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