xref: /linux/drivers/dma/at_xdmac.c (revision ab520be8cd5d56867fc95cfbc34b90880faf1f9d)
1 /*
2  * Driver for the Atmel Extensible DMA Controller (aka XDMAC on AT91 systems)
3  *
4  * Copyright (C) 2014 Atmel Corporation
5  *
6  * Author: Ludovic Desroches <ludovic.desroches@atmel.com>
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms of the GNU General Public License version 2 as published by
10  * the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful, but WITHOUT
13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
15  * more details.
16  *
17  * You should have received a copy of the GNU General Public License along with
18  * this program.  If not, see <http://www.gnu.org/licenses/>.
19  */
20 
21 #include <asm/barrier.h>
22 #include <dt-bindings/dma/at91.h>
23 #include <linux/clk.h>
24 #include <linux/dmaengine.h>
25 #include <linux/dmapool.h>
26 #include <linux/interrupt.h>
27 #include <linux/irq.h>
28 #include <linux/kernel.h>
29 #include <linux/list.h>
30 #include <linux/module.h>
31 #include <linux/of_dma.h>
32 #include <linux/of_platform.h>
33 #include <linux/platform_device.h>
34 #include <linux/pm.h>
35 
36 #include "dmaengine.h"
37 
38 /* Global registers */
39 #define AT_XDMAC_GTYPE		0x00	/* Global Type Register */
40 #define		AT_XDMAC_NB_CH(i)	(((i) & 0x1F) + 1)		/* Number of Channels Minus One */
41 #define		AT_XDMAC_FIFO_SZ(i)	(((i) >> 5) & 0x7FF)		/* Number of Bytes */
42 #define		AT_XDMAC_NB_REQ(i)	((((i) >> 16) & 0x3F) + 1)	/* Number of Peripheral Requests Minus One */
43 #define AT_XDMAC_GCFG		0x04	/* Global Configuration Register */
44 #define AT_XDMAC_GWAC		0x08	/* Global Weighted Arbiter Configuration Register */
45 #define AT_XDMAC_GIE		0x0C	/* Global Interrupt Enable Register */
46 #define AT_XDMAC_GID		0x10	/* Global Interrupt Disable Register */
47 #define AT_XDMAC_GIM		0x14	/* Global Interrupt Mask Register */
48 #define AT_XDMAC_GIS		0x18	/* Global Interrupt Status Register */
49 #define AT_XDMAC_GE		0x1C	/* Global Channel Enable Register */
50 #define AT_XDMAC_GD		0x20	/* Global Channel Disable Register */
51 #define AT_XDMAC_GS		0x24	/* Global Channel Status Register */
52 #define AT_XDMAC_GRS		0x28	/* Global Channel Read Suspend Register */
53 #define AT_XDMAC_GWS		0x2C	/* Global Write Suspend Register */
54 #define AT_XDMAC_GRWS		0x30	/* Global Channel Read Write Suspend Register */
55 #define AT_XDMAC_GRWR		0x34	/* Global Channel Read Write Resume Register */
56 #define AT_XDMAC_GSWR		0x38	/* Global Channel Software Request Register */
57 #define AT_XDMAC_GSWS		0x3C	/* Global channel Software Request Status Register */
58 #define AT_XDMAC_GSWF		0x40	/* Global Channel Software Flush Request Register */
59 #define AT_XDMAC_VERSION	0xFFC	/* XDMAC Version Register */
60 
61 /* Channel relative registers offsets */
62 #define AT_XDMAC_CIE		0x00	/* Channel Interrupt Enable Register */
63 #define		AT_XDMAC_CIE_BIE	BIT(0)	/* End of Block Interrupt Enable Bit */
64 #define		AT_XDMAC_CIE_LIE	BIT(1)	/* End of Linked List Interrupt Enable Bit */
65 #define		AT_XDMAC_CIE_DIE	BIT(2)	/* End of Disable Interrupt Enable Bit */
66 #define		AT_XDMAC_CIE_FIE	BIT(3)	/* End of Flush Interrupt Enable Bit */
67 #define		AT_XDMAC_CIE_RBEIE	BIT(4)	/* Read Bus Error Interrupt Enable Bit */
68 #define		AT_XDMAC_CIE_WBEIE	BIT(5)	/* Write Bus Error Interrupt Enable Bit */
69 #define		AT_XDMAC_CIE_ROIE	BIT(6)	/* Request Overflow Interrupt Enable Bit */
70 #define AT_XDMAC_CID		0x04	/* Channel Interrupt Disable Register */
71 #define		AT_XDMAC_CID_BID	BIT(0)	/* End of Block Interrupt Disable Bit */
72 #define		AT_XDMAC_CID_LID	BIT(1)	/* End of Linked List Interrupt Disable Bit */
73 #define		AT_XDMAC_CID_DID	BIT(2)	/* End of Disable Interrupt Disable Bit */
74 #define		AT_XDMAC_CID_FID	BIT(3)	/* End of Flush Interrupt Disable Bit */
75 #define		AT_XDMAC_CID_RBEID	BIT(4)	/* Read Bus Error Interrupt Disable Bit */
76 #define		AT_XDMAC_CID_WBEID	BIT(5)	/* Write Bus Error Interrupt Disable Bit */
77 #define		AT_XDMAC_CID_ROID	BIT(6)	/* Request Overflow Interrupt Disable Bit */
78 #define AT_XDMAC_CIM		0x08	/* Channel Interrupt Mask Register */
79 #define		AT_XDMAC_CIM_BIM	BIT(0)	/* End of Block Interrupt Mask Bit */
80 #define		AT_XDMAC_CIM_LIM	BIT(1)	/* End of Linked List Interrupt Mask Bit */
81 #define		AT_XDMAC_CIM_DIM	BIT(2)	/* End of Disable Interrupt Mask Bit */
82 #define		AT_XDMAC_CIM_FIM	BIT(3)	/* End of Flush Interrupt Mask Bit */
83 #define		AT_XDMAC_CIM_RBEIM	BIT(4)	/* Read Bus Error Interrupt Mask Bit */
84 #define		AT_XDMAC_CIM_WBEIM	BIT(5)	/* Write Bus Error Interrupt Mask Bit */
85 #define		AT_XDMAC_CIM_ROIM	BIT(6)	/* Request Overflow Interrupt Mask Bit */
86 #define AT_XDMAC_CIS		0x0C	/* Channel Interrupt Status Register */
87 #define		AT_XDMAC_CIS_BIS	BIT(0)	/* End of Block Interrupt Status Bit */
88 #define		AT_XDMAC_CIS_LIS	BIT(1)	/* End of Linked List Interrupt Status Bit */
89 #define		AT_XDMAC_CIS_DIS	BIT(2)	/* End of Disable Interrupt Status Bit */
90 #define		AT_XDMAC_CIS_FIS	BIT(3)	/* End of Flush Interrupt Status Bit */
91 #define		AT_XDMAC_CIS_RBEIS	BIT(4)	/* Read Bus Error Interrupt Status Bit */
92 #define		AT_XDMAC_CIS_WBEIS	BIT(5)	/* Write Bus Error Interrupt Status Bit */
93 #define		AT_XDMAC_CIS_ROIS	BIT(6)	/* Request Overflow Interrupt Status Bit */
94 #define AT_XDMAC_CSA		0x10	/* Channel Source Address Register */
95 #define AT_XDMAC_CDA		0x14	/* Channel Destination Address Register */
96 #define AT_XDMAC_CNDA		0x18	/* Channel Next Descriptor Address Register */
97 #define		AT_XDMAC_CNDA_NDAIF(i)	((i) & 0x1)			/* Channel x Next Descriptor Interface */
98 #define		AT_XDMAC_CNDA_NDA(i)	((i) & 0xfffffffc)		/* Channel x Next Descriptor Address */
99 #define AT_XDMAC_CNDC		0x1C	/* Channel Next Descriptor Control Register */
100 #define		AT_XDMAC_CNDC_NDE		(0x1 << 0)		/* Channel x Next Descriptor Enable */
101 #define		AT_XDMAC_CNDC_NDSUP		(0x1 << 1)		/* Channel x Next Descriptor Source Update */
102 #define		AT_XDMAC_CNDC_NDDUP		(0x1 << 2)		/* Channel x Next Descriptor Destination Update */
103 #define		AT_XDMAC_CNDC_NDVIEW_NDV0	(0x0 << 3)		/* Channel x Next Descriptor View 0 */
104 #define		AT_XDMAC_CNDC_NDVIEW_NDV1	(0x1 << 3)		/* Channel x Next Descriptor View 1 */
105 #define		AT_XDMAC_CNDC_NDVIEW_NDV2	(0x2 << 3)		/* Channel x Next Descriptor View 2 */
106 #define		AT_XDMAC_CNDC_NDVIEW_NDV3	(0x3 << 3)		/* Channel x Next Descriptor View 3 */
107 #define AT_XDMAC_CUBC		0x20	/* Channel Microblock Control Register */
108 #define AT_XDMAC_CBC		0x24	/* Channel Block Control Register */
109 #define AT_XDMAC_CC		0x28	/* Channel Configuration Register */
110 #define		AT_XDMAC_CC_TYPE	(0x1 << 0)	/* Channel Transfer Type */
111 #define			AT_XDMAC_CC_TYPE_MEM_TRAN	(0x0 << 0)	/* Memory to Memory Transfer */
112 #define			AT_XDMAC_CC_TYPE_PER_TRAN	(0x1 << 0)	/* Peripheral to Memory or Memory to Peripheral Transfer */
113 #define		AT_XDMAC_CC_MBSIZE_MASK	(0x3 << 1)
114 #define			AT_XDMAC_CC_MBSIZE_SINGLE	(0x0 << 1)
115 #define			AT_XDMAC_CC_MBSIZE_FOUR		(0x1 << 1)
116 #define			AT_XDMAC_CC_MBSIZE_EIGHT	(0x2 << 1)
117 #define			AT_XDMAC_CC_MBSIZE_SIXTEEN	(0x3 << 1)
118 #define		AT_XDMAC_CC_DSYNC	(0x1 << 4)	/* Channel Synchronization */
119 #define			AT_XDMAC_CC_DSYNC_PER2MEM	(0x0 << 4)
120 #define			AT_XDMAC_CC_DSYNC_MEM2PER	(0x1 << 4)
121 #define		AT_XDMAC_CC_PROT	(0x1 << 5)	/* Channel Protection */
122 #define			AT_XDMAC_CC_PROT_SEC		(0x0 << 5)
123 #define			AT_XDMAC_CC_PROT_UNSEC		(0x1 << 5)
124 #define		AT_XDMAC_CC_SWREQ	(0x1 << 6)	/* Channel Software Request Trigger */
125 #define			AT_XDMAC_CC_SWREQ_HWR_CONNECTED	(0x0 << 6)
126 #define			AT_XDMAC_CC_SWREQ_SWR_CONNECTED	(0x1 << 6)
127 #define		AT_XDMAC_CC_MEMSET	(0x1 << 7)	/* Channel Fill Block of memory */
128 #define			AT_XDMAC_CC_MEMSET_NORMAL_MODE	(0x0 << 7)
129 #define			AT_XDMAC_CC_MEMSET_HW_MODE	(0x1 << 7)
130 #define		AT_XDMAC_CC_CSIZE(i)	((0x7 & (i)) << 8)	/* Channel Chunk Size */
131 #define		AT_XDMAC_CC_DWIDTH_OFFSET	11
132 #define		AT_XDMAC_CC_DWIDTH_MASK	(0x3 << AT_XDMAC_CC_DWIDTH_OFFSET)
133 #define		AT_XDMAC_CC_DWIDTH(i)	((0x3 & (i)) << AT_XDMAC_CC_DWIDTH_OFFSET)	/* Channel Data Width */
134 #define			AT_XDMAC_CC_DWIDTH_BYTE		0x0
135 #define			AT_XDMAC_CC_DWIDTH_HALFWORD	0x1
136 #define			AT_XDMAC_CC_DWIDTH_WORD		0x2
137 #define			AT_XDMAC_CC_DWIDTH_DWORD	0x3
138 #define		AT_XDMAC_CC_SIF(i)	((0x1 & (i)) << 13)	/* Channel Source Interface Identifier */
139 #define		AT_XDMAC_CC_DIF(i)	((0x1 & (i)) << 14)	/* Channel Destination Interface Identifier */
140 #define		AT_XDMAC_CC_SAM_MASK	(0x3 << 16)	/* Channel Source Addressing Mode */
141 #define			AT_XDMAC_CC_SAM_FIXED_AM	(0x0 << 16)
142 #define			AT_XDMAC_CC_SAM_INCREMENTED_AM	(0x1 << 16)
143 #define			AT_XDMAC_CC_SAM_UBS_AM		(0x2 << 16)
144 #define			AT_XDMAC_CC_SAM_UBS_DS_AM	(0x3 << 16)
145 #define		AT_XDMAC_CC_DAM_MASK	(0x3 << 18)	/* Channel Source Addressing Mode */
146 #define			AT_XDMAC_CC_DAM_FIXED_AM	(0x0 << 18)
147 #define			AT_XDMAC_CC_DAM_INCREMENTED_AM	(0x1 << 18)
148 #define			AT_XDMAC_CC_DAM_UBS_AM		(0x2 << 18)
149 #define			AT_XDMAC_CC_DAM_UBS_DS_AM	(0x3 << 18)
150 #define		AT_XDMAC_CC_INITD	(0x1 << 21)	/* Channel Initialization Terminated (read only) */
151 #define			AT_XDMAC_CC_INITD_TERMINATED	(0x0 << 21)
152 #define			AT_XDMAC_CC_INITD_IN_PROGRESS	(0x1 << 21)
153 #define		AT_XDMAC_CC_RDIP	(0x1 << 22)	/* Read in Progress (read only) */
154 #define			AT_XDMAC_CC_RDIP_DONE		(0x0 << 22)
155 #define			AT_XDMAC_CC_RDIP_IN_PROGRESS	(0x1 << 22)
156 #define		AT_XDMAC_CC_WRIP	(0x1 << 23)	/* Write in Progress (read only) */
157 #define			AT_XDMAC_CC_WRIP_DONE		(0x0 << 23)
158 #define			AT_XDMAC_CC_WRIP_IN_PROGRESS	(0x1 << 23)
159 #define		AT_XDMAC_CC_PERID(i)	(0x7f & (i) << 24)	/* Channel Peripheral Identifier */
160 #define AT_XDMAC_CDS_MSP	0x2C	/* Channel Data Stride Memory Set Pattern */
161 #define AT_XDMAC_CSUS		0x30	/* Channel Source Microblock Stride */
162 #define AT_XDMAC_CDUS		0x34	/* Channel Destination Microblock Stride */
163 
164 #define AT_XDMAC_CHAN_REG_BASE	0x50	/* Channel registers base address */
165 
166 /* Microblock control members */
167 #define AT_XDMAC_MBR_UBC_UBLEN_MAX	0xFFFFFFUL	/* Maximum Microblock Length */
168 #define AT_XDMAC_MBR_UBC_NDE		(0x1 << 24)	/* Next Descriptor Enable */
169 #define AT_XDMAC_MBR_UBC_NSEN		(0x1 << 25)	/* Next Descriptor Source Update */
170 #define AT_XDMAC_MBR_UBC_NDEN		(0x1 << 26)	/* Next Descriptor Destination Update */
171 #define AT_XDMAC_MBR_UBC_NDV0		(0x0 << 27)	/* Next Descriptor View 0 */
172 #define AT_XDMAC_MBR_UBC_NDV1		(0x1 << 27)	/* Next Descriptor View 1 */
173 #define AT_XDMAC_MBR_UBC_NDV2		(0x2 << 27)	/* Next Descriptor View 2 */
174 #define AT_XDMAC_MBR_UBC_NDV3		(0x3 << 27)	/* Next Descriptor View 3 */
175 
176 #define AT_XDMAC_MAX_CHAN	0x20
177 #define AT_XDMAC_MAX_CSIZE	16	/* 16 data */
178 #define AT_XDMAC_MAX_DWIDTH	8	/* 64 bits */
179 #define AT_XDMAC_RESIDUE_MAX_RETRIES	5
180 
181 #define AT_XDMAC_DMA_BUSWIDTHS\
182 	(BIT(DMA_SLAVE_BUSWIDTH_UNDEFINED) |\
183 	BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |\
184 	BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |\
185 	BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |\
186 	BIT(DMA_SLAVE_BUSWIDTH_8_BYTES))
187 
188 enum atc_status {
189 	AT_XDMAC_CHAN_IS_CYCLIC = 0,
190 	AT_XDMAC_CHAN_IS_PAUSED,
191 };
192 
193 /* ----- Channels ----- */
194 struct at_xdmac_chan {
195 	struct dma_chan			chan;
196 	void __iomem			*ch_regs;
197 	u32				mask;		/* Channel Mask */
198 	u32				cfg;		/* Channel Configuration Register */
199 	u8				perid;		/* Peripheral ID */
200 	u8				perif;		/* Peripheral Interface */
201 	u8				memif;		/* Memory Interface */
202 	u32				save_cc;
203 	u32				save_cim;
204 	u32				save_cnda;
205 	u32				save_cndc;
206 	unsigned long			status;
207 	struct tasklet_struct		tasklet;
208 	struct dma_slave_config		sconfig;
209 
210 	spinlock_t			lock;
211 
212 	struct list_head		xfers_list;
213 	struct list_head		free_descs_list;
214 };
215 
216 
217 /* ----- Controller ----- */
218 struct at_xdmac {
219 	struct dma_device	dma;
220 	void __iomem		*regs;
221 	int			irq;
222 	struct clk		*clk;
223 	u32			save_gim;
224 	struct dma_pool		*at_xdmac_desc_pool;
225 	struct at_xdmac_chan	chan[0];
226 };
227 
228 
229 /* ----- Descriptors ----- */
230 
231 /* Linked List Descriptor */
232 struct at_xdmac_lld {
233 	dma_addr_t	mbr_nda;	/* Next Descriptor Member */
234 	u32		mbr_ubc;	/* Microblock Control Member */
235 	dma_addr_t	mbr_sa;		/* Source Address Member */
236 	dma_addr_t	mbr_da;		/* Destination Address Member */
237 	u32		mbr_cfg;	/* Configuration Register */
238 	u32		mbr_bc;		/* Block Control Register */
239 	u32		mbr_ds;		/* Data Stride Register */
240 	u32		mbr_sus;	/* Source Microblock Stride Register */
241 	u32		mbr_dus;	/* Destination Microblock Stride Register */
242 };
243 
244 /* 64-bit alignment needed to update CNDA and CUBC registers in an atomic way. */
245 struct at_xdmac_desc {
246 	struct at_xdmac_lld		lld;
247 	enum dma_transfer_direction	direction;
248 	struct dma_async_tx_descriptor	tx_dma_desc;
249 	struct list_head		desc_node;
250 	/* Following members are only used by the first descriptor */
251 	bool				active_xfer;
252 	unsigned int			xfer_size;
253 	struct list_head		descs_list;
254 	struct list_head		xfer_node;
255 } __aligned(sizeof(u64));
256 
257 static inline void __iomem *at_xdmac_chan_reg_base(struct at_xdmac *atxdmac, unsigned int chan_nb)
258 {
259 	return atxdmac->regs + (AT_XDMAC_CHAN_REG_BASE + chan_nb * 0x40);
260 }
261 
262 #define at_xdmac_read(atxdmac, reg) readl_relaxed((atxdmac)->regs + (reg))
263 #define at_xdmac_write(atxdmac, reg, value) \
264 	writel_relaxed((value), (atxdmac)->regs + (reg))
265 
266 #define at_xdmac_chan_read(atchan, reg) readl_relaxed((atchan)->ch_regs + (reg))
267 #define at_xdmac_chan_write(atchan, reg, value) writel_relaxed((value), (atchan)->ch_regs + (reg))
268 
269 static inline struct at_xdmac_chan *to_at_xdmac_chan(struct dma_chan *dchan)
270 {
271 	return container_of(dchan, struct at_xdmac_chan, chan);
272 }
273 
274 static struct device *chan2dev(struct dma_chan *chan)
275 {
276 	return &chan->dev->device;
277 }
278 
279 static inline struct at_xdmac *to_at_xdmac(struct dma_device *ddev)
280 {
281 	return container_of(ddev, struct at_xdmac, dma);
282 }
283 
284 static inline struct at_xdmac_desc *txd_to_at_desc(struct dma_async_tx_descriptor *txd)
285 {
286 	return container_of(txd, struct at_xdmac_desc, tx_dma_desc);
287 }
288 
289 static inline int at_xdmac_chan_is_cyclic(struct at_xdmac_chan *atchan)
290 {
291 	return test_bit(AT_XDMAC_CHAN_IS_CYCLIC, &atchan->status);
292 }
293 
294 static inline int at_xdmac_chan_is_paused(struct at_xdmac_chan *atchan)
295 {
296 	return test_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status);
297 }
298 
299 static inline int at_xdmac_csize(u32 maxburst)
300 {
301 	int csize;
302 
303 	csize = ffs(maxburst) - 1;
304 	if (csize > 4)
305 		csize = -EINVAL;
306 
307 	return csize;
308 };
309 
310 static inline u8 at_xdmac_get_dwidth(u32 cfg)
311 {
312 	return (cfg & AT_XDMAC_CC_DWIDTH_MASK) >> AT_XDMAC_CC_DWIDTH_OFFSET;
313 };
314 
315 static unsigned int init_nr_desc_per_channel = 64;
316 module_param(init_nr_desc_per_channel, uint, 0644);
317 MODULE_PARM_DESC(init_nr_desc_per_channel,
318 		 "initial descriptors per channel (default: 64)");
319 
320 
321 static bool at_xdmac_chan_is_enabled(struct at_xdmac_chan *atchan)
322 {
323 	return at_xdmac_chan_read(atchan, AT_XDMAC_GS) & atchan->mask;
324 }
325 
326 static void at_xdmac_off(struct at_xdmac *atxdmac)
327 {
328 	at_xdmac_write(atxdmac, AT_XDMAC_GD, -1L);
329 
330 	/* Wait that all chans are disabled. */
331 	while (at_xdmac_read(atxdmac, AT_XDMAC_GS))
332 		cpu_relax();
333 
334 	at_xdmac_write(atxdmac, AT_XDMAC_GID, -1L);
335 }
336 
337 /* Call with lock hold. */
338 static void at_xdmac_start_xfer(struct at_xdmac_chan *atchan,
339 				struct at_xdmac_desc *first)
340 {
341 	struct at_xdmac	*atxdmac = to_at_xdmac(atchan->chan.device);
342 	u32		reg;
343 
344 	dev_vdbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, first);
345 
346 	if (at_xdmac_chan_is_enabled(atchan))
347 		return;
348 
349 	/* Set transfer as active to not try to start it again. */
350 	first->active_xfer = true;
351 
352 	/* Tell xdmac where to get the first descriptor. */
353 	reg = AT_XDMAC_CNDA_NDA(first->tx_dma_desc.phys)
354 	      | AT_XDMAC_CNDA_NDAIF(atchan->memif);
355 	at_xdmac_chan_write(atchan, AT_XDMAC_CNDA, reg);
356 
357 	/*
358 	 * When doing non cyclic transfer we need to use the next
359 	 * descriptor view 2 since some fields of the configuration register
360 	 * depend on transfer size and src/dest addresses.
361 	 */
362 	if (at_xdmac_chan_is_cyclic(atchan))
363 		reg = AT_XDMAC_CNDC_NDVIEW_NDV1;
364 	else if (first->lld.mbr_ubc & AT_XDMAC_MBR_UBC_NDV3)
365 		reg = AT_XDMAC_CNDC_NDVIEW_NDV3;
366 	else
367 		reg = AT_XDMAC_CNDC_NDVIEW_NDV2;
368 	/*
369 	 * Even if the register will be updated from the configuration in the
370 	 * descriptor when using view 2 or higher, the PROT bit won't be set
371 	 * properly. This bit can be modified only by using the channel
372 	 * configuration register.
373 	 */
374 	at_xdmac_chan_write(atchan, AT_XDMAC_CC, first->lld.mbr_cfg);
375 
376 	reg |= AT_XDMAC_CNDC_NDDUP
377 	       | AT_XDMAC_CNDC_NDSUP
378 	       | AT_XDMAC_CNDC_NDE;
379 	at_xdmac_chan_write(atchan, AT_XDMAC_CNDC, reg);
380 
381 	dev_vdbg(chan2dev(&atchan->chan),
382 		 "%s: CC=0x%08x CNDA=0x%08x, CNDC=0x%08x, CSA=0x%08x, CDA=0x%08x, CUBC=0x%08x\n",
383 		 __func__, at_xdmac_chan_read(atchan, AT_XDMAC_CC),
384 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDA),
385 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDC),
386 		 at_xdmac_chan_read(atchan, AT_XDMAC_CSA),
387 		 at_xdmac_chan_read(atchan, AT_XDMAC_CDA),
388 		 at_xdmac_chan_read(atchan, AT_XDMAC_CUBC));
389 
390 	at_xdmac_chan_write(atchan, AT_XDMAC_CID, 0xffffffff);
391 	reg = AT_XDMAC_CIE_RBEIE | AT_XDMAC_CIE_WBEIE | AT_XDMAC_CIE_ROIE;
392 	/*
393 	 * There is no end of list when doing cyclic dma, we need to get
394 	 * an interrupt after each periods.
395 	 */
396 	if (at_xdmac_chan_is_cyclic(atchan))
397 		at_xdmac_chan_write(atchan, AT_XDMAC_CIE,
398 				    reg | AT_XDMAC_CIE_BIE);
399 	else
400 		at_xdmac_chan_write(atchan, AT_XDMAC_CIE,
401 				    reg | AT_XDMAC_CIE_LIE);
402 	at_xdmac_write(atxdmac, AT_XDMAC_GIE, atchan->mask);
403 	dev_vdbg(chan2dev(&atchan->chan),
404 		 "%s: enable channel (0x%08x)\n", __func__, atchan->mask);
405 	wmb();
406 	at_xdmac_write(atxdmac, AT_XDMAC_GE, atchan->mask);
407 
408 	dev_vdbg(chan2dev(&atchan->chan),
409 		 "%s: CC=0x%08x CNDA=0x%08x, CNDC=0x%08x, CSA=0x%08x, CDA=0x%08x, CUBC=0x%08x\n",
410 		 __func__, at_xdmac_chan_read(atchan, AT_XDMAC_CC),
411 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDA),
412 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDC),
413 		 at_xdmac_chan_read(atchan, AT_XDMAC_CSA),
414 		 at_xdmac_chan_read(atchan, AT_XDMAC_CDA),
415 		 at_xdmac_chan_read(atchan, AT_XDMAC_CUBC));
416 
417 }
418 
419 static dma_cookie_t at_xdmac_tx_submit(struct dma_async_tx_descriptor *tx)
420 {
421 	struct at_xdmac_desc	*desc = txd_to_at_desc(tx);
422 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(tx->chan);
423 	dma_cookie_t		cookie;
424 	unsigned long		irqflags;
425 
426 	spin_lock_irqsave(&atchan->lock, irqflags);
427 	cookie = dma_cookie_assign(tx);
428 
429 	dev_vdbg(chan2dev(tx->chan), "%s: atchan 0x%p, add desc 0x%p to xfers_list\n",
430 		 __func__, atchan, desc);
431 	list_add_tail(&desc->xfer_node, &atchan->xfers_list);
432 	if (list_is_singular(&atchan->xfers_list))
433 		at_xdmac_start_xfer(atchan, desc);
434 
435 	spin_unlock_irqrestore(&atchan->lock, irqflags);
436 	return cookie;
437 }
438 
439 static struct at_xdmac_desc *at_xdmac_alloc_desc(struct dma_chan *chan,
440 						 gfp_t gfp_flags)
441 {
442 	struct at_xdmac_desc	*desc;
443 	struct at_xdmac		*atxdmac = to_at_xdmac(chan->device);
444 	dma_addr_t		phys;
445 
446 	desc = dma_pool_zalloc(atxdmac->at_xdmac_desc_pool, gfp_flags, &phys);
447 	if (desc) {
448 		INIT_LIST_HEAD(&desc->descs_list);
449 		dma_async_tx_descriptor_init(&desc->tx_dma_desc, chan);
450 		desc->tx_dma_desc.tx_submit = at_xdmac_tx_submit;
451 		desc->tx_dma_desc.phys = phys;
452 	}
453 
454 	return desc;
455 }
456 
457 static void at_xdmac_init_used_desc(struct at_xdmac_desc *desc)
458 {
459 	memset(&desc->lld, 0, sizeof(desc->lld));
460 	INIT_LIST_HEAD(&desc->descs_list);
461 	desc->direction = DMA_TRANS_NONE;
462 	desc->xfer_size = 0;
463 	desc->active_xfer = false;
464 }
465 
466 /* Call must be protected by lock. */
467 static struct at_xdmac_desc *at_xdmac_get_desc(struct at_xdmac_chan *atchan)
468 {
469 	struct at_xdmac_desc *desc;
470 
471 	if (list_empty(&atchan->free_descs_list)) {
472 		desc = at_xdmac_alloc_desc(&atchan->chan, GFP_NOWAIT);
473 	} else {
474 		desc = list_first_entry(&atchan->free_descs_list,
475 					struct at_xdmac_desc, desc_node);
476 		list_del(&desc->desc_node);
477 		at_xdmac_init_used_desc(desc);
478 	}
479 
480 	return desc;
481 }
482 
483 static void at_xdmac_queue_desc(struct dma_chan *chan,
484 				struct at_xdmac_desc *prev,
485 				struct at_xdmac_desc *desc)
486 {
487 	if (!prev || !desc)
488 		return;
489 
490 	prev->lld.mbr_nda = desc->tx_dma_desc.phys;
491 	prev->lld.mbr_ubc |= AT_XDMAC_MBR_UBC_NDE;
492 
493 	dev_dbg(chan2dev(chan),	"%s: chain lld: prev=0x%p, mbr_nda=%pad\n",
494 		__func__, prev, &prev->lld.mbr_nda);
495 }
496 
497 static inline void at_xdmac_increment_block_count(struct dma_chan *chan,
498 						  struct at_xdmac_desc *desc)
499 {
500 	if (!desc)
501 		return;
502 
503 	desc->lld.mbr_bc++;
504 
505 	dev_dbg(chan2dev(chan),
506 		"%s: incrementing the block count of the desc 0x%p\n",
507 		__func__, desc);
508 }
509 
510 static struct dma_chan *at_xdmac_xlate(struct of_phandle_args *dma_spec,
511 				       struct of_dma *of_dma)
512 {
513 	struct at_xdmac		*atxdmac = of_dma->of_dma_data;
514 	struct at_xdmac_chan	*atchan;
515 	struct dma_chan		*chan;
516 	struct device		*dev = atxdmac->dma.dev;
517 
518 	if (dma_spec->args_count != 1) {
519 		dev_err(dev, "dma phandler args: bad number of args\n");
520 		return NULL;
521 	}
522 
523 	chan = dma_get_any_slave_channel(&atxdmac->dma);
524 	if (!chan) {
525 		dev_err(dev, "can't get a dma channel\n");
526 		return NULL;
527 	}
528 
529 	atchan = to_at_xdmac_chan(chan);
530 	atchan->memif = AT91_XDMAC_DT_GET_MEM_IF(dma_spec->args[0]);
531 	atchan->perif = AT91_XDMAC_DT_GET_PER_IF(dma_spec->args[0]);
532 	atchan->perid = AT91_XDMAC_DT_GET_PERID(dma_spec->args[0]);
533 	dev_dbg(dev, "chan dt cfg: memif=%u perif=%u perid=%u\n",
534 		 atchan->memif, atchan->perif, atchan->perid);
535 
536 	return chan;
537 }
538 
539 static int at_xdmac_compute_chan_conf(struct dma_chan *chan,
540 				      enum dma_transfer_direction direction)
541 {
542 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
543 	int			csize, dwidth;
544 
545 	if (direction == DMA_DEV_TO_MEM) {
546 		atchan->cfg =
547 			AT91_XDMAC_DT_PERID(atchan->perid)
548 			| AT_XDMAC_CC_DAM_INCREMENTED_AM
549 			| AT_XDMAC_CC_SAM_FIXED_AM
550 			| AT_XDMAC_CC_DIF(atchan->memif)
551 			| AT_XDMAC_CC_SIF(atchan->perif)
552 			| AT_XDMAC_CC_SWREQ_HWR_CONNECTED
553 			| AT_XDMAC_CC_DSYNC_PER2MEM
554 			| AT_XDMAC_CC_MBSIZE_SIXTEEN
555 			| AT_XDMAC_CC_TYPE_PER_TRAN;
556 		csize = ffs(atchan->sconfig.src_maxburst) - 1;
557 		if (csize < 0) {
558 			dev_err(chan2dev(chan), "invalid src maxburst value\n");
559 			return -EINVAL;
560 		}
561 		atchan->cfg |= AT_XDMAC_CC_CSIZE(csize);
562 		dwidth = ffs(atchan->sconfig.src_addr_width) - 1;
563 		if (dwidth < 0) {
564 			dev_err(chan2dev(chan), "invalid src addr width value\n");
565 			return -EINVAL;
566 		}
567 		atchan->cfg |= AT_XDMAC_CC_DWIDTH(dwidth);
568 	} else if (direction == DMA_MEM_TO_DEV) {
569 		atchan->cfg =
570 			AT91_XDMAC_DT_PERID(atchan->perid)
571 			| AT_XDMAC_CC_DAM_FIXED_AM
572 			| AT_XDMAC_CC_SAM_INCREMENTED_AM
573 			| AT_XDMAC_CC_DIF(atchan->perif)
574 			| AT_XDMAC_CC_SIF(atchan->memif)
575 			| AT_XDMAC_CC_SWREQ_HWR_CONNECTED
576 			| AT_XDMAC_CC_DSYNC_MEM2PER
577 			| AT_XDMAC_CC_MBSIZE_SIXTEEN
578 			| AT_XDMAC_CC_TYPE_PER_TRAN;
579 		csize = ffs(atchan->sconfig.dst_maxburst) - 1;
580 		if (csize < 0) {
581 			dev_err(chan2dev(chan), "invalid src maxburst value\n");
582 			return -EINVAL;
583 		}
584 		atchan->cfg |= AT_XDMAC_CC_CSIZE(csize);
585 		dwidth = ffs(atchan->sconfig.dst_addr_width) - 1;
586 		if (dwidth < 0) {
587 			dev_err(chan2dev(chan), "invalid dst addr width value\n");
588 			return -EINVAL;
589 		}
590 		atchan->cfg |= AT_XDMAC_CC_DWIDTH(dwidth);
591 	}
592 
593 	dev_dbg(chan2dev(chan),	"%s: cfg=0x%08x\n", __func__, atchan->cfg);
594 
595 	return 0;
596 }
597 
598 /*
599  * Only check that maxburst and addr width values are supported by the
600  * the controller but not that the configuration is good to perform the
601  * transfer since we don't know the direction at this stage.
602  */
603 static int at_xdmac_check_slave_config(struct dma_slave_config *sconfig)
604 {
605 	if ((sconfig->src_maxburst > AT_XDMAC_MAX_CSIZE)
606 	    || (sconfig->dst_maxburst > AT_XDMAC_MAX_CSIZE))
607 		return -EINVAL;
608 
609 	if ((sconfig->src_addr_width > AT_XDMAC_MAX_DWIDTH)
610 	    || (sconfig->dst_addr_width > AT_XDMAC_MAX_DWIDTH))
611 		return -EINVAL;
612 
613 	return 0;
614 }
615 
616 static int at_xdmac_set_slave_config(struct dma_chan *chan,
617 				      struct dma_slave_config *sconfig)
618 {
619 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
620 
621 	if (at_xdmac_check_slave_config(sconfig)) {
622 		dev_err(chan2dev(chan), "invalid slave configuration\n");
623 		return -EINVAL;
624 	}
625 
626 	memcpy(&atchan->sconfig, sconfig, sizeof(atchan->sconfig));
627 
628 	return 0;
629 }
630 
631 static struct dma_async_tx_descriptor *
632 at_xdmac_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
633 		       unsigned int sg_len, enum dma_transfer_direction direction,
634 		       unsigned long flags, void *context)
635 {
636 	struct at_xdmac_chan		*atchan = to_at_xdmac_chan(chan);
637 	struct at_xdmac_desc		*first = NULL, *prev = NULL;
638 	struct scatterlist		*sg;
639 	int				i;
640 	unsigned int			xfer_size = 0;
641 	unsigned long			irqflags;
642 	struct dma_async_tx_descriptor	*ret = NULL;
643 
644 	if (!sgl)
645 		return NULL;
646 
647 	if (!is_slave_direction(direction)) {
648 		dev_err(chan2dev(chan), "invalid DMA direction\n");
649 		return NULL;
650 	}
651 
652 	dev_dbg(chan2dev(chan), "%s: sg_len=%d, dir=%s, flags=0x%lx\n",
653 		 __func__, sg_len,
654 		 direction == DMA_MEM_TO_DEV ? "to device" : "from device",
655 		 flags);
656 
657 	/* Protect dma_sconfig field that can be modified by set_slave_conf. */
658 	spin_lock_irqsave(&atchan->lock, irqflags);
659 
660 	if (at_xdmac_compute_chan_conf(chan, direction))
661 		goto spin_unlock;
662 
663 	/* Prepare descriptors. */
664 	for_each_sg(sgl, sg, sg_len, i) {
665 		struct at_xdmac_desc	*desc = NULL;
666 		u32			len, mem, dwidth, fixed_dwidth;
667 
668 		len = sg_dma_len(sg);
669 		mem = sg_dma_address(sg);
670 		if (unlikely(!len)) {
671 			dev_err(chan2dev(chan), "sg data length is zero\n");
672 			goto spin_unlock;
673 		}
674 		dev_dbg(chan2dev(chan), "%s: * sg%d len=%u, mem=0x%08x\n",
675 			 __func__, i, len, mem);
676 
677 		desc = at_xdmac_get_desc(atchan);
678 		if (!desc) {
679 			dev_err(chan2dev(chan), "can't get descriptor\n");
680 			if (first)
681 				list_splice_init(&first->descs_list, &atchan->free_descs_list);
682 			goto spin_unlock;
683 		}
684 
685 		/* Linked list descriptor setup. */
686 		if (direction == DMA_DEV_TO_MEM) {
687 			desc->lld.mbr_sa = atchan->sconfig.src_addr;
688 			desc->lld.mbr_da = mem;
689 		} else {
690 			desc->lld.mbr_sa = mem;
691 			desc->lld.mbr_da = atchan->sconfig.dst_addr;
692 		}
693 		dwidth = at_xdmac_get_dwidth(atchan->cfg);
694 		fixed_dwidth = IS_ALIGNED(len, 1 << dwidth)
695 			       ? dwidth
696 			       : AT_XDMAC_CC_DWIDTH_BYTE;
697 		desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV2			/* next descriptor view */
698 			| AT_XDMAC_MBR_UBC_NDEN					/* next descriptor dst parameter update */
699 			| AT_XDMAC_MBR_UBC_NSEN					/* next descriptor src parameter update */
700 			| (len >> fixed_dwidth);				/* microblock length */
701 		desc->lld.mbr_cfg = (atchan->cfg & ~AT_XDMAC_CC_DWIDTH_MASK) |
702 				    AT_XDMAC_CC_DWIDTH(fixed_dwidth);
703 		dev_dbg(chan2dev(chan),
704 			 "%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x\n",
705 			 __func__, &desc->lld.mbr_sa, &desc->lld.mbr_da, desc->lld.mbr_ubc);
706 
707 		/* Chain lld. */
708 		if (prev)
709 			at_xdmac_queue_desc(chan, prev, desc);
710 
711 		prev = desc;
712 		if (!first)
713 			first = desc;
714 
715 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
716 			 __func__, desc, first);
717 		list_add_tail(&desc->desc_node, &first->descs_list);
718 		xfer_size += len;
719 	}
720 
721 
722 	first->tx_dma_desc.flags = flags;
723 	first->xfer_size = xfer_size;
724 	first->direction = direction;
725 	ret = &first->tx_dma_desc;
726 
727 spin_unlock:
728 	spin_unlock_irqrestore(&atchan->lock, irqflags);
729 	return ret;
730 }
731 
732 static struct dma_async_tx_descriptor *
733 at_xdmac_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
734 			 size_t buf_len, size_t period_len,
735 			 enum dma_transfer_direction direction,
736 			 unsigned long flags)
737 {
738 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
739 	struct at_xdmac_desc	*first = NULL, *prev = NULL;
740 	unsigned int		periods = buf_len / period_len;
741 	int			i;
742 	unsigned long		irqflags;
743 
744 	dev_dbg(chan2dev(chan), "%s: buf_addr=%pad, buf_len=%zd, period_len=%zd, dir=%s, flags=0x%lx\n",
745 		__func__, &buf_addr, buf_len, period_len,
746 		direction == DMA_MEM_TO_DEV ? "mem2per" : "per2mem", flags);
747 
748 	if (!is_slave_direction(direction)) {
749 		dev_err(chan2dev(chan), "invalid DMA direction\n");
750 		return NULL;
751 	}
752 
753 	if (test_and_set_bit(AT_XDMAC_CHAN_IS_CYCLIC, &atchan->status)) {
754 		dev_err(chan2dev(chan), "channel currently used\n");
755 		return NULL;
756 	}
757 
758 	if (at_xdmac_compute_chan_conf(chan, direction))
759 		return NULL;
760 
761 	for (i = 0; i < periods; i++) {
762 		struct at_xdmac_desc	*desc = NULL;
763 
764 		spin_lock_irqsave(&atchan->lock, irqflags);
765 		desc = at_xdmac_get_desc(atchan);
766 		if (!desc) {
767 			dev_err(chan2dev(chan), "can't get descriptor\n");
768 			if (first)
769 				list_splice_init(&first->descs_list, &atchan->free_descs_list);
770 			spin_unlock_irqrestore(&atchan->lock, irqflags);
771 			return NULL;
772 		}
773 		spin_unlock_irqrestore(&atchan->lock, irqflags);
774 		dev_dbg(chan2dev(chan),
775 			"%s: desc=0x%p, tx_dma_desc.phys=%pad\n",
776 			__func__, desc, &desc->tx_dma_desc.phys);
777 
778 		if (direction == DMA_DEV_TO_MEM) {
779 			desc->lld.mbr_sa = atchan->sconfig.src_addr;
780 			desc->lld.mbr_da = buf_addr + i * period_len;
781 		} else {
782 			desc->lld.mbr_sa = buf_addr + i * period_len;
783 			desc->lld.mbr_da = atchan->sconfig.dst_addr;
784 		}
785 		desc->lld.mbr_cfg = atchan->cfg;
786 		desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV1
787 			| AT_XDMAC_MBR_UBC_NDEN
788 			| AT_XDMAC_MBR_UBC_NSEN
789 			| period_len >> at_xdmac_get_dwidth(desc->lld.mbr_cfg);
790 
791 		dev_dbg(chan2dev(chan),
792 			 "%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x\n",
793 			 __func__, &desc->lld.mbr_sa, &desc->lld.mbr_da, desc->lld.mbr_ubc);
794 
795 		/* Chain lld. */
796 		if (prev)
797 			at_xdmac_queue_desc(chan, prev, desc);
798 
799 		prev = desc;
800 		if (!first)
801 			first = desc;
802 
803 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
804 			 __func__, desc, first);
805 		list_add_tail(&desc->desc_node, &first->descs_list);
806 	}
807 
808 	at_xdmac_queue_desc(chan, prev, first);
809 	first->tx_dma_desc.flags = flags;
810 	first->xfer_size = buf_len;
811 	first->direction = direction;
812 
813 	return &first->tx_dma_desc;
814 }
815 
816 static inline u32 at_xdmac_align_width(struct dma_chan *chan, dma_addr_t addr)
817 {
818 	u32 width;
819 
820 	/*
821 	 * Check address alignment to select the greater data width we
822 	 * can use.
823 	 *
824 	 * Some XDMAC implementations don't provide dword transfer, in
825 	 * this case selecting dword has the same behavior as
826 	 * selecting word transfers.
827 	 */
828 	if (!(addr & 7)) {
829 		width = AT_XDMAC_CC_DWIDTH_DWORD;
830 		dev_dbg(chan2dev(chan), "%s: dwidth: double word\n", __func__);
831 	} else if (!(addr & 3)) {
832 		width = AT_XDMAC_CC_DWIDTH_WORD;
833 		dev_dbg(chan2dev(chan), "%s: dwidth: word\n", __func__);
834 	} else if (!(addr & 1)) {
835 		width = AT_XDMAC_CC_DWIDTH_HALFWORD;
836 		dev_dbg(chan2dev(chan), "%s: dwidth: half word\n", __func__);
837 	} else {
838 		width = AT_XDMAC_CC_DWIDTH_BYTE;
839 		dev_dbg(chan2dev(chan), "%s: dwidth: byte\n", __func__);
840 	}
841 
842 	return width;
843 }
844 
845 static struct at_xdmac_desc *
846 at_xdmac_interleaved_queue_desc(struct dma_chan *chan,
847 				struct at_xdmac_chan *atchan,
848 				struct at_xdmac_desc *prev,
849 				dma_addr_t src, dma_addr_t dst,
850 				struct dma_interleaved_template *xt,
851 				struct data_chunk *chunk)
852 {
853 	struct at_xdmac_desc	*desc;
854 	u32			dwidth;
855 	unsigned long		flags;
856 	size_t			ublen;
857 	/*
858 	 * WARNING: The channel configuration is set here since there is no
859 	 * dmaengine_slave_config call in this case. Moreover we don't know the
860 	 * direction, it involves we can't dynamically set the source and dest
861 	 * interface so we have to use the same one. Only interface 0 allows EBI
862 	 * access. Hopefully we can access DDR through both ports (at least on
863 	 * SAMA5D4x), so we can use the same interface for source and dest,
864 	 * that solves the fact we don't know the direction.
865 	 * ERRATA: Even if useless for memory transfers, the PERID has to not
866 	 * match the one of another channel. If not, it could lead to spurious
867 	 * flag status.
868 	 */
869 	u32			chan_cc = AT_XDMAC_CC_PERID(0x3f)
870 					| AT_XDMAC_CC_DIF(0)
871 					| AT_XDMAC_CC_SIF(0)
872 					| AT_XDMAC_CC_MBSIZE_SIXTEEN
873 					| AT_XDMAC_CC_TYPE_MEM_TRAN;
874 
875 	dwidth = at_xdmac_align_width(chan, src | dst | chunk->size);
876 	if (chunk->size >= (AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth)) {
877 		dev_dbg(chan2dev(chan),
878 			"%s: chunk too big (%d, max size %lu)...\n",
879 			__func__, chunk->size,
880 			AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth);
881 		return NULL;
882 	}
883 
884 	if (prev)
885 		dev_dbg(chan2dev(chan),
886 			"Adding items at the end of desc 0x%p\n", prev);
887 
888 	if (xt->src_inc) {
889 		if (xt->src_sgl)
890 			chan_cc |=  AT_XDMAC_CC_SAM_UBS_AM;
891 		else
892 			chan_cc |=  AT_XDMAC_CC_SAM_INCREMENTED_AM;
893 	}
894 
895 	if (xt->dst_inc) {
896 		if (xt->dst_sgl)
897 			chan_cc |=  AT_XDMAC_CC_DAM_UBS_AM;
898 		else
899 			chan_cc |=  AT_XDMAC_CC_DAM_INCREMENTED_AM;
900 	}
901 
902 	spin_lock_irqsave(&atchan->lock, flags);
903 	desc = at_xdmac_get_desc(atchan);
904 	spin_unlock_irqrestore(&atchan->lock, flags);
905 	if (!desc) {
906 		dev_err(chan2dev(chan), "can't get descriptor\n");
907 		return NULL;
908 	}
909 
910 	chan_cc |= AT_XDMAC_CC_DWIDTH(dwidth);
911 
912 	ublen = chunk->size >> dwidth;
913 
914 	desc->lld.mbr_sa = src;
915 	desc->lld.mbr_da = dst;
916 	desc->lld.mbr_sus = dmaengine_get_src_icg(xt, chunk);
917 	desc->lld.mbr_dus = dmaengine_get_dst_icg(xt, chunk);
918 
919 	desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV3
920 		| AT_XDMAC_MBR_UBC_NDEN
921 		| AT_XDMAC_MBR_UBC_NSEN
922 		| ublen;
923 	desc->lld.mbr_cfg = chan_cc;
924 
925 	dev_dbg(chan2dev(chan),
926 		"%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x, mbr_cfg=0x%08x\n",
927 		__func__, &desc->lld.mbr_sa, &desc->lld.mbr_da,
928 		desc->lld.mbr_ubc, desc->lld.mbr_cfg);
929 
930 	/* Chain lld. */
931 	if (prev)
932 		at_xdmac_queue_desc(chan, prev, desc);
933 
934 	return desc;
935 }
936 
937 static struct dma_async_tx_descriptor *
938 at_xdmac_prep_interleaved(struct dma_chan *chan,
939 			  struct dma_interleaved_template *xt,
940 			  unsigned long flags)
941 {
942 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
943 	struct at_xdmac_desc	*prev = NULL, *first = NULL;
944 	dma_addr_t		dst_addr, src_addr;
945 	size_t			src_skip = 0, dst_skip = 0, len = 0;
946 	struct data_chunk	*chunk;
947 	int			i;
948 
949 	if (!xt || !xt->numf || (xt->dir != DMA_MEM_TO_MEM))
950 		return NULL;
951 
952 	/*
953 	 * TODO: Handle the case where we have to repeat a chain of
954 	 * descriptors...
955 	 */
956 	if ((xt->numf > 1) && (xt->frame_size > 1))
957 		return NULL;
958 
959 	dev_dbg(chan2dev(chan), "%s: src=%pad, dest=%pad, numf=%d, frame_size=%d, flags=0x%lx\n",
960 		__func__, &xt->src_start, &xt->dst_start,	xt->numf,
961 		xt->frame_size, flags);
962 
963 	src_addr = xt->src_start;
964 	dst_addr = xt->dst_start;
965 
966 	if (xt->numf > 1) {
967 		first = at_xdmac_interleaved_queue_desc(chan, atchan,
968 							NULL,
969 							src_addr, dst_addr,
970 							xt, xt->sgl);
971 
972 		/* Length of the block is (BLEN+1) microblocks. */
973 		for (i = 0; i < xt->numf - 1; i++)
974 			at_xdmac_increment_block_count(chan, first);
975 
976 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
977 			__func__, first, first);
978 		list_add_tail(&first->desc_node, &first->descs_list);
979 	} else {
980 		for (i = 0; i < xt->frame_size; i++) {
981 			size_t src_icg = 0, dst_icg = 0;
982 			struct at_xdmac_desc *desc;
983 
984 			chunk = xt->sgl + i;
985 
986 			dst_icg = dmaengine_get_dst_icg(xt, chunk);
987 			src_icg = dmaengine_get_src_icg(xt, chunk);
988 
989 			src_skip = chunk->size + src_icg;
990 			dst_skip = chunk->size + dst_icg;
991 
992 			dev_dbg(chan2dev(chan),
993 				"%s: chunk size=%d, src icg=%d, dst icg=%d\n",
994 				__func__, chunk->size, src_icg, dst_icg);
995 
996 			desc = at_xdmac_interleaved_queue_desc(chan, atchan,
997 							       prev,
998 							       src_addr, dst_addr,
999 							       xt, chunk);
1000 			if (!desc) {
1001 				list_splice_init(&first->descs_list,
1002 						 &atchan->free_descs_list);
1003 				return NULL;
1004 			}
1005 
1006 			if (!first)
1007 				first = desc;
1008 
1009 			dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
1010 				__func__, desc, first);
1011 			list_add_tail(&desc->desc_node, &first->descs_list);
1012 
1013 			if (xt->src_sgl)
1014 				src_addr += src_skip;
1015 
1016 			if (xt->dst_sgl)
1017 				dst_addr += dst_skip;
1018 
1019 			len += chunk->size;
1020 			prev = desc;
1021 		}
1022 	}
1023 
1024 	first->tx_dma_desc.cookie = -EBUSY;
1025 	first->tx_dma_desc.flags = flags;
1026 	first->xfer_size = len;
1027 
1028 	return &first->tx_dma_desc;
1029 }
1030 
1031 static struct dma_async_tx_descriptor *
1032 at_xdmac_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1033 			 size_t len, unsigned long flags)
1034 {
1035 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1036 	struct at_xdmac_desc	*first = NULL, *prev = NULL;
1037 	size_t			remaining_size = len, xfer_size = 0, ublen;
1038 	dma_addr_t		src_addr = src, dst_addr = dest;
1039 	u32			dwidth;
1040 	/*
1041 	 * WARNING: We don't know the direction, it involves we can't
1042 	 * dynamically set the source and dest interface so we have to use the
1043 	 * same one. Only interface 0 allows EBI access. Hopefully we can
1044 	 * access DDR through both ports (at least on SAMA5D4x), so we can use
1045 	 * the same interface for source and dest, that solves the fact we
1046 	 * don't know the direction.
1047 	 * ERRATA: Even if useless for memory transfers, the PERID has to not
1048 	 * match the one of another channel. If not, it could lead to spurious
1049 	 * flag status.
1050 	 */
1051 	u32			chan_cc = AT_XDMAC_CC_PERID(0x3f)
1052 					| AT_XDMAC_CC_DAM_INCREMENTED_AM
1053 					| AT_XDMAC_CC_SAM_INCREMENTED_AM
1054 					| AT_XDMAC_CC_DIF(0)
1055 					| AT_XDMAC_CC_SIF(0)
1056 					| AT_XDMAC_CC_MBSIZE_SIXTEEN
1057 					| AT_XDMAC_CC_TYPE_MEM_TRAN;
1058 	unsigned long		irqflags;
1059 
1060 	dev_dbg(chan2dev(chan), "%s: src=%pad, dest=%pad, len=%zd, flags=0x%lx\n",
1061 		__func__, &src, &dest, len, flags);
1062 
1063 	if (unlikely(!len))
1064 		return NULL;
1065 
1066 	dwidth = at_xdmac_align_width(chan, src_addr | dst_addr);
1067 
1068 	/* Prepare descriptors. */
1069 	while (remaining_size) {
1070 		struct at_xdmac_desc	*desc = NULL;
1071 
1072 		dev_dbg(chan2dev(chan), "%s: remaining_size=%zu\n", __func__, remaining_size);
1073 
1074 		spin_lock_irqsave(&atchan->lock, irqflags);
1075 		desc = at_xdmac_get_desc(atchan);
1076 		spin_unlock_irqrestore(&atchan->lock, irqflags);
1077 		if (!desc) {
1078 			dev_err(chan2dev(chan), "can't get descriptor\n");
1079 			if (first)
1080 				list_splice_init(&first->descs_list, &atchan->free_descs_list);
1081 			return NULL;
1082 		}
1083 
1084 		/* Update src and dest addresses. */
1085 		src_addr += xfer_size;
1086 		dst_addr += xfer_size;
1087 
1088 		if (remaining_size >= AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth)
1089 			xfer_size = AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth;
1090 		else
1091 			xfer_size = remaining_size;
1092 
1093 		dev_dbg(chan2dev(chan), "%s: xfer_size=%zu\n", __func__, xfer_size);
1094 
1095 		/* Check remaining length and change data width if needed. */
1096 		dwidth = at_xdmac_align_width(chan,
1097 					      src_addr | dst_addr | xfer_size);
1098 		chan_cc &= ~AT_XDMAC_CC_DWIDTH_MASK;
1099 		chan_cc |= AT_XDMAC_CC_DWIDTH(dwidth);
1100 
1101 		ublen = xfer_size >> dwidth;
1102 		remaining_size -= xfer_size;
1103 
1104 		desc->lld.mbr_sa = src_addr;
1105 		desc->lld.mbr_da = dst_addr;
1106 		desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV2
1107 			| AT_XDMAC_MBR_UBC_NDEN
1108 			| AT_XDMAC_MBR_UBC_NSEN
1109 			| ublen;
1110 		desc->lld.mbr_cfg = chan_cc;
1111 
1112 		dev_dbg(chan2dev(chan),
1113 			 "%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x, mbr_cfg=0x%08x\n",
1114 			 __func__, &desc->lld.mbr_sa, &desc->lld.mbr_da, desc->lld.mbr_ubc, desc->lld.mbr_cfg);
1115 
1116 		/* Chain lld. */
1117 		if (prev)
1118 			at_xdmac_queue_desc(chan, prev, desc);
1119 
1120 		prev = desc;
1121 		if (!first)
1122 			first = desc;
1123 
1124 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
1125 			 __func__, desc, first);
1126 		list_add_tail(&desc->desc_node, &first->descs_list);
1127 	}
1128 
1129 	first->tx_dma_desc.flags = flags;
1130 	first->xfer_size = len;
1131 
1132 	return &first->tx_dma_desc;
1133 }
1134 
1135 static struct at_xdmac_desc *at_xdmac_memset_create_desc(struct dma_chan *chan,
1136 							 struct at_xdmac_chan *atchan,
1137 							 dma_addr_t dst_addr,
1138 							 size_t len,
1139 							 int value)
1140 {
1141 	struct at_xdmac_desc	*desc;
1142 	unsigned long		flags;
1143 	size_t			ublen;
1144 	u32			dwidth;
1145 	/*
1146 	 * WARNING: The channel configuration is set here since there is no
1147 	 * dmaengine_slave_config call in this case. Moreover we don't know the
1148 	 * direction, it involves we can't dynamically set the source and dest
1149 	 * interface so we have to use the same one. Only interface 0 allows EBI
1150 	 * access. Hopefully we can access DDR through both ports (at least on
1151 	 * SAMA5D4x), so we can use the same interface for source and dest,
1152 	 * that solves the fact we don't know the direction.
1153 	 * ERRATA: Even if useless for memory transfers, the PERID has to not
1154 	 * match the one of another channel. If not, it could lead to spurious
1155 	 * flag status.
1156 	 */
1157 	u32			chan_cc = AT_XDMAC_CC_PERID(0x3f)
1158 					| AT_XDMAC_CC_DAM_UBS_AM
1159 					| AT_XDMAC_CC_SAM_INCREMENTED_AM
1160 					| AT_XDMAC_CC_DIF(0)
1161 					| AT_XDMAC_CC_SIF(0)
1162 					| AT_XDMAC_CC_MBSIZE_SIXTEEN
1163 					| AT_XDMAC_CC_MEMSET_HW_MODE
1164 					| AT_XDMAC_CC_TYPE_MEM_TRAN;
1165 
1166 	dwidth = at_xdmac_align_width(chan, dst_addr);
1167 
1168 	if (len >= (AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth)) {
1169 		dev_err(chan2dev(chan),
1170 			"%s: Transfer too large, aborting...\n",
1171 			__func__);
1172 		return NULL;
1173 	}
1174 
1175 	spin_lock_irqsave(&atchan->lock, flags);
1176 	desc = at_xdmac_get_desc(atchan);
1177 	spin_unlock_irqrestore(&atchan->lock, flags);
1178 	if (!desc) {
1179 		dev_err(chan2dev(chan), "can't get descriptor\n");
1180 		return NULL;
1181 	}
1182 
1183 	chan_cc |= AT_XDMAC_CC_DWIDTH(dwidth);
1184 
1185 	ublen = len >> dwidth;
1186 
1187 	desc->lld.mbr_da = dst_addr;
1188 	desc->lld.mbr_ds = value;
1189 	desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV3
1190 		| AT_XDMAC_MBR_UBC_NDEN
1191 		| AT_XDMAC_MBR_UBC_NSEN
1192 		| ublen;
1193 	desc->lld.mbr_cfg = chan_cc;
1194 
1195 	dev_dbg(chan2dev(chan),
1196 		"%s: lld: mbr_da=%pad, mbr_ds=0x%08x, mbr_ubc=0x%08x, mbr_cfg=0x%08x\n",
1197 		__func__, &desc->lld.mbr_da, desc->lld.mbr_ds, desc->lld.mbr_ubc,
1198 		desc->lld.mbr_cfg);
1199 
1200 	return desc;
1201 }
1202 
1203 static struct dma_async_tx_descriptor *
1204 at_xdmac_prep_dma_memset(struct dma_chan *chan, dma_addr_t dest, int value,
1205 			 size_t len, unsigned long flags)
1206 {
1207 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1208 	struct at_xdmac_desc	*desc;
1209 
1210 	dev_dbg(chan2dev(chan), "%s: dest=%pad, len=%d, pattern=0x%x, flags=0x%lx\n",
1211 		__func__, &dest, len, value, flags);
1212 
1213 	if (unlikely(!len))
1214 		return NULL;
1215 
1216 	desc = at_xdmac_memset_create_desc(chan, atchan, dest, len, value);
1217 	list_add_tail(&desc->desc_node, &desc->descs_list);
1218 
1219 	desc->tx_dma_desc.cookie = -EBUSY;
1220 	desc->tx_dma_desc.flags = flags;
1221 	desc->xfer_size = len;
1222 
1223 	return &desc->tx_dma_desc;
1224 }
1225 
1226 static struct dma_async_tx_descriptor *
1227 at_xdmac_prep_dma_memset_sg(struct dma_chan *chan, struct scatterlist *sgl,
1228 			    unsigned int sg_len, int value,
1229 			    unsigned long flags)
1230 {
1231 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1232 	struct at_xdmac_desc	*desc, *pdesc = NULL,
1233 				*ppdesc = NULL, *first = NULL;
1234 	struct scatterlist	*sg, *psg = NULL, *ppsg = NULL;
1235 	size_t			stride = 0, pstride = 0, len = 0;
1236 	int			i;
1237 
1238 	if (!sgl)
1239 		return NULL;
1240 
1241 	dev_dbg(chan2dev(chan), "%s: sg_len=%d, value=0x%x, flags=0x%lx\n",
1242 		__func__, sg_len, value, flags);
1243 
1244 	/* Prepare descriptors. */
1245 	for_each_sg(sgl, sg, sg_len, i) {
1246 		dev_dbg(chan2dev(chan), "%s: dest=%pad, len=%d, pattern=0x%x, flags=0x%lx\n",
1247 			__func__, &sg_dma_address(sg), sg_dma_len(sg),
1248 			value, flags);
1249 		desc = at_xdmac_memset_create_desc(chan, atchan,
1250 						   sg_dma_address(sg),
1251 						   sg_dma_len(sg),
1252 						   value);
1253 		if (!desc && first)
1254 			list_splice_init(&first->descs_list,
1255 					 &atchan->free_descs_list);
1256 
1257 		if (!first)
1258 			first = desc;
1259 
1260 		/* Update our strides */
1261 		pstride = stride;
1262 		if (psg)
1263 			stride = sg_dma_address(sg) -
1264 				(sg_dma_address(psg) + sg_dma_len(psg));
1265 
1266 		/*
1267 		 * The scatterlist API gives us only the address and
1268 		 * length of each elements.
1269 		 *
1270 		 * Unfortunately, we don't have the stride, which we
1271 		 * will need to compute.
1272 		 *
1273 		 * That make us end up in a situation like this one:
1274 		 *    len    stride    len    stride    len
1275 		 * +-------+        +-------+        +-------+
1276 		 * |  N-2  |        |  N-1  |        |   N   |
1277 		 * +-------+        +-------+        +-------+
1278 		 *
1279 		 * We need all these three elements (N-2, N-1 and N)
1280 		 * to actually take the decision on whether we need to
1281 		 * queue N-1 or reuse N-2.
1282 		 *
1283 		 * We will only consider N if it is the last element.
1284 		 */
1285 		if (ppdesc && pdesc) {
1286 			if ((stride == pstride) &&
1287 			    (sg_dma_len(ppsg) == sg_dma_len(psg))) {
1288 				dev_dbg(chan2dev(chan),
1289 					"%s: desc 0x%p can be merged with desc 0x%p\n",
1290 					__func__, pdesc, ppdesc);
1291 
1292 				/*
1293 				 * Increment the block count of the
1294 				 * N-2 descriptor
1295 				 */
1296 				at_xdmac_increment_block_count(chan, ppdesc);
1297 				ppdesc->lld.mbr_dus = stride;
1298 
1299 				/*
1300 				 * Put back the N-1 descriptor in the
1301 				 * free descriptor list
1302 				 */
1303 				list_add_tail(&pdesc->desc_node,
1304 					      &atchan->free_descs_list);
1305 
1306 				/*
1307 				 * Make our N-1 descriptor pointer
1308 				 * point to the N-2 since they were
1309 				 * actually merged.
1310 				 */
1311 				pdesc = ppdesc;
1312 
1313 			/*
1314 			 * Rule out the case where we don't have
1315 			 * pstride computed yet (our second sg
1316 			 * element)
1317 			 *
1318 			 * We also want to catch the case where there
1319 			 * would be a negative stride,
1320 			 */
1321 			} else if (pstride ||
1322 				   sg_dma_address(sg) < sg_dma_address(psg)) {
1323 				/*
1324 				 * Queue the N-1 descriptor after the
1325 				 * N-2
1326 				 */
1327 				at_xdmac_queue_desc(chan, ppdesc, pdesc);
1328 
1329 				/*
1330 				 * Add the N-1 descriptor to the list
1331 				 * of the descriptors used for this
1332 				 * transfer
1333 				 */
1334 				list_add_tail(&desc->desc_node,
1335 					      &first->descs_list);
1336 				dev_dbg(chan2dev(chan),
1337 					"%s: add desc 0x%p to descs_list 0x%p\n",
1338 					__func__, desc, first);
1339 			}
1340 		}
1341 
1342 		/*
1343 		 * If we are the last element, just see if we have the
1344 		 * same size than the previous element.
1345 		 *
1346 		 * If so, we can merge it with the previous descriptor
1347 		 * since we don't care about the stride anymore.
1348 		 */
1349 		if ((i == (sg_len - 1)) &&
1350 		    sg_dma_len(psg) == sg_dma_len(sg)) {
1351 			dev_dbg(chan2dev(chan),
1352 				"%s: desc 0x%p can be merged with desc 0x%p\n",
1353 				__func__, desc, pdesc);
1354 
1355 			/*
1356 			 * Increment the block count of the N-1
1357 			 * descriptor
1358 			 */
1359 			at_xdmac_increment_block_count(chan, pdesc);
1360 			pdesc->lld.mbr_dus = stride;
1361 
1362 			/*
1363 			 * Put back the N descriptor in the free
1364 			 * descriptor list
1365 			 */
1366 			list_add_tail(&desc->desc_node,
1367 				      &atchan->free_descs_list);
1368 		}
1369 
1370 		/* Update our descriptors */
1371 		ppdesc = pdesc;
1372 		pdesc = desc;
1373 
1374 		/* Update our scatter pointers */
1375 		ppsg = psg;
1376 		psg = sg;
1377 
1378 		len += sg_dma_len(sg);
1379 	}
1380 
1381 	first->tx_dma_desc.cookie = -EBUSY;
1382 	first->tx_dma_desc.flags = flags;
1383 	first->xfer_size = len;
1384 
1385 	return &first->tx_dma_desc;
1386 }
1387 
1388 static enum dma_status
1389 at_xdmac_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
1390 		struct dma_tx_state *txstate)
1391 {
1392 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1393 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1394 	struct at_xdmac_desc	*desc, *_desc;
1395 	struct list_head	*descs_list;
1396 	enum dma_status		ret;
1397 	int			residue, retry;
1398 	u32			cur_nda, check_nda, cur_ubc, mask, value;
1399 	u8			dwidth = 0;
1400 	unsigned long		flags;
1401 	bool			initd;
1402 
1403 	ret = dma_cookie_status(chan, cookie, txstate);
1404 	if (ret == DMA_COMPLETE)
1405 		return ret;
1406 
1407 	if (!txstate)
1408 		return ret;
1409 
1410 	spin_lock_irqsave(&atchan->lock, flags);
1411 
1412 	desc = list_first_entry(&atchan->xfers_list, struct at_xdmac_desc, xfer_node);
1413 
1414 	/*
1415 	 * If the transfer has not been started yet, don't need to compute the
1416 	 * residue, it's the transfer length.
1417 	 */
1418 	if (!desc->active_xfer) {
1419 		dma_set_residue(txstate, desc->xfer_size);
1420 		goto spin_unlock;
1421 	}
1422 
1423 	residue = desc->xfer_size;
1424 	/*
1425 	 * Flush FIFO: only relevant when the transfer is source peripheral
1426 	 * synchronized. Flush is needed before reading CUBC because data in
1427 	 * the FIFO are not reported by CUBC. Reporting a residue of the
1428 	 * transfer length while we have data in FIFO can cause issue.
1429 	 * Usecase: atmel USART has a timeout which means I have received
1430 	 * characters but there is no more character received for a while. On
1431 	 * timeout, it requests the residue. If the data are in the DMA FIFO,
1432 	 * we will return a residue of the transfer length. It means no data
1433 	 * received. If an application is waiting for these data, it will hang
1434 	 * since we won't have another USART timeout without receiving new
1435 	 * data.
1436 	 */
1437 	mask = AT_XDMAC_CC_TYPE | AT_XDMAC_CC_DSYNC;
1438 	value = AT_XDMAC_CC_TYPE_PER_TRAN | AT_XDMAC_CC_DSYNC_PER2MEM;
1439 	if ((desc->lld.mbr_cfg & mask) == value) {
1440 		at_xdmac_write(atxdmac, AT_XDMAC_GSWF, atchan->mask);
1441 		while (!(at_xdmac_chan_read(atchan, AT_XDMAC_CIS) & AT_XDMAC_CIS_FIS))
1442 			cpu_relax();
1443 	}
1444 
1445 	/*
1446 	 * The easiest way to compute the residue should be to pause the DMA
1447 	 * but doing this can lead to miss some data as some devices don't
1448 	 * have FIFO.
1449 	 * We need to read several registers because:
1450 	 * - DMA is running therefore a descriptor change is possible while
1451 	 * reading these registers
1452 	 * - When the block transfer is done, the value of the CUBC register
1453 	 * is set to its initial value until the fetch of the next descriptor.
1454 	 * This value will corrupt the residue calculation so we have to skip
1455 	 * it.
1456 	 *
1457 	 * INITD --------                    ------------
1458 	 *              |____________________|
1459 	 *       _______________________  _______________
1460 	 * NDA       @desc2             \/   @desc3
1461 	 *       _______________________/\_______________
1462 	 *       __________  ___________  _______________
1463 	 * CUBC       0    \/ MAX desc1 \/  MAX desc2
1464 	 *       __________/\___________/\_______________
1465 	 *
1466 	 * Since descriptors are aligned on 64 bits, we can assume that
1467 	 * the update of NDA and CUBC is atomic.
1468 	 * Memory barriers are used to ensure the read order of the registers.
1469 	 * A max number of retries is set because unlikely it could never ends.
1470 	 */
1471 	for (retry = 0; retry < AT_XDMAC_RESIDUE_MAX_RETRIES; retry++) {
1472 		check_nda = at_xdmac_chan_read(atchan, AT_XDMAC_CNDA) & 0xfffffffc;
1473 		rmb();
1474 		initd = !!(at_xdmac_chan_read(atchan, AT_XDMAC_CC) & AT_XDMAC_CC_INITD);
1475 		rmb();
1476 		cur_ubc = at_xdmac_chan_read(atchan, AT_XDMAC_CUBC);
1477 		rmb();
1478 		cur_nda = at_xdmac_chan_read(atchan, AT_XDMAC_CNDA) & 0xfffffffc;
1479 		rmb();
1480 
1481 		if ((check_nda == cur_nda) && initd)
1482 			break;
1483 	}
1484 
1485 	if (unlikely(retry >= AT_XDMAC_RESIDUE_MAX_RETRIES)) {
1486 		ret = DMA_ERROR;
1487 		goto spin_unlock;
1488 	}
1489 
1490 	/*
1491 	 * Flush FIFO: only relevant when the transfer is source peripheral
1492 	 * synchronized. Another flush is needed here because CUBC is updated
1493 	 * when the controller sends the data write command. It can lead to
1494 	 * report data that are not written in the memory or the device. The
1495 	 * FIFO flush ensures that data are really written.
1496 	 */
1497 	if ((desc->lld.mbr_cfg & mask) == value) {
1498 		at_xdmac_write(atxdmac, AT_XDMAC_GSWF, atchan->mask);
1499 		while (!(at_xdmac_chan_read(atchan, AT_XDMAC_CIS) & AT_XDMAC_CIS_FIS))
1500 			cpu_relax();
1501 	}
1502 
1503 	/*
1504 	 * Remove size of all microblocks already transferred and the current
1505 	 * one. Then add the remaining size to transfer of the current
1506 	 * microblock.
1507 	 */
1508 	descs_list = &desc->descs_list;
1509 	list_for_each_entry_safe(desc, _desc, descs_list, desc_node) {
1510 		dwidth = at_xdmac_get_dwidth(desc->lld.mbr_cfg);
1511 		residue -= (desc->lld.mbr_ubc & 0xffffff) << dwidth;
1512 		if ((desc->lld.mbr_nda & 0xfffffffc) == cur_nda)
1513 			break;
1514 	}
1515 	residue += cur_ubc << dwidth;
1516 
1517 	dma_set_residue(txstate, residue);
1518 
1519 	dev_dbg(chan2dev(chan),
1520 		 "%s: desc=0x%p, tx_dma_desc.phys=%pad, tx_status=%d, cookie=%d, residue=%d\n",
1521 		 __func__, desc, &desc->tx_dma_desc.phys, ret, cookie, residue);
1522 
1523 spin_unlock:
1524 	spin_unlock_irqrestore(&atchan->lock, flags);
1525 	return ret;
1526 }
1527 
1528 /* Call must be protected by lock. */
1529 static void at_xdmac_remove_xfer(struct at_xdmac_chan *atchan,
1530 				    struct at_xdmac_desc *desc)
1531 {
1532 	dev_dbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, desc);
1533 
1534 	/*
1535 	 * Remove the transfer from the transfer list then move the transfer
1536 	 * descriptors into the free descriptors list.
1537 	 */
1538 	list_del(&desc->xfer_node);
1539 	list_splice_init(&desc->descs_list, &atchan->free_descs_list);
1540 }
1541 
1542 static void at_xdmac_advance_work(struct at_xdmac_chan *atchan)
1543 {
1544 	struct at_xdmac_desc	*desc;
1545 	unsigned long		flags;
1546 
1547 	spin_lock_irqsave(&atchan->lock, flags);
1548 
1549 	/*
1550 	 * If channel is enabled, do nothing, advance_work will be triggered
1551 	 * after the interruption.
1552 	 */
1553 	if (!at_xdmac_chan_is_enabled(atchan) && !list_empty(&atchan->xfers_list)) {
1554 		desc = list_first_entry(&atchan->xfers_list,
1555 					struct at_xdmac_desc,
1556 					xfer_node);
1557 		dev_vdbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, desc);
1558 		if (!desc->active_xfer)
1559 			at_xdmac_start_xfer(atchan, desc);
1560 	}
1561 
1562 	spin_unlock_irqrestore(&atchan->lock, flags);
1563 }
1564 
1565 static void at_xdmac_handle_cyclic(struct at_xdmac_chan *atchan)
1566 {
1567 	struct at_xdmac_desc		*desc;
1568 	struct dma_async_tx_descriptor	*txd;
1569 
1570 	desc = list_first_entry(&atchan->xfers_list, struct at_xdmac_desc, xfer_node);
1571 	txd = &desc->tx_dma_desc;
1572 
1573 	if (txd->flags & DMA_PREP_INTERRUPT)
1574 		dmaengine_desc_get_callback_invoke(txd, NULL);
1575 }
1576 
1577 static void at_xdmac_tasklet(unsigned long data)
1578 {
1579 	struct at_xdmac_chan	*atchan = (struct at_xdmac_chan *)data;
1580 	struct at_xdmac_desc	*desc;
1581 	u32			error_mask;
1582 
1583 	dev_dbg(chan2dev(&atchan->chan), "%s: status=0x%08lx\n",
1584 		 __func__, atchan->status);
1585 
1586 	error_mask = AT_XDMAC_CIS_RBEIS
1587 		     | AT_XDMAC_CIS_WBEIS
1588 		     | AT_XDMAC_CIS_ROIS;
1589 
1590 	if (at_xdmac_chan_is_cyclic(atchan)) {
1591 		at_xdmac_handle_cyclic(atchan);
1592 	} else if ((atchan->status & AT_XDMAC_CIS_LIS)
1593 		   || (atchan->status & error_mask)) {
1594 		struct dma_async_tx_descriptor  *txd;
1595 
1596 		if (atchan->status & AT_XDMAC_CIS_RBEIS)
1597 			dev_err(chan2dev(&atchan->chan), "read bus error!!!");
1598 		if (atchan->status & AT_XDMAC_CIS_WBEIS)
1599 			dev_err(chan2dev(&atchan->chan), "write bus error!!!");
1600 		if (atchan->status & AT_XDMAC_CIS_ROIS)
1601 			dev_err(chan2dev(&atchan->chan), "request overflow error!!!");
1602 
1603 		spin_lock_bh(&atchan->lock);
1604 		desc = list_first_entry(&atchan->xfers_list,
1605 					struct at_xdmac_desc,
1606 					xfer_node);
1607 		dev_vdbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, desc);
1608 		BUG_ON(!desc->active_xfer);
1609 
1610 		txd = &desc->tx_dma_desc;
1611 
1612 		at_xdmac_remove_xfer(atchan, desc);
1613 		spin_unlock_bh(&atchan->lock);
1614 
1615 		if (!at_xdmac_chan_is_cyclic(atchan)) {
1616 			dma_cookie_complete(txd);
1617 			if (txd->flags & DMA_PREP_INTERRUPT)
1618 				dmaengine_desc_get_callback_invoke(txd, NULL);
1619 		}
1620 
1621 		dma_run_dependencies(txd);
1622 
1623 		at_xdmac_advance_work(atchan);
1624 	}
1625 }
1626 
1627 static irqreturn_t at_xdmac_interrupt(int irq, void *dev_id)
1628 {
1629 	struct at_xdmac		*atxdmac = (struct at_xdmac *)dev_id;
1630 	struct at_xdmac_chan	*atchan;
1631 	u32			imr, status, pending;
1632 	u32			chan_imr, chan_status;
1633 	int			i, ret = IRQ_NONE;
1634 
1635 	do {
1636 		imr = at_xdmac_read(atxdmac, AT_XDMAC_GIM);
1637 		status = at_xdmac_read(atxdmac, AT_XDMAC_GIS);
1638 		pending = status & imr;
1639 
1640 		dev_vdbg(atxdmac->dma.dev,
1641 			 "%s: status=0x%08x, imr=0x%08x, pending=0x%08x\n",
1642 			 __func__, status, imr, pending);
1643 
1644 		if (!pending)
1645 			break;
1646 
1647 		/* We have to find which channel has generated the interrupt. */
1648 		for (i = 0; i < atxdmac->dma.chancnt; i++) {
1649 			if (!((1 << i) & pending))
1650 				continue;
1651 
1652 			atchan = &atxdmac->chan[i];
1653 			chan_imr = at_xdmac_chan_read(atchan, AT_XDMAC_CIM);
1654 			chan_status = at_xdmac_chan_read(atchan, AT_XDMAC_CIS);
1655 			atchan->status = chan_status & chan_imr;
1656 			dev_vdbg(atxdmac->dma.dev,
1657 				 "%s: chan%d: imr=0x%x, status=0x%x\n",
1658 				 __func__, i, chan_imr, chan_status);
1659 			dev_vdbg(chan2dev(&atchan->chan),
1660 				 "%s: CC=0x%08x CNDA=0x%08x, CNDC=0x%08x, CSA=0x%08x, CDA=0x%08x, CUBC=0x%08x\n",
1661 				 __func__,
1662 				 at_xdmac_chan_read(atchan, AT_XDMAC_CC),
1663 				 at_xdmac_chan_read(atchan, AT_XDMAC_CNDA),
1664 				 at_xdmac_chan_read(atchan, AT_XDMAC_CNDC),
1665 				 at_xdmac_chan_read(atchan, AT_XDMAC_CSA),
1666 				 at_xdmac_chan_read(atchan, AT_XDMAC_CDA),
1667 				 at_xdmac_chan_read(atchan, AT_XDMAC_CUBC));
1668 
1669 			if (atchan->status & (AT_XDMAC_CIS_RBEIS | AT_XDMAC_CIS_WBEIS))
1670 				at_xdmac_write(atxdmac, AT_XDMAC_GD, atchan->mask);
1671 
1672 			tasklet_schedule(&atchan->tasklet);
1673 			ret = IRQ_HANDLED;
1674 		}
1675 
1676 	} while (pending);
1677 
1678 	return ret;
1679 }
1680 
1681 static void at_xdmac_issue_pending(struct dma_chan *chan)
1682 {
1683 	struct at_xdmac_chan *atchan = to_at_xdmac_chan(chan);
1684 
1685 	dev_dbg(chan2dev(&atchan->chan), "%s\n", __func__);
1686 
1687 	if (!at_xdmac_chan_is_cyclic(atchan))
1688 		at_xdmac_advance_work(atchan);
1689 
1690 	return;
1691 }
1692 
1693 static int at_xdmac_device_config(struct dma_chan *chan,
1694 				  struct dma_slave_config *config)
1695 {
1696 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1697 	int ret;
1698 	unsigned long		flags;
1699 
1700 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1701 
1702 	spin_lock_irqsave(&atchan->lock, flags);
1703 	ret = at_xdmac_set_slave_config(chan, config);
1704 	spin_unlock_irqrestore(&atchan->lock, flags);
1705 
1706 	return ret;
1707 }
1708 
1709 static int at_xdmac_device_pause(struct dma_chan *chan)
1710 {
1711 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1712 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1713 	unsigned long		flags;
1714 
1715 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1716 
1717 	if (test_and_set_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status))
1718 		return 0;
1719 
1720 	spin_lock_irqsave(&atchan->lock, flags);
1721 	at_xdmac_write(atxdmac, AT_XDMAC_GRWS, atchan->mask);
1722 	while (at_xdmac_chan_read(atchan, AT_XDMAC_CC)
1723 	       & (AT_XDMAC_CC_WRIP | AT_XDMAC_CC_RDIP))
1724 		cpu_relax();
1725 	spin_unlock_irqrestore(&atchan->lock, flags);
1726 
1727 	return 0;
1728 }
1729 
1730 static int at_xdmac_device_resume(struct dma_chan *chan)
1731 {
1732 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1733 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1734 	unsigned long		flags;
1735 
1736 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1737 
1738 	spin_lock_irqsave(&atchan->lock, flags);
1739 	if (!at_xdmac_chan_is_paused(atchan)) {
1740 		spin_unlock_irqrestore(&atchan->lock, flags);
1741 		return 0;
1742 	}
1743 
1744 	at_xdmac_write(atxdmac, AT_XDMAC_GRWR, atchan->mask);
1745 	clear_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status);
1746 	spin_unlock_irqrestore(&atchan->lock, flags);
1747 
1748 	return 0;
1749 }
1750 
1751 static int at_xdmac_device_terminate_all(struct dma_chan *chan)
1752 {
1753 	struct at_xdmac_desc	*desc, *_desc;
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 	spin_lock_irqsave(&atchan->lock, flags);
1761 	at_xdmac_write(atxdmac, AT_XDMAC_GD, atchan->mask);
1762 	while (at_xdmac_read(atxdmac, AT_XDMAC_GS) & atchan->mask)
1763 		cpu_relax();
1764 
1765 	/* Cancel all pending transfers. */
1766 	list_for_each_entry_safe(desc, _desc, &atchan->xfers_list, xfer_node)
1767 		at_xdmac_remove_xfer(atchan, desc);
1768 
1769 	clear_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status);
1770 	clear_bit(AT_XDMAC_CHAN_IS_CYCLIC, &atchan->status);
1771 	spin_unlock_irqrestore(&atchan->lock, flags);
1772 
1773 	return 0;
1774 }
1775 
1776 static int at_xdmac_alloc_chan_resources(struct dma_chan *chan)
1777 {
1778 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1779 	struct at_xdmac_desc	*desc;
1780 	int			i;
1781 	unsigned long		flags;
1782 
1783 	spin_lock_irqsave(&atchan->lock, flags);
1784 
1785 	if (at_xdmac_chan_is_enabled(atchan)) {
1786 		dev_err(chan2dev(chan),
1787 			"can't allocate channel resources (channel enabled)\n");
1788 		i = -EIO;
1789 		goto spin_unlock;
1790 	}
1791 
1792 	if (!list_empty(&atchan->free_descs_list)) {
1793 		dev_err(chan2dev(chan),
1794 			"can't allocate channel resources (channel not free from a previous use)\n");
1795 		i = -EIO;
1796 		goto spin_unlock;
1797 	}
1798 
1799 	for (i = 0; i < init_nr_desc_per_channel; i++) {
1800 		desc = at_xdmac_alloc_desc(chan, GFP_ATOMIC);
1801 		if (!desc) {
1802 			dev_warn(chan2dev(chan),
1803 				"only %d descriptors have been allocated\n", i);
1804 			break;
1805 		}
1806 		list_add_tail(&desc->desc_node, &atchan->free_descs_list);
1807 	}
1808 
1809 	dma_cookie_init(chan);
1810 
1811 	dev_dbg(chan2dev(chan), "%s: allocated %d descriptors\n", __func__, i);
1812 
1813 spin_unlock:
1814 	spin_unlock_irqrestore(&atchan->lock, flags);
1815 	return i;
1816 }
1817 
1818 static void at_xdmac_free_chan_resources(struct dma_chan *chan)
1819 {
1820 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1821 	struct at_xdmac		*atxdmac = to_at_xdmac(chan->device);
1822 	struct at_xdmac_desc	*desc, *_desc;
1823 
1824 	list_for_each_entry_safe(desc, _desc, &atchan->free_descs_list, desc_node) {
1825 		dev_dbg(chan2dev(chan), "%s: freeing descriptor %p\n", __func__, desc);
1826 		list_del(&desc->desc_node);
1827 		dma_pool_free(atxdmac->at_xdmac_desc_pool, desc, desc->tx_dma_desc.phys);
1828 	}
1829 
1830 	return;
1831 }
1832 
1833 #ifdef CONFIG_PM
1834 static int atmel_xdmac_prepare(struct device *dev)
1835 {
1836 	struct platform_device	*pdev = to_platform_device(dev);
1837 	struct at_xdmac		*atxdmac = platform_get_drvdata(pdev);
1838 	struct dma_chan		*chan, *_chan;
1839 
1840 	list_for_each_entry_safe(chan, _chan, &atxdmac->dma.channels, device_node) {
1841 		struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1842 
1843 		/* Wait for transfer completion, except in cyclic case. */
1844 		if (at_xdmac_chan_is_enabled(atchan) && !at_xdmac_chan_is_cyclic(atchan))
1845 			return -EAGAIN;
1846 	}
1847 	return 0;
1848 }
1849 #else
1850 #	define atmel_xdmac_prepare NULL
1851 #endif
1852 
1853 #ifdef CONFIG_PM_SLEEP
1854 static int atmel_xdmac_suspend(struct device *dev)
1855 {
1856 	struct platform_device	*pdev = to_platform_device(dev);
1857 	struct at_xdmac		*atxdmac = platform_get_drvdata(pdev);
1858 	struct dma_chan		*chan, *_chan;
1859 
1860 	list_for_each_entry_safe(chan, _chan, &atxdmac->dma.channels, device_node) {
1861 		struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1862 
1863 		atchan->save_cc = at_xdmac_chan_read(atchan, AT_XDMAC_CC);
1864 		if (at_xdmac_chan_is_cyclic(atchan)) {
1865 			if (!at_xdmac_chan_is_paused(atchan))
1866 				at_xdmac_device_pause(chan);
1867 			atchan->save_cim = at_xdmac_chan_read(atchan, AT_XDMAC_CIM);
1868 			atchan->save_cnda = at_xdmac_chan_read(atchan, AT_XDMAC_CNDA);
1869 			atchan->save_cndc = at_xdmac_chan_read(atchan, AT_XDMAC_CNDC);
1870 		}
1871 	}
1872 	atxdmac->save_gim = at_xdmac_read(atxdmac, AT_XDMAC_GIM);
1873 
1874 	at_xdmac_off(atxdmac);
1875 	clk_disable_unprepare(atxdmac->clk);
1876 	return 0;
1877 }
1878 
1879 static int atmel_xdmac_resume(struct device *dev)
1880 {
1881 	struct platform_device	*pdev = to_platform_device(dev);
1882 	struct at_xdmac		*atxdmac = platform_get_drvdata(pdev);
1883 	struct at_xdmac_chan	*atchan;
1884 	struct dma_chan		*chan, *_chan;
1885 	int			i;
1886 
1887 	clk_prepare_enable(atxdmac->clk);
1888 
1889 	/* Clear pending interrupts. */
1890 	for (i = 0; i < atxdmac->dma.chancnt; i++) {
1891 		atchan = &atxdmac->chan[i];
1892 		while (at_xdmac_chan_read(atchan, AT_XDMAC_CIS))
1893 			cpu_relax();
1894 	}
1895 
1896 	at_xdmac_write(atxdmac, AT_XDMAC_GIE, atxdmac->save_gim);
1897 	list_for_each_entry_safe(chan, _chan, &atxdmac->dma.channels, device_node) {
1898 		atchan = to_at_xdmac_chan(chan);
1899 		at_xdmac_chan_write(atchan, AT_XDMAC_CC, atchan->save_cc);
1900 		if (at_xdmac_chan_is_cyclic(atchan)) {
1901 			if (at_xdmac_chan_is_paused(atchan))
1902 				at_xdmac_device_resume(chan);
1903 			at_xdmac_chan_write(atchan, AT_XDMAC_CNDA, atchan->save_cnda);
1904 			at_xdmac_chan_write(atchan, AT_XDMAC_CNDC, atchan->save_cndc);
1905 			at_xdmac_chan_write(atchan, AT_XDMAC_CIE, atchan->save_cim);
1906 			wmb();
1907 			at_xdmac_write(atxdmac, AT_XDMAC_GE, atchan->mask);
1908 		}
1909 	}
1910 	return 0;
1911 }
1912 #endif /* CONFIG_PM_SLEEP */
1913 
1914 static int at_xdmac_probe(struct platform_device *pdev)
1915 {
1916 	struct resource	*res;
1917 	struct at_xdmac	*atxdmac;
1918 	int		irq, size, nr_channels, i, ret;
1919 	void __iomem	*base;
1920 	u32		reg;
1921 
1922 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1923 	if (!res)
1924 		return -EINVAL;
1925 
1926 	irq = platform_get_irq(pdev, 0);
1927 	if (irq < 0)
1928 		return irq;
1929 
1930 	base = devm_ioremap_resource(&pdev->dev, res);
1931 	if (IS_ERR(base))
1932 		return PTR_ERR(base);
1933 
1934 	/*
1935 	 * Read number of xdmac channels, read helper function can't be used
1936 	 * since atxdmac is not yet allocated and we need to know the number
1937 	 * of channels to do the allocation.
1938 	 */
1939 	reg = readl_relaxed(base + AT_XDMAC_GTYPE);
1940 	nr_channels = AT_XDMAC_NB_CH(reg);
1941 	if (nr_channels > AT_XDMAC_MAX_CHAN) {
1942 		dev_err(&pdev->dev, "invalid number of channels (%u)\n",
1943 			nr_channels);
1944 		return -EINVAL;
1945 	}
1946 
1947 	size = sizeof(*atxdmac);
1948 	size += nr_channels * sizeof(struct at_xdmac_chan);
1949 	atxdmac = devm_kzalloc(&pdev->dev, size, GFP_KERNEL);
1950 	if (!atxdmac) {
1951 		dev_err(&pdev->dev, "can't allocate at_xdmac structure\n");
1952 		return -ENOMEM;
1953 	}
1954 
1955 	atxdmac->regs = base;
1956 	atxdmac->irq = irq;
1957 
1958 	atxdmac->clk = devm_clk_get(&pdev->dev, "dma_clk");
1959 	if (IS_ERR(atxdmac->clk)) {
1960 		dev_err(&pdev->dev, "can't get dma_clk\n");
1961 		return PTR_ERR(atxdmac->clk);
1962 	}
1963 
1964 	/* Do not use dev res to prevent races with tasklet */
1965 	ret = request_irq(atxdmac->irq, at_xdmac_interrupt, 0, "at_xdmac", atxdmac);
1966 	if (ret) {
1967 		dev_err(&pdev->dev, "can't request irq\n");
1968 		return ret;
1969 	}
1970 
1971 	ret = clk_prepare_enable(atxdmac->clk);
1972 	if (ret) {
1973 		dev_err(&pdev->dev, "can't prepare or enable clock\n");
1974 		goto err_free_irq;
1975 	}
1976 
1977 	atxdmac->at_xdmac_desc_pool =
1978 		dmam_pool_create(dev_name(&pdev->dev), &pdev->dev,
1979 				sizeof(struct at_xdmac_desc), 4, 0);
1980 	if (!atxdmac->at_xdmac_desc_pool) {
1981 		dev_err(&pdev->dev, "no memory for descriptors dma pool\n");
1982 		ret = -ENOMEM;
1983 		goto err_clk_disable;
1984 	}
1985 
1986 	dma_cap_set(DMA_CYCLIC, atxdmac->dma.cap_mask);
1987 	dma_cap_set(DMA_INTERLEAVE, atxdmac->dma.cap_mask);
1988 	dma_cap_set(DMA_MEMCPY, atxdmac->dma.cap_mask);
1989 	dma_cap_set(DMA_MEMSET, atxdmac->dma.cap_mask);
1990 	dma_cap_set(DMA_MEMSET_SG, atxdmac->dma.cap_mask);
1991 	dma_cap_set(DMA_SLAVE, atxdmac->dma.cap_mask);
1992 	/*
1993 	 * Without DMA_PRIVATE the driver is not able to allocate more than
1994 	 * one channel, second allocation fails in private_candidate.
1995 	 */
1996 	dma_cap_set(DMA_PRIVATE, atxdmac->dma.cap_mask);
1997 	atxdmac->dma.dev				= &pdev->dev;
1998 	atxdmac->dma.device_alloc_chan_resources	= at_xdmac_alloc_chan_resources;
1999 	atxdmac->dma.device_free_chan_resources		= at_xdmac_free_chan_resources;
2000 	atxdmac->dma.device_tx_status			= at_xdmac_tx_status;
2001 	atxdmac->dma.device_issue_pending		= at_xdmac_issue_pending;
2002 	atxdmac->dma.device_prep_dma_cyclic		= at_xdmac_prep_dma_cyclic;
2003 	atxdmac->dma.device_prep_interleaved_dma	= at_xdmac_prep_interleaved;
2004 	atxdmac->dma.device_prep_dma_memcpy		= at_xdmac_prep_dma_memcpy;
2005 	atxdmac->dma.device_prep_dma_memset		= at_xdmac_prep_dma_memset;
2006 	atxdmac->dma.device_prep_dma_memset_sg		= at_xdmac_prep_dma_memset_sg;
2007 	atxdmac->dma.device_prep_slave_sg		= at_xdmac_prep_slave_sg;
2008 	atxdmac->dma.device_config			= at_xdmac_device_config;
2009 	atxdmac->dma.device_pause			= at_xdmac_device_pause;
2010 	atxdmac->dma.device_resume			= at_xdmac_device_resume;
2011 	atxdmac->dma.device_terminate_all		= at_xdmac_device_terminate_all;
2012 	atxdmac->dma.src_addr_widths = AT_XDMAC_DMA_BUSWIDTHS;
2013 	atxdmac->dma.dst_addr_widths = AT_XDMAC_DMA_BUSWIDTHS;
2014 	atxdmac->dma.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2015 	atxdmac->dma.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2016 
2017 	/* Disable all chans and interrupts. */
2018 	at_xdmac_off(atxdmac);
2019 
2020 	/* Init channels. */
2021 	INIT_LIST_HEAD(&atxdmac->dma.channels);
2022 	for (i = 0; i < nr_channels; i++) {
2023 		struct at_xdmac_chan *atchan = &atxdmac->chan[i];
2024 
2025 		atchan->chan.device = &atxdmac->dma;
2026 		list_add_tail(&atchan->chan.device_node,
2027 			      &atxdmac->dma.channels);
2028 
2029 		atchan->ch_regs = at_xdmac_chan_reg_base(atxdmac, i);
2030 		atchan->mask = 1 << i;
2031 
2032 		spin_lock_init(&atchan->lock);
2033 		INIT_LIST_HEAD(&atchan->xfers_list);
2034 		INIT_LIST_HEAD(&atchan->free_descs_list);
2035 		tasklet_init(&atchan->tasklet, at_xdmac_tasklet,
2036 			     (unsigned long)atchan);
2037 
2038 		/* Clear pending interrupts. */
2039 		while (at_xdmac_chan_read(atchan, AT_XDMAC_CIS))
2040 			cpu_relax();
2041 	}
2042 	platform_set_drvdata(pdev, atxdmac);
2043 
2044 	ret = dma_async_device_register(&atxdmac->dma);
2045 	if (ret) {
2046 		dev_err(&pdev->dev, "fail to register DMA engine device\n");
2047 		goto err_clk_disable;
2048 	}
2049 
2050 	ret = of_dma_controller_register(pdev->dev.of_node,
2051 					 at_xdmac_xlate, atxdmac);
2052 	if (ret) {
2053 		dev_err(&pdev->dev, "could not register of dma controller\n");
2054 		goto err_dma_unregister;
2055 	}
2056 
2057 	dev_info(&pdev->dev, "%d channels, mapped at 0x%p\n",
2058 		 nr_channels, atxdmac->regs);
2059 
2060 	return 0;
2061 
2062 err_dma_unregister:
2063 	dma_async_device_unregister(&atxdmac->dma);
2064 err_clk_disable:
2065 	clk_disable_unprepare(atxdmac->clk);
2066 err_free_irq:
2067 	free_irq(atxdmac->irq, atxdmac);
2068 	return ret;
2069 }
2070 
2071 static int at_xdmac_remove(struct platform_device *pdev)
2072 {
2073 	struct at_xdmac	*atxdmac = (struct at_xdmac *)platform_get_drvdata(pdev);
2074 	int		i;
2075 
2076 	at_xdmac_off(atxdmac);
2077 	of_dma_controller_free(pdev->dev.of_node);
2078 	dma_async_device_unregister(&atxdmac->dma);
2079 	clk_disable_unprepare(atxdmac->clk);
2080 
2081 	free_irq(atxdmac->irq, atxdmac);
2082 
2083 	for (i = 0; i < atxdmac->dma.chancnt; i++) {
2084 		struct at_xdmac_chan *atchan = &atxdmac->chan[i];
2085 
2086 		tasklet_kill(&atchan->tasklet);
2087 		at_xdmac_free_chan_resources(&atchan->chan);
2088 	}
2089 
2090 	return 0;
2091 }
2092 
2093 static const struct dev_pm_ops atmel_xdmac_dev_pm_ops = {
2094 	.prepare	= atmel_xdmac_prepare,
2095 	SET_LATE_SYSTEM_SLEEP_PM_OPS(atmel_xdmac_suspend, atmel_xdmac_resume)
2096 };
2097 
2098 static const struct of_device_id atmel_xdmac_dt_ids[] = {
2099 	{
2100 		.compatible = "atmel,sama5d4-dma",
2101 	}, {
2102 		/* sentinel */
2103 	}
2104 };
2105 MODULE_DEVICE_TABLE(of, atmel_xdmac_dt_ids);
2106 
2107 static struct platform_driver at_xdmac_driver = {
2108 	.probe		= at_xdmac_probe,
2109 	.remove		= at_xdmac_remove,
2110 	.driver = {
2111 		.name		= "at_xdmac",
2112 		.of_match_table	= of_match_ptr(atmel_xdmac_dt_ids),
2113 		.pm		= &atmel_xdmac_dev_pm_ops,
2114 	}
2115 };
2116 
2117 static int __init at_xdmac_init(void)
2118 {
2119 	return platform_driver_probe(&at_xdmac_driver, at_xdmac_probe);
2120 }
2121 subsys_initcall(at_xdmac_init);
2122 
2123 MODULE_DESCRIPTION("Atmel Extended DMA Controller driver");
2124 MODULE_AUTHOR("Ludovic Desroches <ludovic.desroches@atmel.com>");
2125 MODULE_LICENSE("GPL");
2126