xref: /openbmc/linux/drivers/dma/stm32-dma.c (revision e82c878d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Driver for STM32 DMA controller
4  *
5  * Inspired by dma-jz4740.c and tegra20-apb-dma.c
6  *
7  * Copyright (C) M'boumba Cedric Madianga 2015
8  * Author: M'boumba Cedric Madianga <cedric.madianga@gmail.com>
9  *         Pierre-Yves Mordret <pierre-yves.mordret@st.com>
10  */
11 
12 #include <linux/clk.h>
13 #include <linux/delay.h>
14 #include <linux/dmaengine.h>
15 #include <linux/dma-mapping.h>
16 #include <linux/err.h>
17 #include <linux/init.h>
18 #include <linux/iopoll.h>
19 #include <linux/jiffies.h>
20 #include <linux/list.h>
21 #include <linux/module.h>
22 #include <linux/of.h>
23 #include <linux/of_device.h>
24 #include <linux/of_dma.h>
25 #include <linux/platform_device.h>
26 #include <linux/pm_runtime.h>
27 #include <linux/reset.h>
28 #include <linux/sched.h>
29 #include <linux/slab.h>
30 
31 #include "virt-dma.h"
32 
33 #define STM32_DMA_LISR			0x0000 /* DMA Low Int Status Reg */
34 #define STM32_DMA_HISR			0x0004 /* DMA High Int Status Reg */
35 #define STM32_DMA_LIFCR			0x0008 /* DMA Low Int Flag Clear Reg */
36 #define STM32_DMA_HIFCR			0x000c /* DMA High Int Flag Clear Reg */
37 #define STM32_DMA_TCI			BIT(5) /* Transfer Complete Interrupt */
38 #define STM32_DMA_HTI			BIT(4) /* Half Transfer Interrupt */
39 #define STM32_DMA_TEI			BIT(3) /* Transfer Error Interrupt */
40 #define STM32_DMA_DMEI			BIT(2) /* Direct Mode Error Interrupt */
41 #define STM32_DMA_FEI			BIT(0) /* FIFO Error Interrupt */
42 #define STM32_DMA_MASKI			(STM32_DMA_TCI \
43 					 | STM32_DMA_TEI \
44 					 | STM32_DMA_DMEI \
45 					 | STM32_DMA_FEI)
46 
47 /* DMA Stream x Configuration Register */
48 #define STM32_DMA_SCR(x)		(0x0010 + 0x18 * (x)) /* x = 0..7 */
49 #define STM32_DMA_SCR_REQ(n)		((n & 0x7) << 25)
50 #define STM32_DMA_SCR_MBURST_MASK	GENMASK(24, 23)
51 #define STM32_DMA_SCR_MBURST(n)	        ((n & 0x3) << 23)
52 #define STM32_DMA_SCR_PBURST_MASK	GENMASK(22, 21)
53 #define STM32_DMA_SCR_PBURST(n)	        ((n & 0x3) << 21)
54 #define STM32_DMA_SCR_PL_MASK		GENMASK(17, 16)
55 #define STM32_DMA_SCR_PL(n)		((n & 0x3) << 16)
56 #define STM32_DMA_SCR_MSIZE_MASK	GENMASK(14, 13)
57 #define STM32_DMA_SCR_MSIZE(n)		((n & 0x3) << 13)
58 #define STM32_DMA_SCR_PSIZE_MASK	GENMASK(12, 11)
59 #define STM32_DMA_SCR_PSIZE(n)		((n & 0x3) << 11)
60 #define STM32_DMA_SCR_PSIZE_GET(n)	((n & STM32_DMA_SCR_PSIZE_MASK) >> 11)
61 #define STM32_DMA_SCR_DIR_MASK		GENMASK(7, 6)
62 #define STM32_DMA_SCR_DIR(n)		((n & 0x3) << 6)
63 #define STM32_DMA_SCR_CT		BIT(19) /* Target in double buffer */
64 #define STM32_DMA_SCR_DBM		BIT(18) /* Double Buffer Mode */
65 #define STM32_DMA_SCR_PINCOS		BIT(15) /* Peripheral inc offset size */
66 #define STM32_DMA_SCR_MINC		BIT(10) /* Memory increment mode */
67 #define STM32_DMA_SCR_PINC		BIT(9) /* Peripheral increment mode */
68 #define STM32_DMA_SCR_CIRC		BIT(8) /* Circular mode */
69 #define STM32_DMA_SCR_PFCTRL		BIT(5) /* Peripheral Flow Controller */
70 #define STM32_DMA_SCR_TCIE		BIT(4) /* Transfer Complete Int Enable
71 						*/
72 #define STM32_DMA_SCR_TEIE		BIT(2) /* Transfer Error Int Enable */
73 #define STM32_DMA_SCR_DMEIE		BIT(1) /* Direct Mode Err Int Enable */
74 #define STM32_DMA_SCR_EN		BIT(0) /* Stream Enable */
75 #define STM32_DMA_SCR_CFG_MASK		(STM32_DMA_SCR_PINC \
76 					| STM32_DMA_SCR_MINC \
77 					| STM32_DMA_SCR_PINCOS \
78 					| STM32_DMA_SCR_PL_MASK)
79 #define STM32_DMA_SCR_IRQ_MASK		(STM32_DMA_SCR_TCIE \
80 					| STM32_DMA_SCR_TEIE \
81 					| STM32_DMA_SCR_DMEIE)
82 
83 /* DMA Stream x number of data register */
84 #define STM32_DMA_SNDTR(x)		(0x0014 + 0x18 * (x))
85 
86 /* DMA stream peripheral address register */
87 #define STM32_DMA_SPAR(x)		(0x0018 + 0x18 * (x))
88 
89 /* DMA stream x memory 0 address register */
90 #define STM32_DMA_SM0AR(x)		(0x001c + 0x18 * (x))
91 
92 /* DMA stream x memory 1 address register */
93 #define STM32_DMA_SM1AR(x)		(0x0020 + 0x18 * (x))
94 
95 /* DMA stream x FIFO control register */
96 #define STM32_DMA_SFCR(x)		(0x0024 + 0x18 * (x))
97 #define STM32_DMA_SFCR_FTH_MASK		GENMASK(1, 0)
98 #define STM32_DMA_SFCR_FTH(n)		(n & STM32_DMA_SFCR_FTH_MASK)
99 #define STM32_DMA_SFCR_FEIE		BIT(7) /* FIFO error interrupt enable */
100 #define STM32_DMA_SFCR_DMDIS		BIT(2) /* Direct mode disable */
101 #define STM32_DMA_SFCR_MASK		(STM32_DMA_SFCR_FEIE \
102 					| STM32_DMA_SFCR_DMDIS)
103 
104 /* DMA direction */
105 #define STM32_DMA_DEV_TO_MEM		0x00
106 #define	STM32_DMA_MEM_TO_DEV		0x01
107 #define	STM32_DMA_MEM_TO_MEM		0x02
108 
109 /* DMA priority level */
110 #define STM32_DMA_PRIORITY_LOW		0x00
111 #define STM32_DMA_PRIORITY_MEDIUM	0x01
112 #define STM32_DMA_PRIORITY_HIGH		0x02
113 #define STM32_DMA_PRIORITY_VERY_HIGH	0x03
114 
115 /* DMA FIFO threshold selection */
116 #define STM32_DMA_FIFO_THRESHOLD_1QUARTERFULL		0x00
117 #define STM32_DMA_FIFO_THRESHOLD_HALFFULL		0x01
118 #define STM32_DMA_FIFO_THRESHOLD_3QUARTERSFULL		0x02
119 #define STM32_DMA_FIFO_THRESHOLD_FULL			0x03
120 
121 #define STM32_DMA_MAX_DATA_ITEMS	0xffff
122 /*
123  * Valid transfer starts from @0 to @0xFFFE leading to unaligned scatter
124  * gather at boundary. Thus it's safer to round down this value on FIFO
125  * size (16 Bytes)
126  */
127 #define STM32_DMA_ALIGNED_MAX_DATA_ITEMS	\
128 	ALIGN_DOWN(STM32_DMA_MAX_DATA_ITEMS, 16)
129 #define STM32_DMA_MAX_CHANNELS		0x08
130 #define STM32_DMA_MAX_REQUEST_ID	0x08
131 #define STM32_DMA_MAX_DATA_PARAM	0x03
132 #define STM32_DMA_FIFO_SIZE		16	/* FIFO is 16 bytes */
133 #define STM32_DMA_MIN_BURST		4
134 #define STM32_DMA_MAX_BURST		16
135 
136 /* DMA Features */
137 #define STM32_DMA_THRESHOLD_FTR_MASK	GENMASK(1, 0)
138 #define STM32_DMA_THRESHOLD_FTR_GET(n)	((n) & STM32_DMA_THRESHOLD_FTR_MASK)
139 
140 enum stm32_dma_width {
141 	STM32_DMA_BYTE,
142 	STM32_DMA_HALF_WORD,
143 	STM32_DMA_WORD,
144 };
145 
146 enum stm32_dma_burst_size {
147 	STM32_DMA_BURST_SINGLE,
148 	STM32_DMA_BURST_INCR4,
149 	STM32_DMA_BURST_INCR8,
150 	STM32_DMA_BURST_INCR16,
151 };
152 
153 /**
154  * struct stm32_dma_cfg - STM32 DMA custom configuration
155  * @channel_id: channel ID
156  * @request_line: DMA request
157  * @stream_config: 32bit mask specifying the DMA channel configuration
158  * @features: 32bit mask specifying the DMA Feature list
159  */
160 struct stm32_dma_cfg {
161 	u32 channel_id;
162 	u32 request_line;
163 	u32 stream_config;
164 	u32 features;
165 };
166 
167 struct stm32_dma_chan_reg {
168 	u32 dma_lisr;
169 	u32 dma_hisr;
170 	u32 dma_lifcr;
171 	u32 dma_hifcr;
172 	u32 dma_scr;
173 	u32 dma_sndtr;
174 	u32 dma_spar;
175 	u32 dma_sm0ar;
176 	u32 dma_sm1ar;
177 	u32 dma_sfcr;
178 };
179 
180 struct stm32_dma_sg_req {
181 	u32 len;
182 	struct stm32_dma_chan_reg chan_reg;
183 };
184 
185 struct stm32_dma_desc {
186 	struct virt_dma_desc vdesc;
187 	bool cyclic;
188 	u32 num_sgs;
189 	struct stm32_dma_sg_req sg_req[];
190 };
191 
192 struct stm32_dma_chan {
193 	struct virt_dma_chan vchan;
194 	bool config_init;
195 	bool busy;
196 	u32 id;
197 	u32 irq;
198 	struct stm32_dma_desc *desc;
199 	u32 next_sg;
200 	struct dma_slave_config	dma_sconfig;
201 	struct stm32_dma_chan_reg chan_reg;
202 	u32 threshold;
203 	u32 mem_burst;
204 	u32 mem_width;
205 };
206 
207 struct stm32_dma_device {
208 	struct dma_device ddev;
209 	void __iomem *base;
210 	struct clk *clk;
211 	bool mem2mem;
212 	struct stm32_dma_chan chan[STM32_DMA_MAX_CHANNELS];
213 };
214 
215 static struct stm32_dma_device *stm32_dma_get_dev(struct stm32_dma_chan *chan)
216 {
217 	return container_of(chan->vchan.chan.device, struct stm32_dma_device,
218 			    ddev);
219 }
220 
221 static struct stm32_dma_chan *to_stm32_dma_chan(struct dma_chan *c)
222 {
223 	return container_of(c, struct stm32_dma_chan, vchan.chan);
224 }
225 
226 static struct stm32_dma_desc *to_stm32_dma_desc(struct virt_dma_desc *vdesc)
227 {
228 	return container_of(vdesc, struct stm32_dma_desc, vdesc);
229 }
230 
231 static struct device *chan2dev(struct stm32_dma_chan *chan)
232 {
233 	return &chan->vchan.chan.dev->device;
234 }
235 
236 static u32 stm32_dma_read(struct stm32_dma_device *dmadev, u32 reg)
237 {
238 	return readl_relaxed(dmadev->base + reg);
239 }
240 
241 static void stm32_dma_write(struct stm32_dma_device *dmadev, u32 reg, u32 val)
242 {
243 	writel_relaxed(val, dmadev->base + reg);
244 }
245 
246 static int stm32_dma_get_width(struct stm32_dma_chan *chan,
247 			       enum dma_slave_buswidth width)
248 {
249 	switch (width) {
250 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
251 		return STM32_DMA_BYTE;
252 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
253 		return STM32_DMA_HALF_WORD;
254 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
255 		return STM32_DMA_WORD;
256 	default:
257 		dev_err(chan2dev(chan), "Dma bus width not supported\n");
258 		return -EINVAL;
259 	}
260 }
261 
262 static enum dma_slave_buswidth stm32_dma_get_max_width(u32 buf_len,
263 						       u32 threshold)
264 {
265 	enum dma_slave_buswidth max_width;
266 
267 	if (threshold == STM32_DMA_FIFO_THRESHOLD_FULL)
268 		max_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
269 	else
270 		max_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
271 
272 	while ((buf_len < max_width  || buf_len % max_width) &&
273 	       max_width > DMA_SLAVE_BUSWIDTH_1_BYTE)
274 		max_width = max_width >> 1;
275 
276 	return max_width;
277 }
278 
279 static bool stm32_dma_fifo_threshold_is_allowed(u32 burst, u32 threshold,
280 						enum dma_slave_buswidth width)
281 {
282 	u32 remaining;
283 
284 	if (width != DMA_SLAVE_BUSWIDTH_UNDEFINED) {
285 		if (burst != 0) {
286 			/*
287 			 * If number of beats fit in several whole bursts
288 			 * this configuration is allowed.
289 			 */
290 			remaining = ((STM32_DMA_FIFO_SIZE / width) *
291 				     (threshold + 1) / 4) % burst;
292 
293 			if (remaining == 0)
294 				return true;
295 		} else {
296 			return true;
297 		}
298 	}
299 
300 	return false;
301 }
302 
303 static bool stm32_dma_is_burst_possible(u32 buf_len, u32 threshold)
304 {
305 	/*
306 	 * Buffer or period length has to be aligned on FIFO depth.
307 	 * Otherwise bytes may be stuck within FIFO at buffer or period
308 	 * length.
309 	 */
310 	return ((buf_len % ((threshold + 1) * 4)) == 0);
311 }
312 
313 static u32 stm32_dma_get_best_burst(u32 buf_len, u32 max_burst, u32 threshold,
314 				    enum dma_slave_buswidth width)
315 {
316 	u32 best_burst = max_burst;
317 
318 	if (best_burst == 1 || !stm32_dma_is_burst_possible(buf_len, threshold))
319 		return 0;
320 
321 	while ((buf_len < best_burst * width && best_burst > 1) ||
322 	       !stm32_dma_fifo_threshold_is_allowed(best_burst, threshold,
323 						    width)) {
324 		if (best_burst > STM32_DMA_MIN_BURST)
325 			best_burst = best_burst >> 1;
326 		else
327 			best_burst = 0;
328 	}
329 
330 	return best_burst;
331 }
332 
333 static int stm32_dma_get_burst(struct stm32_dma_chan *chan, u32 maxburst)
334 {
335 	switch (maxburst) {
336 	case 0:
337 	case 1:
338 		return STM32_DMA_BURST_SINGLE;
339 	case 4:
340 		return STM32_DMA_BURST_INCR4;
341 	case 8:
342 		return STM32_DMA_BURST_INCR8;
343 	case 16:
344 		return STM32_DMA_BURST_INCR16;
345 	default:
346 		dev_err(chan2dev(chan), "Dma burst size not supported\n");
347 		return -EINVAL;
348 	}
349 }
350 
351 static void stm32_dma_set_fifo_config(struct stm32_dma_chan *chan,
352 				      u32 src_burst, u32 dst_burst)
353 {
354 	chan->chan_reg.dma_sfcr &= ~STM32_DMA_SFCR_MASK;
355 	chan->chan_reg.dma_scr &= ~STM32_DMA_SCR_DMEIE;
356 
357 	if (!src_burst && !dst_burst) {
358 		/* Using direct mode */
359 		chan->chan_reg.dma_scr |= STM32_DMA_SCR_DMEIE;
360 	} else {
361 		/* Using FIFO mode */
362 		chan->chan_reg.dma_sfcr |= STM32_DMA_SFCR_MASK;
363 	}
364 }
365 
366 static int stm32_dma_slave_config(struct dma_chan *c,
367 				  struct dma_slave_config *config)
368 {
369 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
370 
371 	memcpy(&chan->dma_sconfig, config, sizeof(*config));
372 
373 	chan->config_init = true;
374 
375 	return 0;
376 }
377 
378 static u32 stm32_dma_irq_status(struct stm32_dma_chan *chan)
379 {
380 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
381 	u32 flags, dma_isr;
382 
383 	/*
384 	 * Read "flags" from DMA_xISR register corresponding to the selected
385 	 * DMA channel at the correct bit offset inside that register.
386 	 *
387 	 * If (ch % 4) is 2 or 3, left shift the mask by 16 bits.
388 	 * If (ch % 4) is 1 or 3, additionally left shift the mask by 6 bits.
389 	 */
390 
391 	if (chan->id & 4)
392 		dma_isr = stm32_dma_read(dmadev, STM32_DMA_HISR);
393 	else
394 		dma_isr = stm32_dma_read(dmadev, STM32_DMA_LISR);
395 
396 	flags = dma_isr >> (((chan->id & 2) << 3) | ((chan->id & 1) * 6));
397 
398 	return flags & STM32_DMA_MASKI;
399 }
400 
401 static void stm32_dma_irq_clear(struct stm32_dma_chan *chan, u32 flags)
402 {
403 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
404 	u32 dma_ifcr;
405 
406 	/*
407 	 * Write "flags" to the DMA_xIFCR register corresponding to the selected
408 	 * DMA channel at the correct bit offset inside that register.
409 	 *
410 	 * If (ch % 4) is 2 or 3, left shift the mask by 16 bits.
411 	 * If (ch % 4) is 1 or 3, additionally left shift the mask by 6 bits.
412 	 */
413 	flags &= STM32_DMA_MASKI;
414 	dma_ifcr = flags << (((chan->id & 2) << 3) | ((chan->id & 1) * 6));
415 
416 	if (chan->id & 4)
417 		stm32_dma_write(dmadev, STM32_DMA_HIFCR, dma_ifcr);
418 	else
419 		stm32_dma_write(dmadev, STM32_DMA_LIFCR, dma_ifcr);
420 }
421 
422 static int stm32_dma_disable_chan(struct stm32_dma_chan *chan)
423 {
424 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
425 	u32 dma_scr, id, reg;
426 
427 	id = chan->id;
428 	reg = STM32_DMA_SCR(id);
429 	dma_scr = stm32_dma_read(dmadev, reg);
430 
431 	if (dma_scr & STM32_DMA_SCR_EN) {
432 		dma_scr &= ~STM32_DMA_SCR_EN;
433 		stm32_dma_write(dmadev, reg, dma_scr);
434 
435 		return readl_relaxed_poll_timeout_atomic(dmadev->base + reg,
436 					dma_scr, !(dma_scr & STM32_DMA_SCR_EN),
437 					10, 1000000);
438 	}
439 
440 	return 0;
441 }
442 
443 static void stm32_dma_stop(struct stm32_dma_chan *chan)
444 {
445 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
446 	u32 dma_scr, dma_sfcr, status;
447 	int ret;
448 
449 	/* Disable interrupts */
450 	dma_scr = stm32_dma_read(dmadev, STM32_DMA_SCR(chan->id));
451 	dma_scr &= ~STM32_DMA_SCR_IRQ_MASK;
452 	stm32_dma_write(dmadev, STM32_DMA_SCR(chan->id), dma_scr);
453 	dma_sfcr = stm32_dma_read(dmadev, STM32_DMA_SFCR(chan->id));
454 	dma_sfcr &= ~STM32_DMA_SFCR_FEIE;
455 	stm32_dma_write(dmadev, STM32_DMA_SFCR(chan->id), dma_sfcr);
456 
457 	/* Disable DMA */
458 	ret = stm32_dma_disable_chan(chan);
459 	if (ret < 0)
460 		return;
461 
462 	/* Clear interrupt status if it is there */
463 	status = stm32_dma_irq_status(chan);
464 	if (status) {
465 		dev_dbg(chan2dev(chan), "%s(): clearing interrupt: 0x%08x\n",
466 			__func__, status);
467 		stm32_dma_irq_clear(chan, status);
468 	}
469 
470 	chan->busy = false;
471 }
472 
473 static int stm32_dma_terminate_all(struct dma_chan *c)
474 {
475 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
476 	unsigned long flags;
477 	LIST_HEAD(head);
478 
479 	spin_lock_irqsave(&chan->vchan.lock, flags);
480 
481 	if (chan->desc) {
482 		vchan_terminate_vdesc(&chan->desc->vdesc);
483 		if (chan->busy)
484 			stm32_dma_stop(chan);
485 		chan->desc = NULL;
486 	}
487 
488 	vchan_get_all_descriptors(&chan->vchan, &head);
489 	spin_unlock_irqrestore(&chan->vchan.lock, flags);
490 	vchan_dma_desc_free_list(&chan->vchan, &head);
491 
492 	return 0;
493 }
494 
495 static void stm32_dma_synchronize(struct dma_chan *c)
496 {
497 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
498 
499 	vchan_synchronize(&chan->vchan);
500 }
501 
502 static void stm32_dma_dump_reg(struct stm32_dma_chan *chan)
503 {
504 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
505 	u32 scr = stm32_dma_read(dmadev, STM32_DMA_SCR(chan->id));
506 	u32 ndtr = stm32_dma_read(dmadev, STM32_DMA_SNDTR(chan->id));
507 	u32 spar = stm32_dma_read(dmadev, STM32_DMA_SPAR(chan->id));
508 	u32 sm0ar = stm32_dma_read(dmadev, STM32_DMA_SM0AR(chan->id));
509 	u32 sm1ar = stm32_dma_read(dmadev, STM32_DMA_SM1AR(chan->id));
510 	u32 sfcr = stm32_dma_read(dmadev, STM32_DMA_SFCR(chan->id));
511 
512 	dev_dbg(chan2dev(chan), "SCR:   0x%08x\n", scr);
513 	dev_dbg(chan2dev(chan), "NDTR:  0x%08x\n", ndtr);
514 	dev_dbg(chan2dev(chan), "SPAR:  0x%08x\n", spar);
515 	dev_dbg(chan2dev(chan), "SM0AR: 0x%08x\n", sm0ar);
516 	dev_dbg(chan2dev(chan), "SM1AR: 0x%08x\n", sm1ar);
517 	dev_dbg(chan2dev(chan), "SFCR:  0x%08x\n", sfcr);
518 }
519 
520 static void stm32_dma_configure_next_sg(struct stm32_dma_chan *chan);
521 
522 static void stm32_dma_start_transfer(struct stm32_dma_chan *chan)
523 {
524 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
525 	struct virt_dma_desc *vdesc;
526 	struct stm32_dma_sg_req *sg_req;
527 	struct stm32_dma_chan_reg *reg;
528 	u32 status;
529 	int ret;
530 
531 	ret = stm32_dma_disable_chan(chan);
532 	if (ret < 0)
533 		return;
534 
535 	if (!chan->desc) {
536 		vdesc = vchan_next_desc(&chan->vchan);
537 		if (!vdesc)
538 			return;
539 
540 		list_del(&vdesc->node);
541 
542 		chan->desc = to_stm32_dma_desc(vdesc);
543 		chan->next_sg = 0;
544 	}
545 
546 	if (chan->next_sg == chan->desc->num_sgs)
547 		chan->next_sg = 0;
548 
549 	sg_req = &chan->desc->sg_req[chan->next_sg];
550 	reg = &sg_req->chan_reg;
551 
552 	reg->dma_scr &= ~STM32_DMA_SCR_EN;
553 	stm32_dma_write(dmadev, STM32_DMA_SCR(chan->id), reg->dma_scr);
554 	stm32_dma_write(dmadev, STM32_DMA_SPAR(chan->id), reg->dma_spar);
555 	stm32_dma_write(dmadev, STM32_DMA_SM0AR(chan->id), reg->dma_sm0ar);
556 	stm32_dma_write(dmadev, STM32_DMA_SFCR(chan->id), reg->dma_sfcr);
557 	stm32_dma_write(dmadev, STM32_DMA_SM1AR(chan->id), reg->dma_sm1ar);
558 	stm32_dma_write(dmadev, STM32_DMA_SNDTR(chan->id), reg->dma_sndtr);
559 
560 	chan->next_sg++;
561 
562 	/* Clear interrupt status if it is there */
563 	status = stm32_dma_irq_status(chan);
564 	if (status)
565 		stm32_dma_irq_clear(chan, status);
566 
567 	if (chan->desc->cyclic)
568 		stm32_dma_configure_next_sg(chan);
569 
570 	stm32_dma_dump_reg(chan);
571 
572 	/* Start DMA */
573 	reg->dma_scr |= STM32_DMA_SCR_EN;
574 	stm32_dma_write(dmadev, STM32_DMA_SCR(chan->id), reg->dma_scr);
575 
576 	chan->busy = true;
577 
578 	dev_dbg(chan2dev(chan), "vchan %pK: started\n", &chan->vchan);
579 }
580 
581 static void stm32_dma_configure_next_sg(struct stm32_dma_chan *chan)
582 {
583 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
584 	struct stm32_dma_sg_req *sg_req;
585 	u32 dma_scr, dma_sm0ar, dma_sm1ar, id;
586 
587 	id = chan->id;
588 	dma_scr = stm32_dma_read(dmadev, STM32_DMA_SCR(id));
589 
590 	if (dma_scr & STM32_DMA_SCR_DBM) {
591 		if (chan->next_sg == chan->desc->num_sgs)
592 			chan->next_sg = 0;
593 
594 		sg_req = &chan->desc->sg_req[chan->next_sg];
595 
596 		if (dma_scr & STM32_DMA_SCR_CT) {
597 			dma_sm0ar = sg_req->chan_reg.dma_sm0ar;
598 			stm32_dma_write(dmadev, STM32_DMA_SM0AR(id), dma_sm0ar);
599 			dev_dbg(chan2dev(chan), "CT=1 <=> SM0AR: 0x%08x\n",
600 				stm32_dma_read(dmadev, STM32_DMA_SM0AR(id)));
601 		} else {
602 			dma_sm1ar = sg_req->chan_reg.dma_sm1ar;
603 			stm32_dma_write(dmadev, STM32_DMA_SM1AR(id), dma_sm1ar);
604 			dev_dbg(chan2dev(chan), "CT=0 <=> SM1AR: 0x%08x\n",
605 				stm32_dma_read(dmadev, STM32_DMA_SM1AR(id)));
606 		}
607 	}
608 }
609 
610 static void stm32_dma_handle_chan_done(struct stm32_dma_chan *chan)
611 {
612 	if (chan->desc) {
613 		if (chan->desc->cyclic) {
614 			vchan_cyclic_callback(&chan->desc->vdesc);
615 			chan->next_sg++;
616 			stm32_dma_configure_next_sg(chan);
617 		} else {
618 			chan->busy = false;
619 			if (chan->next_sg == chan->desc->num_sgs) {
620 				vchan_cookie_complete(&chan->desc->vdesc);
621 				chan->desc = NULL;
622 			}
623 			stm32_dma_start_transfer(chan);
624 		}
625 	}
626 }
627 
628 static irqreturn_t stm32_dma_chan_irq(int irq, void *devid)
629 {
630 	struct stm32_dma_chan *chan = devid;
631 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
632 	u32 status, scr, sfcr;
633 
634 	spin_lock(&chan->vchan.lock);
635 
636 	status = stm32_dma_irq_status(chan);
637 	scr = stm32_dma_read(dmadev, STM32_DMA_SCR(chan->id));
638 	sfcr = stm32_dma_read(dmadev, STM32_DMA_SFCR(chan->id));
639 
640 	if (status & STM32_DMA_TCI) {
641 		stm32_dma_irq_clear(chan, STM32_DMA_TCI);
642 		if (scr & STM32_DMA_SCR_TCIE)
643 			stm32_dma_handle_chan_done(chan);
644 		status &= ~STM32_DMA_TCI;
645 	}
646 	if (status & STM32_DMA_HTI) {
647 		stm32_dma_irq_clear(chan, STM32_DMA_HTI);
648 		status &= ~STM32_DMA_HTI;
649 	}
650 	if (status & STM32_DMA_FEI) {
651 		stm32_dma_irq_clear(chan, STM32_DMA_FEI);
652 		status &= ~STM32_DMA_FEI;
653 		if (sfcr & STM32_DMA_SFCR_FEIE) {
654 			if (!(scr & STM32_DMA_SCR_EN))
655 				dev_err(chan2dev(chan), "FIFO Error\n");
656 			else
657 				dev_dbg(chan2dev(chan), "FIFO over/underrun\n");
658 		}
659 	}
660 	if (status) {
661 		stm32_dma_irq_clear(chan, status);
662 		dev_err(chan2dev(chan), "DMA error: status=0x%08x\n", status);
663 		if (!(scr & STM32_DMA_SCR_EN))
664 			dev_err(chan2dev(chan), "chan disabled by HW\n");
665 	}
666 
667 	spin_unlock(&chan->vchan.lock);
668 
669 	return IRQ_HANDLED;
670 }
671 
672 static void stm32_dma_issue_pending(struct dma_chan *c)
673 {
674 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
675 	unsigned long flags;
676 
677 	spin_lock_irqsave(&chan->vchan.lock, flags);
678 	if (vchan_issue_pending(&chan->vchan) && !chan->desc && !chan->busy) {
679 		dev_dbg(chan2dev(chan), "vchan %pK: issued\n", &chan->vchan);
680 		stm32_dma_start_transfer(chan);
681 
682 	}
683 	spin_unlock_irqrestore(&chan->vchan.lock, flags);
684 }
685 
686 static int stm32_dma_set_xfer_param(struct stm32_dma_chan *chan,
687 				    enum dma_transfer_direction direction,
688 				    enum dma_slave_buswidth *buswidth,
689 				    u32 buf_len)
690 {
691 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
692 	int src_bus_width, dst_bus_width;
693 	int src_burst_size, dst_burst_size;
694 	u32 src_maxburst, dst_maxburst, src_best_burst, dst_best_burst;
695 	u32 dma_scr, threshold;
696 
697 	src_addr_width = chan->dma_sconfig.src_addr_width;
698 	dst_addr_width = chan->dma_sconfig.dst_addr_width;
699 	src_maxburst = chan->dma_sconfig.src_maxburst;
700 	dst_maxburst = chan->dma_sconfig.dst_maxburst;
701 	threshold = chan->threshold;
702 
703 	switch (direction) {
704 	case DMA_MEM_TO_DEV:
705 		/* Set device data size */
706 		dst_bus_width = stm32_dma_get_width(chan, dst_addr_width);
707 		if (dst_bus_width < 0)
708 			return dst_bus_width;
709 
710 		/* Set device burst size */
711 		dst_best_burst = stm32_dma_get_best_burst(buf_len,
712 							  dst_maxburst,
713 							  threshold,
714 							  dst_addr_width);
715 
716 		dst_burst_size = stm32_dma_get_burst(chan, dst_best_burst);
717 		if (dst_burst_size < 0)
718 			return dst_burst_size;
719 
720 		/* Set memory data size */
721 		src_addr_width = stm32_dma_get_max_width(buf_len, threshold);
722 		chan->mem_width = src_addr_width;
723 		src_bus_width = stm32_dma_get_width(chan, src_addr_width);
724 		if (src_bus_width < 0)
725 			return src_bus_width;
726 
727 		/* Set memory burst size */
728 		src_maxburst = STM32_DMA_MAX_BURST;
729 		src_best_burst = stm32_dma_get_best_burst(buf_len,
730 							  src_maxburst,
731 							  threshold,
732 							  src_addr_width);
733 		src_burst_size = stm32_dma_get_burst(chan, src_best_burst);
734 		if (src_burst_size < 0)
735 			return src_burst_size;
736 
737 		dma_scr = STM32_DMA_SCR_DIR(STM32_DMA_MEM_TO_DEV) |
738 			STM32_DMA_SCR_PSIZE(dst_bus_width) |
739 			STM32_DMA_SCR_MSIZE(src_bus_width) |
740 			STM32_DMA_SCR_PBURST(dst_burst_size) |
741 			STM32_DMA_SCR_MBURST(src_burst_size);
742 
743 		/* Set FIFO threshold */
744 		chan->chan_reg.dma_sfcr &= ~STM32_DMA_SFCR_FTH_MASK;
745 		chan->chan_reg.dma_sfcr |= STM32_DMA_SFCR_FTH(threshold);
746 
747 		/* Set peripheral address */
748 		chan->chan_reg.dma_spar = chan->dma_sconfig.dst_addr;
749 		*buswidth = dst_addr_width;
750 		break;
751 
752 	case DMA_DEV_TO_MEM:
753 		/* Set device data size */
754 		src_bus_width = stm32_dma_get_width(chan, src_addr_width);
755 		if (src_bus_width < 0)
756 			return src_bus_width;
757 
758 		/* Set device burst size */
759 		src_best_burst = stm32_dma_get_best_burst(buf_len,
760 							  src_maxburst,
761 							  threshold,
762 							  src_addr_width);
763 		chan->mem_burst = src_best_burst;
764 		src_burst_size = stm32_dma_get_burst(chan, src_best_burst);
765 		if (src_burst_size < 0)
766 			return src_burst_size;
767 
768 		/* Set memory data size */
769 		dst_addr_width = stm32_dma_get_max_width(buf_len, threshold);
770 		chan->mem_width = dst_addr_width;
771 		dst_bus_width = stm32_dma_get_width(chan, dst_addr_width);
772 		if (dst_bus_width < 0)
773 			return dst_bus_width;
774 
775 		/* Set memory burst size */
776 		dst_maxburst = STM32_DMA_MAX_BURST;
777 		dst_best_burst = stm32_dma_get_best_burst(buf_len,
778 							  dst_maxburst,
779 							  threshold,
780 							  dst_addr_width);
781 		chan->mem_burst = dst_best_burst;
782 		dst_burst_size = stm32_dma_get_burst(chan, dst_best_burst);
783 		if (dst_burst_size < 0)
784 			return dst_burst_size;
785 
786 		dma_scr = STM32_DMA_SCR_DIR(STM32_DMA_DEV_TO_MEM) |
787 			STM32_DMA_SCR_PSIZE(src_bus_width) |
788 			STM32_DMA_SCR_MSIZE(dst_bus_width) |
789 			STM32_DMA_SCR_PBURST(src_burst_size) |
790 			STM32_DMA_SCR_MBURST(dst_burst_size);
791 
792 		/* Set FIFO threshold */
793 		chan->chan_reg.dma_sfcr &= ~STM32_DMA_SFCR_FTH_MASK;
794 		chan->chan_reg.dma_sfcr |= STM32_DMA_SFCR_FTH(threshold);
795 
796 		/* Set peripheral address */
797 		chan->chan_reg.dma_spar = chan->dma_sconfig.src_addr;
798 		*buswidth = chan->dma_sconfig.src_addr_width;
799 		break;
800 
801 	default:
802 		dev_err(chan2dev(chan), "Dma direction is not supported\n");
803 		return -EINVAL;
804 	}
805 
806 	stm32_dma_set_fifo_config(chan, src_best_burst, dst_best_burst);
807 
808 	/* Set DMA control register */
809 	chan->chan_reg.dma_scr &= ~(STM32_DMA_SCR_DIR_MASK |
810 			STM32_DMA_SCR_PSIZE_MASK | STM32_DMA_SCR_MSIZE_MASK |
811 			STM32_DMA_SCR_PBURST_MASK | STM32_DMA_SCR_MBURST_MASK);
812 	chan->chan_reg.dma_scr |= dma_scr;
813 
814 	return 0;
815 }
816 
817 static void stm32_dma_clear_reg(struct stm32_dma_chan_reg *regs)
818 {
819 	memset(regs, 0, sizeof(struct stm32_dma_chan_reg));
820 }
821 
822 static struct dma_async_tx_descriptor *stm32_dma_prep_slave_sg(
823 	struct dma_chan *c, struct scatterlist *sgl,
824 	u32 sg_len, enum dma_transfer_direction direction,
825 	unsigned long flags, void *context)
826 {
827 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
828 	struct stm32_dma_desc *desc;
829 	struct scatterlist *sg;
830 	enum dma_slave_buswidth buswidth;
831 	u32 nb_data_items;
832 	int i, ret;
833 
834 	if (!chan->config_init) {
835 		dev_err(chan2dev(chan), "dma channel is not configured\n");
836 		return NULL;
837 	}
838 
839 	if (sg_len < 1) {
840 		dev_err(chan2dev(chan), "Invalid segment length %d\n", sg_len);
841 		return NULL;
842 	}
843 
844 	desc = kzalloc(struct_size(desc, sg_req, sg_len), GFP_NOWAIT);
845 	if (!desc)
846 		return NULL;
847 
848 	/* Set peripheral flow controller */
849 	if (chan->dma_sconfig.device_fc)
850 		chan->chan_reg.dma_scr |= STM32_DMA_SCR_PFCTRL;
851 	else
852 		chan->chan_reg.dma_scr &= ~STM32_DMA_SCR_PFCTRL;
853 
854 	for_each_sg(sgl, sg, sg_len, i) {
855 		ret = stm32_dma_set_xfer_param(chan, direction, &buswidth,
856 					       sg_dma_len(sg));
857 		if (ret < 0)
858 			goto err;
859 
860 		desc->sg_req[i].len = sg_dma_len(sg);
861 
862 		nb_data_items = desc->sg_req[i].len / buswidth;
863 		if (nb_data_items > STM32_DMA_ALIGNED_MAX_DATA_ITEMS) {
864 			dev_err(chan2dev(chan), "nb items not supported\n");
865 			goto err;
866 		}
867 
868 		stm32_dma_clear_reg(&desc->sg_req[i].chan_reg);
869 		desc->sg_req[i].chan_reg.dma_scr = chan->chan_reg.dma_scr;
870 		desc->sg_req[i].chan_reg.dma_sfcr = chan->chan_reg.dma_sfcr;
871 		desc->sg_req[i].chan_reg.dma_spar = chan->chan_reg.dma_spar;
872 		desc->sg_req[i].chan_reg.dma_sm0ar = sg_dma_address(sg);
873 		desc->sg_req[i].chan_reg.dma_sm1ar = sg_dma_address(sg);
874 		desc->sg_req[i].chan_reg.dma_sndtr = nb_data_items;
875 	}
876 
877 	desc->num_sgs = sg_len;
878 	desc->cyclic = false;
879 
880 	return vchan_tx_prep(&chan->vchan, &desc->vdesc, flags);
881 
882 err:
883 	kfree(desc);
884 	return NULL;
885 }
886 
887 static struct dma_async_tx_descriptor *stm32_dma_prep_dma_cyclic(
888 	struct dma_chan *c, dma_addr_t buf_addr, size_t buf_len,
889 	size_t period_len, enum dma_transfer_direction direction,
890 	unsigned long flags)
891 {
892 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
893 	struct stm32_dma_desc *desc;
894 	enum dma_slave_buswidth buswidth;
895 	u32 num_periods, nb_data_items;
896 	int i, ret;
897 
898 	if (!buf_len || !period_len) {
899 		dev_err(chan2dev(chan), "Invalid buffer/period len\n");
900 		return NULL;
901 	}
902 
903 	if (!chan->config_init) {
904 		dev_err(chan2dev(chan), "dma channel is not configured\n");
905 		return NULL;
906 	}
907 
908 	if (buf_len % period_len) {
909 		dev_err(chan2dev(chan), "buf_len not multiple of period_len\n");
910 		return NULL;
911 	}
912 
913 	/*
914 	 * We allow to take more number of requests till DMA is
915 	 * not started. The driver will loop over all requests.
916 	 * Once DMA is started then new requests can be queued only after
917 	 * terminating the DMA.
918 	 */
919 	if (chan->busy) {
920 		dev_err(chan2dev(chan), "Request not allowed when dma busy\n");
921 		return NULL;
922 	}
923 
924 	ret = stm32_dma_set_xfer_param(chan, direction, &buswidth, period_len);
925 	if (ret < 0)
926 		return NULL;
927 
928 	nb_data_items = period_len / buswidth;
929 	if (nb_data_items > STM32_DMA_ALIGNED_MAX_DATA_ITEMS) {
930 		dev_err(chan2dev(chan), "number of items not supported\n");
931 		return NULL;
932 	}
933 
934 	/*  Enable Circular mode or double buffer mode */
935 	if (buf_len == period_len)
936 		chan->chan_reg.dma_scr |= STM32_DMA_SCR_CIRC;
937 	else
938 		chan->chan_reg.dma_scr |= STM32_DMA_SCR_DBM;
939 
940 	/* Clear periph ctrl if client set it */
941 	chan->chan_reg.dma_scr &= ~STM32_DMA_SCR_PFCTRL;
942 
943 	num_periods = buf_len / period_len;
944 
945 	desc = kzalloc(struct_size(desc, sg_req, num_periods), GFP_NOWAIT);
946 	if (!desc)
947 		return NULL;
948 
949 	for (i = 0; i < num_periods; i++) {
950 		desc->sg_req[i].len = period_len;
951 
952 		stm32_dma_clear_reg(&desc->sg_req[i].chan_reg);
953 		desc->sg_req[i].chan_reg.dma_scr = chan->chan_reg.dma_scr;
954 		desc->sg_req[i].chan_reg.dma_sfcr = chan->chan_reg.dma_sfcr;
955 		desc->sg_req[i].chan_reg.dma_spar = chan->chan_reg.dma_spar;
956 		desc->sg_req[i].chan_reg.dma_sm0ar = buf_addr;
957 		desc->sg_req[i].chan_reg.dma_sm1ar = buf_addr;
958 		desc->sg_req[i].chan_reg.dma_sndtr = nb_data_items;
959 		buf_addr += period_len;
960 	}
961 
962 	desc->num_sgs = num_periods;
963 	desc->cyclic = true;
964 
965 	return vchan_tx_prep(&chan->vchan, &desc->vdesc, flags);
966 }
967 
968 static struct dma_async_tx_descriptor *stm32_dma_prep_dma_memcpy(
969 	struct dma_chan *c, dma_addr_t dest,
970 	dma_addr_t src, size_t len, unsigned long flags)
971 {
972 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
973 	enum dma_slave_buswidth max_width;
974 	struct stm32_dma_desc *desc;
975 	size_t xfer_count, offset;
976 	u32 num_sgs, best_burst, dma_burst, threshold;
977 	int i;
978 
979 	num_sgs = DIV_ROUND_UP(len, STM32_DMA_ALIGNED_MAX_DATA_ITEMS);
980 	desc = kzalloc(struct_size(desc, sg_req, num_sgs), GFP_NOWAIT);
981 	if (!desc)
982 		return NULL;
983 
984 	threshold = chan->threshold;
985 
986 	for (offset = 0, i = 0; offset < len; offset += xfer_count, i++) {
987 		xfer_count = min_t(size_t, len - offset,
988 				   STM32_DMA_ALIGNED_MAX_DATA_ITEMS);
989 
990 		/* Compute best burst size */
991 		max_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
992 		best_burst = stm32_dma_get_best_burst(len, STM32_DMA_MAX_BURST,
993 						      threshold, max_width);
994 		dma_burst = stm32_dma_get_burst(chan, best_burst);
995 
996 		stm32_dma_clear_reg(&desc->sg_req[i].chan_reg);
997 		desc->sg_req[i].chan_reg.dma_scr =
998 			STM32_DMA_SCR_DIR(STM32_DMA_MEM_TO_MEM) |
999 			STM32_DMA_SCR_PBURST(dma_burst) |
1000 			STM32_DMA_SCR_MBURST(dma_burst) |
1001 			STM32_DMA_SCR_MINC |
1002 			STM32_DMA_SCR_PINC |
1003 			STM32_DMA_SCR_TCIE |
1004 			STM32_DMA_SCR_TEIE;
1005 		desc->sg_req[i].chan_reg.dma_sfcr |= STM32_DMA_SFCR_MASK;
1006 		desc->sg_req[i].chan_reg.dma_sfcr |=
1007 			STM32_DMA_SFCR_FTH(threshold);
1008 		desc->sg_req[i].chan_reg.dma_spar = src + offset;
1009 		desc->sg_req[i].chan_reg.dma_sm0ar = dest + offset;
1010 		desc->sg_req[i].chan_reg.dma_sndtr = xfer_count;
1011 		desc->sg_req[i].len = xfer_count;
1012 	}
1013 
1014 	desc->num_sgs = num_sgs;
1015 	desc->cyclic = false;
1016 
1017 	return vchan_tx_prep(&chan->vchan, &desc->vdesc, flags);
1018 }
1019 
1020 static u32 stm32_dma_get_remaining_bytes(struct stm32_dma_chan *chan)
1021 {
1022 	u32 dma_scr, width, ndtr;
1023 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
1024 
1025 	dma_scr = stm32_dma_read(dmadev, STM32_DMA_SCR(chan->id));
1026 	width = STM32_DMA_SCR_PSIZE_GET(dma_scr);
1027 	ndtr = stm32_dma_read(dmadev, STM32_DMA_SNDTR(chan->id));
1028 
1029 	return ndtr << width;
1030 }
1031 
1032 /**
1033  * stm32_dma_is_current_sg - check that expected sg_req is currently transferred
1034  * @chan: dma channel
1035  *
1036  * This function called when IRQ are disable, checks that the hardware has not
1037  * switched on the next transfer in double buffer mode. The test is done by
1038  * comparing the next_sg memory address with the hardware related register
1039  * (based on CT bit value).
1040  *
1041  * Returns true if expected current transfer is still running or double
1042  * buffer mode is not activated.
1043  */
1044 static bool stm32_dma_is_current_sg(struct stm32_dma_chan *chan)
1045 {
1046 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
1047 	struct stm32_dma_sg_req *sg_req;
1048 	u32 dma_scr, dma_smar, id;
1049 
1050 	id = chan->id;
1051 	dma_scr = stm32_dma_read(dmadev, STM32_DMA_SCR(id));
1052 
1053 	if (!(dma_scr & STM32_DMA_SCR_DBM))
1054 		return true;
1055 
1056 	sg_req = &chan->desc->sg_req[chan->next_sg];
1057 
1058 	if (dma_scr & STM32_DMA_SCR_CT) {
1059 		dma_smar = stm32_dma_read(dmadev, STM32_DMA_SM0AR(id));
1060 		return (dma_smar == sg_req->chan_reg.dma_sm0ar);
1061 	}
1062 
1063 	dma_smar = stm32_dma_read(dmadev, STM32_DMA_SM1AR(id));
1064 
1065 	return (dma_smar == sg_req->chan_reg.dma_sm1ar);
1066 }
1067 
1068 static size_t stm32_dma_desc_residue(struct stm32_dma_chan *chan,
1069 				     struct stm32_dma_desc *desc,
1070 				     u32 next_sg)
1071 {
1072 	u32 modulo, burst_size;
1073 	u32 residue;
1074 	u32 n_sg = next_sg;
1075 	struct stm32_dma_sg_req *sg_req = &chan->desc->sg_req[chan->next_sg];
1076 	int i;
1077 
1078 	/*
1079 	 * Calculate the residue means compute the descriptors
1080 	 * information:
1081 	 * - the sg_req currently transferred
1082 	 * - the Hardware remaining position in this sg (NDTR bits field).
1083 	 *
1084 	 * A race condition may occur if DMA is running in cyclic or double
1085 	 * buffer mode, since the DMA register are automatically reloaded at end
1086 	 * of period transfer. The hardware may have switched to the next
1087 	 * transfer (CT bit updated) just before the position (SxNDTR reg) is
1088 	 * read.
1089 	 * In this case the SxNDTR reg could (or not) correspond to the new
1090 	 * transfer position, and not the expected one.
1091 	 * The strategy implemented in the stm32 driver is to:
1092 	 *  - read the SxNDTR register
1093 	 *  - crosscheck that hardware is still in current transfer.
1094 	 * In case of switch, we can assume that the DMA is at the beginning of
1095 	 * the next transfer. So we approximate the residue in consequence, by
1096 	 * pointing on the beginning of next transfer.
1097 	 *
1098 	 * This race condition doesn't apply for none cyclic mode, as double
1099 	 * buffer is not used. In such situation registers are updated by the
1100 	 * software.
1101 	 */
1102 
1103 	residue = stm32_dma_get_remaining_bytes(chan);
1104 
1105 	if (!stm32_dma_is_current_sg(chan)) {
1106 		n_sg++;
1107 		if (n_sg == chan->desc->num_sgs)
1108 			n_sg = 0;
1109 		residue = sg_req->len;
1110 	}
1111 
1112 	/*
1113 	 * In cyclic mode, for the last period, residue = remaining bytes
1114 	 * from NDTR,
1115 	 * else for all other periods in cyclic mode, and in sg mode,
1116 	 * residue = remaining bytes from NDTR + remaining
1117 	 * periods/sg to be transferred
1118 	 */
1119 	if (!chan->desc->cyclic || n_sg != 0)
1120 		for (i = n_sg; i < desc->num_sgs; i++)
1121 			residue += desc->sg_req[i].len;
1122 
1123 	if (!chan->mem_burst)
1124 		return residue;
1125 
1126 	burst_size = chan->mem_burst * chan->mem_width;
1127 	modulo = residue % burst_size;
1128 	if (modulo)
1129 		residue = residue - modulo + burst_size;
1130 
1131 	return residue;
1132 }
1133 
1134 static enum dma_status stm32_dma_tx_status(struct dma_chan *c,
1135 					   dma_cookie_t cookie,
1136 					   struct dma_tx_state *state)
1137 {
1138 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
1139 	struct virt_dma_desc *vdesc;
1140 	enum dma_status status;
1141 	unsigned long flags;
1142 	u32 residue = 0;
1143 
1144 	status = dma_cookie_status(c, cookie, state);
1145 	if (status == DMA_COMPLETE || !state)
1146 		return status;
1147 
1148 	spin_lock_irqsave(&chan->vchan.lock, flags);
1149 	vdesc = vchan_find_desc(&chan->vchan, cookie);
1150 	if (chan->desc && cookie == chan->desc->vdesc.tx.cookie)
1151 		residue = stm32_dma_desc_residue(chan, chan->desc,
1152 						 chan->next_sg);
1153 	else if (vdesc)
1154 		residue = stm32_dma_desc_residue(chan,
1155 						 to_stm32_dma_desc(vdesc), 0);
1156 	dma_set_residue(state, residue);
1157 
1158 	spin_unlock_irqrestore(&chan->vchan.lock, flags);
1159 
1160 	return status;
1161 }
1162 
1163 static int stm32_dma_alloc_chan_resources(struct dma_chan *c)
1164 {
1165 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
1166 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
1167 	int ret;
1168 
1169 	chan->config_init = false;
1170 
1171 	ret = pm_runtime_get_sync(dmadev->ddev.dev);
1172 	if (ret < 0)
1173 		return ret;
1174 
1175 	ret = stm32_dma_disable_chan(chan);
1176 	if (ret < 0)
1177 		pm_runtime_put(dmadev->ddev.dev);
1178 
1179 	return ret;
1180 }
1181 
1182 static void stm32_dma_free_chan_resources(struct dma_chan *c)
1183 {
1184 	struct stm32_dma_chan *chan = to_stm32_dma_chan(c);
1185 	struct stm32_dma_device *dmadev = stm32_dma_get_dev(chan);
1186 	unsigned long flags;
1187 
1188 	dev_dbg(chan2dev(chan), "Freeing channel %d\n", chan->id);
1189 
1190 	if (chan->busy) {
1191 		spin_lock_irqsave(&chan->vchan.lock, flags);
1192 		stm32_dma_stop(chan);
1193 		chan->desc = NULL;
1194 		spin_unlock_irqrestore(&chan->vchan.lock, flags);
1195 	}
1196 
1197 	pm_runtime_put(dmadev->ddev.dev);
1198 
1199 	vchan_free_chan_resources(to_virt_chan(c));
1200 }
1201 
1202 static void stm32_dma_desc_free(struct virt_dma_desc *vdesc)
1203 {
1204 	kfree(container_of(vdesc, struct stm32_dma_desc, vdesc));
1205 }
1206 
1207 static void stm32_dma_set_config(struct stm32_dma_chan *chan,
1208 				 struct stm32_dma_cfg *cfg)
1209 {
1210 	stm32_dma_clear_reg(&chan->chan_reg);
1211 
1212 	chan->chan_reg.dma_scr = cfg->stream_config & STM32_DMA_SCR_CFG_MASK;
1213 	chan->chan_reg.dma_scr |= STM32_DMA_SCR_REQ(cfg->request_line);
1214 
1215 	/* Enable Interrupts  */
1216 	chan->chan_reg.dma_scr |= STM32_DMA_SCR_TEIE | STM32_DMA_SCR_TCIE;
1217 
1218 	chan->threshold = STM32_DMA_THRESHOLD_FTR_GET(cfg->features);
1219 }
1220 
1221 static struct dma_chan *stm32_dma_of_xlate(struct of_phandle_args *dma_spec,
1222 					   struct of_dma *ofdma)
1223 {
1224 	struct stm32_dma_device *dmadev = ofdma->of_dma_data;
1225 	struct device *dev = dmadev->ddev.dev;
1226 	struct stm32_dma_cfg cfg;
1227 	struct stm32_dma_chan *chan;
1228 	struct dma_chan *c;
1229 
1230 	if (dma_spec->args_count < 4) {
1231 		dev_err(dev, "Bad number of cells\n");
1232 		return NULL;
1233 	}
1234 
1235 	cfg.channel_id = dma_spec->args[0];
1236 	cfg.request_line = dma_spec->args[1];
1237 	cfg.stream_config = dma_spec->args[2];
1238 	cfg.features = dma_spec->args[3];
1239 
1240 	if (cfg.channel_id >= STM32_DMA_MAX_CHANNELS ||
1241 	    cfg.request_line >= STM32_DMA_MAX_REQUEST_ID) {
1242 		dev_err(dev, "Bad channel and/or request id\n");
1243 		return NULL;
1244 	}
1245 
1246 	chan = &dmadev->chan[cfg.channel_id];
1247 
1248 	c = dma_get_slave_channel(&chan->vchan.chan);
1249 	if (!c) {
1250 		dev_err(dev, "No more channels available\n");
1251 		return NULL;
1252 	}
1253 
1254 	stm32_dma_set_config(chan, &cfg);
1255 
1256 	return c;
1257 }
1258 
1259 static const struct of_device_id stm32_dma_of_match[] = {
1260 	{ .compatible = "st,stm32-dma", },
1261 	{ /* sentinel */ },
1262 };
1263 MODULE_DEVICE_TABLE(of, stm32_dma_of_match);
1264 
1265 static int stm32_dma_probe(struct platform_device *pdev)
1266 {
1267 	struct stm32_dma_chan *chan;
1268 	struct stm32_dma_device *dmadev;
1269 	struct dma_device *dd;
1270 	const struct of_device_id *match;
1271 	struct resource *res;
1272 	struct reset_control *rst;
1273 	int i, ret;
1274 
1275 	match = of_match_device(stm32_dma_of_match, &pdev->dev);
1276 	if (!match) {
1277 		dev_err(&pdev->dev, "Error: No device match found\n");
1278 		return -ENODEV;
1279 	}
1280 
1281 	dmadev = devm_kzalloc(&pdev->dev, sizeof(*dmadev), GFP_KERNEL);
1282 	if (!dmadev)
1283 		return -ENOMEM;
1284 
1285 	dd = &dmadev->ddev;
1286 
1287 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1288 	dmadev->base = devm_ioremap_resource(&pdev->dev, res);
1289 	if (IS_ERR(dmadev->base))
1290 		return PTR_ERR(dmadev->base);
1291 
1292 	dmadev->clk = devm_clk_get(&pdev->dev, NULL);
1293 	if (IS_ERR(dmadev->clk)) {
1294 		ret = PTR_ERR(dmadev->clk);
1295 		if (ret != -EPROBE_DEFER)
1296 			dev_err(&pdev->dev, "Can't get clock\n");
1297 		return ret;
1298 	}
1299 
1300 	ret = clk_prepare_enable(dmadev->clk);
1301 	if (ret < 0) {
1302 		dev_err(&pdev->dev, "clk_prep_enable error: %d\n", ret);
1303 		return ret;
1304 	}
1305 
1306 	dmadev->mem2mem = of_property_read_bool(pdev->dev.of_node,
1307 						"st,mem2mem");
1308 
1309 	rst = devm_reset_control_get(&pdev->dev, NULL);
1310 	if (IS_ERR(rst)) {
1311 		ret = PTR_ERR(rst);
1312 		if (ret == -EPROBE_DEFER)
1313 			goto clk_free;
1314 	} else {
1315 		reset_control_assert(rst);
1316 		udelay(2);
1317 		reset_control_deassert(rst);
1318 	}
1319 
1320 	dma_set_max_seg_size(&pdev->dev, STM32_DMA_ALIGNED_MAX_DATA_ITEMS);
1321 
1322 	dma_cap_set(DMA_SLAVE, dd->cap_mask);
1323 	dma_cap_set(DMA_PRIVATE, dd->cap_mask);
1324 	dma_cap_set(DMA_CYCLIC, dd->cap_mask);
1325 	dd->device_alloc_chan_resources = stm32_dma_alloc_chan_resources;
1326 	dd->device_free_chan_resources = stm32_dma_free_chan_resources;
1327 	dd->device_tx_status = stm32_dma_tx_status;
1328 	dd->device_issue_pending = stm32_dma_issue_pending;
1329 	dd->device_prep_slave_sg = stm32_dma_prep_slave_sg;
1330 	dd->device_prep_dma_cyclic = stm32_dma_prep_dma_cyclic;
1331 	dd->device_config = stm32_dma_slave_config;
1332 	dd->device_terminate_all = stm32_dma_terminate_all;
1333 	dd->device_synchronize = stm32_dma_synchronize;
1334 	dd->src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1335 		BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1336 		BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
1337 	dd->dst_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1338 		BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1339 		BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
1340 	dd->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
1341 	dd->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1342 	dd->copy_align = DMAENGINE_ALIGN_32_BYTES;
1343 	dd->max_burst = STM32_DMA_MAX_BURST;
1344 	dd->descriptor_reuse = true;
1345 	dd->dev = &pdev->dev;
1346 	INIT_LIST_HEAD(&dd->channels);
1347 
1348 	if (dmadev->mem2mem) {
1349 		dma_cap_set(DMA_MEMCPY, dd->cap_mask);
1350 		dd->device_prep_dma_memcpy = stm32_dma_prep_dma_memcpy;
1351 		dd->directions |= BIT(DMA_MEM_TO_MEM);
1352 	}
1353 
1354 	for (i = 0; i < STM32_DMA_MAX_CHANNELS; i++) {
1355 		chan = &dmadev->chan[i];
1356 		chan->id = i;
1357 		chan->vchan.desc_free = stm32_dma_desc_free;
1358 		vchan_init(&chan->vchan, dd);
1359 	}
1360 
1361 	ret = dma_async_device_register(dd);
1362 	if (ret)
1363 		goto clk_free;
1364 
1365 	for (i = 0; i < STM32_DMA_MAX_CHANNELS; i++) {
1366 		chan = &dmadev->chan[i];
1367 		ret = platform_get_irq(pdev, i);
1368 		if (ret < 0)
1369 			goto err_unregister;
1370 		chan->irq = ret;
1371 
1372 		ret = devm_request_irq(&pdev->dev, chan->irq,
1373 				       stm32_dma_chan_irq, 0,
1374 				       dev_name(chan2dev(chan)), chan);
1375 		if (ret) {
1376 			dev_err(&pdev->dev,
1377 				"request_irq failed with err %d channel %d\n",
1378 				ret, i);
1379 			goto err_unregister;
1380 		}
1381 	}
1382 
1383 	ret = of_dma_controller_register(pdev->dev.of_node,
1384 					 stm32_dma_of_xlate, dmadev);
1385 	if (ret < 0) {
1386 		dev_err(&pdev->dev,
1387 			"STM32 DMA DMA OF registration failed %d\n", ret);
1388 		goto err_unregister;
1389 	}
1390 
1391 	platform_set_drvdata(pdev, dmadev);
1392 
1393 	pm_runtime_set_active(&pdev->dev);
1394 	pm_runtime_enable(&pdev->dev);
1395 	pm_runtime_get_noresume(&pdev->dev);
1396 	pm_runtime_put(&pdev->dev);
1397 
1398 	dev_info(&pdev->dev, "STM32 DMA driver registered\n");
1399 
1400 	return 0;
1401 
1402 err_unregister:
1403 	dma_async_device_unregister(dd);
1404 clk_free:
1405 	clk_disable_unprepare(dmadev->clk);
1406 
1407 	return ret;
1408 }
1409 
1410 #ifdef CONFIG_PM
1411 static int stm32_dma_runtime_suspend(struct device *dev)
1412 {
1413 	struct stm32_dma_device *dmadev = dev_get_drvdata(dev);
1414 
1415 	clk_disable_unprepare(dmadev->clk);
1416 
1417 	return 0;
1418 }
1419 
1420 static int stm32_dma_runtime_resume(struct device *dev)
1421 {
1422 	struct stm32_dma_device *dmadev = dev_get_drvdata(dev);
1423 	int ret;
1424 
1425 	ret = clk_prepare_enable(dmadev->clk);
1426 	if (ret) {
1427 		dev_err(dev, "failed to prepare_enable clock\n");
1428 		return ret;
1429 	}
1430 
1431 	return 0;
1432 }
1433 #endif
1434 
1435 #ifdef CONFIG_PM_SLEEP
1436 static int stm32_dma_suspend(struct device *dev)
1437 {
1438 	struct stm32_dma_device *dmadev = dev_get_drvdata(dev);
1439 	int id, ret, scr;
1440 
1441 	ret = pm_runtime_get_sync(dev);
1442 	if (ret < 0)
1443 		return ret;
1444 
1445 	for (id = 0; id < STM32_DMA_MAX_CHANNELS; id++) {
1446 		scr = stm32_dma_read(dmadev, STM32_DMA_SCR(id));
1447 		if (scr & STM32_DMA_SCR_EN) {
1448 			dev_warn(dev, "Suspend is prevented by Chan %i\n", id);
1449 			return -EBUSY;
1450 		}
1451 	}
1452 
1453 	pm_runtime_put_sync(dev);
1454 
1455 	pm_runtime_force_suspend(dev);
1456 
1457 	return 0;
1458 }
1459 
1460 static int stm32_dma_resume(struct device *dev)
1461 {
1462 	return pm_runtime_force_resume(dev);
1463 }
1464 #endif
1465 
1466 static const struct dev_pm_ops stm32_dma_pm_ops = {
1467 	SET_SYSTEM_SLEEP_PM_OPS(stm32_dma_suspend, stm32_dma_resume)
1468 	SET_RUNTIME_PM_OPS(stm32_dma_runtime_suspend,
1469 			   stm32_dma_runtime_resume, NULL)
1470 };
1471 
1472 static struct platform_driver stm32_dma_driver = {
1473 	.driver = {
1474 		.name = "stm32-dma",
1475 		.of_match_table = stm32_dma_of_match,
1476 		.pm = &stm32_dma_pm_ops,
1477 	},
1478 	.probe = stm32_dma_probe,
1479 };
1480 
1481 static int __init stm32_dma_init(void)
1482 {
1483 	return platform_driver_register(&stm32_dma_driver);
1484 }
1485 subsys_initcall(stm32_dma_init);
1486