xref: /linux/drivers/dma/xgene-dma.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * Applied Micro X-Gene SoC DMA engine Driver
3  *
4  * Copyright (c) 2015, Applied Micro Circuits Corporation
5  * Authors: Rameshwar Prasad Sahu <rsahu@apm.com>
6  *	    Loc Ho <lho@apm.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 as published by the
10  * Free Software Foundation;  either version 2 of the  License, or (at your
11  * option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
20  *
21  * NOTE: PM support is currently not available.
22  */
23 
24 #include <linux/acpi.h>
25 #include <linux/clk.h>
26 #include <linux/delay.h>
27 #include <linux/dma-mapping.h>
28 #include <linux/dmaengine.h>
29 #include <linux/dmapool.h>
30 #include <linux/interrupt.h>
31 #include <linux/io.h>
32 #include <linux/module.h>
33 #include <linux/of_device.h>
34 
35 #include "dmaengine.h"
36 
37 /* X-Gene DMA ring csr registers and bit definations */
38 #define XGENE_DMA_RING_CONFIG			0x04
39 #define XGENE_DMA_RING_ENABLE			BIT(31)
40 #define XGENE_DMA_RING_ID			0x08
41 #define XGENE_DMA_RING_ID_SETUP(v)		((v) | BIT(31))
42 #define XGENE_DMA_RING_ID_BUF			0x0C
43 #define XGENE_DMA_RING_ID_BUF_SETUP(v)		(((v) << 9) | BIT(21))
44 #define XGENE_DMA_RING_THRESLD0_SET1		0x30
45 #define XGENE_DMA_RING_THRESLD0_SET1_VAL	0X64
46 #define XGENE_DMA_RING_THRESLD1_SET1		0x34
47 #define XGENE_DMA_RING_THRESLD1_SET1_VAL	0xC8
48 #define XGENE_DMA_RING_HYSTERESIS		0x68
49 #define XGENE_DMA_RING_HYSTERESIS_VAL		0xFFFFFFFF
50 #define XGENE_DMA_RING_STATE			0x6C
51 #define XGENE_DMA_RING_STATE_WR_BASE		0x70
52 #define XGENE_DMA_RING_NE_INT_MODE		0x017C
53 #define XGENE_DMA_RING_NE_INT_MODE_SET(m, v)	\
54 	((m) = ((m) & ~BIT(31 - (v))) | BIT(31 - (v)))
55 #define XGENE_DMA_RING_NE_INT_MODE_RESET(m, v)	\
56 	((m) &= (~BIT(31 - (v))))
57 #define XGENE_DMA_RING_CLKEN			0xC208
58 #define XGENE_DMA_RING_SRST			0xC200
59 #define XGENE_DMA_RING_MEM_RAM_SHUTDOWN		0xD070
60 #define XGENE_DMA_RING_BLK_MEM_RDY		0xD074
61 #define XGENE_DMA_RING_BLK_MEM_RDY_VAL		0xFFFFFFFF
62 #define XGENE_DMA_RING_ID_GET(owner, num)	(((owner) << 6) | (num))
63 #define XGENE_DMA_RING_DST_ID(v)		((1 << 10) | (v))
64 #define XGENE_DMA_RING_CMD_OFFSET		0x2C
65 #define XGENE_DMA_RING_CMD_BASE_OFFSET(v)	((v) << 6)
66 #define XGENE_DMA_RING_COHERENT_SET(m)		\
67 	(((u32 *)(m))[2] |= BIT(4))
68 #define XGENE_DMA_RING_ADDRL_SET(m, v)		\
69 	(((u32 *)(m))[2] |= (((v) >> 8) << 5))
70 #define XGENE_DMA_RING_ADDRH_SET(m, v)		\
71 	(((u32 *)(m))[3] |= ((v) >> 35))
72 #define XGENE_DMA_RING_ACCEPTLERR_SET(m)	\
73 	(((u32 *)(m))[3] |= BIT(19))
74 #define XGENE_DMA_RING_SIZE_SET(m, v)		\
75 	(((u32 *)(m))[3] |= ((v) << 23))
76 #define XGENE_DMA_RING_RECOMBBUF_SET(m)		\
77 	(((u32 *)(m))[3] |= BIT(27))
78 #define XGENE_DMA_RING_RECOMTIMEOUTL_SET(m)	\
79 	(((u32 *)(m))[3] |= (0x7 << 28))
80 #define XGENE_DMA_RING_RECOMTIMEOUTH_SET(m)	\
81 	(((u32 *)(m))[4] |= 0x3)
82 #define XGENE_DMA_RING_SELTHRSH_SET(m)		\
83 	(((u32 *)(m))[4] |= BIT(3))
84 #define XGENE_DMA_RING_TYPE_SET(m, v)		\
85 	(((u32 *)(m))[4] |= ((v) << 19))
86 
87 /* X-Gene DMA device csr registers and bit definitions */
88 #define XGENE_DMA_IPBRR				0x0
89 #define XGENE_DMA_DEV_ID_RD(v)			((v) & 0x00000FFF)
90 #define XGENE_DMA_BUS_ID_RD(v)			(((v) >> 12) & 3)
91 #define XGENE_DMA_REV_NO_RD(v)			(((v) >> 14) & 3)
92 #define XGENE_DMA_GCR				0x10
93 #define XGENE_DMA_CH_SETUP(v)			\
94 	((v) = ((v) & ~0x000FFFFF) | 0x000AAFFF)
95 #define XGENE_DMA_ENABLE(v)			((v) |= BIT(31))
96 #define XGENE_DMA_DISABLE(v)			((v) &= ~BIT(31))
97 #define XGENE_DMA_RAID6_CONT			0x14
98 #define XGENE_DMA_RAID6_MULTI_CTRL(v)		((v) << 24)
99 #define XGENE_DMA_INT				0x70
100 #define XGENE_DMA_INT_MASK			0x74
101 #define XGENE_DMA_INT_ALL_MASK			0xFFFFFFFF
102 #define XGENE_DMA_INT_ALL_UNMASK		0x0
103 #define XGENE_DMA_INT_MASK_SHIFT		0x14
104 #define XGENE_DMA_RING_INT0_MASK		0x90A0
105 #define XGENE_DMA_RING_INT1_MASK		0x90A8
106 #define XGENE_DMA_RING_INT2_MASK		0x90B0
107 #define XGENE_DMA_RING_INT3_MASK		0x90B8
108 #define XGENE_DMA_RING_INT4_MASK		0x90C0
109 #define XGENE_DMA_CFG_RING_WQ_ASSOC		0x90E0
110 #define XGENE_DMA_ASSOC_RING_MNGR1		0xFFFFFFFF
111 #define XGENE_DMA_MEM_RAM_SHUTDOWN		0xD070
112 #define XGENE_DMA_BLK_MEM_RDY			0xD074
113 #define XGENE_DMA_BLK_MEM_RDY_VAL		0xFFFFFFFF
114 #define XGENE_DMA_RING_CMD_SM_OFFSET		0x8000
115 
116 /* X-Gene SoC EFUSE csr register and bit defination */
117 #define XGENE_SOC_JTAG1_SHADOW			0x18
118 #define XGENE_DMA_PQ_DISABLE_MASK		BIT(13)
119 
120 /* X-Gene DMA Descriptor format */
121 #define XGENE_DMA_DESC_NV_BIT			BIT_ULL(50)
122 #define XGENE_DMA_DESC_IN_BIT			BIT_ULL(55)
123 #define XGENE_DMA_DESC_C_BIT			BIT_ULL(63)
124 #define XGENE_DMA_DESC_DR_BIT			BIT_ULL(61)
125 #define XGENE_DMA_DESC_ELERR_POS		46
126 #define XGENE_DMA_DESC_RTYPE_POS		56
127 #define XGENE_DMA_DESC_LERR_POS			60
128 #define XGENE_DMA_DESC_BUFLEN_POS		48
129 #define XGENE_DMA_DESC_HOENQ_NUM_POS		48
130 #define XGENE_DMA_DESC_ELERR_RD(m)		\
131 	(((m) >> XGENE_DMA_DESC_ELERR_POS) & 0x3)
132 #define XGENE_DMA_DESC_LERR_RD(m)		\
133 	(((m) >> XGENE_DMA_DESC_LERR_POS) & 0x7)
134 #define XGENE_DMA_DESC_STATUS(elerr, lerr)	\
135 	(((elerr) << 4) | (lerr))
136 
137 /* X-Gene DMA descriptor empty s/w signature */
138 #define XGENE_DMA_DESC_EMPTY_SIGNATURE		~0ULL
139 
140 /* X-Gene DMA configurable parameters defines */
141 #define XGENE_DMA_RING_NUM		512
142 #define XGENE_DMA_BUFNUM		0x0
143 #define XGENE_DMA_CPU_BUFNUM		0x18
144 #define XGENE_DMA_RING_OWNER_DMA	0x03
145 #define XGENE_DMA_RING_OWNER_CPU	0x0F
146 #define XGENE_DMA_RING_TYPE_REGULAR	0x01
147 #define XGENE_DMA_RING_WQ_DESC_SIZE	32	/* 32 Bytes */
148 #define XGENE_DMA_RING_NUM_CONFIG	5
149 #define XGENE_DMA_MAX_CHANNEL		4
150 #define XGENE_DMA_XOR_CHANNEL		0
151 #define XGENE_DMA_PQ_CHANNEL		1
152 #define XGENE_DMA_MAX_BYTE_CNT		0x4000	/* 16 KB */
153 #define XGENE_DMA_MAX_64B_DESC_BYTE_CNT	0x14000	/* 80 KB */
154 #define XGENE_DMA_MAX_XOR_SRC		5
155 #define XGENE_DMA_16K_BUFFER_LEN_CODE	0x0
156 #define XGENE_DMA_INVALID_LEN_CODE	0x7800000000000000ULL
157 
158 /* X-Gene DMA descriptor error codes */
159 #define ERR_DESC_AXI			0x01
160 #define ERR_BAD_DESC			0x02
161 #define ERR_READ_DATA_AXI		0x03
162 #define ERR_WRITE_DATA_AXI		0x04
163 #define ERR_FBP_TIMEOUT			0x05
164 #define ERR_ECC				0x06
165 #define ERR_DIFF_SIZE			0x08
166 #define ERR_SCT_GAT_LEN			0x09
167 #define ERR_CRC_ERR			0x11
168 #define ERR_CHKSUM			0x12
169 #define ERR_DIF				0x13
170 
171 /* X-Gene DMA error interrupt codes */
172 #define ERR_DIF_SIZE_INT		0x0
173 #define ERR_GS_ERR_INT			0x1
174 #define ERR_FPB_TIMEO_INT		0x2
175 #define ERR_WFIFO_OVF_INT		0x3
176 #define ERR_RFIFO_OVF_INT		0x4
177 #define ERR_WR_TIMEO_INT		0x5
178 #define ERR_RD_TIMEO_INT		0x6
179 #define ERR_WR_ERR_INT			0x7
180 #define ERR_RD_ERR_INT			0x8
181 #define ERR_BAD_DESC_INT		0x9
182 #define ERR_DESC_DST_INT		0xA
183 #define ERR_DESC_SRC_INT		0xB
184 
185 /* X-Gene DMA flyby operation code */
186 #define FLYBY_2SRC_XOR			0x80
187 #define FLYBY_3SRC_XOR			0x90
188 #define FLYBY_4SRC_XOR			0xA0
189 #define FLYBY_5SRC_XOR			0xB0
190 
191 /* X-Gene DMA SW descriptor flags */
192 #define XGENE_DMA_FLAG_64B_DESC		BIT(0)
193 
194 /* Define to dump X-Gene DMA descriptor */
195 #define XGENE_DMA_DESC_DUMP(desc, m)	\
196 	print_hex_dump(KERN_ERR, (m),	\
197 			DUMP_PREFIX_ADDRESS, 16, 8, (desc), 32, 0)
198 
199 #define to_dma_desc_sw(tx)		\
200 	container_of(tx, struct xgene_dma_desc_sw, tx)
201 #define to_dma_chan(dchan)		\
202 	container_of(dchan, struct xgene_dma_chan, dma_chan)
203 
204 #define chan_dbg(chan, fmt, arg...)	\
205 	dev_dbg(chan->dev, "%s: " fmt, chan->name, ##arg)
206 #define chan_err(chan, fmt, arg...)	\
207 	dev_err(chan->dev, "%s: " fmt, chan->name, ##arg)
208 
209 struct xgene_dma_desc_hw {
210 	__le64 m0;
211 	__le64 m1;
212 	__le64 m2;
213 	__le64 m3;
214 };
215 
216 enum xgene_dma_ring_cfgsize {
217 	XGENE_DMA_RING_CFG_SIZE_512B,
218 	XGENE_DMA_RING_CFG_SIZE_2KB,
219 	XGENE_DMA_RING_CFG_SIZE_16KB,
220 	XGENE_DMA_RING_CFG_SIZE_64KB,
221 	XGENE_DMA_RING_CFG_SIZE_512KB,
222 	XGENE_DMA_RING_CFG_SIZE_INVALID
223 };
224 
225 struct xgene_dma_ring {
226 	struct xgene_dma *pdma;
227 	u8 buf_num;
228 	u16 id;
229 	u16 num;
230 	u16 head;
231 	u16 owner;
232 	u16 slots;
233 	u16 dst_ring_num;
234 	u32 size;
235 	void __iomem *cmd;
236 	void __iomem *cmd_base;
237 	dma_addr_t desc_paddr;
238 	u32 state[XGENE_DMA_RING_NUM_CONFIG];
239 	enum xgene_dma_ring_cfgsize cfgsize;
240 	union {
241 		void *desc_vaddr;
242 		struct xgene_dma_desc_hw *desc_hw;
243 	};
244 };
245 
246 struct xgene_dma_desc_sw {
247 	struct xgene_dma_desc_hw desc1;
248 	struct xgene_dma_desc_hw desc2;
249 	u32 flags;
250 	struct list_head node;
251 	struct list_head tx_list;
252 	struct dma_async_tx_descriptor tx;
253 };
254 
255 /**
256  * struct xgene_dma_chan - internal representation of an X-Gene DMA channel
257  * @dma_chan: dmaengine channel object member
258  * @pdma: X-Gene DMA device structure reference
259  * @dev: struct device reference for dma mapping api
260  * @id: raw id of this channel
261  * @rx_irq: channel IRQ
262  * @name: name of X-Gene DMA channel
263  * @lock: serializes enqueue/dequeue operations to the descriptor pool
264  * @pending: number of transaction request pushed to DMA controller for
265  *	execution, but still waiting for completion,
266  * @max_outstanding: max number of outstanding request we can push to channel
267  * @ld_pending: descriptors which are queued to run, but have not yet been
268  *	submitted to the hardware for execution
269  * @ld_running: descriptors which are currently being executing by the hardware
270  * @ld_completed: descriptors which have finished execution by the hardware.
271  *	These descriptors have already had their cleanup actions run. They
272  *	are waiting for the ACK bit to be set by the async tx API.
273  * @desc_pool: descriptor pool for DMA operations
274  * @tasklet: bottom half where all completed descriptors cleans
275  * @tx_ring: transmit ring descriptor that we use to prepare actual
276  *	descriptors for further executions
277  * @rx_ring: receive ring descriptor that we use to get completed DMA
278  *	descriptors during cleanup time
279  */
280 struct xgene_dma_chan {
281 	struct dma_chan dma_chan;
282 	struct xgene_dma *pdma;
283 	struct device *dev;
284 	int id;
285 	int rx_irq;
286 	char name[10];
287 	spinlock_t lock;
288 	int pending;
289 	int max_outstanding;
290 	struct list_head ld_pending;
291 	struct list_head ld_running;
292 	struct list_head ld_completed;
293 	struct dma_pool *desc_pool;
294 	struct tasklet_struct tasklet;
295 	struct xgene_dma_ring tx_ring;
296 	struct xgene_dma_ring rx_ring;
297 };
298 
299 /**
300  * struct xgene_dma - internal representation of an X-Gene DMA device
301  * @err_irq: DMA error irq number
302  * @ring_num: start id number for DMA ring
303  * @csr_dma: base for DMA register access
304  * @csr_ring: base for DMA ring register access
305  * @csr_ring_cmd: base for DMA ring command register access
306  * @csr_efuse: base for efuse register access
307  * @dma_dev: embedded struct dma_device
308  * @chan: reference to X-Gene DMA channels
309  */
310 struct xgene_dma {
311 	struct device *dev;
312 	struct clk *clk;
313 	int err_irq;
314 	int ring_num;
315 	void __iomem *csr_dma;
316 	void __iomem *csr_ring;
317 	void __iomem *csr_ring_cmd;
318 	void __iomem *csr_efuse;
319 	struct dma_device dma_dev[XGENE_DMA_MAX_CHANNEL];
320 	struct xgene_dma_chan chan[XGENE_DMA_MAX_CHANNEL];
321 };
322 
323 static const char * const xgene_dma_desc_err[] = {
324 	[ERR_DESC_AXI] = "AXI error when reading src/dst link list",
325 	[ERR_BAD_DESC] = "ERR or El_ERR fields not set to zero in desc",
326 	[ERR_READ_DATA_AXI] = "AXI error when reading data",
327 	[ERR_WRITE_DATA_AXI] = "AXI error when writing data",
328 	[ERR_FBP_TIMEOUT] = "Timeout on bufpool fetch",
329 	[ERR_ECC] = "ECC double bit error",
330 	[ERR_DIFF_SIZE] = "Bufpool too small to hold all the DIF result",
331 	[ERR_SCT_GAT_LEN] = "Gather and scatter data length not same",
332 	[ERR_CRC_ERR] = "CRC error",
333 	[ERR_CHKSUM] = "Checksum error",
334 	[ERR_DIF] = "DIF error",
335 };
336 
337 static const char * const xgene_dma_err[] = {
338 	[ERR_DIF_SIZE_INT] = "DIF size error",
339 	[ERR_GS_ERR_INT] = "Gather scatter not same size error",
340 	[ERR_FPB_TIMEO_INT] = "Free pool time out error",
341 	[ERR_WFIFO_OVF_INT] = "Write FIFO over flow error",
342 	[ERR_RFIFO_OVF_INT] = "Read FIFO over flow error",
343 	[ERR_WR_TIMEO_INT] = "Write time out error",
344 	[ERR_RD_TIMEO_INT] = "Read time out error",
345 	[ERR_WR_ERR_INT] = "HBF bus write error",
346 	[ERR_RD_ERR_INT] = "HBF bus read error",
347 	[ERR_BAD_DESC_INT] = "Ring descriptor HE0 not set error",
348 	[ERR_DESC_DST_INT] = "HFB reading dst link address error",
349 	[ERR_DESC_SRC_INT] = "HFB reading src link address error",
350 };
351 
352 static bool is_pq_enabled(struct xgene_dma *pdma)
353 {
354 	u32 val;
355 
356 	val = ioread32(pdma->csr_efuse + XGENE_SOC_JTAG1_SHADOW);
357 	return !(val & XGENE_DMA_PQ_DISABLE_MASK);
358 }
359 
360 static u64 xgene_dma_encode_len(size_t len)
361 {
362 	return (len < XGENE_DMA_MAX_BYTE_CNT) ?
363 		((u64)len << XGENE_DMA_DESC_BUFLEN_POS) :
364 		XGENE_DMA_16K_BUFFER_LEN_CODE;
365 }
366 
367 static u8 xgene_dma_encode_xor_flyby(u32 src_cnt)
368 {
369 	static u8 flyby_type[] = {
370 		FLYBY_2SRC_XOR, /* Dummy */
371 		FLYBY_2SRC_XOR, /* Dummy */
372 		FLYBY_2SRC_XOR,
373 		FLYBY_3SRC_XOR,
374 		FLYBY_4SRC_XOR,
375 		FLYBY_5SRC_XOR
376 	};
377 
378 	return flyby_type[src_cnt];
379 }
380 
381 static void xgene_dma_set_src_buffer(__le64 *ext8, size_t *len,
382 				     dma_addr_t *paddr)
383 {
384 	size_t nbytes = (*len < XGENE_DMA_MAX_BYTE_CNT) ?
385 			*len : XGENE_DMA_MAX_BYTE_CNT;
386 
387 	*ext8 |= cpu_to_le64(*paddr);
388 	*ext8 |= cpu_to_le64(xgene_dma_encode_len(nbytes));
389 	*len -= nbytes;
390 	*paddr += nbytes;
391 }
392 
393 static void xgene_dma_invalidate_buffer(__le64 *ext8)
394 {
395 	*ext8 |= cpu_to_le64(XGENE_DMA_INVALID_LEN_CODE);
396 }
397 
398 static __le64 *xgene_dma_lookup_ext8(struct xgene_dma_desc_hw *desc, int idx)
399 {
400 	switch (idx) {
401 	case 0:
402 		return &desc->m1;
403 	case 1:
404 		return &desc->m0;
405 	case 2:
406 		return &desc->m3;
407 	case 3:
408 		return &desc->m2;
409 	default:
410 		pr_err("Invalid dma descriptor index\n");
411 	}
412 
413 	return NULL;
414 }
415 
416 static void xgene_dma_init_desc(struct xgene_dma_desc_hw *desc,
417 				u16 dst_ring_num)
418 {
419 	desc->m0 |= cpu_to_le64(XGENE_DMA_DESC_IN_BIT);
420 	desc->m0 |= cpu_to_le64((u64)XGENE_DMA_RING_OWNER_DMA <<
421 				XGENE_DMA_DESC_RTYPE_POS);
422 	desc->m1 |= cpu_to_le64(XGENE_DMA_DESC_C_BIT);
423 	desc->m3 |= cpu_to_le64((u64)dst_ring_num <<
424 				XGENE_DMA_DESC_HOENQ_NUM_POS);
425 }
426 
427 static void xgene_dma_prep_cpy_desc(struct xgene_dma_chan *chan,
428 				    struct xgene_dma_desc_sw *desc_sw,
429 				    dma_addr_t dst, dma_addr_t src,
430 				    size_t len)
431 {
432 	struct xgene_dma_desc_hw *desc1, *desc2;
433 	int i;
434 
435 	/* Get 1st descriptor */
436 	desc1 = &desc_sw->desc1;
437 	xgene_dma_init_desc(desc1, chan->tx_ring.dst_ring_num);
438 
439 	/* Set destination address */
440 	desc1->m2 |= cpu_to_le64(XGENE_DMA_DESC_DR_BIT);
441 	desc1->m3 |= cpu_to_le64(dst);
442 
443 	/* Set 1st source address */
444 	xgene_dma_set_src_buffer(&desc1->m1, &len, &src);
445 
446 	if (!len)
447 		return;
448 
449 	/*
450 	 * We need to split this source buffer,
451 	 * and need to use 2nd descriptor
452 	 */
453 	desc2 = &desc_sw->desc2;
454 	desc1->m0 |= cpu_to_le64(XGENE_DMA_DESC_NV_BIT);
455 
456 	/* Set 2nd to 5th source address */
457 	for (i = 0; i < 4 && len; i++)
458 		xgene_dma_set_src_buffer(xgene_dma_lookup_ext8(desc2, i),
459 					 &len, &src);
460 
461 	/* Invalidate unused source address field */
462 	for (; i < 4; i++)
463 		xgene_dma_invalidate_buffer(xgene_dma_lookup_ext8(desc2, i));
464 
465 	/* Updated flag that we have prepared 64B descriptor */
466 	desc_sw->flags |= XGENE_DMA_FLAG_64B_DESC;
467 }
468 
469 static void xgene_dma_prep_xor_desc(struct xgene_dma_chan *chan,
470 				    struct xgene_dma_desc_sw *desc_sw,
471 				    dma_addr_t *dst, dma_addr_t *src,
472 				    u32 src_cnt, size_t *nbytes,
473 				    const u8 *scf)
474 {
475 	struct xgene_dma_desc_hw *desc1, *desc2;
476 	size_t len = *nbytes;
477 	int i;
478 
479 	desc1 = &desc_sw->desc1;
480 	desc2 = &desc_sw->desc2;
481 
482 	/* Initialize DMA descriptor */
483 	xgene_dma_init_desc(desc1, chan->tx_ring.dst_ring_num);
484 
485 	/* Set destination address */
486 	desc1->m2 |= cpu_to_le64(XGENE_DMA_DESC_DR_BIT);
487 	desc1->m3 |= cpu_to_le64(*dst);
488 
489 	/* We have multiple source addresses, so need to set NV bit*/
490 	desc1->m0 |= cpu_to_le64(XGENE_DMA_DESC_NV_BIT);
491 
492 	/* Set flyby opcode */
493 	desc1->m2 |= cpu_to_le64(xgene_dma_encode_xor_flyby(src_cnt));
494 
495 	/* Set 1st to 5th source addresses */
496 	for (i = 0; i < src_cnt; i++) {
497 		len = *nbytes;
498 		xgene_dma_set_src_buffer((i == 0) ? &desc1->m1 :
499 					 xgene_dma_lookup_ext8(desc2, i - 1),
500 					 &len, &src[i]);
501 		desc1->m2 |= cpu_to_le64((scf[i] << ((i + 1) * 8)));
502 	}
503 
504 	/* Update meta data */
505 	*nbytes = len;
506 	*dst += XGENE_DMA_MAX_BYTE_CNT;
507 
508 	/* We need always 64B descriptor to perform xor or pq operations */
509 	desc_sw->flags |= XGENE_DMA_FLAG_64B_DESC;
510 }
511 
512 static dma_cookie_t xgene_dma_tx_submit(struct dma_async_tx_descriptor *tx)
513 {
514 	struct xgene_dma_desc_sw *desc;
515 	struct xgene_dma_chan *chan;
516 	dma_cookie_t cookie;
517 
518 	if (unlikely(!tx))
519 		return -EINVAL;
520 
521 	chan = to_dma_chan(tx->chan);
522 	desc = to_dma_desc_sw(tx);
523 
524 	spin_lock_bh(&chan->lock);
525 
526 	cookie = dma_cookie_assign(tx);
527 
528 	/* Add this transaction list onto the tail of the pending queue */
529 	list_splice_tail_init(&desc->tx_list, &chan->ld_pending);
530 
531 	spin_unlock_bh(&chan->lock);
532 
533 	return cookie;
534 }
535 
536 static void xgene_dma_clean_descriptor(struct xgene_dma_chan *chan,
537 				       struct xgene_dma_desc_sw *desc)
538 {
539 	list_del(&desc->node);
540 	chan_dbg(chan, "LD %p free\n", desc);
541 	dma_pool_free(chan->desc_pool, desc, desc->tx.phys);
542 }
543 
544 static struct xgene_dma_desc_sw *xgene_dma_alloc_descriptor(
545 				 struct xgene_dma_chan *chan)
546 {
547 	struct xgene_dma_desc_sw *desc;
548 	dma_addr_t phys;
549 
550 	desc = dma_pool_alloc(chan->desc_pool, GFP_NOWAIT, &phys);
551 	if (!desc) {
552 		chan_err(chan, "Failed to allocate LDs\n");
553 		return NULL;
554 	}
555 
556 	memset(desc, 0, sizeof(*desc));
557 
558 	INIT_LIST_HEAD(&desc->tx_list);
559 	desc->tx.phys = phys;
560 	desc->tx.tx_submit = xgene_dma_tx_submit;
561 	dma_async_tx_descriptor_init(&desc->tx, &chan->dma_chan);
562 
563 	chan_dbg(chan, "LD %p allocated\n", desc);
564 
565 	return desc;
566 }
567 
568 /**
569  * xgene_dma_clean_completed_descriptor - free all descriptors which
570  * has been completed and acked
571  * @chan: X-Gene DMA channel
572  *
573  * This function is used on all completed and acked descriptors.
574  */
575 static void xgene_dma_clean_completed_descriptor(struct xgene_dma_chan *chan)
576 {
577 	struct xgene_dma_desc_sw *desc, *_desc;
578 
579 	/* Run the callback for each descriptor, in order */
580 	list_for_each_entry_safe(desc, _desc, &chan->ld_completed, node) {
581 		if (async_tx_test_ack(&desc->tx))
582 			xgene_dma_clean_descriptor(chan, desc);
583 	}
584 }
585 
586 /**
587  * xgene_dma_run_tx_complete_actions - cleanup a single link descriptor
588  * @chan: X-Gene DMA channel
589  * @desc: descriptor to cleanup and free
590  *
591  * This function is used on a descriptor which has been executed by the DMA
592  * controller. It will run any callbacks, submit any dependencies.
593  */
594 static void xgene_dma_run_tx_complete_actions(struct xgene_dma_chan *chan,
595 					      struct xgene_dma_desc_sw *desc)
596 {
597 	struct dma_async_tx_descriptor *tx = &desc->tx;
598 
599 	/*
600 	 * If this is not the last transaction in the group,
601 	 * then no need to complete cookie and run any callback as
602 	 * this is not the tx_descriptor which had been sent to caller
603 	 * of this DMA request
604 	 */
605 
606 	if (tx->cookie == 0)
607 		return;
608 
609 	dma_cookie_complete(tx);
610 
611 	/* Run the link descriptor callback function */
612 	if (tx->callback)
613 		tx->callback(tx->callback_param);
614 
615 	dma_descriptor_unmap(tx);
616 
617 	/* Run any dependencies */
618 	dma_run_dependencies(tx);
619 }
620 
621 /**
622  * xgene_dma_clean_running_descriptor - move the completed descriptor from
623  * ld_running to ld_completed
624  * @chan: X-Gene DMA channel
625  * @desc: the descriptor which is completed
626  *
627  * Free the descriptor directly if acked by async_tx api,
628  * else move it to queue ld_completed.
629  */
630 static void xgene_dma_clean_running_descriptor(struct xgene_dma_chan *chan,
631 					       struct xgene_dma_desc_sw *desc)
632 {
633 	/* Remove from the list of running transactions */
634 	list_del(&desc->node);
635 
636 	/*
637 	 * the client is allowed to attach dependent operations
638 	 * until 'ack' is set
639 	 */
640 	if (!async_tx_test_ack(&desc->tx)) {
641 		/*
642 		 * Move this descriptor to the list of descriptors which is
643 		 * completed, but still awaiting the 'ack' bit to be set.
644 		 */
645 		list_add_tail(&desc->node, &chan->ld_completed);
646 		return;
647 	}
648 
649 	chan_dbg(chan, "LD %p free\n", desc);
650 	dma_pool_free(chan->desc_pool, desc, desc->tx.phys);
651 }
652 
653 static void xgene_chan_xfer_request(struct xgene_dma_chan *chan,
654 				    struct xgene_dma_desc_sw *desc_sw)
655 {
656 	struct xgene_dma_ring *ring = &chan->tx_ring;
657 	struct xgene_dma_desc_hw *desc_hw;
658 
659 	/* Get hw descriptor from DMA tx ring */
660 	desc_hw = &ring->desc_hw[ring->head];
661 
662 	/*
663 	 * Increment the head count to point next
664 	 * descriptor for next time
665 	 */
666 	if (++ring->head == ring->slots)
667 		ring->head = 0;
668 
669 	/* Copy prepared sw descriptor data to hw descriptor */
670 	memcpy(desc_hw, &desc_sw->desc1, sizeof(*desc_hw));
671 
672 	/*
673 	 * Check if we have prepared 64B descriptor,
674 	 * in this case we need one more hw descriptor
675 	 */
676 	if (desc_sw->flags & XGENE_DMA_FLAG_64B_DESC) {
677 		desc_hw = &ring->desc_hw[ring->head];
678 
679 		if (++ring->head == ring->slots)
680 			ring->head = 0;
681 
682 		memcpy(desc_hw, &desc_sw->desc2, sizeof(*desc_hw));
683 	}
684 
685 	/* Increment the pending transaction count */
686 	chan->pending += ((desc_sw->flags &
687 			  XGENE_DMA_FLAG_64B_DESC) ? 2 : 1);
688 
689 	/* Notify the hw that we have descriptor ready for execution */
690 	iowrite32((desc_sw->flags & XGENE_DMA_FLAG_64B_DESC) ?
691 		  2 : 1, ring->cmd);
692 }
693 
694 /**
695  * xgene_chan_xfer_ld_pending - push any pending transactions to hw
696  * @chan : X-Gene DMA channel
697  *
698  * LOCKING: must hold chan->lock
699  */
700 static void xgene_chan_xfer_ld_pending(struct xgene_dma_chan *chan)
701 {
702 	struct xgene_dma_desc_sw *desc_sw, *_desc_sw;
703 
704 	/*
705 	 * If the list of pending descriptors is empty, then we
706 	 * don't need to do any work at all
707 	 */
708 	if (list_empty(&chan->ld_pending)) {
709 		chan_dbg(chan, "No pending LDs\n");
710 		return;
711 	}
712 
713 	/*
714 	 * Move elements from the queue of pending transactions onto the list
715 	 * of running transactions and push it to hw for further executions
716 	 */
717 	list_for_each_entry_safe(desc_sw, _desc_sw, &chan->ld_pending, node) {
718 		/*
719 		 * Check if have pushed max number of transactions to hw
720 		 * as capable, so let's stop here and will push remaining
721 		 * elements from pening ld queue after completing some
722 		 * descriptors that we have already pushed
723 		 */
724 		if (chan->pending >= chan->max_outstanding)
725 			return;
726 
727 		xgene_chan_xfer_request(chan, desc_sw);
728 
729 		/*
730 		 * Delete this element from ld pending queue and append it to
731 		 * ld running queue
732 		 */
733 		list_move_tail(&desc_sw->node, &chan->ld_running);
734 	}
735 }
736 
737 /**
738  * xgene_dma_cleanup_descriptors - cleanup link descriptors which are completed
739  * and move them to ld_completed to free until flag 'ack' is set
740  * @chan: X-Gene DMA channel
741  *
742  * This function is used on descriptors which have been executed by the DMA
743  * controller. It will run any callbacks, submit any dependencies, then
744  * free these descriptors if flag 'ack' is set.
745  */
746 static void xgene_dma_cleanup_descriptors(struct xgene_dma_chan *chan)
747 {
748 	struct xgene_dma_ring *ring = &chan->rx_ring;
749 	struct xgene_dma_desc_sw *desc_sw, *_desc_sw;
750 	struct xgene_dma_desc_hw *desc_hw;
751 	struct list_head ld_completed;
752 	u8 status;
753 
754 	INIT_LIST_HEAD(&ld_completed);
755 
756 	spin_lock_bh(&chan->lock);
757 
758 	/* Clean already completed and acked descriptors */
759 	xgene_dma_clean_completed_descriptor(chan);
760 
761 	/* Move all completed descriptors to ld completed queue, in order */
762 	list_for_each_entry_safe(desc_sw, _desc_sw, &chan->ld_running, node) {
763 		/* Get subsequent hw descriptor from DMA rx ring */
764 		desc_hw = &ring->desc_hw[ring->head];
765 
766 		/* Check if this descriptor has been completed */
767 		if (unlikely(le64_to_cpu(desc_hw->m0) ==
768 			     XGENE_DMA_DESC_EMPTY_SIGNATURE))
769 			break;
770 
771 		if (++ring->head == ring->slots)
772 			ring->head = 0;
773 
774 		/* Check if we have any error with DMA transactions */
775 		status = XGENE_DMA_DESC_STATUS(
776 				XGENE_DMA_DESC_ELERR_RD(le64_to_cpu(
777 							desc_hw->m0)),
778 				XGENE_DMA_DESC_LERR_RD(le64_to_cpu(
779 						       desc_hw->m0)));
780 		if (status) {
781 			/* Print the DMA error type */
782 			chan_err(chan, "%s\n", xgene_dma_desc_err[status]);
783 
784 			/*
785 			 * We have DMA transactions error here. Dump DMA Tx
786 			 * and Rx descriptors for this request */
787 			XGENE_DMA_DESC_DUMP(&desc_sw->desc1,
788 					    "X-Gene DMA TX DESC1: ");
789 
790 			if (desc_sw->flags & XGENE_DMA_FLAG_64B_DESC)
791 				XGENE_DMA_DESC_DUMP(&desc_sw->desc2,
792 						    "X-Gene DMA TX DESC2: ");
793 
794 			XGENE_DMA_DESC_DUMP(desc_hw,
795 					    "X-Gene DMA RX ERR DESC: ");
796 		}
797 
798 		/* Notify the hw about this completed descriptor */
799 		iowrite32(-1, ring->cmd);
800 
801 		/* Mark this hw descriptor as processed */
802 		desc_hw->m0 = cpu_to_le64(XGENE_DMA_DESC_EMPTY_SIGNATURE);
803 
804 		/*
805 		 * Decrement the pending transaction count
806 		 * as we have processed one
807 		 */
808 		chan->pending -= ((desc_sw->flags &
809 				  XGENE_DMA_FLAG_64B_DESC) ? 2 : 1);
810 
811 		/*
812 		 * Delete this node from ld running queue and append it to
813 		 * ld completed queue for further processing
814 		 */
815 		list_move_tail(&desc_sw->node, &ld_completed);
816 	}
817 
818 	/*
819 	 * Start any pending transactions automatically
820 	 * In the ideal case, we keep the DMA controller busy while we go
821 	 * ahead and free the descriptors below.
822 	 */
823 	xgene_chan_xfer_ld_pending(chan);
824 
825 	spin_unlock_bh(&chan->lock);
826 
827 	/* Run the callback for each descriptor, in order */
828 	list_for_each_entry_safe(desc_sw, _desc_sw, &ld_completed, node) {
829 		xgene_dma_run_tx_complete_actions(chan, desc_sw);
830 		xgene_dma_clean_running_descriptor(chan, desc_sw);
831 	}
832 }
833 
834 static int xgene_dma_alloc_chan_resources(struct dma_chan *dchan)
835 {
836 	struct xgene_dma_chan *chan = to_dma_chan(dchan);
837 
838 	/* Has this channel already been allocated? */
839 	if (chan->desc_pool)
840 		return 1;
841 
842 	chan->desc_pool = dma_pool_create(chan->name, chan->dev,
843 					  sizeof(struct xgene_dma_desc_sw),
844 					  0, 0);
845 	if (!chan->desc_pool) {
846 		chan_err(chan, "Failed to allocate descriptor pool\n");
847 		return -ENOMEM;
848 	}
849 
850 	chan_dbg(chan, "Allocate descripto pool\n");
851 
852 	return 1;
853 }
854 
855 /**
856  * xgene_dma_free_desc_list - Free all descriptors in a queue
857  * @chan: X-Gene DMA channel
858  * @list: the list to free
859  *
860  * LOCKING: must hold chan->lock
861  */
862 static void xgene_dma_free_desc_list(struct xgene_dma_chan *chan,
863 				     struct list_head *list)
864 {
865 	struct xgene_dma_desc_sw *desc, *_desc;
866 
867 	list_for_each_entry_safe(desc, _desc, list, node)
868 		xgene_dma_clean_descriptor(chan, desc);
869 }
870 
871 static void xgene_dma_free_chan_resources(struct dma_chan *dchan)
872 {
873 	struct xgene_dma_chan *chan = to_dma_chan(dchan);
874 
875 	chan_dbg(chan, "Free all resources\n");
876 
877 	if (!chan->desc_pool)
878 		return;
879 
880 	/* Process all running descriptor */
881 	xgene_dma_cleanup_descriptors(chan);
882 
883 	spin_lock_bh(&chan->lock);
884 
885 	/* Clean all link descriptor queues */
886 	xgene_dma_free_desc_list(chan, &chan->ld_pending);
887 	xgene_dma_free_desc_list(chan, &chan->ld_running);
888 	xgene_dma_free_desc_list(chan, &chan->ld_completed);
889 
890 	spin_unlock_bh(&chan->lock);
891 
892 	/* Delete this channel DMA pool */
893 	dma_pool_destroy(chan->desc_pool);
894 	chan->desc_pool = NULL;
895 }
896 
897 static struct dma_async_tx_descriptor *xgene_dma_prep_memcpy(
898 	struct dma_chan *dchan, dma_addr_t dst, dma_addr_t src,
899 	size_t len, unsigned long flags)
900 {
901 	struct xgene_dma_desc_sw *first = NULL, *new;
902 	struct xgene_dma_chan *chan;
903 	size_t copy;
904 
905 	if (unlikely(!dchan || !len))
906 		return NULL;
907 
908 	chan = to_dma_chan(dchan);
909 
910 	do {
911 		/* Allocate the link descriptor from DMA pool */
912 		new = xgene_dma_alloc_descriptor(chan);
913 		if (!new)
914 			goto fail;
915 
916 		/* Create the largest transaction possible */
917 		copy = min_t(size_t, len, XGENE_DMA_MAX_64B_DESC_BYTE_CNT);
918 
919 		/* Prepare DMA descriptor */
920 		xgene_dma_prep_cpy_desc(chan, new, dst, src, copy);
921 
922 		if (!first)
923 			first = new;
924 
925 		new->tx.cookie = 0;
926 		async_tx_ack(&new->tx);
927 
928 		/* Update metadata */
929 		len -= copy;
930 		dst += copy;
931 		src += copy;
932 
933 		/* Insert the link descriptor to the LD ring */
934 		list_add_tail(&new->node, &first->tx_list);
935 	} while (len);
936 
937 	new->tx.flags = flags; /* client is in control of this ack */
938 	new->tx.cookie = -EBUSY;
939 	list_splice(&first->tx_list, &new->tx_list);
940 
941 	return &new->tx;
942 
943 fail:
944 	if (!first)
945 		return NULL;
946 
947 	xgene_dma_free_desc_list(chan, &first->tx_list);
948 	return NULL;
949 }
950 
951 static struct dma_async_tx_descriptor *xgene_dma_prep_sg(
952 	struct dma_chan *dchan, struct scatterlist *dst_sg,
953 	u32 dst_nents, struct scatterlist *src_sg,
954 	u32 src_nents, unsigned long flags)
955 {
956 	struct xgene_dma_desc_sw *first = NULL, *new = NULL;
957 	struct xgene_dma_chan *chan;
958 	size_t dst_avail, src_avail;
959 	dma_addr_t dst, src;
960 	size_t len;
961 
962 	if (unlikely(!dchan))
963 		return NULL;
964 
965 	if (unlikely(!dst_nents || !src_nents))
966 		return NULL;
967 
968 	if (unlikely(!dst_sg || !src_sg))
969 		return NULL;
970 
971 	chan = to_dma_chan(dchan);
972 
973 	/* Get prepared for the loop */
974 	dst_avail = sg_dma_len(dst_sg);
975 	src_avail = sg_dma_len(src_sg);
976 	dst_nents--;
977 	src_nents--;
978 
979 	/* Run until we are out of scatterlist entries */
980 	while (true) {
981 		/* Create the largest transaction possible */
982 		len = min_t(size_t, src_avail, dst_avail);
983 		len = min_t(size_t, len, XGENE_DMA_MAX_64B_DESC_BYTE_CNT);
984 		if (len == 0)
985 			goto fetch;
986 
987 		dst = sg_dma_address(dst_sg) + sg_dma_len(dst_sg) - dst_avail;
988 		src = sg_dma_address(src_sg) + sg_dma_len(src_sg) - src_avail;
989 
990 		/* Allocate the link descriptor from DMA pool */
991 		new = xgene_dma_alloc_descriptor(chan);
992 		if (!new)
993 			goto fail;
994 
995 		/* Prepare DMA descriptor */
996 		xgene_dma_prep_cpy_desc(chan, new, dst, src, len);
997 
998 		if (!first)
999 			first = new;
1000 
1001 		new->tx.cookie = 0;
1002 		async_tx_ack(&new->tx);
1003 
1004 		/* update metadata */
1005 		dst_avail -= len;
1006 		src_avail -= len;
1007 
1008 		/* Insert the link descriptor to the LD ring */
1009 		list_add_tail(&new->node, &first->tx_list);
1010 
1011 fetch:
1012 		/* fetch the next dst scatterlist entry */
1013 		if (dst_avail == 0) {
1014 			/* no more entries: we're done */
1015 			if (dst_nents == 0)
1016 				break;
1017 
1018 			/* fetch the next entry: if there are no more: done */
1019 			dst_sg = sg_next(dst_sg);
1020 			if (!dst_sg)
1021 				break;
1022 
1023 			dst_nents--;
1024 			dst_avail = sg_dma_len(dst_sg);
1025 		}
1026 
1027 		/* fetch the next src scatterlist entry */
1028 		if (src_avail == 0) {
1029 			/* no more entries: we're done */
1030 			if (src_nents == 0)
1031 				break;
1032 
1033 			/* fetch the next entry: if there are no more: done */
1034 			src_sg = sg_next(src_sg);
1035 			if (!src_sg)
1036 				break;
1037 
1038 			src_nents--;
1039 			src_avail = sg_dma_len(src_sg);
1040 		}
1041 	}
1042 
1043 	if (!new)
1044 		return NULL;
1045 
1046 	new->tx.flags = flags; /* client is in control of this ack */
1047 	new->tx.cookie = -EBUSY;
1048 	list_splice(&first->tx_list, &new->tx_list);
1049 
1050 	return &new->tx;
1051 fail:
1052 	if (!first)
1053 		return NULL;
1054 
1055 	xgene_dma_free_desc_list(chan, &first->tx_list);
1056 	return NULL;
1057 }
1058 
1059 static struct dma_async_tx_descriptor *xgene_dma_prep_xor(
1060 	struct dma_chan *dchan, dma_addr_t dst,	dma_addr_t *src,
1061 	u32 src_cnt, size_t len, unsigned long flags)
1062 {
1063 	struct xgene_dma_desc_sw *first = NULL, *new;
1064 	struct xgene_dma_chan *chan;
1065 	static u8 multi[XGENE_DMA_MAX_XOR_SRC] = {
1066 				0x01, 0x01, 0x01, 0x01, 0x01};
1067 
1068 	if (unlikely(!dchan || !len))
1069 		return NULL;
1070 
1071 	chan = to_dma_chan(dchan);
1072 
1073 	do {
1074 		/* Allocate the link descriptor from DMA pool */
1075 		new = xgene_dma_alloc_descriptor(chan);
1076 		if (!new)
1077 			goto fail;
1078 
1079 		/* Prepare xor DMA descriptor */
1080 		xgene_dma_prep_xor_desc(chan, new, &dst, src,
1081 					src_cnt, &len, multi);
1082 
1083 		if (!first)
1084 			first = new;
1085 
1086 		new->tx.cookie = 0;
1087 		async_tx_ack(&new->tx);
1088 
1089 		/* Insert the link descriptor to the LD ring */
1090 		list_add_tail(&new->node, &first->tx_list);
1091 	} while (len);
1092 
1093 	new->tx.flags = flags; /* client is in control of this ack */
1094 	new->tx.cookie = -EBUSY;
1095 	list_splice(&first->tx_list, &new->tx_list);
1096 
1097 	return &new->tx;
1098 
1099 fail:
1100 	if (!first)
1101 		return NULL;
1102 
1103 	xgene_dma_free_desc_list(chan, &first->tx_list);
1104 	return NULL;
1105 }
1106 
1107 static struct dma_async_tx_descriptor *xgene_dma_prep_pq(
1108 	struct dma_chan *dchan, dma_addr_t *dst, dma_addr_t *src,
1109 	u32 src_cnt, const u8 *scf, size_t len, unsigned long flags)
1110 {
1111 	struct xgene_dma_desc_sw *first = NULL, *new;
1112 	struct xgene_dma_chan *chan;
1113 	size_t _len = len;
1114 	dma_addr_t _src[XGENE_DMA_MAX_XOR_SRC];
1115 	static u8 multi[XGENE_DMA_MAX_XOR_SRC] = {0x01, 0x01, 0x01, 0x01, 0x01};
1116 
1117 	if (unlikely(!dchan || !len))
1118 		return NULL;
1119 
1120 	chan = to_dma_chan(dchan);
1121 
1122 	/*
1123 	 * Save source addresses on local variable, may be we have to
1124 	 * prepare two descriptor to generate P and Q if both enabled
1125 	 * in the flags by client
1126 	 */
1127 	memcpy(_src, src, sizeof(*src) * src_cnt);
1128 
1129 	if (flags & DMA_PREP_PQ_DISABLE_P)
1130 		len = 0;
1131 
1132 	if (flags & DMA_PREP_PQ_DISABLE_Q)
1133 		_len = 0;
1134 
1135 	do {
1136 		/* Allocate the link descriptor from DMA pool */
1137 		new = xgene_dma_alloc_descriptor(chan);
1138 		if (!new)
1139 			goto fail;
1140 
1141 		if (!first)
1142 			first = new;
1143 
1144 		new->tx.cookie = 0;
1145 		async_tx_ack(&new->tx);
1146 
1147 		/* Insert the link descriptor to the LD ring */
1148 		list_add_tail(&new->node, &first->tx_list);
1149 
1150 		/*
1151 		 * Prepare DMA descriptor to generate P,
1152 		 * if DMA_PREP_PQ_DISABLE_P flag is not set
1153 		 */
1154 		if (len) {
1155 			xgene_dma_prep_xor_desc(chan, new, &dst[0], src,
1156 						src_cnt, &len, multi);
1157 			continue;
1158 		}
1159 
1160 		/*
1161 		 * Prepare DMA descriptor to generate Q,
1162 		 * if DMA_PREP_PQ_DISABLE_Q flag is not set
1163 		 */
1164 		if (_len) {
1165 			xgene_dma_prep_xor_desc(chan, new, &dst[1], _src,
1166 						src_cnt, &_len, scf);
1167 		}
1168 	} while (len || _len);
1169 
1170 	new->tx.flags = flags; /* client is in control of this ack */
1171 	new->tx.cookie = -EBUSY;
1172 	list_splice(&first->tx_list, &new->tx_list);
1173 
1174 	return &new->tx;
1175 
1176 fail:
1177 	if (!first)
1178 		return NULL;
1179 
1180 	xgene_dma_free_desc_list(chan, &first->tx_list);
1181 	return NULL;
1182 }
1183 
1184 static void xgene_dma_issue_pending(struct dma_chan *dchan)
1185 {
1186 	struct xgene_dma_chan *chan = to_dma_chan(dchan);
1187 
1188 	spin_lock_bh(&chan->lock);
1189 	xgene_chan_xfer_ld_pending(chan);
1190 	spin_unlock_bh(&chan->lock);
1191 }
1192 
1193 static enum dma_status xgene_dma_tx_status(struct dma_chan *dchan,
1194 					   dma_cookie_t cookie,
1195 					   struct dma_tx_state *txstate)
1196 {
1197 	return dma_cookie_status(dchan, cookie, txstate);
1198 }
1199 
1200 static void xgene_dma_tasklet_cb(unsigned long data)
1201 {
1202 	struct xgene_dma_chan *chan = (struct xgene_dma_chan *)data;
1203 
1204 	/* Run all cleanup for descriptors which have been completed */
1205 	xgene_dma_cleanup_descriptors(chan);
1206 
1207 	/* Re-enable DMA channel IRQ */
1208 	enable_irq(chan->rx_irq);
1209 }
1210 
1211 static irqreturn_t xgene_dma_chan_ring_isr(int irq, void *id)
1212 {
1213 	struct xgene_dma_chan *chan = (struct xgene_dma_chan *)id;
1214 
1215 	BUG_ON(!chan);
1216 
1217 	/*
1218 	 * Disable DMA channel IRQ until we process completed
1219 	 * descriptors
1220 	 */
1221 	disable_irq_nosync(chan->rx_irq);
1222 
1223 	/*
1224 	 * Schedule the tasklet to handle all cleanup of the current
1225 	 * transaction. It will start a new transaction if there is
1226 	 * one pending.
1227 	 */
1228 	tasklet_schedule(&chan->tasklet);
1229 
1230 	return IRQ_HANDLED;
1231 }
1232 
1233 static irqreturn_t xgene_dma_err_isr(int irq, void *id)
1234 {
1235 	struct xgene_dma *pdma = (struct xgene_dma *)id;
1236 	unsigned long int_mask;
1237 	u32 val, i;
1238 
1239 	val = ioread32(pdma->csr_dma + XGENE_DMA_INT);
1240 
1241 	/* Clear DMA interrupts */
1242 	iowrite32(val, pdma->csr_dma + XGENE_DMA_INT);
1243 
1244 	/* Print DMA error info */
1245 	int_mask = val >> XGENE_DMA_INT_MASK_SHIFT;
1246 	for_each_set_bit(i, &int_mask, ARRAY_SIZE(xgene_dma_err))
1247 		dev_err(pdma->dev,
1248 			"Interrupt status 0x%08X %s\n", val, xgene_dma_err[i]);
1249 
1250 	return IRQ_HANDLED;
1251 }
1252 
1253 static void xgene_dma_wr_ring_state(struct xgene_dma_ring *ring)
1254 {
1255 	int i;
1256 
1257 	iowrite32(ring->num, ring->pdma->csr_ring + XGENE_DMA_RING_STATE);
1258 
1259 	for (i = 0; i < XGENE_DMA_RING_NUM_CONFIG; i++)
1260 		iowrite32(ring->state[i], ring->pdma->csr_ring +
1261 			  XGENE_DMA_RING_STATE_WR_BASE + (i * 4));
1262 }
1263 
1264 static void xgene_dma_clr_ring_state(struct xgene_dma_ring *ring)
1265 {
1266 	memset(ring->state, 0, sizeof(u32) * XGENE_DMA_RING_NUM_CONFIG);
1267 	xgene_dma_wr_ring_state(ring);
1268 }
1269 
1270 static void xgene_dma_setup_ring(struct xgene_dma_ring *ring)
1271 {
1272 	void *ring_cfg = ring->state;
1273 	u64 addr = ring->desc_paddr;
1274 	u32 i, val;
1275 
1276 	ring->slots = ring->size / XGENE_DMA_RING_WQ_DESC_SIZE;
1277 
1278 	/* Clear DMA ring state */
1279 	xgene_dma_clr_ring_state(ring);
1280 
1281 	/* Set DMA ring type */
1282 	XGENE_DMA_RING_TYPE_SET(ring_cfg, XGENE_DMA_RING_TYPE_REGULAR);
1283 
1284 	if (ring->owner == XGENE_DMA_RING_OWNER_DMA) {
1285 		/* Set recombination buffer and timeout */
1286 		XGENE_DMA_RING_RECOMBBUF_SET(ring_cfg);
1287 		XGENE_DMA_RING_RECOMTIMEOUTL_SET(ring_cfg);
1288 		XGENE_DMA_RING_RECOMTIMEOUTH_SET(ring_cfg);
1289 	}
1290 
1291 	/* Initialize DMA ring state */
1292 	XGENE_DMA_RING_SELTHRSH_SET(ring_cfg);
1293 	XGENE_DMA_RING_ACCEPTLERR_SET(ring_cfg);
1294 	XGENE_DMA_RING_COHERENT_SET(ring_cfg);
1295 	XGENE_DMA_RING_ADDRL_SET(ring_cfg, addr);
1296 	XGENE_DMA_RING_ADDRH_SET(ring_cfg, addr);
1297 	XGENE_DMA_RING_SIZE_SET(ring_cfg, ring->cfgsize);
1298 
1299 	/* Write DMA ring configurations */
1300 	xgene_dma_wr_ring_state(ring);
1301 
1302 	/* Set DMA ring id */
1303 	iowrite32(XGENE_DMA_RING_ID_SETUP(ring->id),
1304 		  ring->pdma->csr_ring + XGENE_DMA_RING_ID);
1305 
1306 	/* Set DMA ring buffer */
1307 	iowrite32(XGENE_DMA_RING_ID_BUF_SETUP(ring->num),
1308 		  ring->pdma->csr_ring + XGENE_DMA_RING_ID_BUF);
1309 
1310 	if (ring->owner != XGENE_DMA_RING_OWNER_CPU)
1311 		return;
1312 
1313 	/* Set empty signature to DMA Rx ring descriptors */
1314 	for (i = 0; i < ring->slots; i++) {
1315 		struct xgene_dma_desc_hw *desc;
1316 
1317 		desc = &ring->desc_hw[i];
1318 		desc->m0 = cpu_to_le64(XGENE_DMA_DESC_EMPTY_SIGNATURE);
1319 	}
1320 
1321 	/* Enable DMA Rx ring interrupt */
1322 	val = ioread32(ring->pdma->csr_ring + XGENE_DMA_RING_NE_INT_MODE);
1323 	XGENE_DMA_RING_NE_INT_MODE_SET(val, ring->buf_num);
1324 	iowrite32(val, ring->pdma->csr_ring + XGENE_DMA_RING_NE_INT_MODE);
1325 }
1326 
1327 static void xgene_dma_clear_ring(struct xgene_dma_ring *ring)
1328 {
1329 	u32 ring_id, val;
1330 
1331 	if (ring->owner == XGENE_DMA_RING_OWNER_CPU) {
1332 		/* Disable DMA Rx ring interrupt */
1333 		val = ioread32(ring->pdma->csr_ring +
1334 			       XGENE_DMA_RING_NE_INT_MODE);
1335 		XGENE_DMA_RING_NE_INT_MODE_RESET(val, ring->buf_num);
1336 		iowrite32(val, ring->pdma->csr_ring +
1337 			  XGENE_DMA_RING_NE_INT_MODE);
1338 	}
1339 
1340 	/* Clear DMA ring state */
1341 	ring_id = XGENE_DMA_RING_ID_SETUP(ring->id);
1342 	iowrite32(ring_id, ring->pdma->csr_ring + XGENE_DMA_RING_ID);
1343 
1344 	iowrite32(0, ring->pdma->csr_ring + XGENE_DMA_RING_ID_BUF);
1345 	xgene_dma_clr_ring_state(ring);
1346 }
1347 
1348 static void xgene_dma_set_ring_cmd(struct xgene_dma_ring *ring)
1349 {
1350 	ring->cmd_base = ring->pdma->csr_ring_cmd +
1351 				XGENE_DMA_RING_CMD_BASE_OFFSET((ring->num -
1352 							  XGENE_DMA_RING_NUM));
1353 
1354 	ring->cmd = ring->cmd_base + XGENE_DMA_RING_CMD_OFFSET;
1355 }
1356 
1357 static int xgene_dma_get_ring_size(struct xgene_dma_chan *chan,
1358 				   enum xgene_dma_ring_cfgsize cfgsize)
1359 {
1360 	int size;
1361 
1362 	switch (cfgsize) {
1363 	case XGENE_DMA_RING_CFG_SIZE_512B:
1364 		size = 0x200;
1365 		break;
1366 	case XGENE_DMA_RING_CFG_SIZE_2KB:
1367 		size = 0x800;
1368 		break;
1369 	case XGENE_DMA_RING_CFG_SIZE_16KB:
1370 		size = 0x4000;
1371 		break;
1372 	case XGENE_DMA_RING_CFG_SIZE_64KB:
1373 		size = 0x10000;
1374 		break;
1375 	case XGENE_DMA_RING_CFG_SIZE_512KB:
1376 		size = 0x80000;
1377 		break;
1378 	default:
1379 		chan_err(chan, "Unsupported cfg ring size %d\n", cfgsize);
1380 		return -EINVAL;
1381 	}
1382 
1383 	return size;
1384 }
1385 
1386 static void xgene_dma_delete_ring_one(struct xgene_dma_ring *ring)
1387 {
1388 	/* Clear DMA ring configurations */
1389 	xgene_dma_clear_ring(ring);
1390 
1391 	/* De-allocate DMA ring descriptor */
1392 	if (ring->desc_vaddr) {
1393 		dma_free_coherent(ring->pdma->dev, ring->size,
1394 				  ring->desc_vaddr, ring->desc_paddr);
1395 		ring->desc_vaddr = NULL;
1396 	}
1397 }
1398 
1399 static void xgene_dma_delete_chan_rings(struct xgene_dma_chan *chan)
1400 {
1401 	xgene_dma_delete_ring_one(&chan->rx_ring);
1402 	xgene_dma_delete_ring_one(&chan->tx_ring);
1403 }
1404 
1405 static int xgene_dma_create_ring_one(struct xgene_dma_chan *chan,
1406 				     struct xgene_dma_ring *ring,
1407 				     enum xgene_dma_ring_cfgsize cfgsize)
1408 {
1409 	int ret;
1410 
1411 	/* Setup DMA ring descriptor variables */
1412 	ring->pdma = chan->pdma;
1413 	ring->cfgsize = cfgsize;
1414 	ring->num = chan->pdma->ring_num++;
1415 	ring->id = XGENE_DMA_RING_ID_GET(ring->owner, ring->buf_num);
1416 
1417 	ret = xgene_dma_get_ring_size(chan, cfgsize);
1418 	if (ret <= 0)
1419 		return ret;
1420 	ring->size = ret;
1421 
1422 	/* Allocate memory for DMA ring descriptor */
1423 	ring->desc_vaddr = dma_zalloc_coherent(chan->dev, ring->size,
1424 					       &ring->desc_paddr, GFP_KERNEL);
1425 	if (!ring->desc_vaddr) {
1426 		chan_err(chan, "Failed to allocate ring desc\n");
1427 		return -ENOMEM;
1428 	}
1429 
1430 	/* Configure and enable DMA ring */
1431 	xgene_dma_set_ring_cmd(ring);
1432 	xgene_dma_setup_ring(ring);
1433 
1434 	return 0;
1435 }
1436 
1437 static int xgene_dma_create_chan_rings(struct xgene_dma_chan *chan)
1438 {
1439 	struct xgene_dma_ring *rx_ring = &chan->rx_ring;
1440 	struct xgene_dma_ring *tx_ring = &chan->tx_ring;
1441 	int ret;
1442 
1443 	/* Create DMA Rx ring descriptor */
1444 	rx_ring->owner = XGENE_DMA_RING_OWNER_CPU;
1445 	rx_ring->buf_num = XGENE_DMA_CPU_BUFNUM + chan->id;
1446 
1447 	ret = xgene_dma_create_ring_one(chan, rx_ring,
1448 					XGENE_DMA_RING_CFG_SIZE_64KB);
1449 	if (ret)
1450 		return ret;
1451 
1452 	chan_dbg(chan, "Rx ring id 0x%X num %d desc 0x%p\n",
1453 		 rx_ring->id, rx_ring->num, rx_ring->desc_vaddr);
1454 
1455 	/* Create DMA Tx ring descriptor */
1456 	tx_ring->owner = XGENE_DMA_RING_OWNER_DMA;
1457 	tx_ring->buf_num = XGENE_DMA_BUFNUM + chan->id;
1458 
1459 	ret = xgene_dma_create_ring_one(chan, tx_ring,
1460 					XGENE_DMA_RING_CFG_SIZE_64KB);
1461 	if (ret) {
1462 		xgene_dma_delete_ring_one(rx_ring);
1463 		return ret;
1464 	}
1465 
1466 	tx_ring->dst_ring_num = XGENE_DMA_RING_DST_ID(rx_ring->num);
1467 
1468 	chan_dbg(chan,
1469 		 "Tx ring id 0x%X num %d desc 0x%p\n",
1470 		 tx_ring->id, tx_ring->num, tx_ring->desc_vaddr);
1471 
1472 	/* Set the max outstanding request possible to this channel */
1473 	chan->max_outstanding = tx_ring->slots;
1474 
1475 	return ret;
1476 }
1477 
1478 static int xgene_dma_init_rings(struct xgene_dma *pdma)
1479 {
1480 	int ret, i, j;
1481 
1482 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
1483 		ret = xgene_dma_create_chan_rings(&pdma->chan[i]);
1484 		if (ret) {
1485 			for (j = 0; j < i; j++)
1486 				xgene_dma_delete_chan_rings(&pdma->chan[j]);
1487 			return ret;
1488 		}
1489 	}
1490 
1491 	return ret;
1492 }
1493 
1494 static void xgene_dma_enable(struct xgene_dma *pdma)
1495 {
1496 	u32 val;
1497 
1498 	/* Configure and enable DMA engine */
1499 	val = ioread32(pdma->csr_dma + XGENE_DMA_GCR);
1500 	XGENE_DMA_CH_SETUP(val);
1501 	XGENE_DMA_ENABLE(val);
1502 	iowrite32(val, pdma->csr_dma + XGENE_DMA_GCR);
1503 }
1504 
1505 static void xgene_dma_disable(struct xgene_dma *pdma)
1506 {
1507 	u32 val;
1508 
1509 	val = ioread32(pdma->csr_dma + XGENE_DMA_GCR);
1510 	XGENE_DMA_DISABLE(val);
1511 	iowrite32(val, pdma->csr_dma + XGENE_DMA_GCR);
1512 }
1513 
1514 static void xgene_dma_mask_interrupts(struct xgene_dma *pdma)
1515 {
1516 	/*
1517 	 * Mask DMA ring overflow, underflow and
1518 	 * AXI write/read error interrupts
1519 	 */
1520 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1521 		  pdma->csr_dma + XGENE_DMA_RING_INT0_MASK);
1522 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1523 		  pdma->csr_dma + XGENE_DMA_RING_INT1_MASK);
1524 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1525 		  pdma->csr_dma + XGENE_DMA_RING_INT2_MASK);
1526 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1527 		  pdma->csr_dma + XGENE_DMA_RING_INT3_MASK);
1528 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1529 		  pdma->csr_dma + XGENE_DMA_RING_INT4_MASK);
1530 
1531 	/* Mask DMA error interrupts */
1532 	iowrite32(XGENE_DMA_INT_ALL_MASK, pdma->csr_dma + XGENE_DMA_INT_MASK);
1533 }
1534 
1535 static void xgene_dma_unmask_interrupts(struct xgene_dma *pdma)
1536 {
1537 	/*
1538 	 * Unmask DMA ring overflow, underflow and
1539 	 * AXI write/read error interrupts
1540 	 */
1541 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1542 		  pdma->csr_dma + XGENE_DMA_RING_INT0_MASK);
1543 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1544 		  pdma->csr_dma + XGENE_DMA_RING_INT1_MASK);
1545 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1546 		  pdma->csr_dma + XGENE_DMA_RING_INT2_MASK);
1547 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1548 		  pdma->csr_dma + XGENE_DMA_RING_INT3_MASK);
1549 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1550 		  pdma->csr_dma + XGENE_DMA_RING_INT4_MASK);
1551 
1552 	/* Unmask DMA error interrupts */
1553 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1554 		  pdma->csr_dma + XGENE_DMA_INT_MASK);
1555 }
1556 
1557 static void xgene_dma_init_hw(struct xgene_dma *pdma)
1558 {
1559 	u32 val;
1560 
1561 	/* Associate DMA ring to corresponding ring HW */
1562 	iowrite32(XGENE_DMA_ASSOC_RING_MNGR1,
1563 		  pdma->csr_dma + XGENE_DMA_CFG_RING_WQ_ASSOC);
1564 
1565 	/* Configure RAID6 polynomial control setting */
1566 	if (is_pq_enabled(pdma))
1567 		iowrite32(XGENE_DMA_RAID6_MULTI_CTRL(0x1D),
1568 			  pdma->csr_dma + XGENE_DMA_RAID6_CONT);
1569 	else
1570 		dev_info(pdma->dev, "PQ is disabled in HW\n");
1571 
1572 	xgene_dma_enable(pdma);
1573 	xgene_dma_unmask_interrupts(pdma);
1574 
1575 	/* Get DMA id and version info */
1576 	val = ioread32(pdma->csr_dma + XGENE_DMA_IPBRR);
1577 
1578 	/* DMA device info */
1579 	dev_info(pdma->dev,
1580 		 "X-Gene DMA v%d.%02d.%02d driver registered %d channels",
1581 		 XGENE_DMA_REV_NO_RD(val), XGENE_DMA_BUS_ID_RD(val),
1582 		 XGENE_DMA_DEV_ID_RD(val), XGENE_DMA_MAX_CHANNEL);
1583 }
1584 
1585 static int xgene_dma_init_ring_mngr(struct xgene_dma *pdma)
1586 {
1587 	if (ioread32(pdma->csr_ring + XGENE_DMA_RING_CLKEN) &&
1588 	    (!ioread32(pdma->csr_ring + XGENE_DMA_RING_SRST)))
1589 		return 0;
1590 
1591 	iowrite32(0x3, pdma->csr_ring + XGENE_DMA_RING_CLKEN);
1592 	iowrite32(0x0, pdma->csr_ring + XGENE_DMA_RING_SRST);
1593 
1594 	/* Bring up memory */
1595 	iowrite32(0x0, pdma->csr_ring + XGENE_DMA_RING_MEM_RAM_SHUTDOWN);
1596 
1597 	/* Force a barrier */
1598 	ioread32(pdma->csr_ring + XGENE_DMA_RING_MEM_RAM_SHUTDOWN);
1599 
1600 	/* reset may take up to 1ms */
1601 	usleep_range(1000, 1100);
1602 
1603 	if (ioread32(pdma->csr_ring + XGENE_DMA_RING_BLK_MEM_RDY)
1604 		!= XGENE_DMA_RING_BLK_MEM_RDY_VAL) {
1605 		dev_err(pdma->dev,
1606 			"Failed to release ring mngr memory from shutdown\n");
1607 		return -ENODEV;
1608 	}
1609 
1610 	/* program threshold set 1 and all hysteresis */
1611 	iowrite32(XGENE_DMA_RING_THRESLD0_SET1_VAL,
1612 		  pdma->csr_ring + XGENE_DMA_RING_THRESLD0_SET1);
1613 	iowrite32(XGENE_DMA_RING_THRESLD1_SET1_VAL,
1614 		  pdma->csr_ring + XGENE_DMA_RING_THRESLD1_SET1);
1615 	iowrite32(XGENE_DMA_RING_HYSTERESIS_VAL,
1616 		  pdma->csr_ring + XGENE_DMA_RING_HYSTERESIS);
1617 
1618 	/* Enable QPcore and assign error queue */
1619 	iowrite32(XGENE_DMA_RING_ENABLE,
1620 		  pdma->csr_ring + XGENE_DMA_RING_CONFIG);
1621 
1622 	return 0;
1623 }
1624 
1625 static int xgene_dma_init_mem(struct xgene_dma *pdma)
1626 {
1627 	int ret;
1628 
1629 	ret = xgene_dma_init_ring_mngr(pdma);
1630 	if (ret)
1631 		return ret;
1632 
1633 	/* Bring up memory */
1634 	iowrite32(0x0, pdma->csr_dma + XGENE_DMA_MEM_RAM_SHUTDOWN);
1635 
1636 	/* Force a barrier */
1637 	ioread32(pdma->csr_dma + XGENE_DMA_MEM_RAM_SHUTDOWN);
1638 
1639 	/* reset may take up to 1ms */
1640 	usleep_range(1000, 1100);
1641 
1642 	if (ioread32(pdma->csr_dma + XGENE_DMA_BLK_MEM_RDY)
1643 		!= XGENE_DMA_BLK_MEM_RDY_VAL) {
1644 		dev_err(pdma->dev,
1645 			"Failed to release DMA memory from shutdown\n");
1646 		return -ENODEV;
1647 	}
1648 
1649 	return 0;
1650 }
1651 
1652 static int xgene_dma_request_irqs(struct xgene_dma *pdma)
1653 {
1654 	struct xgene_dma_chan *chan;
1655 	int ret, i, j;
1656 
1657 	/* Register DMA error irq */
1658 	ret = devm_request_irq(pdma->dev, pdma->err_irq, xgene_dma_err_isr,
1659 			       0, "dma_error", pdma);
1660 	if (ret) {
1661 		dev_err(pdma->dev,
1662 			"Failed to register error IRQ %d\n", pdma->err_irq);
1663 		return ret;
1664 	}
1665 
1666 	/* Register DMA channel rx irq */
1667 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
1668 		chan = &pdma->chan[i];
1669 		ret = devm_request_irq(chan->dev, chan->rx_irq,
1670 				       xgene_dma_chan_ring_isr,
1671 				       0, chan->name, chan);
1672 		if (ret) {
1673 			chan_err(chan, "Failed to register Rx IRQ %d\n",
1674 				 chan->rx_irq);
1675 			devm_free_irq(pdma->dev, pdma->err_irq, pdma);
1676 
1677 			for (j = 0; j < i; j++) {
1678 				chan = &pdma->chan[i];
1679 				devm_free_irq(chan->dev, chan->rx_irq, chan);
1680 			}
1681 
1682 			return ret;
1683 		}
1684 	}
1685 
1686 	return 0;
1687 }
1688 
1689 static void xgene_dma_free_irqs(struct xgene_dma *pdma)
1690 {
1691 	struct xgene_dma_chan *chan;
1692 	int i;
1693 
1694 	/* Free DMA device error irq */
1695 	devm_free_irq(pdma->dev, pdma->err_irq, pdma);
1696 
1697 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
1698 		chan = &pdma->chan[i];
1699 		devm_free_irq(chan->dev, chan->rx_irq, chan);
1700 	}
1701 }
1702 
1703 static void xgene_dma_set_caps(struct xgene_dma_chan *chan,
1704 			       struct dma_device *dma_dev)
1705 {
1706 	/* Initialize DMA device capability mask */
1707 	dma_cap_zero(dma_dev->cap_mask);
1708 
1709 	/* Set DMA device capability */
1710 	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
1711 	dma_cap_set(DMA_SG, dma_dev->cap_mask);
1712 
1713 	/* Basically here, the X-Gene SoC DMA engine channel 0 supports XOR
1714 	 * and channel 1 supports XOR, PQ both. First thing here is we have
1715 	 * mechanism in hw to enable/disable PQ/XOR supports on channel 1,
1716 	 * we can make sure this by reading SoC Efuse register.
1717 	 * Second thing, we have hw errata that if we run channel 0 and
1718 	 * channel 1 simultaneously with executing XOR and PQ request,
1719 	 * suddenly DMA engine hangs, So here we enable XOR on channel 0 only
1720 	 * if XOR and PQ supports on channel 1 is disabled.
1721 	 */
1722 	if ((chan->id == XGENE_DMA_PQ_CHANNEL) &&
1723 	    is_pq_enabled(chan->pdma)) {
1724 		dma_cap_set(DMA_PQ, dma_dev->cap_mask);
1725 		dma_cap_set(DMA_XOR, dma_dev->cap_mask);
1726 	} else if ((chan->id == XGENE_DMA_XOR_CHANNEL) &&
1727 		   !is_pq_enabled(chan->pdma)) {
1728 		dma_cap_set(DMA_XOR, dma_dev->cap_mask);
1729 	}
1730 
1731 	/* Set base and prep routines */
1732 	dma_dev->dev = chan->dev;
1733 	dma_dev->device_alloc_chan_resources = xgene_dma_alloc_chan_resources;
1734 	dma_dev->device_free_chan_resources = xgene_dma_free_chan_resources;
1735 	dma_dev->device_issue_pending = xgene_dma_issue_pending;
1736 	dma_dev->device_tx_status = xgene_dma_tx_status;
1737 	dma_dev->device_prep_dma_memcpy = xgene_dma_prep_memcpy;
1738 	dma_dev->device_prep_dma_sg = xgene_dma_prep_sg;
1739 
1740 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1741 		dma_dev->device_prep_dma_xor = xgene_dma_prep_xor;
1742 		dma_dev->max_xor = XGENE_DMA_MAX_XOR_SRC;
1743 		dma_dev->xor_align = DMAENGINE_ALIGN_64_BYTES;
1744 	}
1745 
1746 	if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1747 		dma_dev->device_prep_dma_pq = xgene_dma_prep_pq;
1748 		dma_dev->max_pq = XGENE_DMA_MAX_XOR_SRC;
1749 		dma_dev->pq_align = DMAENGINE_ALIGN_64_BYTES;
1750 	}
1751 }
1752 
1753 static int xgene_dma_async_register(struct xgene_dma *pdma, int id)
1754 {
1755 	struct xgene_dma_chan *chan = &pdma->chan[id];
1756 	struct dma_device *dma_dev = &pdma->dma_dev[id];
1757 	int ret;
1758 
1759 	chan->dma_chan.device = dma_dev;
1760 
1761 	spin_lock_init(&chan->lock);
1762 	INIT_LIST_HEAD(&chan->ld_pending);
1763 	INIT_LIST_HEAD(&chan->ld_running);
1764 	INIT_LIST_HEAD(&chan->ld_completed);
1765 	tasklet_init(&chan->tasklet, xgene_dma_tasklet_cb,
1766 		     (unsigned long)chan);
1767 
1768 	chan->pending = 0;
1769 	chan->desc_pool = NULL;
1770 	dma_cookie_init(&chan->dma_chan);
1771 
1772 	/* Setup dma device capabilities and prep routines */
1773 	xgene_dma_set_caps(chan, dma_dev);
1774 
1775 	/* Initialize DMA device list head */
1776 	INIT_LIST_HEAD(&dma_dev->channels);
1777 	list_add_tail(&chan->dma_chan.device_node, &dma_dev->channels);
1778 
1779 	/* Register with Linux async DMA framework*/
1780 	ret = dma_async_device_register(dma_dev);
1781 	if (ret) {
1782 		chan_err(chan, "Failed to register async device %d", ret);
1783 		tasklet_kill(&chan->tasklet);
1784 
1785 		return ret;
1786 	}
1787 
1788 	/* DMA capability info */
1789 	dev_info(pdma->dev,
1790 		 "%s: CAPABILITY ( %s%s%s%s)\n", dma_chan_name(&chan->dma_chan),
1791 		 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "MEMCPY " : "",
1792 		 dma_has_cap(DMA_SG, dma_dev->cap_mask) ? "SGCPY " : "",
1793 		 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "XOR " : "",
1794 		 dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "PQ " : "");
1795 
1796 	return 0;
1797 }
1798 
1799 static int xgene_dma_init_async(struct xgene_dma *pdma)
1800 {
1801 	int ret, i, j;
1802 
1803 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL ; i++) {
1804 		ret = xgene_dma_async_register(pdma, i);
1805 		if (ret) {
1806 			for (j = 0; j < i; j++) {
1807 				dma_async_device_unregister(&pdma->dma_dev[j]);
1808 				tasklet_kill(&pdma->chan[j].tasklet);
1809 			}
1810 
1811 			return ret;
1812 		}
1813 	}
1814 
1815 	return ret;
1816 }
1817 
1818 static void xgene_dma_async_unregister(struct xgene_dma *pdma)
1819 {
1820 	int i;
1821 
1822 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++)
1823 		dma_async_device_unregister(&pdma->dma_dev[i]);
1824 }
1825 
1826 static void xgene_dma_init_channels(struct xgene_dma *pdma)
1827 {
1828 	struct xgene_dma_chan *chan;
1829 	int i;
1830 
1831 	pdma->ring_num = XGENE_DMA_RING_NUM;
1832 
1833 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
1834 		chan = &pdma->chan[i];
1835 		chan->dev = pdma->dev;
1836 		chan->pdma = pdma;
1837 		chan->id = i;
1838 		snprintf(chan->name, sizeof(chan->name), "dmachan%d", chan->id);
1839 	}
1840 }
1841 
1842 static int xgene_dma_get_resources(struct platform_device *pdev,
1843 				   struct xgene_dma *pdma)
1844 {
1845 	struct resource *res;
1846 	int irq, i;
1847 
1848 	/* Get DMA csr region */
1849 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1850 	if (!res) {
1851 		dev_err(&pdev->dev, "Failed to get csr region\n");
1852 		return -ENXIO;
1853 	}
1854 
1855 	pdma->csr_dma = devm_ioremap(&pdev->dev, res->start,
1856 				     resource_size(res));
1857 	if (!pdma->csr_dma) {
1858 		dev_err(&pdev->dev, "Failed to ioremap csr region");
1859 		return -ENOMEM;
1860 	}
1861 
1862 	/* Get DMA ring csr region */
1863 	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1864 	if (!res) {
1865 		dev_err(&pdev->dev, "Failed to get ring csr region\n");
1866 		return -ENXIO;
1867 	}
1868 
1869 	pdma->csr_ring =  devm_ioremap(&pdev->dev, res->start,
1870 				       resource_size(res));
1871 	if (!pdma->csr_ring) {
1872 		dev_err(&pdev->dev, "Failed to ioremap ring csr region");
1873 		return -ENOMEM;
1874 	}
1875 
1876 	/* Get DMA ring cmd csr region */
1877 	res = platform_get_resource(pdev, IORESOURCE_MEM, 2);
1878 	if (!res) {
1879 		dev_err(&pdev->dev, "Failed to get ring cmd csr region\n");
1880 		return -ENXIO;
1881 	}
1882 
1883 	pdma->csr_ring_cmd = devm_ioremap(&pdev->dev, res->start,
1884 					  resource_size(res));
1885 	if (!pdma->csr_ring_cmd) {
1886 		dev_err(&pdev->dev, "Failed to ioremap ring cmd csr region");
1887 		return -ENOMEM;
1888 	}
1889 
1890 	pdma->csr_ring_cmd += XGENE_DMA_RING_CMD_SM_OFFSET;
1891 
1892 	/* Get efuse csr region */
1893 	res = platform_get_resource(pdev, IORESOURCE_MEM, 3);
1894 	if (!res) {
1895 		dev_err(&pdev->dev, "Failed to get efuse csr region\n");
1896 		return -ENXIO;
1897 	}
1898 
1899 	pdma->csr_efuse = devm_ioremap(&pdev->dev, res->start,
1900 				       resource_size(res));
1901 	if (!pdma->csr_efuse) {
1902 		dev_err(&pdev->dev, "Failed to ioremap efuse csr region");
1903 		return -ENOMEM;
1904 	}
1905 
1906 	/* Get DMA error interrupt */
1907 	irq = platform_get_irq(pdev, 0);
1908 	if (irq <= 0) {
1909 		dev_err(&pdev->dev, "Failed to get Error IRQ\n");
1910 		return -ENXIO;
1911 	}
1912 
1913 	pdma->err_irq = irq;
1914 
1915 	/* Get DMA Rx ring descriptor interrupts for all DMA channels */
1916 	for (i = 1; i <= XGENE_DMA_MAX_CHANNEL; i++) {
1917 		irq = platform_get_irq(pdev, i);
1918 		if (irq <= 0) {
1919 			dev_err(&pdev->dev, "Failed to get Rx IRQ\n");
1920 			return -ENXIO;
1921 		}
1922 
1923 		pdma->chan[i - 1].rx_irq = irq;
1924 	}
1925 
1926 	return 0;
1927 }
1928 
1929 static int xgene_dma_probe(struct platform_device *pdev)
1930 {
1931 	struct xgene_dma *pdma;
1932 	int ret, i;
1933 
1934 	pdma = devm_kzalloc(&pdev->dev, sizeof(*pdma), GFP_KERNEL);
1935 	if (!pdma)
1936 		return -ENOMEM;
1937 
1938 	pdma->dev = &pdev->dev;
1939 	platform_set_drvdata(pdev, pdma);
1940 
1941 	ret = xgene_dma_get_resources(pdev, pdma);
1942 	if (ret)
1943 		return ret;
1944 
1945 	pdma->clk = devm_clk_get(&pdev->dev, NULL);
1946 	if (IS_ERR(pdma->clk) && !ACPI_COMPANION(&pdev->dev)) {
1947 		dev_err(&pdev->dev, "Failed to get clk\n");
1948 		return PTR_ERR(pdma->clk);
1949 	}
1950 
1951 	/* Enable clk before accessing registers */
1952 	if (!IS_ERR(pdma->clk)) {
1953 		ret = clk_prepare_enable(pdma->clk);
1954 		if (ret) {
1955 			dev_err(&pdev->dev, "Failed to enable clk %d\n", ret);
1956 			return ret;
1957 		}
1958 	}
1959 
1960 	/* Remove DMA RAM out of shutdown */
1961 	ret = xgene_dma_init_mem(pdma);
1962 	if (ret)
1963 		goto err_clk_enable;
1964 
1965 	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(42));
1966 	if (ret) {
1967 		dev_err(&pdev->dev, "No usable DMA configuration\n");
1968 		goto err_dma_mask;
1969 	}
1970 
1971 	/* Initialize DMA channels software state */
1972 	xgene_dma_init_channels(pdma);
1973 
1974 	/* Configue DMA rings */
1975 	ret = xgene_dma_init_rings(pdma);
1976 	if (ret)
1977 		goto err_clk_enable;
1978 
1979 	ret = xgene_dma_request_irqs(pdma);
1980 	if (ret)
1981 		goto err_request_irq;
1982 
1983 	/* Configure and enable DMA engine */
1984 	xgene_dma_init_hw(pdma);
1985 
1986 	/* Register DMA device with linux async framework */
1987 	ret = xgene_dma_init_async(pdma);
1988 	if (ret)
1989 		goto err_async_init;
1990 
1991 	return 0;
1992 
1993 err_async_init:
1994 	xgene_dma_free_irqs(pdma);
1995 
1996 err_request_irq:
1997 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++)
1998 		xgene_dma_delete_chan_rings(&pdma->chan[i]);
1999 
2000 err_dma_mask:
2001 err_clk_enable:
2002 	if (!IS_ERR(pdma->clk))
2003 		clk_disable_unprepare(pdma->clk);
2004 
2005 	return ret;
2006 }
2007 
2008 static int xgene_dma_remove(struct platform_device *pdev)
2009 {
2010 	struct xgene_dma *pdma = platform_get_drvdata(pdev);
2011 	struct xgene_dma_chan *chan;
2012 	int i;
2013 
2014 	xgene_dma_async_unregister(pdma);
2015 
2016 	/* Mask interrupts and disable DMA engine */
2017 	xgene_dma_mask_interrupts(pdma);
2018 	xgene_dma_disable(pdma);
2019 	xgene_dma_free_irqs(pdma);
2020 
2021 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
2022 		chan = &pdma->chan[i];
2023 		tasklet_kill(&chan->tasklet);
2024 		xgene_dma_delete_chan_rings(chan);
2025 	}
2026 
2027 	if (!IS_ERR(pdma->clk))
2028 		clk_disable_unprepare(pdma->clk);
2029 
2030 	return 0;
2031 }
2032 
2033 #ifdef CONFIG_ACPI
2034 static const struct acpi_device_id xgene_dma_acpi_match_ptr[] = {
2035 	{"APMC0D43", 0},
2036 	{},
2037 };
2038 MODULE_DEVICE_TABLE(acpi, xgene_dma_acpi_match_ptr);
2039 #endif
2040 
2041 static const struct of_device_id xgene_dma_of_match_ptr[] = {
2042 	{.compatible = "apm,xgene-storm-dma",},
2043 	{},
2044 };
2045 MODULE_DEVICE_TABLE(of, xgene_dma_of_match_ptr);
2046 
2047 static struct platform_driver xgene_dma_driver = {
2048 	.probe = xgene_dma_probe,
2049 	.remove = xgene_dma_remove,
2050 	.driver = {
2051 		.name = "X-Gene-DMA",
2052 		.of_match_table = xgene_dma_of_match_ptr,
2053 		.acpi_match_table = ACPI_PTR(xgene_dma_acpi_match_ptr),
2054 	},
2055 };
2056 
2057 module_platform_driver(xgene_dma_driver);
2058 
2059 MODULE_DESCRIPTION("APM X-Gene SoC DMA driver");
2060 MODULE_AUTHOR("Rameshwar Prasad Sahu <rsahu@apm.com>");
2061 MODULE_AUTHOR("Loc Ho <lho@apm.com>");
2062 MODULE_LICENSE("GPL");
2063 MODULE_VERSION("1.0");
2064