xref: /openbmc/linux/drivers/dma/pxa_dma.c (revision bc5aa3a0)
1 /*
2  * Copyright 2015 Robert Jarzmik <robert.jarzmik@free.fr>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8 
9 #include <linux/err.h>
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/types.h>
13 #include <linux/interrupt.h>
14 #include <linux/dma-mapping.h>
15 #include <linux/slab.h>
16 #include <linux/dmaengine.h>
17 #include <linux/platform_device.h>
18 #include <linux/device.h>
19 #include <linux/platform_data/mmp_dma.h>
20 #include <linux/dmapool.h>
21 #include <linux/of_device.h>
22 #include <linux/of_dma.h>
23 #include <linux/of.h>
24 #include <linux/wait.h>
25 #include <linux/dma/pxa-dma.h>
26 
27 #include "dmaengine.h"
28 #include "virt-dma.h"
29 
30 #define DCSR(n)		(0x0000 + ((n) << 2))
31 #define DALGN(n)	0x00a0
32 #define DINT		0x00f0
33 #define DDADR(n)	(0x0200 + ((n) << 4))
34 #define DSADR(n)	(0x0204 + ((n) << 4))
35 #define DTADR(n)	(0x0208 + ((n) << 4))
36 #define DCMD(n)		(0x020c + ((n) << 4))
37 
38 #define PXA_DCSR_RUN		BIT(31)	/* Run Bit (read / write) */
39 #define PXA_DCSR_NODESC		BIT(30)	/* No-Descriptor Fetch (read / write) */
40 #define PXA_DCSR_STOPIRQEN	BIT(29)	/* Stop Interrupt Enable (R/W) */
41 #define PXA_DCSR_REQPEND	BIT(8)	/* Request Pending (read-only) */
42 #define PXA_DCSR_STOPSTATE	BIT(3)	/* Stop State (read-only) */
43 #define PXA_DCSR_ENDINTR	BIT(2)	/* End Interrupt (read / write) */
44 #define PXA_DCSR_STARTINTR	BIT(1)	/* Start Interrupt (read / write) */
45 #define PXA_DCSR_BUSERR		BIT(0)	/* Bus Error Interrupt (read / write) */
46 
47 #define PXA_DCSR_EORIRQEN	BIT(28)	/* End of Receive IRQ Enable (R/W) */
48 #define PXA_DCSR_EORJMPEN	BIT(27)	/* Jump to next descriptor on EOR */
49 #define PXA_DCSR_EORSTOPEN	BIT(26)	/* STOP on an EOR */
50 #define PXA_DCSR_SETCMPST	BIT(25)	/* Set Descriptor Compare Status */
51 #define PXA_DCSR_CLRCMPST	BIT(24)	/* Clear Descriptor Compare Status */
52 #define PXA_DCSR_CMPST		BIT(10)	/* The Descriptor Compare Status */
53 #define PXA_DCSR_EORINTR	BIT(9)	/* The end of Receive */
54 
55 #define DRCMR_MAPVLD	BIT(7)	/* Map Valid (read / write) */
56 #define DRCMR_CHLNUM	0x1f	/* mask for Channel Number (read / write) */
57 
58 #define DDADR_DESCADDR	0xfffffff0	/* Address of next descriptor (mask) */
59 #define DDADR_STOP	BIT(0)	/* Stop (read / write) */
60 
61 #define PXA_DCMD_INCSRCADDR	BIT(31)	/* Source Address Increment Setting. */
62 #define PXA_DCMD_INCTRGADDR	BIT(30)	/* Target Address Increment Setting. */
63 #define PXA_DCMD_FLOWSRC	BIT(29)	/* Flow Control by the source. */
64 #define PXA_DCMD_FLOWTRG	BIT(28)	/* Flow Control by the target. */
65 #define PXA_DCMD_STARTIRQEN	BIT(22)	/* Start Interrupt Enable */
66 #define PXA_DCMD_ENDIRQEN	BIT(21)	/* End Interrupt Enable */
67 #define PXA_DCMD_ENDIAN		BIT(18)	/* Device Endian-ness. */
68 #define PXA_DCMD_BURST8		(1 << 16)	/* 8 byte burst */
69 #define PXA_DCMD_BURST16	(2 << 16)	/* 16 byte burst */
70 #define PXA_DCMD_BURST32	(3 << 16)	/* 32 byte burst */
71 #define PXA_DCMD_WIDTH1		(1 << 14)	/* 1 byte width */
72 #define PXA_DCMD_WIDTH2		(2 << 14)	/* 2 byte width (HalfWord) */
73 #define PXA_DCMD_WIDTH4		(3 << 14)	/* 4 byte width (Word) */
74 #define PXA_DCMD_LENGTH		0x01fff		/* length mask (max = 8K - 1) */
75 
76 #define PDMA_ALIGNMENT		3
77 #define PDMA_MAX_DESC_BYTES	(PXA_DCMD_LENGTH & ~((1 << PDMA_ALIGNMENT) - 1))
78 
79 struct pxad_desc_hw {
80 	u32 ddadr;	/* Points to the next descriptor + flags */
81 	u32 dsadr;	/* DSADR value for the current transfer */
82 	u32 dtadr;	/* DTADR value for the current transfer */
83 	u32 dcmd;	/* DCMD value for the current transfer */
84 } __aligned(16);
85 
86 struct pxad_desc_sw {
87 	struct virt_dma_desc	vd;		/* Virtual descriptor */
88 	int			nb_desc;	/* Number of hw. descriptors */
89 	size_t			len;		/* Number of bytes xfered */
90 	dma_addr_t		first;		/* First descriptor's addr */
91 
92 	/* At least one descriptor has an src/dst address not multiple of 8 */
93 	bool			misaligned;
94 	bool			cyclic;
95 	struct dma_pool		*desc_pool;	/* Channel's used allocator */
96 
97 	struct pxad_desc_hw	*hw_desc[];	/* DMA coherent descriptors */
98 };
99 
100 struct pxad_phy {
101 	int			idx;
102 	void __iomem		*base;
103 	struct pxad_chan	*vchan;
104 };
105 
106 struct pxad_chan {
107 	struct virt_dma_chan	vc;		/* Virtual channel */
108 	u32			drcmr;		/* Requestor of the channel */
109 	enum pxad_chan_prio	prio;		/* Required priority of phy */
110 	/*
111 	 * At least one desc_sw in submitted or issued transfers on this channel
112 	 * has one address such as: addr % 8 != 0. This implies the DALGN
113 	 * setting on the phy.
114 	 */
115 	bool			misaligned;
116 	struct dma_slave_config	cfg;		/* Runtime config */
117 
118 	/* protected by vc->lock */
119 	struct pxad_phy		*phy;
120 	struct dma_pool		*desc_pool;	/* Descriptors pool */
121 	dma_cookie_t		bus_error;
122 
123 	wait_queue_head_t	wq_state;
124 };
125 
126 struct pxad_device {
127 	struct dma_device		slave;
128 	int				nr_chans;
129 	int				nr_requestors;
130 	void __iomem			*base;
131 	struct pxad_phy			*phys;
132 	spinlock_t			phy_lock;	/* Phy association */
133 #ifdef CONFIG_DEBUG_FS
134 	struct dentry			*dbgfs_root;
135 	struct dentry			*dbgfs_state;
136 	struct dentry			**dbgfs_chan;
137 #endif
138 };
139 
140 #define tx_to_pxad_desc(tx)					\
141 	container_of(tx, struct pxad_desc_sw, async_tx)
142 #define to_pxad_chan(dchan)					\
143 	container_of(dchan, struct pxad_chan, vc.chan)
144 #define to_pxad_dev(dmadev)					\
145 	container_of(dmadev, struct pxad_device, slave)
146 #define to_pxad_sw_desc(_vd)				\
147 	container_of((_vd), struct pxad_desc_sw, vd)
148 
149 #define _phy_readl_relaxed(phy, _reg)					\
150 	readl_relaxed((phy)->base + _reg((phy)->idx))
151 #define phy_readl_relaxed(phy, _reg)					\
152 	({								\
153 		u32 _v;							\
154 		_v = readl_relaxed((phy)->base + _reg((phy)->idx));	\
155 		dev_vdbg(&phy->vchan->vc.chan.dev->device,		\
156 			 "%s(): readl(%s): 0x%08x\n", __func__, #_reg,	\
157 			  _v);						\
158 		_v;							\
159 	})
160 #define phy_writel(phy, val, _reg)					\
161 	do {								\
162 		writel((val), (phy)->base + _reg((phy)->idx));		\
163 		dev_vdbg(&phy->vchan->vc.chan.dev->device,		\
164 			 "%s(): writel(0x%08x, %s)\n",			\
165 			 __func__, (u32)(val), #_reg);			\
166 	} while (0)
167 #define phy_writel_relaxed(phy, val, _reg)				\
168 	do {								\
169 		writel_relaxed((val), (phy)->base + _reg((phy)->idx));	\
170 		dev_vdbg(&phy->vchan->vc.chan.dev->device,		\
171 			 "%s(): writel_relaxed(0x%08x, %s)\n",		\
172 			 __func__, (u32)(val), #_reg);			\
173 	} while (0)
174 
175 static unsigned int pxad_drcmr(unsigned int line)
176 {
177 	if (line < 64)
178 		return 0x100 + line * 4;
179 	return 0x1000 + line * 4;
180 }
181 
182 /*
183  * Debug fs
184  */
185 #ifdef CONFIG_DEBUG_FS
186 #include <linux/debugfs.h>
187 #include <linux/uaccess.h>
188 #include <linux/seq_file.h>
189 
190 static int dbg_show_requester_chan(struct seq_file *s, void *p)
191 {
192 	struct pxad_phy *phy = s->private;
193 	int i;
194 	u32 drcmr;
195 
196 	seq_printf(s, "DMA channel %d requester :\n", phy->idx);
197 	for (i = 0; i < 70; i++) {
198 		drcmr = readl_relaxed(phy->base + pxad_drcmr(i));
199 		if ((drcmr & DRCMR_CHLNUM) == phy->idx)
200 			seq_printf(s, "\tRequester %d (MAPVLD=%d)\n", i,
201 				   !!(drcmr & DRCMR_MAPVLD));
202 	}
203 	return 0;
204 }
205 
206 static inline int dbg_burst_from_dcmd(u32 dcmd)
207 {
208 	int burst = (dcmd >> 16) & 0x3;
209 
210 	return burst ? 4 << burst : 0;
211 }
212 
213 static int is_phys_valid(unsigned long addr)
214 {
215 	return pfn_valid(__phys_to_pfn(addr));
216 }
217 
218 #define PXA_DCSR_STR(flag) (dcsr & PXA_DCSR_##flag ? #flag" " : "")
219 #define PXA_DCMD_STR(flag) (dcmd & PXA_DCMD_##flag ? #flag" " : "")
220 
221 static int dbg_show_descriptors(struct seq_file *s, void *p)
222 {
223 	struct pxad_phy *phy = s->private;
224 	int i, max_show = 20, burst, width;
225 	u32 dcmd;
226 	unsigned long phys_desc, ddadr;
227 	struct pxad_desc_hw *desc;
228 
229 	phys_desc = ddadr = _phy_readl_relaxed(phy, DDADR);
230 
231 	seq_printf(s, "DMA channel %d descriptors :\n", phy->idx);
232 	seq_printf(s, "[%03d] First descriptor unknown\n", 0);
233 	for (i = 1; i < max_show && is_phys_valid(phys_desc); i++) {
234 		desc = phys_to_virt(phys_desc);
235 		dcmd = desc->dcmd;
236 		burst = dbg_burst_from_dcmd(dcmd);
237 		width = (1 << ((dcmd >> 14) & 0x3)) >> 1;
238 
239 		seq_printf(s, "[%03d] Desc at %08lx(virt %p)\n",
240 			   i, phys_desc, desc);
241 		seq_printf(s, "\tDDADR = %08x\n", desc->ddadr);
242 		seq_printf(s, "\tDSADR = %08x\n", desc->dsadr);
243 		seq_printf(s, "\tDTADR = %08x\n", desc->dtadr);
244 		seq_printf(s, "\tDCMD  = %08x (%s%s%s%s%s%s%sburst=%d width=%d len=%d)\n",
245 			   dcmd,
246 			   PXA_DCMD_STR(INCSRCADDR), PXA_DCMD_STR(INCTRGADDR),
247 			   PXA_DCMD_STR(FLOWSRC), PXA_DCMD_STR(FLOWTRG),
248 			   PXA_DCMD_STR(STARTIRQEN), PXA_DCMD_STR(ENDIRQEN),
249 			   PXA_DCMD_STR(ENDIAN), burst, width,
250 			   dcmd & PXA_DCMD_LENGTH);
251 		phys_desc = desc->ddadr;
252 	}
253 	if (i == max_show)
254 		seq_printf(s, "[%03d] Desc at %08lx ... max display reached\n",
255 			   i, phys_desc);
256 	else
257 		seq_printf(s, "[%03d] Desc at %08lx is %s\n",
258 			   i, phys_desc, phys_desc == DDADR_STOP ?
259 			   "DDADR_STOP" : "invalid");
260 
261 	return 0;
262 }
263 
264 static int dbg_show_chan_state(struct seq_file *s, void *p)
265 {
266 	struct pxad_phy *phy = s->private;
267 	u32 dcsr, dcmd;
268 	int burst, width;
269 	static const char * const str_prio[] = {
270 		"high", "normal", "low", "invalid"
271 	};
272 
273 	dcsr = _phy_readl_relaxed(phy, DCSR);
274 	dcmd = _phy_readl_relaxed(phy, DCMD);
275 	burst = dbg_burst_from_dcmd(dcmd);
276 	width = (1 << ((dcmd >> 14) & 0x3)) >> 1;
277 
278 	seq_printf(s, "DMA channel %d\n", phy->idx);
279 	seq_printf(s, "\tPriority : %s\n",
280 			  str_prio[(phy->idx & 0xf) / 4]);
281 	seq_printf(s, "\tUnaligned transfer bit: %s\n",
282 			  _phy_readl_relaxed(phy, DALGN) & BIT(phy->idx) ?
283 			  "yes" : "no");
284 	seq_printf(s, "\tDCSR  = %08x (%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s)\n",
285 		   dcsr, PXA_DCSR_STR(RUN), PXA_DCSR_STR(NODESC),
286 		   PXA_DCSR_STR(STOPIRQEN), PXA_DCSR_STR(EORIRQEN),
287 		   PXA_DCSR_STR(EORJMPEN), PXA_DCSR_STR(EORSTOPEN),
288 		   PXA_DCSR_STR(SETCMPST), PXA_DCSR_STR(CLRCMPST),
289 		   PXA_DCSR_STR(CMPST), PXA_DCSR_STR(EORINTR),
290 		   PXA_DCSR_STR(REQPEND), PXA_DCSR_STR(STOPSTATE),
291 		   PXA_DCSR_STR(ENDINTR), PXA_DCSR_STR(STARTINTR),
292 		   PXA_DCSR_STR(BUSERR));
293 
294 	seq_printf(s, "\tDCMD  = %08x (%s%s%s%s%s%s%sburst=%d width=%d len=%d)\n",
295 		   dcmd,
296 		   PXA_DCMD_STR(INCSRCADDR), PXA_DCMD_STR(INCTRGADDR),
297 		   PXA_DCMD_STR(FLOWSRC), PXA_DCMD_STR(FLOWTRG),
298 		   PXA_DCMD_STR(STARTIRQEN), PXA_DCMD_STR(ENDIRQEN),
299 		   PXA_DCMD_STR(ENDIAN), burst, width, dcmd & PXA_DCMD_LENGTH);
300 	seq_printf(s, "\tDSADR = %08x\n", _phy_readl_relaxed(phy, DSADR));
301 	seq_printf(s, "\tDTADR = %08x\n", _phy_readl_relaxed(phy, DTADR));
302 	seq_printf(s, "\tDDADR = %08x\n", _phy_readl_relaxed(phy, DDADR));
303 
304 	return 0;
305 }
306 
307 static int dbg_show_state(struct seq_file *s, void *p)
308 {
309 	struct pxad_device *pdev = s->private;
310 
311 	/* basic device status */
312 	seq_puts(s, "DMA engine status\n");
313 	seq_printf(s, "\tChannel number: %d\n", pdev->nr_chans);
314 
315 	return 0;
316 }
317 
318 #define DBGFS_FUNC_DECL(name) \
319 static int dbg_open_##name(struct inode *inode, struct file *file) \
320 { \
321 	return single_open(file, dbg_show_##name, inode->i_private); \
322 } \
323 static const struct file_operations dbg_fops_##name = { \
324 	.open		= dbg_open_##name, \
325 	.llseek		= seq_lseek, \
326 	.read		= seq_read, \
327 	.release	= single_release, \
328 }
329 
330 DBGFS_FUNC_DECL(state);
331 DBGFS_FUNC_DECL(chan_state);
332 DBGFS_FUNC_DECL(descriptors);
333 DBGFS_FUNC_DECL(requester_chan);
334 
335 static struct dentry *pxad_dbg_alloc_chan(struct pxad_device *pdev,
336 					     int ch, struct dentry *chandir)
337 {
338 	char chan_name[11];
339 	struct dentry *chan, *chan_state = NULL, *chan_descr = NULL;
340 	struct dentry *chan_reqs = NULL;
341 	void *dt;
342 
343 	scnprintf(chan_name, sizeof(chan_name), "%d", ch);
344 	chan = debugfs_create_dir(chan_name, chandir);
345 	dt = (void *)&pdev->phys[ch];
346 
347 	if (chan)
348 		chan_state = debugfs_create_file("state", 0400, chan, dt,
349 						 &dbg_fops_chan_state);
350 	if (chan_state)
351 		chan_descr = debugfs_create_file("descriptors", 0400, chan, dt,
352 						 &dbg_fops_descriptors);
353 	if (chan_descr)
354 		chan_reqs = debugfs_create_file("requesters", 0400, chan, dt,
355 						&dbg_fops_requester_chan);
356 	if (!chan_reqs)
357 		goto err_state;
358 
359 	return chan;
360 
361 err_state:
362 	debugfs_remove_recursive(chan);
363 	return NULL;
364 }
365 
366 static void pxad_init_debugfs(struct pxad_device *pdev)
367 {
368 	int i;
369 	struct dentry *chandir;
370 
371 	pdev->dbgfs_root = debugfs_create_dir(dev_name(pdev->slave.dev), NULL);
372 	if (IS_ERR(pdev->dbgfs_root) || !pdev->dbgfs_root)
373 		goto err_root;
374 
375 	pdev->dbgfs_state = debugfs_create_file("state", 0400, pdev->dbgfs_root,
376 						pdev, &dbg_fops_state);
377 	if (!pdev->dbgfs_state)
378 		goto err_state;
379 
380 	pdev->dbgfs_chan =
381 		kmalloc_array(pdev->nr_chans, sizeof(*pdev->dbgfs_state),
382 			      GFP_KERNEL);
383 	if (!pdev->dbgfs_chan)
384 		goto err_alloc;
385 
386 	chandir = debugfs_create_dir("channels", pdev->dbgfs_root);
387 	if (!chandir)
388 		goto err_chandir;
389 
390 	for (i = 0; i < pdev->nr_chans; i++) {
391 		pdev->dbgfs_chan[i] = pxad_dbg_alloc_chan(pdev, i, chandir);
392 		if (!pdev->dbgfs_chan[i])
393 			goto err_chans;
394 	}
395 
396 	return;
397 err_chans:
398 err_chandir:
399 	kfree(pdev->dbgfs_chan);
400 err_alloc:
401 err_state:
402 	debugfs_remove_recursive(pdev->dbgfs_root);
403 err_root:
404 	pr_err("pxad: debugfs is not available\n");
405 }
406 
407 static void pxad_cleanup_debugfs(struct pxad_device *pdev)
408 {
409 	debugfs_remove_recursive(pdev->dbgfs_root);
410 }
411 #else
412 static inline void pxad_init_debugfs(struct pxad_device *pdev) {}
413 static inline void pxad_cleanup_debugfs(struct pxad_device *pdev) {}
414 #endif
415 
416 /*
417  * In the transition phase where legacy pxa handling is done at the same time as
418  * mmp_dma, the DMA physical channel split between the 2 DMA providers is done
419  * through legacy_reserved. Legacy code reserves DMA channels by settings
420  * corresponding bits in legacy_reserved.
421  */
422 static u32 legacy_reserved;
423 static u32 legacy_unavailable;
424 
425 static struct pxad_phy *lookup_phy(struct pxad_chan *pchan)
426 {
427 	int prio, i;
428 	struct pxad_device *pdev = to_pxad_dev(pchan->vc.chan.device);
429 	struct pxad_phy *phy, *found = NULL;
430 	unsigned long flags;
431 
432 	/*
433 	 * dma channel priorities
434 	 * ch 0 - 3,  16 - 19  <--> (0)
435 	 * ch 4 - 7,  20 - 23  <--> (1)
436 	 * ch 8 - 11, 24 - 27  <--> (2)
437 	 * ch 12 - 15, 28 - 31  <--> (3)
438 	 */
439 
440 	spin_lock_irqsave(&pdev->phy_lock, flags);
441 	for (prio = pchan->prio; prio >= PXAD_PRIO_HIGHEST; prio--) {
442 		for (i = 0; i < pdev->nr_chans; i++) {
443 			if (prio != (i & 0xf) >> 2)
444 				continue;
445 			if ((i < 32) && (legacy_reserved & BIT(i)))
446 				continue;
447 			phy = &pdev->phys[i];
448 			if (!phy->vchan) {
449 				phy->vchan = pchan;
450 				found = phy;
451 				if (i < 32)
452 					legacy_unavailable |= BIT(i);
453 				goto out_unlock;
454 			}
455 		}
456 	}
457 
458 out_unlock:
459 	spin_unlock_irqrestore(&pdev->phy_lock, flags);
460 	dev_dbg(&pchan->vc.chan.dev->device,
461 		"%s(): phy=%p(%d)\n", __func__, found,
462 		found ? found->idx : -1);
463 
464 	return found;
465 }
466 
467 static void pxad_free_phy(struct pxad_chan *chan)
468 {
469 	struct pxad_device *pdev = to_pxad_dev(chan->vc.chan.device);
470 	unsigned long flags;
471 	u32 reg;
472 	int i;
473 
474 	dev_dbg(&chan->vc.chan.dev->device,
475 		"%s(): freeing\n", __func__);
476 	if (!chan->phy)
477 		return;
478 
479 	/* clear the channel mapping in DRCMR */
480 	if (chan->drcmr <= pdev->nr_requestors) {
481 		reg = pxad_drcmr(chan->drcmr);
482 		writel_relaxed(0, chan->phy->base + reg);
483 	}
484 
485 	spin_lock_irqsave(&pdev->phy_lock, flags);
486 	for (i = 0; i < 32; i++)
487 		if (chan->phy == &pdev->phys[i])
488 			legacy_unavailable &= ~BIT(i);
489 	chan->phy->vchan = NULL;
490 	chan->phy = NULL;
491 	spin_unlock_irqrestore(&pdev->phy_lock, flags);
492 }
493 
494 static bool is_chan_running(struct pxad_chan *chan)
495 {
496 	u32 dcsr;
497 	struct pxad_phy *phy = chan->phy;
498 
499 	if (!phy)
500 		return false;
501 	dcsr = phy_readl_relaxed(phy, DCSR);
502 	return dcsr & PXA_DCSR_RUN;
503 }
504 
505 static bool is_running_chan_misaligned(struct pxad_chan *chan)
506 {
507 	u32 dalgn;
508 
509 	BUG_ON(!chan->phy);
510 	dalgn = phy_readl_relaxed(chan->phy, DALGN);
511 	return dalgn & (BIT(chan->phy->idx));
512 }
513 
514 static void phy_enable(struct pxad_phy *phy, bool misaligned)
515 {
516 	struct pxad_device *pdev;
517 	u32 reg, dalgn;
518 
519 	if (!phy->vchan)
520 		return;
521 
522 	dev_dbg(&phy->vchan->vc.chan.dev->device,
523 		"%s(); phy=%p(%d) misaligned=%d\n", __func__,
524 		phy, phy->idx, misaligned);
525 
526 	pdev = to_pxad_dev(phy->vchan->vc.chan.device);
527 	if (phy->vchan->drcmr <= pdev->nr_requestors) {
528 		reg = pxad_drcmr(phy->vchan->drcmr);
529 		writel_relaxed(DRCMR_MAPVLD | phy->idx, phy->base + reg);
530 	}
531 
532 	dalgn = phy_readl_relaxed(phy, DALGN);
533 	if (misaligned)
534 		dalgn |= BIT(phy->idx);
535 	else
536 		dalgn &= ~BIT(phy->idx);
537 	phy_writel_relaxed(phy, dalgn, DALGN);
538 
539 	phy_writel(phy, PXA_DCSR_STOPIRQEN | PXA_DCSR_ENDINTR |
540 		   PXA_DCSR_BUSERR | PXA_DCSR_RUN, DCSR);
541 }
542 
543 static void phy_disable(struct pxad_phy *phy)
544 {
545 	u32 dcsr;
546 
547 	if (!phy)
548 		return;
549 
550 	dcsr = phy_readl_relaxed(phy, DCSR);
551 	dev_dbg(&phy->vchan->vc.chan.dev->device,
552 		"%s(): phy=%p(%d)\n", __func__, phy, phy->idx);
553 	phy_writel(phy, dcsr & ~PXA_DCSR_RUN & ~PXA_DCSR_STOPIRQEN, DCSR);
554 }
555 
556 static void pxad_launch_chan(struct pxad_chan *chan,
557 				 struct pxad_desc_sw *desc)
558 {
559 	dev_dbg(&chan->vc.chan.dev->device,
560 		"%s(): desc=%p\n", __func__, desc);
561 	if (!chan->phy) {
562 		chan->phy = lookup_phy(chan);
563 		if (!chan->phy) {
564 			dev_dbg(&chan->vc.chan.dev->device,
565 				"%s(): no free dma channel\n", __func__);
566 			return;
567 		}
568 	}
569 	chan->bus_error = 0;
570 
571 	/*
572 	 * Program the descriptor's address into the DMA controller,
573 	 * then start the DMA transaction
574 	 */
575 	phy_writel(chan->phy, desc->first, DDADR);
576 	phy_enable(chan->phy, chan->misaligned);
577 	wake_up(&chan->wq_state);
578 }
579 
580 static void set_updater_desc(struct pxad_desc_sw *sw_desc,
581 			     unsigned long flags)
582 {
583 	struct pxad_desc_hw *updater =
584 		sw_desc->hw_desc[sw_desc->nb_desc - 1];
585 	dma_addr_t dma = sw_desc->hw_desc[sw_desc->nb_desc - 2]->ddadr;
586 
587 	updater->ddadr = DDADR_STOP;
588 	updater->dsadr = dma;
589 	updater->dtadr = dma + 8;
590 	updater->dcmd = PXA_DCMD_WIDTH4 | PXA_DCMD_BURST32 |
591 		(PXA_DCMD_LENGTH & sizeof(u32));
592 	if (flags & DMA_PREP_INTERRUPT)
593 		updater->dcmd |= PXA_DCMD_ENDIRQEN;
594 	if (sw_desc->cyclic)
595 		sw_desc->hw_desc[sw_desc->nb_desc - 2]->ddadr = sw_desc->first;
596 }
597 
598 static bool is_desc_completed(struct virt_dma_desc *vd)
599 {
600 	struct pxad_desc_sw *sw_desc = to_pxad_sw_desc(vd);
601 	struct pxad_desc_hw *updater =
602 		sw_desc->hw_desc[sw_desc->nb_desc - 1];
603 
604 	return updater->dtadr != (updater->dsadr + 8);
605 }
606 
607 static void pxad_desc_chain(struct virt_dma_desc *vd1,
608 				struct virt_dma_desc *vd2)
609 {
610 	struct pxad_desc_sw *desc1 = to_pxad_sw_desc(vd1);
611 	struct pxad_desc_sw *desc2 = to_pxad_sw_desc(vd2);
612 	dma_addr_t dma_to_chain;
613 
614 	dma_to_chain = desc2->first;
615 	desc1->hw_desc[desc1->nb_desc - 1]->ddadr = dma_to_chain;
616 }
617 
618 static bool pxad_try_hotchain(struct virt_dma_chan *vc,
619 				  struct virt_dma_desc *vd)
620 {
621 	struct virt_dma_desc *vd_last_issued = NULL;
622 	struct pxad_chan *chan = to_pxad_chan(&vc->chan);
623 
624 	/*
625 	 * Attempt to hot chain the tx if the phy is still running. This is
626 	 * considered successful only if either the channel is still running
627 	 * after the chaining, or if the chained transfer is completed after
628 	 * having been hot chained.
629 	 * A change of alignment is not allowed, and forbids hotchaining.
630 	 */
631 	if (is_chan_running(chan)) {
632 		BUG_ON(list_empty(&vc->desc_issued));
633 
634 		if (!is_running_chan_misaligned(chan) &&
635 		    to_pxad_sw_desc(vd)->misaligned)
636 			return false;
637 
638 		vd_last_issued = list_entry(vc->desc_issued.prev,
639 					    struct virt_dma_desc, node);
640 		pxad_desc_chain(vd_last_issued, vd);
641 		if (is_chan_running(chan) || is_desc_completed(vd))
642 			return true;
643 	}
644 
645 	return false;
646 }
647 
648 static unsigned int clear_chan_irq(struct pxad_phy *phy)
649 {
650 	u32 dcsr;
651 	u32 dint = readl(phy->base + DINT);
652 
653 	if (!(dint & BIT(phy->idx)))
654 		return PXA_DCSR_RUN;
655 
656 	/* clear irq */
657 	dcsr = phy_readl_relaxed(phy, DCSR);
658 	phy_writel(phy, dcsr, DCSR);
659 	if ((dcsr & PXA_DCSR_BUSERR) && (phy->vchan))
660 		dev_warn(&phy->vchan->vc.chan.dev->device,
661 			 "%s(chan=%p): PXA_DCSR_BUSERR\n",
662 			 __func__, &phy->vchan);
663 
664 	return dcsr & ~PXA_DCSR_RUN;
665 }
666 
667 static irqreturn_t pxad_chan_handler(int irq, void *dev_id)
668 {
669 	struct pxad_phy *phy = dev_id;
670 	struct pxad_chan *chan = phy->vchan;
671 	struct virt_dma_desc *vd, *tmp;
672 	unsigned int dcsr;
673 	unsigned long flags;
674 	bool vd_completed;
675 	dma_cookie_t last_started = 0;
676 
677 	BUG_ON(!chan);
678 
679 	dcsr = clear_chan_irq(phy);
680 	if (dcsr & PXA_DCSR_RUN)
681 		return IRQ_NONE;
682 
683 	spin_lock_irqsave(&chan->vc.lock, flags);
684 	list_for_each_entry_safe(vd, tmp, &chan->vc.desc_issued, node) {
685 		vd_completed = is_desc_completed(vd);
686 		dev_dbg(&chan->vc.chan.dev->device,
687 			"%s(): checking txd %p[%x]: completed=%d dcsr=0x%x\n",
688 			__func__, vd, vd->tx.cookie, vd_completed,
689 			dcsr);
690 		last_started = vd->tx.cookie;
691 		if (to_pxad_sw_desc(vd)->cyclic) {
692 			vchan_cyclic_callback(vd);
693 			break;
694 		}
695 		if (vd_completed) {
696 			list_del(&vd->node);
697 			vchan_cookie_complete(vd);
698 		} else {
699 			break;
700 		}
701 	}
702 
703 	if (dcsr & PXA_DCSR_BUSERR) {
704 		chan->bus_error = last_started;
705 		phy_disable(phy);
706 	}
707 
708 	if (!chan->bus_error && dcsr & PXA_DCSR_STOPSTATE) {
709 		dev_dbg(&chan->vc.chan.dev->device,
710 		"%s(): channel stopped, submitted_empty=%d issued_empty=%d",
711 			__func__,
712 			list_empty(&chan->vc.desc_submitted),
713 			list_empty(&chan->vc.desc_issued));
714 		phy_writel_relaxed(phy, dcsr & ~PXA_DCSR_STOPIRQEN, DCSR);
715 
716 		if (list_empty(&chan->vc.desc_issued)) {
717 			chan->misaligned =
718 				!list_empty(&chan->vc.desc_submitted);
719 		} else {
720 			vd = list_first_entry(&chan->vc.desc_issued,
721 					      struct virt_dma_desc, node);
722 			pxad_launch_chan(chan, to_pxad_sw_desc(vd));
723 		}
724 	}
725 	spin_unlock_irqrestore(&chan->vc.lock, flags);
726 	wake_up(&chan->wq_state);
727 
728 	return IRQ_HANDLED;
729 }
730 
731 static irqreturn_t pxad_int_handler(int irq, void *dev_id)
732 {
733 	struct pxad_device *pdev = dev_id;
734 	struct pxad_phy *phy;
735 	u32 dint = readl(pdev->base + DINT);
736 	int i, ret = IRQ_NONE;
737 
738 	while (dint) {
739 		i = __ffs(dint);
740 		dint &= (dint - 1);
741 		phy = &pdev->phys[i];
742 		if ((i < 32) && (legacy_reserved & BIT(i)))
743 			continue;
744 		if (pxad_chan_handler(irq, phy) == IRQ_HANDLED)
745 			ret = IRQ_HANDLED;
746 	}
747 
748 	return ret;
749 }
750 
751 static int pxad_alloc_chan_resources(struct dma_chan *dchan)
752 {
753 	struct pxad_chan *chan = to_pxad_chan(dchan);
754 	struct pxad_device *pdev = to_pxad_dev(chan->vc.chan.device);
755 
756 	if (chan->desc_pool)
757 		return 1;
758 
759 	chan->desc_pool = dma_pool_create(dma_chan_name(dchan),
760 					  pdev->slave.dev,
761 					  sizeof(struct pxad_desc_hw),
762 					  __alignof__(struct pxad_desc_hw),
763 					  0);
764 	if (!chan->desc_pool) {
765 		dev_err(&chan->vc.chan.dev->device,
766 			"%s(): unable to allocate descriptor pool\n",
767 			__func__);
768 		return -ENOMEM;
769 	}
770 
771 	return 1;
772 }
773 
774 static void pxad_free_chan_resources(struct dma_chan *dchan)
775 {
776 	struct pxad_chan *chan = to_pxad_chan(dchan);
777 
778 	vchan_free_chan_resources(&chan->vc);
779 	dma_pool_destroy(chan->desc_pool);
780 	chan->desc_pool = NULL;
781 
782 }
783 
784 static void pxad_free_desc(struct virt_dma_desc *vd)
785 {
786 	int i;
787 	dma_addr_t dma;
788 	struct pxad_desc_sw *sw_desc = to_pxad_sw_desc(vd);
789 
790 	BUG_ON(sw_desc->nb_desc == 0);
791 	for (i = sw_desc->nb_desc - 1; i >= 0; i--) {
792 		if (i > 0)
793 			dma = sw_desc->hw_desc[i - 1]->ddadr;
794 		else
795 			dma = sw_desc->first;
796 		dma_pool_free(sw_desc->desc_pool,
797 			      sw_desc->hw_desc[i], dma);
798 	}
799 	sw_desc->nb_desc = 0;
800 	kfree(sw_desc);
801 }
802 
803 static struct pxad_desc_sw *
804 pxad_alloc_desc(struct pxad_chan *chan, unsigned int nb_hw_desc)
805 {
806 	struct pxad_desc_sw *sw_desc;
807 	dma_addr_t dma;
808 	int i;
809 
810 	sw_desc = kzalloc(sizeof(*sw_desc) +
811 			  nb_hw_desc * sizeof(struct pxad_desc_hw *),
812 			  GFP_NOWAIT);
813 	if (!sw_desc)
814 		return NULL;
815 	sw_desc->desc_pool = chan->desc_pool;
816 
817 	for (i = 0; i < nb_hw_desc; i++) {
818 		sw_desc->hw_desc[i] = dma_pool_alloc(sw_desc->desc_pool,
819 						     GFP_NOWAIT, &dma);
820 		if (!sw_desc->hw_desc[i]) {
821 			dev_err(&chan->vc.chan.dev->device,
822 				"%s(): Couldn't allocate the %dth hw_desc from dma_pool %p\n",
823 				__func__, i, sw_desc->desc_pool);
824 			goto err;
825 		}
826 
827 		if (i == 0)
828 			sw_desc->first = dma;
829 		else
830 			sw_desc->hw_desc[i - 1]->ddadr = dma;
831 		sw_desc->nb_desc++;
832 	}
833 
834 	return sw_desc;
835 err:
836 	pxad_free_desc(&sw_desc->vd);
837 	return NULL;
838 }
839 
840 static dma_cookie_t pxad_tx_submit(struct dma_async_tx_descriptor *tx)
841 {
842 	struct virt_dma_chan *vc = to_virt_chan(tx->chan);
843 	struct pxad_chan *chan = to_pxad_chan(&vc->chan);
844 	struct virt_dma_desc *vd_chained = NULL,
845 		*vd = container_of(tx, struct virt_dma_desc, tx);
846 	dma_cookie_t cookie;
847 	unsigned long flags;
848 
849 	set_updater_desc(to_pxad_sw_desc(vd), tx->flags);
850 
851 	spin_lock_irqsave(&vc->lock, flags);
852 	cookie = dma_cookie_assign(tx);
853 
854 	if (list_empty(&vc->desc_submitted) && pxad_try_hotchain(vc, vd)) {
855 		list_move_tail(&vd->node, &vc->desc_issued);
856 		dev_dbg(&chan->vc.chan.dev->device,
857 			"%s(): txd %p[%x]: submitted (hot linked)\n",
858 			__func__, vd, cookie);
859 		goto out;
860 	}
861 
862 	/*
863 	 * Fallback to placing the tx in the submitted queue
864 	 */
865 	if (!list_empty(&vc->desc_submitted)) {
866 		vd_chained = list_entry(vc->desc_submitted.prev,
867 					struct virt_dma_desc, node);
868 		/*
869 		 * Only chain the descriptors if no new misalignment is
870 		 * introduced. If a new misalignment is chained, let the channel
871 		 * stop, and be relaunched in misalign mode from the irq
872 		 * handler.
873 		 */
874 		if (chan->misaligned || !to_pxad_sw_desc(vd)->misaligned)
875 			pxad_desc_chain(vd_chained, vd);
876 		else
877 			vd_chained = NULL;
878 	}
879 	dev_dbg(&chan->vc.chan.dev->device,
880 		"%s(): txd %p[%x]: submitted (%s linked)\n",
881 		__func__, vd, cookie, vd_chained ? "cold" : "not");
882 	list_move_tail(&vd->node, &vc->desc_submitted);
883 	chan->misaligned |= to_pxad_sw_desc(vd)->misaligned;
884 
885 out:
886 	spin_unlock_irqrestore(&vc->lock, flags);
887 	return cookie;
888 }
889 
890 static void pxad_issue_pending(struct dma_chan *dchan)
891 {
892 	struct pxad_chan *chan = to_pxad_chan(dchan);
893 	struct virt_dma_desc *vd_first;
894 	unsigned long flags;
895 
896 	spin_lock_irqsave(&chan->vc.lock, flags);
897 	if (list_empty(&chan->vc.desc_submitted))
898 		goto out;
899 
900 	vd_first = list_first_entry(&chan->vc.desc_submitted,
901 				    struct virt_dma_desc, node);
902 	dev_dbg(&chan->vc.chan.dev->device,
903 		"%s(): txd %p[%x]", __func__, vd_first, vd_first->tx.cookie);
904 
905 	vchan_issue_pending(&chan->vc);
906 	if (!pxad_try_hotchain(&chan->vc, vd_first))
907 		pxad_launch_chan(chan, to_pxad_sw_desc(vd_first));
908 out:
909 	spin_unlock_irqrestore(&chan->vc.lock, flags);
910 }
911 
912 static inline struct dma_async_tx_descriptor *
913 pxad_tx_prep(struct virt_dma_chan *vc, struct virt_dma_desc *vd,
914 		 unsigned long tx_flags)
915 {
916 	struct dma_async_tx_descriptor *tx;
917 	struct pxad_chan *chan = container_of(vc, struct pxad_chan, vc);
918 
919 	INIT_LIST_HEAD(&vd->node);
920 	tx = vchan_tx_prep(vc, vd, tx_flags);
921 	tx->tx_submit = pxad_tx_submit;
922 	dev_dbg(&chan->vc.chan.dev->device,
923 		"%s(): vc=%p txd=%p[%x] flags=0x%lx\n", __func__,
924 		vc, vd, vd->tx.cookie,
925 		tx_flags);
926 
927 	return tx;
928 }
929 
930 static void pxad_get_config(struct pxad_chan *chan,
931 			    enum dma_transfer_direction dir,
932 			    u32 *dcmd, u32 *dev_src, u32 *dev_dst)
933 {
934 	u32 maxburst = 0, dev_addr = 0;
935 	enum dma_slave_buswidth width = DMA_SLAVE_BUSWIDTH_UNDEFINED;
936 	struct pxad_device *pdev = to_pxad_dev(chan->vc.chan.device);
937 
938 	*dcmd = 0;
939 	if (dir == DMA_DEV_TO_MEM) {
940 		maxburst = chan->cfg.src_maxburst;
941 		width = chan->cfg.src_addr_width;
942 		dev_addr = chan->cfg.src_addr;
943 		*dev_src = dev_addr;
944 		*dcmd |= PXA_DCMD_INCTRGADDR;
945 		if (chan->drcmr <= pdev->nr_requestors)
946 			*dcmd |= PXA_DCMD_FLOWSRC;
947 	}
948 	if (dir == DMA_MEM_TO_DEV) {
949 		maxburst = chan->cfg.dst_maxburst;
950 		width = chan->cfg.dst_addr_width;
951 		dev_addr = chan->cfg.dst_addr;
952 		*dev_dst = dev_addr;
953 		*dcmd |= PXA_DCMD_INCSRCADDR;
954 		if (chan->drcmr <= pdev->nr_requestors)
955 			*dcmd |= PXA_DCMD_FLOWTRG;
956 	}
957 	if (dir == DMA_MEM_TO_MEM)
958 		*dcmd |= PXA_DCMD_BURST32 | PXA_DCMD_INCTRGADDR |
959 			PXA_DCMD_INCSRCADDR;
960 
961 	dev_dbg(&chan->vc.chan.dev->device,
962 		"%s(): dev_addr=0x%x maxburst=%d width=%d  dir=%d\n",
963 		__func__, dev_addr, maxburst, width, dir);
964 
965 	if (width == DMA_SLAVE_BUSWIDTH_1_BYTE)
966 		*dcmd |= PXA_DCMD_WIDTH1;
967 	else if (width == DMA_SLAVE_BUSWIDTH_2_BYTES)
968 		*dcmd |= PXA_DCMD_WIDTH2;
969 	else if (width == DMA_SLAVE_BUSWIDTH_4_BYTES)
970 		*dcmd |= PXA_DCMD_WIDTH4;
971 
972 	if (maxburst == 8)
973 		*dcmd |= PXA_DCMD_BURST8;
974 	else if (maxburst == 16)
975 		*dcmd |= PXA_DCMD_BURST16;
976 	else if (maxburst == 32)
977 		*dcmd |= PXA_DCMD_BURST32;
978 
979 	/* FIXME: drivers should be ported over to use the filter
980 	 * function. Once that's done, the following two lines can
981 	 * be removed.
982 	 */
983 	if (chan->cfg.slave_id)
984 		chan->drcmr = chan->cfg.slave_id;
985 }
986 
987 static struct dma_async_tx_descriptor *
988 pxad_prep_memcpy(struct dma_chan *dchan,
989 		 dma_addr_t dma_dst, dma_addr_t dma_src,
990 		 size_t len, unsigned long flags)
991 {
992 	struct pxad_chan *chan = to_pxad_chan(dchan);
993 	struct pxad_desc_sw *sw_desc;
994 	struct pxad_desc_hw *hw_desc;
995 	u32 dcmd;
996 	unsigned int i, nb_desc = 0;
997 	size_t copy;
998 
999 	if (!dchan || !len)
1000 		return NULL;
1001 
1002 	dev_dbg(&chan->vc.chan.dev->device,
1003 		"%s(): dma_dst=0x%lx dma_src=0x%lx len=%zu flags=%lx\n",
1004 		__func__, (unsigned long)dma_dst, (unsigned long)dma_src,
1005 		len, flags);
1006 	pxad_get_config(chan, DMA_MEM_TO_MEM, &dcmd, NULL, NULL);
1007 
1008 	nb_desc = DIV_ROUND_UP(len, PDMA_MAX_DESC_BYTES);
1009 	sw_desc = pxad_alloc_desc(chan, nb_desc + 1);
1010 	if (!sw_desc)
1011 		return NULL;
1012 	sw_desc->len = len;
1013 
1014 	if (!IS_ALIGNED(dma_src, 1 << PDMA_ALIGNMENT) ||
1015 	    !IS_ALIGNED(dma_dst, 1 << PDMA_ALIGNMENT))
1016 		sw_desc->misaligned = true;
1017 
1018 	i = 0;
1019 	do {
1020 		hw_desc = sw_desc->hw_desc[i++];
1021 		copy = min_t(size_t, len, PDMA_MAX_DESC_BYTES);
1022 		hw_desc->dcmd = dcmd | (PXA_DCMD_LENGTH & copy);
1023 		hw_desc->dsadr = dma_src;
1024 		hw_desc->dtadr = dma_dst;
1025 		len -= copy;
1026 		dma_src += copy;
1027 		dma_dst += copy;
1028 	} while (len);
1029 	set_updater_desc(sw_desc, flags);
1030 
1031 	return pxad_tx_prep(&chan->vc, &sw_desc->vd, flags);
1032 }
1033 
1034 static struct dma_async_tx_descriptor *
1035 pxad_prep_slave_sg(struct dma_chan *dchan, struct scatterlist *sgl,
1036 		   unsigned int sg_len, enum dma_transfer_direction dir,
1037 		   unsigned long flags, void *context)
1038 {
1039 	struct pxad_chan *chan = to_pxad_chan(dchan);
1040 	struct pxad_desc_sw *sw_desc;
1041 	size_t len, avail;
1042 	struct scatterlist *sg;
1043 	dma_addr_t dma;
1044 	u32 dcmd, dsadr = 0, dtadr = 0;
1045 	unsigned int nb_desc = 0, i, j = 0;
1046 
1047 	if ((sgl == NULL) || (sg_len == 0))
1048 		return NULL;
1049 
1050 	pxad_get_config(chan, dir, &dcmd, &dsadr, &dtadr);
1051 	dev_dbg(&chan->vc.chan.dev->device,
1052 		"%s(): dir=%d flags=%lx\n", __func__, dir, flags);
1053 
1054 	for_each_sg(sgl, sg, sg_len, i)
1055 		nb_desc += DIV_ROUND_UP(sg_dma_len(sg), PDMA_MAX_DESC_BYTES);
1056 	sw_desc = pxad_alloc_desc(chan, nb_desc + 1);
1057 	if (!sw_desc)
1058 		return NULL;
1059 
1060 	for_each_sg(sgl, sg, sg_len, i) {
1061 		dma = sg_dma_address(sg);
1062 		avail = sg_dma_len(sg);
1063 		sw_desc->len += avail;
1064 
1065 		do {
1066 			len = min_t(size_t, avail, PDMA_MAX_DESC_BYTES);
1067 			if (dma & 0x7)
1068 				sw_desc->misaligned = true;
1069 
1070 			sw_desc->hw_desc[j]->dcmd =
1071 				dcmd | (PXA_DCMD_LENGTH & len);
1072 			sw_desc->hw_desc[j]->dsadr = dsadr ? dsadr : dma;
1073 			sw_desc->hw_desc[j++]->dtadr = dtadr ? dtadr : dma;
1074 
1075 			dma += len;
1076 			avail -= len;
1077 		} while (avail);
1078 	}
1079 	set_updater_desc(sw_desc, flags);
1080 
1081 	return pxad_tx_prep(&chan->vc, &sw_desc->vd, flags);
1082 }
1083 
1084 static struct dma_async_tx_descriptor *
1085 pxad_prep_dma_cyclic(struct dma_chan *dchan,
1086 		     dma_addr_t buf_addr, size_t len, size_t period_len,
1087 		     enum dma_transfer_direction dir, unsigned long flags)
1088 {
1089 	struct pxad_chan *chan = to_pxad_chan(dchan);
1090 	struct pxad_desc_sw *sw_desc;
1091 	struct pxad_desc_hw **phw_desc;
1092 	dma_addr_t dma;
1093 	u32 dcmd, dsadr = 0, dtadr = 0;
1094 	unsigned int nb_desc = 0;
1095 
1096 	if (!dchan || !len || !period_len)
1097 		return NULL;
1098 	if ((dir != DMA_DEV_TO_MEM) && (dir != DMA_MEM_TO_DEV)) {
1099 		dev_err(&chan->vc.chan.dev->device,
1100 			"Unsupported direction for cyclic DMA\n");
1101 		return NULL;
1102 	}
1103 	/* the buffer length must be a multiple of period_len */
1104 	if (len % period_len != 0 || period_len > PDMA_MAX_DESC_BYTES ||
1105 	    !IS_ALIGNED(period_len, 1 << PDMA_ALIGNMENT))
1106 		return NULL;
1107 
1108 	pxad_get_config(chan, dir, &dcmd, &dsadr, &dtadr);
1109 	dcmd |= PXA_DCMD_ENDIRQEN | (PXA_DCMD_LENGTH & period_len);
1110 	dev_dbg(&chan->vc.chan.dev->device,
1111 		"%s(): buf_addr=0x%lx len=%zu period=%zu dir=%d flags=%lx\n",
1112 		__func__, (unsigned long)buf_addr, len, period_len, dir, flags);
1113 
1114 	nb_desc = DIV_ROUND_UP(period_len, PDMA_MAX_DESC_BYTES);
1115 	nb_desc *= DIV_ROUND_UP(len, period_len);
1116 	sw_desc = pxad_alloc_desc(chan, nb_desc + 1);
1117 	if (!sw_desc)
1118 		return NULL;
1119 	sw_desc->cyclic = true;
1120 	sw_desc->len = len;
1121 
1122 	phw_desc = sw_desc->hw_desc;
1123 	dma = buf_addr;
1124 	do {
1125 		phw_desc[0]->dsadr = dsadr ? dsadr : dma;
1126 		phw_desc[0]->dtadr = dtadr ? dtadr : dma;
1127 		phw_desc[0]->dcmd = dcmd;
1128 		phw_desc++;
1129 		dma += period_len;
1130 		len -= period_len;
1131 	} while (len);
1132 	set_updater_desc(sw_desc, flags);
1133 
1134 	return pxad_tx_prep(&chan->vc, &sw_desc->vd, flags);
1135 }
1136 
1137 static int pxad_config(struct dma_chan *dchan,
1138 		       struct dma_slave_config *cfg)
1139 {
1140 	struct pxad_chan *chan = to_pxad_chan(dchan);
1141 
1142 	if (!dchan)
1143 		return -EINVAL;
1144 
1145 	chan->cfg = *cfg;
1146 	return 0;
1147 }
1148 
1149 static int pxad_terminate_all(struct dma_chan *dchan)
1150 {
1151 	struct pxad_chan *chan = to_pxad_chan(dchan);
1152 	struct pxad_device *pdev = to_pxad_dev(chan->vc.chan.device);
1153 	struct virt_dma_desc *vd = NULL;
1154 	unsigned long flags;
1155 	struct pxad_phy *phy;
1156 	LIST_HEAD(head);
1157 
1158 	dev_dbg(&chan->vc.chan.dev->device,
1159 		"%s(): vchan %p: terminate all\n", __func__, &chan->vc);
1160 
1161 	spin_lock_irqsave(&chan->vc.lock, flags);
1162 	vchan_get_all_descriptors(&chan->vc, &head);
1163 
1164 	list_for_each_entry(vd, &head, node) {
1165 		dev_dbg(&chan->vc.chan.dev->device,
1166 			"%s(): cancelling txd %p[%x] (completed=%d)", __func__,
1167 			vd, vd->tx.cookie, is_desc_completed(vd));
1168 	}
1169 
1170 	phy = chan->phy;
1171 	if (phy) {
1172 		phy_disable(chan->phy);
1173 		pxad_free_phy(chan);
1174 		chan->phy = NULL;
1175 		spin_lock(&pdev->phy_lock);
1176 		phy->vchan = NULL;
1177 		spin_unlock(&pdev->phy_lock);
1178 	}
1179 	spin_unlock_irqrestore(&chan->vc.lock, flags);
1180 	vchan_dma_desc_free_list(&chan->vc, &head);
1181 
1182 	return 0;
1183 }
1184 
1185 static unsigned int pxad_residue(struct pxad_chan *chan,
1186 				 dma_cookie_t cookie)
1187 {
1188 	struct virt_dma_desc *vd = NULL;
1189 	struct pxad_desc_sw *sw_desc = NULL;
1190 	struct pxad_desc_hw *hw_desc = NULL;
1191 	u32 curr, start, len, end, residue = 0;
1192 	unsigned long flags;
1193 	bool passed = false;
1194 	int i;
1195 
1196 	/*
1197 	 * If the channel does not have a phy pointer anymore, it has already
1198 	 * been completed. Therefore, its residue is 0.
1199 	 */
1200 	if (!chan->phy)
1201 		return 0;
1202 
1203 	spin_lock_irqsave(&chan->vc.lock, flags);
1204 
1205 	vd = vchan_find_desc(&chan->vc, cookie);
1206 	if (!vd)
1207 		goto out;
1208 
1209 	sw_desc = to_pxad_sw_desc(vd);
1210 	if (sw_desc->hw_desc[0]->dcmd & PXA_DCMD_INCSRCADDR)
1211 		curr = phy_readl_relaxed(chan->phy, DSADR);
1212 	else
1213 		curr = phy_readl_relaxed(chan->phy, DTADR);
1214 
1215 	/*
1216 	 * curr has to be actually read before checking descriptor
1217 	 * completion, so that a curr inside a status updater
1218 	 * descriptor implies the following test returns true, and
1219 	 * preventing reordering of curr load and the test.
1220 	 */
1221 	rmb();
1222 	if (is_desc_completed(vd))
1223 		goto out;
1224 
1225 	for (i = 0; i < sw_desc->nb_desc - 1; i++) {
1226 		hw_desc = sw_desc->hw_desc[i];
1227 		if (sw_desc->hw_desc[0]->dcmd & PXA_DCMD_INCSRCADDR)
1228 			start = hw_desc->dsadr;
1229 		else
1230 			start = hw_desc->dtadr;
1231 		len = hw_desc->dcmd & PXA_DCMD_LENGTH;
1232 		end = start + len;
1233 
1234 		/*
1235 		 * 'passed' will be latched once we found the descriptor
1236 		 * which lies inside the boundaries of the curr
1237 		 * pointer. All descriptors that occur in the list
1238 		 * _after_ we found that partially handled descriptor
1239 		 * are still to be processed and are hence added to the
1240 		 * residual bytes counter.
1241 		 */
1242 
1243 		if (passed) {
1244 			residue += len;
1245 		} else if (curr >= start && curr <= end) {
1246 			residue += end - curr;
1247 			passed = true;
1248 		}
1249 	}
1250 	if (!passed)
1251 		residue = sw_desc->len;
1252 
1253 out:
1254 	spin_unlock_irqrestore(&chan->vc.lock, flags);
1255 	dev_dbg(&chan->vc.chan.dev->device,
1256 		"%s(): txd %p[%x] sw_desc=%p: %d\n",
1257 		__func__, vd, cookie, sw_desc, residue);
1258 	return residue;
1259 }
1260 
1261 static enum dma_status pxad_tx_status(struct dma_chan *dchan,
1262 				      dma_cookie_t cookie,
1263 				      struct dma_tx_state *txstate)
1264 {
1265 	struct pxad_chan *chan = to_pxad_chan(dchan);
1266 	enum dma_status ret;
1267 
1268 	if (cookie == chan->bus_error)
1269 		return DMA_ERROR;
1270 
1271 	ret = dma_cookie_status(dchan, cookie, txstate);
1272 	if (likely(txstate && (ret != DMA_ERROR)))
1273 		dma_set_residue(txstate, pxad_residue(chan, cookie));
1274 
1275 	return ret;
1276 }
1277 
1278 static void pxad_synchronize(struct dma_chan *dchan)
1279 {
1280 	struct pxad_chan *chan = to_pxad_chan(dchan);
1281 
1282 	wait_event(chan->wq_state, !is_chan_running(chan));
1283 	vchan_synchronize(&chan->vc);
1284 }
1285 
1286 static void pxad_free_channels(struct dma_device *dmadev)
1287 {
1288 	struct pxad_chan *c, *cn;
1289 
1290 	list_for_each_entry_safe(c, cn, &dmadev->channels,
1291 				 vc.chan.device_node) {
1292 		list_del(&c->vc.chan.device_node);
1293 		tasklet_kill(&c->vc.task);
1294 	}
1295 }
1296 
1297 static int pxad_remove(struct platform_device *op)
1298 {
1299 	struct pxad_device *pdev = platform_get_drvdata(op);
1300 
1301 	pxad_cleanup_debugfs(pdev);
1302 	pxad_free_channels(&pdev->slave);
1303 	dma_async_device_unregister(&pdev->slave);
1304 	return 0;
1305 }
1306 
1307 static int pxad_init_phys(struct platform_device *op,
1308 			  struct pxad_device *pdev,
1309 			  unsigned int nb_phy_chans)
1310 {
1311 	int irq0, irq, nr_irq = 0, i, ret;
1312 	struct pxad_phy *phy;
1313 
1314 	irq0 = platform_get_irq(op, 0);
1315 	if (irq0 < 0)
1316 		return irq0;
1317 
1318 	pdev->phys = devm_kcalloc(&op->dev, nb_phy_chans,
1319 				  sizeof(pdev->phys[0]), GFP_KERNEL);
1320 	if (!pdev->phys)
1321 		return -ENOMEM;
1322 
1323 	for (i = 0; i < nb_phy_chans; i++)
1324 		if (platform_get_irq(op, i) > 0)
1325 			nr_irq++;
1326 
1327 	for (i = 0; i < nb_phy_chans; i++) {
1328 		phy = &pdev->phys[i];
1329 		phy->base = pdev->base;
1330 		phy->idx = i;
1331 		irq = platform_get_irq(op, i);
1332 		if ((nr_irq > 1) && (irq > 0))
1333 			ret = devm_request_irq(&op->dev, irq,
1334 					       pxad_chan_handler,
1335 					       IRQF_SHARED, "pxa-dma", phy);
1336 		if ((nr_irq == 1) && (i == 0))
1337 			ret = devm_request_irq(&op->dev, irq0,
1338 					       pxad_int_handler,
1339 					       IRQF_SHARED, "pxa-dma", pdev);
1340 		if (ret) {
1341 			dev_err(pdev->slave.dev,
1342 				"%s(): can't request irq %d:%d\n", __func__,
1343 				irq, ret);
1344 			return ret;
1345 		}
1346 	}
1347 
1348 	return 0;
1349 }
1350 
1351 static const struct of_device_id pxad_dt_ids[] = {
1352 	{ .compatible = "marvell,pdma-1.0", },
1353 	{}
1354 };
1355 MODULE_DEVICE_TABLE(of, pxad_dt_ids);
1356 
1357 static struct dma_chan *pxad_dma_xlate(struct of_phandle_args *dma_spec,
1358 					   struct of_dma *ofdma)
1359 {
1360 	struct pxad_device *d = ofdma->of_dma_data;
1361 	struct dma_chan *chan;
1362 
1363 	chan = dma_get_any_slave_channel(&d->slave);
1364 	if (!chan)
1365 		return NULL;
1366 
1367 	to_pxad_chan(chan)->drcmr = dma_spec->args[0];
1368 	to_pxad_chan(chan)->prio = dma_spec->args[1];
1369 
1370 	return chan;
1371 }
1372 
1373 static int pxad_init_dmadev(struct platform_device *op,
1374 			    struct pxad_device *pdev,
1375 			    unsigned int nr_phy_chans,
1376 			    unsigned int nr_requestors)
1377 {
1378 	int ret;
1379 	unsigned int i;
1380 	struct pxad_chan *c;
1381 
1382 	pdev->nr_chans = nr_phy_chans;
1383 	pdev->nr_requestors = nr_requestors;
1384 	INIT_LIST_HEAD(&pdev->slave.channels);
1385 	pdev->slave.device_alloc_chan_resources = pxad_alloc_chan_resources;
1386 	pdev->slave.device_free_chan_resources = pxad_free_chan_resources;
1387 	pdev->slave.device_tx_status = pxad_tx_status;
1388 	pdev->slave.device_issue_pending = pxad_issue_pending;
1389 	pdev->slave.device_config = pxad_config;
1390 	pdev->slave.device_synchronize = pxad_synchronize;
1391 	pdev->slave.device_terminate_all = pxad_terminate_all;
1392 
1393 	if (op->dev.coherent_dma_mask)
1394 		dma_set_mask(&op->dev, op->dev.coherent_dma_mask);
1395 	else
1396 		dma_set_mask(&op->dev, DMA_BIT_MASK(32));
1397 
1398 	ret = pxad_init_phys(op, pdev, nr_phy_chans);
1399 	if (ret)
1400 		return ret;
1401 
1402 	for (i = 0; i < nr_phy_chans; i++) {
1403 		c = devm_kzalloc(&op->dev, sizeof(*c), GFP_KERNEL);
1404 		if (!c)
1405 			return -ENOMEM;
1406 		c->vc.desc_free = pxad_free_desc;
1407 		vchan_init(&c->vc, &pdev->slave);
1408 		init_waitqueue_head(&c->wq_state);
1409 	}
1410 
1411 	return dma_async_device_register(&pdev->slave);
1412 }
1413 
1414 static int pxad_probe(struct platform_device *op)
1415 {
1416 	struct pxad_device *pdev;
1417 	const struct of_device_id *of_id;
1418 	struct mmp_dma_platdata *pdata = dev_get_platdata(&op->dev);
1419 	struct resource *iores;
1420 	int ret, dma_channels = 0, nb_requestors = 0;
1421 	const enum dma_slave_buswidth widths =
1422 		DMA_SLAVE_BUSWIDTH_1_BYTE   | DMA_SLAVE_BUSWIDTH_2_BYTES |
1423 		DMA_SLAVE_BUSWIDTH_4_BYTES;
1424 
1425 	pdev = devm_kzalloc(&op->dev, sizeof(*pdev), GFP_KERNEL);
1426 	if (!pdev)
1427 		return -ENOMEM;
1428 
1429 	spin_lock_init(&pdev->phy_lock);
1430 
1431 	iores = platform_get_resource(op, IORESOURCE_MEM, 0);
1432 	pdev->base = devm_ioremap_resource(&op->dev, iores);
1433 	if (IS_ERR(pdev->base))
1434 		return PTR_ERR(pdev->base);
1435 
1436 	of_id = of_match_device(pxad_dt_ids, &op->dev);
1437 	if (of_id) {
1438 		of_property_read_u32(op->dev.of_node, "#dma-channels",
1439 				     &dma_channels);
1440 		ret = of_property_read_u32(op->dev.of_node, "#dma-requests",
1441 					   &nb_requestors);
1442 		if (ret) {
1443 			dev_warn(pdev->slave.dev,
1444 				 "#dma-requests set to default 32 as missing in OF: %d",
1445 				 ret);
1446 			nb_requestors = 32;
1447 		};
1448 	} else if (pdata && pdata->dma_channels) {
1449 		dma_channels = pdata->dma_channels;
1450 		nb_requestors = pdata->nb_requestors;
1451 	} else {
1452 		dma_channels = 32;	/* default 32 channel */
1453 	}
1454 
1455 	dma_cap_set(DMA_SLAVE, pdev->slave.cap_mask);
1456 	dma_cap_set(DMA_MEMCPY, pdev->slave.cap_mask);
1457 	dma_cap_set(DMA_CYCLIC, pdev->slave.cap_mask);
1458 	dma_cap_set(DMA_PRIVATE, pdev->slave.cap_mask);
1459 	pdev->slave.device_prep_dma_memcpy = pxad_prep_memcpy;
1460 	pdev->slave.device_prep_slave_sg = pxad_prep_slave_sg;
1461 	pdev->slave.device_prep_dma_cyclic = pxad_prep_dma_cyclic;
1462 
1463 	pdev->slave.copy_align = PDMA_ALIGNMENT;
1464 	pdev->slave.src_addr_widths = widths;
1465 	pdev->slave.dst_addr_widths = widths;
1466 	pdev->slave.directions = BIT(DMA_MEM_TO_DEV) | BIT(DMA_DEV_TO_MEM);
1467 	pdev->slave.residue_granularity = DMA_RESIDUE_GRANULARITY_DESCRIPTOR;
1468 	pdev->slave.descriptor_reuse = true;
1469 
1470 	pdev->slave.dev = &op->dev;
1471 	ret = pxad_init_dmadev(op, pdev, dma_channels, nb_requestors);
1472 	if (ret) {
1473 		dev_err(pdev->slave.dev, "unable to register\n");
1474 		return ret;
1475 	}
1476 
1477 	if (op->dev.of_node) {
1478 		/* Device-tree DMA controller registration */
1479 		ret = of_dma_controller_register(op->dev.of_node,
1480 						 pxad_dma_xlate, pdev);
1481 		if (ret < 0) {
1482 			dev_err(pdev->slave.dev,
1483 				"of_dma_controller_register failed\n");
1484 			return ret;
1485 		}
1486 	}
1487 
1488 	platform_set_drvdata(op, pdev);
1489 	pxad_init_debugfs(pdev);
1490 	dev_info(pdev->slave.dev, "initialized %d channels on %d requestors\n",
1491 		 dma_channels, nb_requestors);
1492 	return 0;
1493 }
1494 
1495 static const struct platform_device_id pxad_id_table[] = {
1496 	{ "pxa-dma", },
1497 	{ },
1498 };
1499 
1500 static struct platform_driver pxad_driver = {
1501 	.driver		= {
1502 		.name	= "pxa-dma",
1503 		.of_match_table = pxad_dt_ids,
1504 	},
1505 	.id_table	= pxad_id_table,
1506 	.probe		= pxad_probe,
1507 	.remove		= pxad_remove,
1508 };
1509 
1510 bool pxad_filter_fn(struct dma_chan *chan, void *param)
1511 {
1512 	struct pxad_chan *c = to_pxad_chan(chan);
1513 	struct pxad_param *p = param;
1514 
1515 	if (chan->device->dev->driver != &pxad_driver.driver)
1516 		return false;
1517 
1518 	c->drcmr = p->drcmr;
1519 	c->prio = p->prio;
1520 
1521 	return true;
1522 }
1523 EXPORT_SYMBOL_GPL(pxad_filter_fn);
1524 
1525 int pxad_toggle_reserved_channel(int legacy_channel)
1526 {
1527 	if (legacy_unavailable & (BIT(legacy_channel)))
1528 		return -EBUSY;
1529 	legacy_reserved ^= BIT(legacy_channel);
1530 	return 0;
1531 }
1532 EXPORT_SYMBOL_GPL(pxad_toggle_reserved_channel);
1533 
1534 module_platform_driver(pxad_driver);
1535 
1536 MODULE_DESCRIPTION("Marvell PXA Peripheral DMA Driver");
1537 MODULE_AUTHOR("Robert Jarzmik <robert.jarzmik@free.fr>");
1538 MODULE_LICENSE("GPL v2");
1539