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