Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Driver for Broadcom BCM2835 SPI Controllers
4 *
5 * Copyright (C) 2012 Chris Boot
6 * Copyright (C) 2013 Stephen Warren
7 * Copyright (C) 2015 Martin Sperl
8 *
9 * This driver is inspired by:
10 * spi-ath79.c, Copyright (C) 2009-2011 Gabor Juhos <juhosg@openwrt.org>
11 * spi-atmel.c, Copyright (C) 2006 Atmel Corporation
12 */
13
14#include <linux/cleanup.h>
15#include <linux/clk.h>
16#include <linux/completion.h>
17#include <linux/debugfs.h>
18#include <linux/delay.h>
19#include <linux/dma-mapping.h>
20#include <linux/dmaengine.h>
21#include <linux/err.h>
22#include <linux/interrupt.h>
23#include <linux/io.h>
24#include <linux/kernel.h>
25#include <linux/module.h>
26#include <linux/of.h>
27#include <linux/of_address.h>
28#include <linux/platform_device.h>
29#include <linux/gpio/consumer.h>
30#include <linux/gpio/machine.h> /* FIXME: using GPIO lookup tables */
31#include <linux/of_irq.h>
32#include <linux/overflow.h>
33#include <linux/slab.h>
34#include <linux/spi/spi.h>
35
36/* SPI register offsets */
37#define BCM2835_SPI_CS 0x00
38#define BCM2835_SPI_FIFO 0x04
39#define BCM2835_SPI_CLK 0x08
40#define BCM2835_SPI_DLEN 0x0c
41#define BCM2835_SPI_LTOH 0x10
42#define BCM2835_SPI_DC 0x14
43
44/* Bitfields in CS */
45#define BCM2835_SPI_CS_LEN_LONG 0x02000000
46#define BCM2835_SPI_CS_DMA_LEN 0x01000000
47#define BCM2835_SPI_CS_CSPOL2 0x00800000
48#define BCM2835_SPI_CS_CSPOL1 0x00400000
49#define BCM2835_SPI_CS_CSPOL0 0x00200000
50#define BCM2835_SPI_CS_RXF 0x00100000
51#define BCM2835_SPI_CS_RXR 0x00080000
52#define BCM2835_SPI_CS_TXD 0x00040000
53#define BCM2835_SPI_CS_RXD 0x00020000
54#define BCM2835_SPI_CS_DONE 0x00010000
55#define BCM2835_SPI_CS_LEN 0x00002000
56#define BCM2835_SPI_CS_REN 0x00001000
57#define BCM2835_SPI_CS_ADCS 0x00000800
58#define BCM2835_SPI_CS_INTR 0x00000400
59#define BCM2835_SPI_CS_INTD 0x00000200
60#define BCM2835_SPI_CS_DMAEN 0x00000100
61#define BCM2835_SPI_CS_TA 0x00000080
62#define BCM2835_SPI_CS_CSPOL 0x00000040
63#define BCM2835_SPI_CS_CLEAR_RX 0x00000020
64#define BCM2835_SPI_CS_CLEAR_TX 0x00000010
65#define BCM2835_SPI_CS_CPOL 0x00000008
66#define BCM2835_SPI_CS_CPHA 0x00000004
67#define BCM2835_SPI_CS_CS_10 0x00000002
68#define BCM2835_SPI_CS_CS_01 0x00000001
69
70#define BCM2835_SPI_FIFO_SIZE 64
71#define BCM2835_SPI_FIFO_SIZE_3_4 48
72#define BCM2835_SPI_DMA_MIN_LENGTH 96
73#define BCM2835_SPI_MODE_BITS (SPI_CPOL | SPI_CPHA | SPI_CS_HIGH \
74 | SPI_NO_CS | SPI_3WIRE)
75
76#define DRV_NAME "spi-bcm2835"
77
78/* define polling limits */
79static unsigned int polling_limit_us = 30;
80module_param(polling_limit_us, uint, 0664);
81MODULE_PARM_DESC(polling_limit_us,
82 "time in us to run a transfer in polling mode\n");
83
84/**
85 * struct bcm2835_spi - BCM2835 SPI controller
86 * @regs: base address of register map
87 * @clk: core clock, divided to calculate serial clock
88 * @cs_gpio: chip-select GPIO descriptor
89 * @clk_hz: core clock cached speed
90 * @irq: interrupt, signals TX FIFO empty or RX FIFO ¾ full
91 * @tfr: SPI transfer currently processed
92 * @ctlr: SPI controller reverse lookup
93 * @tx_buf: pointer whence next transmitted byte is read
94 * @rx_buf: pointer where next received byte is written
95 * @tx_len: remaining bytes to transmit
96 * @rx_len: remaining bytes to receive
97 * @tx_prologue: bytes transmitted without DMA if first TX sglist entry's
98 * length is not a multiple of 4 (to overcome hardware limitation)
99 * @rx_prologue: bytes received without DMA if first RX sglist entry's
100 * length is not a multiple of 4 (to overcome hardware limitation)
101 * @tx_spillover: whether @tx_prologue spills over to second TX sglist entry
102 * @debugfs_dir: the debugfs directory - neede to remove debugfs when
103 * unloading the module
104 * @count_transfer_polling: count of how often polling mode is used
105 * @count_transfer_irq: count of how often interrupt mode is used
106 * @count_transfer_irq_after_polling: count of how often we fall back to
107 * interrupt mode after starting in polling mode.
108 * These are counted as well in @count_transfer_polling and
109 * @count_transfer_irq
110 * @count_transfer_dma: count how often dma mode is used
111 * @target: SPI target currently selected
112 * (used by bcm2835_spi_dma_tx_done() to write @clear_rx_cs)
113 * @tx_dma_active: whether a TX DMA descriptor is in progress
114 * @rx_dma_active: whether a RX DMA descriptor is in progress
115 * (used by bcm2835_spi_dma_tx_done() to handle a race)
116 * @fill_tx_desc: preallocated TX DMA descriptor used for RX-only transfers
117 * (cyclically copies from zero page to TX FIFO)
118 * @fill_tx_addr: bus address of zero page
119 */
120struct bcm2835_spi {
121 void __iomem *regs;
122 struct clk *clk;
123 struct gpio_desc *cs_gpio;
124 unsigned long clk_hz;
125 int irq;
126 struct spi_transfer *tfr;
127 struct spi_controller *ctlr;
128 const u8 *tx_buf;
129 u8 *rx_buf;
130 int tx_len;
131 int rx_len;
132 int tx_prologue;
133 int rx_prologue;
134 unsigned int tx_spillover;
135
136 struct dentry *debugfs_dir;
137 u64 count_transfer_polling;
138 u64 count_transfer_irq;
139 u64 count_transfer_irq_after_polling;
140 u64 count_transfer_dma;
141
142 struct bcm2835_spidev *target;
143 unsigned int tx_dma_active;
144 unsigned int rx_dma_active;
145 struct dma_async_tx_descriptor *fill_tx_desc;
146 dma_addr_t fill_tx_addr;
147};
148
149/**
150 * struct bcm2835_spidev - BCM2835 SPI target
151 * @prepare_cs: precalculated CS register value for ->prepare_message()
152 * (uses target-specific clock polarity and phase settings)
153 * @clear_rx_desc: preallocated RX DMA descriptor used for TX-only transfers
154 * (cyclically clears RX FIFO by writing @clear_rx_cs to CS register)
155 * @clear_rx_addr: bus address of @clear_rx_cs
156 * @clear_rx_cs: precalculated CS register value to clear RX FIFO
157 * (uses target-specific clock polarity and phase settings)
158 */
159struct bcm2835_spidev {
160 u32 prepare_cs;
161 struct dma_async_tx_descriptor *clear_rx_desc;
162 dma_addr_t clear_rx_addr;
163 u32 clear_rx_cs ____cacheline_aligned;
164};
165
166#if defined(CONFIG_DEBUG_FS)
167static void bcm2835_debugfs_create(struct bcm2835_spi *bs,
168 const char *dname)
169{
170 char name[64];
171 struct dentry *dir;
172
173 /* get full name */
174 snprintf(name, sizeof(name), "spi-bcm2835-%s", dname);
175
176 /* the base directory */
177 dir = debugfs_create_dir(name, NULL);
178 bs->debugfs_dir = dir;
179
180 /* the counters */
181 debugfs_create_u64("count_transfer_polling", 0444, dir,
182 &bs->count_transfer_polling);
183 debugfs_create_u64("count_transfer_irq", 0444, dir,
184 &bs->count_transfer_irq);
185 debugfs_create_u64("count_transfer_irq_after_polling", 0444, dir,
186 &bs->count_transfer_irq_after_polling);
187 debugfs_create_u64("count_transfer_dma", 0444, dir,
188 &bs->count_transfer_dma);
189}
190
191static void bcm2835_debugfs_remove(struct bcm2835_spi *bs)
192{
193 debugfs_remove_recursive(bs->debugfs_dir);
194 bs->debugfs_dir = NULL;
195}
196#else
197static void bcm2835_debugfs_create(struct bcm2835_spi *bs,
198 const char *dname)
199{
200}
201
202static void bcm2835_debugfs_remove(struct bcm2835_spi *bs)
203{
204}
205#endif /* CONFIG_DEBUG_FS */
206
207static inline u32 bcm2835_rd(struct bcm2835_spi *bs, unsigned int reg)
208{
209 return readl(bs->regs + reg);
210}
211
212static inline void bcm2835_wr(struct bcm2835_spi *bs, unsigned int reg, u32 val)
213{
214 writel(val, bs->regs + reg);
215}
216
217static inline void bcm2835_rd_fifo(struct bcm2835_spi *bs)
218{
219 u8 byte;
220
221 while ((bs->rx_len) &&
222 (bcm2835_rd(bs, BCM2835_SPI_CS) & BCM2835_SPI_CS_RXD)) {
223 byte = bcm2835_rd(bs, BCM2835_SPI_FIFO);
224 if (bs->rx_buf)
225 *bs->rx_buf++ = byte;
226 bs->rx_len--;
227 }
228}
229
230static inline void bcm2835_wr_fifo(struct bcm2835_spi *bs)
231{
232 u8 byte;
233
234 while ((bs->tx_len) &&
235 (bcm2835_rd(bs, BCM2835_SPI_CS) & BCM2835_SPI_CS_TXD)) {
236 byte = bs->tx_buf ? *bs->tx_buf++ : 0;
237 bcm2835_wr(bs, BCM2835_SPI_FIFO, byte);
238 bs->tx_len--;
239 }
240}
241
242/**
243 * bcm2835_rd_fifo_count() - blindly read exactly @count bytes from RX FIFO
244 * @bs: BCM2835 SPI controller
245 * @count: bytes to read from RX FIFO
246 *
247 * The caller must ensure that @bs->rx_len is greater than or equal to @count,
248 * that the RX FIFO contains at least @count bytes and that the DMA Enable flag
249 * in the CS register is set (such that a read from the FIFO register receives
250 * 32-bit instead of just 8-bit). Moreover @bs->rx_buf must not be %NULL.
251 */
252static inline void bcm2835_rd_fifo_count(struct bcm2835_spi *bs, int count)
253{
254 u32 val;
255 int len;
256
257 bs->rx_len -= count;
258
259 do {
260 val = bcm2835_rd(bs, BCM2835_SPI_FIFO);
261 len = min(count, 4);
262 memcpy(bs->rx_buf, &val, len);
263 bs->rx_buf += len;
264 count -= 4;
265 } while (count > 0);
266}
267
268/**
269 * bcm2835_wr_fifo_count() - blindly write exactly @count bytes to TX FIFO
270 * @bs: BCM2835 SPI controller
271 * @count: bytes to write to TX FIFO
272 *
273 * The caller must ensure that @bs->tx_len is greater than or equal to @count,
274 * that the TX FIFO can accommodate @count bytes and that the DMA Enable flag
275 * in the CS register is set (such that a write to the FIFO register transmits
276 * 32-bit instead of just 8-bit).
277 */
278static inline void bcm2835_wr_fifo_count(struct bcm2835_spi *bs, int count)
279{
280 u32 val;
281 int len;
282
283 bs->tx_len -= count;
284
285 do {
286 if (bs->tx_buf) {
287 len = min(count, 4);
288 memcpy(&val, bs->tx_buf, len);
289 bs->tx_buf += len;
290 } else {
291 val = 0;
292 }
293 bcm2835_wr(bs, BCM2835_SPI_FIFO, val);
294 count -= 4;
295 } while (count > 0);
296}
297
298/**
299 * bcm2835_wait_tx_fifo_empty() - busy-wait for TX FIFO to empty
300 * @bs: BCM2835 SPI controller
301 *
302 * The caller must ensure that the RX FIFO can accommodate as many bytes
303 * as have been written to the TX FIFO: Transmission is halted once the
304 * RX FIFO is full, causing this function to spin forever.
305 */
306static inline void bcm2835_wait_tx_fifo_empty(struct bcm2835_spi *bs)
307{
308 while (!(bcm2835_rd(bs, BCM2835_SPI_CS) & BCM2835_SPI_CS_DONE))
309 cpu_relax();
310}
311
312/**
313 * bcm2835_rd_fifo_blind() - blindly read up to @count bytes from RX FIFO
314 * @bs: BCM2835 SPI controller
315 * @count: bytes available for reading in RX FIFO
316 */
317static inline void bcm2835_rd_fifo_blind(struct bcm2835_spi *bs, int count)
318{
319 u8 val;
320
321 count = min(count, bs->rx_len);
322 bs->rx_len -= count;
323
324 do {
325 val = bcm2835_rd(bs, BCM2835_SPI_FIFO);
326 if (bs->rx_buf)
327 *bs->rx_buf++ = val;
328 } while (--count);
329}
330
331/**
332 * bcm2835_wr_fifo_blind() - blindly write up to @count bytes to TX FIFO
333 * @bs: BCM2835 SPI controller
334 * @count: bytes available for writing in TX FIFO
335 */
336static inline void bcm2835_wr_fifo_blind(struct bcm2835_spi *bs, int count)
337{
338 u8 val;
339
340 count = min(count, bs->tx_len);
341 bs->tx_len -= count;
342
343 do {
344 val = bs->tx_buf ? *bs->tx_buf++ : 0;
345 bcm2835_wr(bs, BCM2835_SPI_FIFO, val);
346 } while (--count);
347}
348
349static void bcm2835_spi_reset_hw(struct bcm2835_spi *bs)
350{
351 u32 cs = bcm2835_rd(bs, BCM2835_SPI_CS);
352
353 /* Disable SPI interrupts and transfer */
354 cs &= ~(BCM2835_SPI_CS_INTR |
355 BCM2835_SPI_CS_INTD |
356 BCM2835_SPI_CS_DMAEN |
357 BCM2835_SPI_CS_TA);
358 /*
359 * Transmission sometimes breaks unless the DONE bit is written at the
360 * end of every transfer. The spec says it's a RO bit. Either the
361 * spec is wrong and the bit is actually of type RW1C, or it's a
362 * hardware erratum.
363 */
364 cs |= BCM2835_SPI_CS_DONE;
365 /* and reset RX/TX FIFOS */
366 cs |= BCM2835_SPI_CS_CLEAR_RX | BCM2835_SPI_CS_CLEAR_TX;
367
368 /* and reset the SPI_HW */
369 bcm2835_wr(bs, BCM2835_SPI_CS, cs);
370 /* as well as DLEN */
371 bcm2835_wr(bs, BCM2835_SPI_DLEN, 0);
372}
373
374static irqreturn_t bcm2835_spi_interrupt(int irq, void *dev_id)
375{
376 struct bcm2835_spi *bs = dev_id;
377 u32 cs = bcm2835_rd(bs, BCM2835_SPI_CS);
378
379 /* Bail out early if interrupts are not enabled */
380 if (!(cs & BCM2835_SPI_CS_INTR))
381 return IRQ_NONE;
382
383 /*
384 * An interrupt is signaled either if DONE is set (TX FIFO empty)
385 * or if RXR is set (RX FIFO >= ¾ full).
386 */
387 if (cs & BCM2835_SPI_CS_RXF)
388 bcm2835_rd_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE);
389 else if (cs & BCM2835_SPI_CS_RXR)
390 bcm2835_rd_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE_3_4);
391
392 if (bs->tx_len && cs & BCM2835_SPI_CS_DONE)
393 bcm2835_wr_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE);
394
395 /* Read as many bytes as possible from FIFO */
396 bcm2835_rd_fifo(bs);
397 /* Write as many bytes as possible to FIFO */
398 bcm2835_wr_fifo(bs);
399
400 if (!bs->rx_len) {
401 /* Transfer complete - reset SPI HW */
402 bcm2835_spi_reset_hw(bs);
403 /* wake up the framework */
404 spi_finalize_current_transfer(bs->ctlr);
405 }
406
407 return IRQ_HANDLED;
408}
409
410static int bcm2835_spi_transfer_one_irq(struct spi_controller *ctlr,
411 struct spi_device *spi,
412 struct spi_transfer *tfr,
413 u32 cs, bool fifo_empty)
414{
415 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
416
417 /* update usage statistics */
418 bs->count_transfer_irq++;
419
420 /*
421 * Enable HW block, but with interrupts still disabled.
422 * Otherwise the empty TX FIFO would immediately trigger an interrupt.
423 */
424 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_TA);
425
426 /* fill TX FIFO as much as possible */
427 if (fifo_empty)
428 bcm2835_wr_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE);
429 bcm2835_wr_fifo(bs);
430
431 /* enable interrupts */
432 cs |= BCM2835_SPI_CS_INTR | BCM2835_SPI_CS_INTD | BCM2835_SPI_CS_TA;
433 bcm2835_wr(bs, BCM2835_SPI_CS, cs);
434
435 /* signal that we need to wait for completion */
436 return 1;
437}
438
439/**
440 * bcm2835_spi_transfer_prologue() - transfer first few bytes without DMA
441 * @ctlr: SPI host controller
442 * @tfr: SPI transfer
443 * @bs: BCM2835 SPI controller
444 * @cs: CS register
445 *
446 * A limitation in DMA mode is that the FIFO must be accessed in 4 byte chunks.
447 * Only the final write access is permitted to transmit less than 4 bytes, the
448 * SPI controller deduces its intended size from the DLEN register.
449 *
450 * If a TX or RX sglist contains multiple entries, one per page, and the first
451 * entry starts in the middle of a page, that first entry's length may not be
452 * a multiple of 4. Subsequent entries are fine because they span an entire
453 * page, hence do have a length that's a multiple of 4.
454 *
455 * This cannot happen with kmalloc'ed buffers (which is what most clients use)
456 * because they are contiguous in physical memory and therefore not split on
457 * page boundaries by spi_map_buf(). But it *can* happen with vmalloc'ed
458 * buffers.
459 *
460 * The DMA engine is incapable of combining sglist entries into a continuous
461 * stream of 4 byte chunks, it treats every entry separately: A TX entry is
462 * rounded up a to a multiple of 4 bytes by transmitting surplus bytes, an RX
463 * entry is rounded up by throwing away received bytes.
464 *
465 * Overcome this limitation by transferring the first few bytes without DMA:
466 * E.g. if the first TX sglist entry's length is 23 and the first RX's is 42,
467 * write 3 bytes to the TX FIFO but read only 2 bytes from the RX FIFO.
468 * The residue of 1 byte in the RX FIFO is picked up by DMA. Together with
469 * the rest of the first RX sglist entry it makes up a multiple of 4 bytes.
470 *
471 * Should the RX prologue be larger, say, 3 vis-à-vis a TX prologue of 1,
472 * write 1 + 4 = 5 bytes to the TX FIFO and read 3 bytes from the RX FIFO.
473 * Caution, the additional 4 bytes spill over to the second TX sglist entry
474 * if the length of the first is *exactly* 1.
475 *
476 * At most 6 bytes are written and at most 3 bytes read. Do we know the
477 * transfer has this many bytes? Yes, see BCM2835_SPI_DMA_MIN_LENGTH.
478 *
479 * The FIFO is normally accessed with 8-bit width by the CPU and 32-bit width
480 * by the DMA engine. Toggling the DMA Enable flag in the CS register switches
481 * the width but also garbles the FIFO's contents. The prologue must therefore
482 * be transmitted in 32-bit width to ensure that the following DMA transfer can
483 * pick up the residue in the RX FIFO in ungarbled form.
484 */
485static void bcm2835_spi_transfer_prologue(struct spi_controller *ctlr,
486 struct spi_transfer *tfr,
487 struct bcm2835_spi *bs,
488 u32 cs)
489{
490 int tx_remaining;
491
492 bs->tfr = tfr;
493 bs->tx_prologue = 0;
494 bs->rx_prologue = 0;
495 bs->tx_spillover = false;
496
497 if (bs->tx_buf && !sg_is_last(&tfr->tx_sg.sgl[0]))
498 bs->tx_prologue = sg_dma_len(&tfr->tx_sg.sgl[0]) & 3;
499
500 if (bs->rx_buf && !sg_is_last(&tfr->rx_sg.sgl[0])) {
501 bs->rx_prologue = sg_dma_len(&tfr->rx_sg.sgl[0]) & 3;
502
503 if (bs->rx_prologue > bs->tx_prologue) {
504 if (!bs->tx_buf || sg_is_last(&tfr->tx_sg.sgl[0])) {
505 bs->tx_prologue = bs->rx_prologue;
506 } else {
507 bs->tx_prologue += 4;
508 bs->tx_spillover =
509 !(sg_dma_len(&tfr->tx_sg.sgl[0]) & ~3);
510 }
511 }
512 }
513
514 /* rx_prologue > 0 implies tx_prologue > 0, so check only the latter */
515 if (!bs->tx_prologue)
516 return;
517
518 /* Write and read RX prologue. Adjust first entry in RX sglist. */
519 if (bs->rx_prologue) {
520 bcm2835_wr(bs, BCM2835_SPI_DLEN, bs->rx_prologue);
521 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_TA
522 | BCM2835_SPI_CS_DMAEN);
523 bcm2835_wr_fifo_count(bs, bs->rx_prologue);
524 bcm2835_wait_tx_fifo_empty(bs);
525 bcm2835_rd_fifo_count(bs, bs->rx_prologue);
526 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_CLEAR_RX
527 | BCM2835_SPI_CS_CLEAR_TX
528 | BCM2835_SPI_CS_DONE);
529
530 dma_sync_single_for_device(ctlr->dma_rx->device->dev,
531 sg_dma_address(&tfr->rx_sg.sgl[0]),
532 bs->rx_prologue, DMA_FROM_DEVICE);
533
534 sg_dma_address(&tfr->rx_sg.sgl[0]) += bs->rx_prologue;
535 sg_dma_len(&tfr->rx_sg.sgl[0]) -= bs->rx_prologue;
536 }
537
538 if (!bs->tx_buf)
539 return;
540
541 /*
542 * Write remaining TX prologue. Adjust first entry in TX sglist.
543 * Also adjust second entry if prologue spills over to it.
544 */
545 tx_remaining = bs->tx_prologue - bs->rx_prologue;
546 if (tx_remaining) {
547 bcm2835_wr(bs, BCM2835_SPI_DLEN, tx_remaining);
548 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_TA
549 | BCM2835_SPI_CS_DMAEN);
550 bcm2835_wr_fifo_count(bs, tx_remaining);
551 bcm2835_wait_tx_fifo_empty(bs);
552 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_CLEAR_TX
553 | BCM2835_SPI_CS_DONE);
554 }
555
556 if (likely(!bs->tx_spillover)) {
557 sg_dma_address(&tfr->tx_sg.sgl[0]) += bs->tx_prologue;
558 sg_dma_len(&tfr->tx_sg.sgl[0]) -= bs->tx_prologue;
559 } else {
560 sg_dma_len(&tfr->tx_sg.sgl[0]) = 0;
561 sg_dma_address(&tfr->tx_sg.sgl[1]) += 4;
562 sg_dma_len(&tfr->tx_sg.sgl[1]) -= 4;
563 }
564}
565
566/**
567 * bcm2835_spi_undo_prologue() - reconstruct original sglist state
568 * @bs: BCM2835 SPI controller
569 *
570 * Undo changes which were made to an SPI transfer's sglist when transmitting
571 * the prologue. This is necessary to ensure the same memory ranges are
572 * unmapped that were originally mapped.
573 */
574static void bcm2835_spi_undo_prologue(struct bcm2835_spi *bs)
575{
576 struct spi_transfer *tfr = bs->tfr;
577
578 if (!bs->tx_prologue)
579 return;
580
581 if (bs->rx_prologue) {
582 sg_dma_address(&tfr->rx_sg.sgl[0]) -= bs->rx_prologue;
583 sg_dma_len(&tfr->rx_sg.sgl[0]) += bs->rx_prologue;
584 }
585
586 if (!bs->tx_buf)
587 goto out;
588
589 if (likely(!bs->tx_spillover)) {
590 sg_dma_address(&tfr->tx_sg.sgl[0]) -= bs->tx_prologue;
591 sg_dma_len(&tfr->tx_sg.sgl[0]) += bs->tx_prologue;
592 } else {
593 sg_dma_len(&tfr->tx_sg.sgl[0]) = bs->tx_prologue - 4;
594 sg_dma_address(&tfr->tx_sg.sgl[1]) -= 4;
595 sg_dma_len(&tfr->tx_sg.sgl[1]) += 4;
596 }
597out:
598 bs->tx_prologue = 0;
599}
600
601/**
602 * bcm2835_spi_dma_rx_done() - callback for DMA RX channel
603 * @data: SPI host controller
604 *
605 * Used for bidirectional and RX-only transfers.
606 */
607static void bcm2835_spi_dma_rx_done(void *data)
608{
609 struct spi_controller *ctlr = data;
610 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
611
612 /* terminate tx-dma as we do not have an irq for it
613 * because when the rx dma will terminate and this callback
614 * is called the tx-dma must have finished - can't get to this
615 * situation otherwise...
616 */
617 dmaengine_terminate_async(ctlr->dma_tx);
618 bs->tx_dma_active = false;
619 bs->rx_dma_active = false;
620 bcm2835_spi_undo_prologue(bs);
621
622 /* reset fifo and HW */
623 bcm2835_spi_reset_hw(bs);
624
625 /* and mark as completed */;
626 spi_finalize_current_transfer(ctlr);
627}
628
629/**
630 * bcm2835_spi_dma_tx_done() - callback for DMA TX channel
631 * @data: SPI host controller
632 *
633 * Used for TX-only transfers.
634 */
635static void bcm2835_spi_dma_tx_done(void *data)
636{
637 struct spi_controller *ctlr = data;
638 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
639
640 /* busy-wait for TX FIFO to empty */
641 while (!(bcm2835_rd(bs, BCM2835_SPI_CS) & BCM2835_SPI_CS_DONE))
642 bcm2835_wr(bs, BCM2835_SPI_CS, bs->target->clear_rx_cs);
643
644 bs->tx_dma_active = false;
645 smp_wmb();
646
647 /*
648 * In case of a very short transfer, RX DMA may not have been
649 * issued yet. The onus is then on bcm2835_spi_transfer_one_dma()
650 * to terminate it immediately after issuing.
651 */
652 if (cmpxchg(&bs->rx_dma_active, true, false))
653 dmaengine_terminate_async(ctlr->dma_rx);
654
655 bcm2835_spi_undo_prologue(bs);
656 bcm2835_spi_reset_hw(bs);
657 spi_finalize_current_transfer(ctlr);
658}
659
660/**
661 * bcm2835_spi_prepare_sg() - prepare and submit DMA descriptor for sglist
662 * @ctlr: SPI host controller
663 * @tfr: SPI transfer
664 * @bs: BCM2835 SPI controller
665 * @target: BCM2835 SPI target
666 * @is_tx: whether to submit DMA descriptor for TX or RX sglist
667 *
668 * Prepare and submit a DMA descriptor for the TX or RX sglist of @tfr.
669 * Return 0 on success or a negative error number.
670 */
671static int bcm2835_spi_prepare_sg(struct spi_controller *ctlr,
672 struct spi_transfer *tfr,
673 struct bcm2835_spi *bs,
674 struct bcm2835_spidev *target,
675 bool is_tx)
676{
677 struct dma_chan *chan;
678 struct scatterlist *sgl;
679 unsigned int nents;
680 enum dma_transfer_direction dir;
681 unsigned long flags;
682
683 struct dma_async_tx_descriptor *desc;
684 dma_cookie_t cookie;
685
686 if (is_tx) {
687 dir = DMA_MEM_TO_DEV;
688 chan = ctlr->dma_tx;
689 nents = tfr->tx_sg.nents;
690 sgl = tfr->tx_sg.sgl;
691 flags = tfr->rx_buf ? 0 : DMA_PREP_INTERRUPT;
692 } else {
693 dir = DMA_DEV_TO_MEM;
694 chan = ctlr->dma_rx;
695 nents = tfr->rx_sg.nents;
696 sgl = tfr->rx_sg.sgl;
697 flags = DMA_PREP_INTERRUPT;
698 }
699 /* prepare the channel */
700 desc = dmaengine_prep_slave_sg(chan, sgl, nents, dir, flags);
701 if (!desc)
702 return -EINVAL;
703
704 /*
705 * Completion is signaled by the RX channel for bidirectional and
706 * RX-only transfers; else by the TX channel for TX-only transfers.
707 */
708 if (!is_tx) {
709 desc->callback = bcm2835_spi_dma_rx_done;
710 desc->callback_param = ctlr;
711 } else if (!tfr->rx_buf) {
712 desc->callback = bcm2835_spi_dma_tx_done;
713 desc->callback_param = ctlr;
714 bs->target = target;
715 }
716
717 /* submit it to DMA-engine */
718 cookie = dmaengine_submit(desc);
719
720 return dma_submit_error(cookie);
721}
722
723/**
724 * bcm2835_spi_transfer_one_dma() - perform SPI transfer using DMA engine
725 * @ctlr: SPI host controller
726 * @tfr: SPI transfer
727 * @target: BCM2835 SPI target
728 * @cs: CS register
729 *
730 * For *bidirectional* transfers (both tx_buf and rx_buf are non-%NULL), set up
731 * the TX and RX DMA channel to copy between memory and FIFO register.
732 *
733 * For *TX-only* transfers (rx_buf is %NULL), copying the RX FIFO's contents to
734 * memory is pointless. However not reading the RX FIFO isn't an option either
735 * because transmission is halted once it's full. As a workaround, cyclically
736 * clear the RX FIFO by setting the CLEAR_RX bit in the CS register.
737 *
738 * The CS register value is precalculated in bcm2835_spi_setup(). Normally
739 * this is called only once, on target registration. A DMA descriptor to write
740 * this value is preallocated in bcm2835_dma_init(). All that's left to do
741 * when performing a TX-only transfer is to submit this descriptor to the RX
742 * DMA channel. Latency is thereby minimized. The descriptor does not
743 * generate any interrupts while running. It must be terminated once the
744 * TX DMA channel is done.
745 *
746 * Clearing the RX FIFO is paced by the DREQ signal. The signal is asserted
747 * when the RX FIFO becomes half full, i.e. 32 bytes. (Tuneable with the DC
748 * register.) Reading 32 bytes from the RX FIFO would normally require 8 bus
749 * accesses, whereas clearing it requires only 1 bus access. So an 8-fold
750 * reduction in bus traffic and thus energy consumption is achieved.
751 *
752 * For *RX-only* transfers (tx_buf is %NULL), fill the TX FIFO by cyclically
753 * copying from the zero page. The DMA descriptor to do this is preallocated
754 * in bcm2835_dma_init(). It must be terminated once the RX DMA channel is
755 * done and can then be reused.
756 *
757 * The BCM2835 DMA driver autodetects when a transaction copies from the zero
758 * page and utilizes the DMA controller's ability to synthesize zeroes instead
759 * of copying them from memory. This reduces traffic on the memory bus. The
760 * feature is not available on so-called "lite" channels, but normally TX DMA
761 * is backed by a full-featured channel.
762 *
763 * Zero-filling the TX FIFO is paced by the DREQ signal. Unfortunately the
764 * BCM2835 SPI controller continues to assert DREQ even after the DLEN register
765 * has been counted down to zero (hardware erratum). Thus, when the transfer
766 * has finished, the DMA engine zero-fills the TX FIFO until it is half full.
767 * (Tuneable with the DC register.) So up to 9 gratuitous bus accesses are
768 * performed at the end of an RX-only transfer.
769 */
770static int bcm2835_spi_transfer_one_dma(struct spi_controller *ctlr,
771 struct spi_transfer *tfr,
772 struct bcm2835_spidev *target,
773 u32 cs)
774{
775 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
776 dma_cookie_t cookie;
777 int ret;
778
779 /* update usage statistics */
780 bs->count_transfer_dma++;
781
782 /*
783 * Transfer first few bytes without DMA if length of first TX or RX
784 * sglist entry is not a multiple of 4 bytes (hardware limitation).
785 */
786 bcm2835_spi_transfer_prologue(ctlr, tfr, bs, cs);
787
788 /* setup tx-DMA */
789 if (bs->tx_buf) {
790 ret = bcm2835_spi_prepare_sg(ctlr, tfr, bs, target, true);
791 } else {
792 cookie = dmaengine_submit(bs->fill_tx_desc);
793 ret = dma_submit_error(cookie);
794 }
795 if (ret)
796 goto err_reset_hw;
797
798 /* set the DMA length */
799 bcm2835_wr(bs, BCM2835_SPI_DLEN, bs->tx_len);
800
801 /* start the HW */
802 bcm2835_wr(bs, BCM2835_SPI_CS,
803 cs | BCM2835_SPI_CS_TA | BCM2835_SPI_CS_DMAEN);
804
805 bs->tx_dma_active = true;
806 smp_wmb();
807
808 /* start TX early */
809 dma_async_issue_pending(ctlr->dma_tx);
810
811 /* setup rx-DMA late - to run transfers while
812 * mapping of the rx buffers still takes place
813 * this saves 10us or more.
814 */
815 if (bs->rx_buf) {
816 ret = bcm2835_spi_prepare_sg(ctlr, tfr, bs, target, false);
817 } else {
818 cookie = dmaengine_submit(target->clear_rx_desc);
819 ret = dma_submit_error(cookie);
820 }
821 if (ret) {
822 /* need to reset on errors */
823 dmaengine_terminate_sync(ctlr->dma_tx);
824 bs->tx_dma_active = false;
825 goto err_reset_hw;
826 }
827
828 /* start rx dma late */
829 dma_async_issue_pending(ctlr->dma_rx);
830 bs->rx_dma_active = true;
831 smp_mb();
832
833 /*
834 * In case of a very short TX-only transfer, bcm2835_spi_dma_tx_done()
835 * may run before RX DMA is issued. Terminate RX DMA if so.
836 */
837 if (!bs->rx_buf && !bs->tx_dma_active &&
838 cmpxchg(&bs->rx_dma_active, true, false)) {
839 dmaengine_terminate_async(ctlr->dma_rx);
840 bcm2835_spi_reset_hw(bs);
841 }
842
843 /* wait for wakeup in framework */
844 return 1;
845
846err_reset_hw:
847 bcm2835_spi_reset_hw(bs);
848 bcm2835_spi_undo_prologue(bs);
849 return ret;
850}
851
852static bool bcm2835_spi_can_dma(struct spi_controller *ctlr,
853 struct spi_device *spi,
854 struct spi_transfer *tfr)
855{
856 /* we start DMA efforts only on bigger transfers */
857 if (tfr->len < BCM2835_SPI_DMA_MIN_LENGTH)
858 return false;
859
860 /* return OK */
861 return true;
862}
863
864static void bcm2835_dma_release(struct spi_controller *ctlr,
865 struct bcm2835_spi *bs)
866{
867 if (ctlr->dma_tx) {
868 dmaengine_terminate_sync(ctlr->dma_tx);
869
870 if (bs->fill_tx_desc)
871 dmaengine_desc_free(bs->fill_tx_desc);
872
873 if (bs->fill_tx_addr)
874 dma_unmap_page_attrs(ctlr->dma_tx->device->dev,
875 bs->fill_tx_addr, sizeof(u32),
876 DMA_TO_DEVICE,
877 DMA_ATTR_SKIP_CPU_SYNC);
878
879 dma_release_channel(ctlr->dma_tx);
880 ctlr->dma_tx = NULL;
881 }
882
883 if (ctlr->dma_rx) {
884 dmaengine_terminate_sync(ctlr->dma_rx);
885 dma_release_channel(ctlr->dma_rx);
886 ctlr->dma_rx = NULL;
887 }
888}
889
890static int bcm2835_dma_init(struct spi_controller *ctlr, struct device *dev,
891 struct bcm2835_spi *bs)
892{
893 struct dma_slave_config slave_config;
894 const __be32 *addr;
895 dma_addr_t dma_reg_base;
896 int ret;
897
898 /* base address in dma-space */
899 addr = of_get_address(ctlr->dev.of_node, 0, NULL, NULL);
900 if (!addr) {
901 dev_err(dev, "could not get DMA-register address - not using dma mode\n");
902 /* Fall back to interrupt mode */
903 return 0;
904 }
905 dma_reg_base = be32_to_cpup(addr);
906
907 /* get tx/rx dma */
908 ctlr->dma_tx = dma_request_chan(dev, "tx");
909 if (IS_ERR(ctlr->dma_tx)) {
910 ret = dev_err_probe(dev, PTR_ERR(ctlr->dma_tx),
911 "no tx-dma configuration found - not using dma mode\n");
912 ctlr->dma_tx = NULL;
913 goto err;
914 }
915 ctlr->dma_rx = dma_request_chan(dev, "rx");
916 if (IS_ERR(ctlr->dma_rx)) {
917 ret = dev_err_probe(dev, PTR_ERR(ctlr->dma_rx),
918 "no rx-dma configuration found - not using dma mode\n");
919 ctlr->dma_rx = NULL;
920 goto err_release;
921 }
922
923 /*
924 * The TX DMA channel either copies a transfer's TX buffer to the FIFO
925 * or, in case of an RX-only transfer, cyclically copies from the zero
926 * page to the FIFO using a preallocated, reusable descriptor.
927 */
928 slave_config.dst_addr = (u32)(dma_reg_base + BCM2835_SPI_FIFO);
929 slave_config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
930
931 ret = dmaengine_slave_config(ctlr->dma_tx, &slave_config);
932 if (ret)
933 goto err_config;
934
935 bs->fill_tx_addr = dma_map_page_attrs(ctlr->dma_tx->device->dev,
936 ZERO_PAGE(0), 0, sizeof(u32),
937 DMA_TO_DEVICE,
938 DMA_ATTR_SKIP_CPU_SYNC);
939 if (dma_mapping_error(ctlr->dma_tx->device->dev, bs->fill_tx_addr)) {
940 dev_err(dev, "cannot map zero page - not using DMA mode\n");
941 bs->fill_tx_addr = 0;
942 ret = -ENOMEM;
943 goto err_release;
944 }
945
946 bs->fill_tx_desc = dmaengine_prep_dma_cyclic(ctlr->dma_tx,
947 bs->fill_tx_addr,
948 sizeof(u32), 0,
949 DMA_MEM_TO_DEV, 0);
950 if (!bs->fill_tx_desc) {
951 dev_err(dev, "cannot prepare fill_tx_desc - not using DMA mode\n");
952 ret = -ENOMEM;
953 goto err_release;
954 }
955
956 ret = dmaengine_desc_set_reuse(bs->fill_tx_desc);
957 if (ret) {
958 dev_err(dev, "cannot reuse fill_tx_desc - not using DMA mode\n");
959 goto err_release;
960 }
961
962 /*
963 * The RX DMA channel is used bidirectionally: It either reads the
964 * RX FIFO or, in case of a TX-only transfer, cyclically writes a
965 * precalculated value to the CS register to clear the RX FIFO.
966 */
967 slave_config.src_addr = (u32)(dma_reg_base + BCM2835_SPI_FIFO);
968 slave_config.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
969 slave_config.dst_addr = (u32)(dma_reg_base + BCM2835_SPI_CS);
970 slave_config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
971
972 ret = dmaengine_slave_config(ctlr->dma_rx, &slave_config);
973 if (ret)
974 goto err_config;
975
976 /* all went well, so set can_dma */
977 ctlr->can_dma = bcm2835_spi_can_dma;
978
979 return 0;
980
981err_config:
982 dev_err(dev, "issue configuring dma: %d - not using DMA mode\n",
983 ret);
984err_release:
985 bcm2835_dma_release(ctlr, bs);
986err:
987 /*
988 * Only report error for deferred probing, otherwise fall back to
989 * interrupt mode
990 */
991 if (ret != -EPROBE_DEFER)
992 ret = 0;
993
994 return ret;
995}
996
997static int bcm2835_spi_transfer_one_poll(struct spi_controller *ctlr,
998 struct spi_device *spi,
999 struct spi_transfer *tfr,
1000 u32 cs)
1001{
1002 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1003 unsigned long timeout;
1004
1005 /* update usage statistics */
1006 bs->count_transfer_polling++;
1007
1008 /* enable HW block without interrupts */
1009 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_TA);
1010
1011 /* fill in the fifo before timeout calculations
1012 * if we are interrupted here, then the data is
1013 * getting transferred by the HW while we are interrupted
1014 */
1015 bcm2835_wr_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE);
1016
1017 /* set the timeout to at least 2 jiffies */
1018 timeout = jiffies + 2 + HZ * polling_limit_us / 1000000;
1019
1020 /* loop until finished the transfer */
1021 while (bs->rx_len) {
1022 /* fill in tx fifo with remaining data */
1023 bcm2835_wr_fifo(bs);
1024
1025 /* read from fifo as much as possible */
1026 bcm2835_rd_fifo(bs);
1027
1028 /* if there is still data pending to read
1029 * then check the timeout
1030 */
1031 if (bs->rx_len && time_after(jiffies, timeout)) {
1032 dev_dbg_ratelimited(&spi->dev,
1033 "timeout period reached: jiffies: %lu remaining tx/rx: %d/%d - falling back to interrupt mode\n",
1034 jiffies - timeout,
1035 bs->tx_len, bs->rx_len);
1036 /* fall back to interrupt mode */
1037
1038 /* update usage statistics */
1039 bs->count_transfer_irq_after_polling++;
1040
1041 return bcm2835_spi_transfer_one_irq(ctlr, spi,
1042 tfr, cs, false);
1043 }
1044 }
1045
1046 /* Transfer complete - reset SPI HW */
1047 bcm2835_spi_reset_hw(bs);
1048 /* and return without waiting for completion */
1049 return 0;
1050}
1051
1052static int bcm2835_spi_transfer_one(struct spi_controller *ctlr,
1053 struct spi_device *spi,
1054 struct spi_transfer *tfr)
1055{
1056 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1057 struct bcm2835_spidev *target = spi_get_ctldata(spi);
1058 unsigned long spi_hz, cdiv;
1059 unsigned long hz_per_byte, byte_limit;
1060 u32 cs = target->prepare_cs;
1061
1062 /* set clock */
1063 spi_hz = tfr->speed_hz;
1064
1065 if (spi_hz >= bs->clk_hz / 2) {
1066 cdiv = 2; /* clk_hz/2 is the fastest we can go */
1067 } else if (spi_hz) {
1068 /* CDIV must be a multiple of two */
1069 cdiv = DIV_ROUND_UP(bs->clk_hz, spi_hz);
1070 cdiv += (cdiv % 2);
1071
1072 if (cdiv >= 65536)
1073 cdiv = 0; /* 0 is the slowest we can go */
1074 } else {
1075 cdiv = 0; /* 0 is the slowest we can go */
1076 }
1077 tfr->effective_speed_hz = cdiv ? (bs->clk_hz / cdiv) : (bs->clk_hz / 65536);
1078 bcm2835_wr(bs, BCM2835_SPI_CLK, cdiv);
1079
1080 /* handle all the 3-wire mode */
1081 if (spi->mode & SPI_3WIRE && tfr->rx_buf)
1082 cs |= BCM2835_SPI_CS_REN;
1083
1084 /* set transmit buffers and length */
1085 bs->tx_buf = tfr->tx_buf;
1086 bs->rx_buf = tfr->rx_buf;
1087 bs->tx_len = tfr->len;
1088 bs->rx_len = tfr->len;
1089
1090 /* Calculate the estimated time in us the transfer runs. Note that
1091 * there is 1 idle clocks cycles after each byte getting transferred
1092 * so we have 9 cycles/byte. This is used to find the number of Hz
1093 * per byte per polling limit. E.g., we can transfer 1 byte in 30 us
1094 * per 300,000 Hz of bus clock.
1095 */
1096 hz_per_byte = polling_limit_us ? (9 * 1000000) / polling_limit_us : 0;
1097 byte_limit = hz_per_byte ? tfr->effective_speed_hz / hz_per_byte : 1;
1098
1099 /* run in polling mode for short transfers */
1100 if (tfr->len < byte_limit)
1101 return bcm2835_spi_transfer_one_poll(ctlr, spi, tfr, cs);
1102
1103 /* run in dma mode if conditions are right
1104 * Note that unlike poll or interrupt mode DMA mode does not have
1105 * this 1 idle clock cycle pattern but runs the spi clock without gaps
1106 */
1107 if (ctlr->can_dma && bcm2835_spi_can_dma(ctlr, spi, tfr))
1108 return bcm2835_spi_transfer_one_dma(ctlr, tfr, target, cs);
1109
1110 /* run in interrupt-mode */
1111 return bcm2835_spi_transfer_one_irq(ctlr, spi, tfr, cs, true);
1112}
1113
1114static int bcm2835_spi_prepare_message(struct spi_controller *ctlr,
1115 struct spi_message *msg)
1116{
1117 struct spi_device *spi = msg->spi;
1118 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1119 struct bcm2835_spidev *target = spi_get_ctldata(spi);
1120 int ret;
1121
1122 if (ctlr->can_dma) {
1123 /*
1124 * DMA transfers are limited to 16 bit (0 to 65535 bytes) by
1125 * the SPI HW due to DLEN. Split up transfers (32-bit FIFO
1126 * aligned) if the limit is exceeded.
1127 */
1128 ret = spi_split_transfers_maxsize(ctlr, msg, 65532,
1129 GFP_KERNEL | GFP_DMA);
1130 if (ret)
1131 return ret;
1132 }
1133
1134 /*
1135 * Set up clock polarity before spi_transfer_one_message() asserts
1136 * chip select to avoid a gratuitous clock signal edge.
1137 */
1138 bcm2835_wr(bs, BCM2835_SPI_CS, target->prepare_cs);
1139
1140 return 0;
1141}
1142
1143static void bcm2835_spi_handle_err(struct spi_controller *ctlr,
1144 struct spi_message *msg)
1145{
1146 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1147
1148 /* if an error occurred and we have an active dma, then terminate */
1149 if (ctlr->dma_tx) {
1150 dmaengine_terminate_sync(ctlr->dma_tx);
1151 bs->tx_dma_active = false;
1152 }
1153 if (ctlr->dma_rx) {
1154 dmaengine_terminate_sync(ctlr->dma_rx);
1155 bs->rx_dma_active = false;
1156 }
1157 bcm2835_spi_undo_prologue(bs);
1158
1159 /* and reset */
1160 bcm2835_spi_reset_hw(bs);
1161}
1162
1163static void bcm2835_spi_cleanup(struct spi_device *spi)
1164{
1165 struct bcm2835_spidev *target = spi_get_ctldata(spi);
1166 struct spi_controller *ctlr = spi->controller;
1167 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1168
1169 if (target->clear_rx_desc)
1170 dmaengine_desc_free(target->clear_rx_desc);
1171
1172 if (target->clear_rx_addr)
1173 dma_unmap_single(ctlr->dma_rx->device->dev,
1174 target->clear_rx_addr,
1175 sizeof(u32),
1176 DMA_TO_DEVICE);
1177
1178 gpiod_put(bs->cs_gpio);
1179 spi_set_csgpiod(spi, 0, NULL);
1180
1181 kfree(target);
1182}
1183
1184static int bcm2835_spi_setup_dma(struct spi_controller *ctlr,
1185 struct spi_device *spi,
1186 struct bcm2835_spi *bs,
1187 struct bcm2835_spidev *target)
1188{
1189 int ret;
1190
1191 if (!ctlr->dma_rx)
1192 return 0;
1193
1194 target->clear_rx_addr = dma_map_single(ctlr->dma_rx->device->dev,
1195 &target->clear_rx_cs,
1196 sizeof(u32),
1197 DMA_TO_DEVICE);
1198 if (dma_mapping_error(ctlr->dma_rx->device->dev, target->clear_rx_addr)) {
1199 dev_err(&spi->dev, "cannot map clear_rx_cs\n");
1200 target->clear_rx_addr = 0;
1201 return -ENOMEM;
1202 }
1203
1204 target->clear_rx_desc = dmaengine_prep_dma_cyclic(ctlr->dma_rx,
1205 target->clear_rx_addr,
1206 sizeof(u32), 0,
1207 DMA_MEM_TO_DEV, 0);
1208 if (!target->clear_rx_desc) {
1209 dev_err(&spi->dev, "cannot prepare clear_rx_desc\n");
1210 return -ENOMEM;
1211 }
1212
1213 ret = dmaengine_desc_set_reuse(target->clear_rx_desc);
1214 if (ret) {
1215 dev_err(&spi->dev, "cannot reuse clear_rx_desc\n");
1216 return ret;
1217 }
1218
1219 return 0;
1220}
1221
1222static int bcm2835_spi_setup(struct spi_device *spi)
1223{
1224 struct spi_controller *ctlr = spi->controller;
1225 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1226 struct bcm2835_spidev *target = spi_get_ctldata(spi);
1227 struct gpiod_lookup_table *lookup __free(kfree) = NULL;
1228 int ret;
1229 u32 cs;
1230
1231 if (!target) {
1232 target = kzalloc(ALIGN(sizeof(*target), dma_get_cache_alignment()),
1233 GFP_KERNEL);
1234 if (!target)
1235 return -ENOMEM;
1236
1237 spi_set_ctldata(spi, target);
1238
1239 ret = bcm2835_spi_setup_dma(ctlr, spi, bs, target);
1240 if (ret)
1241 goto err_cleanup;
1242 }
1243
1244 /*
1245 * Precalculate SPI target's CS register value for ->prepare_message():
1246 * The driver always uses software-controlled GPIO chip select, hence
1247 * set the hardware-controlled native chip select to an invalid value
1248 * to prevent it from interfering.
1249 */
1250 cs = BCM2835_SPI_CS_CS_10 | BCM2835_SPI_CS_CS_01;
1251 if (spi->mode & SPI_CPOL)
1252 cs |= BCM2835_SPI_CS_CPOL;
1253 if (spi->mode & SPI_CPHA)
1254 cs |= BCM2835_SPI_CS_CPHA;
1255 target->prepare_cs = cs;
1256
1257 /*
1258 * Precalculate SPI target's CS register value to clear RX FIFO
1259 * in case of a TX-only DMA transfer.
1260 */
1261 if (ctlr->dma_rx) {
1262 target->clear_rx_cs = cs | BCM2835_SPI_CS_TA |
1263 BCM2835_SPI_CS_DMAEN |
1264 BCM2835_SPI_CS_CLEAR_RX;
1265 dma_sync_single_for_device(ctlr->dma_rx->device->dev,
1266 target->clear_rx_addr,
1267 sizeof(u32),
1268 DMA_TO_DEVICE);
1269 }
1270
1271 /*
1272 * sanity checking the native-chipselects
1273 */
1274 if (spi->mode & SPI_NO_CS)
1275 return 0;
1276 /*
1277 * The SPI core has successfully requested the CS GPIO line from the
1278 * device tree, so we are done.
1279 */
1280 if (spi_get_csgpiod(spi, 0))
1281 return 0;
1282 if (spi_get_chipselect(spi, 0) > 1) {
1283 /* error in the case of native CS requested with CS > 1
1284 * officially there is a CS2, but it is not documented
1285 * which GPIO is connected with that...
1286 */
1287 dev_err(&spi->dev,
1288 "setup: only two native chip-selects are supported\n");
1289 ret = -EINVAL;
1290 goto err_cleanup;
1291 }
1292
1293 /*
1294 * TODO: The code below is a slightly better alternative to the utter
1295 * abuse of the GPIO API that I found here before. It creates a
1296 * temporary lookup table, assigns it to the SPI device, gets the GPIO
1297 * descriptor and then releases the lookup table.
1298 *
1299 * More on the problem that it addresses:
1300 * https://www.spinics.net/lists/linux-gpio/msg36218.html
1301 */
1302 lookup = kzalloc(struct_size(lookup, table, 2), GFP_KERNEL);
1303 if (!lookup) {
1304 ret = -ENOMEM;
1305 goto err_cleanup;
1306 }
1307
1308 lookup->dev_id = dev_name(&spi->dev);
1309 lookup->table[0] = GPIO_LOOKUP("pinctrl-bcm2835",
1310 8 - (spi_get_chipselect(spi, 0)),
1311 "cs", GPIO_LOOKUP_FLAGS_DEFAULT);
1312
1313 gpiod_add_lookup_table(lookup);
1314
1315 bs->cs_gpio = gpiod_get(&spi->dev, "cs", GPIOD_OUT_LOW);
1316 gpiod_remove_lookup_table(lookup);
1317 if (IS_ERR(bs->cs_gpio)) {
1318 ret = PTR_ERR(bs->cs_gpio);
1319 goto err_cleanup;
1320 }
1321
1322 spi_set_csgpiod(spi, 0, bs->cs_gpio);
1323
1324 /* and set up the "mode" and level */
1325 dev_info(&spi->dev, "setting up native-CS%i to use GPIO\n",
1326 spi_get_chipselect(spi, 0));
1327
1328 return 0;
1329
1330err_cleanup:
1331 bcm2835_spi_cleanup(spi);
1332 return ret;
1333}
1334
1335static int bcm2835_spi_probe(struct platform_device *pdev)
1336{
1337 struct spi_controller *ctlr;
1338 struct bcm2835_spi *bs;
1339 int err;
1340
1341 ctlr = devm_spi_alloc_host(&pdev->dev, sizeof(*bs));
1342 if (!ctlr)
1343 return -ENOMEM;
1344
1345 platform_set_drvdata(pdev, ctlr);
1346
1347 ctlr->use_gpio_descriptors = true;
1348 ctlr->mode_bits = BCM2835_SPI_MODE_BITS;
1349 ctlr->bits_per_word_mask = SPI_BPW_MASK(8);
1350 ctlr->num_chipselect = 3;
1351 ctlr->setup = bcm2835_spi_setup;
1352 ctlr->cleanup = bcm2835_spi_cleanup;
1353 ctlr->transfer_one = bcm2835_spi_transfer_one;
1354 ctlr->handle_err = bcm2835_spi_handle_err;
1355 ctlr->prepare_message = bcm2835_spi_prepare_message;
1356 ctlr->dev.of_node = pdev->dev.of_node;
1357
1358 bs = spi_controller_get_devdata(ctlr);
1359 bs->ctlr = ctlr;
1360
1361 bs->regs = devm_platform_ioremap_resource(pdev, 0);
1362 if (IS_ERR(bs->regs))
1363 return PTR_ERR(bs->regs);
1364
1365 bs->clk = devm_clk_get_enabled(&pdev->dev, NULL);
1366 if (IS_ERR(bs->clk))
1367 return dev_err_probe(&pdev->dev, PTR_ERR(bs->clk),
1368 "could not get clk\n");
1369
1370 ctlr->max_speed_hz = clk_get_rate(bs->clk) / 2;
1371
1372 bs->irq = platform_get_irq(pdev, 0);
1373 if (bs->irq < 0)
1374 return bs->irq;
1375
1376 bs->clk_hz = clk_get_rate(bs->clk);
1377
1378 err = bcm2835_dma_init(ctlr, &pdev->dev, bs);
1379 if (err)
1380 return err;
1381
1382 /* initialise the hardware with the default polarities */
1383 bcm2835_wr(bs, BCM2835_SPI_CS,
1384 BCM2835_SPI_CS_CLEAR_RX | BCM2835_SPI_CS_CLEAR_TX);
1385
1386 err = devm_request_irq(&pdev->dev, bs->irq, bcm2835_spi_interrupt,
1387 IRQF_SHARED, dev_name(&pdev->dev), bs);
1388 if (err) {
1389 dev_err(&pdev->dev, "could not request IRQ: %d\n", err);
1390 goto out_dma_release;
1391 }
1392
1393 err = spi_register_controller(ctlr);
1394 if (err) {
1395 dev_err(&pdev->dev, "could not register SPI controller: %d\n",
1396 err);
1397 goto out_dma_release;
1398 }
1399
1400 bcm2835_debugfs_create(bs, dev_name(&pdev->dev));
1401
1402 return 0;
1403
1404out_dma_release:
1405 bcm2835_dma_release(ctlr, bs);
1406 return err;
1407}
1408
1409static void bcm2835_spi_remove(struct platform_device *pdev)
1410{
1411 struct spi_controller *ctlr = platform_get_drvdata(pdev);
1412 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1413
1414 bcm2835_debugfs_remove(bs);
1415
1416 spi_unregister_controller(ctlr);
1417
1418 bcm2835_dma_release(ctlr, bs);
1419
1420 /* Clear FIFOs, and disable the HW block */
1421 bcm2835_wr(bs, BCM2835_SPI_CS,
1422 BCM2835_SPI_CS_CLEAR_RX | BCM2835_SPI_CS_CLEAR_TX);
1423}
1424
1425static const struct of_device_id bcm2835_spi_match[] = {
1426 { .compatible = "brcm,bcm2835-spi", },
1427 {}
1428};
1429MODULE_DEVICE_TABLE(of, bcm2835_spi_match);
1430
1431static struct platform_driver bcm2835_spi_driver = {
1432 .driver = {
1433 .name = DRV_NAME,
1434 .of_match_table = bcm2835_spi_match,
1435 },
1436 .probe = bcm2835_spi_probe,
1437 .remove_new = bcm2835_spi_remove,
1438 .shutdown = bcm2835_spi_remove,
1439};
1440module_platform_driver(bcm2835_spi_driver);
1441
1442MODULE_DESCRIPTION("SPI controller driver for Broadcom BCM2835");
1443MODULE_AUTHOR("Chris Boot <bootc@bootc.net>");
1444MODULE_LICENSE("GPL");
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Driver for Broadcom BCM2835 SPI Controllers
4 *
5 * Copyright (C) 2012 Chris Boot
6 * Copyright (C) 2013 Stephen Warren
7 * Copyright (C) 2015 Martin Sperl
8 *
9 * This driver is inspired by:
10 * spi-ath79.c, Copyright (C) 2009-2011 Gabor Juhos <juhosg@openwrt.org>
11 * spi-atmel.c, Copyright (C) 2006 Atmel Corporation
12 */
13
14#include <linux/clk.h>
15#include <linux/completion.h>
16#include <linux/debugfs.h>
17#include <linux/delay.h>
18#include <linux/dma-mapping.h>
19#include <linux/dmaengine.h>
20#include <linux/err.h>
21#include <linux/interrupt.h>
22#include <linux/io.h>
23#include <linux/kernel.h>
24#include <linux/module.h>
25#include <linux/of.h>
26#include <linux/of_address.h>
27#include <linux/of_device.h>
28#include <linux/gpio/consumer.h>
29#include <linux/gpio/machine.h> /* FIXME: using chip internals */
30#include <linux/gpio/driver.h> /* FIXME: using chip internals */
31#include <linux/of_irq.h>
32#include <linux/spi/spi.h>
33
34/* SPI register offsets */
35#define BCM2835_SPI_CS 0x00
36#define BCM2835_SPI_FIFO 0x04
37#define BCM2835_SPI_CLK 0x08
38#define BCM2835_SPI_DLEN 0x0c
39#define BCM2835_SPI_LTOH 0x10
40#define BCM2835_SPI_DC 0x14
41
42/* Bitfields in CS */
43#define BCM2835_SPI_CS_LEN_LONG 0x02000000
44#define BCM2835_SPI_CS_DMA_LEN 0x01000000
45#define BCM2835_SPI_CS_CSPOL2 0x00800000
46#define BCM2835_SPI_CS_CSPOL1 0x00400000
47#define BCM2835_SPI_CS_CSPOL0 0x00200000
48#define BCM2835_SPI_CS_RXF 0x00100000
49#define BCM2835_SPI_CS_RXR 0x00080000
50#define BCM2835_SPI_CS_TXD 0x00040000
51#define BCM2835_SPI_CS_RXD 0x00020000
52#define BCM2835_SPI_CS_DONE 0x00010000
53#define BCM2835_SPI_CS_LEN 0x00002000
54#define BCM2835_SPI_CS_REN 0x00001000
55#define BCM2835_SPI_CS_ADCS 0x00000800
56#define BCM2835_SPI_CS_INTR 0x00000400
57#define BCM2835_SPI_CS_INTD 0x00000200
58#define BCM2835_SPI_CS_DMAEN 0x00000100
59#define BCM2835_SPI_CS_TA 0x00000080
60#define BCM2835_SPI_CS_CSPOL 0x00000040
61#define BCM2835_SPI_CS_CLEAR_RX 0x00000020
62#define BCM2835_SPI_CS_CLEAR_TX 0x00000010
63#define BCM2835_SPI_CS_CPOL 0x00000008
64#define BCM2835_SPI_CS_CPHA 0x00000004
65#define BCM2835_SPI_CS_CS_10 0x00000002
66#define BCM2835_SPI_CS_CS_01 0x00000001
67
68#define BCM2835_SPI_FIFO_SIZE 64
69#define BCM2835_SPI_FIFO_SIZE_3_4 48
70#define BCM2835_SPI_DMA_MIN_LENGTH 96
71#define BCM2835_SPI_NUM_CS 3 /* raise as necessary */
72#define BCM2835_SPI_MODE_BITS (SPI_CPOL | SPI_CPHA | SPI_CS_HIGH \
73 | SPI_NO_CS | SPI_3WIRE)
74
75#define DRV_NAME "spi-bcm2835"
76
77/* define polling limits */
78unsigned int polling_limit_us = 30;
79module_param(polling_limit_us, uint, 0664);
80MODULE_PARM_DESC(polling_limit_us,
81 "time in us to run a transfer in polling mode\n");
82
83/**
84 * struct bcm2835_spi - BCM2835 SPI controller
85 * @regs: base address of register map
86 * @clk: core clock, divided to calculate serial clock
87 * @irq: interrupt, signals TX FIFO empty or RX FIFO ¾ full
88 * @tfr: SPI transfer currently processed
89 * @tx_buf: pointer whence next transmitted byte is read
90 * @rx_buf: pointer where next received byte is written
91 * @tx_len: remaining bytes to transmit
92 * @rx_len: remaining bytes to receive
93 * @tx_prologue: bytes transmitted without DMA if first TX sglist entry's
94 * length is not a multiple of 4 (to overcome hardware limitation)
95 * @rx_prologue: bytes received without DMA if first RX sglist entry's
96 * length is not a multiple of 4 (to overcome hardware limitation)
97 * @tx_spillover: whether @tx_prologue spills over to second TX sglist entry
98 * @prepare_cs: precalculated CS register value for ->prepare_message()
99 * (uses slave-specific clock polarity and phase settings)
100 * @debugfs_dir: the debugfs directory - neede to remove debugfs when
101 * unloading the module
102 * @count_transfer_polling: count of how often polling mode is used
103 * @count_transfer_irq: count of how often interrupt mode is used
104 * @count_transfer_irq_after_polling: count of how often we fall back to
105 * interrupt mode after starting in polling mode.
106 * These are counted as well in @count_transfer_polling and
107 * @count_transfer_irq
108 * @count_transfer_dma: count how often dma mode is used
109 * @chip_select: SPI slave currently selected
110 * (used by bcm2835_spi_dma_tx_done() to write @clear_rx_cs)
111 * @tx_dma_active: whether a TX DMA descriptor is in progress
112 * @rx_dma_active: whether a RX DMA descriptor is in progress
113 * (used by bcm2835_spi_dma_tx_done() to handle a race)
114 * @fill_tx_desc: preallocated TX DMA descriptor used for RX-only transfers
115 * (cyclically copies from zero page to TX FIFO)
116 * @fill_tx_addr: bus address of zero page
117 * @clear_rx_desc: preallocated RX DMA descriptor used for TX-only transfers
118 * (cyclically clears RX FIFO by writing @clear_rx_cs to CS register)
119 * @clear_rx_addr: bus address of @clear_rx_cs
120 * @clear_rx_cs: precalculated CS register value to clear RX FIFO
121 * (uses slave-specific clock polarity and phase settings)
122 */
123struct bcm2835_spi {
124 void __iomem *regs;
125 struct clk *clk;
126 int irq;
127 struct spi_transfer *tfr;
128 const u8 *tx_buf;
129 u8 *rx_buf;
130 int tx_len;
131 int rx_len;
132 int tx_prologue;
133 int rx_prologue;
134 unsigned int tx_spillover;
135 u32 prepare_cs[BCM2835_SPI_NUM_CS];
136
137 struct dentry *debugfs_dir;
138 u64 count_transfer_polling;
139 u64 count_transfer_irq;
140 u64 count_transfer_irq_after_polling;
141 u64 count_transfer_dma;
142
143 u8 chip_select;
144 unsigned int tx_dma_active;
145 unsigned int rx_dma_active;
146 struct dma_async_tx_descriptor *fill_tx_desc;
147 dma_addr_t fill_tx_addr;
148 struct dma_async_tx_descriptor *clear_rx_desc[BCM2835_SPI_NUM_CS];
149 dma_addr_t clear_rx_addr;
150 u32 clear_rx_cs[BCM2835_SPI_NUM_CS] ____cacheline_aligned;
151};
152
153#if defined(CONFIG_DEBUG_FS)
154static void bcm2835_debugfs_create(struct bcm2835_spi *bs,
155 const char *dname)
156{
157 char name[64];
158 struct dentry *dir;
159
160 /* get full name */
161 snprintf(name, sizeof(name), "spi-bcm2835-%s", dname);
162
163 /* the base directory */
164 dir = debugfs_create_dir(name, NULL);
165 bs->debugfs_dir = dir;
166
167 /* the counters */
168 debugfs_create_u64("count_transfer_polling", 0444, dir,
169 &bs->count_transfer_polling);
170 debugfs_create_u64("count_transfer_irq", 0444, dir,
171 &bs->count_transfer_irq);
172 debugfs_create_u64("count_transfer_irq_after_polling", 0444, dir,
173 &bs->count_transfer_irq_after_polling);
174 debugfs_create_u64("count_transfer_dma", 0444, dir,
175 &bs->count_transfer_dma);
176}
177
178static void bcm2835_debugfs_remove(struct bcm2835_spi *bs)
179{
180 debugfs_remove_recursive(bs->debugfs_dir);
181 bs->debugfs_dir = NULL;
182}
183#else
184static void bcm2835_debugfs_create(struct bcm2835_spi *bs,
185 const char *dname)
186{
187}
188
189static void bcm2835_debugfs_remove(struct bcm2835_spi *bs)
190{
191}
192#endif /* CONFIG_DEBUG_FS */
193
194static inline u32 bcm2835_rd(struct bcm2835_spi *bs, unsigned reg)
195{
196 return readl(bs->regs + reg);
197}
198
199static inline void bcm2835_wr(struct bcm2835_spi *bs, unsigned reg, u32 val)
200{
201 writel(val, bs->regs + reg);
202}
203
204static inline void bcm2835_rd_fifo(struct bcm2835_spi *bs)
205{
206 u8 byte;
207
208 while ((bs->rx_len) &&
209 (bcm2835_rd(bs, BCM2835_SPI_CS) & BCM2835_SPI_CS_RXD)) {
210 byte = bcm2835_rd(bs, BCM2835_SPI_FIFO);
211 if (bs->rx_buf)
212 *bs->rx_buf++ = byte;
213 bs->rx_len--;
214 }
215}
216
217static inline void bcm2835_wr_fifo(struct bcm2835_spi *bs)
218{
219 u8 byte;
220
221 while ((bs->tx_len) &&
222 (bcm2835_rd(bs, BCM2835_SPI_CS) & BCM2835_SPI_CS_TXD)) {
223 byte = bs->tx_buf ? *bs->tx_buf++ : 0;
224 bcm2835_wr(bs, BCM2835_SPI_FIFO, byte);
225 bs->tx_len--;
226 }
227}
228
229/**
230 * bcm2835_rd_fifo_count() - blindly read exactly @count bytes from RX FIFO
231 * @bs: BCM2835 SPI controller
232 * @count: bytes to read from RX FIFO
233 *
234 * The caller must ensure that @bs->rx_len is greater than or equal to @count,
235 * that the RX FIFO contains at least @count bytes and that the DMA Enable flag
236 * in the CS register is set (such that a read from the FIFO register receives
237 * 32-bit instead of just 8-bit). Moreover @bs->rx_buf must not be %NULL.
238 */
239static inline void bcm2835_rd_fifo_count(struct bcm2835_spi *bs, int count)
240{
241 u32 val;
242 int len;
243
244 bs->rx_len -= count;
245
246 while (count > 0) {
247 val = bcm2835_rd(bs, BCM2835_SPI_FIFO);
248 len = min(count, 4);
249 memcpy(bs->rx_buf, &val, len);
250 bs->rx_buf += len;
251 count -= 4;
252 }
253}
254
255/**
256 * bcm2835_wr_fifo_count() - blindly write exactly @count bytes to TX FIFO
257 * @bs: BCM2835 SPI controller
258 * @count: bytes to write to TX FIFO
259 *
260 * The caller must ensure that @bs->tx_len is greater than or equal to @count,
261 * that the TX FIFO can accommodate @count bytes and that the DMA Enable flag
262 * in the CS register is set (such that a write to the FIFO register transmits
263 * 32-bit instead of just 8-bit).
264 */
265static inline void bcm2835_wr_fifo_count(struct bcm2835_spi *bs, int count)
266{
267 u32 val;
268 int len;
269
270 bs->tx_len -= count;
271
272 while (count > 0) {
273 if (bs->tx_buf) {
274 len = min(count, 4);
275 memcpy(&val, bs->tx_buf, len);
276 bs->tx_buf += len;
277 } else {
278 val = 0;
279 }
280 bcm2835_wr(bs, BCM2835_SPI_FIFO, val);
281 count -= 4;
282 }
283}
284
285/**
286 * bcm2835_wait_tx_fifo_empty() - busy-wait for TX FIFO to empty
287 * @bs: BCM2835 SPI controller
288 *
289 * The caller must ensure that the RX FIFO can accommodate as many bytes
290 * as have been written to the TX FIFO: Transmission is halted once the
291 * RX FIFO is full, causing this function to spin forever.
292 */
293static inline void bcm2835_wait_tx_fifo_empty(struct bcm2835_spi *bs)
294{
295 while (!(bcm2835_rd(bs, BCM2835_SPI_CS) & BCM2835_SPI_CS_DONE))
296 cpu_relax();
297}
298
299/**
300 * bcm2835_rd_fifo_blind() - blindly read up to @count bytes from RX FIFO
301 * @bs: BCM2835 SPI controller
302 * @count: bytes available for reading in RX FIFO
303 */
304static inline void bcm2835_rd_fifo_blind(struct bcm2835_spi *bs, int count)
305{
306 u8 val;
307
308 count = min(count, bs->rx_len);
309 bs->rx_len -= count;
310
311 while (count) {
312 val = bcm2835_rd(bs, BCM2835_SPI_FIFO);
313 if (bs->rx_buf)
314 *bs->rx_buf++ = val;
315 count--;
316 }
317}
318
319/**
320 * bcm2835_wr_fifo_blind() - blindly write up to @count bytes to TX FIFO
321 * @bs: BCM2835 SPI controller
322 * @count: bytes available for writing in TX FIFO
323 */
324static inline void bcm2835_wr_fifo_blind(struct bcm2835_spi *bs, int count)
325{
326 u8 val;
327
328 count = min(count, bs->tx_len);
329 bs->tx_len -= count;
330
331 while (count) {
332 val = bs->tx_buf ? *bs->tx_buf++ : 0;
333 bcm2835_wr(bs, BCM2835_SPI_FIFO, val);
334 count--;
335 }
336}
337
338static void bcm2835_spi_reset_hw(struct spi_controller *ctlr)
339{
340 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
341 u32 cs = bcm2835_rd(bs, BCM2835_SPI_CS);
342
343 /* Disable SPI interrupts and transfer */
344 cs &= ~(BCM2835_SPI_CS_INTR |
345 BCM2835_SPI_CS_INTD |
346 BCM2835_SPI_CS_DMAEN |
347 BCM2835_SPI_CS_TA);
348 /*
349 * Transmission sometimes breaks unless the DONE bit is written at the
350 * end of every transfer. The spec says it's a RO bit. Either the
351 * spec is wrong and the bit is actually of type RW1C, or it's a
352 * hardware erratum.
353 */
354 cs |= BCM2835_SPI_CS_DONE;
355 /* and reset RX/TX FIFOS */
356 cs |= BCM2835_SPI_CS_CLEAR_RX | BCM2835_SPI_CS_CLEAR_TX;
357
358 /* and reset the SPI_HW */
359 bcm2835_wr(bs, BCM2835_SPI_CS, cs);
360 /* as well as DLEN */
361 bcm2835_wr(bs, BCM2835_SPI_DLEN, 0);
362}
363
364static irqreturn_t bcm2835_spi_interrupt(int irq, void *dev_id)
365{
366 struct spi_controller *ctlr = dev_id;
367 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
368 u32 cs = bcm2835_rd(bs, BCM2835_SPI_CS);
369
370 /*
371 * An interrupt is signaled either if DONE is set (TX FIFO empty)
372 * or if RXR is set (RX FIFO >= ¾ full).
373 */
374 if (cs & BCM2835_SPI_CS_RXF)
375 bcm2835_rd_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE);
376 else if (cs & BCM2835_SPI_CS_RXR)
377 bcm2835_rd_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE_3_4);
378
379 if (bs->tx_len && cs & BCM2835_SPI_CS_DONE)
380 bcm2835_wr_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE);
381
382 /* Read as many bytes as possible from FIFO */
383 bcm2835_rd_fifo(bs);
384 /* Write as many bytes as possible to FIFO */
385 bcm2835_wr_fifo(bs);
386
387 if (!bs->rx_len) {
388 /* Transfer complete - reset SPI HW */
389 bcm2835_spi_reset_hw(ctlr);
390 /* wake up the framework */
391 complete(&ctlr->xfer_completion);
392 }
393
394 return IRQ_HANDLED;
395}
396
397static int bcm2835_spi_transfer_one_irq(struct spi_controller *ctlr,
398 struct spi_device *spi,
399 struct spi_transfer *tfr,
400 u32 cs, bool fifo_empty)
401{
402 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
403
404 /* update usage statistics */
405 bs->count_transfer_irq++;
406
407 /*
408 * Enable HW block, but with interrupts still disabled.
409 * Otherwise the empty TX FIFO would immediately trigger an interrupt.
410 */
411 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_TA);
412
413 /* fill TX FIFO as much as possible */
414 if (fifo_empty)
415 bcm2835_wr_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE);
416 bcm2835_wr_fifo(bs);
417
418 /* enable interrupts */
419 cs |= BCM2835_SPI_CS_INTR | BCM2835_SPI_CS_INTD | BCM2835_SPI_CS_TA;
420 bcm2835_wr(bs, BCM2835_SPI_CS, cs);
421
422 /* signal that we need to wait for completion */
423 return 1;
424}
425
426/**
427 * bcm2835_spi_transfer_prologue() - transfer first few bytes without DMA
428 * @ctlr: SPI master controller
429 * @tfr: SPI transfer
430 * @bs: BCM2835 SPI controller
431 * @cs: CS register
432 *
433 * A limitation in DMA mode is that the FIFO must be accessed in 4 byte chunks.
434 * Only the final write access is permitted to transmit less than 4 bytes, the
435 * SPI controller deduces its intended size from the DLEN register.
436 *
437 * If a TX or RX sglist contains multiple entries, one per page, and the first
438 * entry starts in the middle of a page, that first entry's length may not be
439 * a multiple of 4. Subsequent entries are fine because they span an entire
440 * page, hence do have a length that's a multiple of 4.
441 *
442 * This cannot happen with kmalloc'ed buffers (which is what most clients use)
443 * because they are contiguous in physical memory and therefore not split on
444 * page boundaries by spi_map_buf(). But it *can* happen with vmalloc'ed
445 * buffers.
446 *
447 * The DMA engine is incapable of combining sglist entries into a continuous
448 * stream of 4 byte chunks, it treats every entry separately: A TX entry is
449 * rounded up a to a multiple of 4 bytes by transmitting surplus bytes, an RX
450 * entry is rounded up by throwing away received bytes.
451 *
452 * Overcome this limitation by transferring the first few bytes without DMA:
453 * E.g. if the first TX sglist entry's length is 23 and the first RX's is 42,
454 * write 3 bytes to the TX FIFO but read only 2 bytes from the RX FIFO.
455 * The residue of 1 byte in the RX FIFO is picked up by DMA. Together with
456 * the rest of the first RX sglist entry it makes up a multiple of 4 bytes.
457 *
458 * Should the RX prologue be larger, say, 3 vis-à-vis a TX prologue of 1,
459 * write 1 + 4 = 5 bytes to the TX FIFO and read 3 bytes from the RX FIFO.
460 * Caution, the additional 4 bytes spill over to the second TX sglist entry
461 * if the length of the first is *exactly* 1.
462 *
463 * At most 6 bytes are written and at most 3 bytes read. Do we know the
464 * transfer has this many bytes? Yes, see BCM2835_SPI_DMA_MIN_LENGTH.
465 *
466 * The FIFO is normally accessed with 8-bit width by the CPU and 32-bit width
467 * by the DMA engine. Toggling the DMA Enable flag in the CS register switches
468 * the width but also garbles the FIFO's contents. The prologue must therefore
469 * be transmitted in 32-bit width to ensure that the following DMA transfer can
470 * pick up the residue in the RX FIFO in ungarbled form.
471 */
472static void bcm2835_spi_transfer_prologue(struct spi_controller *ctlr,
473 struct spi_transfer *tfr,
474 struct bcm2835_spi *bs,
475 u32 cs)
476{
477 int tx_remaining;
478
479 bs->tfr = tfr;
480 bs->tx_prologue = 0;
481 bs->rx_prologue = 0;
482 bs->tx_spillover = false;
483
484 if (bs->tx_buf && !sg_is_last(&tfr->tx_sg.sgl[0]))
485 bs->tx_prologue = sg_dma_len(&tfr->tx_sg.sgl[0]) & 3;
486
487 if (bs->rx_buf && !sg_is_last(&tfr->rx_sg.sgl[0])) {
488 bs->rx_prologue = sg_dma_len(&tfr->rx_sg.sgl[0]) & 3;
489
490 if (bs->rx_prologue > bs->tx_prologue) {
491 if (!bs->tx_buf || sg_is_last(&tfr->tx_sg.sgl[0])) {
492 bs->tx_prologue = bs->rx_prologue;
493 } else {
494 bs->tx_prologue += 4;
495 bs->tx_spillover =
496 !(sg_dma_len(&tfr->tx_sg.sgl[0]) & ~3);
497 }
498 }
499 }
500
501 /* rx_prologue > 0 implies tx_prologue > 0, so check only the latter */
502 if (!bs->tx_prologue)
503 return;
504
505 /* Write and read RX prologue. Adjust first entry in RX sglist. */
506 if (bs->rx_prologue) {
507 bcm2835_wr(bs, BCM2835_SPI_DLEN, bs->rx_prologue);
508 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_TA
509 | BCM2835_SPI_CS_DMAEN);
510 bcm2835_wr_fifo_count(bs, bs->rx_prologue);
511 bcm2835_wait_tx_fifo_empty(bs);
512 bcm2835_rd_fifo_count(bs, bs->rx_prologue);
513 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_CLEAR_RX
514 | BCM2835_SPI_CS_CLEAR_TX
515 | BCM2835_SPI_CS_DONE);
516
517 dma_sync_single_for_device(ctlr->dma_rx->device->dev,
518 sg_dma_address(&tfr->rx_sg.sgl[0]),
519 bs->rx_prologue, DMA_FROM_DEVICE);
520
521 sg_dma_address(&tfr->rx_sg.sgl[0]) += bs->rx_prologue;
522 sg_dma_len(&tfr->rx_sg.sgl[0]) -= bs->rx_prologue;
523 }
524
525 if (!bs->tx_buf)
526 return;
527
528 /*
529 * Write remaining TX prologue. Adjust first entry in TX sglist.
530 * Also adjust second entry if prologue spills over to it.
531 */
532 tx_remaining = bs->tx_prologue - bs->rx_prologue;
533 if (tx_remaining) {
534 bcm2835_wr(bs, BCM2835_SPI_DLEN, tx_remaining);
535 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_TA
536 | BCM2835_SPI_CS_DMAEN);
537 bcm2835_wr_fifo_count(bs, tx_remaining);
538 bcm2835_wait_tx_fifo_empty(bs);
539 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_CLEAR_TX
540 | BCM2835_SPI_CS_DONE);
541 }
542
543 if (likely(!bs->tx_spillover)) {
544 sg_dma_address(&tfr->tx_sg.sgl[0]) += bs->tx_prologue;
545 sg_dma_len(&tfr->tx_sg.sgl[0]) -= bs->tx_prologue;
546 } else {
547 sg_dma_len(&tfr->tx_sg.sgl[0]) = 0;
548 sg_dma_address(&tfr->tx_sg.sgl[1]) += 4;
549 sg_dma_len(&tfr->tx_sg.sgl[1]) -= 4;
550 }
551}
552
553/**
554 * bcm2835_spi_undo_prologue() - reconstruct original sglist state
555 * @bs: BCM2835 SPI controller
556 *
557 * Undo changes which were made to an SPI transfer's sglist when transmitting
558 * the prologue. This is necessary to ensure the same memory ranges are
559 * unmapped that were originally mapped.
560 */
561static void bcm2835_spi_undo_prologue(struct bcm2835_spi *bs)
562{
563 struct spi_transfer *tfr = bs->tfr;
564
565 if (!bs->tx_prologue)
566 return;
567
568 if (bs->rx_prologue) {
569 sg_dma_address(&tfr->rx_sg.sgl[0]) -= bs->rx_prologue;
570 sg_dma_len(&tfr->rx_sg.sgl[0]) += bs->rx_prologue;
571 }
572
573 if (!bs->tx_buf)
574 goto out;
575
576 if (likely(!bs->tx_spillover)) {
577 sg_dma_address(&tfr->tx_sg.sgl[0]) -= bs->tx_prologue;
578 sg_dma_len(&tfr->tx_sg.sgl[0]) += bs->tx_prologue;
579 } else {
580 sg_dma_len(&tfr->tx_sg.sgl[0]) = bs->tx_prologue - 4;
581 sg_dma_address(&tfr->tx_sg.sgl[1]) -= 4;
582 sg_dma_len(&tfr->tx_sg.sgl[1]) += 4;
583 }
584out:
585 bs->tx_prologue = 0;
586}
587
588/**
589 * bcm2835_spi_dma_rx_done() - callback for DMA RX channel
590 * @data: SPI master controller
591 *
592 * Used for bidirectional and RX-only transfers.
593 */
594static void bcm2835_spi_dma_rx_done(void *data)
595{
596 struct spi_controller *ctlr = data;
597 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
598
599 /* terminate tx-dma as we do not have an irq for it
600 * because when the rx dma will terminate and this callback
601 * is called the tx-dma must have finished - can't get to this
602 * situation otherwise...
603 */
604 dmaengine_terminate_async(ctlr->dma_tx);
605 bs->tx_dma_active = false;
606 bs->rx_dma_active = false;
607 bcm2835_spi_undo_prologue(bs);
608
609 /* reset fifo and HW */
610 bcm2835_spi_reset_hw(ctlr);
611
612 /* and mark as completed */;
613 complete(&ctlr->xfer_completion);
614}
615
616/**
617 * bcm2835_spi_dma_tx_done() - callback for DMA TX channel
618 * @data: SPI master controller
619 *
620 * Used for TX-only transfers.
621 */
622static void bcm2835_spi_dma_tx_done(void *data)
623{
624 struct spi_controller *ctlr = data;
625 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
626
627 /* busy-wait for TX FIFO to empty */
628 while (!(bcm2835_rd(bs, BCM2835_SPI_CS) & BCM2835_SPI_CS_DONE))
629 bcm2835_wr(bs, BCM2835_SPI_CS,
630 bs->clear_rx_cs[bs->chip_select]);
631
632 bs->tx_dma_active = false;
633 smp_wmb();
634
635 /*
636 * In case of a very short transfer, RX DMA may not have been
637 * issued yet. The onus is then on bcm2835_spi_transfer_one_dma()
638 * to terminate it immediately after issuing.
639 */
640 if (cmpxchg(&bs->rx_dma_active, true, false))
641 dmaengine_terminate_async(ctlr->dma_rx);
642
643 bcm2835_spi_undo_prologue(bs);
644 bcm2835_spi_reset_hw(ctlr);
645 complete(&ctlr->xfer_completion);
646}
647
648/**
649 * bcm2835_spi_prepare_sg() - prepare and submit DMA descriptor for sglist
650 * @ctlr: SPI master controller
651 * @spi: SPI slave
652 * @tfr: SPI transfer
653 * @bs: BCM2835 SPI controller
654 * @is_tx: whether to submit DMA descriptor for TX or RX sglist
655 *
656 * Prepare and submit a DMA descriptor for the TX or RX sglist of @tfr.
657 * Return 0 on success or a negative error number.
658 */
659static int bcm2835_spi_prepare_sg(struct spi_controller *ctlr,
660 struct spi_device *spi,
661 struct spi_transfer *tfr,
662 struct bcm2835_spi *bs,
663 bool is_tx)
664{
665 struct dma_chan *chan;
666 struct scatterlist *sgl;
667 unsigned int nents;
668 enum dma_transfer_direction dir;
669 unsigned long flags;
670
671 struct dma_async_tx_descriptor *desc;
672 dma_cookie_t cookie;
673
674 if (is_tx) {
675 dir = DMA_MEM_TO_DEV;
676 chan = ctlr->dma_tx;
677 nents = tfr->tx_sg.nents;
678 sgl = tfr->tx_sg.sgl;
679 flags = tfr->rx_buf ? 0 : DMA_PREP_INTERRUPT;
680 } else {
681 dir = DMA_DEV_TO_MEM;
682 chan = ctlr->dma_rx;
683 nents = tfr->rx_sg.nents;
684 sgl = tfr->rx_sg.sgl;
685 flags = DMA_PREP_INTERRUPT;
686 }
687 /* prepare the channel */
688 desc = dmaengine_prep_slave_sg(chan, sgl, nents, dir, flags);
689 if (!desc)
690 return -EINVAL;
691
692 /*
693 * Completion is signaled by the RX channel for bidirectional and
694 * RX-only transfers; else by the TX channel for TX-only transfers.
695 */
696 if (!is_tx) {
697 desc->callback = bcm2835_spi_dma_rx_done;
698 desc->callback_param = ctlr;
699 } else if (!tfr->rx_buf) {
700 desc->callback = bcm2835_spi_dma_tx_done;
701 desc->callback_param = ctlr;
702 bs->chip_select = spi->chip_select;
703 }
704
705 /* submit it to DMA-engine */
706 cookie = dmaengine_submit(desc);
707
708 return dma_submit_error(cookie);
709}
710
711/**
712 * bcm2835_spi_transfer_one_dma() - perform SPI transfer using DMA engine
713 * @ctlr: SPI master controller
714 * @spi: SPI slave
715 * @tfr: SPI transfer
716 * @cs: CS register
717 *
718 * For *bidirectional* transfers (both tx_buf and rx_buf are non-%NULL), set up
719 * the TX and RX DMA channel to copy between memory and FIFO register.
720 *
721 * For *TX-only* transfers (rx_buf is %NULL), copying the RX FIFO's contents to
722 * memory is pointless. However not reading the RX FIFO isn't an option either
723 * because transmission is halted once it's full. As a workaround, cyclically
724 * clear the RX FIFO by setting the CLEAR_RX bit in the CS register.
725 *
726 * The CS register value is precalculated in bcm2835_spi_setup(). Normally
727 * this is called only once, on slave registration. A DMA descriptor to write
728 * this value is preallocated in bcm2835_dma_init(). All that's left to do
729 * when performing a TX-only transfer is to submit this descriptor to the RX
730 * DMA channel. Latency is thereby minimized. The descriptor does not
731 * generate any interrupts while running. It must be terminated once the
732 * TX DMA channel is done.
733 *
734 * Clearing the RX FIFO is paced by the DREQ signal. The signal is asserted
735 * when the RX FIFO becomes half full, i.e. 32 bytes. (Tuneable with the DC
736 * register.) Reading 32 bytes from the RX FIFO would normally require 8 bus
737 * accesses, whereas clearing it requires only 1 bus access. So an 8-fold
738 * reduction in bus traffic and thus energy consumption is achieved.
739 *
740 * For *RX-only* transfers (tx_buf is %NULL), fill the TX FIFO by cyclically
741 * copying from the zero page. The DMA descriptor to do this is preallocated
742 * in bcm2835_dma_init(). It must be terminated once the RX DMA channel is
743 * done and can then be reused.
744 *
745 * The BCM2835 DMA driver autodetects when a transaction copies from the zero
746 * page and utilizes the DMA controller's ability to synthesize zeroes instead
747 * of copying them from memory. This reduces traffic on the memory bus. The
748 * feature is not available on so-called "lite" channels, but normally TX DMA
749 * is backed by a full-featured channel.
750 *
751 * Zero-filling the TX FIFO is paced by the DREQ signal. Unfortunately the
752 * BCM2835 SPI controller continues to assert DREQ even after the DLEN register
753 * has been counted down to zero (hardware erratum). Thus, when the transfer
754 * has finished, the DMA engine zero-fills the TX FIFO until it is half full.
755 * (Tuneable with the DC register.) So up to 9 gratuitous bus accesses are
756 * performed at the end of an RX-only transfer.
757 */
758static int bcm2835_spi_transfer_one_dma(struct spi_controller *ctlr,
759 struct spi_device *spi,
760 struct spi_transfer *tfr,
761 u32 cs)
762{
763 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
764 dma_cookie_t cookie;
765 int ret;
766
767 /* update usage statistics */
768 bs->count_transfer_dma++;
769
770 /*
771 * Transfer first few bytes without DMA if length of first TX or RX
772 * sglist entry is not a multiple of 4 bytes (hardware limitation).
773 */
774 bcm2835_spi_transfer_prologue(ctlr, tfr, bs, cs);
775
776 /* setup tx-DMA */
777 if (bs->tx_buf) {
778 ret = bcm2835_spi_prepare_sg(ctlr, spi, tfr, bs, true);
779 } else {
780 cookie = dmaengine_submit(bs->fill_tx_desc);
781 ret = dma_submit_error(cookie);
782 }
783 if (ret)
784 goto err_reset_hw;
785
786 /* set the DMA length */
787 bcm2835_wr(bs, BCM2835_SPI_DLEN, bs->tx_len);
788
789 /* start the HW */
790 bcm2835_wr(bs, BCM2835_SPI_CS,
791 cs | BCM2835_SPI_CS_TA | BCM2835_SPI_CS_DMAEN);
792
793 bs->tx_dma_active = true;
794 smp_wmb();
795
796 /* start TX early */
797 dma_async_issue_pending(ctlr->dma_tx);
798
799 /* setup rx-DMA late - to run transfers while
800 * mapping of the rx buffers still takes place
801 * this saves 10us or more.
802 */
803 if (bs->rx_buf) {
804 ret = bcm2835_spi_prepare_sg(ctlr, spi, tfr, bs, false);
805 } else {
806 cookie = dmaengine_submit(bs->clear_rx_desc[spi->chip_select]);
807 ret = dma_submit_error(cookie);
808 }
809 if (ret) {
810 /* need to reset on errors */
811 dmaengine_terminate_sync(ctlr->dma_tx);
812 bs->tx_dma_active = false;
813 goto err_reset_hw;
814 }
815
816 /* start rx dma late */
817 dma_async_issue_pending(ctlr->dma_rx);
818 bs->rx_dma_active = true;
819 smp_mb();
820
821 /*
822 * In case of a very short TX-only transfer, bcm2835_spi_dma_tx_done()
823 * may run before RX DMA is issued. Terminate RX DMA if so.
824 */
825 if (!bs->rx_buf && !bs->tx_dma_active &&
826 cmpxchg(&bs->rx_dma_active, true, false)) {
827 dmaengine_terminate_async(ctlr->dma_rx);
828 bcm2835_spi_reset_hw(ctlr);
829 }
830
831 /* wait for wakeup in framework */
832 return 1;
833
834err_reset_hw:
835 bcm2835_spi_reset_hw(ctlr);
836 bcm2835_spi_undo_prologue(bs);
837 return ret;
838}
839
840static bool bcm2835_spi_can_dma(struct spi_controller *ctlr,
841 struct spi_device *spi,
842 struct spi_transfer *tfr)
843{
844 /* we start DMA efforts only on bigger transfers */
845 if (tfr->len < BCM2835_SPI_DMA_MIN_LENGTH)
846 return false;
847
848 /* return OK */
849 return true;
850}
851
852static void bcm2835_dma_release(struct spi_controller *ctlr,
853 struct bcm2835_spi *bs)
854{
855 int i;
856
857 if (ctlr->dma_tx) {
858 dmaengine_terminate_sync(ctlr->dma_tx);
859
860 if (bs->fill_tx_desc)
861 dmaengine_desc_free(bs->fill_tx_desc);
862
863 if (bs->fill_tx_addr)
864 dma_unmap_page_attrs(ctlr->dma_tx->device->dev,
865 bs->fill_tx_addr, sizeof(u32),
866 DMA_TO_DEVICE,
867 DMA_ATTR_SKIP_CPU_SYNC);
868
869 dma_release_channel(ctlr->dma_tx);
870 ctlr->dma_tx = NULL;
871 }
872
873 if (ctlr->dma_rx) {
874 dmaengine_terminate_sync(ctlr->dma_rx);
875
876 for (i = 0; i < BCM2835_SPI_NUM_CS; i++)
877 if (bs->clear_rx_desc[i])
878 dmaengine_desc_free(bs->clear_rx_desc[i]);
879
880 if (bs->clear_rx_addr)
881 dma_unmap_single(ctlr->dma_rx->device->dev,
882 bs->clear_rx_addr,
883 sizeof(bs->clear_rx_cs),
884 DMA_TO_DEVICE);
885
886 dma_release_channel(ctlr->dma_rx);
887 ctlr->dma_rx = NULL;
888 }
889}
890
891static void bcm2835_dma_init(struct spi_controller *ctlr, struct device *dev,
892 struct bcm2835_spi *bs)
893{
894 struct dma_slave_config slave_config;
895 const __be32 *addr;
896 dma_addr_t dma_reg_base;
897 int ret, i;
898
899 /* base address in dma-space */
900 addr = of_get_address(ctlr->dev.of_node, 0, NULL, NULL);
901 if (!addr) {
902 dev_err(dev, "could not get DMA-register address - not using dma mode\n");
903 goto err;
904 }
905 dma_reg_base = be32_to_cpup(addr);
906
907 /* get tx/rx dma */
908 ctlr->dma_tx = dma_request_slave_channel(dev, "tx");
909 if (!ctlr->dma_tx) {
910 dev_err(dev, "no tx-dma configuration found - not using dma mode\n");
911 goto err;
912 }
913 ctlr->dma_rx = dma_request_slave_channel(dev, "rx");
914 if (!ctlr->dma_rx) {
915 dev_err(dev, "no rx-dma configuration found - not using dma mode\n");
916 goto err_release;
917 }
918
919 /*
920 * The TX DMA channel either copies a transfer's TX buffer to the FIFO
921 * or, in case of an RX-only transfer, cyclically copies from the zero
922 * page to the FIFO using a preallocated, reusable descriptor.
923 */
924 slave_config.dst_addr = (u32)(dma_reg_base + BCM2835_SPI_FIFO);
925 slave_config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
926
927 ret = dmaengine_slave_config(ctlr->dma_tx, &slave_config);
928 if (ret)
929 goto err_config;
930
931 bs->fill_tx_addr = dma_map_page_attrs(ctlr->dma_tx->device->dev,
932 ZERO_PAGE(0), 0, sizeof(u32),
933 DMA_TO_DEVICE,
934 DMA_ATTR_SKIP_CPU_SYNC);
935 if (dma_mapping_error(ctlr->dma_tx->device->dev, bs->fill_tx_addr)) {
936 dev_err(dev, "cannot map zero page - not using DMA mode\n");
937 bs->fill_tx_addr = 0;
938 goto err_release;
939 }
940
941 bs->fill_tx_desc = dmaengine_prep_dma_cyclic(ctlr->dma_tx,
942 bs->fill_tx_addr,
943 sizeof(u32), 0,
944 DMA_MEM_TO_DEV, 0);
945 if (!bs->fill_tx_desc) {
946 dev_err(dev, "cannot prepare fill_tx_desc - not using DMA mode\n");
947 goto err_release;
948 }
949
950 ret = dmaengine_desc_set_reuse(bs->fill_tx_desc);
951 if (ret) {
952 dev_err(dev, "cannot reuse fill_tx_desc - not using DMA mode\n");
953 goto err_release;
954 }
955
956 /*
957 * The RX DMA channel is used bidirectionally: It either reads the
958 * RX FIFO or, in case of a TX-only transfer, cyclically writes a
959 * precalculated value to the CS register to clear the RX FIFO.
960 */
961 slave_config.src_addr = (u32)(dma_reg_base + BCM2835_SPI_FIFO);
962 slave_config.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
963 slave_config.dst_addr = (u32)(dma_reg_base + BCM2835_SPI_CS);
964 slave_config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
965
966 ret = dmaengine_slave_config(ctlr->dma_rx, &slave_config);
967 if (ret)
968 goto err_config;
969
970 bs->clear_rx_addr = dma_map_single(ctlr->dma_rx->device->dev,
971 bs->clear_rx_cs,
972 sizeof(bs->clear_rx_cs),
973 DMA_TO_DEVICE);
974 if (dma_mapping_error(ctlr->dma_rx->device->dev, bs->clear_rx_addr)) {
975 dev_err(dev, "cannot map clear_rx_cs - not using DMA mode\n");
976 bs->clear_rx_addr = 0;
977 goto err_release;
978 }
979
980 for (i = 0; i < BCM2835_SPI_NUM_CS; i++) {
981 bs->clear_rx_desc[i] = dmaengine_prep_dma_cyclic(ctlr->dma_rx,
982 bs->clear_rx_addr + i * sizeof(u32),
983 sizeof(u32), 0,
984 DMA_MEM_TO_DEV, 0);
985 if (!bs->clear_rx_desc[i]) {
986 dev_err(dev, "cannot prepare clear_rx_desc - not using DMA mode\n");
987 goto err_release;
988 }
989
990 ret = dmaengine_desc_set_reuse(bs->clear_rx_desc[i]);
991 if (ret) {
992 dev_err(dev, "cannot reuse clear_rx_desc - not using DMA mode\n");
993 goto err_release;
994 }
995 }
996
997 /* all went well, so set can_dma */
998 ctlr->can_dma = bcm2835_spi_can_dma;
999
1000 return;
1001
1002err_config:
1003 dev_err(dev, "issue configuring dma: %d - not using DMA mode\n",
1004 ret);
1005err_release:
1006 bcm2835_dma_release(ctlr, bs);
1007err:
1008 return;
1009}
1010
1011static int bcm2835_spi_transfer_one_poll(struct spi_controller *ctlr,
1012 struct spi_device *spi,
1013 struct spi_transfer *tfr,
1014 u32 cs)
1015{
1016 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1017 unsigned long timeout;
1018
1019 /* update usage statistics */
1020 bs->count_transfer_polling++;
1021
1022 /* enable HW block without interrupts */
1023 bcm2835_wr(bs, BCM2835_SPI_CS, cs | BCM2835_SPI_CS_TA);
1024
1025 /* fill in the fifo before timeout calculations
1026 * if we are interrupted here, then the data is
1027 * getting transferred by the HW while we are interrupted
1028 */
1029 bcm2835_wr_fifo_blind(bs, BCM2835_SPI_FIFO_SIZE);
1030
1031 /* set the timeout to at least 2 jiffies */
1032 timeout = jiffies + 2 + HZ * polling_limit_us / 1000000;
1033
1034 /* loop until finished the transfer */
1035 while (bs->rx_len) {
1036 /* fill in tx fifo with remaining data */
1037 bcm2835_wr_fifo(bs);
1038
1039 /* read from fifo as much as possible */
1040 bcm2835_rd_fifo(bs);
1041
1042 /* if there is still data pending to read
1043 * then check the timeout
1044 */
1045 if (bs->rx_len && time_after(jiffies, timeout)) {
1046 dev_dbg_ratelimited(&spi->dev,
1047 "timeout period reached: jiffies: %lu remaining tx/rx: %d/%d - falling back to interrupt mode\n",
1048 jiffies - timeout,
1049 bs->tx_len, bs->rx_len);
1050 /* fall back to interrupt mode */
1051
1052 /* update usage statistics */
1053 bs->count_transfer_irq_after_polling++;
1054
1055 return bcm2835_spi_transfer_one_irq(ctlr, spi,
1056 tfr, cs, false);
1057 }
1058 }
1059
1060 /* Transfer complete - reset SPI HW */
1061 bcm2835_spi_reset_hw(ctlr);
1062 /* and return without waiting for completion */
1063 return 0;
1064}
1065
1066static int bcm2835_spi_transfer_one(struct spi_controller *ctlr,
1067 struct spi_device *spi,
1068 struct spi_transfer *tfr)
1069{
1070 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1071 unsigned long spi_hz, clk_hz, cdiv, spi_used_hz;
1072 unsigned long hz_per_byte, byte_limit;
1073 u32 cs = bs->prepare_cs[spi->chip_select];
1074
1075 /* set clock */
1076 spi_hz = tfr->speed_hz;
1077 clk_hz = clk_get_rate(bs->clk);
1078
1079 if (spi_hz >= clk_hz / 2) {
1080 cdiv = 2; /* clk_hz/2 is the fastest we can go */
1081 } else if (spi_hz) {
1082 /* CDIV must be a multiple of two */
1083 cdiv = DIV_ROUND_UP(clk_hz, spi_hz);
1084 cdiv += (cdiv % 2);
1085
1086 if (cdiv >= 65536)
1087 cdiv = 0; /* 0 is the slowest we can go */
1088 } else {
1089 cdiv = 0; /* 0 is the slowest we can go */
1090 }
1091 spi_used_hz = cdiv ? (clk_hz / cdiv) : (clk_hz / 65536);
1092 bcm2835_wr(bs, BCM2835_SPI_CLK, cdiv);
1093
1094 /* handle all the 3-wire mode */
1095 if (spi->mode & SPI_3WIRE && tfr->rx_buf)
1096 cs |= BCM2835_SPI_CS_REN;
1097
1098 /* set transmit buffers and length */
1099 bs->tx_buf = tfr->tx_buf;
1100 bs->rx_buf = tfr->rx_buf;
1101 bs->tx_len = tfr->len;
1102 bs->rx_len = tfr->len;
1103
1104 /* Calculate the estimated time in us the transfer runs. Note that
1105 * there is 1 idle clocks cycles after each byte getting transferred
1106 * so we have 9 cycles/byte. This is used to find the number of Hz
1107 * per byte per polling limit. E.g., we can transfer 1 byte in 30 us
1108 * per 300,000 Hz of bus clock.
1109 */
1110 hz_per_byte = polling_limit_us ? (9 * 1000000) / polling_limit_us : 0;
1111 byte_limit = hz_per_byte ? spi_used_hz / hz_per_byte : 1;
1112
1113 /* run in polling mode for short transfers */
1114 if (tfr->len < byte_limit)
1115 return bcm2835_spi_transfer_one_poll(ctlr, spi, tfr, cs);
1116
1117 /* run in dma mode if conditions are right
1118 * Note that unlike poll or interrupt mode DMA mode does not have
1119 * this 1 idle clock cycle pattern but runs the spi clock without gaps
1120 */
1121 if (ctlr->can_dma && bcm2835_spi_can_dma(ctlr, spi, tfr))
1122 return bcm2835_spi_transfer_one_dma(ctlr, spi, tfr, cs);
1123
1124 /* run in interrupt-mode */
1125 return bcm2835_spi_transfer_one_irq(ctlr, spi, tfr, cs, true);
1126}
1127
1128static int bcm2835_spi_prepare_message(struct spi_controller *ctlr,
1129 struct spi_message *msg)
1130{
1131 struct spi_device *spi = msg->spi;
1132 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1133 int ret;
1134
1135 if (ctlr->can_dma) {
1136 /*
1137 * DMA transfers are limited to 16 bit (0 to 65535 bytes) by
1138 * the SPI HW due to DLEN. Split up transfers (32-bit FIFO
1139 * aligned) if the limit is exceeded.
1140 */
1141 ret = spi_split_transfers_maxsize(ctlr, msg, 65532,
1142 GFP_KERNEL | GFP_DMA);
1143 if (ret)
1144 return ret;
1145 }
1146
1147 /*
1148 * Set up clock polarity before spi_transfer_one_message() asserts
1149 * chip select to avoid a gratuitous clock signal edge.
1150 */
1151 bcm2835_wr(bs, BCM2835_SPI_CS, bs->prepare_cs[spi->chip_select]);
1152
1153 return 0;
1154}
1155
1156static void bcm2835_spi_handle_err(struct spi_controller *ctlr,
1157 struct spi_message *msg)
1158{
1159 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1160
1161 /* if an error occurred and we have an active dma, then terminate */
1162 dmaengine_terminate_sync(ctlr->dma_tx);
1163 bs->tx_dma_active = false;
1164 dmaengine_terminate_sync(ctlr->dma_rx);
1165 bs->rx_dma_active = false;
1166 bcm2835_spi_undo_prologue(bs);
1167
1168 /* and reset */
1169 bcm2835_spi_reset_hw(ctlr);
1170}
1171
1172static int chip_match_name(struct gpio_chip *chip, void *data)
1173{
1174 return !strcmp(chip->label, data);
1175}
1176
1177static int bcm2835_spi_setup(struct spi_device *spi)
1178{
1179 struct spi_controller *ctlr = spi->controller;
1180 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1181 struct gpio_chip *chip;
1182 enum gpio_lookup_flags lflags;
1183 u32 cs;
1184
1185 /*
1186 * Precalculate SPI slave's CS register value for ->prepare_message():
1187 * The driver always uses software-controlled GPIO chip select, hence
1188 * set the hardware-controlled native chip select to an invalid value
1189 * to prevent it from interfering.
1190 */
1191 cs = BCM2835_SPI_CS_CS_10 | BCM2835_SPI_CS_CS_01;
1192 if (spi->mode & SPI_CPOL)
1193 cs |= BCM2835_SPI_CS_CPOL;
1194 if (spi->mode & SPI_CPHA)
1195 cs |= BCM2835_SPI_CS_CPHA;
1196 bs->prepare_cs[spi->chip_select] = cs;
1197
1198 /*
1199 * Precalculate SPI slave's CS register value to clear RX FIFO
1200 * in case of a TX-only DMA transfer.
1201 */
1202 if (ctlr->dma_rx) {
1203 bs->clear_rx_cs[spi->chip_select] = cs |
1204 BCM2835_SPI_CS_TA |
1205 BCM2835_SPI_CS_DMAEN |
1206 BCM2835_SPI_CS_CLEAR_RX;
1207 dma_sync_single_for_device(ctlr->dma_rx->device->dev,
1208 bs->clear_rx_addr,
1209 sizeof(bs->clear_rx_cs),
1210 DMA_TO_DEVICE);
1211 }
1212
1213 /*
1214 * sanity checking the native-chipselects
1215 */
1216 if (spi->mode & SPI_NO_CS)
1217 return 0;
1218 /*
1219 * The SPI core has successfully requested the CS GPIO line from the
1220 * device tree, so we are done.
1221 */
1222 if (spi->cs_gpiod)
1223 return 0;
1224 if (spi->chip_select > 1) {
1225 /* error in the case of native CS requested with CS > 1
1226 * officially there is a CS2, but it is not documented
1227 * which GPIO is connected with that...
1228 */
1229 dev_err(&spi->dev,
1230 "setup: only two native chip-selects are supported\n");
1231 return -EINVAL;
1232 }
1233
1234 /*
1235 * Translate native CS to GPIO
1236 *
1237 * FIXME: poking around in the gpiolib internals like this is
1238 * not very good practice. Find a way to locate the real problem
1239 * and fix it. Why is the GPIO descriptor in spi->cs_gpiod
1240 * sometimes not assigned correctly? Erroneous device trees?
1241 */
1242
1243 /* get the gpio chip for the base */
1244 chip = gpiochip_find("pinctrl-bcm2835", chip_match_name);
1245 if (!chip)
1246 return 0;
1247
1248 /*
1249 * Retrieve the corresponding GPIO line used for CS.
1250 * The inversion semantics will be handled by the GPIO core
1251 * code, so we pass GPIOS_OUT_LOW for "unasserted" and
1252 * the correct flag for inversion semantics. The SPI_CS_HIGH
1253 * on spi->mode cannot be checked for polarity in this case
1254 * as the flag use_gpio_descriptors enforces SPI_CS_HIGH.
1255 */
1256 if (of_property_read_bool(spi->dev.of_node, "spi-cs-high"))
1257 lflags = GPIO_ACTIVE_HIGH;
1258 else
1259 lflags = GPIO_ACTIVE_LOW;
1260 spi->cs_gpiod = gpiochip_request_own_desc(chip, 8 - spi->chip_select,
1261 DRV_NAME,
1262 lflags,
1263 GPIOD_OUT_LOW);
1264 if (IS_ERR(spi->cs_gpiod))
1265 return PTR_ERR(spi->cs_gpiod);
1266
1267 /* and set up the "mode" and level */
1268 dev_info(&spi->dev, "setting up native-CS%i to use GPIO\n",
1269 spi->chip_select);
1270
1271 return 0;
1272}
1273
1274static int bcm2835_spi_probe(struct platform_device *pdev)
1275{
1276 struct spi_controller *ctlr;
1277 struct bcm2835_spi *bs;
1278 int err;
1279
1280 ctlr = spi_alloc_master(&pdev->dev, ALIGN(sizeof(*bs),
1281 dma_get_cache_alignment()));
1282 if (!ctlr)
1283 return -ENOMEM;
1284
1285 platform_set_drvdata(pdev, ctlr);
1286
1287 ctlr->use_gpio_descriptors = true;
1288 ctlr->mode_bits = BCM2835_SPI_MODE_BITS;
1289 ctlr->bits_per_word_mask = SPI_BPW_MASK(8);
1290 ctlr->num_chipselect = BCM2835_SPI_NUM_CS;
1291 ctlr->setup = bcm2835_spi_setup;
1292 ctlr->transfer_one = bcm2835_spi_transfer_one;
1293 ctlr->handle_err = bcm2835_spi_handle_err;
1294 ctlr->prepare_message = bcm2835_spi_prepare_message;
1295 ctlr->dev.of_node = pdev->dev.of_node;
1296
1297 bs = spi_controller_get_devdata(ctlr);
1298
1299 bs->regs = devm_platform_ioremap_resource(pdev, 0);
1300 if (IS_ERR(bs->regs)) {
1301 err = PTR_ERR(bs->regs);
1302 goto out_controller_put;
1303 }
1304
1305 bs->clk = devm_clk_get(&pdev->dev, NULL);
1306 if (IS_ERR(bs->clk)) {
1307 err = PTR_ERR(bs->clk);
1308 dev_err(&pdev->dev, "could not get clk: %d\n", err);
1309 goto out_controller_put;
1310 }
1311
1312 bs->irq = platform_get_irq(pdev, 0);
1313 if (bs->irq <= 0) {
1314 err = bs->irq ? bs->irq : -ENODEV;
1315 goto out_controller_put;
1316 }
1317
1318 clk_prepare_enable(bs->clk);
1319
1320 bcm2835_dma_init(ctlr, &pdev->dev, bs);
1321
1322 /* initialise the hardware with the default polarities */
1323 bcm2835_wr(bs, BCM2835_SPI_CS,
1324 BCM2835_SPI_CS_CLEAR_RX | BCM2835_SPI_CS_CLEAR_TX);
1325
1326 err = devm_request_irq(&pdev->dev, bs->irq, bcm2835_spi_interrupt, 0,
1327 dev_name(&pdev->dev), ctlr);
1328 if (err) {
1329 dev_err(&pdev->dev, "could not request IRQ: %d\n", err);
1330 goto out_clk_disable;
1331 }
1332
1333 err = devm_spi_register_controller(&pdev->dev, ctlr);
1334 if (err) {
1335 dev_err(&pdev->dev, "could not register SPI controller: %d\n",
1336 err);
1337 goto out_clk_disable;
1338 }
1339
1340 bcm2835_debugfs_create(bs, dev_name(&pdev->dev));
1341
1342 return 0;
1343
1344out_clk_disable:
1345 clk_disable_unprepare(bs->clk);
1346out_controller_put:
1347 spi_controller_put(ctlr);
1348 return err;
1349}
1350
1351static int bcm2835_spi_remove(struct platform_device *pdev)
1352{
1353 struct spi_controller *ctlr = platform_get_drvdata(pdev);
1354 struct bcm2835_spi *bs = spi_controller_get_devdata(ctlr);
1355
1356 bcm2835_debugfs_remove(bs);
1357
1358 /* Clear FIFOs, and disable the HW block */
1359 bcm2835_wr(bs, BCM2835_SPI_CS,
1360 BCM2835_SPI_CS_CLEAR_RX | BCM2835_SPI_CS_CLEAR_TX);
1361
1362 clk_disable_unprepare(bs->clk);
1363
1364 bcm2835_dma_release(ctlr, bs);
1365
1366 return 0;
1367}
1368
1369static const struct of_device_id bcm2835_spi_match[] = {
1370 { .compatible = "brcm,bcm2835-spi", },
1371 {}
1372};
1373MODULE_DEVICE_TABLE(of, bcm2835_spi_match);
1374
1375static struct platform_driver bcm2835_spi_driver = {
1376 .driver = {
1377 .name = DRV_NAME,
1378 .of_match_table = bcm2835_spi_match,
1379 },
1380 .probe = bcm2835_spi_probe,
1381 .remove = bcm2835_spi_remove,
1382};
1383module_platform_driver(bcm2835_spi_driver);
1384
1385MODULE_DESCRIPTION("SPI controller driver for Broadcom BCM2835");
1386MODULE_AUTHOR("Chris Boot <bootc@bootc.net>");
1387MODULE_LICENSE("GPL");