xref: /openbmc/linux/drivers/spi/spi-qcom-qspi.c (revision 3d40aed8)
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/of_platform.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)
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_master *master,
251 				 struct spi_message *msg)
252 {
253 	u32 int_status;
254 	struct qcom_qspi *ctrl = spi_master_get_devdata(master);
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_KERNEL | __GFP_ZERO, &dma_cmd_desc);
312 	if (!virt_cmd_desc)
313 		return -ENOMEM;
314 
315 	ctrl->virt_cmd_desc[ctrl->n_cmd_desc] = virt_cmd_desc;
316 	ctrl->dma_cmd_desc[ctrl->n_cmd_desc] = dma_cmd_desc;
317 	ctrl->n_cmd_desc++;
318 
319 	/* setup cmd descriptor */
320 	virt_cmd_desc->data_address = dma_ptr;
321 	virt_cmd_desc->direction = ctrl->xfer.dir;
322 	virt_cmd_desc->multi_io_mode = qspi_buswidth_to_iomode(ctrl, ctrl->xfer.buswidth);
323 	virt_cmd_desc->fragment = !ctrl->xfer.is_last;
324 	virt_cmd_desc->length = n_bytes;
325 
326 	/* update previous descriptor */
327 	if (ctrl->n_cmd_desc >= 2) {
328 		prev = (ctrl->virt_cmd_desc)[ctrl->n_cmd_desc - 2];
329 		prev->next_descriptor = dma_cmd_desc;
330 		prev->fragment = 1;
331 	}
332 
333 	return 0;
334 }
335 
336 static int qcom_qspi_setup_dma_desc(struct qcom_qspi *ctrl,
337 				struct spi_transfer *xfer)
338 {
339 	int ret;
340 	struct sg_table *sgt;
341 	dma_addr_t dma_ptr_sg;
342 	unsigned int dma_len_sg;
343 	int i;
344 
345 	if (ctrl->n_cmd_desc) {
346 		dev_err(ctrl->dev, "Remnant dma buffers n_cmd_desc-%d\n", ctrl->n_cmd_desc);
347 		return -EIO;
348 	}
349 
350 	sgt = (ctrl->xfer.dir == QSPI_READ) ? &xfer->rx_sg : &xfer->tx_sg;
351 	if (!sgt->nents || sgt->nents > QSPI_MAX_SG) {
352 		dev_warn_once(ctrl->dev, "Cannot handle %d entries in scatter list\n", sgt->nents);
353 		return -EAGAIN;
354 	}
355 
356 	for (i = 0; i < sgt->nents; i++) {
357 		dma_ptr_sg = sg_dma_address(sgt->sgl + i);
358 		if (!IS_ALIGNED(dma_ptr_sg, QSPI_ALIGN_REQ)) {
359 			dev_warn_once(ctrl->dev, "dma_address not aligned to %d\n", QSPI_ALIGN_REQ);
360 			return -EAGAIN;
361 		}
362 	}
363 
364 	for (i = 0; i < sgt->nents; i++) {
365 		dma_ptr_sg = sg_dma_address(sgt->sgl + i);
366 		dma_len_sg = sg_dma_len(sgt->sgl + i);
367 
368 		ret = qcom_qspi_alloc_desc(ctrl, dma_ptr_sg, dma_len_sg);
369 		if (ret)
370 			goto cleanup;
371 	}
372 	return 0;
373 
374 cleanup:
375 	for (i = 0; i < ctrl->n_cmd_desc; i++)
376 		dma_pool_free(ctrl->dma_cmd_pool, ctrl->virt_cmd_desc[i],
377 				  ctrl->dma_cmd_desc[i]);
378 	ctrl->n_cmd_desc = 0;
379 	return ret;
380 }
381 
382 static void qcom_qspi_dma_xfer(struct qcom_qspi *ctrl)
383 {
384 	/* Setup new interrupts */
385 	writel(DMA_CHAIN_DONE, ctrl->base + MSTR_INT_EN);
386 
387 	/* kick off transfer */
388 	writel((u32)((ctrl->dma_cmd_desc)[0]), ctrl->base + NEXT_DMA_DESC_ADDR);
389 }
390 
391 /* Switch to DMA if transfer length exceeds this */
392 #define QSPI_MAX_BYTES_FIFO 64
393 
394 static bool qcom_qspi_can_dma(struct spi_controller *ctlr,
395 			 struct spi_device *slv, struct spi_transfer *xfer)
396 {
397 	return xfer->len > QSPI_MAX_BYTES_FIFO;
398 }
399 
400 static int qcom_qspi_transfer_one(struct spi_master *master,
401 				  struct spi_device *slv,
402 				  struct spi_transfer *xfer)
403 {
404 	struct qcom_qspi *ctrl = spi_master_get_devdata(master);
405 	int ret;
406 	unsigned long speed_hz;
407 	unsigned long flags;
408 	u32 mstr_cfg;
409 
410 	speed_hz = slv->max_speed_hz;
411 	if (xfer->speed_hz)
412 		speed_hz = xfer->speed_hz;
413 
414 	ret = qcom_qspi_set_speed(ctrl, speed_hz);
415 	if (ret)
416 		return ret;
417 
418 	spin_lock_irqsave(&ctrl->lock, flags);
419 	mstr_cfg = readl(ctrl->base + MSTR_CONFIG);
420 
421 	/* We are half duplex, so either rx or tx will be set */
422 	if (xfer->rx_buf) {
423 		ctrl->xfer.dir = QSPI_READ;
424 		ctrl->xfer.buswidth = xfer->rx_nbits;
425 		ctrl->xfer.rx_buf = xfer->rx_buf;
426 	} else {
427 		ctrl->xfer.dir = QSPI_WRITE;
428 		ctrl->xfer.buswidth = xfer->tx_nbits;
429 		ctrl->xfer.tx_buf = xfer->tx_buf;
430 	}
431 	ctrl->xfer.is_last = list_is_last(&xfer->transfer_list,
432 					  &master->cur_msg->transfers);
433 	ctrl->xfer.rem_bytes = xfer->len;
434 
435 	if (xfer->rx_sg.nents || xfer->tx_sg.nents) {
436 		/* do DMA transfer */
437 		if (!(mstr_cfg & DMA_ENABLE)) {
438 			mstr_cfg |= DMA_ENABLE;
439 			writel(mstr_cfg, ctrl->base + MSTR_CONFIG);
440 		}
441 
442 		ret = qcom_qspi_setup_dma_desc(ctrl, xfer);
443 		if (ret != -EAGAIN) {
444 			if (!ret)
445 				qcom_qspi_dma_xfer(ctrl);
446 			goto exit;
447 		}
448 		dev_warn_once(ctrl->dev, "DMA failure, falling back to PIO\n");
449 		ret = 0; /* We'll retry w/ PIO */
450 	}
451 
452 	if (mstr_cfg & DMA_ENABLE) {
453 		mstr_cfg &= ~DMA_ENABLE;
454 		writel(mstr_cfg, ctrl->base + MSTR_CONFIG);
455 	}
456 	qcom_qspi_pio_xfer(ctrl);
457 
458 exit:
459 	spin_unlock_irqrestore(&ctrl->lock, flags);
460 
461 	if (ret)
462 		return ret;
463 
464 	/* We'll call spi_finalize_current_transfer() when done */
465 	return 1;
466 }
467 
468 static int qcom_qspi_prepare_message(struct spi_master *master,
469 				     struct spi_message *message)
470 {
471 	u32 mstr_cfg;
472 	struct qcom_qspi *ctrl;
473 	int tx_data_oe_delay = 1;
474 	int tx_data_delay = 1;
475 	unsigned long flags;
476 
477 	ctrl = spi_master_get_devdata(master);
478 	spin_lock_irqsave(&ctrl->lock, flags);
479 
480 	mstr_cfg = readl(ctrl->base + MSTR_CONFIG);
481 	mstr_cfg &= ~CHIP_SELECT_NUM;
482 	if (spi_get_chipselect(message->spi, 0))
483 		mstr_cfg |= CHIP_SELECT_NUM;
484 
485 	mstr_cfg |= FB_CLK_EN | PIN_WPN | PIN_HOLDN | SBL_EN | FULL_CYCLE_MODE;
486 	mstr_cfg &= ~(SPI_MODE_MSK | TX_DATA_OE_DELAY_MSK | TX_DATA_DELAY_MSK);
487 	mstr_cfg |= message->spi->mode << SPI_MODE_SHFT;
488 	mstr_cfg |= tx_data_oe_delay << TX_DATA_OE_DELAY_SHFT;
489 	mstr_cfg |= tx_data_delay << TX_DATA_DELAY_SHFT;
490 	mstr_cfg &= ~DMA_ENABLE;
491 
492 	writel(mstr_cfg, ctrl->base + MSTR_CONFIG);
493 	spin_unlock_irqrestore(&ctrl->lock, flags);
494 
495 	return 0;
496 }
497 
498 static int qcom_qspi_alloc_dma(struct qcom_qspi *ctrl)
499 {
500 	ctrl->dma_cmd_pool = dmam_pool_create("qspi cmd desc pool",
501 		ctrl->dev, sizeof(struct qspi_cmd_desc), 0, 0);
502 	if (!ctrl->dma_cmd_pool)
503 		return -ENOMEM;
504 
505 	return 0;
506 }
507 
508 static irqreturn_t pio_read(struct qcom_qspi *ctrl)
509 {
510 	u32 rd_fifo_status;
511 	u32 rd_fifo;
512 	unsigned int wr_cnts;
513 	unsigned int bytes_to_read;
514 	unsigned int words_to_read;
515 	u32 *word_buf;
516 	u8 *byte_buf;
517 	int i;
518 
519 	rd_fifo_status = readl(ctrl->base + RD_FIFO_STATUS);
520 
521 	if (!(rd_fifo_status & FIFO_RDY)) {
522 		dev_dbg(ctrl->dev, "Spurious IRQ %#x\n", rd_fifo_status);
523 		return IRQ_NONE;
524 	}
525 
526 	wr_cnts = (rd_fifo_status & WR_CNTS_MSK) >> WR_CNTS_SHFT;
527 	wr_cnts = min(wr_cnts, ctrl->xfer.rem_bytes);
528 
529 	words_to_read = wr_cnts / QSPI_BYTES_PER_WORD;
530 	bytes_to_read = wr_cnts % QSPI_BYTES_PER_WORD;
531 
532 	if (words_to_read) {
533 		word_buf = ctrl->xfer.rx_buf;
534 		ctrl->xfer.rem_bytes -= words_to_read * QSPI_BYTES_PER_WORD;
535 		ioread32_rep(ctrl->base + RD_FIFO, word_buf, words_to_read);
536 		ctrl->xfer.rx_buf = word_buf + words_to_read;
537 	}
538 
539 	if (bytes_to_read) {
540 		byte_buf = ctrl->xfer.rx_buf;
541 		rd_fifo = readl(ctrl->base + RD_FIFO);
542 		ctrl->xfer.rem_bytes -= bytes_to_read;
543 		for (i = 0; i < bytes_to_read; i++)
544 			*byte_buf++ = rd_fifo >> (i * BITS_PER_BYTE);
545 		ctrl->xfer.rx_buf = byte_buf;
546 	}
547 
548 	return IRQ_HANDLED;
549 }
550 
551 static irqreturn_t pio_write(struct qcom_qspi *ctrl)
552 {
553 	const void *xfer_buf = ctrl->xfer.tx_buf;
554 	const int *word_buf;
555 	const char *byte_buf;
556 	unsigned int wr_fifo_bytes;
557 	unsigned int wr_fifo_words;
558 	unsigned int wr_size;
559 	unsigned int rem_words;
560 
561 	wr_fifo_bytes = readl(ctrl->base + PIO_XFER_STATUS);
562 	wr_fifo_bytes >>= WR_FIFO_BYTES_SHFT;
563 
564 	if (ctrl->xfer.rem_bytes < QSPI_BYTES_PER_WORD) {
565 		/* Process the last 1-3 bytes */
566 		wr_size = min(wr_fifo_bytes, ctrl->xfer.rem_bytes);
567 		ctrl->xfer.rem_bytes -= wr_size;
568 
569 		byte_buf = xfer_buf;
570 		while (wr_size--)
571 			writel(*byte_buf++,
572 			       ctrl->base + PIO_DATAOUT_1B);
573 		ctrl->xfer.tx_buf = byte_buf;
574 	} else {
575 		/*
576 		 * Process all the whole words; to keep things simple we'll
577 		 * just wait for the next interrupt to handle the last 1-3
578 		 * bytes if we don't have an even number of words.
579 		 */
580 		rem_words = ctrl->xfer.rem_bytes / QSPI_BYTES_PER_WORD;
581 		wr_fifo_words = wr_fifo_bytes / QSPI_BYTES_PER_WORD;
582 
583 		wr_size = min(rem_words, wr_fifo_words);
584 		ctrl->xfer.rem_bytes -= wr_size * QSPI_BYTES_PER_WORD;
585 
586 		word_buf = xfer_buf;
587 		iowrite32_rep(ctrl->base + PIO_DATAOUT_4B, word_buf, wr_size);
588 		ctrl->xfer.tx_buf = word_buf + wr_size;
589 
590 	}
591 
592 	return IRQ_HANDLED;
593 }
594 
595 static irqreturn_t qcom_qspi_irq(int irq, void *dev_id)
596 {
597 	u32 int_status;
598 	struct qcom_qspi *ctrl = dev_id;
599 	irqreturn_t ret = IRQ_NONE;
600 
601 	spin_lock(&ctrl->lock);
602 
603 	int_status = readl(ctrl->base + MSTR_INT_STATUS);
604 	writel(int_status, ctrl->base + MSTR_INT_STATUS);
605 
606 	/* PIO mode handling */
607 	if (ctrl->xfer.dir == QSPI_WRITE) {
608 		if (int_status & WR_FIFO_EMPTY)
609 			ret = pio_write(ctrl);
610 	} else {
611 		if (int_status & RESP_FIFO_RDY)
612 			ret = pio_read(ctrl);
613 	}
614 
615 	if (int_status & QSPI_ERR_IRQS) {
616 		if (int_status & RESP_FIFO_UNDERRUN)
617 			dev_err(ctrl->dev, "IRQ error: FIFO underrun\n");
618 		if (int_status & WR_FIFO_OVERRUN)
619 			dev_err(ctrl->dev, "IRQ error: FIFO overrun\n");
620 		if (int_status & HRESP_FROM_NOC_ERR)
621 			dev_err(ctrl->dev, "IRQ error: NOC response error\n");
622 		ret = IRQ_HANDLED;
623 	}
624 
625 	if (!ctrl->xfer.rem_bytes) {
626 		writel(0, ctrl->base + MSTR_INT_EN);
627 		spi_finalize_current_transfer(dev_get_drvdata(ctrl->dev));
628 	}
629 
630 	/* DMA mode handling */
631 	if (int_status & DMA_CHAIN_DONE) {
632 		int i;
633 
634 		writel(0, ctrl->base + MSTR_INT_EN);
635 		ctrl->xfer.rem_bytes = 0;
636 
637 		for (i = 0; i < ctrl->n_cmd_desc; i++)
638 			dma_pool_free(ctrl->dma_cmd_pool, ctrl->virt_cmd_desc[i],
639 					  ctrl->dma_cmd_desc[i]);
640 		ctrl->n_cmd_desc = 0;
641 
642 		ret = IRQ_HANDLED;
643 		spi_finalize_current_transfer(dev_get_drvdata(ctrl->dev));
644 	}
645 
646 	spin_unlock(&ctrl->lock);
647 	return ret;
648 }
649 
650 static int qcom_qspi_probe(struct platform_device *pdev)
651 {
652 	int ret;
653 	struct device *dev;
654 	struct spi_master *master;
655 	struct qcom_qspi *ctrl;
656 
657 	dev = &pdev->dev;
658 
659 	master = devm_spi_alloc_master(dev, sizeof(*ctrl));
660 	if (!master)
661 		return -ENOMEM;
662 
663 	platform_set_drvdata(pdev, master);
664 
665 	ctrl = spi_master_get_devdata(master);
666 
667 	spin_lock_init(&ctrl->lock);
668 	ctrl->dev = dev;
669 	ctrl->base = devm_platform_ioremap_resource(pdev, 0);
670 	if (IS_ERR(ctrl->base))
671 		return PTR_ERR(ctrl->base);
672 
673 	ctrl->clks = devm_kcalloc(dev, QSPI_NUM_CLKS,
674 				  sizeof(*ctrl->clks), GFP_KERNEL);
675 	if (!ctrl->clks)
676 		return -ENOMEM;
677 
678 	ctrl->clks[QSPI_CLK_CORE].id = "core";
679 	ctrl->clks[QSPI_CLK_IFACE].id = "iface";
680 	ret = devm_clk_bulk_get(dev, QSPI_NUM_CLKS, ctrl->clks);
681 	if (ret)
682 		return ret;
683 
684 	ctrl->icc_path_cpu_to_qspi = devm_of_icc_get(dev, "qspi-config");
685 	if (IS_ERR(ctrl->icc_path_cpu_to_qspi))
686 		return dev_err_probe(dev, PTR_ERR(ctrl->icc_path_cpu_to_qspi),
687 				     "Failed to get cpu path\n");
688 
689 	/* Set BW vote for register access */
690 	ret = icc_set_bw(ctrl->icc_path_cpu_to_qspi, Bps_to_icc(1000),
691 				Bps_to_icc(1000));
692 	if (ret) {
693 		dev_err(ctrl->dev, "%s: ICC BW voting failed for cpu: %d\n",
694 				__func__, ret);
695 		return ret;
696 	}
697 
698 	ret = icc_disable(ctrl->icc_path_cpu_to_qspi);
699 	if (ret) {
700 		dev_err(ctrl->dev, "%s: ICC disable failed for cpu: %d\n",
701 				__func__, ret);
702 		return ret;
703 	}
704 
705 	ret = platform_get_irq(pdev, 0);
706 	if (ret < 0)
707 		return ret;
708 	ret = devm_request_irq(dev, ret, qcom_qspi_irq, 0, dev_name(dev), ctrl);
709 	if (ret) {
710 		dev_err(dev, "Failed to request irq %d\n", ret);
711 		return ret;
712 	}
713 
714 	ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
715 	if (ret)
716 		return dev_err_probe(dev, ret, "could not set DMA mask\n");
717 
718 	master->max_speed_hz = 300000000;
719 	master->max_dma_len = 65536; /* as per HPG */
720 	master->dma_alignment = QSPI_ALIGN_REQ;
721 	master->num_chipselect = QSPI_NUM_CS;
722 	master->bus_num = -1;
723 	master->dev.of_node = pdev->dev.of_node;
724 	master->mode_bits = SPI_MODE_0 |
725 			    SPI_TX_DUAL | SPI_RX_DUAL |
726 			    SPI_TX_QUAD | SPI_RX_QUAD;
727 	master->flags = SPI_MASTER_HALF_DUPLEX;
728 	master->prepare_message = qcom_qspi_prepare_message;
729 	master->transfer_one = qcom_qspi_transfer_one;
730 	master->handle_err = qcom_qspi_handle_err;
731 	if (of_property_read_bool(pdev->dev.of_node, "iommus"))
732 		master->can_dma = qcom_qspi_can_dma;
733 	master->auto_runtime_pm = true;
734 
735 	ret = devm_pm_opp_set_clkname(&pdev->dev, "core");
736 	if (ret)
737 		return ret;
738 	/* OPP table is optional */
739 	ret = devm_pm_opp_of_add_table(&pdev->dev);
740 	if (ret && ret != -ENODEV) {
741 		dev_err(&pdev->dev, "invalid OPP table in device tree\n");
742 		return ret;
743 	}
744 
745 	ret = qcom_qspi_alloc_dma(ctrl);
746 	if (ret)
747 		return ret;
748 
749 	pm_runtime_use_autosuspend(dev);
750 	pm_runtime_set_autosuspend_delay(dev, 250);
751 	pm_runtime_enable(dev);
752 
753 	ret = spi_register_master(master);
754 	if (!ret)
755 		return 0;
756 
757 	pm_runtime_disable(dev);
758 
759 	return ret;
760 }
761 
762 static void qcom_qspi_remove(struct platform_device *pdev)
763 {
764 	struct spi_master *master = platform_get_drvdata(pdev);
765 
766 	/* Unregister _before_ disabling pm_runtime() so we stop transfers */
767 	spi_unregister_master(master);
768 
769 	pm_runtime_disable(&pdev->dev);
770 }
771 
772 static int __maybe_unused qcom_qspi_runtime_suspend(struct device *dev)
773 {
774 	struct spi_master *master = dev_get_drvdata(dev);
775 	struct qcom_qspi *ctrl = spi_master_get_devdata(master);
776 	int ret;
777 
778 	/* Drop the performance state vote */
779 	dev_pm_opp_set_rate(dev, 0);
780 	clk_bulk_disable_unprepare(QSPI_NUM_CLKS, ctrl->clks);
781 
782 	ret = icc_disable(ctrl->icc_path_cpu_to_qspi);
783 	if (ret) {
784 		dev_err_ratelimited(ctrl->dev, "%s: ICC disable failed for cpu: %d\n",
785 			__func__, ret);
786 		return ret;
787 	}
788 
789 	pinctrl_pm_select_sleep_state(dev);
790 
791 	return 0;
792 }
793 
794 static int __maybe_unused qcom_qspi_runtime_resume(struct device *dev)
795 {
796 	struct spi_master *master = dev_get_drvdata(dev);
797 	struct qcom_qspi *ctrl = spi_master_get_devdata(master);
798 	int ret;
799 
800 	pinctrl_pm_select_default_state(dev);
801 
802 	ret = icc_enable(ctrl->icc_path_cpu_to_qspi);
803 	if (ret) {
804 		dev_err_ratelimited(ctrl->dev, "%s: ICC enable failed for cpu: %d\n",
805 			__func__, ret);
806 		return ret;
807 	}
808 
809 	ret = clk_bulk_prepare_enable(QSPI_NUM_CLKS, ctrl->clks);
810 	if (ret)
811 		return ret;
812 
813 	return dev_pm_opp_set_rate(dev, ctrl->last_speed * 4);
814 }
815 
816 static int __maybe_unused qcom_qspi_suspend(struct device *dev)
817 {
818 	struct spi_master *master = dev_get_drvdata(dev);
819 	int ret;
820 
821 	ret = spi_master_suspend(master);
822 	if (ret)
823 		return ret;
824 
825 	ret = pm_runtime_force_suspend(dev);
826 	if (ret)
827 		spi_master_resume(master);
828 
829 	return ret;
830 }
831 
832 static int __maybe_unused qcom_qspi_resume(struct device *dev)
833 {
834 	struct spi_master *master = dev_get_drvdata(dev);
835 	int ret;
836 
837 	ret = pm_runtime_force_resume(dev);
838 	if (ret)
839 		return ret;
840 
841 	ret = spi_master_resume(master);
842 	if (ret)
843 		pm_runtime_force_suspend(dev);
844 
845 	return ret;
846 }
847 
848 static const struct dev_pm_ops qcom_qspi_dev_pm_ops = {
849 	SET_RUNTIME_PM_OPS(qcom_qspi_runtime_suspend,
850 			   qcom_qspi_runtime_resume, NULL)
851 	SET_SYSTEM_SLEEP_PM_OPS(qcom_qspi_suspend, qcom_qspi_resume)
852 };
853 
854 static const struct of_device_id qcom_qspi_dt_match[] = {
855 	{ .compatible = "qcom,qspi-v1", },
856 	{ }
857 };
858 MODULE_DEVICE_TABLE(of, qcom_qspi_dt_match);
859 
860 static struct platform_driver qcom_qspi_driver = {
861 	.driver = {
862 		.name		= "qcom_qspi",
863 		.pm		= &qcom_qspi_dev_pm_ops,
864 		.of_match_table = qcom_qspi_dt_match,
865 	},
866 	.probe = qcom_qspi_probe,
867 	.remove_new = qcom_qspi_remove,
868 };
869 module_platform_driver(qcom_qspi_driver);
870 
871 MODULE_DESCRIPTION("SPI driver for QSPI cores");
872 MODULE_LICENSE("GPL v2");
873