZE				1
125 #define SPI_TXFFF_OFFSET			25
126 #define SPI_TXFFF_SIZE				1
127 #define SPI_TXFTHF_OFFSET			26
128 #define SPI_TXFTHF_SIZE				1
129 #define SPI_RXFEF_OFFSET			27
130 #define SPI_RXFEF_SIZE				1
131 #define SPI_RXFFF_OFFSET			28
132 #define SPI_RXFFF_SIZE				1
133 #define SPI_RXFTHF_OFFSET			29
134 #define SPI_RXFTHF_SIZE				1
135 #define SPI_TXFPTEF_OFFSET			30
136 #define SPI_TXFPTEF_SIZE			1
137 #define SPI_RXFPTEF_OFFSET			31
138 #define SPI_RXFPTEF_SIZE			1
139 
140 /* Bitfields in CSR0 */
141 #define SPI_CPOL_OFFSET				0
142 #define SPI_CPOL_SIZE				1
143 #define SPI_NCPHA_OFFSET			1
144 #define SPI_NCPHA_SIZE				1
145 #define SPI_CSAAT_OFFSET			3
146 #define SPI_CSAAT_SIZE				1
147 #define SPI_BITS_OFFSET				4
148 #define SPI_BITS_SIZE				4
149 #define SPI_SCBR_OFFSET				8
150 #define SPI_SCBR_SIZE				8
151 #define SPI_DLYBS_OFFSET			16
152 #define SPI_DLYBS_SIZE				8
153 #define SPI_DLYBCT_OFFSET			24
154 #define SPI_DLYBCT_SIZE				8
155 
156 /* Bitfields in RCR */
157 #define SPI_RXCTR_OFFSET			0
158 #define SPI_RXCTR_SIZE				16
159 
160 /* Bitfields in TCR */
161 #define SPI_TXCTR_OFFSET			0
162 #define SPI_TXCTR_SIZE				16
163 
164 /* Bitfields in RNCR */
165 #define SPI_RXNCR_OFFSET			0
166 #define SPI_RXNCR_SIZE				16
167 
168 /* Bitfields in TNCR */
169 #define SPI_TXNCR_OFFSET			0
170 #define SPI_TXNCR_SIZE				16
171 
172 /* Bitfields in PTCR */
173 #define SPI_RXTEN_OFFSET			0
174 #define SPI_RXTEN_SIZE				1
175 #define SPI_RXTDIS_OFFSET			1
176 #define SPI_RXTDIS_SIZE				1
177 #define SPI_TXTEN_OFFSET			8
178 #define SPI_TXTEN_SIZE				1
179 #define SPI_TXTDIS_OFFSET			9
180 #define SPI_TXTDIS_SIZE				1
181 
182 /* Bitfields in FMR */
183 #define SPI_TXRDYM_OFFSET			0
184 #define SPI_TXRDYM_SIZE				2
185 #define SPI_RXRDYM_OFFSET			4
186 #define SPI_RXRDYM_SIZE				2
187 #define SPI_TXFTHRES_OFFSET			16
188 #define SPI_TXFTHRES_SIZE			6
189 #define SPI_RXFTHRES_OFFSET			24
190 #define SPI_RXFTHRES_SIZE			6
191 
192 /* Bitfields in FLR */
193 #define SPI_TXFL_OFFSET				0
194 #define SPI_TXFL_SIZE				6
195 #define SPI_RXFL_OFFSET				16
196 #define SPI_RXFL_SIZE				6
197 
198 /* Constants for BITS */
199 #define SPI_BITS_8_BPT				0
200 #define SPI_BITS_9_BPT				1
201 #define SPI_BITS_10_BPT				2
202 #define SPI_BITS_11_BPT				3
203 #define SPI_BITS_12_BPT				4
204 #define SPI_BITS_13_BPT				5
205 #define SPI_BITS_14_BPT				6
206 #define SPI_BITS_15_BPT				7
207 #define SPI_BITS_16_BPT				8
208 #define SPI_ONE_DATA				0
209 #define SPI_TWO_DATA				1
210 #define SPI_FOUR_DATA				2
211 
212 /* Bit manipulation macros */
213 #define SPI_BIT(name) \
214 	(1 << SPI_##name##_OFFSET)
215 #define SPI_BF(name, value) \
216 	(((value) & ((1 << SPI_##name##_SIZE) - 1)) << SPI_##name##_OFFSET)
217 #define SPI_BFEXT(name, value) \
218 	(((value) >> SPI_##name##_OFFSET) & ((1 << SPI_##name##_SIZE) - 1))
219 #define SPI_BFINS(name, value, old) \
220 	(((old) & ~(((1 << SPI_##name##_SIZE) - 1) << SPI_##name##_OFFSET)) \
221 	  | SPI_BF(name, value))
222 
223 /* Register access macros */
224 #ifdef CONFIG_AVR32
225 #define spi_readl(port, reg) \
226 	__raw_readl((port)->regs + SPI_##reg)
227 #define spi_writel(port, reg, value) \
228 	__raw_writel((value), (port)->regs + SPI_##reg)
229 
230 #define spi_readw(port, reg) \
231 	__raw_readw((port)->regs + SPI_##reg)
232 #define spi_writew(port, reg, value) \
233 	__raw_writew((value), (port)->regs + SPI_##reg)
234 
235 #define spi_readb(port, reg) \
236 	__raw_readb((port)->regs + SPI_##reg)
237 #define spi_writeb(port, reg, value) \
238 	__raw_writeb((value), (port)->regs + SPI_##reg)
239 #else
240 #define spi_readl(port, reg) \
241 	readl_relaxed((port)->regs + SPI_##reg)
242 #define spi_writel(port, reg, value) \
243 	writel_relaxed((value), (port)->regs + SPI_##reg)
244 
245 #define spi_readw(port, reg) \
246 	readw_relaxed((port)->regs + SPI_##reg)
247 #define spi_writew(port, reg, value) \
248 	writew_relaxed((value), (port)->regs + SPI_##reg)
249 
250 #define spi_readb(port, reg) \
251 	readb_relaxed((port)->regs + SPI_##reg)
252 #define spi_writeb(port, reg, value) \
253 	writeb_relaxed((value), (port)->regs + SPI_##reg)
254 #endif
255 /* use PIO for small transfers, avoiding DMA setup/teardown overhead and
256  * cache operations; better heuristics consider wordsize and bitrate.
257  */
258 #define DMA_MIN_BYTES	16
259 
260 #define SPI_DMA_TIMEOUT		(msecs_to_jiffies(1000))
261 
262 #define AUTOSUSPEND_TIMEOUT	2000
263 
264 struct atmel_spi_caps {
265 	bool	is_spi2;
266 	bool	has_wdrbt;
267 	bool	has_dma_support;
268 	bool	has_pdc_support;
269 };
270 
271 /*
272  * The core SPI transfer engine just talks to a register bank to set up
273  * DMA transfers; transfer queue progress is driven by IRQs.  The clock
274  * framework provides the base clock, subdivided for each spi_device.
275  */
276 struct atmel_spi {
277 	spinlock_t		lock;
278 	unsigned long		flags;
279 
280 	phys_addr_t		phybase;
281 	void __iomem		*regs;
282 	int			irq;
283 	struct clk		*clk;
284 	struct platform_device	*pdev;
285 	unsigned long		spi_clk;
286 
287 	struct spi_transfer	*current_transfer;
288 	int			current_remaining_bytes;
289 	int			done_status;
290 	dma_addr_t		dma_addr_rx_bbuf;
291 	dma_addr_t		dma_addr_tx_bbuf;
292 	void			*addr_rx_bbuf;
293 	void			*addr_tx_bbuf;
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 	struct gpio_desc	*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 	u32 mr;
355 
356 	if (atmel_spi_is_v2(as)) {
357 		spi_writel(as, CSR0 + 4 * spi->chip_select, asd->csr);
358 		/* For the low SPI version, there is a issue that PDC transfer
359 		 * on CS1,2,3 needs SPI_CSR0.BITS config as SPI_CSR1,2,3.BITS
360 		 */
361 		spi_writel(as, CSR0, asd->csr);
362 		if (as->caps.has_wdrbt) {
363 			spi_writel(as, MR,
364 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
365 					| SPI_BIT(WDRBT)
366 					| SPI_BIT(MODFDIS)
367 					| SPI_BIT(MSTR));
368 		} else {
369 			spi_writel(as, MR,
370 					SPI_BF(PCS, ~(0x01 << spi->chip_select))
371 					| SPI_BIT(MODFDIS)
372 					| SPI_BIT(MSTR));
373 		}
374 
375 		mr = spi_readl(as, MR);
376 		if (as->use_cs_gpios)
377 			gpiod_set_value(asd->npcs_pin, 1);
378 	} else {
379 		u32 cpol = (spi->mode & SPI_CPOL) ? SPI_BIT(CPOL) : 0;
380 		int i;
381 		u32 csr;
382 
383 		/* Make sure clock polarity is correct */
384 		for (i = 0; i < spi->master->num_chipselect; i++) {
385 			csr = spi_readl(as, CSR0 + 4 * i);
386 			if ((csr ^ cpol) & SPI_BIT(CPOL))
387 				spi_writel(as, CSR0 + 4 * i,
388 						csr ^ SPI_BIT(CPOL));
389 		}
390 
391 		mr = spi_readl(as, MR);
392 		mr = SPI_BFINS(PCS, ~(1 << spi->chip_select), mr);
393 		if (as->use_cs_gpios && spi->chip_select != 0)
394 			gpiod_set_value(asd->npcs_pin, 1);
395 		spi_writel(as, MR, mr);
396 	}
397 
398 	dev_dbg(&spi->dev, "activate NPCS, mr %08x\n", mr);
399 }
400 
401 static void cs_deactivate(struct atmel_spi *as, struct spi_device *spi)
402 {
403 	struct atmel_spi_device *asd = spi->controller_state;
404 	u32 mr;
405 
406 	/* only deactivate *this* device; sometimes transfers to
407 	 * another device may be active when this routine is called.
408 	 */
409 	mr = spi_readl(as, MR);
410 	if (~SPI_BFEXT(PCS, mr) & (1 << spi->chip_select)) {
411 		mr = SPI_BFINS(PCS, 0xf, mr);
412 		spi_writel(as, MR, mr);
413 	}
414 
415 	dev_dbg(&spi->dev, "DEactivate NPCS, mr %08x\n", mr);
416 
417 	if (!as->use_cs_gpios)
418 		spi_writel(as, CR, SPI_BIT(LASTXFER));
419 	else if (atmel_spi_is_v2(as) || spi->chip_select != 0)
420 		gpiod_set_value(asd->npcs_pin, 0);
421 }
422 
423 static void atmel_spi_lock(struct atmel_spi *as) __acquires(&as->lock)
424 {
425 	spin_lock_irqsave(&as->lock, as->flags);
426 }
427 
428 static void atmel_spi_unlock(struct atmel_spi *as) __releases(&as->lock)
429 {
430 	spin_unlock_irqrestore(&as->lock, as->flags);
431 }
432 
433 static inline bool atmel_spi_is_vmalloc_xfer(struct spi_transfer *xfer)
434 {
435 	return is_vmalloc_addr(xfer->tx_buf) || is_vmalloc_addr(xfer->rx_buf);
436 }
437 
438 static inline bool atmel_spi_use_dma(struct atmel_spi *as,
439 				struct spi_transfer *xfer)
440 {
441 	return as->use_dma && xfer->len >= DMA_MIN_BYTES;
442 }
443 
444 static bool atmel_spi_can_dma(struct spi_master *master,
445 			      struct spi_device *spi,
446 			      struct spi_transfer *xfer)
447 {
448 	struct atmel_spi *as = spi_master_get_devdata(master);
449 
450 	if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5))
451 		return atmel_spi_use_dma(as, xfer) &&
452 			!atmel_spi_is_vmalloc_xfer(xfer);
453 	else
454 		return atmel_spi_use_dma(as, xfer);
455 
456 }
457 
458 static int atmel_spi_dma_slave_config(struct atmel_spi *as,
459 				struct dma_slave_config *slave_config,
460 				u8 bits_per_word)
461 {
462 	struct spi_master *master = platform_get_drvdata(as->pdev);
463 	int err = 0;
464 
465 	if (bits_per_word > 8) {
466 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
467 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
468 	} else {
469 		slave_config->dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
470 		slave_config->src_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
471 	}
472 
473 	slave_config->dst_addr = (dma_addr_t)as->phybase + SPI_TDR;
474 	slave_config->src_addr = (dma_addr_t)as->phybase + SPI_RDR;
475 	slave_config->src_maxburst = 1;
476 	slave_config->dst_maxburst = 1;
477 	slave_config->device_fc = false;
478 
479 	/*
480 	 * This driver uses fixed peripheral select mode (PS bit set to '0' in
481 	 * the Mode Register).
482 	 * So according to the datasheet, when FIFOs are available (and
483 	 * enabled), the Transmit FIFO operates in Multiple Data Mode.
484 	 * In this mode, up to 2 data, not 4, can be written into the Transmit
485 	 * Data Register in a single access.
486 	 * However, the first data has to be written into the lowest 16 bits and
487 	 * the second data into the highest 16 bits of the Transmit
488 	 * Data Register. For 8bit data (the most frequent case), it would
489 	 * require to rework tx_buf so each data would actualy fit 16 bits.
490 	 * So we'd rather write only one data at the time. Hence the transmit
491 	 * path works the same whether FIFOs are available (and enabled) or not.
492 	 */
493 	slave_config->direction = DMA_MEM_TO_DEV;
494 	if (dmaengine_slave_config(master->dma_tx, slave_config)) {
495 		dev_err(&as->pdev->dev,
496 			"failed to configure tx dma channel\n");
497 		err = -EINVAL;
498 	}
499 
500 	/*
501 	 * This driver configures the spi controller for master mode (MSTR bit
502 	 * set to '1' in the Mode Register).
503 	 * So according to the datasheet, when FIFOs are available (and
504 	 * enabled), the Receive FIFO operates in Single Data Mode.
505 	 * So the receive path works the same whether FIFOs are available (and
506 	 * enabled) or not.
507 	 */
508 	slave_config->direction = DMA_DEV_TO_MEM;
509 	if (dmaengine_slave_config(master->dma_rx, slave_config)) {
510 		dev_err(&as->pdev->dev,
511 			"failed to configure rx dma channel\n");
512 		err = -EINVAL;
513 	}
514 
515 	return err;
516 }
517 
518 static int atmel_spi_configure_dma(struct spi_master *master,
519 				   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 	master->dma_tx = dma_request_slave_channel_reason(dev, "tx");
530 	if (IS_ERR(master->dma_tx)) {
531 		err = PTR_ERR(master->dma_tx);
532 		if (err == -EPROBE_DEFER) {
533 			dev_warn(dev, "no DMA channel available at the moment\n");
534 			goto error_clear;
535 		}
536 		dev_err(dev,
537 			"DMA TX channel not available, SPI unable to use DMA\n");
538 		err = -EBUSY;
539 		goto error_clear;
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 	master->dma_rx = dma_request_slave_channel(dev, "rx");
547 
548 	if (!master->dma_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(master->dma_tx),
562 			dma_chan_name(master->dma_rx));
563 
564 	return 0;
565 error:
566 	if (master->dma_rx)
567 		dma_release_channel(master->dma_rx);
568 	if (!IS_ERR(master->dma_tx))
569 		dma_release_channel(master->dma_tx);
570 error_clear:
571 	master->dma_tx = master->dma_rx = NULL;
572 	return err;
573 }
574 
575 static void atmel_spi_stop_dma(struct spi_master *master)
576 {
577 	if (master->dma_rx)
578 		dmaengine_terminate_all(master->dma_rx);
579 	if (master->dma_tx)
580 		dmaengine_terminate_all(master->dma_tx);
581 }
582 
583 static void atmel_spi_release_dma(struct spi_master *master)
584 {
585 	if (master->dma_rx) {
586 		dma_release_channel(master->dma_rx);
587 		master->dma_rx = NULL;
588 	}
589 	if (master->dma_tx) {
590 		dma_release_channel(master->dma_tx);
591 		master->dma_tx = NULL;
592 	}
593 }
594 
595 /* This function is called by the DMA driver from tasklet context */
596 static void dma_callback(void *data)
597 {
598 	struct spi_master	*master = data;
599 	struct atmel_spi	*as = spi_master_get_devdata(master);
600 
601 	if (is_vmalloc_addr(as->current_transfer->rx_buf) &&
602 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
603 		memcpy(as->current_transfer->rx_buf, as->addr_rx_bbuf,
604 		       as->current_transfer->len);
605 	}
606 	complete(&as->xfer_completion);
607 }
608 
609 /*
610  * Next transfer using PIO without FIFO.
611  */
612 static void atmel_spi_next_xfer_single(struct spi_master *master,
613 				       struct spi_transfer *xfer)
614 {
615 	struct atmel_spi	*as = spi_master_get_devdata(master);
616 	unsigned long xfer_pos = xfer->len - as->current_remaining_bytes;
617 
618 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_pio\n");
619 
620 	/* Make sure data is not remaining in RDR */
621 	spi_readl(as, RDR);
622 	while (spi_readl(as, SR) & SPI_BIT(RDRF)) {
623 		spi_readl(as, RDR);
624 		cpu_relax();
625 	}
626 
627 	if (xfer->bits_per_word > 8)
628 		spi_writel(as, TDR, *(u16 *)(xfer->tx_buf + xfer_pos));
629 	else
630 		spi_writel(as, TDR, *(u8 *)(xfer->tx_buf + xfer_pos));
631 
632 	dev_dbg(master->dev.parent,
633 		"  start pio xfer %p: len %u tx %p rx %p bitpw %d\n",
634 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
635 		xfer->bits_per_word);
636 
637 	/* Enable relevant interrupts */
638 	spi_writel(as, IER, SPI_BIT(RDRF) | SPI_BIT(OVRES));
639 }
640 
641 /*
642  * Next transfer using PIO with FIFO.
643  */
644 static void atmel_spi_next_xfer_fifo(struct spi_master *master,
645 				     struct spi_transfer *xfer)
646 {
647 	struct atmel_spi *as = spi_master_get_devdata(master);
648 	u32 current_remaining_data, num_data;
649 	u32 offset = xfer->len - as->current_remaining_bytes;
650 	const u16 *words = (const u16 *)((u8 *)xfer->tx_buf + offset);
651 	const u8  *bytes = (const u8  *)((u8 *)xfer->tx_buf + offset);
652 	u16 td0, td1;
653 	u32 fifomr;
654 
655 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_fifo\n");
656 
657 	/* Compute the number of data to transfer in the current iteration */
658 	current_remaining_data = ((xfer->bits_per_word > 8) ?
659 				  ((u32)as->current_remaining_bytes >> 1) :
660 				  (u32)as->current_remaining_bytes);
661 	num_data = min(current_remaining_data, as->fifo_size);
662 
663 	/* Flush RX and TX FIFOs */
664 	spi_writel(as, CR, SPI_BIT(RXFCLR) | SPI_BIT(TXFCLR));
665 	while (spi_readl(as, FLR))
666 		cpu_relax();
667 
668 	/* Set RX FIFO Threshold to the number of data to transfer */
669 	fifomr = spi_readl(as, FMR);
670 	spi_writel(as, FMR, SPI_BFINS(RXFTHRES, num_data, fifomr));
671 
672 	/* Clear FIFO flags in the Status Register, especially RXFTHF */
673 	(void)spi_readl(as, SR);
674 
675 	/* Fill TX FIFO */
676 	while (num_data >= 2) {
677 		if (xfer->bits_per_word > 8) {
678 			td0 = *words++;
679 			td1 = *words++;
680 		} else {
681 			td0 = *bytes++;
682 			td1 = *bytes++;
683 		}
684 
685 		spi_writel(as, TDR, (td1 << 16) | td0);
686 		num_data -= 2;
687 	}
688 
689 	if (num_data) {
690 		if (xfer->bits_per_word > 8)
691 			td0 = *words++;
692 		else
693 			td0 = *bytes++;
694 
695 		spi_writew(as, TDR, td0);
696 		num_data--;
697 	}
698 
699 	dev_dbg(master->dev.parent,
700 		"  start fifo xfer %p: len %u tx %p rx %p bitpw %d\n",
701 		xfer, xfer->len, xfer->tx_buf, xfer->rx_buf,
702 		xfer->bits_per_word);
703 
704 	/*
705 	 * Enable RX FIFO Threshold Flag interrupt to be notified about
706 	 * transfer completion.
707 	 */
708 	spi_writel(as, IER, SPI_BIT(RXFTHF) | SPI_BIT(OVRES));
709 }
710 
711 /*
712  * Next transfer using PIO.
713  */
714 static void atmel_spi_next_xfer_pio(struct spi_master *master,
715 				    struct spi_transfer *xfer)
716 {
717 	struct atmel_spi *as = spi_master_get_devdata(master);
718 
719 	if (as->fifo_size)
720 		atmel_spi_next_xfer_fifo(master, xfer);
721 	else
722 		atmel_spi_next_xfer_single(master, xfer);
723 }
724 
725 /*
726  * Submit next transfer for DMA.
727  */
728 static int atmel_spi_next_xfer_dma_submit(struct spi_master *master,
729 				struct spi_transfer *xfer,
730 				u32 *plen)
731 {
732 	struct atmel_spi	*as = spi_master_get_devdata(master);
733 	struct dma_chan		*rxchan = master->dma_rx;
734 	struct dma_chan		*txchan = master->dma_tx;
735 	struct dma_async_tx_descriptor *rxdesc;
736 	struct dma_async_tx_descriptor *txdesc;
737 	struct dma_slave_config	slave_config;
738 	dma_cookie_t		cookie;
739 
740 	dev_vdbg(master->dev.parent, "atmel_spi_next_xfer_dma_submit\n");
741 
742 	/* Check that the channels are available */
743 	if (!rxchan || !txchan)
744 		return -ENODEV;
745 
746 	/* release lock for DMA operations */
747 	atmel_spi_unlock(as);
748 
749 	*plen = xfer->len;
750 
751 	if (atmel_spi_dma_slave_config(as, &slave_config,
752 				       xfer->bits_per_word))
753 		goto err_exit;
754 
755 	/* Send both scatterlists */
756 	if (atmel_spi_is_vmalloc_xfer(xfer) &&
757 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
758 		rxdesc = dmaengine_prep_slave_single(rxchan,
759 						     as->dma_addr_rx_bbuf,
760 						     xfer->len,
761 						     DMA_DEV_TO_MEM,
762 						     DMA_PREP_INTERRUPT |
763 						     DMA_CTRL_ACK);
764 	} else {
765 		rxdesc = dmaengine_prep_slave_sg(rxchan,
766 						 xfer->rx_sg.sgl,
767 						 xfer->rx_sg.nents,
768 						 DMA_DEV_TO_MEM,
769 						 DMA_PREP_INTERRUPT |
770 						 DMA_CTRL_ACK);
771 	}
772 	if (!rxdesc)
773 		goto err_dma;
774 
775 	if (atmel_spi_is_vmalloc_xfer(xfer) &&
776 	    IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
777 		memcpy(as->addr_tx_bbuf, xfer->tx_buf, xfer->len);
778 		txdesc = dmaengine_prep_slave_single(txchan,
779 						     as->dma_addr_tx_bbuf,
780 						     xfer->len, DMA_MEM_TO_DEV,
781 						     DMA_PREP_INTERRUPT |
782 						     DMA_CTRL_ACK);
783 	} else {
784 		txdesc = dmaengine_prep_slave_sg(txchan,
785 						 xfer->tx_sg.sgl,
786 						 xfer->tx_sg.nents,
787 						 DMA_MEM_TO_DEV,
788 						 DMA_PREP_INTERRUPT |
789 						 DMA_CTRL_ACK);
790 	}
791 	if (!txdesc)
792 		goto err_dma;
793 
794 	dev_dbg(master->dev.parent,
795 		"  start dma xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
796 		xfer, xfer->len, xfer->tx_buf, (unsigned long long)xfer->tx_dma,
797 		xfer->rx_buf, (unsigned long long)xfer->rx_dma);
798 
799 	/* Enable relevant interrupts */
800 	spi_writel(as, IER, SPI_BIT(OVRES));
801 
802 	/* Put the callback on the RX transfer only, that should finish last */
803 	rxdesc->callback = dma_callback;
804 	rxdesc->callback_param = master;
805 
806 	/* Submit and fire RX and TX with TX last so we're ready to read! */
807 	cookie = rxdesc->tx_submit(rxdesc);
808 	if (dma_submit_error(cookie))
809 		goto err_dma;
810 	cookie = txdesc->tx_submit(txdesc);
811 	if (dma_submit_error(cookie))
812 		goto err_dma;
813 	rxchan->device->device_issue_pending(rxchan);
814 	txchan->device->device_issue_pending(txchan);
815 
816 	/* take back lock */
817 	atmel_spi_lock(as);
818 	return 0;
819 
820 err_dma:
821 	spi_writel(as, IDR, SPI_BIT(OVRES));
822 	atmel_spi_stop_dma(master);
823 err_exit:
824 	atmel_spi_lock(as);
825 	return -ENOMEM;
826 }
827 
828 static void atmel_spi_next_xfer_data(struct spi_master *master,
829 				struct spi_transfer *xfer,
830 				dma_addr_t *tx_dma,
831 				dma_addr_t *rx_dma,
832 				u32 *plen)
833 {
834 	*rx_dma = xfer->rx_dma + xfer->len - *plen;
835 	*tx_dma = xfer->tx_dma + xfer->len - *plen;
836 	if (*plen > master->max_dma_len)
837 		*plen = master->max_dma_len;
838 }
839 
840 static int atmel_spi_set_xfer_speed(struct atmel_spi *as,
841 				    struct spi_device *spi,
842 				    struct spi_transfer *xfer)
843 {
844 	u32			scbr, csr;
845 	unsigned long		bus_hz;
846 
847 	/* v1 chips start out at half the peripheral bus speed. */
848 	bus_hz = as->spi_clk;
849 	if (!atmel_spi_is_v2(as))
850 		bus_hz /= 2;
851 
852 	/*
853 	 * Calculate the lowest divider that satisfies the
854 	 * constraint, assuming div32/fdiv/mbz == 0.
855 	 */
856 	scbr = DIV_ROUND_UP(bus_hz, xfer->speed_hz);
857 
858 	/*
859 	 * If the resulting divider doesn't fit into the
860 	 * register bitfield, we can't satisfy the constraint.
861 	 */
862 	if (scbr >= (1 << SPI_SCBR_SIZE)) {
863 		dev_err(&spi->dev,
864 			"setup: %d Hz too slow, scbr %u; min %ld Hz\n",
865 			xfer->speed_hz, scbr, bus_hz/255);
866 		return -EINVAL;
867 	}
868 	if (scbr == 0) {
869 		dev_err(&spi->dev,
870 			"setup: %d Hz too high, scbr %u; max %ld Hz\n",
871 			xfer->speed_hz, scbr, bus_hz);
872 		return -EINVAL;
873 	}
874 	csr = spi_readl(as, CSR0 + 4 * spi->chip_select);
875 	csr = SPI_BFINS(SCBR, scbr, csr);
876 	spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
877 
878 	return 0;
879 }
880 
881 /*
882  * Submit next transfer for PDC.
883  * lock is held, spi irq is blocked
884  */
885 static void atmel_spi_pdc_next_xfer(struct spi_master *master,
886 					struct spi_message *msg,
887 					struct spi_transfer *xfer)
888 {
889 	struct atmel_spi	*as = spi_master_get_devdata(master);
890 	u32			len;
891 	dma_addr_t		tx_dma, rx_dma;
892 
893 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
894 
895 	len = as->current_remaining_bytes;
896 	atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
897 	as->current_remaining_bytes -= len;
898 
899 	spi_writel(as, RPR, rx_dma);
900 	spi_writel(as, TPR, tx_dma);
901 
902 	if (msg->spi->bits_per_word > 8)
903 		len >>= 1;
904 	spi_writel(as, RCR, len);
905 	spi_writel(as, TCR, len);
906 
907 	dev_dbg(&msg->spi->dev,
908 		"  start xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
909 		xfer, xfer->len, xfer->tx_buf,
910 		(unsigned long long)xfer->tx_dma, xfer->rx_buf,
911 		(unsigned long long)xfer->rx_dma);
912 
913 	if (as->current_remaining_bytes) {
914 		len = as->current_remaining_bytes;
915 		atmel_spi_next_xfer_data(master, xfer, &tx_dma, &rx_dma, &len);
916 		as->current_remaining_bytes -= len;
917 
918 		spi_writel(as, RNPR, rx_dma);
919 		spi_writel(as, TNPR, tx_dma);
920 
921 		if (msg->spi->bits_per_word > 8)
922 			len >>= 1;
923 		spi_writel(as, RNCR, len);
924 		spi_writel(as, TNCR, len);
925 
926 		dev_dbg(&msg->spi->dev,
927 			"  next xfer %p: len %u tx %p/%08llx rx %p/%08llx\n",
928 			xfer, xfer->len, xfer->tx_buf,
929 			(unsigned long long)xfer->tx_dma, xfer->rx_buf,
930 			(unsigned long long)xfer->rx_dma);
931 	}
932 
933 	/* REVISIT: We're waiting for RXBUFF before we start the next
934 	 * transfer because we need to handle some difficult timing
935 	 * issues otherwise. If we wait for TXBUFE in one transfer and
936 	 * then starts waiting for RXBUFF in the next, it's difficult
937 	 * to tell the difference between the RXBUFF interrupt we're
938 	 * actually waiting for and the RXBUFF interrupt of the
939 	 * previous transfer.
940 	 *
941 	 * It should be doable, though. Just not now...
942 	 */
943 	spi_writel(as, IER, SPI_BIT(RXBUFF) | SPI_BIT(OVRES));
944 	spi_writel(as, PTCR, SPI_BIT(TXTEN) | SPI_BIT(RXTEN));
945 }
946 
947 /*
948  * For DMA, tx_buf/tx_dma have the same relationship as rx_buf/rx_dma:
949  *  - The buffer is either valid for CPU access, else NULL
950  *  - If the buffer is valid, so is its DMA address
951  *
952  * This driver manages the dma address unless message->is_dma_mapped.
953  */
954 static int
955 atmel_spi_dma_map_xfer(struct atmel_spi *as, struct spi_transfer *xfer)
956 {
957 	struct device	*dev = &as->pdev->dev;
958 
959 	xfer->tx_dma = xfer->rx_dma = INVALID_DMA_ADDRESS;
960 	if (xfer->tx_buf) {
961 		/* tx_buf is a const void* where we need a void * for the dma
962 		 * mapping */
963 		void *nonconst_tx = (void *)xfer->tx_buf;
964 
965 		xfer->tx_dma = dma_map_single(dev,
966 				nonconst_tx, xfer->len,
967 				DMA_TO_DEVICE);
968 		if (dma_mapping_error(dev, xfer->tx_dma))
969 			return -ENOMEM;
970 	}
971 	if (xfer->rx_buf) {
972 		xfer->rx_dma = dma_map_single(dev,
973 				xfer->rx_buf, xfer->len,
974 				DMA_FROM_DEVICE);
975 		if (dma_mapping_error(dev, xfer->rx_dma)) {
976 			if (xfer->tx_buf)
977 				dma_unmap_single(dev,
978 						xfer->tx_dma, xfer->len,
979 						DMA_TO_DEVICE);
980 			return -ENOMEM;
981 		}
982 	}
983 	return 0;
984 }
985 
986 static void atmel_spi_dma_unmap_xfer(struct spi_master *master,
987 				     struct spi_transfer *xfer)
988 {
989 	if (xfer->tx_dma != INVALID_DMA_ADDRESS)
990 		dma_unmap_single(master->dev.parent, xfer->tx_dma,
991 				 xfer->len, DMA_TO_DEVICE);
992 	if (xfer->rx_dma != INVALID_DMA_ADDRESS)
993 		dma_unmap_single(master->dev.parent, xfer->rx_dma,
994 				 xfer->len, DMA_FROM_DEVICE);
995 }
996 
997 static void atmel_spi_disable_pdc_transfer(struct atmel_spi *as)
998 {
999 	spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1000 }
1001 
1002 static void
1003 atmel_spi_pump_single_data(struct atmel_spi *as, struct spi_transfer *xfer)
1004 {
1005 	u8		*rxp;
1006 	u16		*rxp16;
1007 	unsigned long	xfer_pos = xfer->len - as->current_remaining_bytes;
1008 
1009 	if (xfer->bits_per_word > 8) {
1010 		rxp16 = (u16 *)(((u8 *)xfer->rx_buf) + xfer_pos);
1011 		*rxp16 = spi_readl(as, RDR);
1012 	} else {
1013 		rxp = ((u8 *)xfer->rx_buf) + xfer_pos;
1014 		*rxp = spi_readl(as, RDR);
1015 	}
1016 	if (xfer->bits_per_word > 8) {
1017 		if (as->current_remaining_bytes > 2)
1018 			as->current_remaining_bytes -= 2;
1019 		else
1020 			as->current_remaining_bytes = 0;
1021 	} else {
1022 		as->current_remaining_bytes--;
1023 	}
1024 }
1025 
1026 static void
1027 atmel_spi_pump_fifo_data(struct atmel_spi *as, struct spi_transfer *xfer)
1028 {
1029 	u32 fifolr = spi_readl(as, FLR);
1030 	u32 num_bytes, num_data = SPI_BFEXT(RXFL, fifolr);
1031 	u32 offset = xfer->len - as->current_remaining_bytes;
1032 	u16 *words = (u16 *)((u8 *)xfer->rx_buf + offset);
1033 	u8  *bytes = (u8  *)((u8 *)xfer->rx_buf + offset);
1034 	u16 rd; /* RD field is the lowest 16 bits of RDR */
1035 
1036 	/* Update the number of remaining bytes to transfer */
1037 	num_bytes = ((xfer->bits_per_word > 8) ?
1038 		     (num_data << 1) :
1039 		     num_data);
1040 
1041 	if (as->current_remaining_bytes > num_bytes)
1042 		as->current_remaining_bytes -= num_bytes;
1043 	else
1044 		as->current_remaining_bytes = 0;
1045 
1046 	/* Handle odd number of bytes when data are more than 8bit width */
1047 	if (xfer->bits_per_word > 8)
1048 		as->current_remaining_bytes &= ~0x1;
1049 
1050 	/* Read data */
1051 	while (num_data) {
1052 		rd = spi_readl(as, RDR);
1053 		if (xfer->bits_per_word > 8)
1054 			*words++ = rd;
1055 		else
1056 			*bytes++ = rd;
1057 		num_data--;
1058 	}
1059 }
1060 
1061 /* Called from IRQ
1062  *
1063  * Must update "current_remaining_bytes" to keep track of data
1064  * to transfer.
1065  */
1066 static void
1067 atmel_spi_pump_pio_data(struct atmel_spi *as, struct spi_transfer *xfer)
1068 {
1069 	if (as->fifo_size)
1070 		atmel_spi_pump_fifo_data(as, xfer);
1071 	else
1072 		atmel_spi_pump_single_data(as, xfer);
1073 }
1074 
1075 /* Interrupt
1076  *
1077  * No need for locking in this Interrupt handler: done_status is the
1078  * only information modified.
1079  */
1080 static irqreturn_t
1081 atmel_spi_pio_interrupt(int irq, void *dev_id)
1082 {
1083 	struct spi_master	*master = dev_id;
1084 	struct atmel_spi	*as = spi_master_get_devdata(master);
1085 	u32			status, pending, imr;
1086 	struct spi_transfer	*xfer;
1087 	int			ret = IRQ_NONE;
1088 
1089 	imr = spi_readl(as, IMR);
1090 	status = spi_readl(as, SR);
1091 	pending = status & imr;
1092 
1093 	if (pending & SPI_BIT(OVRES)) {
1094 		ret = IRQ_HANDLED;
1095 		spi_writel(as, IDR, SPI_BIT(OVRES));
1096 		dev_warn(master->dev.parent, "overrun\n");
1097 
1098 		/*
1099 		 * When we get an overrun, we disregard the current
1100 		 * transfer. Data will not be copied back from any
1101 		 * bounce buffer and msg->actual_len will not be
1102 		 * updated with the last xfer.
1103 		 *
1104 		 * We will also not process any remaning transfers in
1105 		 * the message.
1106 		 */
1107 		as->done_status = -EIO;
1108 		smp_wmb();
1109 
1110 		/* Clear any overrun happening while cleaning up */
1111 		spi_readl(as, SR);
1112 
1113 		complete(&as->xfer_completion);
1114 
1115 	} else if (pending & (SPI_BIT(RDRF) | SPI_BIT(RXFTHF))) {
1116 		atmel_spi_lock(as);
1117 
1118 		if (as->current_remaining_bytes) {
1119 			ret = IRQ_HANDLED;
1120 			xfer = as->current_transfer;
1121 			atmel_spi_pump_pio_data(as, xfer);
1122 			if (!as->current_remaining_bytes)
1123 				spi_writel(as, IDR, pending);
1124 
1125 			complete(&as->xfer_completion);
1126 		}
1127 
1128 		atmel_spi_unlock(as);
1129 	} else {
1130 		WARN_ONCE(pending, "IRQ not handled, pending = %x\n", pending);
1131 		ret = IRQ_HANDLED;
1132 		spi_writel(as, IDR, pending);
1133 	}
1134 
1135 	return ret;
1136 }
1137 
1138 static irqreturn_t
1139 atmel_spi_pdc_interrupt(int irq, void *dev_id)
1140 {
1141 	struct spi_master	*master = dev_id;
1142 	struct atmel_spi	*as = spi_master_get_devdata(master);
1143 	u32			status, pending, imr;
1144 	int			ret = IRQ_NONE;
1145 
1146 	imr = spi_readl(as, IMR);
1147 	status = spi_readl(as, SR);
1148 	pending = status & imr;
1149 
1150 	if (pending & SPI_BIT(OVRES)) {
1151 
1152 		ret = IRQ_HANDLED;
1153 
1154 		spi_writel(as, IDR, (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX)
1155 				     | SPI_BIT(OVRES)));
1156 
1157 		/* Clear any overrun happening while cleaning up */
1158 		spi_readl(as, SR);
1159 
1160 		as->done_status = -EIO;
1161 
1162 		complete(&as->xfer_completion);
1163 
1164 	} else if (pending & (SPI_BIT(RXBUFF) | SPI_BIT(ENDRX))) {
1165 		ret = IRQ_HANDLED;
1166 
1167 		spi_writel(as, IDR, pending);
1168 
1169 		complete(&as->xfer_completion);
1170 	}
1171 
1172 	return ret;
1173 }
1174 
1175 static int atmel_spi_setup(struct spi_device *spi)
1176 {
1177 	struct atmel_spi	*as;
1178 	struct atmel_spi_device	*asd;
1179 	u32			csr;
1180 	unsigned int		bits = spi->bits_per_word;
1181 
1182 	as = spi_master_get_devdata(spi->master);
1183 
1184 	/* see notes above re chipselect */
1185 	if (!atmel_spi_is_v2(as)
1186 			&& spi->chip_select == 0
1187 			&& (spi->mode & SPI_CS_HIGH)) {
1188 		dev_dbg(&spi->dev, "setup: can't be active-high\n");
1189 		return -EINVAL;
1190 	}
1191 
1192 	csr = SPI_BF(BITS, bits - 8);
1193 	if (spi->mode & SPI_CPOL)
1194 		csr |= SPI_BIT(CPOL);
1195 	if (!(spi->mode & SPI_CPHA))
1196 		csr |= SPI_BIT(NCPHA);
1197 	if (!as->use_cs_gpios)
1198 		csr |= SPI_BIT(CSAAT);
1199 
1200 	/* DLYBS is mostly irrelevant since we manage chipselect using GPIOs.
1201 	 */
1202 	csr |= SPI_BF(DLYBS, 0);
1203 
1204 	/* DLYBCT adds delays between words.  This is useful for slow devices
1205 	 * that need a bit of time to setup the next transfer.
1206 	 */
1207 	csr |= SPI_BF(DLYBCT,
1208 			(as->spi_clk / 1000000 * spi->word_delay_usecs) >> 5);
1209 
1210 	asd = spi->controller_state;
1211 	if (!asd) {
1212 		asd = kzalloc(sizeof(struct atmel_spi_device), GFP_KERNEL);
1213 		if (!asd)
1214 			return -ENOMEM;
1215 
1216 		/*
1217 		 * If use_cs_gpios is true this means that we have "cs-gpios"
1218 		 * defined in the device tree node so we should have
1219 		 * gotten the GPIO lines from the device tree inside the
1220 		 * SPI core. Warn if this is not the case but continue since
1221 		 * CS GPIOs are after all optional.
1222 		 */
1223 		if (as->use_cs_gpios) {
1224 			if (!spi->cs_gpiod) {
1225 				dev_err(&spi->dev,
1226 					"host claims to use CS GPIOs but no CS found in DT by the SPI core\n");
1227 			}
1228 			asd->npcs_pin = spi->cs_gpiod;
1229 		}
1230 
1231 		spi->controller_state = asd;
1232 	}
1233 
1234 	asd->csr = csr;
1235 
1236 	dev_dbg(&spi->dev,
1237 		"setup: bpw %u mode 0x%x -> csr%d %08x\n",
1238 		bits, spi->mode, spi->chip_select, csr);
1239 
1240 	if (!atmel_spi_is_v2(as))
1241 		spi_writel(as, CSR0 + 4 * spi->chip_select, csr);
1242 
1243 	return 0;
1244 }
1245 
1246 static int atmel_spi_one_transfer(struct spi_master *master,
1247 					struct spi_message *msg,
1248 					struct spi_transfer *xfer)
1249 {
1250 	struct atmel_spi	*as;
1251 	struct spi_device	*spi = msg->spi;
1252 	u8			bits;
1253 	u32			len;
1254 	struct atmel_spi_device	*asd;
1255 	int			timeout;
1256 	int			ret;
1257 	unsigned long		dma_timeout;
1258 
1259 	as = spi_master_get_devdata(master);
1260 
1261 	if (!(xfer->tx_buf || xfer->rx_buf) && xfer->len) {
1262 		dev_dbg(&spi->dev, "missing rx or tx buf\n");
1263 		return -EINVAL;
1264 	}
1265 
1266 	asd = spi->controller_state;
1267 	bits = (asd->csr >> 4) & 0xf;
1268 	if (bits != xfer->bits_per_word - 8) {
1269 		dev_dbg(&spi->dev,
1270 			"you can't yet change bits_per_word in transfers\n");
1271 		return -ENOPROTOOPT;
1272 	}
1273 
1274 	/*
1275 	 * DMA map early, for performance (empties dcache ASAP) and
1276 	 * better fault reporting.
1277 	 */
1278 	if ((!msg->is_dma_mapped)
1279 		&& as->use_pdc) {
1280 		if (atmel_spi_dma_map_xfer(as, xfer) < 0)
1281 			return -ENOMEM;
1282 	}
1283 
1284 	atmel_spi_set_xfer_speed(as, msg->spi, xfer);
1285 
1286 	as->done_status = 0;
1287 	as->current_transfer = xfer;
1288 	as->current_remaining_bytes = xfer->len;
1289 	while (as->current_remaining_bytes) {
1290 		reinit_completion(&as->xfer_completion);
1291 
1292 		if (as->use_pdc) {
1293 			atmel_spi_pdc_next_xfer(master, msg, xfer);
1294 		} else if (atmel_spi_use_dma(as, xfer)) {
1295 			len = as->current_remaining_bytes;
1296 			ret = atmel_spi_next_xfer_dma_submit(master,
1297 								xfer, &len);
1298 			if (ret) {
1299 				dev_err(&spi->dev,
1300 					"unable to use DMA, fallback to PIO\n");
1301 				atmel_spi_next_xfer_pio(master, xfer);
1302 			} else {
1303 				as->current_remaining_bytes -= len;
1304 				if (as->current_remaining_bytes < 0)
1305 					as->current_remaining_bytes = 0;
1306 			}
1307 		} else {
1308 			atmel_spi_next_xfer_pio(master, xfer);
1309 		}
1310 
1311 		/* interrupts are disabled, so free the lock for schedule */
1312 		atmel_spi_unlock(as);
1313 		dma_timeout = wait_for_completion_timeout(&as->xfer_completion,
1314 							  SPI_DMA_TIMEOUT);
1315 		atmel_spi_lock(as);
1316 		if (WARN_ON(dma_timeout == 0)) {
1317 			dev_err(&spi->dev, "spi transfer timeout\n");
1318 			as->done_status = -EIO;
1319 		}
1320 
1321 		if (as->done_status)
1322 			break;
1323 	}
1324 
1325 	if (as->done_status) {
1326 		if (as->use_pdc) {
1327 			dev_warn(master->dev.parent,
1328 				"overrun (%u/%u remaining)\n",
1329 				spi_readl(as, TCR), spi_readl(as, RCR));
1330 
1331 			/*
1332 			 * Clean up DMA registers and make sure the data
1333 			 * registers are empty.
1334 			 */
1335 			spi_writel(as, RNCR, 0);
1336 			spi_writel(as, TNCR, 0);
1337 			spi_writel(as, RCR, 0);
1338 			spi_writel(as, TCR, 0);
1339 			for (timeout = 1000; timeout; timeout--)
1340 				if (spi_readl(as, SR) & SPI_BIT(TXEMPTY))
1341 					break;
1342 			if (!timeout)
1343 				dev_warn(master->dev.parent,
1344 					 "timeout waiting for TXEMPTY");
1345 			while (spi_readl(as, SR) & SPI_BIT(RDRF))
1346 				spi_readl(as, RDR);
1347 
1348 			/* Clear any overrun happening while cleaning up */
1349 			spi_readl(as, SR);
1350 
1351 		} else if (atmel_spi_use_dma(as, xfer)) {
1352 			atmel_spi_stop_dma(master);
1353 		}
1354 
1355 		if (!msg->is_dma_mapped
1356 			&& as->use_pdc)
1357 			atmel_spi_dma_unmap_xfer(master, xfer);
1358 
1359 		return 0;
1360 
1361 	} else {
1362 		/* only update length if no error */
1363 		msg->actual_length += xfer->len;
1364 	}
1365 
1366 	if (!msg->is_dma_mapped
1367 		&& as->use_pdc)
1368 		atmel_spi_dma_unmap_xfer(master, xfer);
1369 
1370 	if (xfer->delay_usecs)
1371 		udelay(xfer->delay_usecs);
1372 
1373 	if (xfer->cs_change) {
1374 		if (list_is_last(&xfer->transfer_list,
1375 				 &msg->transfers)) {
1376 			as->keep_cs = true;
1377 		} else {
1378 			as->cs_active = !as->cs_active;
1379 			if (as->cs_active)
1380 				cs_activate(as, msg->spi);
1381 			else
1382 				cs_deactivate(as, msg->spi);
1383 		}
1384 	}
1385 
1386 	return 0;
1387 }
1388 
1389 static int atmel_spi_transfer_one_message(struct spi_master *master,
1390 						struct spi_message *msg)
1391 {
1392 	struct atmel_spi *as;
1393 	struct spi_transfer *xfer;
1394 	struct spi_device *spi = msg->spi;
1395 	int ret = 0;
1396 
1397 	as = spi_master_get_devdata(master);
1398 
1399 	dev_dbg(&spi->dev, "new message %p submitted for %s\n",
1400 					msg, dev_name(&spi->dev));
1401 
1402 	atmel_spi_lock(as);
1403 	cs_activate(as, spi);
1404 
1405 	as->cs_active = true;
1406 	as->keep_cs = false;
1407 
1408 	msg->status = 0;
1409 	msg->actual_length = 0;
1410 
1411 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1412 		ret = atmel_spi_one_transfer(master, msg, xfer);
1413 		if (ret)
1414 			goto msg_done;
1415 	}
1416 
1417 	if (as->use_pdc)
1418 		atmel_spi_disable_pdc_transfer(as);
1419 
1420 	list_for_each_entry(xfer, &msg->transfers, transfer_list) {
1421 		dev_dbg(&spi->dev,
1422 			"  xfer %p: len %u tx %p/%pad rx %p/%pad\n",
1423 			xfer, xfer->len,
1424 			xfer->tx_buf, &xfer->tx_dma,
1425 			xfer->rx_buf, &xfer->rx_dma);
1426 	}
1427 
1428 msg_done:
1429 	if (!as->keep_cs)
1430 		cs_deactivate(as, msg->spi);
1431 
1432 	atmel_spi_unlock(as);
1433 
1434 	msg->status = as->done_status;
1435 	spi_finalize_current_message(spi->master);
1436 
1437 	return ret;
1438 }
1439 
1440 static void atmel_spi_cleanup(struct spi_device *spi)
1441 {
1442 	struct atmel_spi_device	*asd = spi->controller_state;
1443 
1444 	if (!asd)
1445 		return;
1446 
1447 	spi->controller_state = NULL;
1448 	kfree(asd);
1449 }
1450 
1451 static inline unsigned int atmel_get_version(struct atmel_spi *as)
1452 {
1453 	return spi_readl(as, VERSION) & 0x00000fff;
1454 }
1455 
1456 static void atmel_get_caps(struct atmel_spi *as)
1457 {
1458 	unsigned int version;
1459 
1460 	version = atmel_get_version(as);
1461 
1462 	as->caps.is_spi2 = version > 0x121;
1463 	as->caps.has_wdrbt = version >= 0x210;
1464 	as->caps.has_dma_support = version >= 0x212;
1465 	as->caps.has_pdc_support = version < 0x212;
1466 }
1467 
1468 static void atmel_spi_init(struct atmel_spi *as)
1469 {
1470 	spi_writel(as, CR, SPI_BIT(SWRST));
1471 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1472 
1473 	/* It is recommended to enable FIFOs first thing after reset */
1474 	if (as->fifo_size)
1475 		spi_writel(as, CR, SPI_BIT(FIFOEN));
1476 
1477 	if (as->caps.has_wdrbt) {
1478 		spi_writel(as, MR, SPI_BIT(WDRBT) | SPI_BIT(MODFDIS)
1479 				| SPI_BIT(MSTR));
1480 	} else {
1481 		spi_writel(as, MR, SPI_BIT(MSTR) | SPI_BIT(MODFDIS));
1482 	}
1483 
1484 	if (as->use_pdc)
1485 		spi_writel(as, PTCR, SPI_BIT(RXTDIS) | SPI_BIT(TXTDIS));
1486 	spi_writel(as, CR, SPI_BIT(SPIEN));
1487 }
1488 
1489 static int atmel_spi_probe(struct platform_device *pdev)
1490 {
1491 	struct resource		*regs;
1492 	int			irq;
1493 	struct clk		*clk;
1494 	int			ret;
1495 	struct spi_master	*master;
1496 	struct atmel_spi	*as;
1497 
1498 	/* Select default pin state */
1499 	pinctrl_pm_select_default_state(&pdev->dev);
1500 
1501 	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1502 	if (!regs)
1503 		return -ENXIO;
1504 
1505 	irq = platform_get_irq(pdev, 0);
1506 	if (irq < 0)
1507 		return irq;
1508 
1509 	clk = devm_clk_get(&pdev->dev, "spi_clk");
1510 	if (IS_ERR(clk))
1511 		return PTR_ERR(clk);
1512 
1513 	/* setup spi core then atmel-specific driver state */
1514 	ret = -ENOMEM;
1515 	master = spi_alloc_master(&pdev->dev, sizeof(*as));
1516 	if (!master)
1517 		goto out_free;
1518 
1519 	/* the spi->mode bits understood by this driver: */
1520 	master->use_gpio_descriptors = true;
1521 	master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH;
1522 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(8, 16);
1523 	master->dev.of_node = pdev->dev.of_node;
1524 	master->bus_num = pdev->id;
1525 	master->num_chipselect = master->dev.of_node ? 0 : 4;
1526 	master->setup = atmel_spi_setup;
1527 	master->flags = (SPI_MASTER_MUST_RX | SPI_MASTER_MUST_TX);
1528 	master->transfer_one_message = atmel_spi_transfer_one_message;
1529 	master->cleanup = atmel_spi_cleanup;
1530 	master->auto_runtime_pm = true;
1531 	master->max_dma_len = SPI_MAX_DMA_XFER;
1532 	master->can_dma = atmel_spi_can_dma;
1533 	platform_set_drvdata(pdev, master);
1534 
1535 	as = spi_master_get_devdata(master);
1536 
1537 	spin_lock_init(&as->lock);
1538 
1539 	as->pdev = pdev;
1540 	as->regs = devm_ioremap_resource(&pdev->dev, regs);
1541 	if (IS_ERR(as->regs)) {
1542 		ret = PTR_ERR(as->regs);
1543 		goto out_unmap_regs;
1544 	}
1545 	as->phybase = regs->start;
1546 	as->irq = irq;
1547 	as->clk = clk;
1548 
1549 	init_completion(&as->xfer_completion);
1550 
1551 	atmel_get_caps(as);
1552 
1553 	/*
1554 	 * If there are chip selects in the device tree, those will be
1555 	 * discovered by the SPI core when registering the SPI master
1556 	 * and assigned to each SPI device.
1557 	 */
1558 	as->use_cs_gpios = true;
1559 	if (atmel_spi_is_v2(as) &&
1560 	    pdev->dev.of_node &&
1561 	    !of_get_property(pdev->dev.of_node, "cs-gpios", NULL)) {
1562 		as->use_cs_gpios = false;
1563 		master->num_chipselect = 4;
1564 	}
1565 
1566 	as->use_dma = false;
1567 	as->use_pdc = false;
1568 	if (as->caps.has_dma_support) {
1569 		ret = atmel_spi_configure_dma(master, as);
1570 		if (ret == 0) {
1571 			as->use_dma = true;
1572 		} else if (ret == -EPROBE_DEFER) {
1573 			return ret;
1574 		}
1575 	} else if (as->caps.has_pdc_support) {
1576 		as->use_pdc = true;
1577 	}
1578 
1579 	if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
1580 		as->addr_rx_bbuf = dma_alloc_coherent(&pdev->dev,
1581 						      SPI_MAX_DMA_XFER,
1582 						      &as->dma_addr_rx_bbuf,
1583 						      GFP_KERNEL | GFP_DMA);
1584 		if (!as->addr_rx_bbuf) {
1585 			as->use_dma = false;
1586 		} else {
1587 			as->addr_tx_bbuf = dma_alloc_coherent(&pdev->dev,
1588 					SPI_MAX_DMA_XFER,
1589 					&as->dma_addr_tx_bbuf,
1590 					GFP_KERNEL | GFP_DMA);
1591 			if (!as->addr_tx_bbuf) {
1592 				as->use_dma = false;
1593 				dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1594 						  as->addr_rx_bbuf,
1595 						  as->dma_addr_rx_bbuf);
1596 			}
1597 		}
1598 		if (!as->use_dma)
1599 			dev_info(master->dev.parent,
1600 				 "  can not allocate dma coherent memory\n");
1601 	}
1602 
1603 	if (as->caps.has_dma_support && !as->use_dma)
1604 		dev_info(&pdev->dev, "Atmel SPI Controller using PIO only\n");
1605 
1606 	if (as->use_pdc) {
1607 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pdc_interrupt,
1608 					0, dev_name(&pdev->dev), master);
1609 	} else {
1610 		ret = devm_request_irq(&pdev->dev, irq, atmel_spi_pio_interrupt,
1611 					0, dev_name(&pdev->dev), master);
1612 	}
1613 	if (ret)
1614 		goto out_unmap_regs;
1615 
1616 	/* Initialize the hardware */
1617 	ret = clk_prepare_enable(clk);
1618 	if (ret)
1619 		goto out_free_irq;
1620 
1621 	as->spi_clk = clk_get_rate(clk);
1622 
1623 	as->fifo_size = 0;
1624 	if (!of_property_read_u32(pdev->dev.of_node, "atmel,fifo-size",
1625 				  &as->fifo_size)) {
1626 		dev_info(&pdev->dev, "Using FIFO (%u data)\n", as->fifo_size);
1627 	}
1628 
1629 	atmel_spi_init(as);
1630 
1631 	pm_runtime_set_autosuspend_delay(&pdev->dev, AUTOSUSPEND_TIMEOUT);
1632 	pm_runtime_use_autosuspend(&pdev->dev);
1633 	pm_runtime_set_active(&pdev->dev);
1634 	pm_runtime_enable(&pdev->dev);
1635 
1636 	ret = devm_spi_register_master(&pdev->dev, master);
1637 	if (ret)
1638 		goto out_free_dma;
1639 
1640 	/* go! */
1641 	dev_info(&pdev->dev, "Atmel SPI Controller version 0x%x at 0x%08lx (irq %d)\n",
1642 			atmel_get_version(as), (unsigned long)regs->start,
1643 			irq);
1644 
1645 	return 0;
1646 
1647 out_free_dma:
1648 	pm_runtime_disable(&pdev->dev);
1649 	pm_runtime_set_suspended(&pdev->dev);
1650 
1651 	if (as->use_dma)
1652 		atmel_spi_release_dma(master);
1653 
1654 	spi_writel(as, CR, SPI_BIT(SWRST));
1655 	spi_writel(as, CR, SPI_BIT(SWRST)); /* AT91SAM9263 Rev B workaround */
1656 	clk_disable_unprepare(clk);
1657 out_free_irq:
1658 out_unmap_regs:
1659 out_free:
1660 	spi_master_put(master);
1661 	return ret;
1662 }
1663 
1664 static int atmel_spi_remove(struct platform_device *pdev)
1665 {
1666 	struct spi_master	*master = platform_get_drvdata(pdev);
1667 	struct atmel_spi	*as = spi_master_get_devdata(master);
1668 
1669 	pm_runtime_get_sync(&pdev->dev);
1670 
1671 	/* reset the hardware and block queue progress */
1672 	if (as->use_dma) {
1673 		atmel_spi_stop_dma(master);
1674 		atmel_spi_release_dma(master);
1675 		if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
1676 			dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1677 					  as->addr_tx_bbuf,
1678 					  as->dma_addr_tx_bbuf);
1679 			dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
1680 					  as->addr_rx_bbuf,
1681 					  as->dma_addr_rx_bbuf);
1682 		}
1683 	}
1684 
1685 	spin_lock_irq(&as->lock);
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 	clk_disable_unprepare(as->clk);
1692 
1693 	pm_runtime_put_noidle(&pdev->dev);
1694 	pm_runtime_disable(&pdev->dev);
1695 
1696 	return 0;
1697 }
1698 
1699 #ifdef CONFIG_PM
1700 static int atmel_spi_runtime_suspend(struct device *dev)
1701 {
1702 	struct spi_master *master = dev_get_drvdata(dev);
1703 	struct atmel_spi *as = spi_master_get_devdata(master);
1704 
1705 	clk_disable_unprepare(as->clk);
1706 	pinctrl_pm_select_sleep_state(dev);
1707 
1708 	return 0;
1709 }
1710 
1711 static int atmel_spi_runtime_resume(struct device *dev)
1712 {
1713 	struct spi_master *master = dev_get_drvdata(dev);
1714 	struct atmel_spi *as = spi_master_get_devdata(master);
1715 
1716 	pinctrl_pm_select_default_state(dev);
1717 
1718 	return clk_prepare_enable(as->clk);
1719 }
1720 
1721 #ifdef CONFIG_PM_SLEEP
1722 static int atmel_spi_suspend(struct device *dev)
1723 {
1724 	struct spi_master *master = dev_get_drvdata(dev);
1725 	int ret;
1726 
1727 	/* Stop the queue running */
1728 	ret = spi_master_suspend(master);
1729 	if (ret)
1730 		return ret;
1731 
1732 	if (!pm_runtime_suspended(dev))
1733 		atmel_spi_runtime_suspend(dev);
1734 
1735 	return 0;
1736 }
1737 
1738 static int atmel_spi_resume(struct device *dev)
1739 {
1740 	struct spi_master *master = dev_get_drvdata(dev);
1741 	struct atmel_spi *as = spi_master_get_devdata(master);
1742 	int ret;
1743 
1744 	ret = clk_prepare_enable(as->clk);
1745 	if (ret)
1746 		return ret;
1747 
1748 	atmel_spi_init(as);
1749 
1750 	clk_disable_unprepare(as->clk);
1751 
1752 	if (!pm_runtime_suspended(dev)) {
1753 		ret = atmel_spi_runtime_resume(dev);
1754 		if (ret)
1755 			return ret;
1756 	}
1757 
1758 	/* Start the queue running */
1759 	return spi_master_resume(master);
1760 }
1761 #endif
1762 
1763 static const struct dev_pm_ops atmel_spi_pm_ops = {
1764 	SET_SYSTEM_SLEEP_PM_OPS(atmel_spi_suspend, atmel_spi_resume)
1765 	SET_RUNTIME_PM_OPS(atmel_spi_runtime_suspend,
1766 			   atmel_spi_runtime_resume, NULL)
1767 };
1768 #define ATMEL_SPI_PM_OPS	(&atmel_spi_pm_ops)
1769 #else
1770 #define ATMEL_SPI_PM_OPS	NULL
1771 #endif
1772 
1773 #if defined(CONFIG_OF)
1774 static const struct of_device_id atmel_spi_dt_ids[] = {
1775 	{ .compatible = "atmel,at91rm9200-spi" },
1776 	{ /* sentinel */ }
1777 };
1778 
1779 MODULE_DEVICE_TABLE(of, atmel_spi_dt_ids);
1780 #endif
1781 
1782 static struct platform_driver atmel_spi_driver = {
1783 	.driver		= {
1784 		.name	= "atmel_spi",
1785 		.pm	= ATMEL_SPI_PM_OPS,
1786 		.of_match_table	= of_match_ptr(atmel_spi_dt_ids),
1787 	},
1788 	.probe		= atmel_spi_probe,
1789 	.remove		= atmel_spi_remove,
1790 };
1791 module_platform_driver(atmel_spi_driver);
1792 
1793 MODULE_DESCRIPTION("Atmel AT32/AT91 SPI Controller driver");
1794 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1795 MODULE_LICENSE("GPL");
1796 MODULE_ALIAS("platform:atmel_spi");
1797