xref: /linux/drivers/spi/spi-qcom-qspi.c (revision 4f77c3462308c62ffe7129cc18b9ac937f44b5a5)
1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (c) 2017-2018, The Linux foundation. All rights reserved.
3 
4 #include <linux/clk.h>
5 #include <linux/dmapool.h>
6 #include <linux/dma-mapping.h>
7 #include <linux/interconnect.h>
8 #include <linux/interrupt.h>
9 #include <linux/io.h>
10 #include <linux/module.h>
11 #include <linux/of.h>
12 #include <linux/platform_device.h>
13 #include <linux/pinctrl/consumer.h>
14 #include <linux/pm_runtime.h>
15 #include <linux/pm_opp.h>
16 #include <linux/spi/spi.h>
17 #include <linux/spi/spi-mem.h>
18 
19 
20 #define QSPI_NUM_CS		2
21 #define QSPI_BYTES_PER_WORD	4
22 
23 #define MSTR_CONFIG		0x0000
24 #define FULL_CYCLE_MODE		BIT(3)
25 #define FB_CLK_EN		BIT(4)
26 #define PIN_HOLDN		BIT(6)
27 #define PIN_WPN			BIT(7)
28 #define DMA_ENABLE		BIT(8)
29 #define BIG_ENDIAN_MODE		BIT(9)
30 #define SPI_MODE_MSK		0xc00
31 #define SPI_MODE_SHFT		10
32 #define CHIP_SELECT_NUM		BIT(12)
33 #define SBL_EN			BIT(13)
34 #define LPA_BASE_MSK		0x3c000
35 #define LPA_BASE_SHFT		14
36 #define TX_DATA_DELAY_MSK	0xc0000
37 #define TX_DATA_DELAY_SHFT	18
38 #define TX_CLK_DELAY_MSK	0x300000
39 #define TX_CLK_DELAY_SHFT	20
40 #define TX_CS_N_DELAY_MSK	0xc00000
41 #define TX_CS_N_DELAY_SHFT	22
42 #define TX_DATA_OE_DELAY_MSK	0x3000000
43 #define TX_DATA_OE_DELAY_SHFT	24
44 
45 #define AHB_MASTER_CFG				0x0004
46 #define HMEM_TYPE_START_MID_TRANS_MSK		0x7
47 #define HMEM_TYPE_START_MID_TRANS_SHFT		0
48 #define HMEM_TYPE_LAST_TRANS_MSK		0x38
49 #define HMEM_TYPE_LAST_TRANS_SHFT		3
50 #define USE_HMEMTYPE_LAST_ON_DESC_OR_CHAIN_MSK	0xc0
51 #define USE_HMEMTYPE_LAST_ON_DESC_OR_CHAIN_SHFT	6
52 #define HMEMTYPE_READ_TRANS_MSK			0x700
53 #define HMEMTYPE_READ_TRANS_SHFT		8
54 #define HSHARED					BIT(11)
55 #define HINNERSHARED				BIT(12)
56 
57 #define MSTR_INT_EN		0x000C
58 #define MSTR_INT_STATUS		0x0010
59 #define RESP_FIFO_UNDERRUN	BIT(0)
60 #define RESP_FIFO_NOT_EMPTY	BIT(1)
61 #define RESP_FIFO_RDY		BIT(2)
62 #define HRESP_FROM_NOC_ERR	BIT(3)
63 #define WR_FIFO_EMPTY		BIT(9)
64 #define WR_FIFO_FULL		BIT(10)
65 #define WR_FIFO_OVERRUN		BIT(11)
66 #define TRANSACTION_DONE	BIT(16)
67 #define DMA_CHAIN_DONE		BIT(31)
68 #define QSPI_ERR_IRQS		(RESP_FIFO_UNDERRUN | HRESP_FROM_NOC_ERR | \
69 				 WR_FIFO_OVERRUN)
70 #define QSPI_ALL_IRQS		(QSPI_ERR_IRQS | RESP_FIFO_RDY | \
71 				 WR_FIFO_EMPTY | WR_FIFO_FULL | \
72 				 TRANSACTION_DONE | DMA_CHAIN_DONE)
73 
74 #define PIO_XFER_CTRL		0x0014
75 #define REQUEST_COUNT_MSK	0xffff
76 
77 #define PIO_XFER_CFG		0x0018
78 #define TRANSFER_DIRECTION	BIT(0)
79 #define MULTI_IO_MODE_MSK	0xe
80 #define MULTI_IO_MODE_SHFT	1
81 #define TRANSFER_FRAGMENT	BIT(8)
82 #define SDR_1BIT		1
83 #define SDR_2BIT		2
84 #define SDR_4BIT		3
85 #define DDR_1BIT		5
86 #define DDR_2BIT		6
87 #define DDR_4BIT		7
88 #define DMA_DESC_SINGLE_SPI	1
89 #define DMA_DESC_DUAL_SPI	2
90 #define DMA_DESC_QUAD_SPI	3
91 
92 #define PIO_XFER_STATUS		0x001c
93 #define WR_FIFO_BYTES_MSK	0xffff0000
94 #define WR_FIFO_BYTES_SHFT	16
95 
96 #define PIO_DATAOUT_1B		0x0020
97 #define PIO_DATAOUT_4B		0x0024
98 
99 #define RD_FIFO_CFG		0x0028
100 #define CONTINUOUS_MODE		BIT(0)
101 
102 #define RD_FIFO_STATUS	0x002c
103 #define FIFO_EMPTY	BIT(11)
104 #define WR_CNTS_MSK	0x7f0
105 #define WR_CNTS_SHFT	4
106 #define RDY_64BYTE	BIT(3)
107 #define RDY_32BYTE	BIT(2)
108 #define RDY_16BYTE	BIT(1)
109 #define FIFO_RDY	BIT(0)
110 
111 #define RD_FIFO_RESET		0x0030
112 #define RESET_FIFO		BIT(0)
113 
114 #define NEXT_DMA_DESC_ADDR	0x0040
115 #define CURRENT_DMA_DESC_ADDR	0x0044
116 #define CURRENT_MEM_ADDR	0x0048
117 
118 #define CUR_MEM_ADDR		0x0048
119 #define HW_VERSION		0x004c
120 #define RD_FIFO			0x0050
121 #define SAMPLING_CLK_CFG	0x0090
122 #define SAMPLING_CLK_STATUS	0x0094
123 
124 #define QSPI_ALIGN_REQ	32
125 
126 enum qspi_dir {
127 	QSPI_READ,
128 	QSPI_WRITE,
129 };
130 
131 struct qspi_cmd_desc {
132 	u32 data_address;
133 	u32 next_descriptor;
134 	u32 direction:1;
135 	u32 multi_io_mode:3;
136 	u32 reserved1:4;
137 	u32 fragment:1;
138 	u32 reserved2:7;
139 	u32 length:16;
140 };
141 
142 struct qspi_xfer {
143 	union {
144 		const void *tx_buf;
145 		void *rx_buf;
146 	};
147 	unsigned int rem_bytes;
148 	unsigned int buswidth;
149 	enum qspi_dir dir;
150 	bool is_last;
151 };
152 
153 enum qspi_clocks {
154 	QSPI_CLK_CORE,
155 	QSPI_CLK_IFACE,
156 	QSPI_NUM_CLKS
157 };
158 
159 /*
160  * Number of entries in sgt returned from spi framework that-
161  * will be supported. Can be modified as required.
162  * In practice, given max_dma_len is 64KB, the number of
163  * entries is not expected to exceed 1.
164  */
165 #define QSPI_MAX_SG 5
166 
167 struct qcom_qspi {
168 	void __iomem *base;
169 	struct device *dev;
170 	struct clk_bulk_data *clks;
171 	struct qspi_xfer xfer;
172 	struct dma_pool *dma_cmd_pool;
173 	dma_addr_t dma_cmd_desc[QSPI_MAX_SG];
174 	void *virt_cmd_desc[QSPI_MAX_SG];
175 	unsigned int n_cmd_desc;
176 	struct icc_path *icc_path_cpu_to_qspi;
177 	unsigned long last_speed;
178 	/* Lock to protect data accessed by IRQs */
179 	spinlock_t lock;
180 };
181 
182 static u32 qspi_buswidth_to_iomode(struct qcom_qspi *ctrl,
183 				   unsigned int buswidth)
184 {
185 	switch (buswidth) {
186 	case 1:
187 		return SDR_1BIT;
188 	case 2:
189 		return SDR_2BIT;
190 	case 4:
191 		return SDR_4BIT;
192 	default:
193 		dev_warn_once(ctrl->dev,
194 				"Unexpected bus width: %u\n", buswidth);
195 		return SDR_1BIT;
196 	}
197 }
198 
199 static void qcom_qspi_pio_xfer_cfg(struct qcom_qspi *ctrl)
200 {
201 	u32 pio_xfer_cfg;
202 	u32 iomode;
203 	const struct qspi_xfer *xfer;
204 
205 	xfer = &ctrl->xfer;
206 	pio_xfer_cfg = readl(ctrl->base + PIO_XFER_CFG);
207 	pio_xfer_cfg &= ~TRANSFER_DIRECTION;
208 	pio_xfer_cfg |= xfer->dir;
209 	if (xfer->is_last)
210 		pio_xfer_cfg &= ~TRANSFER_FRAGMENT;
211 	else
212 		pio_xfer_cfg |= TRANSFER_FRAGMENT;
213 	pio_xfer_cfg &= ~MULTI_IO_MODE_MSK;
214 	iomode = qspi_buswidth_to_iomode(ctrl, xfer->buswidth);
215 	pio_xfer_cfg |= iomode << MULTI_IO_MODE_SHFT;
216 
217 	writel(pio_xfer_cfg, ctrl->base + PIO_XFER_CFG);
218 }
219 
220 static void qcom_qspi_pio_xfer_ctrl(struct qcom_qspi *ctrl)
221 {
222 	u32 pio_xfer_ctrl;
223 
224 	pio_xfer_ctrl = readl(ctrl->base + PIO_XFER_CTRL);
225 	pio_xfer_ctrl &= ~REQUEST_COUNT_MSK;
226 	pio_xfer_ctrl |= ctrl->xfer.rem_bytes;
227 	writel(pio_xfer_ctrl, ctrl->base + PIO_XFER_CTRL);
228 }
229 
230 static void qcom_qspi_pio_xfer(struct qcom_qspi *ctrl)
231 {
232 	u32 ints;
233 
234 	qcom_qspi_pio_xfer_cfg(ctrl);
235 
236 	/* Ack any previous interrupts that might be hanging around */
237 	writel(QSPI_ALL_IRQS, ctrl->base + MSTR_INT_STATUS);
238 
239 	/* Setup new interrupts */
240 	if (ctrl->xfer.dir == QSPI_WRITE)
241 		ints = QSPI_ERR_IRQS | WR_FIFO_EMPTY;
242 	else
243 		ints = QSPI_ERR_IRQS | RESP_FIFO_RDY;
244 	writel(ints, ctrl->base + MSTR_INT_EN);
245 
246 	/* Kick off the transfer */
247 	qcom_qspi_pio_xfer_ctrl(ctrl);
248 }
249 
250 static void qcom_qspi_handle_err(struct spi_controller *host,
251 				 struct spi_message *msg)
252 {
253 	u32 int_status;
254 	struct qcom_qspi *ctrl = spi_controller_get_devdata(host);
255 	unsigned long flags;
256 	int i;
257 
258 	spin_lock_irqsave(&ctrl->lock, flags);
259 	writel(0, ctrl->base + MSTR_INT_EN);
260 	int_status = readl(ctrl->base + MSTR_INT_STATUS);
261 	writel(int_status, ctrl->base + MSTR_INT_STATUS);
262 	ctrl->xfer.rem_bytes = 0;
263 
264 	/* free cmd descriptors if they are around (DMA mode) */
265 	for (i = 0; i < ctrl->n_cmd_desc; i++)
266 		dma_pool_free(ctrl->dma_cmd_pool, ctrl->virt_cmd_desc[i],
267 				  ctrl->dma_cmd_desc[i]);
268 	ctrl->n_cmd_desc = 0;
269 	spin_unlock_irqrestore(&ctrl->lock, flags);
270 }
271 
272 static int qcom_qspi_set_speed(struct qcom_qspi *ctrl, unsigned long speed_hz)
273 {
274 	int ret;
275 	unsigned int avg_bw_cpu;
276 
277 	if (speed_hz == ctrl->last_speed)
278 		return 0;
279 
280 	/* In regular operation (SBL_EN=1) core must be 4x transfer clock */
281 	ret = dev_pm_opp_set_rate(ctrl->dev, speed_hz * 4);
282 	if (ret) {
283 		dev_err(ctrl->dev, "Failed to set core clk %d\n", ret);
284 		return ret;
285 	}
286 
287 	/*
288 	 * Set BW quota for CPU.
289 	 * We don't have explicit peak requirement so keep it equal to avg_bw.
290 	 */
291 	avg_bw_cpu = Bps_to_icc(speed_hz);
292 	ret = icc_set_bw(ctrl->icc_path_cpu_to_qspi, avg_bw_cpu, avg_bw_cpu);
293 	if (ret) {
294 		dev_err(ctrl->dev, "%s: ICC BW voting failed for cpu: %d\n",
295 			__func__, ret);
296 		return ret;
297 	}
298 
299 	ctrl->last_speed = speed_hz;
300 
301 	return 0;
302 }
303 
304 static int qcom_qspi_alloc_desc(struct qcom_qspi *ctrl, dma_addr_t dma_ptr,
305 			uint32_t n_bytes)
306 {
307 	struct qspi_cmd_desc *virt_cmd_desc, *prev;
308 	dma_addr_t dma_cmd_desc;
309 
310 	/* allocate for dma cmd descriptor */
311 	virt_cmd_desc = dma_pool_alloc(ctrl->dma_cmd_pool, GFP_ATOMIC | __GFP_ZERO, &dma_cmd_desc);
312 	if (!virt_cmd_desc) {
313 		dev_warn_once(ctrl->dev, "Couldn't find memory for descriptor\n");
314 		return -EAGAIN;
315 	}
316 
317 	ctrl->virt_cmd_desc[ctrl->n_cmd_desc] = virt_cmd_desc;
318 	ctrl->dma_cmd_desc[ctrl->n_cmd_desc] = dma_cmd_desc;
319 	ctrl->n_cmd_desc++;
320 
321 	/* setup cmd descriptor */
322 	virt_cmd_desc->data_address = dma_ptr;
323 	virt_cmd_desc->direction = ctrl->xfer.dir;
324 	virt_cmd_desc->multi_io_mode = qspi_buswidth_to_iomode(ctrl, ctrl->xfer.buswidth);
325 	virt_cmd_desc->fragment = !ctrl->xfer.is_last;
326 	virt_cmd_desc->length = n_bytes;
327 
328 	/* update previous descriptor */
329 	if (ctrl->n_cmd_desc >= 2) {
330 		prev = (ctrl->virt_cmd_desc)[ctrl->n_cmd_desc - 2];
331 		prev->next_descriptor = dma_cmd_desc;
332 		prev->fragment = 1;
333 	}
334 
335 	return 0;
336 }
337 
338 static int qcom_qspi_setup_dma_desc(struct qcom_qspi *ctrl,
339 				struct spi_transfer *xfer)
340 {
341 	int ret;
342 	struct sg_table *sgt;
343 	dma_addr_t dma_ptr_sg;
344 	unsigned int dma_len_sg;
345 	int i;
346 
347 	if (ctrl->n_cmd_desc) {
348 		dev_err(ctrl->dev, "Remnant dma buffers n_cmd_desc-%d\n", ctrl->n_cmd_desc);
349 		return -EIO;
350 	}
351 
352 	sgt = (ctrl->xfer.dir == QSPI_READ) ? &xfer->rx_sg : &xfer->tx_sg;
353 	if (!sgt->nents || sgt->nents > QSPI_MAX_SG) {
354 		dev_warn_once(ctrl->dev, "Cannot handle %d entries in scatter list\n", sgt->nents);
355 		return -EAGAIN;
356 	}
357 
358 	for (i = 0; i < sgt->nents; i++) {
359 		dma_ptr_sg = sg_dma_address(sgt->sgl + i);
360 		dma_len_sg = sg_dma_len(sgt->sgl + i);
361 		if (!IS_ALIGNED(dma_ptr_sg, QSPI_ALIGN_REQ)) {
362 			dev_warn_once(ctrl->dev, "dma_address not aligned to %d\n", QSPI_ALIGN_REQ);
363 			return -EAGAIN;
364 		}
365 		/*
366 		 * When reading with DMA the controller writes to memory 1 word
367 		 * at a time. If the length isn't a multiple of 4 bytes then
368 		 * the controller can clobber the things later in memory.
369 		 * Fallback to PIO to be safe.
370 		 */
371 		if (ctrl->xfer.dir == QSPI_READ && (dma_len_sg & 0x03)) {
372 			dev_warn_once(ctrl->dev, "fallback to PIO for read of size %#010x\n",
373 				      dma_len_sg);
374 			return -EAGAIN;
375 		}
376 	}
377 
378 	for (i = 0; i < sgt->nents; i++) {
379 		dma_ptr_sg = sg_dma_address(sgt->sgl + i);
380 		dma_len_sg = sg_dma_len(sgt->sgl + i);
381 
382 		ret = qcom_qspi_alloc_desc(ctrl, dma_ptr_sg, dma_len_sg);
383 		if (ret)
384 			goto cleanup;
385 	}
386 	return 0;
387 
388 cleanup:
389 	for (i = 0; i < ctrl->n_cmd_desc; i++)
390 		dma_pool_free(ctrl->dma_cmd_pool, ctrl->virt_cmd_desc[i],
391 				  ctrl->dma_cmd_desc[i]);
392 	ctrl->n_cmd_desc = 0;
393 	return ret;
394 }
395 
396 static void qcom_qspi_dma_xfer(struct qcom_qspi *ctrl)
397 {
398 	/* Setup new interrupts */
399 	writel(DMA_CHAIN_DONE, ctrl->base + MSTR_INT_EN);
400 
401 	/* kick off transfer */
402 	writel((u32)((ctrl->dma_cmd_desc)[0]), ctrl->base + NEXT_DMA_DESC_ADDR);
403 }
404 
405 /* Switch to DMA if transfer length exceeds this */
406 #define QSPI_MAX_BYTES_FIFO 64
407 
408 static bool qcom_qspi_can_dma(struct spi_controller *ctlr,
409 			 struct spi_device *slv, struct spi_transfer *xfer)
410 {
411 	return xfer->len > QSPI_MAX_BYTES_FIFO;
412 }
413 
414 static int qcom_qspi_transfer_one(struct spi_controller *host,
415 				  struct spi_device *slv,
416 				  struct spi_transfer *xfer)
417 {
418 	struct qcom_qspi *ctrl = spi_controller_get_devdata(host);
419 	int ret;
420 	unsigned long speed_hz;
421 	unsigned long flags;
422 	u32 mstr_cfg;
423 
424 	speed_hz = slv->max_speed_hz;
425 	if (xfer->speed_hz)
426 		speed_hz = xfer->speed_hz;
427 
428 	ret = qcom_qspi_set_speed(ctrl, speed_hz);
429 	if (ret)
430 		return ret;
431 
432 	spin_lock_irqsave(&ctrl->lock, flags);
433 	mstr_cfg = readl(ctrl->base + MSTR_CONFIG);
434 
435 	/* We are half duplex, so either rx or tx will be set */
436 	if (xfer->rx_buf) {
437 		ctrl->xfer.dir = QSPI_READ;
438 		ctrl->xfer.buswidth = xfer->rx_nbits;
439 		ctrl->xfer.rx_buf = xfer->rx_buf;
440 	} else {
441 		ctrl->xfer.dir = QSPI_WRITE;
442 		ctrl->xfer.buswidth = xfer->tx_nbits;
443 		ctrl->xfer.tx_buf = xfer->tx_buf;
444 	}
445 	ctrl->xfer.is_last = list_is_last(&xfer->transfer_list,
446 					  &host->cur_msg->transfers);
447 	ctrl->xfer.rem_bytes = xfer->len;
448 
449 	if (xfer->rx_sg.nents || xfer->tx_sg.nents) {
450 		/* do DMA transfer */
451 		if (!(mstr_cfg & DMA_ENABLE)) {
452 			mstr_cfg |= DMA_ENABLE;
453 			writel(mstr_cfg, ctrl->base + MSTR_CONFIG);
454 		}
455 
456 		ret = qcom_qspi_setup_dma_desc(ctrl, xfer);
457 		if (ret != -EAGAIN) {
458 			if (!ret) {
459 				dma_wmb();
460 				qcom_qspi_dma_xfer(ctrl);
461 			}
462 			goto exit;
463 		}
464 		dev_warn_once(ctrl->dev, "DMA failure, falling back to PIO\n");
465 		ret = 0; /* We'll retry w/ PIO */
466 	}
467 
468 	if (mstr_cfg & DMA_ENABLE) {
469 		mstr_cfg &= ~DMA_ENABLE;
470 		writel(mstr_cfg, ctrl->base + MSTR_CONFIG);
471 	}
472 	qcom_qspi_pio_xfer(ctrl);
473 
474 exit:
475 	spin_unlock_irqrestore(&ctrl->lock, flags);
476 
477 	if (ret)
478 		return ret;
479 
480 	/* We'll call spi_finalize_current_transfer() when done */
481 	return 1;
482 }
483 
484 static int qcom_qspi_prepare_message(struct spi_controller *host,
485 				     struct spi_message *message)
486 {
487 	u32 mstr_cfg;
488 	struct qcom_qspi *ctrl;
489 	int tx_data_oe_delay = 1;
490 	int tx_data_delay = 1;
491 	unsigned long flags;
492 
493 	ctrl = spi_controller_get_devdata(host);
494 	spin_lock_irqsave(&ctrl->lock, flags);
495 
496 	mstr_cfg = readl(ctrl->base + MSTR_CONFIG);
497 	mstr_cfg &= ~CHIP_SELECT_NUM;
498 	if (spi_get_chipselect(message->spi, 0))
499 		mstr_cfg |= CHIP_SELECT_NUM;
500 
501 	mstr_cfg |= FB_CLK_EN | PIN_WPN | PIN_HOLDN | SBL_EN | FULL_CYCLE_MODE;
502 	mstr_cfg &= ~(SPI_MODE_MSK | TX_DATA_OE_DELAY_MSK | TX_DATA_DELAY_MSK);
503 	mstr_cfg |= message->spi->mode << SPI_MODE_SHFT;
504 	mstr_cfg |= tx_data_oe_delay << TX_DATA_OE_DELAY_SHFT;
505 	mstr_cfg |= tx_data_delay << TX_DATA_DELAY_SHFT;
506 	mstr_cfg &= ~DMA_ENABLE;
507 
508 	writel(mstr_cfg, ctrl->base + MSTR_CONFIG);
509 	spin_unlock_irqrestore(&ctrl->lock, flags);
510 
511 	return 0;
512 }
513 
514 static int qcom_qspi_alloc_dma(struct qcom_qspi *ctrl)
515 {
516 	ctrl->dma_cmd_pool = dmam_pool_create("qspi cmd desc pool",
517 		ctrl->dev, sizeof(struct qspi_cmd_desc), 0, 0);
518 	if (!ctrl->dma_cmd_pool)
519 		return -ENOMEM;
520 
521 	return 0;
522 }
523 
524 static irqreturn_t pio_read(struct qcom_qspi *ctrl)
525 {
526 	u32 rd_fifo_status;
527 	u32 rd_fifo;
528 	unsigned int wr_cnts;
529 	unsigned int bytes_to_read;
530 	unsigned int words_to_read;
531 	u32 *word_buf;
532 	u8 *byte_buf;
533 	int i;
534 
535 	rd_fifo_status = readl(ctrl->base + RD_FIFO_STATUS);
536 
537 	if (!(rd_fifo_status & FIFO_RDY)) {
538 		dev_dbg(ctrl->dev, "Spurious IRQ %#x\n", rd_fifo_status);
539 		return IRQ_NONE;
540 	}
541 
542 	wr_cnts = (rd_fifo_status & WR_CNTS_MSK) >> WR_CNTS_SHFT;
543 	wr_cnts = min(wr_cnts, ctrl->xfer.rem_bytes);
544 
545 	words_to_read = wr_cnts / QSPI_BYTES_PER_WORD;
546 	bytes_to_read = wr_cnts % QSPI_BYTES_PER_WORD;
547 
548 	if (words_to_read) {
549 		word_buf = ctrl->xfer.rx_buf;
550 		ctrl->xfer.rem_bytes -= words_to_read * QSPI_BYTES_PER_WORD;
551 		ioread32_rep(ctrl->base + RD_FIFO, word_buf, words_to_read);
552 		ctrl->xfer.rx_buf = word_buf + words_to_read;
553 	}
554 
555 	if (bytes_to_read) {
556 		byte_buf = ctrl->xfer.rx_buf;
557 		rd_fifo = readl(ctrl->base + RD_FIFO);
558 		ctrl->xfer.rem_bytes -= bytes_to_read;
559 		for (i = 0; i < bytes_to_read; i++)
560 			*byte_buf++ = rd_fifo >> (i * BITS_PER_BYTE);
561 		ctrl->xfer.rx_buf = byte_buf;
562 	}
563 
564 	return IRQ_HANDLED;
565 }
566 
567 static irqreturn_t pio_write(struct qcom_qspi *ctrl)
568 {
569 	const void *xfer_buf = ctrl->xfer.tx_buf;
570 	const int *word_buf;
571 	const char *byte_buf;
572 	unsigned int wr_fifo_bytes;
573 	unsigned int wr_fifo_words;
574 	unsigned int wr_size;
575 	unsigned int rem_words;
576 
577 	wr_fifo_bytes = readl(ctrl->base + PIO_XFER_STATUS);
578 	wr_fifo_bytes >>= WR_FIFO_BYTES_SHFT;
579 
580 	if (ctrl->xfer.rem_bytes < QSPI_BYTES_PER_WORD) {
581 		/* Process the last 1-3 bytes */
582 		wr_size = min(wr_fifo_bytes, ctrl->xfer.rem_bytes);
583 		ctrl->xfer.rem_bytes -= wr_size;
584 
585 		byte_buf = xfer_buf;
586 		while (wr_size--)
587 			writel(*byte_buf++,
588 			       ctrl->base + PIO_DATAOUT_1B);
589 		ctrl->xfer.tx_buf = byte_buf;
590 	} else {
591 		/*
592 		 * Process all the whole words; to keep things simple we'll
593 		 * just wait for the next interrupt to handle the last 1-3
594 		 * bytes if we don't have an even number of words.
595 		 */
596 		rem_words = ctrl->xfer.rem_bytes / QSPI_BYTES_PER_WORD;
597 		wr_fifo_words = wr_fifo_bytes / QSPI_BYTES_PER_WORD;
598 
599 		wr_size = min(rem_words, wr_fifo_words);
600 		ctrl->xfer.rem_bytes -= wr_size * QSPI_BYTES_PER_WORD;
601 
602 		word_buf = xfer_buf;
603 		iowrite32_rep(ctrl->base + PIO_DATAOUT_4B, word_buf, wr_size);
604 		ctrl->xfer.tx_buf = word_buf + wr_size;
605 
606 	}
607 
608 	return IRQ_HANDLED;
609 }
610 
611 static irqreturn_t qcom_qspi_irq(int irq, void *dev_id)
612 {
613 	u32 int_status;
614 	struct qcom_qspi *ctrl = dev_id;
615 	irqreturn_t ret = IRQ_NONE;
616 
617 	spin_lock(&ctrl->lock);
618 
619 	int_status = readl(ctrl->base + MSTR_INT_STATUS);
620 	writel(int_status, ctrl->base + MSTR_INT_STATUS);
621 
622 	/* Ignore disabled interrupts */
623 	int_status &= readl(ctrl->base + MSTR_INT_EN);
624 
625 	/* PIO mode handling */
626 	if (ctrl->xfer.dir == QSPI_WRITE) {
627 		if (int_status & WR_FIFO_EMPTY)
628 			ret = pio_write(ctrl);
629 	} else {
630 		if (int_status & RESP_FIFO_RDY)
631 			ret = pio_read(ctrl);
632 	}
633 
634 	if (int_status & QSPI_ERR_IRQS) {
635 		if (int_status & RESP_FIFO_UNDERRUN)
636 			dev_err(ctrl->dev, "IRQ error: FIFO underrun\n");
637 		if (int_status & WR_FIFO_OVERRUN)
638 			dev_err(ctrl->dev, "IRQ error: FIFO overrun\n");
639 		if (int_status & HRESP_FROM_NOC_ERR)
640 			dev_err(ctrl->dev, "IRQ error: NOC response error\n");
641 		ret = IRQ_HANDLED;
642 	}
643 
644 	if (!ctrl->xfer.rem_bytes) {
645 		writel(0, ctrl->base + MSTR_INT_EN);
646 		spi_finalize_current_transfer(dev_get_drvdata(ctrl->dev));
647 	}
648 
649 	/* DMA mode handling */
650 	if (int_status & DMA_CHAIN_DONE) {
651 		int i;
652 
653 		writel(0, ctrl->base + MSTR_INT_EN);
654 		ctrl->xfer.rem_bytes = 0;
655 
656 		for (i = 0; i < ctrl->n_cmd_desc; i++)
657 			dma_pool_free(ctrl->dma_cmd_pool, ctrl->virt_cmd_desc[i],
658 					  ctrl->dma_cmd_desc[i]);
659 		ctrl->n_cmd_desc = 0;
660 
661 		ret = IRQ_HANDLED;
662 		spi_finalize_current_transfer(dev_get_drvdata(ctrl->dev));
663 	}
664 
665 	spin_unlock(&ctrl->lock);
666 	return ret;
667 }
668 
669 static int qcom_qspi_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
670 {
671 	/*
672 	 * If qcom_qspi_can_dma() is going to return false we don't need to
673 	 * adjust anything.
674 	 */
675 	if (op->data.nbytes <= QSPI_MAX_BYTES_FIFO)
676 		return 0;
677 
678 	/*
679 	 * When reading, the transfer needs to be a multiple of 4 bytes so
680 	 * shrink the transfer if that's not true. The caller will then do a
681 	 * second transfer to finish things up.
682 	 */
683 	if (op->data.dir == SPI_MEM_DATA_IN && (op->data.nbytes & 0x3))
684 		op->data.nbytes &= ~0x3;
685 
686 	return 0;
687 }
688 
689 static const struct spi_controller_mem_ops qcom_qspi_mem_ops = {
690 	.adjust_op_size = qcom_qspi_adjust_op_size,
691 };
692 
693 static int qcom_qspi_probe(struct platform_device *pdev)
694 {
695 	int ret;
696 	struct device *dev;
697 	struct spi_controller *host;
698 	struct qcom_qspi *ctrl;
699 
700 	dev = &pdev->dev;
701 
702 	host = devm_spi_alloc_host(dev, sizeof(*ctrl));
703 	if (!host)
704 		return -ENOMEM;
705 
706 	platform_set_drvdata(pdev, host);
707 
708 	ctrl = spi_controller_get_devdata(host);
709 
710 	spin_lock_init(&ctrl->lock);
711 	ctrl->dev = dev;
712 	ctrl->base = devm_platform_ioremap_resource(pdev, 0);
713 	if (IS_ERR(ctrl->base))
714 		return PTR_ERR(ctrl->base);
715 
716 	ctrl->clks = devm_kcalloc(dev, QSPI_NUM_CLKS,
717 				  sizeof(*ctrl->clks), GFP_KERNEL);
718 	if (!ctrl->clks)
719 		return -ENOMEM;
720 
721 	ctrl->clks[QSPI_CLK_CORE].id = "core";
722 	ctrl->clks[QSPI_CLK_IFACE].id = "iface";
723 	ret = devm_clk_bulk_get(dev, QSPI_NUM_CLKS, ctrl->clks);
724 	if (ret)
725 		return ret;
726 
727 	ctrl->icc_path_cpu_to_qspi = devm_of_icc_get(dev, "qspi-config");
728 	if (IS_ERR(ctrl->icc_path_cpu_to_qspi))
729 		return dev_err_probe(dev, PTR_ERR(ctrl->icc_path_cpu_to_qspi),
730 				     "Failed to get cpu path\n");
731 
732 	/* Set BW vote for register access */
733 	ret = icc_set_bw(ctrl->icc_path_cpu_to_qspi, Bps_to_icc(1000),
734 				Bps_to_icc(1000));
735 	if (ret) {
736 		dev_err(ctrl->dev, "%s: ICC BW voting failed for cpu: %d\n",
737 				__func__, ret);
738 		return ret;
739 	}
740 
741 	ret = icc_disable(ctrl->icc_path_cpu_to_qspi);
742 	if (ret) {
743 		dev_err(ctrl->dev, "%s: ICC disable failed for cpu: %d\n",
744 				__func__, ret);
745 		return ret;
746 	}
747 
748 	ret = platform_get_irq(pdev, 0);
749 	if (ret < 0)
750 		return ret;
751 	ret = devm_request_irq(dev, ret, qcom_qspi_irq, 0, dev_name(dev), ctrl);
752 	if (ret) {
753 		dev_err(dev, "Failed to request irq %d\n", ret);
754 		return ret;
755 	}
756 
757 	ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
758 	if (ret)
759 		return dev_err_probe(dev, ret, "could not set DMA mask\n");
760 
761 	host->max_speed_hz = 300000000;
762 	host->max_dma_len = 65536; /* as per HPG */
763 	host->dma_alignment = QSPI_ALIGN_REQ;
764 	host->num_chipselect = QSPI_NUM_CS;
765 	host->bus_num = -1;
766 	host->dev.of_node = pdev->dev.of_node;
767 	host->mode_bits = SPI_MODE_0 |
768 			  SPI_TX_DUAL | SPI_RX_DUAL |
769 			  SPI_TX_QUAD | SPI_RX_QUAD;
770 	host->flags = SPI_CONTROLLER_HALF_DUPLEX;
771 	host->prepare_message = qcom_qspi_prepare_message;
772 	host->transfer_one = qcom_qspi_transfer_one;
773 	host->handle_err = qcom_qspi_handle_err;
774 	if (of_property_read_bool(pdev->dev.of_node, "iommus"))
775 		host->can_dma = qcom_qspi_can_dma;
776 	host->auto_runtime_pm = true;
777 	host->mem_ops = &qcom_qspi_mem_ops;
778 
779 	ret = devm_pm_opp_set_clkname(&pdev->dev, "core");
780 	if (ret)
781 		return ret;
782 	/* OPP table is optional */
783 	ret = devm_pm_opp_of_add_table(&pdev->dev);
784 	if (ret && ret != -ENODEV) {
785 		dev_err(&pdev->dev, "invalid OPP table in device tree\n");
786 		return ret;
787 	}
788 
789 	ret = qcom_qspi_alloc_dma(ctrl);
790 	if (ret)
791 		return ret;
792 
793 	pm_runtime_use_autosuspend(dev);
794 	pm_runtime_set_autosuspend_delay(dev, 250);
795 	pm_runtime_enable(dev);
796 
797 	ret = spi_register_controller(host);
798 	if (!ret)
799 		return 0;
800 
801 	pm_runtime_disable(dev);
802 
803 	return ret;
804 }
805 
806 static void qcom_qspi_remove(struct platform_device *pdev)
807 {
808 	struct spi_controller *host = platform_get_drvdata(pdev);
809 
810 	/* Unregister _before_ disabling pm_runtime() so we stop transfers */
811 	spi_unregister_controller(host);
812 
813 	pm_runtime_disable(&pdev->dev);
814 }
815 
816 static int __maybe_unused qcom_qspi_runtime_suspend(struct device *dev)
817 {
818 	struct spi_controller *host = dev_get_drvdata(dev);
819 	struct qcom_qspi *ctrl = spi_controller_get_devdata(host);
820 	int ret;
821 
822 	/* Drop the performance state vote */
823 	dev_pm_opp_set_rate(dev, 0);
824 	clk_bulk_disable_unprepare(QSPI_NUM_CLKS, ctrl->clks);
825 
826 	ret = icc_disable(ctrl->icc_path_cpu_to_qspi);
827 	if (ret) {
828 		dev_err_ratelimited(ctrl->dev, "%s: ICC disable failed for cpu: %d\n",
829 			__func__, ret);
830 		return ret;
831 	}
832 
833 	pinctrl_pm_select_sleep_state(dev);
834 
835 	return 0;
836 }
837 
838 static int __maybe_unused qcom_qspi_runtime_resume(struct device *dev)
839 {
840 	struct spi_controller *host = dev_get_drvdata(dev);
841 	struct qcom_qspi *ctrl = spi_controller_get_devdata(host);
842 	int ret;
843 
844 	pinctrl_pm_select_default_state(dev);
845 
846 	ret = icc_enable(ctrl->icc_path_cpu_to_qspi);
847 	if (ret) {
848 		dev_err_ratelimited(ctrl->dev, "%s: ICC enable failed for cpu: %d\n",
849 			__func__, ret);
850 		return ret;
851 	}
852 
853 	ret = clk_bulk_prepare_enable(QSPI_NUM_CLKS, ctrl->clks);
854 	if (ret)
855 		return ret;
856 
857 	return dev_pm_opp_set_rate(dev, ctrl->last_speed * 4);
858 }
859 
860 static int __maybe_unused qcom_qspi_suspend(struct device *dev)
861 {
862 	struct spi_controller *host = dev_get_drvdata(dev);
863 	int ret;
864 
865 	ret = spi_controller_suspend(host);
866 	if (ret)
867 		return ret;
868 
869 	ret = pm_runtime_force_suspend(dev);
870 	if (ret)
871 		spi_controller_resume(host);
872 
873 	return ret;
874 }
875 
876 static int __maybe_unused qcom_qspi_resume(struct device *dev)
877 {
878 	struct spi_controller *host = dev_get_drvdata(dev);
879 	int ret;
880 
881 	ret = pm_runtime_force_resume(dev);
882 	if (ret)
883 		return ret;
884 
885 	ret = spi_controller_resume(host);
886 	if (ret)
887 		pm_runtime_force_suspend(dev);
888 
889 	return ret;
890 }
891 
892 static const struct dev_pm_ops qcom_qspi_dev_pm_ops = {
893 	SET_RUNTIME_PM_OPS(qcom_qspi_runtime_suspend,
894 			   qcom_qspi_runtime_resume, NULL)
895 	SET_SYSTEM_SLEEP_PM_OPS(qcom_qspi_suspend, qcom_qspi_resume)
896 };
897 
898 static const struct of_device_id qcom_qspi_dt_match[] = {
899 	{ .compatible = "qcom,qspi-v1", },
900 	{ }
901 };
902 MODULE_DEVICE_TABLE(of, qcom_qspi_dt_match);
903 
904 static struct platform_driver qcom_qspi_driver = {
905 	.driver = {
906 		.name		= "qcom_qspi",
907 		.pm		= &qcom_qspi_dev_pm_ops,
908 		.of_match_table = qcom_qspi_dt_match,
909 	},
910 	.probe = qcom_qspi_probe,
911 	.remove_new = qcom_qspi_remove,
912 };
913 module_platform_driver(qcom_qspi_driver);
914 
915 MODULE_DESCRIPTION("SPI driver for QSPI cores");
916 MODULE_LICENSE("GPL v2");
917