xref: /linux/drivers/dma/at_xdmac.c (revision c5d3cdad688ed75fb311a3a671eb30ba7106d7d3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Driver for the Atmel Extensible DMA Controller (aka XDMAC on AT91 systems)
4  *
5  * Copyright (C) 2014 Atmel Corporation
6  *
7  * Author: Ludovic Desroches <ludovic.desroches@atmel.com>
8  */
9 
10 #include <asm/barrier.h>
11 #include <dt-bindings/dma/at91.h>
12 #include <linux/clk.h>
13 #include <linux/dmaengine.h>
14 #include <linux/dmapool.h>
15 #include <linux/interrupt.h>
16 #include <linux/irq.h>
17 #include <linux/kernel.h>
18 #include <linux/list.h>
19 #include <linux/module.h>
20 #include <linux/of_dma.h>
21 #include <linux/of_platform.h>
22 #include <linux/platform_device.h>
23 #include <linux/pm.h>
24 
25 #include "dmaengine.h"
26 
27 /* Global registers */
28 #define AT_XDMAC_GTYPE		0x00	/* Global Type Register */
29 #define		AT_XDMAC_NB_CH(i)	(((i) & 0x1F) + 1)		/* Number of Channels Minus One */
30 #define		AT_XDMAC_FIFO_SZ(i)	(((i) >> 5) & 0x7FF)		/* Number of Bytes */
31 #define		AT_XDMAC_NB_REQ(i)	((((i) >> 16) & 0x3F) + 1)	/* Number of Peripheral Requests Minus One */
32 #define AT_XDMAC_GCFG		0x04	/* Global Configuration Register */
33 #define AT_XDMAC_GWAC		0x08	/* Global Weighted Arbiter Configuration Register */
34 #define AT_XDMAC_GIE		0x0C	/* Global Interrupt Enable Register */
35 #define AT_XDMAC_GID		0x10	/* Global Interrupt Disable Register */
36 #define AT_XDMAC_GIM		0x14	/* Global Interrupt Mask Register */
37 #define AT_XDMAC_GIS		0x18	/* Global Interrupt Status Register */
38 #define AT_XDMAC_GE		0x1C	/* Global Channel Enable Register */
39 #define AT_XDMAC_GD		0x20	/* Global Channel Disable Register */
40 #define AT_XDMAC_GS		0x24	/* Global Channel Status Register */
41 #define AT_XDMAC_GRS		0x28	/* Global Channel Read Suspend Register */
42 #define AT_XDMAC_GWS		0x2C	/* Global Write Suspend Register */
43 #define AT_XDMAC_GRWS		0x30	/* Global Channel Read Write Suspend Register */
44 #define AT_XDMAC_GRWR		0x34	/* Global Channel Read Write Resume Register */
45 #define AT_XDMAC_GSWR		0x38	/* Global Channel Software Request Register */
46 #define AT_XDMAC_GSWS		0x3C	/* Global channel Software Request Status Register */
47 #define AT_XDMAC_GSWF		0x40	/* Global Channel Software Flush Request Register */
48 #define AT_XDMAC_VERSION	0xFFC	/* XDMAC Version Register */
49 
50 /* Channel relative registers offsets */
51 #define AT_XDMAC_CIE		0x00	/* Channel Interrupt Enable Register */
52 #define		AT_XDMAC_CIE_BIE	BIT(0)	/* End of Block Interrupt Enable Bit */
53 #define		AT_XDMAC_CIE_LIE	BIT(1)	/* End of Linked List Interrupt Enable Bit */
54 #define		AT_XDMAC_CIE_DIE	BIT(2)	/* End of Disable Interrupt Enable Bit */
55 #define		AT_XDMAC_CIE_FIE	BIT(3)	/* End of Flush Interrupt Enable Bit */
56 #define		AT_XDMAC_CIE_RBEIE	BIT(4)	/* Read Bus Error Interrupt Enable Bit */
57 #define		AT_XDMAC_CIE_WBEIE	BIT(5)	/* Write Bus Error Interrupt Enable Bit */
58 #define		AT_XDMAC_CIE_ROIE	BIT(6)	/* Request Overflow Interrupt Enable Bit */
59 #define AT_XDMAC_CID		0x04	/* Channel Interrupt Disable Register */
60 #define		AT_XDMAC_CID_BID	BIT(0)	/* End of Block Interrupt Disable Bit */
61 #define		AT_XDMAC_CID_LID	BIT(1)	/* End of Linked List Interrupt Disable Bit */
62 #define		AT_XDMAC_CID_DID	BIT(2)	/* End of Disable Interrupt Disable Bit */
63 #define		AT_XDMAC_CID_FID	BIT(3)	/* End of Flush Interrupt Disable Bit */
64 #define		AT_XDMAC_CID_RBEID	BIT(4)	/* Read Bus Error Interrupt Disable Bit */
65 #define		AT_XDMAC_CID_WBEID	BIT(5)	/* Write Bus Error Interrupt Disable Bit */
66 #define		AT_XDMAC_CID_ROID	BIT(6)	/* Request Overflow Interrupt Disable Bit */
67 #define AT_XDMAC_CIM		0x08	/* Channel Interrupt Mask Register */
68 #define		AT_XDMAC_CIM_BIM	BIT(0)	/* End of Block Interrupt Mask Bit */
69 #define		AT_XDMAC_CIM_LIM	BIT(1)	/* End of Linked List Interrupt Mask Bit */
70 #define		AT_XDMAC_CIM_DIM	BIT(2)	/* End of Disable Interrupt Mask Bit */
71 #define		AT_XDMAC_CIM_FIM	BIT(3)	/* End of Flush Interrupt Mask Bit */
72 #define		AT_XDMAC_CIM_RBEIM	BIT(4)	/* Read Bus Error Interrupt Mask Bit */
73 #define		AT_XDMAC_CIM_WBEIM	BIT(5)	/* Write Bus Error Interrupt Mask Bit */
74 #define		AT_XDMAC_CIM_ROIM	BIT(6)	/* Request Overflow Interrupt Mask Bit */
75 #define AT_XDMAC_CIS		0x0C	/* Channel Interrupt Status Register */
76 #define		AT_XDMAC_CIS_BIS	BIT(0)	/* End of Block Interrupt Status Bit */
77 #define		AT_XDMAC_CIS_LIS	BIT(1)	/* End of Linked List Interrupt Status Bit */
78 #define		AT_XDMAC_CIS_DIS	BIT(2)	/* End of Disable Interrupt Status Bit */
79 #define		AT_XDMAC_CIS_FIS	BIT(3)	/* End of Flush Interrupt Status Bit */
80 #define		AT_XDMAC_CIS_RBEIS	BIT(4)	/* Read Bus Error Interrupt Status Bit */
81 #define		AT_XDMAC_CIS_WBEIS	BIT(5)	/* Write Bus Error Interrupt Status Bit */
82 #define		AT_XDMAC_CIS_ROIS	BIT(6)	/* Request Overflow Interrupt Status Bit */
83 #define AT_XDMAC_CSA		0x10	/* Channel Source Address Register */
84 #define AT_XDMAC_CDA		0x14	/* Channel Destination Address Register */
85 #define AT_XDMAC_CNDA		0x18	/* Channel Next Descriptor Address Register */
86 #define		AT_XDMAC_CNDA_NDAIF(i)	((i) & 0x1)			/* Channel x Next Descriptor Interface */
87 #define		AT_XDMAC_CNDA_NDA(i)	((i) & 0xfffffffc)		/* Channel x Next Descriptor Address */
88 #define AT_XDMAC_CNDC		0x1C	/* Channel Next Descriptor Control Register */
89 #define		AT_XDMAC_CNDC_NDE		(0x1 << 0)		/* Channel x Next Descriptor Enable */
90 #define		AT_XDMAC_CNDC_NDSUP		(0x1 << 1)		/* Channel x Next Descriptor Source Update */
91 #define		AT_XDMAC_CNDC_NDDUP		(0x1 << 2)		/* Channel x Next Descriptor Destination Update */
92 #define		AT_XDMAC_CNDC_NDVIEW_NDV0	(0x0 << 3)		/* Channel x Next Descriptor View 0 */
93 #define		AT_XDMAC_CNDC_NDVIEW_NDV1	(0x1 << 3)		/* Channel x Next Descriptor View 1 */
94 #define		AT_XDMAC_CNDC_NDVIEW_NDV2	(0x2 << 3)		/* Channel x Next Descriptor View 2 */
95 #define		AT_XDMAC_CNDC_NDVIEW_NDV3	(0x3 << 3)		/* Channel x Next Descriptor View 3 */
96 #define AT_XDMAC_CUBC		0x20	/* Channel Microblock Control Register */
97 #define AT_XDMAC_CBC		0x24	/* Channel Block Control Register */
98 #define AT_XDMAC_CC		0x28	/* Channel Configuration Register */
99 #define		AT_XDMAC_CC_TYPE	(0x1 << 0)	/* Channel Transfer Type */
100 #define			AT_XDMAC_CC_TYPE_MEM_TRAN	(0x0 << 0)	/* Memory to Memory Transfer */
101 #define			AT_XDMAC_CC_TYPE_PER_TRAN	(0x1 << 0)	/* Peripheral to Memory or Memory to Peripheral Transfer */
102 #define		AT_XDMAC_CC_MBSIZE_MASK	(0x3 << 1)
103 #define			AT_XDMAC_CC_MBSIZE_SINGLE	(0x0 << 1)
104 #define			AT_XDMAC_CC_MBSIZE_FOUR		(0x1 << 1)
105 #define			AT_XDMAC_CC_MBSIZE_EIGHT	(0x2 << 1)
106 #define			AT_XDMAC_CC_MBSIZE_SIXTEEN	(0x3 << 1)
107 #define		AT_XDMAC_CC_DSYNC	(0x1 << 4)	/* Channel Synchronization */
108 #define			AT_XDMAC_CC_DSYNC_PER2MEM	(0x0 << 4)
109 #define			AT_XDMAC_CC_DSYNC_MEM2PER	(0x1 << 4)
110 #define		AT_XDMAC_CC_PROT	(0x1 << 5)	/* Channel Protection */
111 #define			AT_XDMAC_CC_PROT_SEC		(0x0 << 5)
112 #define			AT_XDMAC_CC_PROT_UNSEC		(0x1 << 5)
113 #define		AT_XDMAC_CC_SWREQ	(0x1 << 6)	/* Channel Software Request Trigger */
114 #define			AT_XDMAC_CC_SWREQ_HWR_CONNECTED	(0x0 << 6)
115 #define			AT_XDMAC_CC_SWREQ_SWR_CONNECTED	(0x1 << 6)
116 #define		AT_XDMAC_CC_MEMSET	(0x1 << 7)	/* Channel Fill Block of memory */
117 #define			AT_XDMAC_CC_MEMSET_NORMAL_MODE	(0x0 << 7)
118 #define			AT_XDMAC_CC_MEMSET_HW_MODE	(0x1 << 7)
119 #define		AT_XDMAC_CC_CSIZE(i)	((0x7 & (i)) << 8)	/* Channel Chunk Size */
120 #define		AT_XDMAC_CC_DWIDTH_OFFSET	11
121 #define		AT_XDMAC_CC_DWIDTH_MASK	(0x3 << AT_XDMAC_CC_DWIDTH_OFFSET)
122 #define		AT_XDMAC_CC_DWIDTH(i)	((0x3 & (i)) << AT_XDMAC_CC_DWIDTH_OFFSET)	/* Channel Data Width */
123 #define			AT_XDMAC_CC_DWIDTH_BYTE		0x0
124 #define			AT_XDMAC_CC_DWIDTH_HALFWORD	0x1
125 #define			AT_XDMAC_CC_DWIDTH_WORD		0x2
126 #define			AT_XDMAC_CC_DWIDTH_DWORD	0x3
127 #define		AT_XDMAC_CC_SIF(i)	((0x1 & (i)) << 13)	/* Channel Source Interface Identifier */
128 #define		AT_XDMAC_CC_DIF(i)	((0x1 & (i)) << 14)	/* Channel Destination Interface Identifier */
129 #define		AT_XDMAC_CC_SAM_MASK	(0x3 << 16)	/* Channel Source Addressing Mode */
130 #define			AT_XDMAC_CC_SAM_FIXED_AM	(0x0 << 16)
131 #define			AT_XDMAC_CC_SAM_INCREMENTED_AM	(0x1 << 16)
132 #define			AT_XDMAC_CC_SAM_UBS_AM		(0x2 << 16)
133 #define			AT_XDMAC_CC_SAM_UBS_DS_AM	(0x3 << 16)
134 #define		AT_XDMAC_CC_DAM_MASK	(0x3 << 18)	/* Channel Source Addressing Mode */
135 #define			AT_XDMAC_CC_DAM_FIXED_AM	(0x0 << 18)
136 #define			AT_XDMAC_CC_DAM_INCREMENTED_AM	(0x1 << 18)
137 #define			AT_XDMAC_CC_DAM_UBS_AM		(0x2 << 18)
138 #define			AT_XDMAC_CC_DAM_UBS_DS_AM	(0x3 << 18)
139 #define		AT_XDMAC_CC_INITD	(0x1 << 21)	/* Channel Initialization Terminated (read only) */
140 #define			AT_XDMAC_CC_INITD_TERMINATED	(0x0 << 21)
141 #define			AT_XDMAC_CC_INITD_IN_PROGRESS	(0x1 << 21)
142 #define		AT_XDMAC_CC_RDIP	(0x1 << 22)	/* Read in Progress (read only) */
143 #define			AT_XDMAC_CC_RDIP_DONE		(0x0 << 22)
144 #define			AT_XDMAC_CC_RDIP_IN_PROGRESS	(0x1 << 22)
145 #define		AT_XDMAC_CC_WRIP	(0x1 << 23)	/* Write in Progress (read only) */
146 #define			AT_XDMAC_CC_WRIP_DONE		(0x0 << 23)
147 #define			AT_XDMAC_CC_WRIP_IN_PROGRESS	(0x1 << 23)
148 #define		AT_XDMAC_CC_PERID(i)	(0x7f & (i) << 24)	/* Channel Peripheral Identifier */
149 #define AT_XDMAC_CDS_MSP	0x2C	/* Channel Data Stride Memory Set Pattern */
150 #define AT_XDMAC_CSUS		0x30	/* Channel Source Microblock Stride */
151 #define AT_XDMAC_CDUS		0x34	/* Channel Destination Microblock Stride */
152 
153 #define AT_XDMAC_CHAN_REG_BASE	0x50	/* Channel registers base address */
154 
155 /* Microblock control members */
156 #define AT_XDMAC_MBR_UBC_UBLEN_MAX	0xFFFFFFUL	/* Maximum Microblock Length */
157 #define AT_XDMAC_MBR_UBC_NDE		(0x1 << 24)	/* Next Descriptor Enable */
158 #define AT_XDMAC_MBR_UBC_NSEN		(0x1 << 25)	/* Next Descriptor Source Update */
159 #define AT_XDMAC_MBR_UBC_NDEN		(0x1 << 26)	/* Next Descriptor Destination Update */
160 #define AT_XDMAC_MBR_UBC_NDV0		(0x0 << 27)	/* Next Descriptor View 0 */
161 #define AT_XDMAC_MBR_UBC_NDV1		(0x1 << 27)	/* Next Descriptor View 1 */
162 #define AT_XDMAC_MBR_UBC_NDV2		(0x2 << 27)	/* Next Descriptor View 2 */
163 #define AT_XDMAC_MBR_UBC_NDV3		(0x3 << 27)	/* Next Descriptor View 3 */
164 
165 #define AT_XDMAC_MAX_CHAN	0x20
166 #define AT_XDMAC_MAX_CSIZE	16	/* 16 data */
167 #define AT_XDMAC_MAX_DWIDTH	8	/* 64 bits */
168 #define AT_XDMAC_RESIDUE_MAX_RETRIES	5
169 
170 #define AT_XDMAC_DMA_BUSWIDTHS\
171 	(BIT(DMA_SLAVE_BUSWIDTH_UNDEFINED) |\
172 	BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |\
173 	BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |\
174 	BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |\
175 	BIT(DMA_SLAVE_BUSWIDTH_8_BYTES))
176 
177 enum atc_status {
178 	AT_XDMAC_CHAN_IS_CYCLIC = 0,
179 	AT_XDMAC_CHAN_IS_PAUSED,
180 };
181 
182 /* ----- Channels ----- */
183 struct at_xdmac_chan {
184 	struct dma_chan			chan;
185 	void __iomem			*ch_regs;
186 	u32				mask;		/* Channel Mask */
187 	u32				cfg;		/* Channel Configuration Register */
188 	u8				perid;		/* Peripheral ID */
189 	u8				perif;		/* Peripheral Interface */
190 	u8				memif;		/* Memory Interface */
191 	u32				save_cc;
192 	u32				save_cim;
193 	u32				save_cnda;
194 	u32				save_cndc;
195 	u32				irq_status;
196 	unsigned long			status;
197 	struct tasklet_struct		tasklet;
198 	struct dma_slave_config		sconfig;
199 
200 	spinlock_t			lock;
201 
202 	struct list_head		xfers_list;
203 	struct list_head		free_descs_list;
204 };
205 
206 
207 /* ----- Controller ----- */
208 struct at_xdmac {
209 	struct dma_device	dma;
210 	void __iomem		*regs;
211 	int			irq;
212 	struct clk		*clk;
213 	u32			save_gim;
214 	struct dma_pool		*at_xdmac_desc_pool;
215 	struct at_xdmac_chan	chan[0];
216 };
217 
218 
219 /* ----- Descriptors ----- */
220 
221 /* Linked List Descriptor */
222 struct at_xdmac_lld {
223 	dma_addr_t	mbr_nda;	/* Next Descriptor Member */
224 	u32		mbr_ubc;	/* Microblock Control Member */
225 	dma_addr_t	mbr_sa;		/* Source Address Member */
226 	dma_addr_t	mbr_da;		/* Destination Address Member */
227 	u32		mbr_cfg;	/* Configuration Register */
228 	u32		mbr_bc;		/* Block Control Register */
229 	u32		mbr_ds;		/* Data Stride Register */
230 	u32		mbr_sus;	/* Source Microblock Stride Register */
231 	u32		mbr_dus;	/* Destination Microblock Stride Register */
232 };
233 
234 /* 64-bit alignment needed to update CNDA and CUBC registers in an atomic way. */
235 struct at_xdmac_desc {
236 	struct at_xdmac_lld		lld;
237 	enum dma_transfer_direction	direction;
238 	struct dma_async_tx_descriptor	tx_dma_desc;
239 	struct list_head		desc_node;
240 	/* Following members are only used by the first descriptor */
241 	bool				active_xfer;
242 	unsigned int			xfer_size;
243 	struct list_head		descs_list;
244 	struct list_head		xfer_node;
245 } __aligned(sizeof(u64));
246 
247 static inline void __iomem *at_xdmac_chan_reg_base(struct at_xdmac *atxdmac, unsigned int chan_nb)
248 {
249 	return atxdmac->regs + (AT_XDMAC_CHAN_REG_BASE + chan_nb * 0x40);
250 }
251 
252 #define at_xdmac_read(atxdmac, reg) readl_relaxed((atxdmac)->regs + (reg))
253 #define at_xdmac_write(atxdmac, reg, value) \
254 	writel_relaxed((value), (atxdmac)->regs + (reg))
255 
256 #define at_xdmac_chan_read(atchan, reg) readl_relaxed((atchan)->ch_regs + (reg))
257 #define at_xdmac_chan_write(atchan, reg, value) writel_relaxed((value), (atchan)->ch_regs + (reg))
258 
259 static inline struct at_xdmac_chan *to_at_xdmac_chan(struct dma_chan *dchan)
260 {
261 	return container_of(dchan, struct at_xdmac_chan, chan);
262 }
263 
264 static struct device *chan2dev(struct dma_chan *chan)
265 {
266 	return &chan->dev->device;
267 }
268 
269 static inline struct at_xdmac *to_at_xdmac(struct dma_device *ddev)
270 {
271 	return container_of(ddev, struct at_xdmac, dma);
272 }
273 
274 static inline struct at_xdmac_desc *txd_to_at_desc(struct dma_async_tx_descriptor *txd)
275 {
276 	return container_of(txd, struct at_xdmac_desc, tx_dma_desc);
277 }
278 
279 static inline int at_xdmac_chan_is_cyclic(struct at_xdmac_chan *atchan)
280 {
281 	return test_bit(AT_XDMAC_CHAN_IS_CYCLIC, &atchan->status);
282 }
283 
284 static inline int at_xdmac_chan_is_paused(struct at_xdmac_chan *atchan)
285 {
286 	return test_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status);
287 }
288 
289 static inline int at_xdmac_csize(u32 maxburst)
290 {
291 	int csize;
292 
293 	csize = ffs(maxburst) - 1;
294 	if (csize > 4)
295 		csize = -EINVAL;
296 
297 	return csize;
298 };
299 
300 static inline bool at_xdmac_chan_is_peripheral_xfer(u32 cfg)
301 {
302 	return cfg & AT_XDMAC_CC_TYPE_PER_TRAN;
303 }
304 
305 static inline u8 at_xdmac_get_dwidth(u32 cfg)
306 {
307 	return (cfg & AT_XDMAC_CC_DWIDTH_MASK) >> AT_XDMAC_CC_DWIDTH_OFFSET;
308 };
309 
310 static unsigned int init_nr_desc_per_channel = 64;
311 module_param(init_nr_desc_per_channel, uint, 0644);
312 MODULE_PARM_DESC(init_nr_desc_per_channel,
313 		 "initial descriptors per channel (default: 64)");
314 
315 
316 static bool at_xdmac_chan_is_enabled(struct at_xdmac_chan *atchan)
317 {
318 	return at_xdmac_chan_read(atchan, AT_XDMAC_GS) & atchan->mask;
319 }
320 
321 static void at_xdmac_off(struct at_xdmac *atxdmac)
322 {
323 	at_xdmac_write(atxdmac, AT_XDMAC_GD, -1L);
324 
325 	/* Wait that all chans are disabled. */
326 	while (at_xdmac_read(atxdmac, AT_XDMAC_GS))
327 		cpu_relax();
328 
329 	at_xdmac_write(atxdmac, AT_XDMAC_GID, -1L);
330 }
331 
332 /* Call with lock hold. */
333 static void at_xdmac_start_xfer(struct at_xdmac_chan *atchan,
334 				struct at_xdmac_desc *first)
335 {
336 	struct at_xdmac	*atxdmac = to_at_xdmac(atchan->chan.device);
337 	u32		reg;
338 
339 	dev_vdbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, first);
340 
341 	if (at_xdmac_chan_is_enabled(atchan))
342 		return;
343 
344 	/* Set transfer as active to not try to start it again. */
345 	first->active_xfer = true;
346 
347 	/* Tell xdmac where to get the first descriptor. */
348 	reg = AT_XDMAC_CNDA_NDA(first->tx_dma_desc.phys)
349 	      | AT_XDMAC_CNDA_NDAIF(atchan->memif);
350 	at_xdmac_chan_write(atchan, AT_XDMAC_CNDA, reg);
351 
352 	/*
353 	 * When doing non cyclic transfer we need to use the next
354 	 * descriptor view 2 since some fields of the configuration register
355 	 * depend on transfer size and src/dest addresses.
356 	 */
357 	if (at_xdmac_chan_is_cyclic(atchan))
358 		reg = AT_XDMAC_CNDC_NDVIEW_NDV1;
359 	else if (first->lld.mbr_ubc & AT_XDMAC_MBR_UBC_NDV3)
360 		reg = AT_XDMAC_CNDC_NDVIEW_NDV3;
361 	else
362 		reg = AT_XDMAC_CNDC_NDVIEW_NDV2;
363 	/*
364 	 * Even if the register will be updated from the configuration in the
365 	 * descriptor when using view 2 or higher, the PROT bit won't be set
366 	 * properly. This bit can be modified only by using the channel
367 	 * configuration register.
368 	 */
369 	at_xdmac_chan_write(atchan, AT_XDMAC_CC, first->lld.mbr_cfg);
370 
371 	reg |= AT_XDMAC_CNDC_NDDUP
372 	       | AT_XDMAC_CNDC_NDSUP
373 	       | AT_XDMAC_CNDC_NDE;
374 	at_xdmac_chan_write(atchan, AT_XDMAC_CNDC, reg);
375 
376 	dev_vdbg(chan2dev(&atchan->chan),
377 		 "%s: CC=0x%08x CNDA=0x%08x, CNDC=0x%08x, CSA=0x%08x, CDA=0x%08x, CUBC=0x%08x\n",
378 		 __func__, at_xdmac_chan_read(atchan, AT_XDMAC_CC),
379 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDA),
380 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDC),
381 		 at_xdmac_chan_read(atchan, AT_XDMAC_CSA),
382 		 at_xdmac_chan_read(atchan, AT_XDMAC_CDA),
383 		 at_xdmac_chan_read(atchan, AT_XDMAC_CUBC));
384 
385 	at_xdmac_chan_write(atchan, AT_XDMAC_CID, 0xffffffff);
386 	reg = AT_XDMAC_CIE_RBEIE | AT_XDMAC_CIE_WBEIE;
387 	/*
388 	 * Request Overflow Error is only for peripheral synchronized transfers
389 	 */
390 	if (at_xdmac_chan_is_peripheral_xfer(first->lld.mbr_cfg))
391 		reg |= AT_XDMAC_CIE_ROIE;
392 
393 	/*
394 	 * There is no end of list when doing cyclic dma, we need to get
395 	 * an interrupt after each periods.
396 	 */
397 	if (at_xdmac_chan_is_cyclic(atchan))
398 		at_xdmac_chan_write(atchan, AT_XDMAC_CIE,
399 				    reg | AT_XDMAC_CIE_BIE);
400 	else
401 		at_xdmac_chan_write(atchan, AT_XDMAC_CIE,
402 				    reg | AT_XDMAC_CIE_LIE);
403 	at_xdmac_write(atxdmac, AT_XDMAC_GIE, atchan->mask);
404 	dev_vdbg(chan2dev(&atchan->chan),
405 		 "%s: enable channel (0x%08x)\n", __func__, atchan->mask);
406 	wmb();
407 	at_xdmac_write(atxdmac, AT_XDMAC_GE, atchan->mask);
408 
409 	dev_vdbg(chan2dev(&atchan->chan),
410 		 "%s: CC=0x%08x CNDA=0x%08x, CNDC=0x%08x, CSA=0x%08x, CDA=0x%08x, CUBC=0x%08x\n",
411 		 __func__, at_xdmac_chan_read(atchan, AT_XDMAC_CC),
412 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDA),
413 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDC),
414 		 at_xdmac_chan_read(atchan, AT_XDMAC_CSA),
415 		 at_xdmac_chan_read(atchan, AT_XDMAC_CDA),
416 		 at_xdmac_chan_read(atchan, AT_XDMAC_CUBC));
417 
418 }
419 
420 static dma_cookie_t at_xdmac_tx_submit(struct dma_async_tx_descriptor *tx)
421 {
422 	struct at_xdmac_desc	*desc = txd_to_at_desc(tx);
423 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(tx->chan);
424 	dma_cookie_t		cookie;
425 	unsigned long		irqflags;
426 
427 	spin_lock_irqsave(&atchan->lock, irqflags);
428 	cookie = dma_cookie_assign(tx);
429 
430 	dev_vdbg(chan2dev(tx->chan), "%s: atchan 0x%p, add desc 0x%p to xfers_list\n",
431 		 __func__, atchan, desc);
432 	list_add_tail(&desc->xfer_node, &atchan->xfers_list);
433 	if (list_is_singular(&atchan->xfers_list))
434 		at_xdmac_start_xfer(atchan, desc);
435 
436 	spin_unlock_irqrestore(&atchan->lock, irqflags);
437 	return cookie;
438 }
439 
440 static struct at_xdmac_desc *at_xdmac_alloc_desc(struct dma_chan *chan,
441 						 gfp_t gfp_flags)
442 {
443 	struct at_xdmac_desc	*desc;
444 	struct at_xdmac		*atxdmac = to_at_xdmac(chan->device);
445 	dma_addr_t		phys;
446 
447 	desc = dma_pool_zalloc(atxdmac->at_xdmac_desc_pool, gfp_flags, &phys);
448 	if (desc) {
449 		INIT_LIST_HEAD(&desc->descs_list);
450 		dma_async_tx_descriptor_init(&desc->tx_dma_desc, chan);
451 		desc->tx_dma_desc.tx_submit = at_xdmac_tx_submit;
452 		desc->tx_dma_desc.phys = phys;
453 	}
454 
455 	return desc;
456 }
457 
458 static void at_xdmac_init_used_desc(struct at_xdmac_desc *desc)
459 {
460 	memset(&desc->lld, 0, sizeof(desc->lld));
461 	INIT_LIST_HEAD(&desc->descs_list);
462 	desc->direction = DMA_TRANS_NONE;
463 	desc->xfer_size = 0;
464 	desc->active_xfer = false;
465 }
466 
467 /* Call must be protected by lock. */
468 static struct at_xdmac_desc *at_xdmac_get_desc(struct at_xdmac_chan *atchan)
469 {
470 	struct at_xdmac_desc *desc;
471 
472 	if (list_empty(&atchan->free_descs_list)) {
473 		desc = at_xdmac_alloc_desc(&atchan->chan, GFP_NOWAIT);
474 	} else {
475 		desc = list_first_entry(&atchan->free_descs_list,
476 					struct at_xdmac_desc, desc_node);
477 		list_del(&desc->desc_node);
478 		at_xdmac_init_used_desc(desc);
479 	}
480 
481 	return desc;
482 }
483 
484 static void at_xdmac_queue_desc(struct dma_chan *chan,
485 				struct at_xdmac_desc *prev,
486 				struct at_xdmac_desc *desc)
487 {
488 	if (!prev || !desc)
489 		return;
490 
491 	prev->lld.mbr_nda = desc->tx_dma_desc.phys;
492 	prev->lld.mbr_ubc |= AT_XDMAC_MBR_UBC_NDE;
493 
494 	dev_dbg(chan2dev(chan),	"%s: chain lld: prev=0x%p, mbr_nda=%pad\n",
495 		__func__, prev, &prev->lld.mbr_nda);
496 }
497 
498 static inline void at_xdmac_increment_block_count(struct dma_chan *chan,
499 						  struct at_xdmac_desc *desc)
500 {
501 	if (!desc)
502 		return;
503 
504 	desc->lld.mbr_bc++;
505 
506 	dev_dbg(chan2dev(chan),
507 		"%s: incrementing the block count of the desc 0x%p\n",
508 		__func__, desc);
509 }
510 
511 static struct dma_chan *at_xdmac_xlate(struct of_phandle_args *dma_spec,
512 				       struct of_dma *of_dma)
513 {
514 	struct at_xdmac		*atxdmac = of_dma->of_dma_data;
515 	struct at_xdmac_chan	*atchan;
516 	struct dma_chan		*chan;
517 	struct device		*dev = atxdmac->dma.dev;
518 
519 	if (dma_spec->args_count != 1) {
520 		dev_err(dev, "dma phandler args: bad number of args\n");
521 		return NULL;
522 	}
523 
524 	chan = dma_get_any_slave_channel(&atxdmac->dma);
525 	if (!chan) {
526 		dev_err(dev, "can't get a dma channel\n");
527 		return NULL;
528 	}
529 
530 	atchan = to_at_xdmac_chan(chan);
531 	atchan->memif = AT91_XDMAC_DT_GET_MEM_IF(dma_spec->args[0]);
532 	atchan->perif = AT91_XDMAC_DT_GET_PER_IF(dma_spec->args[0]);
533 	atchan->perid = AT91_XDMAC_DT_GET_PERID(dma_spec->args[0]);
534 	dev_dbg(dev, "chan dt cfg: memif=%u perif=%u perid=%u\n",
535 		 atchan->memif, atchan->perif, atchan->perid);
536 
537 	return chan;
538 }
539 
540 static int at_xdmac_compute_chan_conf(struct dma_chan *chan,
541 				      enum dma_transfer_direction direction)
542 {
543 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
544 	int			csize, dwidth;
545 
546 	if (direction == DMA_DEV_TO_MEM) {
547 		atchan->cfg =
548 			AT91_XDMAC_DT_PERID(atchan->perid)
549 			| AT_XDMAC_CC_DAM_INCREMENTED_AM
550 			| AT_XDMAC_CC_SAM_FIXED_AM
551 			| AT_XDMAC_CC_DIF(atchan->memif)
552 			| AT_XDMAC_CC_SIF(atchan->perif)
553 			| AT_XDMAC_CC_SWREQ_HWR_CONNECTED
554 			| AT_XDMAC_CC_DSYNC_PER2MEM
555 			| AT_XDMAC_CC_MBSIZE_SIXTEEN
556 			| AT_XDMAC_CC_TYPE_PER_TRAN;
557 		csize = ffs(atchan->sconfig.src_maxburst) - 1;
558 		if (csize < 0) {
559 			dev_err(chan2dev(chan), "invalid src maxburst value\n");
560 			return -EINVAL;
561 		}
562 		atchan->cfg |= AT_XDMAC_CC_CSIZE(csize);
563 		dwidth = ffs(atchan->sconfig.src_addr_width) - 1;
564 		if (dwidth < 0) {
565 			dev_err(chan2dev(chan), "invalid src addr width value\n");
566 			return -EINVAL;
567 		}
568 		atchan->cfg |= AT_XDMAC_CC_DWIDTH(dwidth);
569 	} else if (direction == DMA_MEM_TO_DEV) {
570 		atchan->cfg =
571 			AT91_XDMAC_DT_PERID(atchan->perid)
572 			| AT_XDMAC_CC_DAM_FIXED_AM
573 			| AT_XDMAC_CC_SAM_INCREMENTED_AM
574 			| AT_XDMAC_CC_DIF(atchan->perif)
575 			| AT_XDMAC_CC_SIF(atchan->memif)
576 			| AT_XDMAC_CC_SWREQ_HWR_CONNECTED
577 			| AT_XDMAC_CC_DSYNC_MEM2PER
578 			| AT_XDMAC_CC_MBSIZE_SIXTEEN
579 			| AT_XDMAC_CC_TYPE_PER_TRAN;
580 		csize = ffs(atchan->sconfig.dst_maxburst) - 1;
581 		if (csize < 0) {
582 			dev_err(chan2dev(chan), "invalid src maxburst value\n");
583 			return -EINVAL;
584 		}
585 		atchan->cfg |= AT_XDMAC_CC_CSIZE(csize);
586 		dwidth = ffs(atchan->sconfig.dst_addr_width) - 1;
587 		if (dwidth < 0) {
588 			dev_err(chan2dev(chan), "invalid dst addr width value\n");
589 			return -EINVAL;
590 		}
591 		atchan->cfg |= AT_XDMAC_CC_DWIDTH(dwidth);
592 	}
593 
594 	dev_dbg(chan2dev(chan),	"%s: cfg=0x%08x\n", __func__, atchan->cfg);
595 
596 	return 0;
597 }
598 
599 /*
600  * Only check that maxburst and addr width values are supported by the
601  * the controller but not that the configuration is good to perform the
602  * transfer since we don't know the direction at this stage.
603  */
604 static int at_xdmac_check_slave_config(struct dma_slave_config *sconfig)
605 {
606 	if ((sconfig->src_maxburst > AT_XDMAC_MAX_CSIZE)
607 	    || (sconfig->dst_maxburst > AT_XDMAC_MAX_CSIZE))
608 		return -EINVAL;
609 
610 	if ((sconfig->src_addr_width > AT_XDMAC_MAX_DWIDTH)
611 	    || (sconfig->dst_addr_width > AT_XDMAC_MAX_DWIDTH))
612 		return -EINVAL;
613 
614 	return 0;
615 }
616 
617 static int at_xdmac_set_slave_config(struct dma_chan *chan,
618 				      struct dma_slave_config *sconfig)
619 {
620 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
621 
622 	if (at_xdmac_check_slave_config(sconfig)) {
623 		dev_err(chan2dev(chan), "invalid slave configuration\n");
624 		return -EINVAL;
625 	}
626 
627 	memcpy(&atchan->sconfig, sconfig, sizeof(atchan->sconfig));
628 
629 	return 0;
630 }
631 
632 static struct dma_async_tx_descriptor *
633 at_xdmac_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
634 		       unsigned int sg_len, enum dma_transfer_direction direction,
635 		       unsigned long flags, void *context)
636 {
637 	struct at_xdmac_chan		*atchan = to_at_xdmac_chan(chan);
638 	struct at_xdmac_desc		*first = NULL, *prev = NULL;
639 	struct scatterlist		*sg;
640 	int				i;
641 	unsigned int			xfer_size = 0;
642 	unsigned long			irqflags;
643 	struct dma_async_tx_descriptor	*ret = NULL;
644 
645 	if (!sgl)
646 		return NULL;
647 
648 	if (!is_slave_direction(direction)) {
649 		dev_err(chan2dev(chan), "invalid DMA direction\n");
650 		return NULL;
651 	}
652 
653 	dev_dbg(chan2dev(chan), "%s: sg_len=%d, dir=%s, flags=0x%lx\n",
654 		 __func__, sg_len,
655 		 direction == DMA_MEM_TO_DEV ? "to device" : "from device",
656 		 flags);
657 
658 	/* Protect dma_sconfig field that can be modified by set_slave_conf. */
659 	spin_lock_irqsave(&atchan->lock, irqflags);
660 
661 	if (at_xdmac_compute_chan_conf(chan, direction))
662 		goto spin_unlock;
663 
664 	/* Prepare descriptors. */
665 	for_each_sg(sgl, sg, sg_len, i) {
666 		struct at_xdmac_desc	*desc = NULL;
667 		u32			len, mem, dwidth, fixed_dwidth;
668 
669 		len = sg_dma_len(sg);
670 		mem = sg_dma_address(sg);
671 		if (unlikely(!len)) {
672 			dev_err(chan2dev(chan), "sg data length is zero\n");
673 			goto spin_unlock;
674 		}
675 		dev_dbg(chan2dev(chan), "%s: * sg%d len=%u, mem=0x%08x\n",
676 			 __func__, i, len, mem);
677 
678 		desc = at_xdmac_get_desc(atchan);
679 		if (!desc) {
680 			dev_err(chan2dev(chan), "can't get descriptor\n");
681 			if (first)
682 				list_splice_init(&first->descs_list, &atchan->free_descs_list);
683 			goto spin_unlock;
684 		}
685 
686 		/* Linked list descriptor setup. */
687 		if (direction == DMA_DEV_TO_MEM) {
688 			desc->lld.mbr_sa = atchan->sconfig.src_addr;
689 			desc->lld.mbr_da = mem;
690 		} else {
691 			desc->lld.mbr_sa = mem;
692 			desc->lld.mbr_da = atchan->sconfig.dst_addr;
693 		}
694 		dwidth = at_xdmac_get_dwidth(atchan->cfg);
695 		fixed_dwidth = IS_ALIGNED(len, 1 << dwidth)
696 			       ? dwidth
697 			       : AT_XDMAC_CC_DWIDTH_BYTE;
698 		desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV2			/* next descriptor view */
699 			| AT_XDMAC_MBR_UBC_NDEN					/* next descriptor dst parameter update */
700 			| AT_XDMAC_MBR_UBC_NSEN					/* next descriptor src parameter update */
701 			| (len >> fixed_dwidth);				/* microblock length */
702 		desc->lld.mbr_cfg = (atchan->cfg & ~AT_XDMAC_CC_DWIDTH_MASK) |
703 				    AT_XDMAC_CC_DWIDTH(fixed_dwidth);
704 		dev_dbg(chan2dev(chan),
705 			 "%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x\n",
706 			 __func__, &desc->lld.mbr_sa, &desc->lld.mbr_da, desc->lld.mbr_ubc);
707 
708 		/* Chain lld. */
709 		if (prev)
710 			at_xdmac_queue_desc(chan, prev, desc);
711 
712 		prev = desc;
713 		if (!first)
714 			first = desc;
715 
716 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
717 			 __func__, desc, first);
718 		list_add_tail(&desc->desc_node, &first->descs_list);
719 		xfer_size += len;
720 	}
721 
722 
723 	first->tx_dma_desc.flags = flags;
724 	first->xfer_size = xfer_size;
725 	first->direction = direction;
726 	ret = &first->tx_dma_desc;
727 
728 spin_unlock:
729 	spin_unlock_irqrestore(&atchan->lock, irqflags);
730 	return ret;
731 }
732 
733 static struct dma_async_tx_descriptor *
734 at_xdmac_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
735 			 size_t buf_len, size_t period_len,
736 			 enum dma_transfer_direction direction,
737 			 unsigned long flags)
738 {
739 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
740 	struct at_xdmac_desc	*first = NULL, *prev = NULL;
741 	unsigned int		periods = buf_len / period_len;
742 	int			i;
743 	unsigned long		irqflags;
744 
745 	dev_dbg(chan2dev(chan), "%s: buf_addr=%pad, buf_len=%zd, period_len=%zd, dir=%s, flags=0x%lx\n",
746 		__func__, &buf_addr, buf_len, period_len,
747 		direction == DMA_MEM_TO_DEV ? "mem2per" : "per2mem", flags);
748 
749 	if (!is_slave_direction(direction)) {
750 		dev_err(chan2dev(chan), "invalid DMA direction\n");
751 		return NULL;
752 	}
753 
754 	if (test_and_set_bit(AT_XDMAC_CHAN_IS_CYCLIC, &atchan->status)) {
755 		dev_err(chan2dev(chan), "channel currently used\n");
756 		return NULL;
757 	}
758 
759 	if (at_xdmac_compute_chan_conf(chan, direction))
760 		return NULL;
761 
762 	for (i = 0; i < periods; i++) {
763 		struct at_xdmac_desc	*desc = NULL;
764 
765 		spin_lock_irqsave(&atchan->lock, irqflags);
766 		desc = at_xdmac_get_desc(atchan);
767 		if (!desc) {
768 			dev_err(chan2dev(chan), "can't get descriptor\n");
769 			if (first)
770 				list_splice_init(&first->descs_list, &atchan->free_descs_list);
771 			spin_unlock_irqrestore(&atchan->lock, irqflags);
772 			return NULL;
773 		}
774 		spin_unlock_irqrestore(&atchan->lock, irqflags);
775 		dev_dbg(chan2dev(chan),
776 			"%s: desc=0x%p, tx_dma_desc.phys=%pad\n",
777 			__func__, desc, &desc->tx_dma_desc.phys);
778 
779 		if (direction == DMA_DEV_TO_MEM) {
780 			desc->lld.mbr_sa = atchan->sconfig.src_addr;
781 			desc->lld.mbr_da = buf_addr + i * period_len;
782 		} else {
783 			desc->lld.mbr_sa = buf_addr + i * period_len;
784 			desc->lld.mbr_da = atchan->sconfig.dst_addr;
785 		}
786 		desc->lld.mbr_cfg = atchan->cfg;
787 		desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV1
788 			| AT_XDMAC_MBR_UBC_NDEN
789 			| AT_XDMAC_MBR_UBC_NSEN
790 			| period_len >> at_xdmac_get_dwidth(desc->lld.mbr_cfg);
791 
792 		dev_dbg(chan2dev(chan),
793 			 "%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x\n",
794 			 __func__, &desc->lld.mbr_sa, &desc->lld.mbr_da, desc->lld.mbr_ubc);
795 
796 		/* Chain lld. */
797 		if (prev)
798 			at_xdmac_queue_desc(chan, prev, desc);
799 
800 		prev = desc;
801 		if (!first)
802 			first = desc;
803 
804 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
805 			 __func__, desc, first);
806 		list_add_tail(&desc->desc_node, &first->descs_list);
807 	}
808 
809 	at_xdmac_queue_desc(chan, prev, first);
810 	first->tx_dma_desc.flags = flags;
811 	first->xfer_size = buf_len;
812 	first->direction = direction;
813 
814 	return &first->tx_dma_desc;
815 }
816 
817 static inline u32 at_xdmac_align_width(struct dma_chan *chan, dma_addr_t addr)
818 {
819 	u32 width;
820 
821 	/*
822 	 * Check address alignment to select the greater data width we
823 	 * can use.
824 	 *
825 	 * Some XDMAC implementations don't provide dword transfer, in
826 	 * this case selecting dword has the same behavior as
827 	 * selecting word transfers.
828 	 */
829 	if (!(addr & 7)) {
830 		width = AT_XDMAC_CC_DWIDTH_DWORD;
831 		dev_dbg(chan2dev(chan), "%s: dwidth: double word\n", __func__);
832 	} else if (!(addr & 3)) {
833 		width = AT_XDMAC_CC_DWIDTH_WORD;
834 		dev_dbg(chan2dev(chan), "%s: dwidth: word\n", __func__);
835 	} else if (!(addr & 1)) {
836 		width = AT_XDMAC_CC_DWIDTH_HALFWORD;
837 		dev_dbg(chan2dev(chan), "%s: dwidth: half word\n", __func__);
838 	} else {
839 		width = AT_XDMAC_CC_DWIDTH_BYTE;
840 		dev_dbg(chan2dev(chan), "%s: dwidth: byte\n", __func__);
841 	}
842 
843 	return width;
844 }
845 
846 static struct at_xdmac_desc *
847 at_xdmac_interleaved_queue_desc(struct dma_chan *chan,
848 				struct at_xdmac_chan *atchan,
849 				struct at_xdmac_desc *prev,
850 				dma_addr_t src, dma_addr_t dst,
851 				struct dma_interleaved_template *xt,
852 				struct data_chunk *chunk)
853 {
854 	struct at_xdmac_desc	*desc;
855 	u32			dwidth;
856 	unsigned long		flags;
857 	size_t			ublen;
858 	/*
859 	 * WARNING: The channel configuration is set here since there is no
860 	 * dmaengine_slave_config call in this case. Moreover we don't know the
861 	 * direction, it involves we can't dynamically set the source and dest
862 	 * interface so we have to use the same one. Only interface 0 allows EBI
863 	 * access. Hopefully we can access DDR through both ports (at least on
864 	 * SAMA5D4x), so we can use the same interface for source and dest,
865 	 * that solves the fact we don't know the direction.
866 	 * ERRATA: Even if useless for memory transfers, the PERID has to not
867 	 * match the one of another channel. If not, it could lead to spurious
868 	 * flag status.
869 	 */
870 	u32			chan_cc = AT_XDMAC_CC_PERID(0x3f)
871 					| AT_XDMAC_CC_DIF(0)
872 					| AT_XDMAC_CC_SIF(0)
873 					| AT_XDMAC_CC_MBSIZE_SIXTEEN
874 					| AT_XDMAC_CC_TYPE_MEM_TRAN;
875 
876 	dwidth = at_xdmac_align_width(chan, src | dst | chunk->size);
877 	if (chunk->size >= (AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth)) {
878 		dev_dbg(chan2dev(chan),
879 			"%s: chunk too big (%zu, max size %lu)...\n",
880 			__func__, chunk->size,
881 			AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth);
882 		return NULL;
883 	}
884 
885 	if (prev)
886 		dev_dbg(chan2dev(chan),
887 			"Adding items at the end of desc 0x%p\n", prev);
888 
889 	if (xt->src_inc) {
890 		if (xt->src_sgl)
891 			chan_cc |=  AT_XDMAC_CC_SAM_UBS_AM;
892 		else
893 			chan_cc |=  AT_XDMAC_CC_SAM_INCREMENTED_AM;
894 	}
895 
896 	if (xt->dst_inc) {
897 		if (xt->dst_sgl)
898 			chan_cc |=  AT_XDMAC_CC_DAM_UBS_AM;
899 		else
900 			chan_cc |=  AT_XDMAC_CC_DAM_INCREMENTED_AM;
901 	}
902 
903 	spin_lock_irqsave(&atchan->lock, flags);
904 	desc = at_xdmac_get_desc(atchan);
905 	spin_unlock_irqrestore(&atchan->lock, flags);
906 	if (!desc) {
907 		dev_err(chan2dev(chan), "can't get descriptor\n");
908 		return NULL;
909 	}
910 
911 	chan_cc |= AT_XDMAC_CC_DWIDTH(dwidth);
912 
913 	ublen = chunk->size >> dwidth;
914 
915 	desc->lld.mbr_sa = src;
916 	desc->lld.mbr_da = dst;
917 	desc->lld.mbr_sus = dmaengine_get_src_icg(xt, chunk);
918 	desc->lld.mbr_dus = dmaengine_get_dst_icg(xt, chunk);
919 
920 	desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV3
921 		| AT_XDMAC_MBR_UBC_NDEN
922 		| AT_XDMAC_MBR_UBC_NSEN
923 		| ublen;
924 	desc->lld.mbr_cfg = chan_cc;
925 
926 	dev_dbg(chan2dev(chan),
927 		"%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x, mbr_cfg=0x%08x\n",
928 		__func__, &desc->lld.mbr_sa, &desc->lld.mbr_da,
929 		desc->lld.mbr_ubc, desc->lld.mbr_cfg);
930 
931 	/* Chain lld. */
932 	if (prev)
933 		at_xdmac_queue_desc(chan, prev, desc);
934 
935 	return desc;
936 }
937 
938 static struct dma_async_tx_descriptor *
939 at_xdmac_prep_interleaved(struct dma_chan *chan,
940 			  struct dma_interleaved_template *xt,
941 			  unsigned long flags)
942 {
943 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
944 	struct at_xdmac_desc	*prev = NULL, *first = NULL;
945 	dma_addr_t		dst_addr, src_addr;
946 	size_t			src_skip = 0, dst_skip = 0, len = 0;
947 	struct data_chunk	*chunk;
948 	int			i;
949 
950 	if (!xt || !xt->numf || (xt->dir != DMA_MEM_TO_MEM))
951 		return NULL;
952 
953 	/*
954 	 * TODO: Handle the case where we have to repeat a chain of
955 	 * descriptors...
956 	 */
957 	if ((xt->numf > 1) && (xt->frame_size > 1))
958 		return NULL;
959 
960 	dev_dbg(chan2dev(chan), "%s: src=%pad, dest=%pad, numf=%zu, frame_size=%zu, flags=0x%lx\n",
961 		__func__, &xt->src_start, &xt->dst_start,	xt->numf,
962 		xt->frame_size, flags);
963 
964 	src_addr = xt->src_start;
965 	dst_addr = xt->dst_start;
966 
967 	if (xt->numf > 1) {
968 		first = at_xdmac_interleaved_queue_desc(chan, atchan,
969 							NULL,
970 							src_addr, dst_addr,
971 							xt, xt->sgl);
972 
973 		/* Length of the block is (BLEN+1) microblocks. */
974 		for (i = 0; i < xt->numf - 1; i++)
975 			at_xdmac_increment_block_count(chan, first);
976 
977 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
978 			__func__, first, first);
979 		list_add_tail(&first->desc_node, &first->descs_list);
980 	} else {
981 		for (i = 0; i < xt->frame_size; i++) {
982 			size_t src_icg = 0, dst_icg = 0;
983 			struct at_xdmac_desc *desc;
984 
985 			chunk = xt->sgl + i;
986 
987 			dst_icg = dmaengine_get_dst_icg(xt, chunk);
988 			src_icg = dmaengine_get_src_icg(xt, chunk);
989 
990 			src_skip = chunk->size + src_icg;
991 			dst_skip = chunk->size + dst_icg;
992 
993 			dev_dbg(chan2dev(chan),
994 				"%s: chunk size=%zu, src icg=%zu, dst icg=%zu\n",
995 				__func__, chunk->size, src_icg, dst_icg);
996 
997 			desc = at_xdmac_interleaved_queue_desc(chan, atchan,
998 							       prev,
999 							       src_addr, dst_addr,
1000 							       xt, chunk);
1001 			if (!desc) {
1002 				list_splice_init(&first->descs_list,
1003 						 &atchan->free_descs_list);
1004 				return NULL;
1005 			}
1006 
1007 			if (!first)
1008 				first = desc;
1009 
1010 			dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
1011 				__func__, desc, first);
1012 			list_add_tail(&desc->desc_node, &first->descs_list);
1013 
1014 			if (xt->src_sgl)
1015 				src_addr += src_skip;
1016 
1017 			if (xt->dst_sgl)
1018 				dst_addr += dst_skip;
1019 
1020 			len += chunk->size;
1021 			prev = desc;
1022 		}
1023 	}
1024 
1025 	first->tx_dma_desc.cookie = -EBUSY;
1026 	first->tx_dma_desc.flags = flags;
1027 	first->xfer_size = len;
1028 
1029 	return &first->tx_dma_desc;
1030 }
1031 
1032 static struct dma_async_tx_descriptor *
1033 at_xdmac_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1034 			 size_t len, unsigned long flags)
1035 {
1036 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1037 	struct at_xdmac_desc	*first = NULL, *prev = NULL;
1038 	size_t			remaining_size = len, xfer_size = 0, ublen;
1039 	dma_addr_t		src_addr = src, dst_addr = dest;
1040 	u32			dwidth;
1041 	/*
1042 	 * WARNING: We don't know the direction, it involves we can't
1043 	 * dynamically set the source and dest interface so we have to use the
1044 	 * same one. Only interface 0 allows EBI access. Hopefully we can
1045 	 * access DDR through both ports (at least on SAMA5D4x), so we can use
1046 	 * the same interface for source and dest, that solves the fact we
1047 	 * don't know the direction.
1048 	 * ERRATA: Even if useless for memory transfers, the PERID has to not
1049 	 * match the one of another channel. If not, it could lead to spurious
1050 	 * flag status.
1051 	 */
1052 	u32			chan_cc = AT_XDMAC_CC_PERID(0x3f)
1053 					| AT_XDMAC_CC_DAM_INCREMENTED_AM
1054 					| AT_XDMAC_CC_SAM_INCREMENTED_AM
1055 					| AT_XDMAC_CC_DIF(0)
1056 					| AT_XDMAC_CC_SIF(0)
1057 					| AT_XDMAC_CC_MBSIZE_SIXTEEN
1058 					| AT_XDMAC_CC_TYPE_MEM_TRAN;
1059 	unsigned long		irqflags;
1060 
1061 	dev_dbg(chan2dev(chan), "%s: src=%pad, dest=%pad, len=%zd, flags=0x%lx\n",
1062 		__func__, &src, &dest, len, flags);
1063 
1064 	if (unlikely(!len))
1065 		return NULL;
1066 
1067 	dwidth = at_xdmac_align_width(chan, src_addr | dst_addr);
1068 
1069 	/* Prepare descriptors. */
1070 	while (remaining_size) {
1071 		struct at_xdmac_desc	*desc = NULL;
1072 
1073 		dev_dbg(chan2dev(chan), "%s: remaining_size=%zu\n", __func__, remaining_size);
1074 
1075 		spin_lock_irqsave(&atchan->lock, irqflags);
1076 		desc = at_xdmac_get_desc(atchan);
1077 		spin_unlock_irqrestore(&atchan->lock, irqflags);
1078 		if (!desc) {
1079 			dev_err(chan2dev(chan), "can't get descriptor\n");
1080 			if (first)
1081 				list_splice_init(&first->descs_list, &atchan->free_descs_list);
1082 			return NULL;
1083 		}
1084 
1085 		/* Update src and dest addresses. */
1086 		src_addr += xfer_size;
1087 		dst_addr += xfer_size;
1088 
1089 		if (remaining_size >= AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth)
1090 			xfer_size = AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth;
1091 		else
1092 			xfer_size = remaining_size;
1093 
1094 		dev_dbg(chan2dev(chan), "%s: xfer_size=%zu\n", __func__, xfer_size);
1095 
1096 		/* Check remaining length and change data width if needed. */
1097 		dwidth = at_xdmac_align_width(chan,
1098 					      src_addr | dst_addr | xfer_size);
1099 		chan_cc &= ~AT_XDMAC_CC_DWIDTH_MASK;
1100 		chan_cc |= AT_XDMAC_CC_DWIDTH(dwidth);
1101 
1102 		ublen = xfer_size >> dwidth;
1103 		remaining_size -= xfer_size;
1104 
1105 		desc->lld.mbr_sa = src_addr;
1106 		desc->lld.mbr_da = dst_addr;
1107 		desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV2
1108 			| AT_XDMAC_MBR_UBC_NDEN
1109 			| AT_XDMAC_MBR_UBC_NSEN
1110 			| ublen;
1111 		desc->lld.mbr_cfg = chan_cc;
1112 
1113 		dev_dbg(chan2dev(chan),
1114 			 "%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x, mbr_cfg=0x%08x\n",
1115 			 __func__, &desc->lld.mbr_sa, &desc->lld.mbr_da, desc->lld.mbr_ubc, desc->lld.mbr_cfg);
1116 
1117 		/* Chain lld. */
1118 		if (prev)
1119 			at_xdmac_queue_desc(chan, prev, desc);
1120 
1121 		prev = desc;
1122 		if (!first)
1123 			first = desc;
1124 
1125 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
1126 			 __func__, desc, first);
1127 		list_add_tail(&desc->desc_node, &first->descs_list);
1128 	}
1129 
1130 	first->tx_dma_desc.flags = flags;
1131 	first->xfer_size = len;
1132 
1133 	return &first->tx_dma_desc;
1134 }
1135 
1136 static struct at_xdmac_desc *at_xdmac_memset_create_desc(struct dma_chan *chan,
1137 							 struct at_xdmac_chan *atchan,
1138 							 dma_addr_t dst_addr,
1139 							 size_t len,
1140 							 int value)
1141 {
1142 	struct at_xdmac_desc	*desc;
1143 	unsigned long		flags;
1144 	size_t			ublen;
1145 	u32			dwidth;
1146 	/*
1147 	 * WARNING: The channel configuration is set here since there is no
1148 	 * dmaengine_slave_config call in this case. Moreover we don't know the
1149 	 * direction, it involves we can't dynamically set the source and dest
1150 	 * interface so we have to use the same one. Only interface 0 allows EBI
1151 	 * access. Hopefully we can access DDR through both ports (at least on
1152 	 * SAMA5D4x), so we can use the same interface for source and dest,
1153 	 * that solves the fact we don't know the direction.
1154 	 * ERRATA: Even if useless for memory transfers, the PERID has to not
1155 	 * match the one of another channel. If not, it could lead to spurious
1156 	 * flag status.
1157 	 */
1158 	u32			chan_cc = AT_XDMAC_CC_PERID(0x3f)
1159 					| AT_XDMAC_CC_DAM_UBS_AM
1160 					| AT_XDMAC_CC_SAM_INCREMENTED_AM
1161 					| AT_XDMAC_CC_DIF(0)
1162 					| AT_XDMAC_CC_SIF(0)
1163 					| AT_XDMAC_CC_MBSIZE_SIXTEEN
1164 					| AT_XDMAC_CC_MEMSET_HW_MODE
1165 					| AT_XDMAC_CC_TYPE_MEM_TRAN;
1166 
1167 	dwidth = at_xdmac_align_width(chan, dst_addr);
1168 
1169 	if (len >= (AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth)) {
1170 		dev_err(chan2dev(chan),
1171 			"%s: Transfer too large, aborting...\n",
1172 			__func__);
1173 		return NULL;
1174 	}
1175 
1176 	spin_lock_irqsave(&atchan->lock, flags);
1177 	desc = at_xdmac_get_desc(atchan);
1178 	spin_unlock_irqrestore(&atchan->lock, flags);
1179 	if (!desc) {
1180 		dev_err(chan2dev(chan), "can't get descriptor\n");
1181 		return NULL;
1182 	}
1183 
1184 	chan_cc |= AT_XDMAC_CC_DWIDTH(dwidth);
1185 
1186 	ublen = len >> dwidth;
1187 
1188 	desc->lld.mbr_da = dst_addr;
1189 	desc->lld.mbr_ds = value;
1190 	desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV3
1191 		| AT_XDMAC_MBR_UBC_NDEN
1192 		| AT_XDMAC_MBR_UBC_NSEN
1193 		| ublen;
1194 	desc->lld.mbr_cfg = chan_cc;
1195 
1196 	dev_dbg(chan2dev(chan),
1197 		"%s: lld: mbr_da=%pad, mbr_ds=0x%08x, mbr_ubc=0x%08x, mbr_cfg=0x%08x\n",
1198 		__func__, &desc->lld.mbr_da, desc->lld.mbr_ds, desc->lld.mbr_ubc,
1199 		desc->lld.mbr_cfg);
1200 
1201 	return desc;
1202 }
1203 
1204 static struct dma_async_tx_descriptor *
1205 at_xdmac_prep_dma_memset(struct dma_chan *chan, dma_addr_t dest, int value,
1206 			 size_t len, unsigned long flags)
1207 {
1208 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1209 	struct at_xdmac_desc	*desc;
1210 
1211 	dev_dbg(chan2dev(chan), "%s: dest=%pad, len=%zu, pattern=0x%x, flags=0x%lx\n",
1212 		__func__, &dest, len, value, flags);
1213 
1214 	if (unlikely(!len))
1215 		return NULL;
1216 
1217 	desc = at_xdmac_memset_create_desc(chan, atchan, dest, len, value);
1218 	list_add_tail(&desc->desc_node, &desc->descs_list);
1219 
1220 	desc->tx_dma_desc.cookie = -EBUSY;
1221 	desc->tx_dma_desc.flags = flags;
1222 	desc->xfer_size = len;
1223 
1224 	return &desc->tx_dma_desc;
1225 }
1226 
1227 static struct dma_async_tx_descriptor *
1228 at_xdmac_prep_dma_memset_sg(struct dma_chan *chan, struct scatterlist *sgl,
1229 			    unsigned int sg_len, int value,
1230 			    unsigned long flags)
1231 {
1232 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1233 	struct at_xdmac_desc	*desc, *pdesc = NULL,
1234 				*ppdesc = NULL, *first = NULL;
1235 	struct scatterlist	*sg, *psg = NULL, *ppsg = NULL;
1236 	size_t			stride = 0, pstride = 0, len = 0;
1237 	int			i;
1238 
1239 	if (!sgl)
1240 		return NULL;
1241 
1242 	dev_dbg(chan2dev(chan), "%s: sg_len=%d, value=0x%x, flags=0x%lx\n",
1243 		__func__, sg_len, value, flags);
1244 
1245 	/* Prepare descriptors. */
1246 	for_each_sg(sgl, sg, sg_len, i) {
1247 		dev_dbg(chan2dev(chan), "%s: dest=%pad, len=%d, pattern=0x%x, flags=0x%lx\n",
1248 			__func__, &sg_dma_address(sg), sg_dma_len(sg),
1249 			value, flags);
1250 		desc = at_xdmac_memset_create_desc(chan, atchan,
1251 						   sg_dma_address(sg),
1252 						   sg_dma_len(sg),
1253 						   value);
1254 		if (!desc && first)
1255 			list_splice_init(&first->descs_list,
1256 					 &atchan->free_descs_list);
1257 
1258 		if (!first)
1259 			first = desc;
1260 
1261 		/* Update our strides */
1262 		pstride = stride;
1263 		if (psg)
1264 			stride = sg_dma_address(sg) -
1265 				(sg_dma_address(psg) + sg_dma_len(psg));
1266 
1267 		/*
1268 		 * The scatterlist API gives us only the address and
1269 		 * length of each elements.
1270 		 *
1271 		 * Unfortunately, we don't have the stride, which we
1272 		 * will need to compute.
1273 		 *
1274 		 * That make us end up in a situation like this one:
1275 		 *    len    stride    len    stride    len
1276 		 * +-------+        +-------+        +-------+
1277 		 * |  N-2  |        |  N-1  |        |   N   |
1278 		 * +-------+        +-------+        +-------+
1279 		 *
1280 		 * We need all these three elements (N-2, N-1 and N)
1281 		 * to actually take the decision on whether we need to
1282 		 * queue N-1 or reuse N-2.
1283 		 *
1284 		 * We will only consider N if it is the last element.
1285 		 */
1286 		if (ppdesc && pdesc) {
1287 			if ((stride == pstride) &&
1288 			    (sg_dma_len(ppsg) == sg_dma_len(psg))) {
1289 				dev_dbg(chan2dev(chan),
1290 					"%s: desc 0x%p can be merged with desc 0x%p\n",
1291 					__func__, pdesc, ppdesc);
1292 
1293 				/*
1294 				 * Increment the block count of the
1295 				 * N-2 descriptor
1296 				 */
1297 				at_xdmac_increment_block_count(chan, ppdesc);
1298 				ppdesc->lld.mbr_dus = stride;
1299 
1300 				/*
1301 				 * Put back the N-1 descriptor in the
1302 				 * free descriptor list
1303 				 */
1304 				list_add_tail(&pdesc->desc_node,
1305 					      &atchan->free_descs_list);
1306 
1307 				/*
1308 				 * Make our N-1 descriptor pointer
1309 				 * point to the N-2 since they were
1310 				 * actually merged.
1311 				 */
1312 				pdesc = ppdesc;
1313 
1314 			/*
1315 			 * Rule out the case where we don't have
1316 			 * pstride computed yet (our second sg
1317 			 * element)
1318 			 *
1319 			 * We also want to catch the case where there
1320 			 * would be a negative stride,
1321 			 */
1322 			} else if (pstride ||
1323 				   sg_dma_address(sg) < sg_dma_address(psg)) {
1324 				/*
1325 				 * Queue the N-1 descriptor after the
1326 				 * N-2
1327 				 */
1328 				at_xdmac_queue_desc(chan, ppdesc, pdesc);
1329 
1330 				/*
1331 				 * Add the N-1 descriptor to the list
1332 				 * of the descriptors used for this
1333 				 * transfer
1334 				 */
1335 				list_add_tail(&desc->desc_node,
1336 					      &first->descs_list);
1337 				dev_dbg(chan2dev(chan),
1338 					"%s: add desc 0x%p to descs_list 0x%p\n",
1339 					__func__, desc, first);
1340 			}
1341 		}
1342 
1343 		/*
1344 		 * If we are the last element, just see if we have the
1345 		 * same size than the previous element.
1346 		 *
1347 		 * If so, we can merge it with the previous descriptor
1348 		 * since we don't care about the stride anymore.
1349 		 */
1350 		if ((i == (sg_len - 1)) &&
1351 		    sg_dma_len(psg) == sg_dma_len(sg)) {
1352 			dev_dbg(chan2dev(chan),
1353 				"%s: desc 0x%p can be merged with desc 0x%p\n",
1354 				__func__, desc, pdesc);
1355 
1356 			/*
1357 			 * Increment the block count of the N-1
1358 			 * descriptor
1359 			 */
1360 			at_xdmac_increment_block_count(chan, pdesc);
1361 			pdesc->lld.mbr_dus = stride;
1362 
1363 			/*
1364 			 * Put back the N descriptor in the free
1365 			 * descriptor list
1366 			 */
1367 			list_add_tail(&desc->desc_node,
1368 				      &atchan->free_descs_list);
1369 		}
1370 
1371 		/* Update our descriptors */
1372 		ppdesc = pdesc;
1373 		pdesc = desc;
1374 
1375 		/* Update our scatter pointers */
1376 		ppsg = psg;
1377 		psg = sg;
1378 
1379 		len += sg_dma_len(sg);
1380 	}
1381 
1382 	first->tx_dma_desc.cookie = -EBUSY;
1383 	first->tx_dma_desc.flags = flags;
1384 	first->xfer_size = len;
1385 
1386 	return &first->tx_dma_desc;
1387 }
1388 
1389 static enum dma_status
1390 at_xdmac_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
1391 		struct dma_tx_state *txstate)
1392 {
1393 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1394 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1395 	struct at_xdmac_desc	*desc, *_desc;
1396 	struct list_head	*descs_list;
1397 	enum dma_status		ret;
1398 	int			residue, retry;
1399 	u32			cur_nda, check_nda, cur_ubc, mask, value;
1400 	u8			dwidth = 0;
1401 	unsigned long		flags;
1402 	bool			initd;
1403 
1404 	ret = dma_cookie_status(chan, cookie, txstate);
1405 	if (ret == DMA_COMPLETE)
1406 		return ret;
1407 
1408 	if (!txstate)
1409 		return ret;
1410 
1411 	spin_lock_irqsave(&atchan->lock, flags);
1412 
1413 	desc = list_first_entry(&atchan->xfers_list, struct at_xdmac_desc, xfer_node);
1414 
1415 	/*
1416 	 * If the transfer has not been started yet, don't need to compute the
1417 	 * residue, it's the transfer length.
1418 	 */
1419 	if (!desc->active_xfer) {
1420 		dma_set_residue(txstate, desc->xfer_size);
1421 		goto spin_unlock;
1422 	}
1423 
1424 	residue = desc->xfer_size;
1425 	/*
1426 	 * Flush FIFO: only relevant when the transfer is source peripheral
1427 	 * synchronized. Flush is needed before reading CUBC because data in
1428 	 * the FIFO are not reported by CUBC. Reporting a residue of the
1429 	 * transfer length while we have data in FIFO can cause issue.
1430 	 * Usecase: atmel USART has a timeout which means I have received
1431 	 * characters but there is no more character received for a while. On
1432 	 * timeout, it requests the residue. If the data are in the DMA FIFO,
1433 	 * we will return a residue of the transfer length. It means no data
1434 	 * received. If an application is waiting for these data, it will hang
1435 	 * since we won't have another USART timeout without receiving new
1436 	 * data.
1437 	 */
1438 	mask = AT_XDMAC_CC_TYPE | AT_XDMAC_CC_DSYNC;
1439 	value = AT_XDMAC_CC_TYPE_PER_TRAN | AT_XDMAC_CC_DSYNC_PER2MEM;
1440 	if ((desc->lld.mbr_cfg & mask) == value) {
1441 		at_xdmac_write(atxdmac, AT_XDMAC_GSWF, atchan->mask);
1442 		while (!(at_xdmac_chan_read(atchan, AT_XDMAC_CIS) & AT_XDMAC_CIS_FIS))
1443 			cpu_relax();
1444 	}
1445 
1446 	/*
1447 	 * The easiest way to compute the residue should be to pause the DMA
1448 	 * but doing this can lead to miss some data as some devices don't
1449 	 * have FIFO.
1450 	 * We need to read several registers because:
1451 	 * - DMA is running therefore a descriptor change is possible while
1452 	 * reading these registers
1453 	 * - When the block transfer is done, the value of the CUBC register
1454 	 * is set to its initial value until the fetch of the next descriptor.
1455 	 * This value will corrupt the residue calculation so we have to skip
1456 	 * it.
1457 	 *
1458 	 * INITD --------                    ------------
1459 	 *              |____________________|
1460 	 *       _______________________  _______________
1461 	 * NDA       @desc2             \/   @desc3
1462 	 *       _______________________/\_______________
1463 	 *       __________  ___________  _______________
1464 	 * CUBC       0    \/ MAX desc1 \/  MAX desc2
1465 	 *       __________/\___________/\_______________
1466 	 *
1467 	 * Since descriptors are aligned on 64 bits, we can assume that
1468 	 * the update of NDA and CUBC is atomic.
1469 	 * Memory barriers are used to ensure the read order of the registers.
1470 	 * A max number of retries is set because unlikely it could never ends.
1471 	 */
1472 	for (retry = 0; retry < AT_XDMAC_RESIDUE_MAX_RETRIES; retry++) {
1473 		check_nda = at_xdmac_chan_read(atchan, AT_XDMAC_CNDA) & 0xfffffffc;
1474 		rmb();
1475 		cur_ubc = at_xdmac_chan_read(atchan, AT_XDMAC_CUBC);
1476 		rmb();
1477 		initd = !!(at_xdmac_chan_read(atchan, AT_XDMAC_CC) & AT_XDMAC_CC_INITD);
1478 		rmb();
1479 		cur_nda = at_xdmac_chan_read(atchan, AT_XDMAC_CNDA) & 0xfffffffc;
1480 		rmb();
1481 
1482 		if ((check_nda == cur_nda) && initd)
1483 			break;
1484 	}
1485 
1486 	if (unlikely(retry >= AT_XDMAC_RESIDUE_MAX_RETRIES)) {
1487 		ret = DMA_ERROR;
1488 		goto spin_unlock;
1489 	}
1490 
1491 	/*
1492 	 * Flush FIFO: only relevant when the transfer is source peripheral
1493 	 * synchronized. Another flush is needed here because CUBC is updated
1494 	 * when the controller sends the data write command. It can lead to
1495 	 * report data that are not written in the memory or the device. The
1496 	 * FIFO flush ensures that data are really written.
1497 	 */
1498 	if ((desc->lld.mbr_cfg & mask) == value) {
1499 		at_xdmac_write(atxdmac, AT_XDMAC_GSWF, atchan->mask);
1500 		while (!(at_xdmac_chan_read(atchan, AT_XDMAC_CIS) & AT_XDMAC_CIS_FIS))
1501 			cpu_relax();
1502 	}
1503 
1504 	/*
1505 	 * Remove size of all microblocks already transferred and the current
1506 	 * one. Then add the remaining size to transfer of the current
1507 	 * microblock.
1508 	 */
1509 	descs_list = &desc->descs_list;
1510 	list_for_each_entry_safe(desc, _desc, descs_list, desc_node) {
1511 		dwidth = at_xdmac_get_dwidth(desc->lld.mbr_cfg);
1512 		residue -= (desc->lld.mbr_ubc & 0xffffff) << dwidth;
1513 		if ((desc->lld.mbr_nda & 0xfffffffc) == cur_nda)
1514 			break;
1515 	}
1516 	residue += cur_ubc << dwidth;
1517 
1518 	dma_set_residue(txstate, residue);
1519 
1520 	dev_dbg(chan2dev(chan),
1521 		 "%s: desc=0x%p, tx_dma_desc.phys=%pad, tx_status=%d, cookie=%d, residue=%d\n",
1522 		 __func__, desc, &desc->tx_dma_desc.phys, ret, cookie, residue);
1523 
1524 spin_unlock:
1525 	spin_unlock_irqrestore(&atchan->lock, flags);
1526 	return ret;
1527 }
1528 
1529 /* Call must be protected by lock. */
1530 static void at_xdmac_remove_xfer(struct at_xdmac_chan *atchan,
1531 				    struct at_xdmac_desc *desc)
1532 {
1533 	dev_dbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, desc);
1534 
1535 	/*
1536 	 * Remove the transfer from the transfer list then move the transfer
1537 	 * descriptors into the free descriptors list.
1538 	 */
1539 	list_del(&desc->xfer_node);
1540 	list_splice_init(&desc->descs_list, &atchan->free_descs_list);
1541 }
1542 
1543 static void at_xdmac_advance_work(struct at_xdmac_chan *atchan)
1544 {
1545 	struct at_xdmac_desc	*desc;
1546 
1547 	/*
1548 	 * If channel is enabled, do nothing, advance_work will be triggered
1549 	 * after the interruption.
1550 	 */
1551 	if (!at_xdmac_chan_is_enabled(atchan) && !list_empty(&atchan->xfers_list)) {
1552 		desc = list_first_entry(&atchan->xfers_list,
1553 					struct at_xdmac_desc,
1554 					xfer_node);
1555 		dev_vdbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, desc);
1556 		if (!desc->active_xfer)
1557 			at_xdmac_start_xfer(atchan, desc);
1558 	}
1559 }
1560 
1561 static void at_xdmac_handle_cyclic(struct at_xdmac_chan *atchan)
1562 {
1563 	struct at_xdmac_desc		*desc;
1564 	struct dma_async_tx_descriptor	*txd;
1565 
1566 	if (!list_empty(&atchan->xfers_list)) {
1567 		desc = list_first_entry(&atchan->xfers_list,
1568 					struct at_xdmac_desc, xfer_node);
1569 		txd = &desc->tx_dma_desc;
1570 
1571 		if (txd->flags & DMA_PREP_INTERRUPT)
1572 			dmaengine_desc_get_callback_invoke(txd, NULL);
1573 	}
1574 }
1575 
1576 static void at_xdmac_handle_error(struct at_xdmac_chan *atchan)
1577 {
1578 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1579 	struct at_xdmac_desc	*bad_desc;
1580 
1581 	/*
1582 	 * The descriptor currently at the head of the active list is
1583 	 * broken. Since we don't have any way to report errors, we'll
1584 	 * just have to scream loudly and try to continue with other
1585 	 * descriptors queued (if any).
1586 	 */
1587 	if (atchan->irq_status & AT_XDMAC_CIS_RBEIS)
1588 		dev_err(chan2dev(&atchan->chan), "read bus error!!!");
1589 	if (atchan->irq_status & AT_XDMAC_CIS_WBEIS)
1590 		dev_err(chan2dev(&atchan->chan), "write bus error!!!");
1591 	if (atchan->irq_status & AT_XDMAC_CIS_ROIS)
1592 		dev_err(chan2dev(&atchan->chan), "request overflow error!!!");
1593 
1594 	spin_lock_irq(&atchan->lock);
1595 
1596 	/* Channel must be disabled first as it's not done automatically */
1597 	at_xdmac_write(atxdmac, AT_XDMAC_GD, atchan->mask);
1598 	while (at_xdmac_read(atxdmac, AT_XDMAC_GS) & atchan->mask)
1599 		cpu_relax();
1600 
1601 	bad_desc = list_first_entry(&atchan->xfers_list,
1602 				    struct at_xdmac_desc,
1603 				    xfer_node);
1604 
1605 	spin_unlock_irq(&atchan->lock);
1606 
1607 	/* Print bad descriptor's details if needed */
1608 	dev_dbg(chan2dev(&atchan->chan),
1609 		"%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x\n",
1610 		__func__, &bad_desc->lld.mbr_sa, &bad_desc->lld.mbr_da,
1611 		bad_desc->lld.mbr_ubc);
1612 
1613 	/* Then continue with usual descriptor management */
1614 }
1615 
1616 static void at_xdmac_tasklet(unsigned long data)
1617 {
1618 	struct at_xdmac_chan	*atchan = (struct at_xdmac_chan *)data;
1619 	struct at_xdmac_desc	*desc;
1620 	u32			error_mask;
1621 
1622 	dev_dbg(chan2dev(&atchan->chan), "%s: status=0x%08x\n",
1623 		__func__, atchan->irq_status);
1624 
1625 	error_mask = AT_XDMAC_CIS_RBEIS
1626 		     | AT_XDMAC_CIS_WBEIS
1627 		     | AT_XDMAC_CIS_ROIS;
1628 
1629 	if (at_xdmac_chan_is_cyclic(atchan)) {
1630 		at_xdmac_handle_cyclic(atchan);
1631 	} else if ((atchan->irq_status & AT_XDMAC_CIS_LIS)
1632 		   || (atchan->irq_status & error_mask)) {
1633 		struct dma_async_tx_descriptor  *txd;
1634 
1635 		if (atchan->irq_status & error_mask)
1636 			at_xdmac_handle_error(atchan);
1637 
1638 		spin_lock_irq(&atchan->lock);
1639 		desc = list_first_entry(&atchan->xfers_list,
1640 					struct at_xdmac_desc,
1641 					xfer_node);
1642 		dev_vdbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, desc);
1643 		if (!desc->active_xfer) {
1644 			dev_err(chan2dev(&atchan->chan), "Xfer not active: exiting");
1645 			spin_unlock_irq(&atchan->lock);
1646 			return;
1647 		}
1648 
1649 		txd = &desc->tx_dma_desc;
1650 
1651 		at_xdmac_remove_xfer(atchan, desc);
1652 		spin_unlock_irq(&atchan->lock);
1653 
1654 		dma_cookie_complete(txd);
1655 		if (txd->flags & DMA_PREP_INTERRUPT)
1656 			dmaengine_desc_get_callback_invoke(txd, NULL);
1657 
1658 		dma_run_dependencies(txd);
1659 
1660 		spin_lock_irq(&atchan->lock);
1661 		at_xdmac_advance_work(atchan);
1662 		spin_unlock_irq(&atchan->lock);
1663 	}
1664 }
1665 
1666 static irqreturn_t at_xdmac_interrupt(int irq, void *dev_id)
1667 {
1668 	struct at_xdmac		*atxdmac = (struct at_xdmac *)dev_id;
1669 	struct at_xdmac_chan	*atchan;
1670 	u32			imr, status, pending;
1671 	u32			chan_imr, chan_status;
1672 	int			i, ret = IRQ_NONE;
1673 
1674 	do {
1675 		imr = at_xdmac_read(atxdmac, AT_XDMAC_GIM);
1676 		status = at_xdmac_read(atxdmac, AT_XDMAC_GIS);
1677 		pending = status & imr;
1678 
1679 		dev_vdbg(atxdmac->dma.dev,
1680 			 "%s: status=0x%08x, imr=0x%08x, pending=0x%08x\n",
1681 			 __func__, status, imr, pending);
1682 
1683 		if (!pending)
1684 			break;
1685 
1686 		/* We have to find which channel has generated the interrupt. */
1687 		for (i = 0; i < atxdmac->dma.chancnt; i++) {
1688 			if (!((1 << i) & pending))
1689 				continue;
1690 
1691 			atchan = &atxdmac->chan[i];
1692 			chan_imr = at_xdmac_chan_read(atchan, AT_XDMAC_CIM);
1693 			chan_status = at_xdmac_chan_read(atchan, AT_XDMAC_CIS);
1694 			atchan->irq_status = chan_status & chan_imr;
1695 			dev_vdbg(atxdmac->dma.dev,
1696 				 "%s: chan%d: imr=0x%x, status=0x%x\n",
1697 				 __func__, i, chan_imr, chan_status);
1698 			dev_vdbg(chan2dev(&atchan->chan),
1699 				 "%s: CC=0x%08x CNDA=0x%08x, CNDC=0x%08x, CSA=0x%08x, CDA=0x%08x, CUBC=0x%08x\n",
1700 				 __func__,
1701 				 at_xdmac_chan_read(atchan, AT_XDMAC_CC),
1702 				 at_xdmac_chan_read(atchan, AT_XDMAC_CNDA),
1703 				 at_xdmac_chan_read(atchan, AT_XDMAC_CNDC),
1704 				 at_xdmac_chan_read(atchan, AT_XDMAC_CSA),
1705 				 at_xdmac_chan_read(atchan, AT_XDMAC_CDA),
1706 				 at_xdmac_chan_read(atchan, AT_XDMAC_CUBC));
1707 
1708 			if (atchan->irq_status & (AT_XDMAC_CIS_RBEIS | AT_XDMAC_CIS_WBEIS))
1709 				at_xdmac_write(atxdmac, AT_XDMAC_GD, atchan->mask);
1710 
1711 			tasklet_schedule(&atchan->tasklet);
1712 			ret = IRQ_HANDLED;
1713 		}
1714 
1715 	} while (pending);
1716 
1717 	return ret;
1718 }
1719 
1720 static void at_xdmac_issue_pending(struct dma_chan *chan)
1721 {
1722 	struct at_xdmac_chan *atchan = to_at_xdmac_chan(chan);
1723 	unsigned long flags;
1724 
1725 	dev_dbg(chan2dev(&atchan->chan), "%s\n", __func__);
1726 
1727 	if (!at_xdmac_chan_is_cyclic(atchan)) {
1728 		spin_lock_irqsave(&atchan->lock, flags);
1729 		at_xdmac_advance_work(atchan);
1730 		spin_unlock_irqrestore(&atchan->lock, flags);
1731 	}
1732 
1733 	return;
1734 }
1735 
1736 static int at_xdmac_device_config(struct dma_chan *chan,
1737 				  struct dma_slave_config *config)
1738 {
1739 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1740 	int ret;
1741 	unsigned long		flags;
1742 
1743 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1744 
1745 	spin_lock_irqsave(&atchan->lock, flags);
1746 	ret = at_xdmac_set_slave_config(chan, config);
1747 	spin_unlock_irqrestore(&atchan->lock, flags);
1748 
1749 	return ret;
1750 }
1751 
1752 static int at_xdmac_device_pause(struct dma_chan *chan)
1753 {
1754 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1755 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1756 	unsigned long		flags;
1757 
1758 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1759 
1760 	if (test_and_set_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status))
1761 		return 0;
1762 
1763 	spin_lock_irqsave(&atchan->lock, flags);
1764 	at_xdmac_write(atxdmac, AT_XDMAC_GRWS, atchan->mask);
1765 	while (at_xdmac_chan_read(atchan, AT_XDMAC_CC)
1766 	       & (AT_XDMAC_CC_WRIP | AT_XDMAC_CC_RDIP))
1767 		cpu_relax();
1768 	spin_unlock_irqrestore(&atchan->lock, flags);
1769 
1770 	return 0;
1771 }
1772 
1773 static int at_xdmac_device_resume(struct dma_chan *chan)
1774 {
1775 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1776 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1777 	unsigned long		flags;
1778 
1779 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1780 
1781 	spin_lock_irqsave(&atchan->lock, flags);
1782 	if (!at_xdmac_chan_is_paused(atchan)) {
1783 		spin_unlock_irqrestore(&atchan->lock, flags);
1784 		return 0;
1785 	}
1786 
1787 	at_xdmac_write(atxdmac, AT_XDMAC_GRWR, atchan->mask);
1788 	clear_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status);
1789 	spin_unlock_irqrestore(&atchan->lock, flags);
1790 
1791 	return 0;
1792 }
1793 
1794 static int at_xdmac_device_terminate_all(struct dma_chan *chan)
1795 {
1796 	struct at_xdmac_desc	*desc, *_desc;
1797 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1798 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1799 	unsigned long		flags;
1800 
1801 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1802 
1803 	spin_lock_irqsave(&atchan->lock, flags);
1804 	at_xdmac_write(atxdmac, AT_XDMAC_GD, atchan->mask);
1805 	while (at_xdmac_read(atxdmac, AT_XDMAC_GS) & atchan->mask)
1806 		cpu_relax();
1807 
1808 	/* Cancel all pending transfers. */
1809 	list_for_each_entry_safe(desc, _desc, &atchan->xfers_list, xfer_node)
1810 		at_xdmac_remove_xfer(atchan, desc);
1811 
1812 	clear_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status);
1813 	clear_bit(AT_XDMAC_CHAN_IS_CYCLIC, &atchan->status);
1814 	spin_unlock_irqrestore(&atchan->lock, flags);
1815 
1816 	return 0;
1817 }
1818 
1819 static int at_xdmac_alloc_chan_resources(struct dma_chan *chan)
1820 {
1821 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1822 	struct at_xdmac_desc	*desc;
1823 	int			i;
1824 
1825 	if (at_xdmac_chan_is_enabled(atchan)) {
1826 		dev_err(chan2dev(chan),
1827 			"can't allocate channel resources (channel enabled)\n");
1828 		return -EIO;
1829 	}
1830 
1831 	if (!list_empty(&atchan->free_descs_list)) {
1832 		dev_err(chan2dev(chan),
1833 			"can't allocate channel resources (channel not free from a previous use)\n");
1834 		return -EIO;
1835 	}
1836 
1837 	for (i = 0; i < init_nr_desc_per_channel; i++) {
1838 		desc = at_xdmac_alloc_desc(chan, GFP_KERNEL);
1839 		if (!desc) {
1840 			dev_warn(chan2dev(chan),
1841 				"only %d descriptors have been allocated\n", i);
1842 			break;
1843 		}
1844 		list_add_tail(&desc->desc_node, &atchan->free_descs_list);
1845 	}
1846 
1847 	dma_cookie_init(chan);
1848 
1849 	dev_dbg(chan2dev(chan), "%s: allocated %d descriptors\n", __func__, i);
1850 
1851 	return i;
1852 }
1853 
1854 static void at_xdmac_free_chan_resources(struct dma_chan *chan)
1855 {
1856 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1857 	struct at_xdmac		*atxdmac = to_at_xdmac(chan->device);
1858 	struct at_xdmac_desc	*desc, *_desc;
1859 
1860 	list_for_each_entry_safe(desc, _desc, &atchan->free_descs_list, desc_node) {
1861 		dev_dbg(chan2dev(chan), "%s: freeing descriptor %p\n", __func__, desc);
1862 		list_del(&desc->desc_node);
1863 		dma_pool_free(atxdmac->at_xdmac_desc_pool, desc, desc->tx_dma_desc.phys);
1864 	}
1865 
1866 	return;
1867 }
1868 
1869 #ifdef CONFIG_PM
1870 static int atmel_xdmac_prepare(struct device *dev)
1871 {
1872 	struct at_xdmac		*atxdmac = dev_get_drvdata(dev);
1873 	struct dma_chan		*chan, *_chan;
1874 
1875 	list_for_each_entry_safe(chan, _chan, &atxdmac->dma.channels, device_node) {
1876 		struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1877 
1878 		/* Wait for transfer completion, except in cyclic case. */
1879 		if (at_xdmac_chan_is_enabled(atchan) && !at_xdmac_chan_is_cyclic(atchan))
1880 			return -EAGAIN;
1881 	}
1882 	return 0;
1883 }
1884 #else
1885 #	define atmel_xdmac_prepare NULL
1886 #endif
1887 
1888 #ifdef CONFIG_PM_SLEEP
1889 static int atmel_xdmac_suspend(struct device *dev)
1890 {
1891 	struct at_xdmac		*atxdmac = dev_get_drvdata(dev);
1892 	struct dma_chan		*chan, *_chan;
1893 
1894 	list_for_each_entry_safe(chan, _chan, &atxdmac->dma.channels, device_node) {
1895 		struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1896 
1897 		atchan->save_cc = at_xdmac_chan_read(atchan, AT_XDMAC_CC);
1898 		if (at_xdmac_chan_is_cyclic(atchan)) {
1899 			if (!at_xdmac_chan_is_paused(atchan))
1900 				at_xdmac_device_pause(chan);
1901 			atchan->save_cim = at_xdmac_chan_read(atchan, AT_XDMAC_CIM);
1902 			atchan->save_cnda = at_xdmac_chan_read(atchan, AT_XDMAC_CNDA);
1903 			atchan->save_cndc = at_xdmac_chan_read(atchan, AT_XDMAC_CNDC);
1904 		}
1905 	}
1906 	atxdmac->save_gim = at_xdmac_read(atxdmac, AT_XDMAC_GIM);
1907 
1908 	at_xdmac_off(atxdmac);
1909 	clk_disable_unprepare(atxdmac->clk);
1910 	return 0;
1911 }
1912 
1913 static int atmel_xdmac_resume(struct device *dev)
1914 {
1915 	struct at_xdmac		*atxdmac = dev_get_drvdata(dev);
1916 	struct at_xdmac_chan	*atchan;
1917 	struct dma_chan		*chan, *_chan;
1918 	int			i;
1919 	int ret;
1920 
1921 	ret = clk_prepare_enable(atxdmac->clk);
1922 	if (ret)
1923 		return ret;
1924 
1925 	/* Clear pending interrupts. */
1926 	for (i = 0; i < atxdmac->dma.chancnt; i++) {
1927 		atchan = &atxdmac->chan[i];
1928 		while (at_xdmac_chan_read(atchan, AT_XDMAC_CIS))
1929 			cpu_relax();
1930 	}
1931 
1932 	at_xdmac_write(atxdmac, AT_XDMAC_GIE, atxdmac->save_gim);
1933 	list_for_each_entry_safe(chan, _chan, &atxdmac->dma.channels, device_node) {
1934 		atchan = to_at_xdmac_chan(chan);
1935 		at_xdmac_chan_write(atchan, AT_XDMAC_CC, atchan->save_cc);
1936 		if (at_xdmac_chan_is_cyclic(atchan)) {
1937 			if (at_xdmac_chan_is_paused(atchan))
1938 				at_xdmac_device_resume(chan);
1939 			at_xdmac_chan_write(atchan, AT_XDMAC_CNDA, atchan->save_cnda);
1940 			at_xdmac_chan_write(atchan, AT_XDMAC_CNDC, atchan->save_cndc);
1941 			at_xdmac_chan_write(atchan, AT_XDMAC_CIE, atchan->save_cim);
1942 			wmb();
1943 			at_xdmac_write(atxdmac, AT_XDMAC_GE, atchan->mask);
1944 		}
1945 	}
1946 	return 0;
1947 }
1948 #endif /* CONFIG_PM_SLEEP */
1949 
1950 static int at_xdmac_probe(struct platform_device *pdev)
1951 {
1952 	struct at_xdmac	*atxdmac;
1953 	int		irq, size, nr_channels, i, ret;
1954 	void __iomem	*base;
1955 	u32		reg;
1956 
1957 	irq = platform_get_irq(pdev, 0);
1958 	if (irq < 0)
1959 		return irq;
1960 
1961 	base = devm_platform_ioremap_resource(pdev, 0);
1962 	if (IS_ERR(base))
1963 		return PTR_ERR(base);
1964 
1965 	/*
1966 	 * Read number of xdmac channels, read helper function can't be used
1967 	 * since atxdmac is not yet allocated and we need to know the number
1968 	 * of channels to do the allocation.
1969 	 */
1970 	reg = readl_relaxed(base + AT_XDMAC_GTYPE);
1971 	nr_channels = AT_XDMAC_NB_CH(reg);
1972 	if (nr_channels > AT_XDMAC_MAX_CHAN) {
1973 		dev_err(&pdev->dev, "invalid number of channels (%u)\n",
1974 			nr_channels);
1975 		return -EINVAL;
1976 	}
1977 
1978 	size = sizeof(*atxdmac);
1979 	size += nr_channels * sizeof(struct at_xdmac_chan);
1980 	atxdmac = devm_kzalloc(&pdev->dev, size, GFP_KERNEL);
1981 	if (!atxdmac) {
1982 		dev_err(&pdev->dev, "can't allocate at_xdmac structure\n");
1983 		return -ENOMEM;
1984 	}
1985 
1986 	atxdmac->regs = base;
1987 	atxdmac->irq = irq;
1988 
1989 	atxdmac->clk = devm_clk_get(&pdev->dev, "dma_clk");
1990 	if (IS_ERR(atxdmac->clk)) {
1991 		dev_err(&pdev->dev, "can't get dma_clk\n");
1992 		return PTR_ERR(atxdmac->clk);
1993 	}
1994 
1995 	/* Do not use dev res to prevent races with tasklet */
1996 	ret = request_irq(atxdmac->irq, at_xdmac_interrupt, 0, "at_xdmac", atxdmac);
1997 	if (ret) {
1998 		dev_err(&pdev->dev, "can't request irq\n");
1999 		return ret;
2000 	}
2001 
2002 	ret = clk_prepare_enable(atxdmac->clk);
2003 	if (ret) {
2004 		dev_err(&pdev->dev, "can't prepare or enable clock\n");
2005 		goto err_free_irq;
2006 	}
2007 
2008 	atxdmac->at_xdmac_desc_pool =
2009 		dmam_pool_create(dev_name(&pdev->dev), &pdev->dev,
2010 				sizeof(struct at_xdmac_desc), 4, 0);
2011 	if (!atxdmac->at_xdmac_desc_pool) {
2012 		dev_err(&pdev->dev, "no memory for descriptors dma pool\n");
2013 		ret = -ENOMEM;
2014 		goto err_clk_disable;
2015 	}
2016 
2017 	dma_cap_set(DMA_CYCLIC, atxdmac->dma.cap_mask);
2018 	dma_cap_set(DMA_INTERLEAVE, atxdmac->dma.cap_mask);
2019 	dma_cap_set(DMA_MEMCPY, atxdmac->dma.cap_mask);
2020 	dma_cap_set(DMA_MEMSET, atxdmac->dma.cap_mask);
2021 	dma_cap_set(DMA_MEMSET_SG, atxdmac->dma.cap_mask);
2022 	dma_cap_set(DMA_SLAVE, atxdmac->dma.cap_mask);
2023 	/*
2024 	 * Without DMA_PRIVATE the driver is not able to allocate more than
2025 	 * one channel, second allocation fails in private_candidate.
2026 	 */
2027 	dma_cap_set(DMA_PRIVATE, atxdmac->dma.cap_mask);
2028 	atxdmac->dma.dev				= &pdev->dev;
2029 	atxdmac->dma.device_alloc_chan_resources	= at_xdmac_alloc_chan_resources;
2030 	atxdmac->dma.device_free_chan_resources		= at_xdmac_free_chan_resources;
2031 	atxdmac->dma.device_tx_status			= at_xdmac_tx_status;
2032 	atxdmac->dma.device_issue_pending		= at_xdmac_issue_pending;
2033 	atxdmac->dma.device_prep_dma_cyclic		= at_xdmac_prep_dma_cyclic;
2034 	atxdmac->dma.device_prep_interleaved_dma	= at_xdmac_prep_interleaved;
2035 	atxdmac->dma.device_prep_dma_memcpy		= at_xdmac_prep_dma_memcpy;
2036 	atxdmac->dma.device_prep_dma_memset		= at_xdmac_prep_dma_memset;
2037 	atxdmac->dma.device_prep_dma_memset_sg		= at_xdmac_prep_dma_memset_sg;
2038 	atxdmac->dma.device_prep_slave_sg		= at_xdmac_prep_slave_sg;
2039 	atxdmac->dma.device_config			= at_xdmac_device_config;
2040 	atxdmac->dma.device_pause			= at_xdmac_device_pause;
2041 	atxdmac->dma.device_resume			= at_xdmac_device_resume;
2042 	atxdmac->dma.device_terminate_all		= at_xdmac_device_terminate_all;
2043 	atxdmac->dma.src_addr_widths = AT_XDMAC_DMA_BUSWIDTHS;
2044 	atxdmac->dma.dst_addr_widths = AT_XDMAC_DMA_BUSWIDTHS;
2045 	atxdmac->dma.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2046 	atxdmac->dma.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2047 
2048 	/* Disable all chans and interrupts. */
2049 	at_xdmac_off(atxdmac);
2050 
2051 	/* Init channels. */
2052 	INIT_LIST_HEAD(&atxdmac->dma.channels);
2053 	for (i = 0; i < nr_channels; i++) {
2054 		struct at_xdmac_chan *atchan = &atxdmac->chan[i];
2055 
2056 		atchan->chan.device = &atxdmac->dma;
2057 		list_add_tail(&atchan->chan.device_node,
2058 			      &atxdmac->dma.channels);
2059 
2060 		atchan->ch_regs = at_xdmac_chan_reg_base(atxdmac, i);
2061 		atchan->mask = 1 << i;
2062 
2063 		spin_lock_init(&atchan->lock);
2064 		INIT_LIST_HEAD(&atchan->xfers_list);
2065 		INIT_LIST_HEAD(&atchan->free_descs_list);
2066 		tasklet_init(&atchan->tasklet, at_xdmac_tasklet,
2067 			     (unsigned long)atchan);
2068 
2069 		/* Clear pending interrupts. */
2070 		while (at_xdmac_chan_read(atchan, AT_XDMAC_CIS))
2071 			cpu_relax();
2072 	}
2073 	platform_set_drvdata(pdev, atxdmac);
2074 
2075 	ret = dma_async_device_register(&atxdmac->dma);
2076 	if (ret) {
2077 		dev_err(&pdev->dev, "fail to register DMA engine device\n");
2078 		goto err_clk_disable;
2079 	}
2080 
2081 	ret = of_dma_controller_register(pdev->dev.of_node,
2082 					 at_xdmac_xlate, atxdmac);
2083 	if (ret) {
2084 		dev_err(&pdev->dev, "could not register of dma controller\n");
2085 		goto err_dma_unregister;
2086 	}
2087 
2088 	dev_info(&pdev->dev, "%d channels, mapped at 0x%p\n",
2089 		 nr_channels, atxdmac->regs);
2090 
2091 	return 0;
2092 
2093 err_dma_unregister:
2094 	dma_async_device_unregister(&atxdmac->dma);
2095 err_clk_disable:
2096 	clk_disable_unprepare(atxdmac->clk);
2097 err_free_irq:
2098 	free_irq(atxdmac->irq, atxdmac);
2099 	return ret;
2100 }
2101 
2102 static int at_xdmac_remove(struct platform_device *pdev)
2103 {
2104 	struct at_xdmac	*atxdmac = (struct at_xdmac *)platform_get_drvdata(pdev);
2105 	int		i;
2106 
2107 	at_xdmac_off(atxdmac);
2108 	of_dma_controller_free(pdev->dev.of_node);
2109 	dma_async_device_unregister(&atxdmac->dma);
2110 	clk_disable_unprepare(atxdmac->clk);
2111 
2112 	free_irq(atxdmac->irq, atxdmac);
2113 
2114 	for (i = 0; i < atxdmac->dma.chancnt; i++) {
2115 		struct at_xdmac_chan *atchan = &atxdmac->chan[i];
2116 
2117 		tasklet_kill(&atchan->tasklet);
2118 		at_xdmac_free_chan_resources(&atchan->chan);
2119 	}
2120 
2121 	return 0;
2122 }
2123 
2124 static const struct dev_pm_ops atmel_xdmac_dev_pm_ops = {
2125 	.prepare	= atmel_xdmac_prepare,
2126 	SET_LATE_SYSTEM_SLEEP_PM_OPS(atmel_xdmac_suspend, atmel_xdmac_resume)
2127 };
2128 
2129 static const struct of_device_id atmel_xdmac_dt_ids[] = {
2130 	{
2131 		.compatible = "atmel,sama5d4-dma",
2132 	}, {
2133 		/* sentinel */
2134 	}
2135 };
2136 MODULE_DEVICE_TABLE(of, atmel_xdmac_dt_ids);
2137 
2138 static struct platform_driver at_xdmac_driver = {
2139 	.probe		= at_xdmac_probe,
2140 	.remove		= at_xdmac_remove,
2141 	.driver = {
2142 		.name		= "at_xdmac",
2143 		.of_match_table	= of_match_ptr(atmel_xdmac_dt_ids),
2144 		.pm		= &atmel_xdmac_dev_pm_ops,
2145 	}
2146 };
2147 
2148 static int __init at_xdmac_init(void)
2149 {
2150 	return platform_driver_probe(&at_xdmac_driver, at_xdmac_probe);
2151 }
2152 subsys_initcall(at_xdmac_init);
2153 
2154 MODULE_DESCRIPTION("Atmel Extended DMA Controller driver");
2155 MODULE_AUTHOR("Ludovic Desroches <ludovic.desroches@atmel.com>");
2156 MODULE_LICENSE("GPL");
2157