xref: /linux/drivers/dma/at_xdmac.c (revision 3bdab16c55f57a24245c97d707241dd9b48d1a91)
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 	u32				irq_status;
207 	unsigned long			status;
208 	struct tasklet_struct		tasklet;
209 	struct dma_slave_config		sconfig;
210 
211 	spinlock_t			lock;
212 
213 	struct list_head		xfers_list;
214 	struct list_head		free_descs_list;
215 };
216 
217 
218 /* ----- Controller ----- */
219 struct at_xdmac {
220 	struct dma_device	dma;
221 	void __iomem		*regs;
222 	int			irq;
223 	struct clk		*clk;
224 	u32			save_gim;
225 	struct dma_pool		*at_xdmac_desc_pool;
226 	struct at_xdmac_chan	chan[0];
227 };
228 
229 
230 /* ----- Descriptors ----- */
231 
232 /* Linked List Descriptor */
233 struct at_xdmac_lld {
234 	dma_addr_t	mbr_nda;	/* Next Descriptor Member */
235 	u32		mbr_ubc;	/* Microblock Control Member */
236 	dma_addr_t	mbr_sa;		/* Source Address Member */
237 	dma_addr_t	mbr_da;		/* Destination Address Member */
238 	u32		mbr_cfg;	/* Configuration Register */
239 	u32		mbr_bc;		/* Block Control Register */
240 	u32		mbr_ds;		/* Data Stride Register */
241 	u32		mbr_sus;	/* Source Microblock Stride Register */
242 	u32		mbr_dus;	/* Destination Microblock Stride Register */
243 };
244 
245 /* 64-bit alignment needed to update CNDA and CUBC registers in an atomic way. */
246 struct at_xdmac_desc {
247 	struct at_xdmac_lld		lld;
248 	enum dma_transfer_direction	direction;
249 	struct dma_async_tx_descriptor	tx_dma_desc;
250 	struct list_head		desc_node;
251 	/* Following members are only used by the first descriptor */
252 	bool				active_xfer;
253 	unsigned int			xfer_size;
254 	struct list_head		descs_list;
255 	struct list_head		xfer_node;
256 } __aligned(sizeof(u64));
257 
258 static inline void __iomem *at_xdmac_chan_reg_base(struct at_xdmac *atxdmac, unsigned int chan_nb)
259 {
260 	return atxdmac->regs + (AT_XDMAC_CHAN_REG_BASE + chan_nb * 0x40);
261 }
262 
263 #define at_xdmac_read(atxdmac, reg) readl_relaxed((atxdmac)->regs + (reg))
264 #define at_xdmac_write(atxdmac, reg, value) \
265 	writel_relaxed((value), (atxdmac)->regs + (reg))
266 
267 #define at_xdmac_chan_read(atchan, reg) readl_relaxed((atchan)->ch_regs + (reg))
268 #define at_xdmac_chan_write(atchan, reg, value) writel_relaxed((value), (atchan)->ch_regs + (reg))
269 
270 static inline struct at_xdmac_chan *to_at_xdmac_chan(struct dma_chan *dchan)
271 {
272 	return container_of(dchan, struct at_xdmac_chan, chan);
273 }
274 
275 static struct device *chan2dev(struct dma_chan *chan)
276 {
277 	return &chan->dev->device;
278 }
279 
280 static inline struct at_xdmac *to_at_xdmac(struct dma_device *ddev)
281 {
282 	return container_of(ddev, struct at_xdmac, dma);
283 }
284 
285 static inline struct at_xdmac_desc *txd_to_at_desc(struct dma_async_tx_descriptor *txd)
286 {
287 	return container_of(txd, struct at_xdmac_desc, tx_dma_desc);
288 }
289 
290 static inline int at_xdmac_chan_is_cyclic(struct at_xdmac_chan *atchan)
291 {
292 	return test_bit(AT_XDMAC_CHAN_IS_CYCLIC, &atchan->status);
293 }
294 
295 static inline int at_xdmac_chan_is_paused(struct at_xdmac_chan *atchan)
296 {
297 	return test_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status);
298 }
299 
300 static inline int at_xdmac_csize(u32 maxburst)
301 {
302 	int csize;
303 
304 	csize = ffs(maxburst) - 1;
305 	if (csize > 4)
306 		csize = -EINVAL;
307 
308 	return csize;
309 };
310 
311 static inline bool at_xdmac_chan_is_peripheral_xfer(u32 cfg)
312 {
313 	return cfg & AT_XDMAC_CC_TYPE_PER_TRAN;
314 }
315 
316 static inline u8 at_xdmac_get_dwidth(u32 cfg)
317 {
318 	return (cfg & AT_XDMAC_CC_DWIDTH_MASK) >> AT_XDMAC_CC_DWIDTH_OFFSET;
319 };
320 
321 static unsigned int init_nr_desc_per_channel = 64;
322 module_param(init_nr_desc_per_channel, uint, 0644);
323 MODULE_PARM_DESC(init_nr_desc_per_channel,
324 		 "initial descriptors per channel (default: 64)");
325 
326 
327 static bool at_xdmac_chan_is_enabled(struct at_xdmac_chan *atchan)
328 {
329 	return at_xdmac_chan_read(atchan, AT_XDMAC_GS) & atchan->mask;
330 }
331 
332 static void at_xdmac_off(struct at_xdmac *atxdmac)
333 {
334 	at_xdmac_write(atxdmac, AT_XDMAC_GD, -1L);
335 
336 	/* Wait that all chans are disabled. */
337 	while (at_xdmac_read(atxdmac, AT_XDMAC_GS))
338 		cpu_relax();
339 
340 	at_xdmac_write(atxdmac, AT_XDMAC_GID, -1L);
341 }
342 
343 /* Call with lock hold. */
344 static void at_xdmac_start_xfer(struct at_xdmac_chan *atchan,
345 				struct at_xdmac_desc *first)
346 {
347 	struct at_xdmac	*atxdmac = to_at_xdmac(atchan->chan.device);
348 	u32		reg;
349 
350 	dev_vdbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, first);
351 
352 	if (at_xdmac_chan_is_enabled(atchan))
353 		return;
354 
355 	/* Set transfer as active to not try to start it again. */
356 	first->active_xfer = true;
357 
358 	/* Tell xdmac where to get the first descriptor. */
359 	reg = AT_XDMAC_CNDA_NDA(first->tx_dma_desc.phys)
360 	      | AT_XDMAC_CNDA_NDAIF(atchan->memif);
361 	at_xdmac_chan_write(atchan, AT_XDMAC_CNDA, reg);
362 
363 	/*
364 	 * When doing non cyclic transfer we need to use the next
365 	 * descriptor view 2 since some fields of the configuration register
366 	 * depend on transfer size and src/dest addresses.
367 	 */
368 	if (at_xdmac_chan_is_cyclic(atchan))
369 		reg = AT_XDMAC_CNDC_NDVIEW_NDV1;
370 	else if (first->lld.mbr_ubc & AT_XDMAC_MBR_UBC_NDV3)
371 		reg = AT_XDMAC_CNDC_NDVIEW_NDV3;
372 	else
373 		reg = AT_XDMAC_CNDC_NDVIEW_NDV2;
374 	/*
375 	 * Even if the register will be updated from the configuration in the
376 	 * descriptor when using view 2 or higher, the PROT bit won't be set
377 	 * properly. This bit can be modified only by using the channel
378 	 * configuration register.
379 	 */
380 	at_xdmac_chan_write(atchan, AT_XDMAC_CC, first->lld.mbr_cfg);
381 
382 	reg |= AT_XDMAC_CNDC_NDDUP
383 	       | AT_XDMAC_CNDC_NDSUP
384 	       | AT_XDMAC_CNDC_NDE;
385 	at_xdmac_chan_write(atchan, AT_XDMAC_CNDC, reg);
386 
387 	dev_vdbg(chan2dev(&atchan->chan),
388 		 "%s: CC=0x%08x CNDA=0x%08x, CNDC=0x%08x, CSA=0x%08x, CDA=0x%08x, CUBC=0x%08x\n",
389 		 __func__, at_xdmac_chan_read(atchan, AT_XDMAC_CC),
390 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDA),
391 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDC),
392 		 at_xdmac_chan_read(atchan, AT_XDMAC_CSA),
393 		 at_xdmac_chan_read(atchan, AT_XDMAC_CDA),
394 		 at_xdmac_chan_read(atchan, AT_XDMAC_CUBC));
395 
396 	at_xdmac_chan_write(atchan, AT_XDMAC_CID, 0xffffffff);
397 	reg = AT_XDMAC_CIE_RBEIE | AT_XDMAC_CIE_WBEIE;
398 	/*
399 	 * Request Overflow Error is only for peripheral synchronized transfers
400 	 */
401 	if (at_xdmac_chan_is_peripheral_xfer(first->lld.mbr_cfg))
402 		reg |= AT_XDMAC_CIE_ROIE;
403 
404 	/*
405 	 * There is no end of list when doing cyclic dma, we need to get
406 	 * an interrupt after each periods.
407 	 */
408 	if (at_xdmac_chan_is_cyclic(atchan))
409 		at_xdmac_chan_write(atchan, AT_XDMAC_CIE,
410 				    reg | AT_XDMAC_CIE_BIE);
411 	else
412 		at_xdmac_chan_write(atchan, AT_XDMAC_CIE,
413 				    reg | AT_XDMAC_CIE_LIE);
414 	at_xdmac_write(atxdmac, AT_XDMAC_GIE, atchan->mask);
415 	dev_vdbg(chan2dev(&atchan->chan),
416 		 "%s: enable channel (0x%08x)\n", __func__, atchan->mask);
417 	wmb();
418 	at_xdmac_write(atxdmac, AT_XDMAC_GE, atchan->mask);
419 
420 	dev_vdbg(chan2dev(&atchan->chan),
421 		 "%s: CC=0x%08x CNDA=0x%08x, CNDC=0x%08x, CSA=0x%08x, CDA=0x%08x, CUBC=0x%08x\n",
422 		 __func__, at_xdmac_chan_read(atchan, AT_XDMAC_CC),
423 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDA),
424 		 at_xdmac_chan_read(atchan, AT_XDMAC_CNDC),
425 		 at_xdmac_chan_read(atchan, AT_XDMAC_CSA),
426 		 at_xdmac_chan_read(atchan, AT_XDMAC_CDA),
427 		 at_xdmac_chan_read(atchan, AT_XDMAC_CUBC));
428 
429 }
430 
431 static dma_cookie_t at_xdmac_tx_submit(struct dma_async_tx_descriptor *tx)
432 {
433 	struct at_xdmac_desc	*desc = txd_to_at_desc(tx);
434 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(tx->chan);
435 	dma_cookie_t		cookie;
436 	unsigned long		irqflags;
437 
438 	spin_lock_irqsave(&atchan->lock, irqflags);
439 	cookie = dma_cookie_assign(tx);
440 
441 	dev_vdbg(chan2dev(tx->chan), "%s: atchan 0x%p, add desc 0x%p to xfers_list\n",
442 		 __func__, atchan, desc);
443 	list_add_tail(&desc->xfer_node, &atchan->xfers_list);
444 	if (list_is_singular(&atchan->xfers_list))
445 		at_xdmac_start_xfer(atchan, desc);
446 
447 	spin_unlock_irqrestore(&atchan->lock, irqflags);
448 	return cookie;
449 }
450 
451 static struct at_xdmac_desc *at_xdmac_alloc_desc(struct dma_chan *chan,
452 						 gfp_t gfp_flags)
453 {
454 	struct at_xdmac_desc	*desc;
455 	struct at_xdmac		*atxdmac = to_at_xdmac(chan->device);
456 	dma_addr_t		phys;
457 
458 	desc = dma_pool_zalloc(atxdmac->at_xdmac_desc_pool, gfp_flags, &phys);
459 	if (desc) {
460 		INIT_LIST_HEAD(&desc->descs_list);
461 		dma_async_tx_descriptor_init(&desc->tx_dma_desc, chan);
462 		desc->tx_dma_desc.tx_submit = at_xdmac_tx_submit;
463 		desc->tx_dma_desc.phys = phys;
464 	}
465 
466 	return desc;
467 }
468 
469 static void at_xdmac_init_used_desc(struct at_xdmac_desc *desc)
470 {
471 	memset(&desc->lld, 0, sizeof(desc->lld));
472 	INIT_LIST_HEAD(&desc->descs_list);
473 	desc->direction = DMA_TRANS_NONE;
474 	desc->xfer_size = 0;
475 	desc->active_xfer = false;
476 }
477 
478 /* Call must be protected by lock. */
479 static struct at_xdmac_desc *at_xdmac_get_desc(struct at_xdmac_chan *atchan)
480 {
481 	struct at_xdmac_desc *desc;
482 
483 	if (list_empty(&atchan->free_descs_list)) {
484 		desc = at_xdmac_alloc_desc(&atchan->chan, GFP_NOWAIT);
485 	} else {
486 		desc = list_first_entry(&atchan->free_descs_list,
487 					struct at_xdmac_desc, desc_node);
488 		list_del(&desc->desc_node);
489 		at_xdmac_init_used_desc(desc);
490 	}
491 
492 	return desc;
493 }
494 
495 static void at_xdmac_queue_desc(struct dma_chan *chan,
496 				struct at_xdmac_desc *prev,
497 				struct at_xdmac_desc *desc)
498 {
499 	if (!prev || !desc)
500 		return;
501 
502 	prev->lld.mbr_nda = desc->tx_dma_desc.phys;
503 	prev->lld.mbr_ubc |= AT_XDMAC_MBR_UBC_NDE;
504 
505 	dev_dbg(chan2dev(chan),	"%s: chain lld: prev=0x%p, mbr_nda=%pad\n",
506 		__func__, prev, &prev->lld.mbr_nda);
507 }
508 
509 static inline void at_xdmac_increment_block_count(struct dma_chan *chan,
510 						  struct at_xdmac_desc *desc)
511 {
512 	if (!desc)
513 		return;
514 
515 	desc->lld.mbr_bc++;
516 
517 	dev_dbg(chan2dev(chan),
518 		"%s: incrementing the block count of the desc 0x%p\n",
519 		__func__, desc);
520 }
521 
522 static struct dma_chan *at_xdmac_xlate(struct of_phandle_args *dma_spec,
523 				       struct of_dma *of_dma)
524 {
525 	struct at_xdmac		*atxdmac = of_dma->of_dma_data;
526 	struct at_xdmac_chan	*atchan;
527 	struct dma_chan		*chan;
528 	struct device		*dev = atxdmac->dma.dev;
529 
530 	if (dma_spec->args_count != 1) {
531 		dev_err(dev, "dma phandler args: bad number of args\n");
532 		return NULL;
533 	}
534 
535 	chan = dma_get_any_slave_channel(&atxdmac->dma);
536 	if (!chan) {
537 		dev_err(dev, "can't get a dma channel\n");
538 		return NULL;
539 	}
540 
541 	atchan = to_at_xdmac_chan(chan);
542 	atchan->memif = AT91_XDMAC_DT_GET_MEM_IF(dma_spec->args[0]);
543 	atchan->perif = AT91_XDMAC_DT_GET_PER_IF(dma_spec->args[0]);
544 	atchan->perid = AT91_XDMAC_DT_GET_PERID(dma_spec->args[0]);
545 	dev_dbg(dev, "chan dt cfg: memif=%u perif=%u perid=%u\n",
546 		 atchan->memif, atchan->perif, atchan->perid);
547 
548 	return chan;
549 }
550 
551 static int at_xdmac_compute_chan_conf(struct dma_chan *chan,
552 				      enum dma_transfer_direction direction)
553 {
554 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
555 	int			csize, dwidth;
556 
557 	if (direction == DMA_DEV_TO_MEM) {
558 		atchan->cfg =
559 			AT91_XDMAC_DT_PERID(atchan->perid)
560 			| AT_XDMAC_CC_DAM_INCREMENTED_AM
561 			| AT_XDMAC_CC_SAM_FIXED_AM
562 			| AT_XDMAC_CC_DIF(atchan->memif)
563 			| AT_XDMAC_CC_SIF(atchan->perif)
564 			| AT_XDMAC_CC_SWREQ_HWR_CONNECTED
565 			| AT_XDMAC_CC_DSYNC_PER2MEM
566 			| AT_XDMAC_CC_MBSIZE_SIXTEEN
567 			| AT_XDMAC_CC_TYPE_PER_TRAN;
568 		csize = ffs(atchan->sconfig.src_maxburst) - 1;
569 		if (csize < 0) {
570 			dev_err(chan2dev(chan), "invalid src maxburst value\n");
571 			return -EINVAL;
572 		}
573 		atchan->cfg |= AT_XDMAC_CC_CSIZE(csize);
574 		dwidth = ffs(atchan->sconfig.src_addr_width) - 1;
575 		if (dwidth < 0) {
576 			dev_err(chan2dev(chan), "invalid src addr width value\n");
577 			return -EINVAL;
578 		}
579 		atchan->cfg |= AT_XDMAC_CC_DWIDTH(dwidth);
580 	} else if (direction == DMA_MEM_TO_DEV) {
581 		atchan->cfg =
582 			AT91_XDMAC_DT_PERID(atchan->perid)
583 			| AT_XDMAC_CC_DAM_FIXED_AM
584 			| AT_XDMAC_CC_SAM_INCREMENTED_AM
585 			| AT_XDMAC_CC_DIF(atchan->perif)
586 			| AT_XDMAC_CC_SIF(atchan->memif)
587 			| AT_XDMAC_CC_SWREQ_HWR_CONNECTED
588 			| AT_XDMAC_CC_DSYNC_MEM2PER
589 			| AT_XDMAC_CC_MBSIZE_SIXTEEN
590 			| AT_XDMAC_CC_TYPE_PER_TRAN;
591 		csize = ffs(atchan->sconfig.dst_maxburst) - 1;
592 		if (csize < 0) {
593 			dev_err(chan2dev(chan), "invalid src maxburst value\n");
594 			return -EINVAL;
595 		}
596 		atchan->cfg |= AT_XDMAC_CC_CSIZE(csize);
597 		dwidth = ffs(atchan->sconfig.dst_addr_width) - 1;
598 		if (dwidth < 0) {
599 			dev_err(chan2dev(chan), "invalid dst addr width value\n");
600 			return -EINVAL;
601 		}
602 		atchan->cfg |= AT_XDMAC_CC_DWIDTH(dwidth);
603 	}
604 
605 	dev_dbg(chan2dev(chan),	"%s: cfg=0x%08x\n", __func__, atchan->cfg);
606 
607 	return 0;
608 }
609 
610 /*
611  * Only check that maxburst and addr width values are supported by the
612  * the controller but not that the configuration is good to perform the
613  * transfer since we don't know the direction at this stage.
614  */
615 static int at_xdmac_check_slave_config(struct dma_slave_config *sconfig)
616 {
617 	if ((sconfig->src_maxburst > AT_XDMAC_MAX_CSIZE)
618 	    || (sconfig->dst_maxburst > AT_XDMAC_MAX_CSIZE))
619 		return -EINVAL;
620 
621 	if ((sconfig->src_addr_width > AT_XDMAC_MAX_DWIDTH)
622 	    || (sconfig->dst_addr_width > AT_XDMAC_MAX_DWIDTH))
623 		return -EINVAL;
624 
625 	return 0;
626 }
627 
628 static int at_xdmac_set_slave_config(struct dma_chan *chan,
629 				      struct dma_slave_config *sconfig)
630 {
631 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
632 
633 	if (at_xdmac_check_slave_config(sconfig)) {
634 		dev_err(chan2dev(chan), "invalid slave configuration\n");
635 		return -EINVAL;
636 	}
637 
638 	memcpy(&atchan->sconfig, sconfig, sizeof(atchan->sconfig));
639 
640 	return 0;
641 }
642 
643 static struct dma_async_tx_descriptor *
644 at_xdmac_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
645 		       unsigned int sg_len, enum dma_transfer_direction direction,
646 		       unsigned long flags, void *context)
647 {
648 	struct at_xdmac_chan		*atchan = to_at_xdmac_chan(chan);
649 	struct at_xdmac_desc		*first = NULL, *prev = NULL;
650 	struct scatterlist		*sg;
651 	int				i;
652 	unsigned int			xfer_size = 0;
653 	unsigned long			irqflags;
654 	struct dma_async_tx_descriptor	*ret = NULL;
655 
656 	if (!sgl)
657 		return NULL;
658 
659 	if (!is_slave_direction(direction)) {
660 		dev_err(chan2dev(chan), "invalid DMA direction\n");
661 		return NULL;
662 	}
663 
664 	dev_dbg(chan2dev(chan), "%s: sg_len=%d, dir=%s, flags=0x%lx\n",
665 		 __func__, sg_len,
666 		 direction == DMA_MEM_TO_DEV ? "to device" : "from device",
667 		 flags);
668 
669 	/* Protect dma_sconfig field that can be modified by set_slave_conf. */
670 	spin_lock_irqsave(&atchan->lock, irqflags);
671 
672 	if (at_xdmac_compute_chan_conf(chan, direction))
673 		goto spin_unlock;
674 
675 	/* Prepare descriptors. */
676 	for_each_sg(sgl, sg, sg_len, i) {
677 		struct at_xdmac_desc	*desc = NULL;
678 		u32			len, mem, dwidth, fixed_dwidth;
679 
680 		len = sg_dma_len(sg);
681 		mem = sg_dma_address(sg);
682 		if (unlikely(!len)) {
683 			dev_err(chan2dev(chan), "sg data length is zero\n");
684 			goto spin_unlock;
685 		}
686 		dev_dbg(chan2dev(chan), "%s: * sg%d len=%u, mem=0x%08x\n",
687 			 __func__, i, len, mem);
688 
689 		desc = at_xdmac_get_desc(atchan);
690 		if (!desc) {
691 			dev_err(chan2dev(chan), "can't get descriptor\n");
692 			if (first)
693 				list_splice_init(&first->descs_list, &atchan->free_descs_list);
694 			goto spin_unlock;
695 		}
696 
697 		/* Linked list descriptor setup. */
698 		if (direction == DMA_DEV_TO_MEM) {
699 			desc->lld.mbr_sa = atchan->sconfig.src_addr;
700 			desc->lld.mbr_da = mem;
701 		} else {
702 			desc->lld.mbr_sa = mem;
703 			desc->lld.mbr_da = atchan->sconfig.dst_addr;
704 		}
705 		dwidth = at_xdmac_get_dwidth(atchan->cfg);
706 		fixed_dwidth = IS_ALIGNED(len, 1 << dwidth)
707 			       ? dwidth
708 			       : AT_XDMAC_CC_DWIDTH_BYTE;
709 		desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV2			/* next descriptor view */
710 			| AT_XDMAC_MBR_UBC_NDEN					/* next descriptor dst parameter update */
711 			| AT_XDMAC_MBR_UBC_NSEN					/* next descriptor src parameter update */
712 			| (len >> fixed_dwidth);				/* microblock length */
713 		desc->lld.mbr_cfg = (atchan->cfg & ~AT_XDMAC_CC_DWIDTH_MASK) |
714 				    AT_XDMAC_CC_DWIDTH(fixed_dwidth);
715 		dev_dbg(chan2dev(chan),
716 			 "%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x\n",
717 			 __func__, &desc->lld.mbr_sa, &desc->lld.mbr_da, desc->lld.mbr_ubc);
718 
719 		/* Chain lld. */
720 		if (prev)
721 			at_xdmac_queue_desc(chan, prev, desc);
722 
723 		prev = desc;
724 		if (!first)
725 			first = desc;
726 
727 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
728 			 __func__, desc, first);
729 		list_add_tail(&desc->desc_node, &first->descs_list);
730 		xfer_size += len;
731 	}
732 
733 
734 	first->tx_dma_desc.flags = flags;
735 	first->xfer_size = xfer_size;
736 	first->direction = direction;
737 	ret = &first->tx_dma_desc;
738 
739 spin_unlock:
740 	spin_unlock_irqrestore(&atchan->lock, irqflags);
741 	return ret;
742 }
743 
744 static struct dma_async_tx_descriptor *
745 at_xdmac_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
746 			 size_t buf_len, size_t period_len,
747 			 enum dma_transfer_direction direction,
748 			 unsigned long flags)
749 {
750 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
751 	struct at_xdmac_desc	*first = NULL, *prev = NULL;
752 	unsigned int		periods = buf_len / period_len;
753 	int			i;
754 	unsigned long		irqflags;
755 
756 	dev_dbg(chan2dev(chan), "%s: buf_addr=%pad, buf_len=%zd, period_len=%zd, dir=%s, flags=0x%lx\n",
757 		__func__, &buf_addr, buf_len, period_len,
758 		direction == DMA_MEM_TO_DEV ? "mem2per" : "per2mem", flags);
759 
760 	if (!is_slave_direction(direction)) {
761 		dev_err(chan2dev(chan), "invalid DMA direction\n");
762 		return NULL;
763 	}
764 
765 	if (test_and_set_bit(AT_XDMAC_CHAN_IS_CYCLIC, &atchan->status)) {
766 		dev_err(chan2dev(chan), "channel currently used\n");
767 		return NULL;
768 	}
769 
770 	if (at_xdmac_compute_chan_conf(chan, direction))
771 		return NULL;
772 
773 	for (i = 0; i < periods; i++) {
774 		struct at_xdmac_desc	*desc = NULL;
775 
776 		spin_lock_irqsave(&atchan->lock, irqflags);
777 		desc = at_xdmac_get_desc(atchan);
778 		if (!desc) {
779 			dev_err(chan2dev(chan), "can't get descriptor\n");
780 			if (first)
781 				list_splice_init(&first->descs_list, &atchan->free_descs_list);
782 			spin_unlock_irqrestore(&atchan->lock, irqflags);
783 			return NULL;
784 		}
785 		spin_unlock_irqrestore(&atchan->lock, irqflags);
786 		dev_dbg(chan2dev(chan),
787 			"%s: desc=0x%p, tx_dma_desc.phys=%pad\n",
788 			__func__, desc, &desc->tx_dma_desc.phys);
789 
790 		if (direction == DMA_DEV_TO_MEM) {
791 			desc->lld.mbr_sa = atchan->sconfig.src_addr;
792 			desc->lld.mbr_da = buf_addr + i * period_len;
793 		} else {
794 			desc->lld.mbr_sa = buf_addr + i * period_len;
795 			desc->lld.mbr_da = atchan->sconfig.dst_addr;
796 		}
797 		desc->lld.mbr_cfg = atchan->cfg;
798 		desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV1
799 			| AT_XDMAC_MBR_UBC_NDEN
800 			| AT_XDMAC_MBR_UBC_NSEN
801 			| period_len >> at_xdmac_get_dwidth(desc->lld.mbr_cfg);
802 
803 		dev_dbg(chan2dev(chan),
804 			 "%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x\n",
805 			 __func__, &desc->lld.mbr_sa, &desc->lld.mbr_da, desc->lld.mbr_ubc);
806 
807 		/* Chain lld. */
808 		if (prev)
809 			at_xdmac_queue_desc(chan, prev, desc);
810 
811 		prev = desc;
812 		if (!first)
813 			first = desc;
814 
815 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
816 			 __func__, desc, first);
817 		list_add_tail(&desc->desc_node, &first->descs_list);
818 	}
819 
820 	at_xdmac_queue_desc(chan, prev, first);
821 	first->tx_dma_desc.flags = flags;
822 	first->xfer_size = buf_len;
823 	first->direction = direction;
824 
825 	return &first->tx_dma_desc;
826 }
827 
828 static inline u32 at_xdmac_align_width(struct dma_chan *chan, dma_addr_t addr)
829 {
830 	u32 width;
831 
832 	/*
833 	 * Check address alignment to select the greater data width we
834 	 * can use.
835 	 *
836 	 * Some XDMAC implementations don't provide dword transfer, in
837 	 * this case selecting dword has the same behavior as
838 	 * selecting word transfers.
839 	 */
840 	if (!(addr & 7)) {
841 		width = AT_XDMAC_CC_DWIDTH_DWORD;
842 		dev_dbg(chan2dev(chan), "%s: dwidth: double word\n", __func__);
843 	} else if (!(addr & 3)) {
844 		width = AT_XDMAC_CC_DWIDTH_WORD;
845 		dev_dbg(chan2dev(chan), "%s: dwidth: word\n", __func__);
846 	} else if (!(addr & 1)) {
847 		width = AT_XDMAC_CC_DWIDTH_HALFWORD;
848 		dev_dbg(chan2dev(chan), "%s: dwidth: half word\n", __func__);
849 	} else {
850 		width = AT_XDMAC_CC_DWIDTH_BYTE;
851 		dev_dbg(chan2dev(chan), "%s: dwidth: byte\n", __func__);
852 	}
853 
854 	return width;
855 }
856 
857 static struct at_xdmac_desc *
858 at_xdmac_interleaved_queue_desc(struct dma_chan *chan,
859 				struct at_xdmac_chan *atchan,
860 				struct at_xdmac_desc *prev,
861 				dma_addr_t src, dma_addr_t dst,
862 				struct dma_interleaved_template *xt,
863 				struct data_chunk *chunk)
864 {
865 	struct at_xdmac_desc	*desc;
866 	u32			dwidth;
867 	unsigned long		flags;
868 	size_t			ublen;
869 	/*
870 	 * WARNING: The channel configuration is set here since there is no
871 	 * dmaengine_slave_config call in this case. Moreover we don't know the
872 	 * direction, it involves we can't dynamically set the source and dest
873 	 * interface so we have to use the same one. Only interface 0 allows EBI
874 	 * access. Hopefully we can access DDR through both ports (at least on
875 	 * SAMA5D4x), so we can use the same interface for source and dest,
876 	 * that solves the fact we don't know the direction.
877 	 * ERRATA: Even if useless for memory transfers, the PERID has to not
878 	 * match the one of another channel. If not, it could lead to spurious
879 	 * flag status.
880 	 */
881 	u32			chan_cc = AT_XDMAC_CC_PERID(0x3f)
882 					| AT_XDMAC_CC_DIF(0)
883 					| AT_XDMAC_CC_SIF(0)
884 					| AT_XDMAC_CC_MBSIZE_SIXTEEN
885 					| AT_XDMAC_CC_TYPE_MEM_TRAN;
886 
887 	dwidth = at_xdmac_align_width(chan, src | dst | chunk->size);
888 	if (chunk->size >= (AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth)) {
889 		dev_dbg(chan2dev(chan),
890 			"%s: chunk too big (%zu, max size %lu)...\n",
891 			__func__, chunk->size,
892 			AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth);
893 		return NULL;
894 	}
895 
896 	if (prev)
897 		dev_dbg(chan2dev(chan),
898 			"Adding items at the end of desc 0x%p\n", prev);
899 
900 	if (xt->src_inc) {
901 		if (xt->src_sgl)
902 			chan_cc |=  AT_XDMAC_CC_SAM_UBS_AM;
903 		else
904 			chan_cc |=  AT_XDMAC_CC_SAM_INCREMENTED_AM;
905 	}
906 
907 	if (xt->dst_inc) {
908 		if (xt->dst_sgl)
909 			chan_cc |=  AT_XDMAC_CC_DAM_UBS_AM;
910 		else
911 			chan_cc |=  AT_XDMAC_CC_DAM_INCREMENTED_AM;
912 	}
913 
914 	spin_lock_irqsave(&atchan->lock, flags);
915 	desc = at_xdmac_get_desc(atchan);
916 	spin_unlock_irqrestore(&atchan->lock, flags);
917 	if (!desc) {
918 		dev_err(chan2dev(chan), "can't get descriptor\n");
919 		return NULL;
920 	}
921 
922 	chan_cc |= AT_XDMAC_CC_DWIDTH(dwidth);
923 
924 	ublen = chunk->size >> dwidth;
925 
926 	desc->lld.mbr_sa = src;
927 	desc->lld.mbr_da = dst;
928 	desc->lld.mbr_sus = dmaengine_get_src_icg(xt, chunk);
929 	desc->lld.mbr_dus = dmaengine_get_dst_icg(xt, chunk);
930 
931 	desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV3
932 		| AT_XDMAC_MBR_UBC_NDEN
933 		| AT_XDMAC_MBR_UBC_NSEN
934 		| ublen;
935 	desc->lld.mbr_cfg = chan_cc;
936 
937 	dev_dbg(chan2dev(chan),
938 		"%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x, mbr_cfg=0x%08x\n",
939 		__func__, &desc->lld.mbr_sa, &desc->lld.mbr_da,
940 		desc->lld.mbr_ubc, desc->lld.mbr_cfg);
941 
942 	/* Chain lld. */
943 	if (prev)
944 		at_xdmac_queue_desc(chan, prev, desc);
945 
946 	return desc;
947 }
948 
949 static struct dma_async_tx_descriptor *
950 at_xdmac_prep_interleaved(struct dma_chan *chan,
951 			  struct dma_interleaved_template *xt,
952 			  unsigned long flags)
953 {
954 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
955 	struct at_xdmac_desc	*prev = NULL, *first = NULL;
956 	dma_addr_t		dst_addr, src_addr;
957 	size_t			src_skip = 0, dst_skip = 0, len = 0;
958 	struct data_chunk	*chunk;
959 	int			i;
960 
961 	if (!xt || !xt->numf || (xt->dir != DMA_MEM_TO_MEM))
962 		return NULL;
963 
964 	/*
965 	 * TODO: Handle the case where we have to repeat a chain of
966 	 * descriptors...
967 	 */
968 	if ((xt->numf > 1) && (xt->frame_size > 1))
969 		return NULL;
970 
971 	dev_dbg(chan2dev(chan), "%s: src=%pad, dest=%pad, numf=%zu, frame_size=%zu, flags=0x%lx\n",
972 		__func__, &xt->src_start, &xt->dst_start,	xt->numf,
973 		xt->frame_size, flags);
974 
975 	src_addr = xt->src_start;
976 	dst_addr = xt->dst_start;
977 
978 	if (xt->numf > 1) {
979 		first = at_xdmac_interleaved_queue_desc(chan, atchan,
980 							NULL,
981 							src_addr, dst_addr,
982 							xt, xt->sgl);
983 
984 		/* Length of the block is (BLEN+1) microblocks. */
985 		for (i = 0; i < xt->numf - 1; i++)
986 			at_xdmac_increment_block_count(chan, first);
987 
988 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
989 			__func__, first, first);
990 		list_add_tail(&first->desc_node, &first->descs_list);
991 	} else {
992 		for (i = 0; i < xt->frame_size; i++) {
993 			size_t src_icg = 0, dst_icg = 0;
994 			struct at_xdmac_desc *desc;
995 
996 			chunk = xt->sgl + i;
997 
998 			dst_icg = dmaengine_get_dst_icg(xt, chunk);
999 			src_icg = dmaengine_get_src_icg(xt, chunk);
1000 
1001 			src_skip = chunk->size + src_icg;
1002 			dst_skip = chunk->size + dst_icg;
1003 
1004 			dev_dbg(chan2dev(chan),
1005 				"%s: chunk size=%zu, src icg=%zu, dst icg=%zu\n",
1006 				__func__, chunk->size, src_icg, dst_icg);
1007 
1008 			desc = at_xdmac_interleaved_queue_desc(chan, atchan,
1009 							       prev,
1010 							       src_addr, dst_addr,
1011 							       xt, chunk);
1012 			if (!desc) {
1013 				list_splice_init(&first->descs_list,
1014 						 &atchan->free_descs_list);
1015 				return NULL;
1016 			}
1017 
1018 			if (!first)
1019 				first = desc;
1020 
1021 			dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
1022 				__func__, desc, first);
1023 			list_add_tail(&desc->desc_node, &first->descs_list);
1024 
1025 			if (xt->src_sgl)
1026 				src_addr += src_skip;
1027 
1028 			if (xt->dst_sgl)
1029 				dst_addr += dst_skip;
1030 
1031 			len += chunk->size;
1032 			prev = desc;
1033 		}
1034 	}
1035 
1036 	first->tx_dma_desc.cookie = -EBUSY;
1037 	first->tx_dma_desc.flags = flags;
1038 	first->xfer_size = len;
1039 
1040 	return &first->tx_dma_desc;
1041 }
1042 
1043 static struct dma_async_tx_descriptor *
1044 at_xdmac_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1045 			 size_t len, unsigned long flags)
1046 {
1047 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1048 	struct at_xdmac_desc	*first = NULL, *prev = NULL;
1049 	size_t			remaining_size = len, xfer_size = 0, ublen;
1050 	dma_addr_t		src_addr = src, dst_addr = dest;
1051 	u32			dwidth;
1052 	/*
1053 	 * WARNING: We don't know the direction, it involves we can't
1054 	 * dynamically set the source and dest interface so we have to use the
1055 	 * same one. Only interface 0 allows EBI access. Hopefully we can
1056 	 * access DDR through both ports (at least on SAMA5D4x), so we can use
1057 	 * the same interface for source and dest, that solves the fact we
1058 	 * don't know the direction.
1059 	 * ERRATA: Even if useless for memory transfers, the PERID has to not
1060 	 * match the one of another channel. If not, it could lead to spurious
1061 	 * flag status.
1062 	 */
1063 	u32			chan_cc = AT_XDMAC_CC_PERID(0x3f)
1064 					| AT_XDMAC_CC_DAM_INCREMENTED_AM
1065 					| AT_XDMAC_CC_SAM_INCREMENTED_AM
1066 					| AT_XDMAC_CC_DIF(0)
1067 					| AT_XDMAC_CC_SIF(0)
1068 					| AT_XDMAC_CC_MBSIZE_SIXTEEN
1069 					| AT_XDMAC_CC_TYPE_MEM_TRAN;
1070 	unsigned long		irqflags;
1071 
1072 	dev_dbg(chan2dev(chan), "%s: src=%pad, dest=%pad, len=%zd, flags=0x%lx\n",
1073 		__func__, &src, &dest, len, flags);
1074 
1075 	if (unlikely(!len))
1076 		return NULL;
1077 
1078 	dwidth = at_xdmac_align_width(chan, src_addr | dst_addr);
1079 
1080 	/* Prepare descriptors. */
1081 	while (remaining_size) {
1082 		struct at_xdmac_desc	*desc = NULL;
1083 
1084 		dev_dbg(chan2dev(chan), "%s: remaining_size=%zu\n", __func__, remaining_size);
1085 
1086 		spin_lock_irqsave(&atchan->lock, irqflags);
1087 		desc = at_xdmac_get_desc(atchan);
1088 		spin_unlock_irqrestore(&atchan->lock, irqflags);
1089 		if (!desc) {
1090 			dev_err(chan2dev(chan), "can't get descriptor\n");
1091 			if (first)
1092 				list_splice_init(&first->descs_list, &atchan->free_descs_list);
1093 			return NULL;
1094 		}
1095 
1096 		/* Update src and dest addresses. */
1097 		src_addr += xfer_size;
1098 		dst_addr += xfer_size;
1099 
1100 		if (remaining_size >= AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth)
1101 			xfer_size = AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth;
1102 		else
1103 			xfer_size = remaining_size;
1104 
1105 		dev_dbg(chan2dev(chan), "%s: xfer_size=%zu\n", __func__, xfer_size);
1106 
1107 		/* Check remaining length and change data width if needed. */
1108 		dwidth = at_xdmac_align_width(chan,
1109 					      src_addr | dst_addr | xfer_size);
1110 		chan_cc &= ~AT_XDMAC_CC_DWIDTH_MASK;
1111 		chan_cc |= AT_XDMAC_CC_DWIDTH(dwidth);
1112 
1113 		ublen = xfer_size >> dwidth;
1114 		remaining_size -= xfer_size;
1115 
1116 		desc->lld.mbr_sa = src_addr;
1117 		desc->lld.mbr_da = dst_addr;
1118 		desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV2
1119 			| AT_XDMAC_MBR_UBC_NDEN
1120 			| AT_XDMAC_MBR_UBC_NSEN
1121 			| ublen;
1122 		desc->lld.mbr_cfg = chan_cc;
1123 
1124 		dev_dbg(chan2dev(chan),
1125 			 "%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x, mbr_cfg=0x%08x\n",
1126 			 __func__, &desc->lld.mbr_sa, &desc->lld.mbr_da, desc->lld.mbr_ubc, desc->lld.mbr_cfg);
1127 
1128 		/* Chain lld. */
1129 		if (prev)
1130 			at_xdmac_queue_desc(chan, prev, desc);
1131 
1132 		prev = desc;
1133 		if (!first)
1134 			first = desc;
1135 
1136 		dev_dbg(chan2dev(chan), "%s: add desc 0x%p to descs_list 0x%p\n",
1137 			 __func__, desc, first);
1138 		list_add_tail(&desc->desc_node, &first->descs_list);
1139 	}
1140 
1141 	first->tx_dma_desc.flags = flags;
1142 	first->xfer_size = len;
1143 
1144 	return &first->tx_dma_desc;
1145 }
1146 
1147 static struct at_xdmac_desc *at_xdmac_memset_create_desc(struct dma_chan *chan,
1148 							 struct at_xdmac_chan *atchan,
1149 							 dma_addr_t dst_addr,
1150 							 size_t len,
1151 							 int value)
1152 {
1153 	struct at_xdmac_desc	*desc;
1154 	unsigned long		flags;
1155 	size_t			ublen;
1156 	u32			dwidth;
1157 	/*
1158 	 * WARNING: The channel configuration is set here since there is no
1159 	 * dmaengine_slave_config call in this case. Moreover we don't know the
1160 	 * direction, it involves we can't dynamically set the source and dest
1161 	 * interface so we have to use the same one. Only interface 0 allows EBI
1162 	 * access. Hopefully we can access DDR through both ports (at least on
1163 	 * SAMA5D4x), so we can use the same interface for source and dest,
1164 	 * that solves the fact we don't know the direction.
1165 	 * ERRATA: Even if useless for memory transfers, the PERID has to not
1166 	 * match the one of another channel. If not, it could lead to spurious
1167 	 * flag status.
1168 	 */
1169 	u32			chan_cc = AT_XDMAC_CC_PERID(0x3f)
1170 					| AT_XDMAC_CC_DAM_UBS_AM
1171 					| AT_XDMAC_CC_SAM_INCREMENTED_AM
1172 					| AT_XDMAC_CC_DIF(0)
1173 					| AT_XDMAC_CC_SIF(0)
1174 					| AT_XDMAC_CC_MBSIZE_SIXTEEN
1175 					| AT_XDMAC_CC_MEMSET_HW_MODE
1176 					| AT_XDMAC_CC_TYPE_MEM_TRAN;
1177 
1178 	dwidth = at_xdmac_align_width(chan, dst_addr);
1179 
1180 	if (len >= (AT_XDMAC_MBR_UBC_UBLEN_MAX << dwidth)) {
1181 		dev_err(chan2dev(chan),
1182 			"%s: Transfer too large, aborting...\n",
1183 			__func__);
1184 		return NULL;
1185 	}
1186 
1187 	spin_lock_irqsave(&atchan->lock, flags);
1188 	desc = at_xdmac_get_desc(atchan);
1189 	spin_unlock_irqrestore(&atchan->lock, flags);
1190 	if (!desc) {
1191 		dev_err(chan2dev(chan), "can't get descriptor\n");
1192 		return NULL;
1193 	}
1194 
1195 	chan_cc |= AT_XDMAC_CC_DWIDTH(dwidth);
1196 
1197 	ublen = len >> dwidth;
1198 
1199 	desc->lld.mbr_da = dst_addr;
1200 	desc->lld.mbr_ds = value;
1201 	desc->lld.mbr_ubc = AT_XDMAC_MBR_UBC_NDV3
1202 		| AT_XDMAC_MBR_UBC_NDEN
1203 		| AT_XDMAC_MBR_UBC_NSEN
1204 		| ublen;
1205 	desc->lld.mbr_cfg = chan_cc;
1206 
1207 	dev_dbg(chan2dev(chan),
1208 		"%s: lld: mbr_da=%pad, mbr_ds=0x%08x, mbr_ubc=0x%08x, mbr_cfg=0x%08x\n",
1209 		__func__, &desc->lld.mbr_da, desc->lld.mbr_ds, desc->lld.mbr_ubc,
1210 		desc->lld.mbr_cfg);
1211 
1212 	return desc;
1213 }
1214 
1215 static struct dma_async_tx_descriptor *
1216 at_xdmac_prep_dma_memset(struct dma_chan *chan, dma_addr_t dest, int value,
1217 			 size_t len, unsigned long flags)
1218 {
1219 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1220 	struct at_xdmac_desc	*desc;
1221 
1222 	dev_dbg(chan2dev(chan), "%s: dest=%pad, len=%zu, pattern=0x%x, flags=0x%lx\n",
1223 		__func__, &dest, len, value, flags);
1224 
1225 	if (unlikely(!len))
1226 		return NULL;
1227 
1228 	desc = at_xdmac_memset_create_desc(chan, atchan, dest, len, value);
1229 	list_add_tail(&desc->desc_node, &desc->descs_list);
1230 
1231 	desc->tx_dma_desc.cookie = -EBUSY;
1232 	desc->tx_dma_desc.flags = flags;
1233 	desc->xfer_size = len;
1234 
1235 	return &desc->tx_dma_desc;
1236 }
1237 
1238 static struct dma_async_tx_descriptor *
1239 at_xdmac_prep_dma_memset_sg(struct dma_chan *chan, struct scatterlist *sgl,
1240 			    unsigned int sg_len, int value,
1241 			    unsigned long flags)
1242 {
1243 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1244 	struct at_xdmac_desc	*desc, *pdesc = NULL,
1245 				*ppdesc = NULL, *first = NULL;
1246 	struct scatterlist	*sg, *psg = NULL, *ppsg = NULL;
1247 	size_t			stride = 0, pstride = 0, len = 0;
1248 	int			i;
1249 
1250 	if (!sgl)
1251 		return NULL;
1252 
1253 	dev_dbg(chan2dev(chan), "%s: sg_len=%d, value=0x%x, flags=0x%lx\n",
1254 		__func__, sg_len, value, flags);
1255 
1256 	/* Prepare descriptors. */
1257 	for_each_sg(sgl, sg, sg_len, i) {
1258 		dev_dbg(chan2dev(chan), "%s: dest=%pad, len=%d, pattern=0x%x, flags=0x%lx\n",
1259 			__func__, &sg_dma_address(sg), sg_dma_len(sg),
1260 			value, flags);
1261 		desc = at_xdmac_memset_create_desc(chan, atchan,
1262 						   sg_dma_address(sg),
1263 						   sg_dma_len(sg),
1264 						   value);
1265 		if (!desc && first)
1266 			list_splice_init(&first->descs_list,
1267 					 &atchan->free_descs_list);
1268 
1269 		if (!first)
1270 			first = desc;
1271 
1272 		/* Update our strides */
1273 		pstride = stride;
1274 		if (psg)
1275 			stride = sg_dma_address(sg) -
1276 				(sg_dma_address(psg) + sg_dma_len(psg));
1277 
1278 		/*
1279 		 * The scatterlist API gives us only the address and
1280 		 * length of each elements.
1281 		 *
1282 		 * Unfortunately, we don't have the stride, which we
1283 		 * will need to compute.
1284 		 *
1285 		 * That make us end up in a situation like this one:
1286 		 *    len    stride    len    stride    len
1287 		 * +-------+        +-------+        +-------+
1288 		 * |  N-2  |        |  N-1  |        |   N   |
1289 		 * +-------+        +-------+        +-------+
1290 		 *
1291 		 * We need all these three elements (N-2, N-1 and N)
1292 		 * to actually take the decision on whether we need to
1293 		 * queue N-1 or reuse N-2.
1294 		 *
1295 		 * We will only consider N if it is the last element.
1296 		 */
1297 		if (ppdesc && pdesc) {
1298 			if ((stride == pstride) &&
1299 			    (sg_dma_len(ppsg) == sg_dma_len(psg))) {
1300 				dev_dbg(chan2dev(chan),
1301 					"%s: desc 0x%p can be merged with desc 0x%p\n",
1302 					__func__, pdesc, ppdesc);
1303 
1304 				/*
1305 				 * Increment the block count of the
1306 				 * N-2 descriptor
1307 				 */
1308 				at_xdmac_increment_block_count(chan, ppdesc);
1309 				ppdesc->lld.mbr_dus = stride;
1310 
1311 				/*
1312 				 * Put back the N-1 descriptor in the
1313 				 * free descriptor list
1314 				 */
1315 				list_add_tail(&pdesc->desc_node,
1316 					      &atchan->free_descs_list);
1317 
1318 				/*
1319 				 * Make our N-1 descriptor pointer
1320 				 * point to the N-2 since they were
1321 				 * actually merged.
1322 				 */
1323 				pdesc = ppdesc;
1324 
1325 			/*
1326 			 * Rule out the case where we don't have
1327 			 * pstride computed yet (our second sg
1328 			 * element)
1329 			 *
1330 			 * We also want to catch the case where there
1331 			 * would be a negative stride,
1332 			 */
1333 			} else if (pstride ||
1334 				   sg_dma_address(sg) < sg_dma_address(psg)) {
1335 				/*
1336 				 * Queue the N-1 descriptor after the
1337 				 * N-2
1338 				 */
1339 				at_xdmac_queue_desc(chan, ppdesc, pdesc);
1340 
1341 				/*
1342 				 * Add the N-1 descriptor to the list
1343 				 * of the descriptors used for this
1344 				 * transfer
1345 				 */
1346 				list_add_tail(&desc->desc_node,
1347 					      &first->descs_list);
1348 				dev_dbg(chan2dev(chan),
1349 					"%s: add desc 0x%p to descs_list 0x%p\n",
1350 					__func__, desc, first);
1351 			}
1352 		}
1353 
1354 		/*
1355 		 * If we are the last element, just see if we have the
1356 		 * same size than the previous element.
1357 		 *
1358 		 * If so, we can merge it with the previous descriptor
1359 		 * since we don't care about the stride anymore.
1360 		 */
1361 		if ((i == (sg_len - 1)) &&
1362 		    sg_dma_len(psg) == sg_dma_len(sg)) {
1363 			dev_dbg(chan2dev(chan),
1364 				"%s: desc 0x%p can be merged with desc 0x%p\n",
1365 				__func__, desc, pdesc);
1366 
1367 			/*
1368 			 * Increment the block count of the N-1
1369 			 * descriptor
1370 			 */
1371 			at_xdmac_increment_block_count(chan, pdesc);
1372 			pdesc->lld.mbr_dus = stride;
1373 
1374 			/*
1375 			 * Put back the N descriptor in the free
1376 			 * descriptor list
1377 			 */
1378 			list_add_tail(&desc->desc_node,
1379 				      &atchan->free_descs_list);
1380 		}
1381 
1382 		/* Update our descriptors */
1383 		ppdesc = pdesc;
1384 		pdesc = desc;
1385 
1386 		/* Update our scatter pointers */
1387 		ppsg = psg;
1388 		psg = sg;
1389 
1390 		len += sg_dma_len(sg);
1391 	}
1392 
1393 	first->tx_dma_desc.cookie = -EBUSY;
1394 	first->tx_dma_desc.flags = flags;
1395 	first->xfer_size = len;
1396 
1397 	return &first->tx_dma_desc;
1398 }
1399 
1400 static enum dma_status
1401 at_xdmac_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
1402 		struct dma_tx_state *txstate)
1403 {
1404 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1405 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1406 	struct at_xdmac_desc	*desc, *_desc;
1407 	struct list_head	*descs_list;
1408 	enum dma_status		ret;
1409 	int			residue, retry;
1410 	u32			cur_nda, check_nda, cur_ubc, mask, value;
1411 	u8			dwidth = 0;
1412 	unsigned long		flags;
1413 	bool			initd;
1414 
1415 	ret = dma_cookie_status(chan, cookie, txstate);
1416 	if (ret == DMA_COMPLETE)
1417 		return ret;
1418 
1419 	if (!txstate)
1420 		return ret;
1421 
1422 	spin_lock_irqsave(&atchan->lock, flags);
1423 
1424 	desc = list_first_entry(&atchan->xfers_list, struct at_xdmac_desc, xfer_node);
1425 
1426 	/*
1427 	 * If the transfer has not been started yet, don't need to compute the
1428 	 * residue, it's the transfer length.
1429 	 */
1430 	if (!desc->active_xfer) {
1431 		dma_set_residue(txstate, desc->xfer_size);
1432 		goto spin_unlock;
1433 	}
1434 
1435 	residue = desc->xfer_size;
1436 	/*
1437 	 * Flush FIFO: only relevant when the transfer is source peripheral
1438 	 * synchronized. Flush is needed before reading CUBC because data in
1439 	 * the FIFO are not reported by CUBC. Reporting a residue of the
1440 	 * transfer length while we have data in FIFO can cause issue.
1441 	 * Usecase: atmel USART has a timeout which means I have received
1442 	 * characters but there is no more character received for a while. On
1443 	 * timeout, it requests the residue. If the data are in the DMA FIFO,
1444 	 * we will return a residue of the transfer length. It means no data
1445 	 * received. If an application is waiting for these data, it will hang
1446 	 * since we won't have another USART timeout without receiving new
1447 	 * data.
1448 	 */
1449 	mask = AT_XDMAC_CC_TYPE | AT_XDMAC_CC_DSYNC;
1450 	value = AT_XDMAC_CC_TYPE_PER_TRAN | AT_XDMAC_CC_DSYNC_PER2MEM;
1451 	if ((desc->lld.mbr_cfg & mask) == value) {
1452 		at_xdmac_write(atxdmac, AT_XDMAC_GSWF, atchan->mask);
1453 		while (!(at_xdmac_chan_read(atchan, AT_XDMAC_CIS) & AT_XDMAC_CIS_FIS))
1454 			cpu_relax();
1455 	}
1456 
1457 	/*
1458 	 * The easiest way to compute the residue should be to pause the DMA
1459 	 * but doing this can lead to miss some data as some devices don't
1460 	 * have FIFO.
1461 	 * We need to read several registers because:
1462 	 * - DMA is running therefore a descriptor change is possible while
1463 	 * reading these registers
1464 	 * - When the block transfer is done, the value of the CUBC register
1465 	 * is set to its initial value until the fetch of the next descriptor.
1466 	 * This value will corrupt the residue calculation so we have to skip
1467 	 * it.
1468 	 *
1469 	 * INITD --------                    ------------
1470 	 *              |____________________|
1471 	 *       _______________________  _______________
1472 	 * NDA       @desc2             \/   @desc3
1473 	 *       _______________________/\_______________
1474 	 *       __________  ___________  _______________
1475 	 * CUBC       0    \/ MAX desc1 \/  MAX desc2
1476 	 *       __________/\___________/\_______________
1477 	 *
1478 	 * Since descriptors are aligned on 64 bits, we can assume that
1479 	 * the update of NDA and CUBC is atomic.
1480 	 * Memory barriers are used to ensure the read order of the registers.
1481 	 * A max number of retries is set because unlikely it could never ends.
1482 	 */
1483 	for (retry = 0; retry < AT_XDMAC_RESIDUE_MAX_RETRIES; retry++) {
1484 		check_nda = at_xdmac_chan_read(atchan, AT_XDMAC_CNDA) & 0xfffffffc;
1485 		rmb();
1486 		cur_ubc = at_xdmac_chan_read(atchan, AT_XDMAC_CUBC);
1487 		rmb();
1488 		initd = !!(at_xdmac_chan_read(atchan, AT_XDMAC_CC) & AT_XDMAC_CC_INITD);
1489 		rmb();
1490 		cur_nda = at_xdmac_chan_read(atchan, AT_XDMAC_CNDA) & 0xfffffffc;
1491 		rmb();
1492 
1493 		if ((check_nda == cur_nda) && initd)
1494 			break;
1495 	}
1496 
1497 	if (unlikely(retry >= AT_XDMAC_RESIDUE_MAX_RETRIES)) {
1498 		ret = DMA_ERROR;
1499 		goto spin_unlock;
1500 	}
1501 
1502 	/*
1503 	 * Flush FIFO: only relevant when the transfer is source peripheral
1504 	 * synchronized. Another flush is needed here because CUBC is updated
1505 	 * when the controller sends the data write command. It can lead to
1506 	 * report data that are not written in the memory or the device. The
1507 	 * FIFO flush ensures that data are really written.
1508 	 */
1509 	if ((desc->lld.mbr_cfg & mask) == value) {
1510 		at_xdmac_write(atxdmac, AT_XDMAC_GSWF, atchan->mask);
1511 		while (!(at_xdmac_chan_read(atchan, AT_XDMAC_CIS) & AT_XDMAC_CIS_FIS))
1512 			cpu_relax();
1513 	}
1514 
1515 	/*
1516 	 * Remove size of all microblocks already transferred and the current
1517 	 * one. Then add the remaining size to transfer of the current
1518 	 * microblock.
1519 	 */
1520 	descs_list = &desc->descs_list;
1521 	list_for_each_entry_safe(desc, _desc, descs_list, desc_node) {
1522 		dwidth = at_xdmac_get_dwidth(desc->lld.mbr_cfg);
1523 		residue -= (desc->lld.mbr_ubc & 0xffffff) << dwidth;
1524 		if ((desc->lld.mbr_nda & 0xfffffffc) == cur_nda)
1525 			break;
1526 	}
1527 	residue += cur_ubc << dwidth;
1528 
1529 	dma_set_residue(txstate, residue);
1530 
1531 	dev_dbg(chan2dev(chan),
1532 		 "%s: desc=0x%p, tx_dma_desc.phys=%pad, tx_status=%d, cookie=%d, residue=%d\n",
1533 		 __func__, desc, &desc->tx_dma_desc.phys, ret, cookie, residue);
1534 
1535 spin_unlock:
1536 	spin_unlock_irqrestore(&atchan->lock, flags);
1537 	return ret;
1538 }
1539 
1540 /* Call must be protected by lock. */
1541 static void at_xdmac_remove_xfer(struct at_xdmac_chan *atchan,
1542 				    struct at_xdmac_desc *desc)
1543 {
1544 	dev_dbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, desc);
1545 
1546 	/*
1547 	 * Remove the transfer from the transfer list then move the transfer
1548 	 * descriptors into the free descriptors list.
1549 	 */
1550 	list_del(&desc->xfer_node);
1551 	list_splice_init(&desc->descs_list, &atchan->free_descs_list);
1552 }
1553 
1554 static void at_xdmac_advance_work(struct at_xdmac_chan *atchan)
1555 {
1556 	struct at_xdmac_desc	*desc;
1557 	unsigned long		flags;
1558 
1559 	spin_lock_irqsave(&atchan->lock, flags);
1560 
1561 	/*
1562 	 * If channel is enabled, do nothing, advance_work will be triggered
1563 	 * after the interruption.
1564 	 */
1565 	if (!at_xdmac_chan_is_enabled(atchan) && !list_empty(&atchan->xfers_list)) {
1566 		desc = list_first_entry(&atchan->xfers_list,
1567 					struct at_xdmac_desc,
1568 					xfer_node);
1569 		dev_vdbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, desc);
1570 		if (!desc->active_xfer)
1571 			at_xdmac_start_xfer(atchan, desc);
1572 	}
1573 
1574 	spin_unlock_irqrestore(&atchan->lock, flags);
1575 }
1576 
1577 static void at_xdmac_handle_cyclic(struct at_xdmac_chan *atchan)
1578 {
1579 	struct at_xdmac_desc		*desc;
1580 	struct dma_async_tx_descriptor	*txd;
1581 
1582 	desc = list_first_entry(&atchan->xfers_list, struct at_xdmac_desc, xfer_node);
1583 	txd = &desc->tx_dma_desc;
1584 
1585 	if (txd->flags & DMA_PREP_INTERRUPT)
1586 		dmaengine_desc_get_callback_invoke(txd, NULL);
1587 }
1588 
1589 static void at_xdmac_handle_error(struct at_xdmac_chan *atchan)
1590 {
1591 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1592 	struct at_xdmac_desc	*bad_desc;
1593 
1594 	/*
1595 	 * The descriptor currently at the head of the active list is
1596 	 * broken. Since we don't have any way to report errors, we'll
1597 	 * just have to scream loudly and try to continue with other
1598 	 * descriptors queued (if any).
1599 	 */
1600 	if (atchan->irq_status & AT_XDMAC_CIS_RBEIS)
1601 		dev_err(chan2dev(&atchan->chan), "read bus error!!!");
1602 	if (atchan->irq_status & AT_XDMAC_CIS_WBEIS)
1603 		dev_err(chan2dev(&atchan->chan), "write bus error!!!");
1604 	if (atchan->irq_status & AT_XDMAC_CIS_ROIS)
1605 		dev_err(chan2dev(&atchan->chan), "request overflow error!!!");
1606 
1607 	spin_lock_bh(&atchan->lock);
1608 
1609 	/* Channel must be disabled first as it's not done automatically */
1610 	at_xdmac_write(atxdmac, AT_XDMAC_GD, atchan->mask);
1611 	while (at_xdmac_read(atxdmac, AT_XDMAC_GS) & atchan->mask)
1612 		cpu_relax();
1613 
1614 	bad_desc = list_first_entry(&atchan->xfers_list,
1615 				    struct at_xdmac_desc,
1616 				    xfer_node);
1617 
1618 	spin_unlock_bh(&atchan->lock);
1619 
1620 	/* Print bad descriptor's details if needed */
1621 	dev_dbg(chan2dev(&atchan->chan),
1622 		"%s: lld: mbr_sa=%pad, mbr_da=%pad, mbr_ubc=0x%08x\n",
1623 		__func__, &bad_desc->lld.mbr_sa, &bad_desc->lld.mbr_da,
1624 		bad_desc->lld.mbr_ubc);
1625 
1626 	/* Then continue with usual descriptor management */
1627 }
1628 
1629 static void at_xdmac_tasklet(unsigned long data)
1630 {
1631 	struct at_xdmac_chan	*atchan = (struct at_xdmac_chan *)data;
1632 	struct at_xdmac_desc	*desc;
1633 	u32			error_mask;
1634 
1635 	dev_dbg(chan2dev(&atchan->chan), "%s: status=0x%08x\n",
1636 		__func__, atchan->irq_status);
1637 
1638 	error_mask = AT_XDMAC_CIS_RBEIS
1639 		     | AT_XDMAC_CIS_WBEIS
1640 		     | AT_XDMAC_CIS_ROIS;
1641 
1642 	if (at_xdmac_chan_is_cyclic(atchan)) {
1643 		at_xdmac_handle_cyclic(atchan);
1644 	} else if ((atchan->irq_status & AT_XDMAC_CIS_LIS)
1645 		   || (atchan->irq_status & error_mask)) {
1646 		struct dma_async_tx_descriptor  *txd;
1647 
1648 		if (atchan->irq_status & error_mask)
1649 			at_xdmac_handle_error(atchan);
1650 
1651 		spin_lock(&atchan->lock);
1652 		desc = list_first_entry(&atchan->xfers_list,
1653 					struct at_xdmac_desc,
1654 					xfer_node);
1655 		dev_vdbg(chan2dev(&atchan->chan), "%s: desc 0x%p\n", __func__, desc);
1656 		if (!desc->active_xfer) {
1657 			dev_err(chan2dev(&atchan->chan), "Xfer not active: exiting");
1658 			spin_unlock(&atchan->lock);
1659 			return;
1660 		}
1661 
1662 		txd = &desc->tx_dma_desc;
1663 
1664 		at_xdmac_remove_xfer(atchan, desc);
1665 		spin_unlock(&atchan->lock);
1666 
1667 		if (!at_xdmac_chan_is_cyclic(atchan)) {
1668 			dma_cookie_complete(txd);
1669 			if (txd->flags & DMA_PREP_INTERRUPT)
1670 				dmaengine_desc_get_callback_invoke(txd, NULL);
1671 		}
1672 
1673 		dma_run_dependencies(txd);
1674 
1675 		at_xdmac_advance_work(atchan);
1676 	}
1677 }
1678 
1679 static irqreturn_t at_xdmac_interrupt(int irq, void *dev_id)
1680 {
1681 	struct at_xdmac		*atxdmac = (struct at_xdmac *)dev_id;
1682 	struct at_xdmac_chan	*atchan;
1683 	u32			imr, status, pending;
1684 	u32			chan_imr, chan_status;
1685 	int			i, ret = IRQ_NONE;
1686 
1687 	do {
1688 		imr = at_xdmac_read(atxdmac, AT_XDMAC_GIM);
1689 		status = at_xdmac_read(atxdmac, AT_XDMAC_GIS);
1690 		pending = status & imr;
1691 
1692 		dev_vdbg(atxdmac->dma.dev,
1693 			 "%s: status=0x%08x, imr=0x%08x, pending=0x%08x\n",
1694 			 __func__, status, imr, pending);
1695 
1696 		if (!pending)
1697 			break;
1698 
1699 		/* We have to find which channel has generated the interrupt. */
1700 		for (i = 0; i < atxdmac->dma.chancnt; i++) {
1701 			if (!((1 << i) & pending))
1702 				continue;
1703 
1704 			atchan = &atxdmac->chan[i];
1705 			chan_imr = at_xdmac_chan_read(atchan, AT_XDMAC_CIM);
1706 			chan_status = at_xdmac_chan_read(atchan, AT_XDMAC_CIS);
1707 			atchan->irq_status = chan_status & chan_imr;
1708 			dev_vdbg(atxdmac->dma.dev,
1709 				 "%s: chan%d: imr=0x%x, status=0x%x\n",
1710 				 __func__, i, chan_imr, chan_status);
1711 			dev_vdbg(chan2dev(&atchan->chan),
1712 				 "%s: CC=0x%08x CNDA=0x%08x, CNDC=0x%08x, CSA=0x%08x, CDA=0x%08x, CUBC=0x%08x\n",
1713 				 __func__,
1714 				 at_xdmac_chan_read(atchan, AT_XDMAC_CC),
1715 				 at_xdmac_chan_read(atchan, AT_XDMAC_CNDA),
1716 				 at_xdmac_chan_read(atchan, AT_XDMAC_CNDC),
1717 				 at_xdmac_chan_read(atchan, AT_XDMAC_CSA),
1718 				 at_xdmac_chan_read(atchan, AT_XDMAC_CDA),
1719 				 at_xdmac_chan_read(atchan, AT_XDMAC_CUBC));
1720 
1721 			if (atchan->irq_status & (AT_XDMAC_CIS_RBEIS | AT_XDMAC_CIS_WBEIS))
1722 				at_xdmac_write(atxdmac, AT_XDMAC_GD, atchan->mask);
1723 
1724 			tasklet_schedule(&atchan->tasklet);
1725 			ret = IRQ_HANDLED;
1726 		}
1727 
1728 	} while (pending);
1729 
1730 	return ret;
1731 }
1732 
1733 static void at_xdmac_issue_pending(struct dma_chan *chan)
1734 {
1735 	struct at_xdmac_chan *atchan = to_at_xdmac_chan(chan);
1736 
1737 	dev_dbg(chan2dev(&atchan->chan), "%s\n", __func__);
1738 
1739 	if (!at_xdmac_chan_is_cyclic(atchan))
1740 		at_xdmac_advance_work(atchan);
1741 
1742 	return;
1743 }
1744 
1745 static int at_xdmac_device_config(struct dma_chan *chan,
1746 				  struct dma_slave_config *config)
1747 {
1748 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1749 	int ret;
1750 	unsigned long		flags;
1751 
1752 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1753 
1754 	spin_lock_irqsave(&atchan->lock, flags);
1755 	ret = at_xdmac_set_slave_config(chan, config);
1756 	spin_unlock_irqrestore(&atchan->lock, flags);
1757 
1758 	return ret;
1759 }
1760 
1761 static int at_xdmac_device_pause(struct dma_chan *chan)
1762 {
1763 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1764 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1765 	unsigned long		flags;
1766 
1767 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1768 
1769 	if (test_and_set_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status))
1770 		return 0;
1771 
1772 	spin_lock_irqsave(&atchan->lock, flags);
1773 	at_xdmac_write(atxdmac, AT_XDMAC_GRWS, atchan->mask);
1774 	while (at_xdmac_chan_read(atchan, AT_XDMAC_CC)
1775 	       & (AT_XDMAC_CC_WRIP | AT_XDMAC_CC_RDIP))
1776 		cpu_relax();
1777 	spin_unlock_irqrestore(&atchan->lock, flags);
1778 
1779 	return 0;
1780 }
1781 
1782 static int at_xdmac_device_resume(struct dma_chan *chan)
1783 {
1784 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1785 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1786 	unsigned long		flags;
1787 
1788 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1789 
1790 	spin_lock_irqsave(&atchan->lock, flags);
1791 	if (!at_xdmac_chan_is_paused(atchan)) {
1792 		spin_unlock_irqrestore(&atchan->lock, flags);
1793 		return 0;
1794 	}
1795 
1796 	at_xdmac_write(atxdmac, AT_XDMAC_GRWR, atchan->mask);
1797 	clear_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status);
1798 	spin_unlock_irqrestore(&atchan->lock, flags);
1799 
1800 	return 0;
1801 }
1802 
1803 static int at_xdmac_device_terminate_all(struct dma_chan *chan)
1804 {
1805 	struct at_xdmac_desc	*desc, *_desc;
1806 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1807 	struct at_xdmac		*atxdmac = to_at_xdmac(atchan->chan.device);
1808 	unsigned long		flags;
1809 
1810 	dev_dbg(chan2dev(chan), "%s\n", __func__);
1811 
1812 	spin_lock_irqsave(&atchan->lock, flags);
1813 	at_xdmac_write(atxdmac, AT_XDMAC_GD, atchan->mask);
1814 	while (at_xdmac_read(atxdmac, AT_XDMAC_GS) & atchan->mask)
1815 		cpu_relax();
1816 
1817 	/* Cancel all pending transfers. */
1818 	list_for_each_entry_safe(desc, _desc, &atchan->xfers_list, xfer_node)
1819 		at_xdmac_remove_xfer(atchan, desc);
1820 
1821 	clear_bit(AT_XDMAC_CHAN_IS_PAUSED, &atchan->status);
1822 	clear_bit(AT_XDMAC_CHAN_IS_CYCLIC, &atchan->status);
1823 	spin_unlock_irqrestore(&atchan->lock, flags);
1824 
1825 	return 0;
1826 }
1827 
1828 static int at_xdmac_alloc_chan_resources(struct dma_chan *chan)
1829 {
1830 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1831 	struct at_xdmac_desc	*desc;
1832 	int			i;
1833 	unsigned long		flags;
1834 
1835 	spin_lock_irqsave(&atchan->lock, flags);
1836 
1837 	if (at_xdmac_chan_is_enabled(atchan)) {
1838 		dev_err(chan2dev(chan),
1839 			"can't allocate channel resources (channel enabled)\n");
1840 		i = -EIO;
1841 		goto spin_unlock;
1842 	}
1843 
1844 	if (!list_empty(&atchan->free_descs_list)) {
1845 		dev_err(chan2dev(chan),
1846 			"can't allocate channel resources (channel not free from a previous use)\n");
1847 		i = -EIO;
1848 		goto spin_unlock;
1849 	}
1850 
1851 	for (i = 0; i < init_nr_desc_per_channel; i++) {
1852 		desc = at_xdmac_alloc_desc(chan, GFP_ATOMIC);
1853 		if (!desc) {
1854 			dev_warn(chan2dev(chan),
1855 				"only %d descriptors have been allocated\n", i);
1856 			break;
1857 		}
1858 		list_add_tail(&desc->desc_node, &atchan->free_descs_list);
1859 	}
1860 
1861 	dma_cookie_init(chan);
1862 
1863 	dev_dbg(chan2dev(chan), "%s: allocated %d descriptors\n", __func__, i);
1864 
1865 spin_unlock:
1866 	spin_unlock_irqrestore(&atchan->lock, flags);
1867 	return i;
1868 }
1869 
1870 static void at_xdmac_free_chan_resources(struct dma_chan *chan)
1871 {
1872 	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1873 	struct at_xdmac		*atxdmac = to_at_xdmac(chan->device);
1874 	struct at_xdmac_desc	*desc, *_desc;
1875 
1876 	list_for_each_entry_safe(desc, _desc, &atchan->free_descs_list, desc_node) {
1877 		dev_dbg(chan2dev(chan), "%s: freeing descriptor %p\n", __func__, desc);
1878 		list_del(&desc->desc_node);
1879 		dma_pool_free(atxdmac->at_xdmac_desc_pool, desc, desc->tx_dma_desc.phys);
1880 	}
1881 
1882 	return;
1883 }
1884 
1885 #ifdef CONFIG_PM
1886 static int atmel_xdmac_prepare(struct device *dev)
1887 {
1888 	struct at_xdmac		*atxdmac = dev_get_drvdata(dev);
1889 	struct dma_chan		*chan, *_chan;
1890 
1891 	list_for_each_entry_safe(chan, _chan, &atxdmac->dma.channels, device_node) {
1892 		struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1893 
1894 		/* Wait for transfer completion, except in cyclic case. */
1895 		if (at_xdmac_chan_is_enabled(atchan) && !at_xdmac_chan_is_cyclic(atchan))
1896 			return -EAGAIN;
1897 	}
1898 	return 0;
1899 }
1900 #else
1901 #	define atmel_xdmac_prepare NULL
1902 #endif
1903 
1904 #ifdef CONFIG_PM_SLEEP
1905 static int atmel_xdmac_suspend(struct device *dev)
1906 {
1907 	struct at_xdmac		*atxdmac = dev_get_drvdata(dev);
1908 	struct dma_chan		*chan, *_chan;
1909 
1910 	list_for_each_entry_safe(chan, _chan, &atxdmac->dma.channels, device_node) {
1911 		struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
1912 
1913 		atchan->save_cc = at_xdmac_chan_read(atchan, AT_XDMAC_CC);
1914 		if (at_xdmac_chan_is_cyclic(atchan)) {
1915 			if (!at_xdmac_chan_is_paused(atchan))
1916 				at_xdmac_device_pause(chan);
1917 			atchan->save_cim = at_xdmac_chan_read(atchan, AT_XDMAC_CIM);
1918 			atchan->save_cnda = at_xdmac_chan_read(atchan, AT_XDMAC_CNDA);
1919 			atchan->save_cndc = at_xdmac_chan_read(atchan, AT_XDMAC_CNDC);
1920 		}
1921 	}
1922 	atxdmac->save_gim = at_xdmac_read(atxdmac, AT_XDMAC_GIM);
1923 
1924 	at_xdmac_off(atxdmac);
1925 	clk_disable_unprepare(atxdmac->clk);
1926 	return 0;
1927 }
1928 
1929 static int atmel_xdmac_resume(struct device *dev)
1930 {
1931 	struct at_xdmac		*atxdmac = dev_get_drvdata(dev);
1932 	struct at_xdmac_chan	*atchan;
1933 	struct dma_chan		*chan, *_chan;
1934 	int			i;
1935 	int ret;
1936 
1937 	ret = clk_prepare_enable(atxdmac->clk);
1938 	if (ret)
1939 		return ret;
1940 
1941 	/* Clear pending interrupts. */
1942 	for (i = 0; i < atxdmac->dma.chancnt; i++) {
1943 		atchan = &atxdmac->chan[i];
1944 		while (at_xdmac_chan_read(atchan, AT_XDMAC_CIS))
1945 			cpu_relax();
1946 	}
1947 
1948 	at_xdmac_write(atxdmac, AT_XDMAC_GIE, atxdmac->save_gim);
1949 	list_for_each_entry_safe(chan, _chan, &atxdmac->dma.channels, device_node) {
1950 		atchan = to_at_xdmac_chan(chan);
1951 		at_xdmac_chan_write(atchan, AT_XDMAC_CC, atchan->save_cc);
1952 		if (at_xdmac_chan_is_cyclic(atchan)) {
1953 			if (at_xdmac_chan_is_paused(atchan))
1954 				at_xdmac_device_resume(chan);
1955 			at_xdmac_chan_write(atchan, AT_XDMAC_CNDA, atchan->save_cnda);
1956 			at_xdmac_chan_write(atchan, AT_XDMAC_CNDC, atchan->save_cndc);
1957 			at_xdmac_chan_write(atchan, AT_XDMAC_CIE, atchan->save_cim);
1958 			wmb();
1959 			at_xdmac_write(atxdmac, AT_XDMAC_GE, atchan->mask);
1960 		}
1961 	}
1962 	return 0;
1963 }
1964 #endif /* CONFIG_PM_SLEEP */
1965 
1966 static int at_xdmac_probe(struct platform_device *pdev)
1967 {
1968 	struct resource	*res;
1969 	struct at_xdmac	*atxdmac;
1970 	int		irq, size, nr_channels, i, ret;
1971 	void __iomem	*base;
1972 	u32		reg;
1973 
1974 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1975 	if (!res)
1976 		return -EINVAL;
1977 
1978 	irq = platform_get_irq(pdev, 0);
1979 	if (irq < 0)
1980 		return irq;
1981 
1982 	base = devm_ioremap_resource(&pdev->dev, res);
1983 	if (IS_ERR(base))
1984 		return PTR_ERR(base);
1985 
1986 	/*
1987 	 * Read number of xdmac channels, read helper function can't be used
1988 	 * since atxdmac is not yet allocated and we need to know the number
1989 	 * of channels to do the allocation.
1990 	 */
1991 	reg = readl_relaxed(base + AT_XDMAC_GTYPE);
1992 	nr_channels = AT_XDMAC_NB_CH(reg);
1993 	if (nr_channels > AT_XDMAC_MAX_CHAN) {
1994 		dev_err(&pdev->dev, "invalid number of channels (%u)\n",
1995 			nr_channels);
1996 		return -EINVAL;
1997 	}
1998 
1999 	size = sizeof(*atxdmac);
2000 	size += nr_channels * sizeof(struct at_xdmac_chan);
2001 	atxdmac = devm_kzalloc(&pdev->dev, size, GFP_KERNEL);
2002 	if (!atxdmac) {
2003 		dev_err(&pdev->dev, "can't allocate at_xdmac structure\n");
2004 		return -ENOMEM;
2005 	}
2006 
2007 	atxdmac->regs = base;
2008 	atxdmac->irq = irq;
2009 
2010 	atxdmac->clk = devm_clk_get(&pdev->dev, "dma_clk");
2011 	if (IS_ERR(atxdmac->clk)) {
2012 		dev_err(&pdev->dev, "can't get dma_clk\n");
2013 		return PTR_ERR(atxdmac->clk);
2014 	}
2015 
2016 	/* Do not use dev res to prevent races with tasklet */
2017 	ret = request_irq(atxdmac->irq, at_xdmac_interrupt, 0, "at_xdmac", atxdmac);
2018 	if (ret) {
2019 		dev_err(&pdev->dev, "can't request irq\n");
2020 		return ret;
2021 	}
2022 
2023 	ret = clk_prepare_enable(atxdmac->clk);
2024 	if (ret) {
2025 		dev_err(&pdev->dev, "can't prepare or enable clock\n");
2026 		goto err_free_irq;
2027 	}
2028 
2029 	atxdmac->at_xdmac_desc_pool =
2030 		dmam_pool_create(dev_name(&pdev->dev), &pdev->dev,
2031 				sizeof(struct at_xdmac_desc), 4, 0);
2032 	if (!atxdmac->at_xdmac_desc_pool) {
2033 		dev_err(&pdev->dev, "no memory for descriptors dma pool\n");
2034 		ret = -ENOMEM;
2035 		goto err_clk_disable;
2036 	}
2037 
2038 	dma_cap_set(DMA_CYCLIC, atxdmac->dma.cap_mask);
2039 	dma_cap_set(DMA_INTERLEAVE, atxdmac->dma.cap_mask);
2040 	dma_cap_set(DMA_MEMCPY, atxdmac->dma.cap_mask);
2041 	dma_cap_set(DMA_MEMSET, atxdmac->dma.cap_mask);
2042 	dma_cap_set(DMA_MEMSET_SG, atxdmac->dma.cap_mask);
2043 	dma_cap_set(DMA_SLAVE, atxdmac->dma.cap_mask);
2044 	/*
2045 	 * Without DMA_PRIVATE the driver is not able to allocate more than
2046 	 * one channel, second allocation fails in private_candidate.
2047 	 */
2048 	dma_cap_set(DMA_PRIVATE, atxdmac->dma.cap_mask);
2049 	atxdmac->dma.dev				= &pdev->dev;
2050 	atxdmac->dma.device_alloc_chan_resources	= at_xdmac_alloc_chan_resources;
2051 	atxdmac->dma.device_free_chan_resources		= at_xdmac_free_chan_resources;
2052 	atxdmac->dma.device_tx_status			= at_xdmac_tx_status;
2053 	atxdmac->dma.device_issue_pending		= at_xdmac_issue_pending;
2054 	atxdmac->dma.device_prep_dma_cyclic		= at_xdmac_prep_dma_cyclic;
2055 	atxdmac->dma.device_prep_interleaved_dma	= at_xdmac_prep_interleaved;
2056 	atxdmac->dma.device_prep_dma_memcpy		= at_xdmac_prep_dma_memcpy;
2057 	atxdmac->dma.device_prep_dma_memset		= at_xdmac_prep_dma_memset;
2058 	atxdmac->dma.device_prep_dma_memset_sg		= at_xdmac_prep_dma_memset_sg;
2059 	atxdmac->dma.device_prep_slave_sg		= at_xdmac_prep_slave_sg;
2060 	atxdmac->dma.device_config			= at_xdmac_device_config;
2061 	atxdmac->dma.device_pause			= at_xdmac_device_pause;
2062 	atxdmac->dma.device_resume			= at_xdmac_device_resume;
2063 	atxdmac->dma.device_terminate_all		= at_xdmac_device_terminate_all;
2064 	atxdmac->dma.src_addr_widths = AT_XDMAC_DMA_BUSWIDTHS;
2065 	atxdmac->dma.dst_addr_widths = AT_XDMAC_DMA_BUSWIDTHS;
2066 	atxdmac->dma.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2067 	atxdmac->dma.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2068 
2069 	/* Disable all chans and interrupts. */
2070 	at_xdmac_off(atxdmac);
2071 
2072 	/* Init channels. */
2073 	INIT_LIST_HEAD(&atxdmac->dma.channels);
2074 	for (i = 0; i < nr_channels; i++) {
2075 		struct at_xdmac_chan *atchan = &atxdmac->chan[i];
2076 
2077 		atchan->chan.device = &atxdmac->dma;
2078 		list_add_tail(&atchan->chan.device_node,
2079 			      &atxdmac->dma.channels);
2080 
2081 		atchan->ch_regs = at_xdmac_chan_reg_base(atxdmac, i);
2082 		atchan->mask = 1 << i;
2083 
2084 		spin_lock_init(&atchan->lock);
2085 		INIT_LIST_HEAD(&atchan->xfers_list);
2086 		INIT_LIST_HEAD(&atchan->free_descs_list);
2087 		tasklet_init(&atchan->tasklet, at_xdmac_tasklet,
2088 			     (unsigned long)atchan);
2089 
2090 		/* Clear pending interrupts. */
2091 		while (at_xdmac_chan_read(atchan, AT_XDMAC_CIS))
2092 			cpu_relax();
2093 	}
2094 	platform_set_drvdata(pdev, atxdmac);
2095 
2096 	ret = dma_async_device_register(&atxdmac->dma);
2097 	if (ret) {
2098 		dev_err(&pdev->dev, "fail to register DMA engine device\n");
2099 		goto err_clk_disable;
2100 	}
2101 
2102 	ret = of_dma_controller_register(pdev->dev.of_node,
2103 					 at_xdmac_xlate, atxdmac);
2104 	if (ret) {
2105 		dev_err(&pdev->dev, "could not register of dma controller\n");
2106 		goto err_dma_unregister;
2107 	}
2108 
2109 	dev_info(&pdev->dev, "%d channels, mapped at 0x%p\n",
2110 		 nr_channels, atxdmac->regs);
2111 
2112 	return 0;
2113 
2114 err_dma_unregister:
2115 	dma_async_device_unregister(&atxdmac->dma);
2116 err_clk_disable:
2117 	clk_disable_unprepare(atxdmac->clk);
2118 err_free_irq:
2119 	free_irq(atxdmac->irq, atxdmac);
2120 	return ret;
2121 }
2122 
2123 static int at_xdmac_remove(struct platform_device *pdev)
2124 {
2125 	struct at_xdmac	*atxdmac = (struct at_xdmac *)platform_get_drvdata(pdev);
2126 	int		i;
2127 
2128 	at_xdmac_off(atxdmac);
2129 	of_dma_controller_free(pdev->dev.of_node);
2130 	dma_async_device_unregister(&atxdmac->dma);
2131 	clk_disable_unprepare(atxdmac->clk);
2132 
2133 	free_irq(atxdmac->irq, atxdmac);
2134 
2135 	for (i = 0; i < atxdmac->dma.chancnt; i++) {
2136 		struct at_xdmac_chan *atchan = &atxdmac->chan[i];
2137 
2138 		tasklet_kill(&atchan->tasklet);
2139 		at_xdmac_free_chan_resources(&atchan->chan);
2140 	}
2141 
2142 	return 0;
2143 }
2144 
2145 static const struct dev_pm_ops atmel_xdmac_dev_pm_ops = {
2146 	.prepare	= atmel_xdmac_prepare,
2147 	SET_LATE_SYSTEM_SLEEP_PM_OPS(atmel_xdmac_suspend, atmel_xdmac_resume)
2148 };
2149 
2150 static const struct of_device_id atmel_xdmac_dt_ids[] = {
2151 	{
2152 		.compatible = "atmel,sama5d4-dma",
2153 	}, {
2154 		/* sentinel */
2155 	}
2156 };
2157 MODULE_DEVICE_TABLE(of, atmel_xdmac_dt_ids);
2158 
2159 static struct platform_driver at_xdmac_driver = {
2160 	.probe		= at_xdmac_probe,
2161 	.remove		= at_xdmac_remove,
2162 	.driver = {
2163 		.name		= "at_xdmac",
2164 		.of_match_table	= of_match_ptr(atmel_xdmac_dt_ids),
2165 		.pm		= &atmel_xdmac_dev_pm_ops,
2166 	}
2167 };
2168 
2169 static int __init at_xdmac_init(void)
2170 {
2171 	return platform_driver_probe(&at_xdmac_driver, at_xdmac_probe);
2172 }
2173 subsys_initcall(at_xdmac_init);
2174 
2175 MODULE_DESCRIPTION("Atmel Extended DMA Controller driver");
2176 MODULE_AUTHOR("Ludovic Desroches <ludovic.desroches@atmel.com>");
2177 MODULE_LICENSE("GPL");
2178