xref: /openbmc/linux/drivers/spi/spi-geni-qcom.c (revision 36de991e)
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 #define MAX_TX_SG		3
75 #define NUM_SPI_XFER		8
76 #define SPI_XFER_TIMEOUT_MS	250
77 
78 struct spi_geni_master {
79 	struct geni_se se;
80 	struct device *dev;
81 	u32 tx_fifo_depth;
82 	u32 fifo_width_bits;
83 	u32 tx_wm;
84 	u32 last_mode;
85 	unsigned long cur_speed_hz;
86 	unsigned long cur_sclk_hz;
87 	unsigned int cur_bits_per_word;
88 	unsigned int tx_rem_bytes;
89 	unsigned int rx_rem_bytes;
90 	const struct spi_transfer *cur_xfer;
91 	struct completion cs_done;
92 	struct completion cancel_done;
93 	struct completion abort_done;
94 	unsigned int oversampling;
95 	spinlock_t lock;
96 	int irq;
97 	bool cs_flag;
98 	bool abort_failed;
99 	struct dma_chan *tx;
100 	struct dma_chan *rx;
101 	int cur_xfer_mode;
102 };
103 
104 static int get_spi_clk_cfg(unsigned int speed_hz,
105 			struct spi_geni_master *mas,
106 			unsigned int *clk_idx,
107 			unsigned int *clk_div)
108 {
109 	unsigned long sclk_freq;
110 	unsigned int actual_hz;
111 	int ret;
112 
113 	ret = geni_se_clk_freq_match(&mas->se,
114 				speed_hz * mas->oversampling,
115 				clk_idx, &sclk_freq, false);
116 	if (ret) {
117 		dev_err(mas->dev, "Failed(%d) to find src clk for %dHz\n",
118 							ret, speed_hz);
119 		return ret;
120 	}
121 
122 	*clk_div = DIV_ROUND_UP(sclk_freq, mas->oversampling * speed_hz);
123 	actual_hz = sclk_freq / (mas->oversampling * *clk_div);
124 
125 	dev_dbg(mas->dev, "req %u=>%u sclk %lu, idx %d, div %d\n", speed_hz,
126 				actual_hz, sclk_freq, *clk_idx, *clk_div);
127 	ret = dev_pm_opp_set_rate(mas->dev, sclk_freq);
128 	if (ret)
129 		dev_err(mas->dev, "dev_pm_opp_set_rate failed %d\n", ret);
130 	else
131 		mas->cur_sclk_hz = sclk_freq;
132 
133 	return ret;
134 }
135 
136 static void handle_fifo_timeout(struct spi_master *spi,
137 				struct spi_message *msg)
138 {
139 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
140 	unsigned long time_left;
141 	struct geni_se *se = &mas->se;
142 
143 	spin_lock_irq(&mas->lock);
144 	reinit_completion(&mas->cancel_done);
145 	writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
146 	mas->cur_xfer = NULL;
147 	geni_se_cancel_m_cmd(se);
148 	spin_unlock_irq(&mas->lock);
149 
150 	time_left = wait_for_completion_timeout(&mas->cancel_done, HZ);
151 	if (time_left)
152 		return;
153 
154 	spin_lock_irq(&mas->lock);
155 	reinit_completion(&mas->abort_done);
156 	geni_se_abort_m_cmd(se);
157 	spin_unlock_irq(&mas->lock);
158 
159 	time_left = wait_for_completion_timeout(&mas->abort_done, HZ);
160 	if (!time_left) {
161 		dev_err(mas->dev, "Failed to cancel/abort m_cmd\n");
162 
163 		/*
164 		 * No need for a lock since SPI core has a lock and we never
165 		 * access this from an interrupt.
166 		 */
167 		mas->abort_failed = true;
168 	}
169 }
170 
171 static bool spi_geni_is_abort_still_pending(struct spi_geni_master *mas)
172 {
173 	struct geni_se *se = &mas->se;
174 	u32 m_irq, m_irq_en;
175 
176 	if (!mas->abort_failed)
177 		return false;
178 
179 	/*
180 	 * The only known case where a transfer times out and then a cancel
181 	 * times out then an abort times out is if something is blocking our
182 	 * interrupt handler from running.  Avoid starting any new transfers
183 	 * until that sorts itself out.
184 	 */
185 	spin_lock_irq(&mas->lock);
186 	m_irq = readl(se->base + SE_GENI_M_IRQ_STATUS);
187 	m_irq_en = readl(se->base + SE_GENI_M_IRQ_EN);
188 	spin_unlock_irq(&mas->lock);
189 
190 	if (m_irq & m_irq_en) {
191 		dev_err(mas->dev, "Interrupts pending after abort: %#010x\n",
192 			m_irq & m_irq_en);
193 		return true;
194 	}
195 
196 	/*
197 	 * If we're here the problem resolved itself so no need to check more
198 	 * on future transfers.
199 	 */
200 	mas->abort_failed = false;
201 
202 	return false;
203 }
204 
205 static void spi_geni_set_cs(struct spi_device *slv, bool set_flag)
206 {
207 	struct spi_geni_master *mas = spi_master_get_devdata(slv->master);
208 	struct spi_master *spi = dev_get_drvdata(mas->dev);
209 	struct geni_se *se = &mas->se;
210 	unsigned long time_left;
211 
212 	if (!(slv->mode & SPI_CS_HIGH))
213 		set_flag = !set_flag;
214 
215 	if (set_flag == mas->cs_flag)
216 		return;
217 
218 	pm_runtime_get_sync(mas->dev);
219 
220 	if (spi_geni_is_abort_still_pending(mas)) {
221 		dev_err(mas->dev, "Can't set chip select\n");
222 		goto exit;
223 	}
224 
225 	spin_lock_irq(&mas->lock);
226 	if (mas->cur_xfer) {
227 		dev_err(mas->dev, "Can't set CS when prev xfer running\n");
228 		spin_unlock_irq(&mas->lock);
229 		goto exit;
230 	}
231 
232 	mas->cs_flag = set_flag;
233 	reinit_completion(&mas->cs_done);
234 	if (set_flag)
235 		geni_se_setup_m_cmd(se, SPI_CS_ASSERT, 0);
236 	else
237 		geni_se_setup_m_cmd(se, SPI_CS_DEASSERT, 0);
238 	spin_unlock_irq(&mas->lock);
239 
240 	time_left = wait_for_completion_timeout(&mas->cs_done, HZ);
241 	if (!time_left) {
242 		dev_warn(mas->dev, "Timeout setting chip select\n");
243 		handle_fifo_timeout(spi, NULL);
244 	}
245 
246 exit:
247 	pm_runtime_put(mas->dev);
248 }
249 
250 static void spi_setup_word_len(struct spi_geni_master *mas, u16 mode,
251 					unsigned int bits_per_word)
252 {
253 	unsigned int pack_words;
254 	bool msb_first = (mode & SPI_LSB_FIRST) ? false : true;
255 	struct geni_se *se = &mas->se;
256 	u32 word_len;
257 
258 	/*
259 	 * If bits_per_word isn't a byte aligned value, set the packing to be
260 	 * 1 SPI word per FIFO word.
261 	 */
262 	if (!(mas->fifo_width_bits % bits_per_word))
263 		pack_words = mas->fifo_width_bits / bits_per_word;
264 	else
265 		pack_words = 1;
266 	geni_se_config_packing(&mas->se, bits_per_word, pack_words, msb_first,
267 								true, true);
268 	word_len = (bits_per_word - MIN_WORD_LEN) & WORD_LEN_MSK;
269 	writel(word_len, se->base + SE_SPI_WORD_LEN);
270 }
271 
272 static int geni_spi_set_clock_and_bw(struct spi_geni_master *mas,
273 					unsigned long clk_hz)
274 {
275 	u32 clk_sel, m_clk_cfg, idx, div;
276 	struct geni_se *se = &mas->se;
277 	int ret;
278 
279 	if (clk_hz == mas->cur_speed_hz)
280 		return 0;
281 
282 	ret = get_spi_clk_cfg(clk_hz, mas, &idx, &div);
283 	if (ret) {
284 		dev_err(mas->dev, "Err setting clk to %lu: %d\n", clk_hz, ret);
285 		return ret;
286 	}
287 
288 	/*
289 	 * SPI core clock gets configured with the requested frequency
290 	 * or the frequency closer to the requested frequency.
291 	 * For that reason requested frequency is stored in the
292 	 * cur_speed_hz and referred in the consecutive transfer instead
293 	 * of calling clk_get_rate() API.
294 	 */
295 	mas->cur_speed_hz = clk_hz;
296 
297 	clk_sel = idx & CLK_SEL_MSK;
298 	m_clk_cfg = (div << CLK_DIV_SHFT) | SER_CLK_EN;
299 	writel(clk_sel, se->base + SE_GENI_CLK_SEL);
300 	writel(m_clk_cfg, se->base + GENI_SER_M_CLK_CFG);
301 
302 	/* Set BW quota for CPU as driver supports FIFO mode only. */
303 	se->icc_paths[CPU_TO_GENI].avg_bw = Bps_to_icc(mas->cur_speed_hz);
304 	ret = geni_icc_set_bw(se);
305 	if (ret)
306 		return ret;
307 
308 	return 0;
309 }
310 
311 static int setup_fifo_params(struct spi_device *spi_slv,
312 					struct spi_master *spi)
313 {
314 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
315 	struct geni_se *se = &mas->se;
316 	u32 loopback_cfg = 0, cpol = 0, cpha = 0, demux_output_inv = 0;
317 	u32 demux_sel;
318 
319 	if (mas->last_mode != spi_slv->mode) {
320 		if (spi_slv->mode & SPI_LOOP)
321 			loopback_cfg = LOOPBACK_ENABLE;
322 
323 		if (spi_slv->mode & SPI_CPOL)
324 			cpol = CPOL;
325 
326 		if (spi_slv->mode & SPI_CPHA)
327 			cpha = CPHA;
328 
329 		if (spi_slv->mode & SPI_CS_HIGH)
330 			demux_output_inv = BIT(spi_slv->chip_select);
331 
332 		demux_sel = spi_slv->chip_select;
333 		mas->cur_bits_per_word = spi_slv->bits_per_word;
334 
335 		spi_setup_word_len(mas, spi_slv->mode, spi_slv->bits_per_word);
336 		writel(loopback_cfg, se->base + SE_SPI_LOOPBACK);
337 		writel(demux_sel, se->base + SE_SPI_DEMUX_SEL);
338 		writel(cpha, se->base + SE_SPI_CPHA);
339 		writel(cpol, se->base + SE_SPI_CPOL);
340 		writel(demux_output_inv, se->base + SE_SPI_DEMUX_OUTPUT_INV);
341 
342 		mas->last_mode = spi_slv->mode;
343 	}
344 
345 	return geni_spi_set_clock_and_bw(mas, spi_slv->max_speed_hz);
346 }
347 
348 static void
349 spi_gsi_callback_result(void *cb, const struct dmaengine_result *result)
350 {
351 	struct spi_master *spi = cb;
352 
353 	if (result->result != DMA_TRANS_NOERROR) {
354 		dev_err(&spi->dev, "DMA txn failed: %d\n", result->result);
355 		return;
356 	}
357 
358 	if (!result->residue) {
359 		dev_dbg(&spi->dev, "DMA txn completed\n");
360 		spi_finalize_current_transfer(spi);
361 	} else {
362 		dev_err(&spi->dev, "DMA xfer has pending: %d\n", result->residue);
363 	}
364 }
365 
366 static int setup_gsi_xfer(struct spi_transfer *xfer, struct spi_geni_master *mas,
367 			  struct spi_device *spi_slv, struct spi_master *spi)
368 {
369 	unsigned long flags = DMA_PREP_INTERRUPT | DMA_CTRL_ACK;
370 	struct dma_slave_config config = {};
371 	struct gpi_spi_config peripheral = {};
372 	struct dma_async_tx_descriptor *tx_desc, *rx_desc;
373 	int ret;
374 
375 	config.peripheral_config = &peripheral;
376 	config.peripheral_size = sizeof(peripheral);
377 	peripheral.set_config = true;
378 
379 	if (xfer->bits_per_word != mas->cur_bits_per_word ||
380 	    xfer->speed_hz != mas->cur_speed_hz) {
381 		mas->cur_bits_per_word = xfer->bits_per_word;
382 		mas->cur_speed_hz = xfer->speed_hz;
383 	}
384 
385 	if (xfer->tx_buf && xfer->rx_buf) {
386 		peripheral.cmd = SPI_DUPLEX;
387 	} else if (xfer->tx_buf) {
388 		peripheral.cmd = SPI_TX;
389 		peripheral.rx_len = 0;
390 	} else if (xfer->rx_buf) {
391 		peripheral.cmd = SPI_RX;
392 		if (!(mas->cur_bits_per_word % MIN_WORD_LEN)) {
393 			peripheral.rx_len = ((xfer->len << 3) / mas->cur_bits_per_word);
394 		} else {
395 			int bytes_per_word = (mas->cur_bits_per_word / BITS_PER_BYTE) + 1;
396 
397 			peripheral.rx_len = (xfer->len / bytes_per_word);
398 		}
399 	}
400 
401 	peripheral.loopback_en = !!(spi_slv->mode & SPI_LOOP);
402 	peripheral.clock_pol_high = !!(spi_slv->mode & SPI_CPOL);
403 	peripheral.data_pol_high = !!(spi_slv->mode & SPI_CPHA);
404 	peripheral.cs = spi_slv->chip_select;
405 	peripheral.pack_en = true;
406 	peripheral.word_len = xfer->bits_per_word - MIN_WORD_LEN;
407 
408 	ret = get_spi_clk_cfg(mas->cur_speed_hz, mas,
409 			      &peripheral.clk_src, &peripheral.clk_div);
410 	if (ret) {
411 		dev_err(mas->dev, "Err in get_spi_clk_cfg() :%d\n", ret);
412 		return ret;
413 	}
414 
415 	if (!xfer->cs_change) {
416 		if (!list_is_last(&xfer->transfer_list, &spi->cur_msg->transfers))
417 			peripheral.fragmentation = FRAGMENTATION;
418 	}
419 
420 	if (peripheral.cmd & SPI_RX) {
421 		dmaengine_slave_config(mas->rx, &config);
422 		rx_desc = dmaengine_prep_slave_sg(mas->rx, xfer->rx_sg.sgl, xfer->rx_sg.nents,
423 						  DMA_DEV_TO_MEM, flags);
424 		if (!rx_desc) {
425 			dev_err(mas->dev, "Err setting up rx desc\n");
426 			return -EIO;
427 		}
428 	}
429 
430 	/*
431 	 * Prepare the TX always, even for RX or tx_buf being null, we would
432 	 * need TX to be prepared per GSI spec
433 	 */
434 	dmaengine_slave_config(mas->tx, &config);
435 	tx_desc = dmaengine_prep_slave_sg(mas->tx, xfer->tx_sg.sgl, xfer->tx_sg.nents,
436 					  DMA_MEM_TO_DEV, flags);
437 	if (!tx_desc) {
438 		dev_err(mas->dev, "Err setting up tx desc\n");
439 		return -EIO;
440 	}
441 
442 	tx_desc->callback_result = spi_gsi_callback_result;
443 	tx_desc->callback_param = spi;
444 
445 	if (peripheral.cmd & SPI_RX)
446 		dmaengine_submit(rx_desc);
447 	dmaengine_submit(tx_desc);
448 
449 	if (peripheral.cmd & SPI_RX)
450 		dma_async_issue_pending(mas->rx);
451 
452 	dma_async_issue_pending(mas->tx);
453 	return 1;
454 }
455 
456 static bool geni_can_dma(struct spi_controller *ctlr,
457 			 struct spi_device *slv, struct spi_transfer *xfer)
458 {
459 	struct spi_geni_master *mas = spi_master_get_devdata(slv->master);
460 
461 	/* check if dma is supported */
462 	return mas->cur_xfer_mode != GENI_SE_FIFO;
463 }
464 
465 static int spi_geni_prepare_message(struct spi_master *spi,
466 					struct spi_message *spi_msg)
467 {
468 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
469 	int ret;
470 
471 	switch (mas->cur_xfer_mode) {
472 	case GENI_SE_FIFO:
473 		if (spi_geni_is_abort_still_pending(mas))
474 			return -EBUSY;
475 		ret = setup_fifo_params(spi_msg->spi, spi);
476 		if (ret)
477 			dev_err(mas->dev, "Couldn't select mode %d\n", ret);
478 		return ret;
479 
480 	case GENI_GPI_DMA:
481 		/* nothing to do for GPI DMA */
482 		return 0;
483 	}
484 
485 	dev_err(mas->dev, "Mode not supported %d", mas->cur_xfer_mode);
486 	return -EINVAL;
487 }
488 
489 static int spi_geni_grab_gpi_chan(struct spi_geni_master *mas)
490 {
491 	int ret;
492 
493 	mas->tx = dma_request_chan(mas->dev, "tx");
494 	if (IS_ERR(mas->tx)) {
495 		ret = dev_err_probe(mas->dev, PTR_ERR(mas->tx),
496 				    "Failed to get tx DMA ch\n");
497 		goto err_tx;
498 	}
499 
500 	mas->rx = dma_request_chan(mas->dev, "rx");
501 	if (IS_ERR(mas->rx)) {
502 		ret = dev_err_probe(mas->dev, PTR_ERR(mas->rx),
503 				    "Failed to get rx DMA ch\n");
504 		goto err_rx;
505 	}
506 
507 	return 0;
508 
509 err_rx:
510 	mas->rx = NULL;
511 	dma_release_channel(mas->tx);
512 err_tx:
513 	mas->tx = NULL;
514 	return ret;
515 }
516 
517 static void spi_geni_release_dma_chan(struct spi_geni_master *mas)
518 {
519 	if (mas->rx) {
520 		dma_release_channel(mas->rx);
521 		mas->rx = NULL;
522 	}
523 
524 	if (mas->tx) {
525 		dma_release_channel(mas->tx);
526 		mas->tx = NULL;
527 	}
528 }
529 
530 static int spi_geni_init(struct spi_geni_master *mas)
531 {
532 	struct geni_se *se = &mas->se;
533 	unsigned int proto, major, minor, ver;
534 	u32 spi_tx_cfg, fifo_disable;
535 	int ret = -ENXIO;
536 
537 	pm_runtime_get_sync(mas->dev);
538 
539 	proto = geni_se_read_proto(se);
540 	if (proto != GENI_SE_SPI) {
541 		dev_err(mas->dev, "Invalid proto %d\n", proto);
542 		goto out_pm;
543 	}
544 	mas->tx_fifo_depth = geni_se_get_tx_fifo_depth(se);
545 
546 	/* Width of Tx and Rx FIFO is same */
547 	mas->fifo_width_bits = geni_se_get_tx_fifo_width(se);
548 
549 	/*
550 	 * Hardware programming guide suggests to configure
551 	 * RX FIFO RFR level to fifo_depth-2.
552 	 */
553 	geni_se_init(se, mas->tx_fifo_depth - 3, mas->tx_fifo_depth - 2);
554 	/* Transmit an entire FIFO worth of data per IRQ */
555 	mas->tx_wm = 1;
556 	ver = geni_se_get_qup_hw_version(se);
557 	major = GENI_SE_VERSION_MAJOR(ver);
558 	minor = GENI_SE_VERSION_MINOR(ver);
559 
560 	if (major == 1 && minor == 0)
561 		mas->oversampling = 2;
562 	else
563 		mas->oversampling = 1;
564 
565 	fifo_disable = readl(se->base + GENI_IF_DISABLE_RO) & FIFO_IF_DISABLE;
566 	switch (fifo_disable) {
567 	case 1:
568 		ret = spi_geni_grab_gpi_chan(mas);
569 		if (!ret) { /* success case */
570 			mas->cur_xfer_mode = GENI_GPI_DMA;
571 			geni_se_select_mode(se, GENI_GPI_DMA);
572 			dev_dbg(mas->dev, "Using GPI DMA mode for SPI\n");
573 			break;
574 		}
575 		/*
576 		 * in case of failure to get dma channel, we can still do the
577 		 * FIFO mode, so fallthrough
578 		 */
579 		dev_warn(mas->dev, "FIFO mode disabled, but couldn't get DMA, fall back to FIFO mode\n");
580 		fallthrough;
581 
582 	case 0:
583 		mas->cur_xfer_mode = GENI_SE_FIFO;
584 		geni_se_select_mode(se, GENI_SE_FIFO);
585 		ret = 0;
586 		break;
587 	}
588 
589 	/* We always control CS manually */
590 	spi_tx_cfg = readl(se->base + SE_SPI_TRANS_CFG);
591 	spi_tx_cfg &= ~CS_TOGGLE;
592 	writel(spi_tx_cfg, se->base + SE_SPI_TRANS_CFG);
593 
594 out_pm:
595 	pm_runtime_put(mas->dev);
596 	return ret;
597 }
598 
599 static unsigned int geni_byte_per_fifo_word(struct spi_geni_master *mas)
600 {
601 	/*
602 	 * Calculate how many bytes we'll put in each FIFO word.  If the
603 	 * transfer words don't pack cleanly into a FIFO word we'll just put
604 	 * one transfer word in each FIFO word.  If they do pack we'll pack 'em.
605 	 */
606 	if (mas->fifo_width_bits % mas->cur_bits_per_word)
607 		return roundup_pow_of_two(DIV_ROUND_UP(mas->cur_bits_per_word,
608 						       BITS_PER_BYTE));
609 
610 	return mas->fifo_width_bits / BITS_PER_BYTE;
611 }
612 
613 static bool geni_spi_handle_tx(struct spi_geni_master *mas)
614 {
615 	struct geni_se *se = &mas->se;
616 	unsigned int max_bytes;
617 	const u8 *tx_buf;
618 	unsigned int bytes_per_fifo_word = geni_byte_per_fifo_word(mas);
619 	unsigned int i = 0;
620 
621 	/* Stop the watermark IRQ if nothing to send */
622 	if (!mas->cur_xfer) {
623 		writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
624 		return false;
625 	}
626 
627 	max_bytes = (mas->tx_fifo_depth - mas->tx_wm) * bytes_per_fifo_word;
628 	if (mas->tx_rem_bytes < max_bytes)
629 		max_bytes = mas->tx_rem_bytes;
630 
631 	tx_buf = mas->cur_xfer->tx_buf + mas->cur_xfer->len - mas->tx_rem_bytes;
632 	while (i < max_bytes) {
633 		unsigned int j;
634 		unsigned int bytes_to_write;
635 		u32 fifo_word = 0;
636 		u8 *fifo_byte = (u8 *)&fifo_word;
637 
638 		bytes_to_write = min(bytes_per_fifo_word, max_bytes - i);
639 		for (j = 0; j < bytes_to_write; j++)
640 			fifo_byte[j] = tx_buf[i++];
641 		iowrite32_rep(se->base + SE_GENI_TX_FIFOn, &fifo_word, 1);
642 	}
643 	mas->tx_rem_bytes -= max_bytes;
644 	if (!mas->tx_rem_bytes) {
645 		writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
646 		return false;
647 	}
648 	return true;
649 }
650 
651 static void geni_spi_handle_rx(struct spi_geni_master *mas)
652 {
653 	struct geni_se *se = &mas->se;
654 	u32 rx_fifo_status;
655 	unsigned int rx_bytes;
656 	unsigned int rx_last_byte_valid;
657 	u8 *rx_buf;
658 	unsigned int bytes_per_fifo_word = geni_byte_per_fifo_word(mas);
659 	unsigned int i = 0;
660 
661 	rx_fifo_status = readl(se->base + SE_GENI_RX_FIFO_STATUS);
662 	rx_bytes = (rx_fifo_status & RX_FIFO_WC_MSK) * bytes_per_fifo_word;
663 	if (rx_fifo_status & RX_LAST) {
664 		rx_last_byte_valid = rx_fifo_status & RX_LAST_BYTE_VALID_MSK;
665 		rx_last_byte_valid >>= RX_LAST_BYTE_VALID_SHFT;
666 		if (rx_last_byte_valid && rx_last_byte_valid < 4)
667 			rx_bytes -= bytes_per_fifo_word - rx_last_byte_valid;
668 	}
669 
670 	/* Clear out the FIFO and bail if nowhere to put it */
671 	if (!mas->cur_xfer) {
672 		for (i = 0; i < DIV_ROUND_UP(rx_bytes, bytes_per_fifo_word); i++)
673 			readl(se->base + SE_GENI_RX_FIFOn);
674 		return;
675 	}
676 
677 	if (mas->rx_rem_bytes < rx_bytes)
678 		rx_bytes = mas->rx_rem_bytes;
679 
680 	rx_buf = mas->cur_xfer->rx_buf + mas->cur_xfer->len - mas->rx_rem_bytes;
681 	while (i < rx_bytes) {
682 		u32 fifo_word = 0;
683 		u8 *fifo_byte = (u8 *)&fifo_word;
684 		unsigned int bytes_to_read;
685 		unsigned int j;
686 
687 		bytes_to_read = min(bytes_per_fifo_word, rx_bytes - i);
688 		ioread32_rep(se->base + SE_GENI_RX_FIFOn, &fifo_word, 1);
689 		for (j = 0; j < bytes_to_read; j++)
690 			rx_buf[i++] = fifo_byte[j];
691 	}
692 	mas->rx_rem_bytes -= rx_bytes;
693 }
694 
695 static void setup_fifo_xfer(struct spi_transfer *xfer,
696 				struct spi_geni_master *mas,
697 				u16 mode, struct spi_master *spi)
698 {
699 	u32 m_cmd = 0;
700 	u32 len;
701 	struct geni_se *se = &mas->se;
702 	int ret;
703 
704 	/*
705 	 * Ensure that our interrupt handler isn't still running from some
706 	 * prior command before we start messing with the hardware behind
707 	 * its back.  We don't need to _keep_ the lock here since we're only
708 	 * worried about racing with out interrupt handler.  The SPI core
709 	 * already handles making sure that we're not trying to do two
710 	 * transfers at once or setting a chip select and doing a transfer
711 	 * concurrently.
712 	 *
713 	 * NOTE: we actually _can't_ hold the lock here because possibly we
714 	 * might call clk_set_rate() which needs to be able to sleep.
715 	 */
716 	spin_lock_irq(&mas->lock);
717 	spin_unlock_irq(&mas->lock);
718 
719 	if (xfer->bits_per_word != mas->cur_bits_per_word) {
720 		spi_setup_word_len(mas, mode, xfer->bits_per_word);
721 		mas->cur_bits_per_word = xfer->bits_per_word;
722 	}
723 
724 	/* Speed and bits per word can be overridden per transfer */
725 	ret = geni_spi_set_clock_and_bw(mas, xfer->speed_hz);
726 	if (ret)
727 		return;
728 
729 	mas->tx_rem_bytes = 0;
730 	mas->rx_rem_bytes = 0;
731 
732 	if (!(mas->cur_bits_per_word % MIN_WORD_LEN))
733 		len = xfer->len * BITS_PER_BYTE / mas->cur_bits_per_word;
734 	else
735 		len = xfer->len / (mas->cur_bits_per_word / BITS_PER_BYTE + 1);
736 	len &= TRANS_LEN_MSK;
737 
738 	mas->cur_xfer = xfer;
739 	if (xfer->tx_buf) {
740 		m_cmd |= SPI_TX_ONLY;
741 		mas->tx_rem_bytes = xfer->len;
742 		writel(len, se->base + SE_SPI_TX_TRANS_LEN);
743 	}
744 
745 	if (xfer->rx_buf) {
746 		m_cmd |= SPI_RX_ONLY;
747 		writel(len, se->base + SE_SPI_RX_TRANS_LEN);
748 		mas->rx_rem_bytes = xfer->len;
749 	}
750 
751 	/*
752 	 * Lock around right before we start the transfer since our
753 	 * interrupt could come in at any time now.
754 	 */
755 	spin_lock_irq(&mas->lock);
756 	geni_se_setup_m_cmd(se, m_cmd, FRAGMENTATION);
757 	if (m_cmd & SPI_TX_ONLY) {
758 		if (geni_spi_handle_tx(mas))
759 			writel(mas->tx_wm, se->base + SE_GENI_TX_WATERMARK_REG);
760 	}
761 	spin_unlock_irq(&mas->lock);
762 }
763 
764 static int spi_geni_transfer_one(struct spi_master *spi,
765 				struct spi_device *slv,
766 				struct spi_transfer *xfer)
767 {
768 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
769 
770 	if (spi_geni_is_abort_still_pending(mas))
771 		return -EBUSY;
772 
773 	/* Terminate and return success for 0 byte length transfer */
774 	if (!xfer->len)
775 		return 0;
776 
777 	if (mas->cur_xfer_mode == GENI_SE_FIFO) {
778 		setup_fifo_xfer(xfer, mas, slv->mode, spi);
779 		return 1;
780 	}
781 	return setup_gsi_xfer(xfer, mas, slv, spi);
782 }
783 
784 static irqreturn_t geni_spi_isr(int irq, void *data)
785 {
786 	struct spi_master *spi = data;
787 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
788 	struct geni_se *se = &mas->se;
789 	u32 m_irq;
790 
791 	m_irq = readl(se->base + SE_GENI_M_IRQ_STATUS);
792 	if (!m_irq)
793 		return IRQ_NONE;
794 
795 	if (m_irq & (M_CMD_OVERRUN_EN | M_ILLEGAL_CMD_EN | M_CMD_FAILURE_EN |
796 		     M_RX_FIFO_RD_ERR_EN | M_RX_FIFO_WR_ERR_EN |
797 		     M_TX_FIFO_RD_ERR_EN | M_TX_FIFO_WR_ERR_EN))
798 		dev_warn(mas->dev, "Unexpected IRQ err status %#010x\n", m_irq);
799 
800 	spin_lock(&mas->lock);
801 
802 	if ((m_irq & M_RX_FIFO_WATERMARK_EN) || (m_irq & M_RX_FIFO_LAST_EN))
803 		geni_spi_handle_rx(mas);
804 
805 	if (m_irq & M_TX_FIFO_WATERMARK_EN)
806 		geni_spi_handle_tx(mas);
807 
808 	if (m_irq & M_CMD_DONE_EN) {
809 		if (mas->cur_xfer) {
810 			spi_finalize_current_transfer(spi);
811 			mas->cur_xfer = NULL;
812 			/*
813 			 * If this happens, then a CMD_DONE came before all the
814 			 * Tx buffer bytes were sent out. This is unusual, log
815 			 * this condition and disable the WM interrupt to
816 			 * prevent the system from stalling due an interrupt
817 			 * storm.
818 			 *
819 			 * If this happens when all Rx bytes haven't been
820 			 * received, log the condition. The only known time
821 			 * this can happen is if bits_per_word != 8 and some
822 			 * registers that expect xfer lengths in num spi_words
823 			 * weren't written correctly.
824 			 */
825 			if (mas->tx_rem_bytes) {
826 				writel(0, se->base + SE_GENI_TX_WATERMARK_REG);
827 				dev_err(mas->dev, "Premature done. tx_rem = %d bpw%d\n",
828 					mas->tx_rem_bytes, mas->cur_bits_per_word);
829 			}
830 			if (mas->rx_rem_bytes)
831 				dev_err(mas->dev, "Premature done. rx_rem = %d bpw%d\n",
832 					mas->rx_rem_bytes, mas->cur_bits_per_word);
833 		} else {
834 			complete(&mas->cs_done);
835 		}
836 	}
837 
838 	if (m_irq & M_CMD_CANCEL_EN)
839 		complete(&mas->cancel_done);
840 	if (m_irq & M_CMD_ABORT_EN)
841 		complete(&mas->abort_done);
842 
843 	/*
844 	 * It's safe or a good idea to Ack all of our interrupts at the end
845 	 * of the function. Specifically:
846 	 * - M_CMD_DONE_EN / M_RX_FIFO_LAST_EN: Edge triggered interrupts and
847 	 *   clearing Acks. Clearing at the end relies on nobody else having
848 	 *   started a new transfer yet or else we could be clearing _their_
849 	 *   done bit, but everyone grabs the spinlock before starting a new
850 	 *   transfer.
851 	 * - M_RX_FIFO_WATERMARK_EN / M_TX_FIFO_WATERMARK_EN: These appear
852 	 *   to be "latched level" interrupts so it's important to clear them
853 	 *   _after_ you've handled the condition and always safe to do so
854 	 *   since they'll re-assert if they're still happening.
855 	 */
856 	writel(m_irq, se->base + SE_GENI_M_IRQ_CLEAR);
857 
858 	spin_unlock(&mas->lock);
859 
860 	return IRQ_HANDLED;
861 }
862 
863 static int spi_geni_probe(struct platform_device *pdev)
864 {
865 	int ret, irq;
866 	struct spi_master *spi;
867 	struct spi_geni_master *mas;
868 	void __iomem *base;
869 	struct clk *clk;
870 	struct device *dev = &pdev->dev;
871 
872 	irq = platform_get_irq(pdev, 0);
873 	if (irq < 0)
874 		return irq;
875 
876 	ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64));
877 	if (ret) {
878 		ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
879 		if (ret)
880 			return dev_err_probe(dev, ret, "could not set DMA mask\n");
881 	}
882 
883 	base = devm_platform_ioremap_resource(pdev, 0);
884 	if (IS_ERR(base))
885 		return PTR_ERR(base);
886 
887 	clk = devm_clk_get(dev, "se");
888 	if (IS_ERR(clk))
889 		return PTR_ERR(clk);
890 
891 	spi = devm_spi_alloc_master(dev, sizeof(*mas));
892 	if (!spi)
893 		return -ENOMEM;
894 
895 	platform_set_drvdata(pdev, spi);
896 	mas = spi_master_get_devdata(spi);
897 	mas->irq = irq;
898 	mas->dev = dev;
899 	mas->se.dev = dev;
900 	mas->se.wrapper = dev_get_drvdata(dev->parent);
901 	mas->se.base = base;
902 	mas->se.clk = clk;
903 
904 	ret = devm_pm_opp_set_clkname(&pdev->dev, "se");
905 	if (ret)
906 		return ret;
907 	/* OPP table is optional */
908 	ret = devm_pm_opp_of_add_table(&pdev->dev);
909 	if (ret && ret != -ENODEV) {
910 		dev_err(&pdev->dev, "invalid OPP table in device tree\n");
911 		return ret;
912 	}
913 
914 	spi->bus_num = -1;
915 	spi->dev.of_node = dev->of_node;
916 	spi->mode_bits = SPI_CPOL | SPI_CPHA | SPI_LOOP | SPI_CS_HIGH;
917 	spi->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
918 	spi->num_chipselect = 4;
919 	spi->max_speed_hz = 50000000;
920 	spi->prepare_message = spi_geni_prepare_message;
921 	spi->transfer_one = spi_geni_transfer_one;
922 	spi->can_dma = geni_can_dma;
923 	spi->dma_map_dev = dev->parent;
924 	spi->auto_runtime_pm = true;
925 	spi->handle_err = handle_fifo_timeout;
926 	spi->use_gpio_descriptors = true;
927 
928 	init_completion(&mas->cs_done);
929 	init_completion(&mas->cancel_done);
930 	init_completion(&mas->abort_done);
931 	spin_lock_init(&mas->lock);
932 	pm_runtime_use_autosuspend(&pdev->dev);
933 	pm_runtime_set_autosuspend_delay(&pdev->dev, 250);
934 	pm_runtime_enable(dev);
935 
936 	ret = geni_icc_get(&mas->se, NULL);
937 	if (ret)
938 		goto spi_geni_probe_runtime_disable;
939 	/* Set the bus quota to a reasonable value for register access */
940 	mas->se.icc_paths[GENI_TO_CORE].avg_bw = Bps_to_icc(CORE_2X_50_MHZ);
941 	mas->se.icc_paths[CPU_TO_GENI].avg_bw = GENI_DEFAULT_BW;
942 
943 	ret = geni_icc_set_bw(&mas->se);
944 	if (ret)
945 		goto spi_geni_probe_runtime_disable;
946 
947 	ret = spi_geni_init(mas);
948 	if (ret)
949 		goto spi_geni_probe_runtime_disable;
950 
951 	/*
952 	 * check the mode supported and set_cs for fifo mode only
953 	 * for dma (gsi) mode, the gsi will set cs based on params passed in
954 	 * TRE
955 	 */
956 	if (mas->cur_xfer_mode == GENI_SE_FIFO)
957 		spi->set_cs = spi_geni_set_cs;
958 
959 	ret = request_irq(mas->irq, geni_spi_isr, 0, dev_name(dev), spi);
960 	if (ret)
961 		goto spi_geni_release_dma;
962 
963 	ret = spi_register_master(spi);
964 	if (ret)
965 		goto spi_geni_probe_free_irq;
966 
967 	return 0;
968 spi_geni_probe_free_irq:
969 	free_irq(mas->irq, spi);
970 spi_geni_release_dma:
971 	spi_geni_release_dma_chan(mas);
972 spi_geni_probe_runtime_disable:
973 	pm_runtime_disable(dev);
974 	return ret;
975 }
976 
977 static int spi_geni_remove(struct platform_device *pdev)
978 {
979 	struct spi_master *spi = platform_get_drvdata(pdev);
980 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
981 
982 	/* Unregister _before_ disabling pm_runtime() so we stop transfers */
983 	spi_unregister_master(spi);
984 
985 	spi_geni_release_dma_chan(mas);
986 
987 	free_irq(mas->irq, spi);
988 	pm_runtime_disable(&pdev->dev);
989 	return 0;
990 }
991 
992 static int __maybe_unused spi_geni_runtime_suspend(struct device *dev)
993 {
994 	struct spi_master *spi = dev_get_drvdata(dev);
995 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
996 	int ret;
997 
998 	/* Drop the performance state vote */
999 	dev_pm_opp_set_rate(dev, 0);
1000 
1001 	ret = geni_se_resources_off(&mas->se);
1002 	if (ret)
1003 		return ret;
1004 
1005 	return geni_icc_disable(&mas->se);
1006 }
1007 
1008 static int __maybe_unused spi_geni_runtime_resume(struct device *dev)
1009 {
1010 	struct spi_master *spi = dev_get_drvdata(dev);
1011 	struct spi_geni_master *mas = spi_master_get_devdata(spi);
1012 	int ret;
1013 
1014 	ret = geni_icc_enable(&mas->se);
1015 	if (ret)
1016 		return ret;
1017 
1018 	ret = geni_se_resources_on(&mas->se);
1019 	if (ret)
1020 		return ret;
1021 
1022 	return dev_pm_opp_set_rate(mas->dev, mas->cur_sclk_hz);
1023 }
1024 
1025 static int __maybe_unused spi_geni_suspend(struct device *dev)
1026 {
1027 	struct spi_master *spi = dev_get_drvdata(dev);
1028 	int ret;
1029 
1030 	ret = spi_master_suspend(spi);
1031 	if (ret)
1032 		return ret;
1033 
1034 	ret = pm_runtime_force_suspend(dev);
1035 	if (ret)
1036 		spi_master_resume(spi);
1037 
1038 	return ret;
1039 }
1040 
1041 static int __maybe_unused spi_geni_resume(struct device *dev)
1042 {
1043 	struct spi_master *spi = dev_get_drvdata(dev);
1044 	int ret;
1045 
1046 	ret = pm_runtime_force_resume(dev);
1047 	if (ret)
1048 		return ret;
1049 
1050 	ret = spi_master_resume(spi);
1051 	if (ret)
1052 		pm_runtime_force_suspend(dev);
1053 
1054 	return ret;
1055 }
1056 
1057 static const struct dev_pm_ops spi_geni_pm_ops = {
1058 	SET_RUNTIME_PM_OPS(spi_geni_runtime_suspend,
1059 					spi_geni_runtime_resume, NULL)
1060 	SET_SYSTEM_SLEEP_PM_OPS(spi_geni_suspend, spi_geni_resume)
1061 };
1062 
1063 static const struct of_device_id spi_geni_dt_match[] = {
1064 	{ .compatible = "qcom,geni-spi" },
1065 	{}
1066 };
1067 MODULE_DEVICE_TABLE(of, spi_geni_dt_match);
1068 
1069 static struct platform_driver spi_geni_driver = {
1070 	.probe  = spi_geni_probe,
1071 	.remove = spi_geni_remove,
1072 	.driver = {
1073 		.name = "geni_spi",
1074 		.pm = &spi_geni_pm_ops,
1075 		.of_match_table = spi_geni_dt_match,
1076 	},
1077 };
1078 module_platform_driver(spi_geni_driver);
1079 
1080 MODULE_DESCRIPTION("SPI driver for GENI based QUP cores");
1081 MODULE_LICENSE("GPL v2");
1082