xref: /openbmc/linux/drivers/spi/spi-geni-qcom.c (revision 92d0c2eb)
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/dmaengine.h>
6 #include <linux/dma-mapping.h>
7 #include <linux/dma/qcom-gpi-dma.h>
8 #include <linux/interrupt.h>
9 #include <linux/io.h>
10 #include <linux/log2.h>
11 #include <linux/module.h>
12 #include <linux/platform_device.h>
13 #include <linux/pm_opp.h>
14 #include <linux/pm_runtime.h>
15 #include <linux/qcom-geni-se.h>
16 #include <linux/spi/spi.h>
17 #include <linux/spinlock.h>
18 
19 /* SPI SE specific registers and respective register fields */
20 #define SE_SPI_CPHA		0x224
21 #define CPHA			BIT(0)
22 
23 #define SE_SPI_LOOPBACK		0x22c
24 #define LOOPBACK_ENABLE		0x1
25 #define NORMAL_MODE		0x0
26 #define LOOPBACK_MSK		GENMASK(1, 0)
27 
28 #define SE_SPI_CPOL		0x230
29 #define CPOL			BIT(2)
30 
31 #define SE_SPI_DEMUX_OUTPUT_INV	0x24c
32 #define CS_DEMUX_OUTPUT_INV_MSK	GENMASK(3, 0)
33 
34 #define SE_SPI_DEMUX_SEL	0x250
35 #define CS_DEMUX_OUTPUT_SEL	GENMASK(3, 0)
36 
37 #define SE_SPI_TRANS_CFG	0x25c
38 #define CS_TOGGLE		BIT(0)
39 
40 #define SE_SPI_WORD_LEN		0x268
41 #define WORD_LEN_MSK		GENMASK(9, 0)
42 #define MIN_WORD_LEN		4
43 
44 #define SE_SPI_TX_TRANS_LEN	0x26c
45 #define SE_SPI_RX_TRANS_LEN	0x270
46 #define TRANS_LEN_MSK		GENMASK(23, 0)
47 
48 #define SE_SPI_PRE_POST_CMD_DLY	0x274
49 
50 #define SE_SPI_DELAY_COUNTERS	0x278
51 #define SPI_INTER_WORDS_DELAY_MSK	GENMASK(9, 0)
52 #define SPI_CS_CLK_DELAY_MSK		GENMASK(19, 10)
53 #define SPI_CS_CLK_DELAY_SHFT		10
54 
55 /* M_CMD OP codes for SPI */
56 #define SPI_TX_ONLY		1
57 #define SPI_RX_ONLY		2
58 #define SPI_TX_RX		7
59 #define SPI_CS_ASSERT		8
60 #define SPI_CS_DEASSERT		9
61 #define SPI_SCK_ONLY		10
62 /* M_CMD params for SPI */
63 #define SPI_PRE_CMD_DELAY	BIT(0)
64 #define TIMESTAMP_BEFORE	BIT(1)
65 #define FRAGMENTATION		BIT(2)
66 #define TIMESTAMP_AFTER		BIT(3)
67 #define POST_CMD_DELAY		BIT(4)
68 
69 #define GSI_LOOPBACK_EN		BIT(0)
70 #define GSI_CS_TOGGLE		BIT(3)
71 #define GSI_CPHA		BIT(4)
72 #define GSI_CPOL		BIT(5)
73 
74 struct spi_geni_master {
75 	struct geni_se se;
76 	struct device *dev;
77 	u32 tx_fifo_depth;
78 	u32 fifo_width_bits;
79 	u32 tx_wm;
80 	u32 last_mode;
81 	unsigned long cur_speed_hz;
82 	unsigned long cur_sclk_hz;
83 	unsigned int cur_bits_per_word;
84 	unsigned int tx_rem_bytes;
85 	unsigned int rx_rem_bytes;
86 	const struct spi_transfer *cur_xfer;
87 	struct completion cs_done;
88 	struct completion cancel_done;
89 	struct completion abort_done;
90 	unsigned int oversampling;
91 	spinlock_t lock;
92 	int irq;
93 	bool cs_flag;
94 	bool abort_failed;
95 	struct dma_chan *tx;
96 	struct dma_chan *rx;
97 	int cur_xfer_mode;
98 };
99 
100 static int get_spi_clk_cfg(unsigned int speed_hz,
101 			struct spi_geni_master *mas,
102 			unsigned int *clk_idx,
103 			unsigned int *clk_div)
104 {
105 	unsigned long sclk_freq;
106 	unsigned int actual_hz;
107 	int ret;
108 
109 	ret = geni_se_clk_freq_match(&mas->se,
110 				speed_hz * mas->oversampling,
111 				clk_idx, &sclk_freq, false);
112 	if (ret) {
113 		dev_err(mas->dev, "Failed(%d) to find src clk for %dHz\n",
114 							ret, speed_hz);
115 		return ret;
116 	}
117 
118 	*clk_div = DIV_ROUND_UP(sclk_freq, mas->oversampling * speed_hz);
119 	actual_hz = sclk_freq / (mas->oversampling * *clk_div);
120 
121 	dev_dbg(mas->dev, "req %u=>%u sclk %lu, idx %d, div %d\n", speed_hz,
122 				actual_hz, sclk_freq, *clk_idx, *clk_div);
123 	ret = dev_pm_opp_set_rate(mas->dev, sclk_freq);
124 	if (ret)
125 		dev_err(mas->dev, "dev_pm_opp_set_rate failed %d\n", ret);
126 	else
127 		mas->cur_sclk_hz = sclk_freq;
128 
129 	return ret;
130 }
131 
132 static void handle_fifo_timeout(struct spi_master *spi,
133 				struct spi_message *msg)
134 {
135 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
136 	unsigned long time_left;
137 	struct geni_se *se = &mas->se;
138 
139 	spin_lock_irq(&mas->lock);
140 	reinit_completion(&mas->cancel_done);
141 	writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
142 	mas->cur_xfer = NULL;
143 	geni_se_cancel_m_cmd(se);
144 	spin_unlock_irq(&mas->lock);
145 
146 	time_left = wait_for_completion_timeout(&mas->cancel_done, HZ);
147 	if (time_left)
148 		return;
149 
150 	spin_lock_irq(&mas->lock);
151 	reinit_completion(&mas->abort_done);
152 	geni_se_abort_m_cmd(se);
153 	spin_unlock_irq(&mas->lock);
154 
155 	time_left = wait_for_completion_timeout(&mas->abort_done, HZ);
156 	if (!time_left) {
157 		dev_err(mas->dev, "Failed to cancel/abort m_cmd\n");
158 
159 		/*
160 		 * No need for a lock since SPI core has a lock and we never
161 		 * access this from an interrupt.
162 		 */
163 		mas->abort_failed = true;
164 	}
165 }
166 
167 static void handle_gpi_timeout(struct spi_master *spi, struct spi_message *msg)
168 {
169 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
170 
171 	dmaengine_terminate_sync(mas->tx);
172 	dmaengine_terminate_sync(mas->rx);
173 }
174 
175 static void spi_geni_handle_err(struct spi_master *spi, struct spi_message *msg)
176 {
177 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
178 
179 	switch (mas->cur_xfer_mode) {
180 	case GENI_SE_FIFO:
181 		handle_fifo_timeout(spi, msg);
182 		break;
183 	case GENI_GPI_DMA:
184 		handle_gpi_timeout(spi, msg);
185 		break;
186 	default:
187 		dev_err(mas->dev, "Abort on Mode:%d not supported", mas->cur_xfer_mode);
188 	}
189 }
190 
191 static bool spi_geni_is_abort_still_pending(struct spi_geni_master *mas)
192 {
193 	struct geni_se *se = &mas->se;
194 	u32 m_irq, m_irq_en;
195 
196 	if (!mas->abort_failed)
197 		return false;
198 
199 	/*
200 	 * The only known case where a transfer times out and then a cancel
201 	 * times out then an abort times out is if something is blocking our
202 	 * interrupt handler from running.  Avoid starting any new transfers
203 	 * until that sorts itself out.
204 	 */
205 	spin_lock_irq(&mas->lock);
206 	m_irq = readl(se->base + SE_GENI_M_IRQ_STATUS);
207 	m_irq_en = readl(se->base + SE_GENI_M_IRQ_EN);
208 	spin_unlock_irq(&mas->lock);
209 
210 	if (m_irq & m_irq_en) {
211 		dev_err(mas->dev, "Interrupts pending after abort: %#010x\n",
212 			m_irq & m_irq_en);
213 		return true;
214 	}
215 
216 	/*
217 	 * If we're here the problem resolved itself so no need to check more
218 	 * on future transfers.
219 	 */
220 	mas->abort_failed = false;
221 
222 	return false;
223 }
224 
225 static void spi_geni_set_cs(struct spi_device *slv, bool set_flag)
226 {
227 	struct spi_geni_master *mas = spi_master_get_devdata(slv->master);
228 	struct spi_master *spi = dev_get_drvdata(mas->dev);
229 	struct geni_se *se = &mas->se;
230 	unsigned long time_left;
231 
232 	if (!(slv->mode & SPI_CS_HIGH))
233 		set_flag = !set_flag;
234 
235 	if (set_flag == mas->cs_flag)
236 		return;
237 
238 	pm_runtime_get_sync(mas->dev);
239 
240 	if (spi_geni_is_abort_still_pending(mas)) {
241 		dev_err(mas->dev, "Can't set chip select\n");
242 		goto exit;
243 	}
244 
245 	spin_lock_irq(&mas->lock);
246 	if (mas->cur_xfer) {
247 		dev_err(mas->dev, "Can't set CS when prev xfer running\n");
248 		spin_unlock_irq(&mas->lock);
249 		goto exit;
250 	}
251 
252 	mas->cs_flag = set_flag;
253 	reinit_completion(&mas->cs_done);
254 	if (set_flag)
255 		geni_se_setup_m_cmd(se, SPI_CS_ASSERT, 0);
256 	else
257 		geni_se_setup_m_cmd(se, SPI_CS_DEASSERT, 0);
258 	spin_unlock_irq(&mas->lock);
259 
260 	time_left = wait_for_completion_timeout(&mas->cs_done, HZ);
261 	if (!time_left) {
262 		dev_warn(mas->dev, "Timeout setting chip select\n");
263 		handle_fifo_timeout(spi, NULL);
264 	}
265 
266 exit:
267 	pm_runtime_put(mas->dev);
268 }
269 
270 static void spi_setup_word_len(struct spi_geni_master *mas, u16 mode,
271 					unsigned int bits_per_word)
272 {
273 	unsigned int pack_words;
274 	bool msb_first = (mode & SPI_LSB_FIRST) ? false : true;
275 	struct geni_se *se = &mas->se;
276 	u32 word_len;
277 
278 	/*
279 	 * If bits_per_word isn't a byte aligned value, set the packing to be
280 	 * 1 SPI word per FIFO word.
281 	 */
282 	if (!(mas->fifo_width_bits % bits_per_word))
283 		pack_words = mas->fifo_width_bits / bits_per_word;
284 	else
285 		pack_words = 1;
286 	geni_se_config_packing(&mas->se, bits_per_word, pack_words, msb_first,
287 								true, true);
288 	word_len = (bits_per_word - MIN_WORD_LEN) & WORD_LEN_MSK;
289 	writel(word_len, se->base + SE_SPI_WORD_LEN);
290 }
291 
292 static int geni_spi_set_clock_and_bw(struct spi_geni_master *mas,
293 					unsigned long clk_hz)
294 {
295 	u32 clk_sel, m_clk_cfg, idx, div;
296 	struct geni_se *se = &mas->se;
297 	int ret;
298 
299 	if (clk_hz == mas->cur_speed_hz)
300 		return 0;
301 
302 	ret = get_spi_clk_cfg(clk_hz, mas, &idx, &div);
303 	if (ret) {
304 		dev_err(mas->dev, "Err setting clk to %lu: %d\n", clk_hz, ret);
305 		return ret;
306 	}
307 
308 	/*
309 	 * SPI core clock gets configured with the requested frequency
310 	 * or the frequency closer to the requested frequency.
311 	 * For that reason requested frequency is stored in the
312 	 * cur_speed_hz and referred in the consecutive transfer instead
313 	 * of calling clk_get_rate() API.
314 	 */
315 	mas->cur_speed_hz = clk_hz;
316 
317 	clk_sel = idx & CLK_SEL_MSK;
318 	m_clk_cfg = (div << CLK_DIV_SHFT) | SER_CLK_EN;
319 	writel(clk_sel, se->base + SE_GENI_CLK_SEL);
320 	writel(m_clk_cfg, se->base + GENI_SER_M_CLK_CFG);
321 
322 	/* Set BW quota for CPU as driver supports FIFO mode only. */
323 	se->icc_paths[CPU_TO_GENI].avg_bw = Bps_to_icc(mas->cur_speed_hz);
324 	ret = geni_icc_set_bw(se);
325 	if (ret)
326 		return ret;
327 
328 	return 0;
329 }
330 
331 static int setup_fifo_params(struct spi_device *spi_slv,
332 					struct spi_master *spi)
333 {
334 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
335 	struct geni_se *se = &mas->se;
336 	u32 loopback_cfg = 0, cpol = 0, cpha = 0, demux_output_inv = 0;
337 	u32 demux_sel;
338 
339 	if (mas->last_mode != spi_slv->mode) {
340 		if (spi_slv->mode & SPI_LOOP)
341 			loopback_cfg = LOOPBACK_ENABLE;
342 
343 		if (spi_slv->mode & SPI_CPOL)
344 			cpol = CPOL;
345 
346 		if (spi_slv->mode & SPI_CPHA)
347 			cpha = CPHA;
348 
349 		if (spi_slv->mode & SPI_CS_HIGH)
350 			demux_output_inv = BIT(spi_slv->chip_select);
351 
352 		demux_sel = spi_slv->chip_select;
353 		mas->cur_bits_per_word = spi_slv->bits_per_word;
354 
355 		spi_setup_word_len(mas, spi_slv->mode, spi_slv->bits_per_word);
356 		writel(loopback_cfg, se->base + SE_SPI_LOOPBACK);
357 		writel(demux_sel, se->base + SE_SPI_DEMUX_SEL);
358 		writel(cpha, se->base + SE_SPI_CPHA);
359 		writel(cpol, se->base + SE_SPI_CPOL);
360 		writel(demux_output_inv, se->base + SE_SPI_DEMUX_OUTPUT_INV);
361 
362 		mas->last_mode = spi_slv->mode;
363 	}
364 
365 	return geni_spi_set_clock_and_bw(mas, spi_slv->max_speed_hz);
366 }
367 
368 static void
369 spi_gsi_callback_result(void *cb, const struct dmaengine_result *result)
370 {
371 	struct spi_master *spi = cb;
372 
373 	spi->cur_msg->status = -EIO;
374 	if (result->result != DMA_TRANS_NOERROR) {
375 		dev_err(&spi->dev, "DMA txn failed: %d\n", result->result);
376 		spi_finalize_current_transfer(spi);
377 		return;
378 	}
379 
380 	if (!result->residue) {
381 		spi->cur_msg->status = 0;
382 		dev_dbg(&spi->dev, "DMA txn completed\n");
383 	} else {
384 		dev_err(&spi->dev, "DMA xfer has pending: %d\n", result->residue);
385 	}
386 
387 	spi_finalize_current_transfer(spi);
388 }
389 
390 static int setup_gsi_xfer(struct spi_transfer *xfer, struct spi_geni_master *mas,
391 			  struct spi_device *spi_slv, struct spi_master *spi)
392 {
393 	unsigned long flags = DMA_PREP_INTERRUPT | DMA_CTRL_ACK;
394 	struct dma_slave_config config = {};
395 	struct gpi_spi_config peripheral = {};
396 	struct dma_async_tx_descriptor *tx_desc, *rx_desc;
397 	int ret;
398 
399 	config.peripheral_config = &peripheral;
400 	config.peripheral_size = sizeof(peripheral);
401 	peripheral.set_config = true;
402 
403 	if (xfer->bits_per_word != mas->cur_bits_per_word ||
404 	    xfer->speed_hz != mas->cur_speed_hz) {
405 		mas->cur_bits_per_word = xfer->bits_per_word;
406 		mas->cur_speed_hz = xfer->speed_hz;
407 	}
408 
409 	if (xfer->tx_buf && xfer->rx_buf) {
410 		peripheral.cmd = SPI_DUPLEX;
411 	} else if (xfer->tx_buf) {
412 		peripheral.cmd = SPI_TX;
413 		peripheral.rx_len = 0;
414 	} else if (xfer->rx_buf) {
415 		peripheral.cmd = SPI_RX;
416 		if (!(mas->cur_bits_per_word % MIN_WORD_LEN)) {
417 			peripheral.rx_len = ((xfer->len << 3) / mas->cur_bits_per_word);
418 		} else {
419 			int bytes_per_word = (mas->cur_bits_per_word / BITS_PER_BYTE) + 1;
420 
421 			peripheral.rx_len = (xfer->len / bytes_per_word);
422 		}
423 	}
424 
425 	peripheral.loopback_en = !!(spi_slv->mode & SPI_LOOP);
426 	peripheral.clock_pol_high = !!(spi_slv->mode & SPI_CPOL);
427 	peripheral.data_pol_high = !!(spi_slv->mode & SPI_CPHA);
428 	peripheral.cs = spi_slv->chip_select;
429 	peripheral.pack_en = true;
430 	peripheral.word_len = xfer->bits_per_word - MIN_WORD_LEN;
431 
432 	ret = get_spi_clk_cfg(mas->cur_speed_hz, mas,
433 			      &peripheral.clk_src, &peripheral.clk_div);
434 	if (ret) {
435 		dev_err(mas->dev, "Err in get_spi_clk_cfg() :%d\n", ret);
436 		return ret;
437 	}
438 
439 	if (!xfer->cs_change) {
440 		if (!list_is_last(&xfer->transfer_list, &spi->cur_msg->transfers))
441 			peripheral.fragmentation = FRAGMENTATION;
442 	}
443 
444 	if (peripheral.cmd & SPI_RX) {
445 		dmaengine_slave_config(mas->rx, &config);
446 		rx_desc = dmaengine_prep_slave_sg(mas->rx, xfer->rx_sg.sgl, xfer->rx_sg.nents,
447 						  DMA_DEV_TO_MEM, flags);
448 		if (!rx_desc) {
449 			dev_err(mas->dev, "Err setting up rx desc\n");
450 			return -EIO;
451 		}
452 	}
453 
454 	/*
455 	 * Prepare the TX always, even for RX or tx_buf being null, we would
456 	 * need TX to be prepared per GSI spec
457 	 */
458 	dmaengine_slave_config(mas->tx, &config);
459 	tx_desc = dmaengine_prep_slave_sg(mas->tx, xfer->tx_sg.sgl, xfer->tx_sg.nents,
460 					  DMA_MEM_TO_DEV, flags);
461 	if (!tx_desc) {
462 		dev_err(mas->dev, "Err setting up tx desc\n");
463 		return -EIO;
464 	}
465 
466 	tx_desc->callback_result = spi_gsi_callback_result;
467 	tx_desc->callback_param = spi;
468 
469 	if (peripheral.cmd & SPI_RX)
470 		dmaengine_submit(rx_desc);
471 	dmaengine_submit(tx_desc);
472 
473 	if (peripheral.cmd & SPI_RX)
474 		dma_async_issue_pending(mas->rx);
475 
476 	dma_async_issue_pending(mas->tx);
477 	return 1;
478 }
479 
480 static bool geni_can_dma(struct spi_controller *ctlr,
481 			 struct spi_device *slv, struct spi_transfer *xfer)
482 {
483 	struct spi_geni_master *mas = spi_master_get_devdata(slv->master);
484 
485 	/* check if dma is supported */
486 	return mas->cur_xfer_mode != GENI_SE_FIFO;
487 }
488 
489 static int spi_geni_prepare_message(struct spi_master *spi,
490 					struct spi_message *spi_msg)
491 {
492 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
493 	int ret;
494 
495 	switch (mas->cur_xfer_mode) {
496 	case GENI_SE_FIFO:
497 		if (spi_geni_is_abort_still_pending(mas))
498 			return -EBUSY;
499 		ret = setup_fifo_params(spi_msg->spi, spi);
500 		if (ret)
501 			dev_err(mas->dev, "Couldn't select mode %d\n", ret);
502 		return ret;
503 
504 	case GENI_GPI_DMA:
505 		/* nothing to do for GPI DMA */
506 		return 0;
507 	}
508 
509 	dev_err(mas->dev, "Mode not supported %d", mas->cur_xfer_mode);
510 	return -EINVAL;
511 }
512 
513 static int spi_geni_grab_gpi_chan(struct spi_geni_master *mas)
514 {
515 	int ret;
516 
517 	mas->tx = dma_request_chan(mas->dev, "tx");
518 	if (IS_ERR(mas->tx)) {
519 		ret = dev_err_probe(mas->dev, PTR_ERR(mas->tx),
520 				    "Failed to get tx DMA ch\n");
521 		goto err_tx;
522 	}
523 
524 	mas->rx = dma_request_chan(mas->dev, "rx");
525 	if (IS_ERR(mas->rx)) {
526 		ret = dev_err_probe(mas->dev, PTR_ERR(mas->rx),
527 				    "Failed to get rx DMA ch\n");
528 		goto err_rx;
529 	}
530 
531 	return 0;
532 
533 err_rx:
534 	mas->rx = NULL;
535 	dma_release_channel(mas->tx);
536 err_tx:
537 	mas->tx = NULL;
538 	return ret;
539 }
540 
541 static void spi_geni_release_dma_chan(struct spi_geni_master *mas)
542 {
543 	if (mas->rx) {
544 		dma_release_channel(mas->rx);
545 		mas->rx = NULL;
546 	}
547 
548 	if (mas->tx) {
549 		dma_release_channel(mas->tx);
550 		mas->tx = NULL;
551 	}
552 }
553 
554 static int spi_geni_init(struct spi_geni_master *mas)
555 {
556 	struct geni_se *se = &mas->se;
557 	unsigned int proto, major, minor, ver;
558 	u32 spi_tx_cfg, fifo_disable;
559 	int ret = -ENXIO;
560 
561 	pm_runtime_get_sync(mas->dev);
562 
563 	proto = geni_se_read_proto(se);
564 	if (proto != GENI_SE_SPI) {
565 		dev_err(mas->dev, "Invalid proto %d\n", proto);
566 		goto out_pm;
567 	}
568 	mas->tx_fifo_depth = geni_se_get_tx_fifo_depth(se);
569 
570 	/* Width of Tx and Rx FIFO is same */
571 	mas->fifo_width_bits = geni_se_get_tx_fifo_width(se);
572 
573 	/*
574 	 * Hardware programming guide suggests to configure
575 	 * RX FIFO RFR level to fifo_depth-2.
576 	 */
577 	geni_se_init(se, mas->tx_fifo_depth - 3, mas->tx_fifo_depth - 2);
578 	/* Transmit an entire FIFO worth of data per IRQ */
579 	mas->tx_wm = 1;
580 	ver = geni_se_get_qup_hw_version(se);
581 	major = GENI_SE_VERSION_MAJOR(ver);
582 	minor = GENI_SE_VERSION_MINOR(ver);
583 
584 	if (major == 1 && minor == 0)
585 		mas->oversampling = 2;
586 	else
587 		mas->oversampling = 1;
588 
589 	fifo_disable = readl(se->base + GENI_IF_DISABLE_RO) & FIFO_IF_DISABLE;
590 	switch (fifo_disable) {
591 	case 1:
592 		ret = spi_geni_grab_gpi_chan(mas);
593 		if (!ret) { /* success case */
594 			mas->cur_xfer_mode = GENI_GPI_DMA;
595 			geni_se_select_mode(se, GENI_GPI_DMA);
596 			dev_dbg(mas->dev, "Using GPI DMA mode for SPI\n");
597 			break;
598 		}
599 		/*
600 		 * in case of failure to get dma channel, we can still do the
601 		 * FIFO mode, so fallthrough
602 		 */
603 		dev_warn(mas->dev, "FIFO mode disabled, but couldn't get DMA, fall back to FIFO mode\n");
604 		fallthrough;
605 
606 	case 0:
607 		mas->cur_xfer_mode = GENI_SE_FIFO;
608 		geni_se_select_mode(se, GENI_SE_FIFO);
609 		ret = 0;
610 		break;
611 	}
612 
613 	/* We always control CS manually */
614 	spi_tx_cfg = readl(se->base + SE_SPI_TRANS_CFG);
615 	spi_tx_cfg &= ~CS_TOGGLE;
616 	writel(spi_tx_cfg, se->base + SE_SPI_TRANS_CFG);
617 
618 out_pm:
619 	pm_runtime_put(mas->dev);
620 	return ret;
621 }
622 
623 static unsigned int geni_byte_per_fifo_word(struct spi_geni_master *mas)
624 {
625 	/*
626 	 * Calculate how many bytes we'll put in each FIFO word.  If the
627 	 * transfer words don't pack cleanly into a FIFO word we'll just put
628 	 * one transfer word in each FIFO word.  If they do pack we'll pack 'em.
629 	 */
630 	if (mas->fifo_width_bits % mas->cur_bits_per_word)
631 		return roundup_pow_of_two(DIV_ROUND_UP(mas->cur_bits_per_word,
632 						       BITS_PER_BYTE));
633 
634 	return mas->fifo_width_bits / BITS_PER_BYTE;
635 }
636 
637 static bool geni_spi_handle_tx(struct spi_geni_master *mas)
638 {
639 	struct geni_se *se = &mas->se;
640 	unsigned int max_bytes;
641 	const u8 *tx_buf;
642 	unsigned int bytes_per_fifo_word = geni_byte_per_fifo_word(mas);
643 	unsigned int i = 0;
644 
645 	/* Stop the watermark IRQ if nothing to send */
646 	if (!mas->cur_xfer) {
647 		writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
648 		return false;
649 	}
650 
651 	max_bytes = (mas->tx_fifo_depth - mas->tx_wm) * bytes_per_fifo_word;
652 	if (mas->tx_rem_bytes < max_bytes)
653 		max_bytes = mas->tx_rem_bytes;
654 
655 	tx_buf = mas->cur_xfer->tx_buf + mas->cur_xfer->len - mas->tx_rem_bytes;
656 	while (i < max_bytes) {
657 		unsigned int j;
658 		unsigned int bytes_to_write;
659 		u32 fifo_word = 0;
660 		u8 *fifo_byte = (u8 *)&fifo_word;
661 
662 		bytes_to_write = min(bytes_per_fifo_word, max_bytes - i);
663 		for (j = 0; j < bytes_to_write; j++)
664 			fifo_byte[j] = tx_buf[i++];
665 		iowrite32_rep(se->base + SE_GENI_TX_FIFOn, &fifo_word, 1);
666 	}
667 	mas->tx_rem_bytes -= max_bytes;
668 	if (!mas->tx_rem_bytes) {
669 		writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
670 		return false;
671 	}
672 	return true;
673 }
674 
675 static void geni_spi_handle_rx(struct spi_geni_master *mas)
676 {
677 	struct geni_se *se = &mas->se;
678 	u32 rx_fifo_status;
679 	unsigned int rx_bytes;
680 	unsigned int rx_last_byte_valid;
681 	u8 *rx_buf;
682 	unsigned int bytes_per_fifo_word = geni_byte_per_fifo_word(mas);
683 	unsigned int i = 0;
684 
685 	rx_fifo_status = readl(se->base + SE_GENI_RX_FIFO_STATUS);
686 	rx_bytes = (rx_fifo_status & RX_FIFO_WC_MSK) * bytes_per_fifo_word;
687 	if (rx_fifo_status & RX_LAST) {
688 		rx_last_byte_valid = rx_fifo_status & RX_LAST_BYTE_VALID_MSK;
689 		rx_last_byte_valid >>= RX_LAST_BYTE_VALID_SHFT;
690 		if (rx_last_byte_valid && rx_last_byte_valid < 4)
691 			rx_bytes -= bytes_per_fifo_word - rx_last_byte_valid;
692 	}
693 
694 	/* Clear out the FIFO and bail if nowhere to put it */
695 	if (!mas->cur_xfer) {
696 		for (i = 0; i < DIV_ROUND_UP(rx_bytes, bytes_per_fifo_word); i++)
697 			readl(se->base + SE_GENI_RX_FIFOn);
698 		return;
699 	}
700 
701 	if (mas->rx_rem_bytes < rx_bytes)
702 		rx_bytes = mas->rx_rem_bytes;
703 
704 	rx_buf = mas->cur_xfer->rx_buf + mas->cur_xfer->len - mas->rx_rem_bytes;
705 	while (i < rx_bytes) {
706 		u32 fifo_word = 0;
707 		u8 *fifo_byte = (u8 *)&fifo_word;
708 		unsigned int bytes_to_read;
709 		unsigned int j;
710 
711 		bytes_to_read = min(bytes_per_fifo_word, rx_bytes - i);
712 		ioread32_rep(se->base + SE_GENI_RX_FIFOn, &fifo_word, 1);
713 		for (j = 0; j < bytes_to_read; j++)
714 			rx_buf[i++] = fifo_byte[j];
715 	}
716 	mas->rx_rem_bytes -= rx_bytes;
717 }
718 
719 static void setup_fifo_xfer(struct spi_transfer *xfer,
720 				struct spi_geni_master *mas,
721 				u16 mode, struct spi_master *spi)
722 {
723 	u32 m_cmd = 0;
724 	u32 len;
725 	struct geni_se *se = &mas->se;
726 	int ret;
727 
728 	/*
729 	 * Ensure that our interrupt handler isn't still running from some
730 	 * prior command before we start messing with the hardware behind
731 	 * its back.  We don't need to _keep_ the lock here since we're only
732 	 * worried about racing with out interrupt handler.  The SPI core
733 	 * already handles making sure that we're not trying to do two
734 	 * transfers at once or setting a chip select and doing a transfer
735 	 * concurrently.
736 	 *
737 	 * NOTE: we actually _can't_ hold the lock here because possibly we
738 	 * might call clk_set_rate() which needs to be able to sleep.
739 	 */
740 	spin_lock_irq(&mas->lock);
741 	spin_unlock_irq(&mas->lock);
742 
743 	if (xfer->bits_per_word != mas->cur_bits_per_word) {
744 		spi_setup_word_len(mas, mode, xfer->bits_per_word);
745 		mas->cur_bits_per_word = xfer->bits_per_word;
746 	}
747 
748 	/* Speed and bits per word can be overridden per transfer */
749 	ret = geni_spi_set_clock_and_bw(mas, xfer->speed_hz);
750 	if (ret)
751 		return;
752 
753 	mas->tx_rem_bytes = 0;
754 	mas->rx_rem_bytes = 0;
755 
756 	if (!(mas->cur_bits_per_word % MIN_WORD_LEN))
757 		len = xfer->len * BITS_PER_BYTE / mas->cur_bits_per_word;
758 	else
759 		len = xfer->len / (mas->cur_bits_per_word / BITS_PER_BYTE + 1);
760 	len &= TRANS_LEN_MSK;
761 
762 	mas->cur_xfer = xfer;
763 	if (xfer->tx_buf) {
764 		m_cmd |= SPI_TX_ONLY;
765 		mas->tx_rem_bytes = xfer->len;
766 		writel(len, se->base + SE_SPI_TX_TRANS_LEN);
767 	}
768 
769 	if (xfer->rx_buf) {
770 		m_cmd |= SPI_RX_ONLY;
771 		writel(len, se->base + SE_SPI_RX_TRANS_LEN);
772 		mas->rx_rem_bytes = xfer->len;
773 	}
774 
775 	/*
776 	 * Lock around right before we start the transfer since our
777 	 * interrupt could come in at any time now.
778 	 */
779 	spin_lock_irq(&mas->lock);
780 	geni_se_setup_m_cmd(se, m_cmd, FRAGMENTATION);
781 	if (m_cmd & SPI_TX_ONLY) {
782 		if (geni_spi_handle_tx(mas))
783 			writel(mas->tx_wm, se->base + SE_GENI_TX_WATERMARK_REG);
784 	}
785 	spin_unlock_irq(&mas->lock);
786 }
787 
788 static int spi_geni_transfer_one(struct spi_master *spi,
789 				struct spi_device *slv,
790 				struct spi_transfer *xfer)
791 {
792 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
793 
794 	if (spi_geni_is_abort_still_pending(mas))
795 		return -EBUSY;
796 
797 	/* Terminate and return success for 0 byte length transfer */
798 	if (!xfer->len)
799 		return 0;
800 
801 	if (mas->cur_xfer_mode == GENI_SE_FIFO) {
802 		setup_fifo_xfer(xfer, mas, slv->mode, spi);
803 		return 1;
804 	}
805 	return setup_gsi_xfer(xfer, mas, slv, spi);
806 }
807 
808 static irqreturn_t geni_spi_isr(int irq, void *data)
809 {
810 	struct spi_master *spi = data;
811 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
812 	struct geni_se *se = &mas->se;
813 	u32 m_irq;
814 
815 	m_irq = readl(se->base + SE_GENI_M_IRQ_STATUS);
816 	if (!m_irq)
817 		return IRQ_NONE;
818 
819 	if (m_irq & (M_CMD_OVERRUN_EN | M_ILLEGAL_CMD_EN | M_CMD_FAILURE_EN |
820 		     M_RX_FIFO_RD_ERR_EN | M_RX_FIFO_WR_ERR_EN |
821 		     M_TX_FIFO_RD_ERR_EN | M_TX_FIFO_WR_ERR_EN))
822 		dev_warn(mas->dev, "Unexpected IRQ err status %#010x\n", m_irq);
823 
824 	spin_lock(&mas->lock);
825 
826 	if ((m_irq & M_RX_FIFO_WATERMARK_EN) || (m_irq & M_RX_FIFO_LAST_EN))
827 		geni_spi_handle_rx(mas);
828 
829 	if (m_irq & M_TX_FIFO_WATERMARK_EN)
830 		geni_spi_handle_tx(mas);
831 
832 	if (m_irq & M_CMD_DONE_EN) {
833 		if (mas->cur_xfer) {
834 			spi_finalize_current_transfer(spi);
835 			mas->cur_xfer = NULL;
836 			/*
837 			 * If this happens, then a CMD_DONE came before all the
838 			 * Tx buffer bytes were sent out. This is unusual, log
839 			 * this condition and disable the WM interrupt to
840 			 * prevent the system from stalling due an interrupt
841 			 * storm.
842 			 *
843 			 * If this happens when all Rx bytes haven't been
844 			 * received, log the condition. The only known time
845 			 * this can happen is if bits_per_word != 8 and some
846 			 * registers that expect xfer lengths in num spi_words
847 			 * weren't written correctly.
848 			 */
849 			if (mas->tx_rem_bytes) {
850 				writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
851 				dev_err(mas->dev, "Premature done. tx_rem = %d bpw%d\n",
852 					mas->tx_rem_bytes, mas->cur_bits_per_word);
853 			}
854 			if (mas->rx_rem_bytes)
855 				dev_err(mas->dev, "Premature done. rx_rem = %d bpw%d\n",
856 					mas->rx_rem_bytes, mas->cur_bits_per_word);
857 		} else {
858 			complete(&mas->cs_done);
859 		}
860 	}
861 
862 	if (m_irq & M_CMD_CANCEL_EN)
863 		complete(&mas->cancel_done);
864 	if (m_irq & M_CMD_ABORT_EN)
865 		complete(&mas->abort_done);
866 
867 	/*
868 	 * It's safe or a good idea to Ack all of our interrupts at the end
869 	 * of the function. Specifically:
870 	 * - M_CMD_DONE_EN / M_RX_FIFO_LAST_EN: Edge triggered interrupts and
871 	 *   clearing Acks. Clearing at the end relies on nobody else having
872 	 *   started a new transfer yet or else we could be clearing _their_
873 	 *   done bit, but everyone grabs the spinlock before starting a new
874 	 *   transfer.
875 	 * - M_RX_FIFO_WATERMARK_EN / M_TX_FIFO_WATERMARK_EN: These appear
876 	 *   to be "latched level" interrupts so it's important to clear them
877 	 *   _after_ you've handled the condition and always safe to do so
878 	 *   since they'll re-assert if they're still happening.
879 	 */
880 	writel(m_irq, se->base + SE_GENI_M_IRQ_CLEAR);
881 
882 	spin_unlock(&mas->lock);
883 
884 	return IRQ_HANDLED;
885 }
886 
887 static int spi_geni_probe(struct platform_device *pdev)
888 {
889 	int ret, irq;
890 	struct spi_master *spi;
891 	struct spi_geni_master *mas;
892 	void __iomem *base;
893 	struct clk *clk;
894 	struct device *dev = &pdev->dev;
895 
896 	irq = platform_get_irq(pdev, 0);
897 	if (irq < 0)
898 		return irq;
899 
900 	ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64));
901 	if (ret) {
902 		ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
903 		if (ret)
904 			return dev_err_probe(dev, ret, "could not set DMA mask\n");
905 	}
906 
907 	base = devm_platform_ioremap_resource(pdev, 0);
908 	if (IS_ERR(base))
909 		return PTR_ERR(base);
910 
911 	clk = devm_clk_get(dev, "se");
912 	if (IS_ERR(clk))
913 		return PTR_ERR(clk);
914 
915 	spi = devm_spi_alloc_master(dev, sizeof(*mas));
916 	if (!spi)
917 		return -ENOMEM;
918 
919 	platform_set_drvdata(pdev, spi);
920 	mas = spi_master_get_devdata(spi);
921 	mas->irq = irq;
922 	mas->dev = dev;
923 	mas->se.dev = dev;
924 	mas->se.wrapper = dev_get_drvdata(dev->parent);
925 	mas->se.base = base;
926 	mas->se.clk = clk;
927 
928 	ret = devm_pm_opp_set_clkname(&pdev->dev, "se");
929 	if (ret)
930 		return ret;
931 	/* OPP table is optional */
932 	ret = devm_pm_opp_of_add_table(&pdev->dev);
933 	if (ret && ret != -ENODEV) {
934 		dev_err(&pdev->dev, "invalid OPP table in device tree\n");
935 		return ret;
936 	}
937 
938 	spi->bus_num = -1;
939 	spi->dev.of_node = dev->of_node;
940 	spi->mode_bits = SPI_CPOL | SPI_CPHA | SPI_LOOP | SPI_CS_HIGH;
941 	spi->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
942 	spi->num_chipselect = 4;
943 	spi->max_speed_hz = 50000000;
944 	spi->prepare_message = spi_geni_prepare_message;
945 	spi->transfer_one = spi_geni_transfer_one;
946 	spi->can_dma = geni_can_dma;
947 	spi->dma_map_dev = dev->parent;
948 	spi->auto_runtime_pm = true;
949 	spi->handle_err = spi_geni_handle_err;
950 	spi->use_gpio_descriptors = true;
951 
952 	init_completion(&mas->cs_done);
953 	init_completion(&mas->cancel_done);
954 	init_completion(&mas->abort_done);
955 	spin_lock_init(&mas->lock);
956 	pm_runtime_use_autosuspend(&pdev->dev);
957 	pm_runtime_set_autosuspend_delay(&pdev->dev, 250);
958 	pm_runtime_enable(dev);
959 
960 	ret = geni_icc_get(&mas->se, NULL);
961 	if (ret)
962 		goto spi_geni_probe_runtime_disable;
963 	/* Set the bus quota to a reasonable value for register access */
964 	mas->se.icc_paths[GENI_TO_CORE].avg_bw = Bps_to_icc(CORE_2X_50_MHZ);
965 	mas->se.icc_paths[CPU_TO_GENI].avg_bw = GENI_DEFAULT_BW;
966 
967 	ret = geni_icc_set_bw(&mas->se);
968 	if (ret)
969 		goto spi_geni_probe_runtime_disable;
970 
971 	ret = spi_geni_init(mas);
972 	if (ret)
973 		goto spi_geni_probe_runtime_disable;
974 
975 	/*
976 	 * check the mode supported and set_cs for fifo mode only
977 	 * for dma (gsi) mode, the gsi will set cs based on params passed in
978 	 * TRE
979 	 */
980 	if (mas->cur_xfer_mode == GENI_SE_FIFO)
981 		spi->set_cs = spi_geni_set_cs;
982 
983 	ret = request_irq(mas->irq, geni_spi_isr, 0, dev_name(dev), spi);
984 	if (ret)
985 		goto spi_geni_release_dma;
986 
987 	ret = spi_register_master(spi);
988 	if (ret)
989 		goto spi_geni_probe_free_irq;
990 
991 	return 0;
992 spi_geni_probe_free_irq:
993 	free_irq(mas->irq, spi);
994 spi_geni_release_dma:
995 	spi_geni_release_dma_chan(mas);
996 spi_geni_probe_runtime_disable:
997 	pm_runtime_disable(dev);
998 	return ret;
999 }
1000 
1001 static int spi_geni_remove(struct platform_device *pdev)
1002 {
1003 	struct spi_master *spi = platform_get_drvdata(pdev);
1004 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
1005 
1006 	/* Unregister _before_ disabling pm_runtime() so we stop transfers */
1007 	spi_unregister_master(spi);
1008 
1009 	spi_geni_release_dma_chan(mas);
1010 
1011 	free_irq(mas->irq, spi);
1012 	pm_runtime_disable(&pdev->dev);
1013 	return 0;
1014 }
1015 
1016 static int __maybe_unused spi_geni_runtime_suspend(struct device *dev)
1017 {
1018 	struct spi_master *spi = dev_get_drvdata(dev);
1019 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
1020 	int ret;
1021 
1022 	/* Drop the performance state vote */
1023 	dev_pm_opp_set_rate(dev, 0);
1024 
1025 	ret = geni_se_resources_off(&mas->se);
1026 	if (ret)
1027 		return ret;
1028 
1029 	return geni_icc_disable(&mas->se);
1030 }
1031 
1032 static int __maybe_unused spi_geni_runtime_resume(struct device *dev)
1033 {
1034 	struct spi_master *spi = dev_get_drvdata(dev);
1035 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
1036 	int ret;
1037 
1038 	ret = geni_icc_enable(&mas->se);
1039 	if (ret)
1040 		return ret;
1041 
1042 	ret = geni_se_resources_on(&mas->se);
1043 	if (ret)
1044 		return ret;
1045 
1046 	return dev_pm_opp_set_rate(mas->dev, mas->cur_sclk_hz);
1047 }
1048 
1049 static int __maybe_unused spi_geni_suspend(struct device *dev)
1050 {
1051 	struct spi_master *spi = dev_get_drvdata(dev);
1052 	int ret;
1053 
1054 	ret = spi_master_suspend(spi);
1055 	if (ret)
1056 		return ret;
1057 
1058 	ret = pm_runtime_force_suspend(dev);
1059 	if (ret)
1060 		spi_master_resume(spi);
1061 
1062 	return ret;
1063 }
1064 
1065 static int __maybe_unused spi_geni_resume(struct device *dev)
1066 {
1067 	struct spi_master *spi = dev_get_drvdata(dev);
1068 	int ret;
1069 
1070 	ret = pm_runtime_force_resume(dev);
1071 	if (ret)
1072 		return ret;
1073 
1074 	ret = spi_master_resume(spi);
1075 	if (ret)
1076 		pm_runtime_force_suspend(dev);
1077 
1078 	return ret;
1079 }
1080 
1081 static const struct dev_pm_ops spi_geni_pm_ops = {
1082 	SET_RUNTIME_PM_OPS(spi_geni_runtime_suspend,
1083 					spi_geni_runtime_resume, NULL)
1084 	SET_SYSTEM_SLEEP_PM_OPS(spi_geni_suspend, spi_geni_resume)
1085 };
1086 
1087 static const struct of_device_id spi_geni_dt_match[] = {
1088 	{ .compatible = "qcom,geni-spi" },
1089 	{}
1090 };
1091 MODULE_DEVICE_TABLE(of, spi_geni_dt_match);
1092 
1093 static struct platform_driver spi_geni_driver = {
1094 	.probe  = spi_geni_probe,
1095 	.remove = spi_geni_remove,
1096 	.driver = {
1097 		.name = "geni_spi",
1098 		.pm = &spi_geni_pm_ops,
1099 		.of_match_table = spi_geni_dt_match,
1100 	},
1101 };
1102 module_platform_driver(spi_geni_driver);
1103 
1104 MODULE_DESCRIPTION("SPI driver for GENI based QUP cores");
1105 MODULE_LICENSE("GPL v2");
1106