xref: /linux/drivers/dma/xgene-dma.c (revision b14132797d8041a42e03f4ffa1e722da1425adfb)
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_DESC_CNT(v)		(((v) & 0x0001FFFE) >> 1)
63 #define XGENE_DMA_RING_ID_GET(owner, num)	(((owner) << 6) | (num))
64 #define XGENE_DMA_RING_DST_ID(v)		((1 << 10) | (v))
65 #define XGENE_DMA_RING_CMD_OFFSET		0x2C
66 #define XGENE_DMA_RING_CMD_BASE_OFFSET(v)	((v) << 6)
67 #define XGENE_DMA_RING_COHERENT_SET(m)		\
68 	(((u32 *)(m))[2] |= BIT(4))
69 #define XGENE_DMA_RING_ADDRL_SET(m, v)		\
70 	(((u32 *)(m))[2] |= (((v) >> 8) << 5))
71 #define XGENE_DMA_RING_ADDRH_SET(m, v)		\
72 	(((u32 *)(m))[3] |= ((v) >> 35))
73 #define XGENE_DMA_RING_ACCEPTLERR_SET(m)	\
74 	(((u32 *)(m))[3] |= BIT(19))
75 #define XGENE_DMA_RING_SIZE_SET(m, v)		\
76 	(((u32 *)(m))[3] |= ((v) << 23))
77 #define XGENE_DMA_RING_RECOMBBUF_SET(m)		\
78 	(((u32 *)(m))[3] |= BIT(27))
79 #define XGENE_DMA_RING_RECOMTIMEOUTL_SET(m)	\
80 	(((u32 *)(m))[3] |= (0x7 << 28))
81 #define XGENE_DMA_RING_RECOMTIMEOUTH_SET(m)	\
82 	(((u32 *)(m))[4] |= 0x3)
83 #define XGENE_DMA_RING_SELTHRSH_SET(m)		\
84 	(((u32 *)(m))[4] |= BIT(3))
85 #define XGENE_DMA_RING_TYPE_SET(m, v)		\
86 	(((u32 *)(m))[4] |= ((v) << 19))
87 
88 /* X-Gene DMA device csr registers and bit definitions */
89 #define XGENE_DMA_IPBRR				0x0
90 #define XGENE_DMA_DEV_ID_RD(v)			((v) & 0x00000FFF)
91 #define XGENE_DMA_BUS_ID_RD(v)			(((v) >> 12) & 3)
92 #define XGENE_DMA_REV_NO_RD(v)			(((v) >> 14) & 3)
93 #define XGENE_DMA_GCR				0x10
94 #define XGENE_DMA_CH_SETUP(v)			\
95 	((v) = ((v) & ~0x000FFFFF) | 0x000AAFFF)
96 #define XGENE_DMA_ENABLE(v)			((v) |= BIT(31))
97 #define XGENE_DMA_DISABLE(v)			((v) &= ~BIT(31))
98 #define XGENE_DMA_RAID6_CONT			0x14
99 #define XGENE_DMA_RAID6_MULTI_CTRL(v)		((v) << 24)
100 #define XGENE_DMA_INT				0x70
101 #define XGENE_DMA_INT_MASK			0x74
102 #define XGENE_DMA_INT_ALL_MASK			0xFFFFFFFF
103 #define XGENE_DMA_INT_ALL_UNMASK		0x0
104 #define XGENE_DMA_INT_MASK_SHIFT		0x14
105 #define XGENE_DMA_RING_INT0_MASK		0x90A0
106 #define XGENE_DMA_RING_INT1_MASK		0x90A8
107 #define XGENE_DMA_RING_INT2_MASK		0x90B0
108 #define XGENE_DMA_RING_INT3_MASK		0x90B8
109 #define XGENE_DMA_RING_INT4_MASK		0x90C0
110 #define XGENE_DMA_CFG_RING_WQ_ASSOC		0x90E0
111 #define XGENE_DMA_ASSOC_RING_MNGR1		0xFFFFFFFF
112 #define XGENE_DMA_MEM_RAM_SHUTDOWN		0xD070
113 #define XGENE_DMA_BLK_MEM_RDY			0xD074
114 #define XGENE_DMA_BLK_MEM_RDY_VAL		0xFFFFFFFF
115 #define XGENE_DMA_RING_CMD_SM_OFFSET		0x8000
116 
117 /* X-Gene SoC EFUSE csr register and bit defination */
118 #define XGENE_SOC_JTAG1_SHADOW			0x18
119 #define XGENE_DMA_PQ_DISABLE_MASK		BIT(13)
120 
121 /* X-Gene DMA Descriptor format */
122 #define XGENE_DMA_DESC_NV_BIT			BIT_ULL(50)
123 #define XGENE_DMA_DESC_IN_BIT			BIT_ULL(55)
124 #define XGENE_DMA_DESC_C_BIT			BIT_ULL(63)
125 #define XGENE_DMA_DESC_DR_BIT			BIT_ULL(61)
126 #define XGENE_DMA_DESC_ELERR_POS		46
127 #define XGENE_DMA_DESC_RTYPE_POS		56
128 #define XGENE_DMA_DESC_LERR_POS			60
129 #define XGENE_DMA_DESC_BUFLEN_POS		48
130 #define XGENE_DMA_DESC_HOENQ_NUM_POS		48
131 #define XGENE_DMA_DESC_ELERR_RD(m)		\
132 	(((m) >> XGENE_DMA_DESC_ELERR_POS) & 0x3)
133 #define XGENE_DMA_DESC_LERR_RD(m)		\
134 	(((m) >> XGENE_DMA_DESC_LERR_POS) & 0x7)
135 #define XGENE_DMA_DESC_STATUS(elerr, lerr)	\
136 	(((elerr) << 4) | (lerr))
137 
138 /* X-Gene DMA descriptor empty s/w signature */
139 #define XGENE_DMA_DESC_EMPTY_SIGNATURE		~0ULL
140 
141 /* X-Gene DMA configurable parameters defines */
142 #define XGENE_DMA_RING_NUM		512
143 #define XGENE_DMA_BUFNUM		0x0
144 #define XGENE_DMA_CPU_BUFNUM		0x18
145 #define XGENE_DMA_RING_OWNER_DMA	0x03
146 #define XGENE_DMA_RING_OWNER_CPU	0x0F
147 #define XGENE_DMA_RING_TYPE_REGULAR	0x01
148 #define XGENE_DMA_RING_WQ_DESC_SIZE	32	/* 32 Bytes */
149 #define XGENE_DMA_RING_NUM_CONFIG	5
150 #define XGENE_DMA_MAX_CHANNEL		4
151 #define XGENE_DMA_XOR_CHANNEL		0
152 #define XGENE_DMA_PQ_CHANNEL		1
153 #define XGENE_DMA_MAX_BYTE_CNT		0x4000	/* 16 KB */
154 #define XGENE_DMA_MAX_64B_DESC_BYTE_CNT	0x14000	/* 80 KB */
155 #define XGENE_DMA_MAX_XOR_SRC		5
156 #define XGENE_DMA_16K_BUFFER_LEN_CODE	0x0
157 #define XGENE_DMA_INVALID_LEN_CODE	0x7800000000000000ULL
158 
159 /* X-Gene DMA descriptor error codes */
160 #define ERR_DESC_AXI			0x01
161 #define ERR_BAD_DESC			0x02
162 #define ERR_READ_DATA_AXI		0x03
163 #define ERR_WRITE_DATA_AXI		0x04
164 #define ERR_FBP_TIMEOUT			0x05
165 #define ERR_ECC				0x06
166 #define ERR_DIFF_SIZE			0x08
167 #define ERR_SCT_GAT_LEN			0x09
168 #define ERR_CRC_ERR			0x11
169 #define ERR_CHKSUM			0x12
170 #define ERR_DIF				0x13
171 
172 /* X-Gene DMA error interrupt codes */
173 #define ERR_DIF_SIZE_INT		0x0
174 #define ERR_GS_ERR_INT			0x1
175 #define ERR_FPB_TIMEO_INT		0x2
176 #define ERR_WFIFO_OVF_INT		0x3
177 #define ERR_RFIFO_OVF_INT		0x4
178 #define ERR_WR_TIMEO_INT		0x5
179 #define ERR_RD_TIMEO_INT		0x6
180 #define ERR_WR_ERR_INT			0x7
181 #define ERR_RD_ERR_INT			0x8
182 #define ERR_BAD_DESC_INT		0x9
183 #define ERR_DESC_DST_INT		0xA
184 #define ERR_DESC_SRC_INT		0xB
185 
186 /* X-Gene DMA flyby operation code */
187 #define FLYBY_2SRC_XOR			0x80
188 #define FLYBY_3SRC_XOR			0x90
189 #define FLYBY_4SRC_XOR			0xA0
190 #define FLYBY_5SRC_XOR			0xB0
191 
192 /* X-Gene DMA SW descriptor flags */
193 #define XGENE_DMA_FLAG_64B_DESC		BIT(0)
194 
195 /* Define to dump X-Gene DMA descriptor */
196 #define XGENE_DMA_DESC_DUMP(desc, m)	\
197 	print_hex_dump(KERN_ERR, (m),	\
198 			DUMP_PREFIX_ADDRESS, 16, 8, (desc), 32, 0)
199 
200 #define to_dma_desc_sw(tx)		\
201 	container_of(tx, struct xgene_dma_desc_sw, tx)
202 #define to_dma_chan(dchan)		\
203 	container_of(dchan, struct xgene_dma_chan, dma_chan)
204 
205 #define chan_dbg(chan, fmt, arg...)	\
206 	dev_dbg(chan->dev, "%s: " fmt, chan->name, ##arg)
207 #define chan_err(chan, fmt, arg...)	\
208 	dev_err(chan->dev, "%s: " fmt, chan->name, ##arg)
209 
210 struct xgene_dma_desc_hw {
211 	__le64 m0;
212 	__le64 m1;
213 	__le64 m2;
214 	__le64 m3;
215 };
216 
217 enum xgene_dma_ring_cfgsize {
218 	XGENE_DMA_RING_CFG_SIZE_512B,
219 	XGENE_DMA_RING_CFG_SIZE_2KB,
220 	XGENE_DMA_RING_CFG_SIZE_16KB,
221 	XGENE_DMA_RING_CFG_SIZE_64KB,
222 	XGENE_DMA_RING_CFG_SIZE_512KB,
223 	XGENE_DMA_RING_CFG_SIZE_INVALID
224 };
225 
226 struct xgene_dma_ring {
227 	struct xgene_dma *pdma;
228 	u8 buf_num;
229 	u16 id;
230 	u16 num;
231 	u16 head;
232 	u16 owner;
233 	u16 slots;
234 	u16 dst_ring_num;
235 	u32 size;
236 	void __iomem *cmd;
237 	void __iomem *cmd_base;
238 	dma_addr_t desc_paddr;
239 	u32 state[XGENE_DMA_RING_NUM_CONFIG];
240 	enum xgene_dma_ring_cfgsize cfgsize;
241 	union {
242 		void *desc_vaddr;
243 		struct xgene_dma_desc_hw *desc_hw;
244 	};
245 };
246 
247 struct xgene_dma_desc_sw {
248 	struct xgene_dma_desc_hw desc1;
249 	struct xgene_dma_desc_hw desc2;
250 	u32 flags;
251 	struct list_head node;
252 	struct list_head tx_list;
253 	struct dma_async_tx_descriptor tx;
254 };
255 
256 /**
257  * struct xgene_dma_chan - internal representation of an X-Gene DMA channel
258  * @dma_chan: dmaengine channel object member
259  * @pdma: X-Gene DMA device structure reference
260  * @dev: struct device reference for dma mapping api
261  * @id: raw id of this channel
262  * @rx_irq: channel IRQ
263  * @name: name of X-Gene DMA channel
264  * @lock: serializes enqueue/dequeue operations to the descriptor pool
265  * @pending: number of transaction request pushed to DMA controller for
266  *	execution, but still waiting for completion,
267  * @max_outstanding: max number of outstanding request we can push to channel
268  * @ld_pending: descriptors which are queued to run, but have not yet been
269  *	submitted to the hardware for execution
270  * @ld_running: descriptors which are currently being executing by the hardware
271  * @ld_completed: descriptors which have finished execution by the hardware.
272  *	These descriptors have already had their cleanup actions run. They
273  *	are waiting for the ACK bit to be set by the async tx API.
274  * @desc_pool: descriptor pool for DMA operations
275  * @tasklet: bottom half where all completed descriptors cleans
276  * @tx_ring: transmit ring descriptor that we use to prepare actual
277  *	descriptors for further executions
278  * @rx_ring: receive ring descriptor that we use to get completed DMA
279  *	descriptors during cleanup time
280  */
281 struct xgene_dma_chan {
282 	struct dma_chan dma_chan;
283 	struct xgene_dma *pdma;
284 	struct device *dev;
285 	int id;
286 	int rx_irq;
287 	char name[10];
288 	spinlock_t lock;
289 	int pending;
290 	int max_outstanding;
291 	struct list_head ld_pending;
292 	struct list_head ld_running;
293 	struct list_head ld_completed;
294 	struct dma_pool *desc_pool;
295 	struct tasklet_struct tasklet;
296 	struct xgene_dma_ring tx_ring;
297 	struct xgene_dma_ring rx_ring;
298 };
299 
300 /**
301  * struct xgene_dma - internal representation of an X-Gene DMA device
302  * @err_irq: DMA error irq number
303  * @ring_num: start id number for DMA ring
304  * @csr_dma: base for DMA register access
305  * @csr_ring: base for DMA ring register access
306  * @csr_ring_cmd: base for DMA ring command register access
307  * @csr_efuse: base for efuse register access
308  * @dma_dev: embedded struct dma_device
309  * @chan: reference to X-Gene DMA channels
310  */
311 struct xgene_dma {
312 	struct device *dev;
313 	struct clk *clk;
314 	int err_irq;
315 	int ring_num;
316 	void __iomem *csr_dma;
317 	void __iomem *csr_ring;
318 	void __iomem *csr_ring_cmd;
319 	void __iomem *csr_efuse;
320 	struct dma_device dma_dev[XGENE_DMA_MAX_CHANNEL];
321 	struct xgene_dma_chan chan[XGENE_DMA_MAX_CHANNEL];
322 };
323 
324 static const char * const xgene_dma_desc_err[] = {
325 	[ERR_DESC_AXI] = "AXI error when reading src/dst link list",
326 	[ERR_BAD_DESC] = "ERR or El_ERR fields not set to zero in desc",
327 	[ERR_READ_DATA_AXI] = "AXI error when reading data",
328 	[ERR_WRITE_DATA_AXI] = "AXI error when writing data",
329 	[ERR_FBP_TIMEOUT] = "Timeout on bufpool fetch",
330 	[ERR_ECC] = "ECC double bit error",
331 	[ERR_DIFF_SIZE] = "Bufpool too small to hold all the DIF result",
332 	[ERR_SCT_GAT_LEN] = "Gather and scatter data length not same",
333 	[ERR_CRC_ERR] = "CRC error",
334 	[ERR_CHKSUM] = "Checksum error",
335 	[ERR_DIF] = "DIF error",
336 };
337 
338 static const char * const xgene_dma_err[] = {
339 	[ERR_DIF_SIZE_INT] = "DIF size error",
340 	[ERR_GS_ERR_INT] = "Gather scatter not same size error",
341 	[ERR_FPB_TIMEO_INT] = "Free pool time out error",
342 	[ERR_WFIFO_OVF_INT] = "Write FIFO over flow error",
343 	[ERR_RFIFO_OVF_INT] = "Read FIFO over flow error",
344 	[ERR_WR_TIMEO_INT] = "Write time out error",
345 	[ERR_RD_TIMEO_INT] = "Read time out error",
346 	[ERR_WR_ERR_INT] = "HBF bus write error",
347 	[ERR_RD_ERR_INT] = "HBF bus read error",
348 	[ERR_BAD_DESC_INT] = "Ring descriptor HE0 not set error",
349 	[ERR_DESC_DST_INT] = "HFB reading dst link address error",
350 	[ERR_DESC_SRC_INT] = "HFB reading src link address error",
351 };
352 
353 static bool is_pq_enabled(struct xgene_dma *pdma)
354 {
355 	u32 val;
356 
357 	val = ioread32(pdma->csr_efuse + XGENE_SOC_JTAG1_SHADOW);
358 	return !(val & XGENE_DMA_PQ_DISABLE_MASK);
359 }
360 
361 static u64 xgene_dma_encode_len(size_t len)
362 {
363 	return (len < XGENE_DMA_MAX_BYTE_CNT) ?
364 		((u64)len << XGENE_DMA_DESC_BUFLEN_POS) :
365 		XGENE_DMA_16K_BUFFER_LEN_CODE;
366 }
367 
368 static u8 xgene_dma_encode_xor_flyby(u32 src_cnt)
369 {
370 	static u8 flyby_type[] = {
371 		FLYBY_2SRC_XOR, /* Dummy */
372 		FLYBY_2SRC_XOR, /* Dummy */
373 		FLYBY_2SRC_XOR,
374 		FLYBY_3SRC_XOR,
375 		FLYBY_4SRC_XOR,
376 		FLYBY_5SRC_XOR
377 	};
378 
379 	return flyby_type[src_cnt];
380 }
381 
382 static u32 xgene_dma_ring_desc_cnt(struct xgene_dma_ring *ring)
383 {
384 	u32 __iomem *cmd_base = ring->cmd_base;
385 	u32 ring_state = ioread32(&cmd_base[1]);
386 
387 	return XGENE_DMA_RING_DESC_CNT(ring_state);
388 }
389 
390 static void xgene_dma_set_src_buffer(__le64 *ext8, size_t *len,
391 				     dma_addr_t *paddr)
392 {
393 	size_t nbytes = (*len < XGENE_DMA_MAX_BYTE_CNT) ?
394 			*len : XGENE_DMA_MAX_BYTE_CNT;
395 
396 	*ext8 |= cpu_to_le64(*paddr);
397 	*ext8 |= cpu_to_le64(xgene_dma_encode_len(nbytes));
398 	*len -= nbytes;
399 	*paddr += nbytes;
400 }
401 
402 static void xgene_dma_invalidate_buffer(__le64 *ext8)
403 {
404 	*ext8 |= cpu_to_le64(XGENE_DMA_INVALID_LEN_CODE);
405 }
406 
407 static __le64 *xgene_dma_lookup_ext8(struct xgene_dma_desc_hw *desc, int idx)
408 {
409 	switch (idx) {
410 	case 0:
411 		return &desc->m1;
412 	case 1:
413 		return &desc->m0;
414 	case 2:
415 		return &desc->m3;
416 	case 3:
417 		return &desc->m2;
418 	default:
419 		pr_err("Invalid dma descriptor index\n");
420 	}
421 
422 	return NULL;
423 }
424 
425 static void xgene_dma_init_desc(struct xgene_dma_desc_hw *desc,
426 				u16 dst_ring_num)
427 {
428 	desc->m0 |= cpu_to_le64(XGENE_DMA_DESC_IN_BIT);
429 	desc->m0 |= cpu_to_le64((u64)XGENE_DMA_RING_OWNER_DMA <<
430 				XGENE_DMA_DESC_RTYPE_POS);
431 	desc->m1 |= cpu_to_le64(XGENE_DMA_DESC_C_BIT);
432 	desc->m3 |= cpu_to_le64((u64)dst_ring_num <<
433 				XGENE_DMA_DESC_HOENQ_NUM_POS);
434 }
435 
436 static void xgene_dma_prep_cpy_desc(struct xgene_dma_chan *chan,
437 				    struct xgene_dma_desc_sw *desc_sw,
438 				    dma_addr_t dst, dma_addr_t src,
439 				    size_t len)
440 {
441 	struct xgene_dma_desc_hw *desc1, *desc2;
442 	int i;
443 
444 	/* Get 1st descriptor */
445 	desc1 = &desc_sw->desc1;
446 	xgene_dma_init_desc(desc1, chan->tx_ring.dst_ring_num);
447 
448 	/* Set destination address */
449 	desc1->m2 |= cpu_to_le64(XGENE_DMA_DESC_DR_BIT);
450 	desc1->m3 |= cpu_to_le64(dst);
451 
452 	/* Set 1st source address */
453 	xgene_dma_set_src_buffer(&desc1->m1, &len, &src);
454 
455 	if (!len)
456 		return;
457 
458 	/*
459 	 * We need to split this source buffer,
460 	 * and need to use 2nd descriptor
461 	 */
462 	desc2 = &desc_sw->desc2;
463 	desc1->m0 |= cpu_to_le64(XGENE_DMA_DESC_NV_BIT);
464 
465 	/* Set 2nd to 5th source address */
466 	for (i = 0; i < 4 && len; i++)
467 		xgene_dma_set_src_buffer(xgene_dma_lookup_ext8(desc2, i),
468 					 &len, &src);
469 
470 	/* Invalidate unused source address field */
471 	for (; i < 4; i++)
472 		xgene_dma_invalidate_buffer(xgene_dma_lookup_ext8(desc2, i));
473 
474 	/* Updated flag that we have prepared 64B descriptor */
475 	desc_sw->flags |= XGENE_DMA_FLAG_64B_DESC;
476 }
477 
478 static void xgene_dma_prep_xor_desc(struct xgene_dma_chan *chan,
479 				    struct xgene_dma_desc_sw *desc_sw,
480 				    dma_addr_t *dst, dma_addr_t *src,
481 				    u32 src_cnt, size_t *nbytes,
482 				    const u8 *scf)
483 {
484 	struct xgene_dma_desc_hw *desc1, *desc2;
485 	size_t len = *nbytes;
486 	int i;
487 
488 	desc1 = &desc_sw->desc1;
489 	desc2 = &desc_sw->desc2;
490 
491 	/* Initialize DMA descriptor */
492 	xgene_dma_init_desc(desc1, chan->tx_ring.dst_ring_num);
493 
494 	/* Set destination address */
495 	desc1->m2 |= cpu_to_le64(XGENE_DMA_DESC_DR_BIT);
496 	desc1->m3 |= cpu_to_le64(*dst);
497 
498 	/* We have multiple source addresses, so need to set NV bit*/
499 	desc1->m0 |= cpu_to_le64(XGENE_DMA_DESC_NV_BIT);
500 
501 	/* Set flyby opcode */
502 	desc1->m2 |= cpu_to_le64(xgene_dma_encode_xor_flyby(src_cnt));
503 
504 	/* Set 1st to 5th source addresses */
505 	for (i = 0; i < src_cnt; i++) {
506 		len = *nbytes;
507 		xgene_dma_set_src_buffer((i == 0) ? &desc1->m1 :
508 					 xgene_dma_lookup_ext8(desc2, i - 1),
509 					 &len, &src[i]);
510 		desc1->m2 |= cpu_to_le64((scf[i] << ((i + 1) * 8)));
511 	}
512 
513 	/* Update meta data */
514 	*nbytes = len;
515 	*dst += XGENE_DMA_MAX_BYTE_CNT;
516 
517 	/* We need always 64B descriptor to perform xor or pq operations */
518 	desc_sw->flags |= XGENE_DMA_FLAG_64B_DESC;
519 }
520 
521 static dma_cookie_t xgene_dma_tx_submit(struct dma_async_tx_descriptor *tx)
522 {
523 	struct xgene_dma_desc_sw *desc;
524 	struct xgene_dma_chan *chan;
525 	dma_cookie_t cookie;
526 
527 	if (unlikely(!tx))
528 		return -EINVAL;
529 
530 	chan = to_dma_chan(tx->chan);
531 	desc = to_dma_desc_sw(tx);
532 
533 	spin_lock_bh(&chan->lock);
534 
535 	cookie = dma_cookie_assign(tx);
536 
537 	/* Add this transaction list onto the tail of the pending queue */
538 	list_splice_tail_init(&desc->tx_list, &chan->ld_pending);
539 
540 	spin_unlock_bh(&chan->lock);
541 
542 	return cookie;
543 }
544 
545 static void xgene_dma_clean_descriptor(struct xgene_dma_chan *chan,
546 				       struct xgene_dma_desc_sw *desc)
547 {
548 	list_del(&desc->node);
549 	chan_dbg(chan, "LD %p free\n", desc);
550 	dma_pool_free(chan->desc_pool, desc, desc->tx.phys);
551 }
552 
553 static struct xgene_dma_desc_sw *xgene_dma_alloc_descriptor(
554 				 struct xgene_dma_chan *chan)
555 {
556 	struct xgene_dma_desc_sw *desc;
557 	dma_addr_t phys;
558 
559 	desc = dma_pool_alloc(chan->desc_pool, GFP_NOWAIT, &phys);
560 	if (!desc) {
561 		chan_err(chan, "Failed to allocate LDs\n");
562 		return NULL;
563 	}
564 
565 	memset(desc, 0, sizeof(*desc));
566 
567 	INIT_LIST_HEAD(&desc->tx_list);
568 	desc->tx.phys = phys;
569 	desc->tx.tx_submit = xgene_dma_tx_submit;
570 	dma_async_tx_descriptor_init(&desc->tx, &chan->dma_chan);
571 
572 	chan_dbg(chan, "LD %p allocated\n", desc);
573 
574 	return desc;
575 }
576 
577 /**
578  * xgene_dma_clean_completed_descriptor - free all descriptors which
579  * has been completed and acked
580  * @chan: X-Gene DMA channel
581  *
582  * This function is used on all completed and acked descriptors.
583  */
584 static void xgene_dma_clean_completed_descriptor(struct xgene_dma_chan *chan)
585 {
586 	struct xgene_dma_desc_sw *desc, *_desc;
587 
588 	/* Run the callback for each descriptor, in order */
589 	list_for_each_entry_safe(desc, _desc, &chan->ld_completed, node) {
590 		if (async_tx_test_ack(&desc->tx))
591 			xgene_dma_clean_descriptor(chan, desc);
592 	}
593 }
594 
595 /**
596  * xgene_dma_run_tx_complete_actions - cleanup a single link descriptor
597  * @chan: X-Gene DMA channel
598  * @desc: descriptor to cleanup and free
599  *
600  * This function is used on a descriptor which has been executed by the DMA
601  * controller. It will run any callbacks, submit any dependencies.
602  */
603 static void xgene_dma_run_tx_complete_actions(struct xgene_dma_chan *chan,
604 					      struct xgene_dma_desc_sw *desc)
605 {
606 	struct dma_async_tx_descriptor *tx = &desc->tx;
607 
608 	/*
609 	 * If this is not the last transaction in the group,
610 	 * then no need to complete cookie and run any callback as
611 	 * this is not the tx_descriptor which had been sent to caller
612 	 * of this DMA request
613 	 */
614 
615 	if (tx->cookie == 0)
616 		return;
617 
618 	dma_cookie_complete(tx);
619 
620 	/* Run the link descriptor callback function */
621 	if (tx->callback)
622 		tx->callback(tx->callback_param);
623 
624 	dma_descriptor_unmap(tx);
625 
626 	/* Run any dependencies */
627 	dma_run_dependencies(tx);
628 }
629 
630 /**
631  * xgene_dma_clean_running_descriptor - move the completed descriptor from
632  * ld_running to ld_completed
633  * @chan: X-Gene DMA channel
634  * @desc: the descriptor which is completed
635  *
636  * Free the descriptor directly if acked by async_tx api,
637  * else move it to queue ld_completed.
638  */
639 static void xgene_dma_clean_running_descriptor(struct xgene_dma_chan *chan,
640 					       struct xgene_dma_desc_sw *desc)
641 {
642 	/* Remove from the list of running transactions */
643 	list_del(&desc->node);
644 
645 	/*
646 	 * the client is allowed to attach dependent operations
647 	 * until 'ack' is set
648 	 */
649 	if (!async_tx_test_ack(&desc->tx)) {
650 		/*
651 		 * Move this descriptor to the list of descriptors which is
652 		 * completed, but still awaiting the 'ack' bit to be set.
653 		 */
654 		list_add_tail(&desc->node, &chan->ld_completed);
655 		return;
656 	}
657 
658 	chan_dbg(chan, "LD %p free\n", desc);
659 	dma_pool_free(chan->desc_pool, desc, desc->tx.phys);
660 }
661 
662 static int xgene_chan_xfer_request(struct xgene_dma_ring *ring,
663 				   struct xgene_dma_desc_sw *desc_sw)
664 {
665 	struct xgene_dma_desc_hw *desc_hw;
666 
667 	/* Check if can push more descriptor to hw for execution */
668 	if (xgene_dma_ring_desc_cnt(ring) > (ring->slots - 2))
669 		return -EBUSY;
670 
671 	/* Get hw descriptor from DMA tx ring */
672 	desc_hw = &ring->desc_hw[ring->head];
673 
674 	/*
675 	 * Increment the head count to point next
676 	 * descriptor for next time
677 	 */
678 	if (++ring->head == ring->slots)
679 		ring->head = 0;
680 
681 	/* Copy prepared sw descriptor data to hw descriptor */
682 	memcpy(desc_hw, &desc_sw->desc1, sizeof(*desc_hw));
683 
684 	/*
685 	 * Check if we have prepared 64B descriptor,
686 	 * in this case we need one more hw descriptor
687 	 */
688 	if (desc_sw->flags & XGENE_DMA_FLAG_64B_DESC) {
689 		desc_hw = &ring->desc_hw[ring->head];
690 
691 		if (++ring->head == ring->slots)
692 			ring->head = 0;
693 
694 		memcpy(desc_hw, &desc_sw->desc2, sizeof(*desc_hw));
695 	}
696 
697 	/* Notify the hw that we have descriptor ready for execution */
698 	iowrite32((desc_sw->flags & XGENE_DMA_FLAG_64B_DESC) ?
699 		  2 : 1, ring->cmd);
700 
701 	return 0;
702 }
703 
704 /**
705  * xgene_chan_xfer_ld_pending - push any pending transactions to hw
706  * @chan : X-Gene DMA channel
707  *
708  * LOCKING: must hold chan->lock
709  */
710 static void xgene_chan_xfer_ld_pending(struct xgene_dma_chan *chan)
711 {
712 	struct xgene_dma_desc_sw *desc_sw, *_desc_sw;
713 	int ret;
714 
715 	/*
716 	 * If the list of pending descriptors is empty, then we
717 	 * don't need to do any work at all
718 	 */
719 	if (list_empty(&chan->ld_pending)) {
720 		chan_dbg(chan, "No pending LDs\n");
721 		return;
722 	}
723 
724 	/*
725 	 * Move elements from the queue of pending transactions onto the list
726 	 * of running transactions and push it to hw for further executions
727 	 */
728 	list_for_each_entry_safe(desc_sw, _desc_sw, &chan->ld_pending, node) {
729 		/*
730 		 * Check if have pushed max number of transactions to hw
731 		 * as capable, so let's stop here and will push remaining
732 		 * elements from pening ld queue after completing some
733 		 * descriptors that we have already pushed
734 		 */
735 		if (chan->pending >= chan->max_outstanding)
736 			return;
737 
738 		ret = xgene_chan_xfer_request(&chan->tx_ring, desc_sw);
739 		if (ret)
740 			return;
741 
742 		/*
743 		 * Delete this element from ld pending queue and append it to
744 		 * ld running queue
745 		 */
746 		list_move_tail(&desc_sw->node, &chan->ld_running);
747 
748 		/* Increment the pending transaction count */
749 		chan->pending++;
750 	}
751 }
752 
753 /**
754  * xgene_dma_cleanup_descriptors - cleanup link descriptors which are completed
755  * and move them to ld_completed to free until flag 'ack' is set
756  * @chan: X-Gene DMA channel
757  *
758  * This function is used on descriptors which have been executed by the DMA
759  * controller. It will run any callbacks, submit any dependencies, then
760  * free these descriptors if flag 'ack' is set.
761  */
762 static void xgene_dma_cleanup_descriptors(struct xgene_dma_chan *chan)
763 {
764 	struct xgene_dma_ring *ring = &chan->rx_ring;
765 	struct xgene_dma_desc_sw *desc_sw, *_desc_sw;
766 	struct xgene_dma_desc_hw *desc_hw;
767 	struct list_head ld_completed;
768 	u8 status;
769 
770 	INIT_LIST_HEAD(&ld_completed);
771 
772 	spin_lock_bh(&chan->lock);
773 
774 	/* Clean already completed and acked descriptors */
775 	xgene_dma_clean_completed_descriptor(chan);
776 
777 	/* Move all completed descriptors to ld completed queue, in order */
778 	list_for_each_entry_safe(desc_sw, _desc_sw, &chan->ld_running, node) {
779 		/* Get subsequent hw descriptor from DMA rx ring */
780 		desc_hw = &ring->desc_hw[ring->head];
781 
782 		/* Check if this descriptor has been completed */
783 		if (unlikely(le64_to_cpu(desc_hw->m0) ==
784 			     XGENE_DMA_DESC_EMPTY_SIGNATURE))
785 			break;
786 
787 		if (++ring->head == ring->slots)
788 			ring->head = 0;
789 
790 		/* Check if we have any error with DMA transactions */
791 		status = XGENE_DMA_DESC_STATUS(
792 				XGENE_DMA_DESC_ELERR_RD(le64_to_cpu(
793 							desc_hw->m0)),
794 				XGENE_DMA_DESC_LERR_RD(le64_to_cpu(
795 						       desc_hw->m0)));
796 		if (status) {
797 			/* Print the DMA error type */
798 			chan_err(chan, "%s\n", xgene_dma_desc_err[status]);
799 
800 			/*
801 			 * We have DMA transactions error here. Dump DMA Tx
802 			 * and Rx descriptors for this request */
803 			XGENE_DMA_DESC_DUMP(&desc_sw->desc1,
804 					    "X-Gene DMA TX DESC1: ");
805 
806 			if (desc_sw->flags & XGENE_DMA_FLAG_64B_DESC)
807 				XGENE_DMA_DESC_DUMP(&desc_sw->desc2,
808 						    "X-Gene DMA TX DESC2: ");
809 
810 			XGENE_DMA_DESC_DUMP(desc_hw,
811 					    "X-Gene DMA RX ERR DESC: ");
812 		}
813 
814 		/* Notify the hw about this completed descriptor */
815 		iowrite32(-1, ring->cmd);
816 
817 		/* Mark this hw descriptor as processed */
818 		desc_hw->m0 = cpu_to_le64(XGENE_DMA_DESC_EMPTY_SIGNATURE);
819 
820 		/*
821 		 * Decrement the pending transaction count
822 		 * as we have processed one
823 		 */
824 		chan->pending--;
825 
826 		/*
827 		 * Delete this node from ld running queue and append it to
828 		 * ld completed queue for further processing
829 		 */
830 		list_move_tail(&desc_sw->node, &ld_completed);
831 	}
832 
833 	/*
834 	 * Start any pending transactions automatically
835 	 * In the ideal case, we keep the DMA controller busy while we go
836 	 * ahead and free the descriptors below.
837 	 */
838 	xgene_chan_xfer_ld_pending(chan);
839 
840 	spin_unlock_bh(&chan->lock);
841 
842 	/* Run the callback for each descriptor, in order */
843 	list_for_each_entry_safe(desc_sw, _desc_sw, &ld_completed, node) {
844 		xgene_dma_run_tx_complete_actions(chan, desc_sw);
845 		xgene_dma_clean_running_descriptor(chan, desc_sw);
846 	}
847 }
848 
849 static int xgene_dma_alloc_chan_resources(struct dma_chan *dchan)
850 {
851 	struct xgene_dma_chan *chan = to_dma_chan(dchan);
852 
853 	/* Has this channel already been allocated? */
854 	if (chan->desc_pool)
855 		return 1;
856 
857 	chan->desc_pool = dma_pool_create(chan->name, chan->dev,
858 					  sizeof(struct xgene_dma_desc_sw),
859 					  0, 0);
860 	if (!chan->desc_pool) {
861 		chan_err(chan, "Failed to allocate descriptor pool\n");
862 		return -ENOMEM;
863 	}
864 
865 	chan_dbg(chan, "Allocate descripto pool\n");
866 
867 	return 1;
868 }
869 
870 /**
871  * xgene_dma_free_desc_list - Free all descriptors in a queue
872  * @chan: X-Gene DMA channel
873  * @list: the list to free
874  *
875  * LOCKING: must hold chan->lock
876  */
877 static void xgene_dma_free_desc_list(struct xgene_dma_chan *chan,
878 				     struct list_head *list)
879 {
880 	struct xgene_dma_desc_sw *desc, *_desc;
881 
882 	list_for_each_entry_safe(desc, _desc, list, node)
883 		xgene_dma_clean_descriptor(chan, desc);
884 }
885 
886 static void xgene_dma_free_chan_resources(struct dma_chan *dchan)
887 {
888 	struct xgene_dma_chan *chan = to_dma_chan(dchan);
889 
890 	chan_dbg(chan, "Free all resources\n");
891 
892 	if (!chan->desc_pool)
893 		return;
894 
895 	/* Process all running descriptor */
896 	xgene_dma_cleanup_descriptors(chan);
897 
898 	spin_lock_bh(&chan->lock);
899 
900 	/* Clean all link descriptor queues */
901 	xgene_dma_free_desc_list(chan, &chan->ld_pending);
902 	xgene_dma_free_desc_list(chan, &chan->ld_running);
903 	xgene_dma_free_desc_list(chan, &chan->ld_completed);
904 
905 	spin_unlock_bh(&chan->lock);
906 
907 	/* Delete this channel DMA pool */
908 	dma_pool_destroy(chan->desc_pool);
909 	chan->desc_pool = NULL;
910 }
911 
912 static struct dma_async_tx_descriptor *xgene_dma_prep_memcpy(
913 	struct dma_chan *dchan, dma_addr_t dst, dma_addr_t src,
914 	size_t len, unsigned long flags)
915 {
916 	struct xgene_dma_desc_sw *first = NULL, *new;
917 	struct xgene_dma_chan *chan;
918 	size_t copy;
919 
920 	if (unlikely(!dchan || !len))
921 		return NULL;
922 
923 	chan = to_dma_chan(dchan);
924 
925 	do {
926 		/* Allocate the link descriptor from DMA pool */
927 		new = xgene_dma_alloc_descriptor(chan);
928 		if (!new)
929 			goto fail;
930 
931 		/* Create the largest transaction possible */
932 		copy = min_t(size_t, len, XGENE_DMA_MAX_64B_DESC_BYTE_CNT);
933 
934 		/* Prepare DMA descriptor */
935 		xgene_dma_prep_cpy_desc(chan, new, dst, src, copy);
936 
937 		if (!first)
938 			first = new;
939 
940 		new->tx.cookie = 0;
941 		async_tx_ack(&new->tx);
942 
943 		/* Update metadata */
944 		len -= copy;
945 		dst += copy;
946 		src += copy;
947 
948 		/* Insert the link descriptor to the LD ring */
949 		list_add_tail(&new->node, &first->tx_list);
950 	} while (len);
951 
952 	new->tx.flags = flags; /* client is in control of this ack */
953 	new->tx.cookie = -EBUSY;
954 	list_splice(&first->tx_list, &new->tx_list);
955 
956 	return &new->tx;
957 
958 fail:
959 	if (!first)
960 		return NULL;
961 
962 	xgene_dma_free_desc_list(chan, &first->tx_list);
963 	return NULL;
964 }
965 
966 static struct dma_async_tx_descriptor *xgene_dma_prep_sg(
967 	struct dma_chan *dchan, struct scatterlist *dst_sg,
968 	u32 dst_nents, struct scatterlist *src_sg,
969 	u32 src_nents, unsigned long flags)
970 {
971 	struct xgene_dma_desc_sw *first = NULL, *new = NULL;
972 	struct xgene_dma_chan *chan;
973 	size_t dst_avail, src_avail;
974 	dma_addr_t dst, src;
975 	size_t len;
976 
977 	if (unlikely(!dchan))
978 		return NULL;
979 
980 	if (unlikely(!dst_nents || !src_nents))
981 		return NULL;
982 
983 	if (unlikely(!dst_sg || !src_sg))
984 		return NULL;
985 
986 	chan = to_dma_chan(dchan);
987 
988 	/* Get prepared for the loop */
989 	dst_avail = sg_dma_len(dst_sg);
990 	src_avail = sg_dma_len(src_sg);
991 	dst_nents--;
992 	src_nents--;
993 
994 	/* Run until we are out of scatterlist entries */
995 	while (true) {
996 		/* Create the largest transaction possible */
997 		len = min_t(size_t, src_avail, dst_avail);
998 		len = min_t(size_t, len, XGENE_DMA_MAX_64B_DESC_BYTE_CNT);
999 		if (len == 0)
1000 			goto fetch;
1001 
1002 		dst = sg_dma_address(dst_sg) + sg_dma_len(dst_sg) - dst_avail;
1003 		src = sg_dma_address(src_sg) + sg_dma_len(src_sg) - src_avail;
1004 
1005 		/* Allocate the link descriptor from DMA pool */
1006 		new = xgene_dma_alloc_descriptor(chan);
1007 		if (!new)
1008 			goto fail;
1009 
1010 		/* Prepare DMA descriptor */
1011 		xgene_dma_prep_cpy_desc(chan, new, dst, src, len);
1012 
1013 		if (!first)
1014 			first = new;
1015 
1016 		new->tx.cookie = 0;
1017 		async_tx_ack(&new->tx);
1018 
1019 		/* update metadata */
1020 		dst_avail -= len;
1021 		src_avail -= len;
1022 
1023 		/* Insert the link descriptor to the LD ring */
1024 		list_add_tail(&new->node, &first->tx_list);
1025 
1026 fetch:
1027 		/* fetch the next dst scatterlist entry */
1028 		if (dst_avail == 0) {
1029 			/* no more entries: we're done */
1030 			if (dst_nents == 0)
1031 				break;
1032 
1033 			/* fetch the next entry: if there are no more: done */
1034 			dst_sg = sg_next(dst_sg);
1035 			if (!dst_sg)
1036 				break;
1037 
1038 			dst_nents--;
1039 			dst_avail = sg_dma_len(dst_sg);
1040 		}
1041 
1042 		/* fetch the next src scatterlist entry */
1043 		if (src_avail == 0) {
1044 			/* no more entries: we're done */
1045 			if (src_nents == 0)
1046 				break;
1047 
1048 			/* fetch the next entry: if there are no more: done */
1049 			src_sg = sg_next(src_sg);
1050 			if (!src_sg)
1051 				break;
1052 
1053 			src_nents--;
1054 			src_avail = sg_dma_len(src_sg);
1055 		}
1056 	}
1057 
1058 	if (!new)
1059 		return NULL;
1060 
1061 	new->tx.flags = flags; /* client is in control of this ack */
1062 	new->tx.cookie = -EBUSY;
1063 	list_splice(&first->tx_list, &new->tx_list);
1064 
1065 	return &new->tx;
1066 fail:
1067 	if (!first)
1068 		return NULL;
1069 
1070 	xgene_dma_free_desc_list(chan, &first->tx_list);
1071 	return NULL;
1072 }
1073 
1074 static struct dma_async_tx_descriptor *xgene_dma_prep_xor(
1075 	struct dma_chan *dchan, dma_addr_t dst,	dma_addr_t *src,
1076 	u32 src_cnt, size_t len, unsigned long flags)
1077 {
1078 	struct xgene_dma_desc_sw *first = NULL, *new;
1079 	struct xgene_dma_chan *chan;
1080 	static u8 multi[XGENE_DMA_MAX_XOR_SRC] = {
1081 				0x01, 0x01, 0x01, 0x01, 0x01};
1082 
1083 	if (unlikely(!dchan || !len))
1084 		return NULL;
1085 
1086 	chan = to_dma_chan(dchan);
1087 
1088 	do {
1089 		/* Allocate the link descriptor from DMA pool */
1090 		new = xgene_dma_alloc_descriptor(chan);
1091 		if (!new)
1092 			goto fail;
1093 
1094 		/* Prepare xor DMA descriptor */
1095 		xgene_dma_prep_xor_desc(chan, new, &dst, src,
1096 					src_cnt, &len, multi);
1097 
1098 		if (!first)
1099 			first = new;
1100 
1101 		new->tx.cookie = 0;
1102 		async_tx_ack(&new->tx);
1103 
1104 		/* Insert the link descriptor to the LD ring */
1105 		list_add_tail(&new->node, &first->tx_list);
1106 	} while (len);
1107 
1108 	new->tx.flags = flags; /* client is in control of this ack */
1109 	new->tx.cookie = -EBUSY;
1110 	list_splice(&first->tx_list, &new->tx_list);
1111 
1112 	return &new->tx;
1113 
1114 fail:
1115 	if (!first)
1116 		return NULL;
1117 
1118 	xgene_dma_free_desc_list(chan, &first->tx_list);
1119 	return NULL;
1120 }
1121 
1122 static struct dma_async_tx_descriptor *xgene_dma_prep_pq(
1123 	struct dma_chan *dchan, dma_addr_t *dst, dma_addr_t *src,
1124 	u32 src_cnt, const u8 *scf, size_t len, unsigned long flags)
1125 {
1126 	struct xgene_dma_desc_sw *first = NULL, *new;
1127 	struct xgene_dma_chan *chan;
1128 	size_t _len = len;
1129 	dma_addr_t _src[XGENE_DMA_MAX_XOR_SRC];
1130 	static u8 multi[XGENE_DMA_MAX_XOR_SRC] = {0x01, 0x01, 0x01, 0x01, 0x01};
1131 
1132 	if (unlikely(!dchan || !len))
1133 		return NULL;
1134 
1135 	chan = to_dma_chan(dchan);
1136 
1137 	/*
1138 	 * Save source addresses on local variable, may be we have to
1139 	 * prepare two descriptor to generate P and Q if both enabled
1140 	 * in the flags by client
1141 	 */
1142 	memcpy(_src, src, sizeof(*src) * src_cnt);
1143 
1144 	if (flags & DMA_PREP_PQ_DISABLE_P)
1145 		len = 0;
1146 
1147 	if (flags & DMA_PREP_PQ_DISABLE_Q)
1148 		_len = 0;
1149 
1150 	do {
1151 		/* Allocate the link descriptor from DMA pool */
1152 		new = xgene_dma_alloc_descriptor(chan);
1153 		if (!new)
1154 			goto fail;
1155 
1156 		if (!first)
1157 			first = new;
1158 
1159 		new->tx.cookie = 0;
1160 		async_tx_ack(&new->tx);
1161 
1162 		/* Insert the link descriptor to the LD ring */
1163 		list_add_tail(&new->node, &first->tx_list);
1164 
1165 		/*
1166 		 * Prepare DMA descriptor to generate P,
1167 		 * if DMA_PREP_PQ_DISABLE_P flag is not set
1168 		 */
1169 		if (len) {
1170 			xgene_dma_prep_xor_desc(chan, new, &dst[0], src,
1171 						src_cnt, &len, multi);
1172 			continue;
1173 		}
1174 
1175 		/*
1176 		 * Prepare DMA descriptor to generate Q,
1177 		 * if DMA_PREP_PQ_DISABLE_Q flag is not set
1178 		 */
1179 		if (_len) {
1180 			xgene_dma_prep_xor_desc(chan, new, &dst[1], _src,
1181 						src_cnt, &_len, scf);
1182 		}
1183 	} while (len || _len);
1184 
1185 	new->tx.flags = flags; /* client is in control of this ack */
1186 	new->tx.cookie = -EBUSY;
1187 	list_splice(&first->tx_list, &new->tx_list);
1188 
1189 	return &new->tx;
1190 
1191 fail:
1192 	if (!first)
1193 		return NULL;
1194 
1195 	xgene_dma_free_desc_list(chan, &first->tx_list);
1196 	return NULL;
1197 }
1198 
1199 static void xgene_dma_issue_pending(struct dma_chan *dchan)
1200 {
1201 	struct xgene_dma_chan *chan = to_dma_chan(dchan);
1202 
1203 	spin_lock_bh(&chan->lock);
1204 	xgene_chan_xfer_ld_pending(chan);
1205 	spin_unlock_bh(&chan->lock);
1206 }
1207 
1208 static enum dma_status xgene_dma_tx_status(struct dma_chan *dchan,
1209 					   dma_cookie_t cookie,
1210 					   struct dma_tx_state *txstate)
1211 {
1212 	return dma_cookie_status(dchan, cookie, txstate);
1213 }
1214 
1215 static void xgene_dma_tasklet_cb(unsigned long data)
1216 {
1217 	struct xgene_dma_chan *chan = (struct xgene_dma_chan *)data;
1218 
1219 	/* Run all cleanup for descriptors which have been completed */
1220 	xgene_dma_cleanup_descriptors(chan);
1221 
1222 	/* Re-enable DMA channel IRQ */
1223 	enable_irq(chan->rx_irq);
1224 }
1225 
1226 static irqreturn_t xgene_dma_chan_ring_isr(int irq, void *id)
1227 {
1228 	struct xgene_dma_chan *chan = (struct xgene_dma_chan *)id;
1229 
1230 	BUG_ON(!chan);
1231 
1232 	/*
1233 	 * Disable DMA channel IRQ until we process completed
1234 	 * descriptors
1235 	 */
1236 	disable_irq_nosync(chan->rx_irq);
1237 
1238 	/*
1239 	 * Schedule the tasklet to handle all cleanup of the current
1240 	 * transaction. It will start a new transaction if there is
1241 	 * one pending.
1242 	 */
1243 	tasklet_schedule(&chan->tasklet);
1244 
1245 	return IRQ_HANDLED;
1246 }
1247 
1248 static irqreturn_t xgene_dma_err_isr(int irq, void *id)
1249 {
1250 	struct xgene_dma *pdma = (struct xgene_dma *)id;
1251 	unsigned long int_mask;
1252 	u32 val, i;
1253 
1254 	val = ioread32(pdma->csr_dma + XGENE_DMA_INT);
1255 
1256 	/* Clear DMA interrupts */
1257 	iowrite32(val, pdma->csr_dma + XGENE_DMA_INT);
1258 
1259 	/* Print DMA error info */
1260 	int_mask = val >> XGENE_DMA_INT_MASK_SHIFT;
1261 	for_each_set_bit(i, &int_mask, ARRAY_SIZE(xgene_dma_err))
1262 		dev_err(pdma->dev,
1263 			"Interrupt status 0x%08X %s\n", val, xgene_dma_err[i]);
1264 
1265 	return IRQ_HANDLED;
1266 }
1267 
1268 static void xgene_dma_wr_ring_state(struct xgene_dma_ring *ring)
1269 {
1270 	int i;
1271 
1272 	iowrite32(ring->num, ring->pdma->csr_ring + XGENE_DMA_RING_STATE);
1273 
1274 	for (i = 0; i < XGENE_DMA_RING_NUM_CONFIG; i++)
1275 		iowrite32(ring->state[i], ring->pdma->csr_ring +
1276 			  XGENE_DMA_RING_STATE_WR_BASE + (i * 4));
1277 }
1278 
1279 static void xgene_dma_clr_ring_state(struct xgene_dma_ring *ring)
1280 {
1281 	memset(ring->state, 0, sizeof(u32) * XGENE_DMA_RING_NUM_CONFIG);
1282 	xgene_dma_wr_ring_state(ring);
1283 }
1284 
1285 static void xgene_dma_setup_ring(struct xgene_dma_ring *ring)
1286 {
1287 	void *ring_cfg = ring->state;
1288 	u64 addr = ring->desc_paddr;
1289 	u32 i, val;
1290 
1291 	ring->slots = ring->size / XGENE_DMA_RING_WQ_DESC_SIZE;
1292 
1293 	/* Clear DMA ring state */
1294 	xgene_dma_clr_ring_state(ring);
1295 
1296 	/* Set DMA ring type */
1297 	XGENE_DMA_RING_TYPE_SET(ring_cfg, XGENE_DMA_RING_TYPE_REGULAR);
1298 
1299 	if (ring->owner == XGENE_DMA_RING_OWNER_DMA) {
1300 		/* Set recombination buffer and timeout */
1301 		XGENE_DMA_RING_RECOMBBUF_SET(ring_cfg);
1302 		XGENE_DMA_RING_RECOMTIMEOUTL_SET(ring_cfg);
1303 		XGENE_DMA_RING_RECOMTIMEOUTH_SET(ring_cfg);
1304 	}
1305 
1306 	/* Initialize DMA ring state */
1307 	XGENE_DMA_RING_SELTHRSH_SET(ring_cfg);
1308 	XGENE_DMA_RING_ACCEPTLERR_SET(ring_cfg);
1309 	XGENE_DMA_RING_COHERENT_SET(ring_cfg);
1310 	XGENE_DMA_RING_ADDRL_SET(ring_cfg, addr);
1311 	XGENE_DMA_RING_ADDRH_SET(ring_cfg, addr);
1312 	XGENE_DMA_RING_SIZE_SET(ring_cfg, ring->cfgsize);
1313 
1314 	/* Write DMA ring configurations */
1315 	xgene_dma_wr_ring_state(ring);
1316 
1317 	/* Set DMA ring id */
1318 	iowrite32(XGENE_DMA_RING_ID_SETUP(ring->id),
1319 		  ring->pdma->csr_ring + XGENE_DMA_RING_ID);
1320 
1321 	/* Set DMA ring buffer */
1322 	iowrite32(XGENE_DMA_RING_ID_BUF_SETUP(ring->num),
1323 		  ring->pdma->csr_ring + XGENE_DMA_RING_ID_BUF);
1324 
1325 	if (ring->owner != XGENE_DMA_RING_OWNER_CPU)
1326 		return;
1327 
1328 	/* Set empty signature to DMA Rx ring descriptors */
1329 	for (i = 0; i < ring->slots; i++) {
1330 		struct xgene_dma_desc_hw *desc;
1331 
1332 		desc = &ring->desc_hw[i];
1333 		desc->m0 = cpu_to_le64(XGENE_DMA_DESC_EMPTY_SIGNATURE);
1334 	}
1335 
1336 	/* Enable DMA Rx ring interrupt */
1337 	val = ioread32(ring->pdma->csr_ring + XGENE_DMA_RING_NE_INT_MODE);
1338 	XGENE_DMA_RING_NE_INT_MODE_SET(val, ring->buf_num);
1339 	iowrite32(val, ring->pdma->csr_ring + XGENE_DMA_RING_NE_INT_MODE);
1340 }
1341 
1342 static void xgene_dma_clear_ring(struct xgene_dma_ring *ring)
1343 {
1344 	u32 ring_id, val;
1345 
1346 	if (ring->owner == XGENE_DMA_RING_OWNER_CPU) {
1347 		/* Disable DMA Rx ring interrupt */
1348 		val = ioread32(ring->pdma->csr_ring +
1349 			       XGENE_DMA_RING_NE_INT_MODE);
1350 		XGENE_DMA_RING_NE_INT_MODE_RESET(val, ring->buf_num);
1351 		iowrite32(val, ring->pdma->csr_ring +
1352 			  XGENE_DMA_RING_NE_INT_MODE);
1353 	}
1354 
1355 	/* Clear DMA ring state */
1356 	ring_id = XGENE_DMA_RING_ID_SETUP(ring->id);
1357 	iowrite32(ring_id, ring->pdma->csr_ring + XGENE_DMA_RING_ID);
1358 
1359 	iowrite32(0, ring->pdma->csr_ring + XGENE_DMA_RING_ID_BUF);
1360 	xgene_dma_clr_ring_state(ring);
1361 }
1362 
1363 static void xgene_dma_set_ring_cmd(struct xgene_dma_ring *ring)
1364 {
1365 	ring->cmd_base = ring->pdma->csr_ring_cmd +
1366 				XGENE_DMA_RING_CMD_BASE_OFFSET((ring->num -
1367 							  XGENE_DMA_RING_NUM));
1368 
1369 	ring->cmd = ring->cmd_base + XGENE_DMA_RING_CMD_OFFSET;
1370 }
1371 
1372 static int xgene_dma_get_ring_size(struct xgene_dma_chan *chan,
1373 				   enum xgene_dma_ring_cfgsize cfgsize)
1374 {
1375 	int size;
1376 
1377 	switch (cfgsize) {
1378 	case XGENE_DMA_RING_CFG_SIZE_512B:
1379 		size = 0x200;
1380 		break;
1381 	case XGENE_DMA_RING_CFG_SIZE_2KB:
1382 		size = 0x800;
1383 		break;
1384 	case XGENE_DMA_RING_CFG_SIZE_16KB:
1385 		size = 0x4000;
1386 		break;
1387 	case XGENE_DMA_RING_CFG_SIZE_64KB:
1388 		size = 0x10000;
1389 		break;
1390 	case XGENE_DMA_RING_CFG_SIZE_512KB:
1391 		size = 0x80000;
1392 		break;
1393 	default:
1394 		chan_err(chan, "Unsupported cfg ring size %d\n", cfgsize);
1395 		return -EINVAL;
1396 	}
1397 
1398 	return size;
1399 }
1400 
1401 static void xgene_dma_delete_ring_one(struct xgene_dma_ring *ring)
1402 {
1403 	/* Clear DMA ring configurations */
1404 	xgene_dma_clear_ring(ring);
1405 
1406 	/* De-allocate DMA ring descriptor */
1407 	if (ring->desc_vaddr) {
1408 		dma_free_coherent(ring->pdma->dev, ring->size,
1409 				  ring->desc_vaddr, ring->desc_paddr);
1410 		ring->desc_vaddr = NULL;
1411 	}
1412 }
1413 
1414 static void xgene_dma_delete_chan_rings(struct xgene_dma_chan *chan)
1415 {
1416 	xgene_dma_delete_ring_one(&chan->rx_ring);
1417 	xgene_dma_delete_ring_one(&chan->tx_ring);
1418 }
1419 
1420 static int xgene_dma_create_ring_one(struct xgene_dma_chan *chan,
1421 				     struct xgene_dma_ring *ring,
1422 				     enum xgene_dma_ring_cfgsize cfgsize)
1423 {
1424 	/* Setup DMA ring descriptor variables */
1425 	ring->pdma = chan->pdma;
1426 	ring->cfgsize = cfgsize;
1427 	ring->num = chan->pdma->ring_num++;
1428 	ring->id = XGENE_DMA_RING_ID_GET(ring->owner, ring->buf_num);
1429 
1430 	ring->size = xgene_dma_get_ring_size(chan, cfgsize);
1431 	if (ring->size <= 0)
1432 		return ring->size;
1433 
1434 	/* Allocate memory for DMA ring descriptor */
1435 	ring->desc_vaddr = dma_zalloc_coherent(chan->dev, ring->size,
1436 					       &ring->desc_paddr, GFP_KERNEL);
1437 	if (!ring->desc_vaddr) {
1438 		chan_err(chan, "Failed to allocate ring desc\n");
1439 		return -ENOMEM;
1440 	}
1441 
1442 	/* Configure and enable DMA ring */
1443 	xgene_dma_set_ring_cmd(ring);
1444 	xgene_dma_setup_ring(ring);
1445 
1446 	return 0;
1447 }
1448 
1449 static int xgene_dma_create_chan_rings(struct xgene_dma_chan *chan)
1450 {
1451 	struct xgene_dma_ring *rx_ring = &chan->rx_ring;
1452 	struct xgene_dma_ring *tx_ring = &chan->tx_ring;
1453 	int ret;
1454 
1455 	/* Create DMA Rx ring descriptor */
1456 	rx_ring->owner = XGENE_DMA_RING_OWNER_CPU;
1457 	rx_ring->buf_num = XGENE_DMA_CPU_BUFNUM + chan->id;
1458 
1459 	ret = xgene_dma_create_ring_one(chan, rx_ring,
1460 					XGENE_DMA_RING_CFG_SIZE_64KB);
1461 	if (ret)
1462 		return ret;
1463 
1464 	chan_dbg(chan, "Rx ring id 0x%X num %d desc 0x%p\n",
1465 		 rx_ring->id, rx_ring->num, rx_ring->desc_vaddr);
1466 
1467 	/* Create DMA Tx ring descriptor */
1468 	tx_ring->owner = XGENE_DMA_RING_OWNER_DMA;
1469 	tx_ring->buf_num = XGENE_DMA_BUFNUM + chan->id;
1470 
1471 	ret = xgene_dma_create_ring_one(chan, tx_ring,
1472 					XGENE_DMA_RING_CFG_SIZE_64KB);
1473 	if (ret) {
1474 		xgene_dma_delete_ring_one(rx_ring);
1475 		return ret;
1476 	}
1477 
1478 	tx_ring->dst_ring_num = XGENE_DMA_RING_DST_ID(rx_ring->num);
1479 
1480 	chan_dbg(chan,
1481 		 "Tx ring id 0x%X num %d desc 0x%p\n",
1482 		 tx_ring->id, tx_ring->num, tx_ring->desc_vaddr);
1483 
1484 	/* Set the max outstanding request possible to this channel */
1485 	chan->max_outstanding = rx_ring->slots;
1486 
1487 	return ret;
1488 }
1489 
1490 static int xgene_dma_init_rings(struct xgene_dma *pdma)
1491 {
1492 	int ret, i, j;
1493 
1494 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
1495 		ret = xgene_dma_create_chan_rings(&pdma->chan[i]);
1496 		if (ret) {
1497 			for (j = 0; j < i; j++)
1498 				xgene_dma_delete_chan_rings(&pdma->chan[j]);
1499 			return ret;
1500 		}
1501 	}
1502 
1503 	return ret;
1504 }
1505 
1506 static void xgene_dma_enable(struct xgene_dma *pdma)
1507 {
1508 	u32 val;
1509 
1510 	/* Configure and enable DMA engine */
1511 	val = ioread32(pdma->csr_dma + XGENE_DMA_GCR);
1512 	XGENE_DMA_CH_SETUP(val);
1513 	XGENE_DMA_ENABLE(val);
1514 	iowrite32(val, pdma->csr_dma + XGENE_DMA_GCR);
1515 }
1516 
1517 static void xgene_dma_disable(struct xgene_dma *pdma)
1518 {
1519 	u32 val;
1520 
1521 	val = ioread32(pdma->csr_dma + XGENE_DMA_GCR);
1522 	XGENE_DMA_DISABLE(val);
1523 	iowrite32(val, pdma->csr_dma + XGENE_DMA_GCR);
1524 }
1525 
1526 static void xgene_dma_mask_interrupts(struct xgene_dma *pdma)
1527 {
1528 	/*
1529 	 * Mask DMA ring overflow, underflow and
1530 	 * AXI write/read error interrupts
1531 	 */
1532 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1533 		  pdma->csr_dma + XGENE_DMA_RING_INT0_MASK);
1534 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1535 		  pdma->csr_dma + XGENE_DMA_RING_INT1_MASK);
1536 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1537 		  pdma->csr_dma + XGENE_DMA_RING_INT2_MASK);
1538 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1539 		  pdma->csr_dma + XGENE_DMA_RING_INT3_MASK);
1540 	iowrite32(XGENE_DMA_INT_ALL_MASK,
1541 		  pdma->csr_dma + XGENE_DMA_RING_INT4_MASK);
1542 
1543 	/* Mask DMA error interrupts */
1544 	iowrite32(XGENE_DMA_INT_ALL_MASK, pdma->csr_dma + XGENE_DMA_INT_MASK);
1545 }
1546 
1547 static void xgene_dma_unmask_interrupts(struct xgene_dma *pdma)
1548 {
1549 	/*
1550 	 * Unmask DMA ring overflow, underflow and
1551 	 * AXI write/read error interrupts
1552 	 */
1553 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1554 		  pdma->csr_dma + XGENE_DMA_RING_INT0_MASK);
1555 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1556 		  pdma->csr_dma + XGENE_DMA_RING_INT1_MASK);
1557 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1558 		  pdma->csr_dma + XGENE_DMA_RING_INT2_MASK);
1559 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1560 		  pdma->csr_dma + XGENE_DMA_RING_INT3_MASK);
1561 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1562 		  pdma->csr_dma + XGENE_DMA_RING_INT4_MASK);
1563 
1564 	/* Unmask DMA error interrupts */
1565 	iowrite32(XGENE_DMA_INT_ALL_UNMASK,
1566 		  pdma->csr_dma + XGENE_DMA_INT_MASK);
1567 }
1568 
1569 static void xgene_dma_init_hw(struct xgene_dma *pdma)
1570 {
1571 	u32 val;
1572 
1573 	/* Associate DMA ring to corresponding ring HW */
1574 	iowrite32(XGENE_DMA_ASSOC_RING_MNGR1,
1575 		  pdma->csr_dma + XGENE_DMA_CFG_RING_WQ_ASSOC);
1576 
1577 	/* Configure RAID6 polynomial control setting */
1578 	if (is_pq_enabled(pdma))
1579 		iowrite32(XGENE_DMA_RAID6_MULTI_CTRL(0x1D),
1580 			  pdma->csr_dma + XGENE_DMA_RAID6_CONT);
1581 	else
1582 		dev_info(pdma->dev, "PQ is disabled in HW\n");
1583 
1584 	xgene_dma_enable(pdma);
1585 	xgene_dma_unmask_interrupts(pdma);
1586 
1587 	/* Get DMA id and version info */
1588 	val = ioread32(pdma->csr_dma + XGENE_DMA_IPBRR);
1589 
1590 	/* DMA device info */
1591 	dev_info(pdma->dev,
1592 		 "X-Gene DMA v%d.%02d.%02d driver registered %d channels",
1593 		 XGENE_DMA_REV_NO_RD(val), XGENE_DMA_BUS_ID_RD(val),
1594 		 XGENE_DMA_DEV_ID_RD(val), XGENE_DMA_MAX_CHANNEL);
1595 }
1596 
1597 static int xgene_dma_init_ring_mngr(struct xgene_dma *pdma)
1598 {
1599 	if (ioread32(pdma->csr_ring + XGENE_DMA_RING_CLKEN) &&
1600 	    (!ioread32(pdma->csr_ring + XGENE_DMA_RING_SRST)))
1601 		return 0;
1602 
1603 	iowrite32(0x3, pdma->csr_ring + XGENE_DMA_RING_CLKEN);
1604 	iowrite32(0x0, pdma->csr_ring + XGENE_DMA_RING_SRST);
1605 
1606 	/* Bring up memory */
1607 	iowrite32(0x0, pdma->csr_ring + XGENE_DMA_RING_MEM_RAM_SHUTDOWN);
1608 
1609 	/* Force a barrier */
1610 	ioread32(pdma->csr_ring + XGENE_DMA_RING_MEM_RAM_SHUTDOWN);
1611 
1612 	/* reset may take up to 1ms */
1613 	usleep_range(1000, 1100);
1614 
1615 	if (ioread32(pdma->csr_ring + XGENE_DMA_RING_BLK_MEM_RDY)
1616 		!= XGENE_DMA_RING_BLK_MEM_RDY_VAL) {
1617 		dev_err(pdma->dev,
1618 			"Failed to release ring mngr memory from shutdown\n");
1619 		return -ENODEV;
1620 	}
1621 
1622 	/* program threshold set 1 and all hysteresis */
1623 	iowrite32(XGENE_DMA_RING_THRESLD0_SET1_VAL,
1624 		  pdma->csr_ring + XGENE_DMA_RING_THRESLD0_SET1);
1625 	iowrite32(XGENE_DMA_RING_THRESLD1_SET1_VAL,
1626 		  pdma->csr_ring + XGENE_DMA_RING_THRESLD1_SET1);
1627 	iowrite32(XGENE_DMA_RING_HYSTERESIS_VAL,
1628 		  pdma->csr_ring + XGENE_DMA_RING_HYSTERESIS);
1629 
1630 	/* Enable QPcore and assign error queue */
1631 	iowrite32(XGENE_DMA_RING_ENABLE,
1632 		  pdma->csr_ring + XGENE_DMA_RING_CONFIG);
1633 
1634 	return 0;
1635 }
1636 
1637 static int xgene_dma_init_mem(struct xgene_dma *pdma)
1638 {
1639 	int ret;
1640 
1641 	ret = xgene_dma_init_ring_mngr(pdma);
1642 	if (ret)
1643 		return ret;
1644 
1645 	/* Bring up memory */
1646 	iowrite32(0x0, pdma->csr_dma + XGENE_DMA_MEM_RAM_SHUTDOWN);
1647 
1648 	/* Force a barrier */
1649 	ioread32(pdma->csr_dma + XGENE_DMA_MEM_RAM_SHUTDOWN);
1650 
1651 	/* reset may take up to 1ms */
1652 	usleep_range(1000, 1100);
1653 
1654 	if (ioread32(pdma->csr_dma + XGENE_DMA_BLK_MEM_RDY)
1655 		!= XGENE_DMA_BLK_MEM_RDY_VAL) {
1656 		dev_err(pdma->dev,
1657 			"Failed to release DMA memory from shutdown\n");
1658 		return -ENODEV;
1659 	}
1660 
1661 	return 0;
1662 }
1663 
1664 static int xgene_dma_request_irqs(struct xgene_dma *pdma)
1665 {
1666 	struct xgene_dma_chan *chan;
1667 	int ret, i, j;
1668 
1669 	/* Register DMA error irq */
1670 	ret = devm_request_irq(pdma->dev, pdma->err_irq, xgene_dma_err_isr,
1671 			       0, "dma_error", pdma);
1672 	if (ret) {
1673 		dev_err(pdma->dev,
1674 			"Failed to register error IRQ %d\n", pdma->err_irq);
1675 		return ret;
1676 	}
1677 
1678 	/* Register DMA channel rx irq */
1679 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
1680 		chan = &pdma->chan[i];
1681 		ret = devm_request_irq(chan->dev, chan->rx_irq,
1682 				       xgene_dma_chan_ring_isr,
1683 				       0, chan->name, chan);
1684 		if (ret) {
1685 			chan_err(chan, "Failed to register Rx IRQ %d\n",
1686 				 chan->rx_irq);
1687 			devm_free_irq(pdma->dev, pdma->err_irq, pdma);
1688 
1689 			for (j = 0; j < i; j++) {
1690 				chan = &pdma->chan[i];
1691 				devm_free_irq(chan->dev, chan->rx_irq, chan);
1692 			}
1693 
1694 			return ret;
1695 		}
1696 	}
1697 
1698 	return 0;
1699 }
1700 
1701 static void xgene_dma_free_irqs(struct xgene_dma *pdma)
1702 {
1703 	struct xgene_dma_chan *chan;
1704 	int i;
1705 
1706 	/* Free DMA device error irq */
1707 	devm_free_irq(pdma->dev, pdma->err_irq, pdma);
1708 
1709 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
1710 		chan = &pdma->chan[i];
1711 		devm_free_irq(chan->dev, chan->rx_irq, chan);
1712 	}
1713 }
1714 
1715 static void xgene_dma_set_caps(struct xgene_dma_chan *chan,
1716 			       struct dma_device *dma_dev)
1717 {
1718 	/* Initialize DMA device capability mask */
1719 	dma_cap_zero(dma_dev->cap_mask);
1720 
1721 	/* Set DMA device capability */
1722 	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
1723 	dma_cap_set(DMA_SG, dma_dev->cap_mask);
1724 
1725 	/* Basically here, the X-Gene SoC DMA engine channel 0 supports XOR
1726 	 * and channel 1 supports XOR, PQ both. First thing here is we have
1727 	 * mechanism in hw to enable/disable PQ/XOR supports on channel 1,
1728 	 * we can make sure this by reading SoC Efuse register.
1729 	 * Second thing, we have hw errata that if we run channel 0 and
1730 	 * channel 1 simultaneously with executing XOR and PQ request,
1731 	 * suddenly DMA engine hangs, So here we enable XOR on channel 0 only
1732 	 * if XOR and PQ supports on channel 1 is disabled.
1733 	 */
1734 	if ((chan->id == XGENE_DMA_PQ_CHANNEL) &&
1735 	    is_pq_enabled(chan->pdma)) {
1736 		dma_cap_set(DMA_PQ, dma_dev->cap_mask);
1737 		dma_cap_set(DMA_XOR, dma_dev->cap_mask);
1738 	} else if ((chan->id == XGENE_DMA_XOR_CHANNEL) &&
1739 		   !is_pq_enabled(chan->pdma)) {
1740 		dma_cap_set(DMA_XOR, dma_dev->cap_mask);
1741 	}
1742 
1743 	/* Set base and prep routines */
1744 	dma_dev->dev = chan->dev;
1745 	dma_dev->device_alloc_chan_resources = xgene_dma_alloc_chan_resources;
1746 	dma_dev->device_free_chan_resources = xgene_dma_free_chan_resources;
1747 	dma_dev->device_issue_pending = xgene_dma_issue_pending;
1748 	dma_dev->device_tx_status = xgene_dma_tx_status;
1749 	dma_dev->device_prep_dma_memcpy = xgene_dma_prep_memcpy;
1750 	dma_dev->device_prep_dma_sg = xgene_dma_prep_sg;
1751 
1752 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1753 		dma_dev->device_prep_dma_xor = xgene_dma_prep_xor;
1754 		dma_dev->max_xor = XGENE_DMA_MAX_XOR_SRC;
1755 		dma_dev->xor_align = DMAENGINE_ALIGN_64_BYTES;
1756 	}
1757 
1758 	if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1759 		dma_dev->device_prep_dma_pq = xgene_dma_prep_pq;
1760 		dma_dev->max_pq = XGENE_DMA_MAX_XOR_SRC;
1761 		dma_dev->pq_align = DMAENGINE_ALIGN_64_BYTES;
1762 	}
1763 }
1764 
1765 static int xgene_dma_async_register(struct xgene_dma *pdma, int id)
1766 {
1767 	struct xgene_dma_chan *chan = &pdma->chan[id];
1768 	struct dma_device *dma_dev = &pdma->dma_dev[id];
1769 	int ret;
1770 
1771 	chan->dma_chan.device = dma_dev;
1772 
1773 	spin_lock_init(&chan->lock);
1774 	INIT_LIST_HEAD(&chan->ld_pending);
1775 	INIT_LIST_HEAD(&chan->ld_running);
1776 	INIT_LIST_HEAD(&chan->ld_completed);
1777 	tasklet_init(&chan->tasklet, xgene_dma_tasklet_cb,
1778 		     (unsigned long)chan);
1779 
1780 	chan->pending = 0;
1781 	chan->desc_pool = NULL;
1782 	dma_cookie_init(&chan->dma_chan);
1783 
1784 	/* Setup dma device capabilities and prep routines */
1785 	xgene_dma_set_caps(chan, dma_dev);
1786 
1787 	/* Initialize DMA device list head */
1788 	INIT_LIST_HEAD(&dma_dev->channels);
1789 	list_add_tail(&chan->dma_chan.device_node, &dma_dev->channels);
1790 
1791 	/* Register with Linux async DMA framework*/
1792 	ret = dma_async_device_register(dma_dev);
1793 	if (ret) {
1794 		chan_err(chan, "Failed to register async device %d", ret);
1795 		tasklet_kill(&chan->tasklet);
1796 
1797 		return ret;
1798 	}
1799 
1800 	/* DMA capability info */
1801 	dev_info(pdma->dev,
1802 		 "%s: CAPABILITY ( %s%s%s%s)\n", dma_chan_name(&chan->dma_chan),
1803 		 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "MEMCPY " : "",
1804 		 dma_has_cap(DMA_SG, dma_dev->cap_mask) ? "SGCPY " : "",
1805 		 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "XOR " : "",
1806 		 dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "PQ " : "");
1807 
1808 	return 0;
1809 }
1810 
1811 static int xgene_dma_init_async(struct xgene_dma *pdma)
1812 {
1813 	int ret, i, j;
1814 
1815 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL ; i++) {
1816 		ret = xgene_dma_async_register(pdma, i);
1817 		if (ret) {
1818 			for (j = 0; j < i; j++) {
1819 				dma_async_device_unregister(&pdma->dma_dev[j]);
1820 				tasklet_kill(&pdma->chan[j].tasklet);
1821 			}
1822 
1823 			return ret;
1824 		}
1825 	}
1826 
1827 	return ret;
1828 }
1829 
1830 static void xgene_dma_async_unregister(struct xgene_dma *pdma)
1831 {
1832 	int i;
1833 
1834 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++)
1835 		dma_async_device_unregister(&pdma->dma_dev[i]);
1836 }
1837 
1838 static void xgene_dma_init_channels(struct xgene_dma *pdma)
1839 {
1840 	struct xgene_dma_chan *chan;
1841 	int i;
1842 
1843 	pdma->ring_num = XGENE_DMA_RING_NUM;
1844 
1845 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
1846 		chan = &pdma->chan[i];
1847 		chan->dev = pdma->dev;
1848 		chan->pdma = pdma;
1849 		chan->id = i;
1850 		snprintf(chan->name, sizeof(chan->name), "dmachan%d", chan->id);
1851 	}
1852 }
1853 
1854 static int xgene_dma_get_resources(struct platform_device *pdev,
1855 				   struct xgene_dma *pdma)
1856 {
1857 	struct resource *res;
1858 	int irq, i;
1859 
1860 	/* Get DMA csr region */
1861 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1862 	if (!res) {
1863 		dev_err(&pdev->dev, "Failed to get csr region\n");
1864 		return -ENXIO;
1865 	}
1866 
1867 	pdma->csr_dma = devm_ioremap(&pdev->dev, res->start,
1868 				     resource_size(res));
1869 	if (!pdma->csr_dma) {
1870 		dev_err(&pdev->dev, "Failed to ioremap csr region");
1871 		return -ENOMEM;
1872 	}
1873 
1874 	/* Get DMA ring csr region */
1875 	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1876 	if (!res) {
1877 		dev_err(&pdev->dev, "Failed to get ring csr region\n");
1878 		return -ENXIO;
1879 	}
1880 
1881 	pdma->csr_ring =  devm_ioremap(&pdev->dev, res->start,
1882 				       resource_size(res));
1883 	if (!pdma->csr_ring) {
1884 		dev_err(&pdev->dev, "Failed to ioremap ring csr region");
1885 		return -ENOMEM;
1886 	}
1887 
1888 	/* Get DMA ring cmd csr region */
1889 	res = platform_get_resource(pdev, IORESOURCE_MEM, 2);
1890 	if (!res) {
1891 		dev_err(&pdev->dev, "Failed to get ring cmd csr region\n");
1892 		return -ENXIO;
1893 	}
1894 
1895 	pdma->csr_ring_cmd = devm_ioremap(&pdev->dev, res->start,
1896 					  resource_size(res));
1897 	if (!pdma->csr_ring_cmd) {
1898 		dev_err(&pdev->dev, "Failed to ioremap ring cmd csr region");
1899 		return -ENOMEM;
1900 	}
1901 
1902 	pdma->csr_ring_cmd += XGENE_DMA_RING_CMD_SM_OFFSET;
1903 
1904 	/* Get efuse csr region */
1905 	res = platform_get_resource(pdev, IORESOURCE_MEM, 3);
1906 	if (!res) {
1907 		dev_err(&pdev->dev, "Failed to get efuse csr region\n");
1908 		return -ENXIO;
1909 	}
1910 
1911 	pdma->csr_efuse = devm_ioremap(&pdev->dev, res->start,
1912 				       resource_size(res));
1913 	if (!pdma->csr_efuse) {
1914 		dev_err(&pdev->dev, "Failed to ioremap efuse csr region");
1915 		return -ENOMEM;
1916 	}
1917 
1918 	/* Get DMA error interrupt */
1919 	irq = platform_get_irq(pdev, 0);
1920 	if (irq <= 0) {
1921 		dev_err(&pdev->dev, "Failed to get Error IRQ\n");
1922 		return -ENXIO;
1923 	}
1924 
1925 	pdma->err_irq = irq;
1926 
1927 	/* Get DMA Rx ring descriptor interrupts for all DMA channels */
1928 	for (i = 1; i <= XGENE_DMA_MAX_CHANNEL; i++) {
1929 		irq = platform_get_irq(pdev, i);
1930 		if (irq <= 0) {
1931 			dev_err(&pdev->dev, "Failed to get Rx IRQ\n");
1932 			return -ENXIO;
1933 		}
1934 
1935 		pdma->chan[i - 1].rx_irq = irq;
1936 	}
1937 
1938 	return 0;
1939 }
1940 
1941 static int xgene_dma_probe(struct platform_device *pdev)
1942 {
1943 	struct xgene_dma *pdma;
1944 	int ret, i;
1945 
1946 	pdma = devm_kzalloc(&pdev->dev, sizeof(*pdma), GFP_KERNEL);
1947 	if (!pdma)
1948 		return -ENOMEM;
1949 
1950 	pdma->dev = &pdev->dev;
1951 	platform_set_drvdata(pdev, pdma);
1952 
1953 	ret = xgene_dma_get_resources(pdev, pdma);
1954 	if (ret)
1955 		return ret;
1956 
1957 	pdma->clk = devm_clk_get(&pdev->dev, NULL);
1958 	if (IS_ERR(pdma->clk) && !ACPI_COMPANION(&pdev->dev)) {
1959 		dev_err(&pdev->dev, "Failed to get clk\n");
1960 		return PTR_ERR(pdma->clk);
1961 	}
1962 
1963 	/* Enable clk before accessing registers */
1964 	if (!IS_ERR(pdma->clk)) {
1965 		ret = clk_prepare_enable(pdma->clk);
1966 		if (ret) {
1967 			dev_err(&pdev->dev, "Failed to enable clk %d\n", ret);
1968 			return ret;
1969 		}
1970 	}
1971 
1972 	/* Remove DMA RAM out of shutdown */
1973 	ret = xgene_dma_init_mem(pdma);
1974 	if (ret)
1975 		goto err_clk_enable;
1976 
1977 	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(42));
1978 	if (ret) {
1979 		dev_err(&pdev->dev, "No usable DMA configuration\n");
1980 		goto err_dma_mask;
1981 	}
1982 
1983 	/* Initialize DMA channels software state */
1984 	xgene_dma_init_channels(pdma);
1985 
1986 	/* Configue DMA rings */
1987 	ret = xgene_dma_init_rings(pdma);
1988 	if (ret)
1989 		goto err_clk_enable;
1990 
1991 	ret = xgene_dma_request_irqs(pdma);
1992 	if (ret)
1993 		goto err_request_irq;
1994 
1995 	/* Configure and enable DMA engine */
1996 	xgene_dma_init_hw(pdma);
1997 
1998 	/* Register DMA device with linux async framework */
1999 	ret = xgene_dma_init_async(pdma);
2000 	if (ret)
2001 		goto err_async_init;
2002 
2003 	return 0;
2004 
2005 err_async_init:
2006 	xgene_dma_free_irqs(pdma);
2007 
2008 err_request_irq:
2009 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++)
2010 		xgene_dma_delete_chan_rings(&pdma->chan[i]);
2011 
2012 err_dma_mask:
2013 err_clk_enable:
2014 	if (!IS_ERR(pdma->clk))
2015 		clk_disable_unprepare(pdma->clk);
2016 
2017 	return ret;
2018 }
2019 
2020 static int xgene_dma_remove(struct platform_device *pdev)
2021 {
2022 	struct xgene_dma *pdma = platform_get_drvdata(pdev);
2023 	struct xgene_dma_chan *chan;
2024 	int i;
2025 
2026 	xgene_dma_async_unregister(pdma);
2027 
2028 	/* Mask interrupts and disable DMA engine */
2029 	xgene_dma_mask_interrupts(pdma);
2030 	xgene_dma_disable(pdma);
2031 	xgene_dma_free_irqs(pdma);
2032 
2033 	for (i = 0; i < XGENE_DMA_MAX_CHANNEL; i++) {
2034 		chan = &pdma->chan[i];
2035 		tasklet_kill(&chan->tasklet);
2036 		xgene_dma_delete_chan_rings(chan);
2037 	}
2038 
2039 	if (!IS_ERR(pdma->clk))
2040 		clk_disable_unprepare(pdma->clk);
2041 
2042 	return 0;
2043 }
2044 
2045 #ifdef CONFIG_ACPI
2046 static const struct acpi_device_id xgene_dma_acpi_match_ptr[] = {
2047 	{"APMC0D43", 0},
2048 	{},
2049 };
2050 MODULE_DEVICE_TABLE(acpi, xgene_dma_acpi_match_ptr);
2051 #endif
2052 
2053 static const struct of_device_id xgene_dma_of_match_ptr[] = {
2054 	{.compatible = "apm,xgene-storm-dma",},
2055 	{},
2056 };
2057 MODULE_DEVICE_TABLE(of, xgene_dma_of_match_ptr);
2058 
2059 static struct platform_driver xgene_dma_driver = {
2060 	.probe = xgene_dma_probe,
2061 	.remove = xgene_dma_remove,
2062 	.driver = {
2063 		.name = "X-Gene-DMA",
2064 		.of_match_table = xgene_dma_of_match_ptr,
2065 		.acpi_match_table = ACPI_PTR(xgene_dma_acpi_match_ptr),
2066 	},
2067 };
2068 
2069 module_platform_driver(xgene_dma_driver);
2070 
2071 MODULE_DESCRIPTION("APM X-Gene SoC DMA driver");
2072 MODULE_AUTHOR("Rameshwar Prasad Sahu <rsahu@apm.com>");
2073 MODULE_AUTHOR("Loc Ho <lho@apm.com>");
2074 MODULE_LICENSE("GPL");
2075 MODULE_VERSION("1.0");
2076