xref: /linux/drivers/dma/pl330.c (revision b7019ac550eb3916f34d79db583e9b7ea2524afa)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
4  *		http://www.samsung.com
5  *
6  * Copyright (C) 2010 Samsung Electronics Co. Ltd.
7  *	Jaswinder Singh <jassi.brar@samsung.com>
8  */
9 
10 #include <linux/debugfs.h>
11 #include <linux/kernel.h>
12 #include <linux/io.h>
13 #include <linux/init.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16 #include <linux/string.h>
17 #include <linux/delay.h>
18 #include <linux/interrupt.h>
19 #include <linux/dma-mapping.h>
20 #include <linux/dmaengine.h>
21 #include <linux/amba/bus.h>
22 #include <linux/scatterlist.h>
23 #include <linux/of.h>
24 #include <linux/of_dma.h>
25 #include <linux/err.h>
26 #include <linux/pm_runtime.h>
27 #include <linux/bug.h>
28 
29 #include "dmaengine.h"
30 #define PL330_MAX_CHAN		8
31 #define PL330_MAX_IRQS		32
32 #define PL330_MAX_PERI		32
33 #define PL330_MAX_BURST         16
34 
35 #define PL330_QUIRK_BROKEN_NO_FLUSHP BIT(0)
36 
37 enum pl330_cachectrl {
38 	CCTRL0,		/* Noncacheable and nonbufferable */
39 	CCTRL1,		/* Bufferable only */
40 	CCTRL2,		/* Cacheable, but do not allocate */
41 	CCTRL3,		/* Cacheable and bufferable, but do not allocate */
42 	INVALID1,	/* AWCACHE = 0x1000 */
43 	INVALID2,
44 	CCTRL6,		/* Cacheable write-through, allocate on writes only */
45 	CCTRL7,		/* Cacheable write-back, allocate on writes only */
46 };
47 
48 enum pl330_byteswap {
49 	SWAP_NO,
50 	SWAP_2,
51 	SWAP_4,
52 	SWAP_8,
53 	SWAP_16,
54 };
55 
56 /* Register and Bit field Definitions */
57 #define DS			0x0
58 #define DS_ST_STOP		0x0
59 #define DS_ST_EXEC		0x1
60 #define DS_ST_CMISS		0x2
61 #define DS_ST_UPDTPC		0x3
62 #define DS_ST_WFE		0x4
63 #define DS_ST_ATBRR		0x5
64 #define DS_ST_QBUSY		0x6
65 #define DS_ST_WFP		0x7
66 #define DS_ST_KILL		0x8
67 #define DS_ST_CMPLT		0x9
68 #define DS_ST_FLTCMP		0xe
69 #define DS_ST_FAULT		0xf
70 
71 #define DPC			0x4
72 #define INTEN			0x20
73 #define ES			0x24
74 #define INTSTATUS		0x28
75 #define INTCLR			0x2c
76 #define FSM			0x30
77 #define FSC			0x34
78 #define FTM			0x38
79 
80 #define _FTC			0x40
81 #define FTC(n)			(_FTC + (n)*0x4)
82 
83 #define _CS			0x100
84 #define CS(n)			(_CS + (n)*0x8)
85 #define CS_CNS			(1 << 21)
86 
87 #define _CPC			0x104
88 #define CPC(n)			(_CPC + (n)*0x8)
89 
90 #define _SA			0x400
91 #define SA(n)			(_SA + (n)*0x20)
92 
93 #define _DA			0x404
94 #define DA(n)			(_DA + (n)*0x20)
95 
96 #define _CC			0x408
97 #define CC(n)			(_CC + (n)*0x20)
98 
99 #define CC_SRCINC		(1 << 0)
100 #define CC_DSTINC		(1 << 14)
101 #define CC_SRCPRI		(1 << 8)
102 #define CC_DSTPRI		(1 << 22)
103 #define CC_SRCNS		(1 << 9)
104 #define CC_DSTNS		(1 << 23)
105 #define CC_SRCIA		(1 << 10)
106 #define CC_DSTIA		(1 << 24)
107 #define CC_SRCBRSTLEN_SHFT	4
108 #define CC_DSTBRSTLEN_SHFT	18
109 #define CC_SRCBRSTSIZE_SHFT	1
110 #define CC_DSTBRSTSIZE_SHFT	15
111 #define CC_SRCCCTRL_SHFT	11
112 #define CC_SRCCCTRL_MASK	0x7
113 #define CC_DSTCCTRL_SHFT	25
114 #define CC_DRCCCTRL_MASK	0x7
115 #define CC_SWAP_SHFT		28
116 
117 #define _LC0			0x40c
118 #define LC0(n)			(_LC0 + (n)*0x20)
119 
120 #define _LC1			0x410
121 #define LC1(n)			(_LC1 + (n)*0x20)
122 
123 #define DBGSTATUS		0xd00
124 #define DBG_BUSY		(1 << 0)
125 
126 #define DBGCMD			0xd04
127 #define DBGINST0		0xd08
128 #define DBGINST1		0xd0c
129 
130 #define CR0			0xe00
131 #define CR1			0xe04
132 #define CR2			0xe08
133 #define CR3			0xe0c
134 #define CR4			0xe10
135 #define CRD			0xe14
136 
137 #define PERIPH_ID		0xfe0
138 #define PERIPH_REV_SHIFT	20
139 #define PERIPH_REV_MASK		0xf
140 #define PERIPH_REV_R0P0		0
141 #define PERIPH_REV_R1P0		1
142 #define PERIPH_REV_R1P1		2
143 
144 #define CR0_PERIPH_REQ_SET	(1 << 0)
145 #define CR0_BOOT_EN_SET		(1 << 1)
146 #define CR0_BOOT_MAN_NS		(1 << 2)
147 #define CR0_NUM_CHANS_SHIFT	4
148 #define CR0_NUM_CHANS_MASK	0x7
149 #define CR0_NUM_PERIPH_SHIFT	12
150 #define CR0_NUM_PERIPH_MASK	0x1f
151 #define CR0_NUM_EVENTS_SHIFT	17
152 #define CR0_NUM_EVENTS_MASK	0x1f
153 
154 #define CR1_ICACHE_LEN_SHIFT	0
155 #define CR1_ICACHE_LEN_MASK	0x7
156 #define CR1_NUM_ICACHELINES_SHIFT	4
157 #define CR1_NUM_ICACHELINES_MASK	0xf
158 
159 #define CRD_DATA_WIDTH_SHIFT	0
160 #define CRD_DATA_WIDTH_MASK	0x7
161 #define CRD_WR_CAP_SHIFT	4
162 #define CRD_WR_CAP_MASK		0x7
163 #define CRD_WR_Q_DEP_SHIFT	8
164 #define CRD_WR_Q_DEP_MASK	0xf
165 #define CRD_RD_CAP_SHIFT	12
166 #define CRD_RD_CAP_MASK		0x7
167 #define CRD_RD_Q_DEP_SHIFT	16
168 #define CRD_RD_Q_DEP_MASK	0xf
169 #define CRD_DATA_BUFF_SHIFT	20
170 #define CRD_DATA_BUFF_MASK	0x3ff
171 
172 #define PART			0x330
173 #define DESIGNER		0x41
174 #define REVISION		0x0
175 #define INTEG_CFG		0x0
176 #define PERIPH_ID_VAL		((PART << 0) | (DESIGNER << 12))
177 
178 #define PL330_STATE_STOPPED		(1 << 0)
179 #define PL330_STATE_EXECUTING		(1 << 1)
180 #define PL330_STATE_WFE			(1 << 2)
181 #define PL330_STATE_FAULTING		(1 << 3)
182 #define PL330_STATE_COMPLETING		(1 << 4)
183 #define PL330_STATE_WFP			(1 << 5)
184 #define PL330_STATE_KILLING		(1 << 6)
185 #define PL330_STATE_FAULT_COMPLETING	(1 << 7)
186 #define PL330_STATE_CACHEMISS		(1 << 8)
187 #define PL330_STATE_UPDTPC		(1 << 9)
188 #define PL330_STATE_ATBARRIER		(1 << 10)
189 #define PL330_STATE_QUEUEBUSY		(1 << 11)
190 #define PL330_STATE_INVALID		(1 << 15)
191 
192 #define PL330_STABLE_STATES (PL330_STATE_STOPPED | PL330_STATE_EXECUTING \
193 				| PL330_STATE_WFE | PL330_STATE_FAULTING)
194 
195 #define CMD_DMAADDH		0x54
196 #define CMD_DMAEND		0x00
197 #define CMD_DMAFLUSHP		0x35
198 #define CMD_DMAGO		0xa0
199 #define CMD_DMALD		0x04
200 #define CMD_DMALDP		0x25
201 #define CMD_DMALP		0x20
202 #define CMD_DMALPEND		0x28
203 #define CMD_DMAKILL		0x01
204 #define CMD_DMAMOV		0xbc
205 #define CMD_DMANOP		0x18
206 #define CMD_DMARMB		0x12
207 #define CMD_DMASEV		0x34
208 #define CMD_DMAST		0x08
209 #define CMD_DMASTP		0x29
210 #define CMD_DMASTZ		0x0c
211 #define CMD_DMAWFE		0x36
212 #define CMD_DMAWFP		0x30
213 #define CMD_DMAWMB		0x13
214 
215 #define SZ_DMAADDH		3
216 #define SZ_DMAEND		1
217 #define SZ_DMAFLUSHP		2
218 #define SZ_DMALD		1
219 #define SZ_DMALDP		2
220 #define SZ_DMALP		2
221 #define SZ_DMALPEND		2
222 #define SZ_DMAKILL		1
223 #define SZ_DMAMOV		6
224 #define SZ_DMANOP		1
225 #define SZ_DMARMB		1
226 #define SZ_DMASEV		2
227 #define SZ_DMAST		1
228 #define SZ_DMASTP		2
229 #define SZ_DMASTZ		1
230 #define SZ_DMAWFE		2
231 #define SZ_DMAWFP		2
232 #define SZ_DMAWMB		1
233 #define SZ_DMAGO		6
234 
235 #define BRST_LEN(ccr)		((((ccr) >> CC_SRCBRSTLEN_SHFT) & 0xf) + 1)
236 #define BRST_SIZE(ccr)		(1 << (((ccr) >> CC_SRCBRSTSIZE_SHFT) & 0x7))
237 
238 #define BYTE_TO_BURST(b, ccr)	((b) / BRST_SIZE(ccr) / BRST_LEN(ccr))
239 #define BURST_TO_BYTE(c, ccr)	((c) * BRST_SIZE(ccr) * BRST_LEN(ccr))
240 
241 /*
242  * With 256 bytes, we can do more than 2.5MB and 5MB xfers per req
243  * at 1byte/burst for P<->M and M<->M respectively.
244  * For typical scenario, at 1word/burst, 10MB and 20MB xfers per req
245  * should be enough for P<->M and M<->M respectively.
246  */
247 #define MCODE_BUFF_PER_REQ	256
248 
249 /* Use this _only_ to wait on transient states */
250 #define UNTIL(t, s)	while (!(_state(t) & (s))) cpu_relax();
251 
252 #ifdef PL330_DEBUG_MCGEN
253 static unsigned cmd_line;
254 #define PL330_DBGCMD_DUMP(off, x...)	do { \
255 						printk("%x:", cmd_line); \
256 						printk(x); \
257 						cmd_line += off; \
258 					} while (0)
259 #define PL330_DBGMC_START(addr)		(cmd_line = addr)
260 #else
261 #define PL330_DBGCMD_DUMP(off, x...)	do {} while (0)
262 #define PL330_DBGMC_START(addr)		do {} while (0)
263 #endif
264 
265 /* The number of default descriptors */
266 
267 #define NR_DEFAULT_DESC	16
268 
269 /* Delay for runtime PM autosuspend, ms */
270 #define PL330_AUTOSUSPEND_DELAY 20
271 
272 /* Populated by the PL330 core driver for DMA API driver's info */
273 struct pl330_config {
274 	u32	periph_id;
275 #define DMAC_MODE_NS	(1 << 0)
276 	unsigned int	mode;
277 	unsigned int	data_bus_width:10; /* In number of bits */
278 	unsigned int	data_buf_dep:11;
279 	unsigned int	num_chan:4;
280 	unsigned int	num_peri:6;
281 	u32		peri_ns;
282 	unsigned int	num_events:6;
283 	u32		irq_ns;
284 };
285 
286 /**
287  * Request Configuration.
288  * The PL330 core does not modify this and uses the last
289  * working configuration if the request doesn't provide any.
290  *
291  * The Client may want to provide this info only for the
292  * first request and a request with new settings.
293  */
294 struct pl330_reqcfg {
295 	/* Address Incrementing */
296 	unsigned dst_inc:1;
297 	unsigned src_inc:1;
298 
299 	/*
300 	 * For now, the SRC & DST protection levels
301 	 * and burst size/length are assumed same.
302 	 */
303 	bool nonsecure;
304 	bool privileged;
305 	bool insnaccess;
306 	unsigned brst_len:5;
307 	unsigned brst_size:3; /* in power of 2 */
308 
309 	enum pl330_cachectrl dcctl;
310 	enum pl330_cachectrl scctl;
311 	enum pl330_byteswap swap;
312 	struct pl330_config *pcfg;
313 };
314 
315 /*
316  * One cycle of DMAC operation.
317  * There may be more than one xfer in a request.
318  */
319 struct pl330_xfer {
320 	u32 src_addr;
321 	u32 dst_addr;
322 	/* Size to xfer */
323 	u32 bytes;
324 };
325 
326 /* The xfer callbacks are made with one of these arguments. */
327 enum pl330_op_err {
328 	/* The all xfers in the request were success. */
329 	PL330_ERR_NONE,
330 	/* If req aborted due to global error. */
331 	PL330_ERR_ABORT,
332 	/* If req failed due to problem with Channel. */
333 	PL330_ERR_FAIL,
334 };
335 
336 enum dmamov_dst {
337 	SAR = 0,
338 	CCR,
339 	DAR,
340 };
341 
342 enum pl330_dst {
343 	SRC = 0,
344 	DST,
345 };
346 
347 enum pl330_cond {
348 	SINGLE,
349 	BURST,
350 	ALWAYS,
351 };
352 
353 struct dma_pl330_desc;
354 
355 struct _pl330_req {
356 	u32 mc_bus;
357 	void *mc_cpu;
358 	struct dma_pl330_desc *desc;
359 };
360 
361 /* ToBeDone for tasklet */
362 struct _pl330_tbd {
363 	bool reset_dmac;
364 	bool reset_mngr;
365 	u8 reset_chan;
366 };
367 
368 /* A DMAC Thread */
369 struct pl330_thread {
370 	u8 id;
371 	int ev;
372 	/* If the channel is not yet acquired by any client */
373 	bool free;
374 	/* Parent DMAC */
375 	struct pl330_dmac *dmac;
376 	/* Only two at a time */
377 	struct _pl330_req req[2];
378 	/* Index of the last enqueued request */
379 	unsigned lstenq;
380 	/* Index of the last submitted request or -1 if the DMA is stopped */
381 	int req_running;
382 };
383 
384 enum pl330_dmac_state {
385 	UNINIT,
386 	INIT,
387 	DYING,
388 };
389 
390 enum desc_status {
391 	/* In the DMAC pool */
392 	FREE,
393 	/*
394 	 * Allocated to some channel during prep_xxx
395 	 * Also may be sitting on the work_list.
396 	 */
397 	PREP,
398 	/*
399 	 * Sitting on the work_list and already submitted
400 	 * to the PL330 core. Not more than two descriptors
401 	 * of a channel can be BUSY at any time.
402 	 */
403 	BUSY,
404 	/*
405 	 * Sitting on the channel work_list but xfer done
406 	 * by PL330 core
407 	 */
408 	DONE,
409 };
410 
411 struct dma_pl330_chan {
412 	/* Schedule desc completion */
413 	struct tasklet_struct task;
414 
415 	/* DMA-Engine Channel */
416 	struct dma_chan chan;
417 
418 	/* List of submitted descriptors */
419 	struct list_head submitted_list;
420 	/* List of issued descriptors */
421 	struct list_head work_list;
422 	/* List of completed descriptors */
423 	struct list_head completed_list;
424 
425 	/* Pointer to the DMAC that manages this channel,
426 	 * NULL if the channel is available to be acquired.
427 	 * As the parent, this DMAC also provides descriptors
428 	 * to the channel.
429 	 */
430 	struct pl330_dmac *dmac;
431 
432 	/* To protect channel manipulation */
433 	spinlock_t lock;
434 
435 	/*
436 	 * Hardware channel thread of PL330 DMAC. NULL if the channel is
437 	 * available.
438 	 */
439 	struct pl330_thread *thread;
440 
441 	/* For D-to-M and M-to-D channels */
442 	int burst_sz; /* the peripheral fifo width */
443 	int burst_len; /* the number of burst */
444 	phys_addr_t fifo_addr;
445 	/* DMA-mapped view of the FIFO; may differ if an IOMMU is present */
446 	dma_addr_t fifo_dma;
447 	enum dma_data_direction dir;
448 	struct dma_slave_config slave_config;
449 
450 	/* for cyclic capability */
451 	bool cyclic;
452 
453 	/* for runtime pm tracking */
454 	bool active;
455 };
456 
457 struct pl330_dmac {
458 	/* DMA-Engine Device */
459 	struct dma_device ddma;
460 
461 	/* Holds info about sg limitations */
462 	struct device_dma_parameters dma_parms;
463 
464 	/* Pool of descriptors available for the DMAC's channels */
465 	struct list_head desc_pool;
466 	/* To protect desc_pool manipulation */
467 	spinlock_t pool_lock;
468 
469 	/* Size of MicroCode buffers for each channel. */
470 	unsigned mcbufsz;
471 	/* ioremap'ed address of PL330 registers. */
472 	void __iomem	*base;
473 	/* Populated by the PL330 core driver during pl330_add */
474 	struct pl330_config	pcfg;
475 
476 	spinlock_t		lock;
477 	/* Maximum possible events/irqs */
478 	int			events[32];
479 	/* BUS address of MicroCode buffer */
480 	dma_addr_t		mcode_bus;
481 	/* CPU address of MicroCode buffer */
482 	void			*mcode_cpu;
483 	/* List of all Channel threads */
484 	struct pl330_thread	*channels;
485 	/* Pointer to the MANAGER thread */
486 	struct pl330_thread	*manager;
487 	/* To handle bad news in interrupt */
488 	struct tasklet_struct	tasks;
489 	struct _pl330_tbd	dmac_tbd;
490 	/* State of DMAC operation */
491 	enum pl330_dmac_state	state;
492 	/* Holds list of reqs with due callbacks */
493 	struct list_head        req_done;
494 
495 	/* Peripheral channels connected to this DMAC */
496 	unsigned int num_peripherals;
497 	struct dma_pl330_chan *peripherals; /* keep at end */
498 	int quirks;
499 };
500 
501 static struct pl330_of_quirks {
502 	char *quirk;
503 	int id;
504 } of_quirks[] = {
505 	{
506 		.quirk = "arm,pl330-broken-no-flushp",
507 		.id = PL330_QUIRK_BROKEN_NO_FLUSHP,
508 	}
509 };
510 
511 struct dma_pl330_desc {
512 	/* To attach to a queue as child */
513 	struct list_head node;
514 
515 	/* Descriptor for the DMA Engine API */
516 	struct dma_async_tx_descriptor txd;
517 
518 	/* Xfer for PL330 core */
519 	struct pl330_xfer px;
520 
521 	struct pl330_reqcfg rqcfg;
522 
523 	enum desc_status status;
524 
525 	int bytes_requested;
526 	bool last;
527 
528 	/* The channel which currently holds this desc */
529 	struct dma_pl330_chan *pchan;
530 
531 	enum dma_transfer_direction rqtype;
532 	/* Index of peripheral for the xfer. */
533 	unsigned peri:5;
534 	/* Hook to attach to DMAC's list of reqs with due callback */
535 	struct list_head rqd;
536 };
537 
538 struct _xfer_spec {
539 	u32 ccr;
540 	struct dma_pl330_desc *desc;
541 };
542 
543 static int pl330_config_write(struct dma_chan *chan,
544 			struct dma_slave_config *slave_config,
545 			enum dma_transfer_direction direction);
546 
547 static inline bool _queue_full(struct pl330_thread *thrd)
548 {
549 	return thrd->req[0].desc != NULL && thrd->req[1].desc != NULL;
550 }
551 
552 static inline bool is_manager(struct pl330_thread *thrd)
553 {
554 	return thrd->dmac->manager == thrd;
555 }
556 
557 /* If manager of the thread is in Non-Secure mode */
558 static inline bool _manager_ns(struct pl330_thread *thrd)
559 {
560 	return (thrd->dmac->pcfg.mode & DMAC_MODE_NS) ? true : false;
561 }
562 
563 static inline u32 get_revision(u32 periph_id)
564 {
565 	return (periph_id >> PERIPH_REV_SHIFT) & PERIPH_REV_MASK;
566 }
567 
568 static inline u32 _emit_END(unsigned dry_run, u8 buf[])
569 {
570 	if (dry_run)
571 		return SZ_DMAEND;
572 
573 	buf[0] = CMD_DMAEND;
574 
575 	PL330_DBGCMD_DUMP(SZ_DMAEND, "\tDMAEND\n");
576 
577 	return SZ_DMAEND;
578 }
579 
580 static inline u32 _emit_FLUSHP(unsigned dry_run, u8 buf[], u8 peri)
581 {
582 	if (dry_run)
583 		return SZ_DMAFLUSHP;
584 
585 	buf[0] = CMD_DMAFLUSHP;
586 
587 	peri &= 0x1f;
588 	peri <<= 3;
589 	buf[1] = peri;
590 
591 	PL330_DBGCMD_DUMP(SZ_DMAFLUSHP, "\tDMAFLUSHP %u\n", peri >> 3);
592 
593 	return SZ_DMAFLUSHP;
594 }
595 
596 static inline u32 _emit_LD(unsigned dry_run, u8 buf[],	enum pl330_cond cond)
597 {
598 	if (dry_run)
599 		return SZ_DMALD;
600 
601 	buf[0] = CMD_DMALD;
602 
603 	if (cond == SINGLE)
604 		buf[0] |= (0 << 1) | (1 << 0);
605 	else if (cond == BURST)
606 		buf[0] |= (1 << 1) | (1 << 0);
607 
608 	PL330_DBGCMD_DUMP(SZ_DMALD, "\tDMALD%c\n",
609 		cond == SINGLE ? 'S' : (cond == BURST ? 'B' : 'A'));
610 
611 	return SZ_DMALD;
612 }
613 
614 static inline u32 _emit_LDP(unsigned dry_run, u8 buf[],
615 		enum pl330_cond cond, u8 peri)
616 {
617 	if (dry_run)
618 		return SZ_DMALDP;
619 
620 	buf[0] = CMD_DMALDP;
621 
622 	if (cond == BURST)
623 		buf[0] |= (1 << 1);
624 
625 	peri &= 0x1f;
626 	peri <<= 3;
627 	buf[1] = peri;
628 
629 	PL330_DBGCMD_DUMP(SZ_DMALDP, "\tDMALDP%c %u\n",
630 		cond == SINGLE ? 'S' : 'B', peri >> 3);
631 
632 	return SZ_DMALDP;
633 }
634 
635 static inline u32 _emit_LP(unsigned dry_run, u8 buf[],
636 		unsigned loop, u8 cnt)
637 {
638 	if (dry_run)
639 		return SZ_DMALP;
640 
641 	buf[0] = CMD_DMALP;
642 
643 	if (loop)
644 		buf[0] |= (1 << 1);
645 
646 	cnt--; /* DMAC increments by 1 internally */
647 	buf[1] = cnt;
648 
649 	PL330_DBGCMD_DUMP(SZ_DMALP, "\tDMALP_%c %u\n", loop ? '1' : '0', cnt);
650 
651 	return SZ_DMALP;
652 }
653 
654 struct _arg_LPEND {
655 	enum pl330_cond cond;
656 	bool forever;
657 	unsigned loop;
658 	u8 bjump;
659 };
660 
661 static inline u32 _emit_LPEND(unsigned dry_run, u8 buf[],
662 		const struct _arg_LPEND *arg)
663 {
664 	enum pl330_cond cond = arg->cond;
665 	bool forever = arg->forever;
666 	unsigned loop = arg->loop;
667 	u8 bjump = arg->bjump;
668 
669 	if (dry_run)
670 		return SZ_DMALPEND;
671 
672 	buf[0] = CMD_DMALPEND;
673 
674 	if (loop)
675 		buf[0] |= (1 << 2);
676 
677 	if (!forever)
678 		buf[0] |= (1 << 4);
679 
680 	if (cond == SINGLE)
681 		buf[0] |= (0 << 1) | (1 << 0);
682 	else if (cond == BURST)
683 		buf[0] |= (1 << 1) | (1 << 0);
684 
685 	buf[1] = bjump;
686 
687 	PL330_DBGCMD_DUMP(SZ_DMALPEND, "\tDMALP%s%c_%c bjmpto_%x\n",
688 			forever ? "FE" : "END",
689 			cond == SINGLE ? 'S' : (cond == BURST ? 'B' : 'A'),
690 			loop ? '1' : '0',
691 			bjump);
692 
693 	return SZ_DMALPEND;
694 }
695 
696 static inline u32 _emit_KILL(unsigned dry_run, u8 buf[])
697 {
698 	if (dry_run)
699 		return SZ_DMAKILL;
700 
701 	buf[0] = CMD_DMAKILL;
702 
703 	return SZ_DMAKILL;
704 }
705 
706 static inline u32 _emit_MOV(unsigned dry_run, u8 buf[],
707 		enum dmamov_dst dst, u32 val)
708 {
709 	if (dry_run)
710 		return SZ_DMAMOV;
711 
712 	buf[0] = CMD_DMAMOV;
713 	buf[1] = dst;
714 	buf[2] = val;
715 	buf[3] = val >> 8;
716 	buf[4] = val >> 16;
717 	buf[5] = val >> 24;
718 
719 	PL330_DBGCMD_DUMP(SZ_DMAMOV, "\tDMAMOV %s 0x%x\n",
720 		dst == SAR ? "SAR" : (dst == DAR ? "DAR" : "CCR"), val);
721 
722 	return SZ_DMAMOV;
723 }
724 
725 static inline u32 _emit_RMB(unsigned dry_run, u8 buf[])
726 {
727 	if (dry_run)
728 		return SZ_DMARMB;
729 
730 	buf[0] = CMD_DMARMB;
731 
732 	PL330_DBGCMD_DUMP(SZ_DMARMB, "\tDMARMB\n");
733 
734 	return SZ_DMARMB;
735 }
736 
737 static inline u32 _emit_SEV(unsigned dry_run, u8 buf[], u8 ev)
738 {
739 	if (dry_run)
740 		return SZ_DMASEV;
741 
742 	buf[0] = CMD_DMASEV;
743 
744 	ev &= 0x1f;
745 	ev <<= 3;
746 	buf[1] = ev;
747 
748 	PL330_DBGCMD_DUMP(SZ_DMASEV, "\tDMASEV %u\n", ev >> 3);
749 
750 	return SZ_DMASEV;
751 }
752 
753 static inline u32 _emit_ST(unsigned dry_run, u8 buf[], enum pl330_cond cond)
754 {
755 	if (dry_run)
756 		return SZ_DMAST;
757 
758 	buf[0] = CMD_DMAST;
759 
760 	if (cond == SINGLE)
761 		buf[0] |= (0 << 1) | (1 << 0);
762 	else if (cond == BURST)
763 		buf[0] |= (1 << 1) | (1 << 0);
764 
765 	PL330_DBGCMD_DUMP(SZ_DMAST, "\tDMAST%c\n",
766 		cond == SINGLE ? 'S' : (cond == BURST ? 'B' : 'A'));
767 
768 	return SZ_DMAST;
769 }
770 
771 static inline u32 _emit_STP(unsigned dry_run, u8 buf[],
772 		enum pl330_cond cond, u8 peri)
773 {
774 	if (dry_run)
775 		return SZ_DMASTP;
776 
777 	buf[0] = CMD_DMASTP;
778 
779 	if (cond == BURST)
780 		buf[0] |= (1 << 1);
781 
782 	peri &= 0x1f;
783 	peri <<= 3;
784 	buf[1] = peri;
785 
786 	PL330_DBGCMD_DUMP(SZ_DMASTP, "\tDMASTP%c %u\n",
787 		cond == SINGLE ? 'S' : 'B', peri >> 3);
788 
789 	return SZ_DMASTP;
790 }
791 
792 static inline u32 _emit_WFP(unsigned dry_run, u8 buf[],
793 		enum pl330_cond cond, u8 peri)
794 {
795 	if (dry_run)
796 		return SZ_DMAWFP;
797 
798 	buf[0] = CMD_DMAWFP;
799 
800 	if (cond == SINGLE)
801 		buf[0] |= (0 << 1) | (0 << 0);
802 	else if (cond == BURST)
803 		buf[0] |= (1 << 1) | (0 << 0);
804 	else
805 		buf[0] |= (0 << 1) | (1 << 0);
806 
807 	peri &= 0x1f;
808 	peri <<= 3;
809 	buf[1] = peri;
810 
811 	PL330_DBGCMD_DUMP(SZ_DMAWFP, "\tDMAWFP%c %u\n",
812 		cond == SINGLE ? 'S' : (cond == BURST ? 'B' : 'P'), peri >> 3);
813 
814 	return SZ_DMAWFP;
815 }
816 
817 static inline u32 _emit_WMB(unsigned dry_run, u8 buf[])
818 {
819 	if (dry_run)
820 		return SZ_DMAWMB;
821 
822 	buf[0] = CMD_DMAWMB;
823 
824 	PL330_DBGCMD_DUMP(SZ_DMAWMB, "\tDMAWMB\n");
825 
826 	return SZ_DMAWMB;
827 }
828 
829 struct _arg_GO {
830 	u8 chan;
831 	u32 addr;
832 	unsigned ns;
833 };
834 
835 static inline u32 _emit_GO(unsigned dry_run, u8 buf[],
836 		const struct _arg_GO *arg)
837 {
838 	u8 chan = arg->chan;
839 	u32 addr = arg->addr;
840 	unsigned ns = arg->ns;
841 
842 	if (dry_run)
843 		return SZ_DMAGO;
844 
845 	buf[0] = CMD_DMAGO;
846 	buf[0] |= (ns << 1);
847 	buf[1] = chan & 0x7;
848 	buf[2] = addr;
849 	buf[3] = addr >> 8;
850 	buf[4] = addr >> 16;
851 	buf[5] = addr >> 24;
852 
853 	return SZ_DMAGO;
854 }
855 
856 #define msecs_to_loops(t) (loops_per_jiffy / 1000 * HZ * t)
857 
858 /* Returns Time-Out */
859 static bool _until_dmac_idle(struct pl330_thread *thrd)
860 {
861 	void __iomem *regs = thrd->dmac->base;
862 	unsigned long loops = msecs_to_loops(5);
863 
864 	do {
865 		/* Until Manager is Idle */
866 		if (!(readl(regs + DBGSTATUS) & DBG_BUSY))
867 			break;
868 
869 		cpu_relax();
870 	} while (--loops);
871 
872 	if (!loops)
873 		return true;
874 
875 	return false;
876 }
877 
878 static inline void _execute_DBGINSN(struct pl330_thread *thrd,
879 		u8 insn[], bool as_manager)
880 {
881 	void __iomem *regs = thrd->dmac->base;
882 	u32 val;
883 
884 	val = (insn[0] << 16) | (insn[1] << 24);
885 	if (!as_manager) {
886 		val |= (1 << 0);
887 		val |= (thrd->id << 8); /* Channel Number */
888 	}
889 	writel(val, regs + DBGINST0);
890 
891 	val = le32_to_cpu(*((__le32 *)&insn[2]));
892 	writel(val, regs + DBGINST1);
893 
894 	/* If timed out due to halted state-machine */
895 	if (_until_dmac_idle(thrd)) {
896 		dev_err(thrd->dmac->ddma.dev, "DMAC halted!\n");
897 		return;
898 	}
899 
900 	/* Get going */
901 	writel(0, regs + DBGCMD);
902 }
903 
904 static inline u32 _state(struct pl330_thread *thrd)
905 {
906 	void __iomem *regs = thrd->dmac->base;
907 	u32 val;
908 
909 	if (is_manager(thrd))
910 		val = readl(regs + DS) & 0xf;
911 	else
912 		val = readl(regs + CS(thrd->id)) & 0xf;
913 
914 	switch (val) {
915 	case DS_ST_STOP:
916 		return PL330_STATE_STOPPED;
917 	case DS_ST_EXEC:
918 		return PL330_STATE_EXECUTING;
919 	case DS_ST_CMISS:
920 		return PL330_STATE_CACHEMISS;
921 	case DS_ST_UPDTPC:
922 		return PL330_STATE_UPDTPC;
923 	case DS_ST_WFE:
924 		return PL330_STATE_WFE;
925 	case DS_ST_FAULT:
926 		return PL330_STATE_FAULTING;
927 	case DS_ST_ATBRR:
928 		if (is_manager(thrd))
929 			return PL330_STATE_INVALID;
930 		else
931 			return PL330_STATE_ATBARRIER;
932 	case DS_ST_QBUSY:
933 		if (is_manager(thrd))
934 			return PL330_STATE_INVALID;
935 		else
936 			return PL330_STATE_QUEUEBUSY;
937 	case DS_ST_WFP:
938 		if (is_manager(thrd))
939 			return PL330_STATE_INVALID;
940 		else
941 			return PL330_STATE_WFP;
942 	case DS_ST_KILL:
943 		if (is_manager(thrd))
944 			return PL330_STATE_INVALID;
945 		else
946 			return PL330_STATE_KILLING;
947 	case DS_ST_CMPLT:
948 		if (is_manager(thrd))
949 			return PL330_STATE_INVALID;
950 		else
951 			return PL330_STATE_COMPLETING;
952 	case DS_ST_FLTCMP:
953 		if (is_manager(thrd))
954 			return PL330_STATE_INVALID;
955 		else
956 			return PL330_STATE_FAULT_COMPLETING;
957 	default:
958 		return PL330_STATE_INVALID;
959 	}
960 }
961 
962 static void _stop(struct pl330_thread *thrd)
963 {
964 	void __iomem *regs = thrd->dmac->base;
965 	u8 insn[6] = {0, 0, 0, 0, 0, 0};
966 	u32 inten = readl(regs + INTEN);
967 
968 	if (_state(thrd) == PL330_STATE_FAULT_COMPLETING)
969 		UNTIL(thrd, PL330_STATE_FAULTING | PL330_STATE_KILLING);
970 
971 	/* Return if nothing needs to be done */
972 	if (_state(thrd) == PL330_STATE_COMPLETING
973 		  || _state(thrd) == PL330_STATE_KILLING
974 		  || _state(thrd) == PL330_STATE_STOPPED)
975 		return;
976 
977 	_emit_KILL(0, insn);
978 
979 	_execute_DBGINSN(thrd, insn, is_manager(thrd));
980 
981 	/* clear the event */
982 	if (inten & (1 << thrd->ev))
983 		writel(1 << thrd->ev, regs + INTCLR);
984 	/* Stop generating interrupts for SEV */
985 	writel(inten & ~(1 << thrd->ev), regs + INTEN);
986 }
987 
988 /* Start doing req 'idx' of thread 'thrd' */
989 static bool _trigger(struct pl330_thread *thrd)
990 {
991 	void __iomem *regs = thrd->dmac->base;
992 	struct _pl330_req *req;
993 	struct dma_pl330_desc *desc;
994 	struct _arg_GO go;
995 	unsigned ns;
996 	u8 insn[6] = {0, 0, 0, 0, 0, 0};
997 	int idx;
998 
999 	/* Return if already ACTIVE */
1000 	if (_state(thrd) != PL330_STATE_STOPPED)
1001 		return true;
1002 
1003 	idx = 1 - thrd->lstenq;
1004 	if (thrd->req[idx].desc != NULL) {
1005 		req = &thrd->req[idx];
1006 	} else {
1007 		idx = thrd->lstenq;
1008 		if (thrd->req[idx].desc != NULL)
1009 			req = &thrd->req[idx];
1010 		else
1011 			req = NULL;
1012 	}
1013 
1014 	/* Return if no request */
1015 	if (!req)
1016 		return true;
1017 
1018 	/* Return if req is running */
1019 	if (idx == thrd->req_running)
1020 		return true;
1021 
1022 	desc = req->desc;
1023 
1024 	ns = desc->rqcfg.nonsecure ? 1 : 0;
1025 
1026 	/* See 'Abort Sources' point-4 at Page 2-25 */
1027 	if (_manager_ns(thrd) && !ns)
1028 		dev_info(thrd->dmac->ddma.dev, "%s:%d Recipe for ABORT!\n",
1029 			__func__, __LINE__);
1030 
1031 	go.chan = thrd->id;
1032 	go.addr = req->mc_bus;
1033 	go.ns = ns;
1034 	_emit_GO(0, insn, &go);
1035 
1036 	/* Set to generate interrupts for SEV */
1037 	writel(readl(regs + INTEN) | (1 << thrd->ev), regs + INTEN);
1038 
1039 	/* Only manager can execute GO */
1040 	_execute_DBGINSN(thrd, insn, true);
1041 
1042 	thrd->req_running = idx;
1043 
1044 	return true;
1045 }
1046 
1047 static bool _start(struct pl330_thread *thrd)
1048 {
1049 	switch (_state(thrd)) {
1050 	case PL330_STATE_FAULT_COMPLETING:
1051 		UNTIL(thrd, PL330_STATE_FAULTING | PL330_STATE_KILLING);
1052 
1053 		if (_state(thrd) == PL330_STATE_KILLING)
1054 			UNTIL(thrd, PL330_STATE_STOPPED)
1055 		/* fall through */
1056 
1057 	case PL330_STATE_FAULTING:
1058 		_stop(thrd);
1059 		/* fall through */
1060 
1061 	case PL330_STATE_KILLING:
1062 	case PL330_STATE_COMPLETING:
1063 		UNTIL(thrd, PL330_STATE_STOPPED)
1064 		/* fall through */
1065 
1066 	case PL330_STATE_STOPPED:
1067 		return _trigger(thrd);
1068 
1069 	case PL330_STATE_WFP:
1070 	case PL330_STATE_QUEUEBUSY:
1071 	case PL330_STATE_ATBARRIER:
1072 	case PL330_STATE_UPDTPC:
1073 	case PL330_STATE_CACHEMISS:
1074 	case PL330_STATE_EXECUTING:
1075 		return true;
1076 
1077 	case PL330_STATE_WFE: /* For RESUME, nothing yet */
1078 	default:
1079 		return false;
1080 	}
1081 }
1082 
1083 static inline int _ldst_memtomem(unsigned dry_run, u8 buf[],
1084 		const struct _xfer_spec *pxs, int cyc)
1085 {
1086 	int off = 0;
1087 	struct pl330_config *pcfg = pxs->desc->rqcfg.pcfg;
1088 
1089 	/* check lock-up free version */
1090 	if (get_revision(pcfg->periph_id) >= PERIPH_REV_R1P0) {
1091 		while (cyc--) {
1092 			off += _emit_LD(dry_run, &buf[off], ALWAYS);
1093 			off += _emit_ST(dry_run, &buf[off], ALWAYS);
1094 		}
1095 	} else {
1096 		while (cyc--) {
1097 			off += _emit_LD(dry_run, &buf[off], ALWAYS);
1098 			off += _emit_RMB(dry_run, &buf[off]);
1099 			off += _emit_ST(dry_run, &buf[off], ALWAYS);
1100 			off += _emit_WMB(dry_run, &buf[off]);
1101 		}
1102 	}
1103 
1104 	return off;
1105 }
1106 
1107 static u32 _emit_load(unsigned int dry_run, u8 buf[],
1108 	enum pl330_cond cond, enum dma_transfer_direction direction,
1109 	u8 peri)
1110 {
1111 	int off = 0;
1112 
1113 	switch (direction) {
1114 	case DMA_MEM_TO_MEM:
1115 		/* fall through */
1116 	case DMA_MEM_TO_DEV:
1117 		off += _emit_LD(dry_run, &buf[off], cond);
1118 		break;
1119 
1120 	case DMA_DEV_TO_MEM:
1121 		if (cond == ALWAYS) {
1122 			off += _emit_LDP(dry_run, &buf[off], SINGLE,
1123 				peri);
1124 			off += _emit_LDP(dry_run, &buf[off], BURST,
1125 				peri);
1126 		} else {
1127 			off += _emit_LDP(dry_run, &buf[off], cond,
1128 				peri);
1129 		}
1130 		break;
1131 
1132 	default:
1133 		/* this code should be unreachable */
1134 		WARN_ON(1);
1135 		break;
1136 	}
1137 
1138 	return off;
1139 }
1140 
1141 static inline u32 _emit_store(unsigned int dry_run, u8 buf[],
1142 	enum pl330_cond cond, enum dma_transfer_direction direction,
1143 	u8 peri)
1144 {
1145 	int off = 0;
1146 
1147 	switch (direction) {
1148 	case DMA_MEM_TO_MEM:
1149 		/* fall through */
1150 	case DMA_DEV_TO_MEM:
1151 		off += _emit_ST(dry_run, &buf[off], cond);
1152 		break;
1153 
1154 	case DMA_MEM_TO_DEV:
1155 		if (cond == ALWAYS) {
1156 			off += _emit_STP(dry_run, &buf[off], SINGLE,
1157 				peri);
1158 			off += _emit_STP(dry_run, &buf[off], BURST,
1159 				peri);
1160 		} else {
1161 			off += _emit_STP(dry_run, &buf[off], cond,
1162 				peri);
1163 		}
1164 		break;
1165 
1166 	default:
1167 		/* this code should be unreachable */
1168 		WARN_ON(1);
1169 		break;
1170 	}
1171 
1172 	return off;
1173 }
1174 
1175 static inline int _ldst_peripheral(struct pl330_dmac *pl330,
1176 				 unsigned dry_run, u8 buf[],
1177 				 const struct _xfer_spec *pxs, int cyc,
1178 				 enum pl330_cond cond)
1179 {
1180 	int off = 0;
1181 
1182 	if (pl330->quirks & PL330_QUIRK_BROKEN_NO_FLUSHP)
1183 		cond = BURST;
1184 
1185 	/*
1186 	 * do FLUSHP at beginning to clear any stale dma requests before the
1187 	 * first WFP.
1188 	 */
1189 	if (!(pl330->quirks & PL330_QUIRK_BROKEN_NO_FLUSHP))
1190 		off += _emit_FLUSHP(dry_run, &buf[off], pxs->desc->peri);
1191 	while (cyc--) {
1192 		off += _emit_WFP(dry_run, &buf[off], cond, pxs->desc->peri);
1193 		off += _emit_load(dry_run, &buf[off], cond, pxs->desc->rqtype,
1194 			pxs->desc->peri);
1195 		off += _emit_store(dry_run, &buf[off], cond, pxs->desc->rqtype,
1196 			pxs->desc->peri);
1197 	}
1198 
1199 	return off;
1200 }
1201 
1202 static int _bursts(struct pl330_dmac *pl330, unsigned dry_run, u8 buf[],
1203 		const struct _xfer_spec *pxs, int cyc)
1204 {
1205 	int off = 0;
1206 	enum pl330_cond cond = BRST_LEN(pxs->ccr) > 1 ? BURST : SINGLE;
1207 
1208 	switch (pxs->desc->rqtype) {
1209 	case DMA_MEM_TO_DEV:
1210 		/* fall through */
1211 	case DMA_DEV_TO_MEM:
1212 		off += _ldst_peripheral(pl330, dry_run, &buf[off], pxs, cyc,
1213 			cond);
1214 		break;
1215 
1216 	case DMA_MEM_TO_MEM:
1217 		off += _ldst_memtomem(dry_run, &buf[off], pxs, cyc);
1218 		break;
1219 
1220 	default:
1221 		/* this code should be unreachable */
1222 		WARN_ON(1);
1223 		break;
1224 	}
1225 
1226 	return off;
1227 }
1228 
1229 /*
1230  * transfer dregs with single transfers to peripheral, or a reduced size burst
1231  * for mem-to-mem.
1232  */
1233 static int _dregs(struct pl330_dmac *pl330, unsigned int dry_run, u8 buf[],
1234 		const struct _xfer_spec *pxs, int transfer_length)
1235 {
1236 	int off = 0;
1237 	int dregs_ccr;
1238 
1239 	if (transfer_length == 0)
1240 		return off;
1241 
1242 	switch (pxs->desc->rqtype) {
1243 	case DMA_MEM_TO_DEV:
1244 		/* fall through */
1245 	case DMA_DEV_TO_MEM:
1246 		off += _ldst_peripheral(pl330, dry_run, &buf[off], pxs,
1247 			transfer_length, SINGLE);
1248 		break;
1249 
1250 	case DMA_MEM_TO_MEM:
1251 		dregs_ccr = pxs->ccr;
1252 		dregs_ccr &= ~((0xf << CC_SRCBRSTLEN_SHFT) |
1253 			(0xf << CC_DSTBRSTLEN_SHFT));
1254 		dregs_ccr |= (((transfer_length - 1) & 0xf) <<
1255 			CC_SRCBRSTLEN_SHFT);
1256 		dregs_ccr |= (((transfer_length - 1) & 0xf) <<
1257 			CC_DSTBRSTLEN_SHFT);
1258 		off += _emit_MOV(dry_run, &buf[off], CCR, dregs_ccr);
1259 		off += _ldst_memtomem(dry_run, &buf[off], pxs, 1);
1260 		break;
1261 
1262 	default:
1263 		/* this code should be unreachable */
1264 		WARN_ON(1);
1265 		break;
1266 	}
1267 
1268 	return off;
1269 }
1270 
1271 /* Returns bytes consumed and updates bursts */
1272 static inline int _loop(struct pl330_dmac *pl330, unsigned dry_run, u8 buf[],
1273 		unsigned long *bursts, const struct _xfer_spec *pxs)
1274 {
1275 	int cyc, cycmax, szlp, szlpend, szbrst, off;
1276 	unsigned lcnt0, lcnt1, ljmp0, ljmp1;
1277 	struct _arg_LPEND lpend;
1278 
1279 	if (*bursts == 1)
1280 		return _bursts(pl330, dry_run, buf, pxs, 1);
1281 
1282 	/* Max iterations possible in DMALP is 256 */
1283 	if (*bursts >= 256*256) {
1284 		lcnt1 = 256;
1285 		lcnt0 = 256;
1286 		cyc = *bursts / lcnt1 / lcnt0;
1287 	} else if (*bursts > 256) {
1288 		lcnt1 = 256;
1289 		lcnt0 = *bursts / lcnt1;
1290 		cyc = 1;
1291 	} else {
1292 		lcnt1 = *bursts;
1293 		lcnt0 = 0;
1294 		cyc = 1;
1295 	}
1296 
1297 	szlp = _emit_LP(1, buf, 0, 0);
1298 	szbrst = _bursts(pl330, 1, buf, pxs, 1);
1299 
1300 	lpend.cond = ALWAYS;
1301 	lpend.forever = false;
1302 	lpend.loop = 0;
1303 	lpend.bjump = 0;
1304 	szlpend = _emit_LPEND(1, buf, &lpend);
1305 
1306 	if (lcnt0) {
1307 		szlp *= 2;
1308 		szlpend *= 2;
1309 	}
1310 
1311 	/*
1312 	 * Max bursts that we can unroll due to limit on the
1313 	 * size of backward jump that can be encoded in DMALPEND
1314 	 * which is 8-bits and hence 255
1315 	 */
1316 	cycmax = (255 - (szlp + szlpend)) / szbrst;
1317 
1318 	cyc = (cycmax < cyc) ? cycmax : cyc;
1319 
1320 	off = 0;
1321 
1322 	if (lcnt0) {
1323 		off += _emit_LP(dry_run, &buf[off], 0, lcnt0);
1324 		ljmp0 = off;
1325 	}
1326 
1327 	off += _emit_LP(dry_run, &buf[off], 1, lcnt1);
1328 	ljmp1 = off;
1329 
1330 	off += _bursts(pl330, dry_run, &buf[off], pxs, cyc);
1331 
1332 	lpend.cond = ALWAYS;
1333 	lpend.forever = false;
1334 	lpend.loop = 1;
1335 	lpend.bjump = off - ljmp1;
1336 	off += _emit_LPEND(dry_run, &buf[off], &lpend);
1337 
1338 	if (lcnt0) {
1339 		lpend.cond = ALWAYS;
1340 		lpend.forever = false;
1341 		lpend.loop = 0;
1342 		lpend.bjump = off - ljmp0;
1343 		off += _emit_LPEND(dry_run, &buf[off], &lpend);
1344 	}
1345 
1346 	*bursts = lcnt1 * cyc;
1347 	if (lcnt0)
1348 		*bursts *= lcnt0;
1349 
1350 	return off;
1351 }
1352 
1353 static inline int _setup_loops(struct pl330_dmac *pl330,
1354 			       unsigned dry_run, u8 buf[],
1355 			       const struct _xfer_spec *pxs)
1356 {
1357 	struct pl330_xfer *x = &pxs->desc->px;
1358 	u32 ccr = pxs->ccr;
1359 	unsigned long c, bursts = BYTE_TO_BURST(x->bytes, ccr);
1360 	int num_dregs = (x->bytes - BURST_TO_BYTE(bursts, ccr)) /
1361 		BRST_SIZE(ccr);
1362 	int off = 0;
1363 
1364 	while (bursts) {
1365 		c = bursts;
1366 		off += _loop(pl330, dry_run, &buf[off], &c, pxs);
1367 		bursts -= c;
1368 	}
1369 	off += _dregs(pl330, dry_run, &buf[off], pxs, num_dregs);
1370 
1371 	return off;
1372 }
1373 
1374 static inline int _setup_xfer(struct pl330_dmac *pl330,
1375 			      unsigned dry_run, u8 buf[],
1376 			      const struct _xfer_spec *pxs)
1377 {
1378 	struct pl330_xfer *x = &pxs->desc->px;
1379 	int off = 0;
1380 
1381 	/* DMAMOV SAR, x->src_addr */
1382 	off += _emit_MOV(dry_run, &buf[off], SAR, x->src_addr);
1383 	/* DMAMOV DAR, x->dst_addr */
1384 	off += _emit_MOV(dry_run, &buf[off], DAR, x->dst_addr);
1385 
1386 	/* Setup Loop(s) */
1387 	off += _setup_loops(pl330, dry_run, &buf[off], pxs);
1388 
1389 	return off;
1390 }
1391 
1392 /*
1393  * A req is a sequence of one or more xfer units.
1394  * Returns the number of bytes taken to setup the MC for the req.
1395  */
1396 static int _setup_req(struct pl330_dmac *pl330, unsigned dry_run,
1397 		      struct pl330_thread *thrd, unsigned index,
1398 		      struct _xfer_spec *pxs)
1399 {
1400 	struct _pl330_req *req = &thrd->req[index];
1401 	u8 *buf = req->mc_cpu;
1402 	int off = 0;
1403 
1404 	PL330_DBGMC_START(req->mc_bus);
1405 
1406 	/* DMAMOV CCR, ccr */
1407 	off += _emit_MOV(dry_run, &buf[off], CCR, pxs->ccr);
1408 
1409 	off += _setup_xfer(pl330, dry_run, &buf[off], pxs);
1410 
1411 	/* DMASEV peripheral/event */
1412 	off += _emit_SEV(dry_run, &buf[off], thrd->ev);
1413 	/* DMAEND */
1414 	off += _emit_END(dry_run, &buf[off]);
1415 
1416 	return off;
1417 }
1418 
1419 static inline u32 _prepare_ccr(const struct pl330_reqcfg *rqc)
1420 {
1421 	u32 ccr = 0;
1422 
1423 	if (rqc->src_inc)
1424 		ccr |= CC_SRCINC;
1425 
1426 	if (rqc->dst_inc)
1427 		ccr |= CC_DSTINC;
1428 
1429 	/* We set same protection levels for Src and DST for now */
1430 	if (rqc->privileged)
1431 		ccr |= CC_SRCPRI | CC_DSTPRI;
1432 	if (rqc->nonsecure)
1433 		ccr |= CC_SRCNS | CC_DSTNS;
1434 	if (rqc->insnaccess)
1435 		ccr |= CC_SRCIA | CC_DSTIA;
1436 
1437 	ccr |= (((rqc->brst_len - 1) & 0xf) << CC_SRCBRSTLEN_SHFT);
1438 	ccr |= (((rqc->brst_len - 1) & 0xf) << CC_DSTBRSTLEN_SHFT);
1439 
1440 	ccr |= (rqc->brst_size << CC_SRCBRSTSIZE_SHFT);
1441 	ccr |= (rqc->brst_size << CC_DSTBRSTSIZE_SHFT);
1442 
1443 	ccr |= (rqc->scctl << CC_SRCCCTRL_SHFT);
1444 	ccr |= (rqc->dcctl << CC_DSTCCTRL_SHFT);
1445 
1446 	ccr |= (rqc->swap << CC_SWAP_SHFT);
1447 
1448 	return ccr;
1449 }
1450 
1451 /*
1452  * Submit a list of xfers after which the client wants notification.
1453  * Client is not notified after each xfer unit, just once after all
1454  * xfer units are done or some error occurs.
1455  */
1456 static int pl330_submit_req(struct pl330_thread *thrd,
1457 	struct dma_pl330_desc *desc)
1458 {
1459 	struct pl330_dmac *pl330 = thrd->dmac;
1460 	struct _xfer_spec xs;
1461 	unsigned long flags;
1462 	unsigned idx;
1463 	u32 ccr;
1464 	int ret = 0;
1465 
1466 	switch (desc->rqtype) {
1467 	case DMA_MEM_TO_DEV:
1468 		break;
1469 
1470 	case DMA_DEV_TO_MEM:
1471 		break;
1472 
1473 	case DMA_MEM_TO_MEM:
1474 		break;
1475 
1476 	default:
1477 		return -ENOTSUPP;
1478 	}
1479 
1480 	if (pl330->state == DYING
1481 		|| pl330->dmac_tbd.reset_chan & (1 << thrd->id)) {
1482 		dev_info(thrd->dmac->ddma.dev, "%s:%d\n",
1483 			__func__, __LINE__);
1484 		return -EAGAIN;
1485 	}
1486 
1487 	/* If request for non-existing peripheral */
1488 	if (desc->rqtype != DMA_MEM_TO_MEM &&
1489 	    desc->peri >= pl330->pcfg.num_peri) {
1490 		dev_info(thrd->dmac->ddma.dev,
1491 				"%s:%d Invalid peripheral(%u)!\n",
1492 				__func__, __LINE__, desc->peri);
1493 		return -EINVAL;
1494 	}
1495 
1496 	spin_lock_irqsave(&pl330->lock, flags);
1497 
1498 	if (_queue_full(thrd)) {
1499 		ret = -EAGAIN;
1500 		goto xfer_exit;
1501 	}
1502 
1503 	/* Prefer Secure Channel */
1504 	if (!_manager_ns(thrd))
1505 		desc->rqcfg.nonsecure = 0;
1506 	else
1507 		desc->rqcfg.nonsecure = 1;
1508 
1509 	ccr = _prepare_ccr(&desc->rqcfg);
1510 
1511 	idx = thrd->req[0].desc == NULL ? 0 : 1;
1512 
1513 	xs.ccr = ccr;
1514 	xs.desc = desc;
1515 
1516 	/* First dry run to check if req is acceptable */
1517 	ret = _setup_req(pl330, 1, thrd, idx, &xs);
1518 	if (ret < 0)
1519 		goto xfer_exit;
1520 
1521 	if (ret > pl330->mcbufsz / 2) {
1522 		dev_info(pl330->ddma.dev, "%s:%d Try increasing mcbufsz (%i/%i)\n",
1523 				__func__, __LINE__, ret, pl330->mcbufsz / 2);
1524 		ret = -ENOMEM;
1525 		goto xfer_exit;
1526 	}
1527 
1528 	/* Hook the request */
1529 	thrd->lstenq = idx;
1530 	thrd->req[idx].desc = desc;
1531 	_setup_req(pl330, 0, thrd, idx, &xs);
1532 
1533 	ret = 0;
1534 
1535 xfer_exit:
1536 	spin_unlock_irqrestore(&pl330->lock, flags);
1537 
1538 	return ret;
1539 }
1540 
1541 static void dma_pl330_rqcb(struct dma_pl330_desc *desc, enum pl330_op_err err)
1542 {
1543 	struct dma_pl330_chan *pch;
1544 	unsigned long flags;
1545 
1546 	if (!desc)
1547 		return;
1548 
1549 	pch = desc->pchan;
1550 
1551 	/* If desc aborted */
1552 	if (!pch)
1553 		return;
1554 
1555 	spin_lock_irqsave(&pch->lock, flags);
1556 
1557 	desc->status = DONE;
1558 
1559 	spin_unlock_irqrestore(&pch->lock, flags);
1560 
1561 	tasklet_schedule(&pch->task);
1562 }
1563 
1564 static void pl330_dotask(unsigned long data)
1565 {
1566 	struct pl330_dmac *pl330 = (struct pl330_dmac *) data;
1567 	unsigned long flags;
1568 	int i;
1569 
1570 	spin_lock_irqsave(&pl330->lock, flags);
1571 
1572 	/* The DMAC itself gone nuts */
1573 	if (pl330->dmac_tbd.reset_dmac) {
1574 		pl330->state = DYING;
1575 		/* Reset the manager too */
1576 		pl330->dmac_tbd.reset_mngr = true;
1577 		/* Clear the reset flag */
1578 		pl330->dmac_tbd.reset_dmac = false;
1579 	}
1580 
1581 	if (pl330->dmac_tbd.reset_mngr) {
1582 		_stop(pl330->manager);
1583 		/* Reset all channels */
1584 		pl330->dmac_tbd.reset_chan = (1 << pl330->pcfg.num_chan) - 1;
1585 		/* Clear the reset flag */
1586 		pl330->dmac_tbd.reset_mngr = false;
1587 	}
1588 
1589 	for (i = 0; i < pl330->pcfg.num_chan; i++) {
1590 
1591 		if (pl330->dmac_tbd.reset_chan & (1 << i)) {
1592 			struct pl330_thread *thrd = &pl330->channels[i];
1593 			void __iomem *regs = pl330->base;
1594 			enum pl330_op_err err;
1595 
1596 			_stop(thrd);
1597 
1598 			if (readl(regs + FSC) & (1 << thrd->id))
1599 				err = PL330_ERR_FAIL;
1600 			else
1601 				err = PL330_ERR_ABORT;
1602 
1603 			spin_unlock_irqrestore(&pl330->lock, flags);
1604 			dma_pl330_rqcb(thrd->req[1 - thrd->lstenq].desc, err);
1605 			dma_pl330_rqcb(thrd->req[thrd->lstenq].desc, err);
1606 			spin_lock_irqsave(&pl330->lock, flags);
1607 
1608 			thrd->req[0].desc = NULL;
1609 			thrd->req[1].desc = NULL;
1610 			thrd->req_running = -1;
1611 
1612 			/* Clear the reset flag */
1613 			pl330->dmac_tbd.reset_chan &= ~(1 << i);
1614 		}
1615 	}
1616 
1617 	spin_unlock_irqrestore(&pl330->lock, flags);
1618 
1619 	return;
1620 }
1621 
1622 /* Returns 1 if state was updated, 0 otherwise */
1623 static int pl330_update(struct pl330_dmac *pl330)
1624 {
1625 	struct dma_pl330_desc *descdone;
1626 	unsigned long flags;
1627 	void __iomem *regs;
1628 	u32 val;
1629 	int id, ev, ret = 0;
1630 
1631 	regs = pl330->base;
1632 
1633 	spin_lock_irqsave(&pl330->lock, flags);
1634 
1635 	val = readl(regs + FSM) & 0x1;
1636 	if (val)
1637 		pl330->dmac_tbd.reset_mngr = true;
1638 	else
1639 		pl330->dmac_tbd.reset_mngr = false;
1640 
1641 	val = readl(regs + FSC) & ((1 << pl330->pcfg.num_chan) - 1);
1642 	pl330->dmac_tbd.reset_chan |= val;
1643 	if (val) {
1644 		int i = 0;
1645 		while (i < pl330->pcfg.num_chan) {
1646 			if (val & (1 << i)) {
1647 				dev_info(pl330->ddma.dev,
1648 					"Reset Channel-%d\t CS-%x FTC-%x\n",
1649 						i, readl(regs + CS(i)),
1650 						readl(regs + FTC(i)));
1651 				_stop(&pl330->channels[i]);
1652 			}
1653 			i++;
1654 		}
1655 	}
1656 
1657 	/* Check which event happened i.e, thread notified */
1658 	val = readl(regs + ES);
1659 	if (pl330->pcfg.num_events < 32
1660 			&& val & ~((1 << pl330->pcfg.num_events) - 1)) {
1661 		pl330->dmac_tbd.reset_dmac = true;
1662 		dev_err(pl330->ddma.dev, "%s:%d Unexpected!\n", __func__,
1663 			__LINE__);
1664 		ret = 1;
1665 		goto updt_exit;
1666 	}
1667 
1668 	for (ev = 0; ev < pl330->pcfg.num_events; ev++) {
1669 		if (val & (1 << ev)) { /* Event occurred */
1670 			struct pl330_thread *thrd;
1671 			u32 inten = readl(regs + INTEN);
1672 			int active;
1673 
1674 			/* Clear the event */
1675 			if (inten & (1 << ev))
1676 				writel(1 << ev, regs + INTCLR);
1677 
1678 			ret = 1;
1679 
1680 			id = pl330->events[ev];
1681 
1682 			thrd = &pl330->channels[id];
1683 
1684 			active = thrd->req_running;
1685 			if (active == -1) /* Aborted */
1686 				continue;
1687 
1688 			/* Detach the req */
1689 			descdone = thrd->req[active].desc;
1690 			thrd->req[active].desc = NULL;
1691 
1692 			thrd->req_running = -1;
1693 
1694 			/* Get going again ASAP */
1695 			_start(thrd);
1696 
1697 			/* For now, just make a list of callbacks to be done */
1698 			list_add_tail(&descdone->rqd, &pl330->req_done);
1699 		}
1700 	}
1701 
1702 	/* Now that we are in no hurry, do the callbacks */
1703 	while (!list_empty(&pl330->req_done)) {
1704 		descdone = list_first_entry(&pl330->req_done,
1705 					    struct dma_pl330_desc, rqd);
1706 		list_del(&descdone->rqd);
1707 		spin_unlock_irqrestore(&pl330->lock, flags);
1708 		dma_pl330_rqcb(descdone, PL330_ERR_NONE);
1709 		spin_lock_irqsave(&pl330->lock, flags);
1710 	}
1711 
1712 updt_exit:
1713 	spin_unlock_irqrestore(&pl330->lock, flags);
1714 
1715 	if (pl330->dmac_tbd.reset_dmac
1716 			|| pl330->dmac_tbd.reset_mngr
1717 			|| pl330->dmac_tbd.reset_chan) {
1718 		ret = 1;
1719 		tasklet_schedule(&pl330->tasks);
1720 	}
1721 
1722 	return ret;
1723 }
1724 
1725 /* Reserve an event */
1726 static inline int _alloc_event(struct pl330_thread *thrd)
1727 {
1728 	struct pl330_dmac *pl330 = thrd->dmac;
1729 	int ev;
1730 
1731 	for (ev = 0; ev < pl330->pcfg.num_events; ev++)
1732 		if (pl330->events[ev] == -1) {
1733 			pl330->events[ev] = thrd->id;
1734 			return ev;
1735 		}
1736 
1737 	return -1;
1738 }
1739 
1740 static bool _chan_ns(const struct pl330_dmac *pl330, int i)
1741 {
1742 	return pl330->pcfg.irq_ns & (1 << i);
1743 }
1744 
1745 /* Upon success, returns IdentityToken for the
1746  * allocated channel, NULL otherwise.
1747  */
1748 static struct pl330_thread *pl330_request_channel(struct pl330_dmac *pl330)
1749 {
1750 	struct pl330_thread *thrd = NULL;
1751 	int chans, i;
1752 
1753 	if (pl330->state == DYING)
1754 		return NULL;
1755 
1756 	chans = pl330->pcfg.num_chan;
1757 
1758 	for (i = 0; i < chans; i++) {
1759 		thrd = &pl330->channels[i];
1760 		if ((thrd->free) && (!_manager_ns(thrd) ||
1761 					_chan_ns(pl330, i))) {
1762 			thrd->ev = _alloc_event(thrd);
1763 			if (thrd->ev >= 0) {
1764 				thrd->free = false;
1765 				thrd->lstenq = 1;
1766 				thrd->req[0].desc = NULL;
1767 				thrd->req[1].desc = NULL;
1768 				thrd->req_running = -1;
1769 				break;
1770 			}
1771 		}
1772 		thrd = NULL;
1773 	}
1774 
1775 	return thrd;
1776 }
1777 
1778 /* Release an event */
1779 static inline void _free_event(struct pl330_thread *thrd, int ev)
1780 {
1781 	struct pl330_dmac *pl330 = thrd->dmac;
1782 
1783 	/* If the event is valid and was held by the thread */
1784 	if (ev >= 0 && ev < pl330->pcfg.num_events
1785 			&& pl330->events[ev] == thrd->id)
1786 		pl330->events[ev] = -1;
1787 }
1788 
1789 static void pl330_release_channel(struct pl330_thread *thrd)
1790 {
1791 	if (!thrd || thrd->free)
1792 		return;
1793 
1794 	_stop(thrd);
1795 
1796 	dma_pl330_rqcb(thrd->req[1 - thrd->lstenq].desc, PL330_ERR_ABORT);
1797 	dma_pl330_rqcb(thrd->req[thrd->lstenq].desc, PL330_ERR_ABORT);
1798 
1799 	_free_event(thrd, thrd->ev);
1800 	thrd->free = true;
1801 }
1802 
1803 /* Initialize the structure for PL330 configuration, that can be used
1804  * by the client driver the make best use of the DMAC
1805  */
1806 static void read_dmac_config(struct pl330_dmac *pl330)
1807 {
1808 	void __iomem *regs = pl330->base;
1809 	u32 val;
1810 
1811 	val = readl(regs + CRD) >> CRD_DATA_WIDTH_SHIFT;
1812 	val &= CRD_DATA_WIDTH_MASK;
1813 	pl330->pcfg.data_bus_width = 8 * (1 << val);
1814 
1815 	val = readl(regs + CRD) >> CRD_DATA_BUFF_SHIFT;
1816 	val &= CRD_DATA_BUFF_MASK;
1817 	pl330->pcfg.data_buf_dep = val + 1;
1818 
1819 	val = readl(regs + CR0) >> CR0_NUM_CHANS_SHIFT;
1820 	val &= CR0_NUM_CHANS_MASK;
1821 	val += 1;
1822 	pl330->pcfg.num_chan = val;
1823 
1824 	val = readl(regs + CR0);
1825 	if (val & CR0_PERIPH_REQ_SET) {
1826 		val = (val >> CR0_NUM_PERIPH_SHIFT) & CR0_NUM_PERIPH_MASK;
1827 		val += 1;
1828 		pl330->pcfg.num_peri = val;
1829 		pl330->pcfg.peri_ns = readl(regs + CR4);
1830 	} else {
1831 		pl330->pcfg.num_peri = 0;
1832 	}
1833 
1834 	val = readl(regs + CR0);
1835 	if (val & CR0_BOOT_MAN_NS)
1836 		pl330->pcfg.mode |= DMAC_MODE_NS;
1837 	else
1838 		pl330->pcfg.mode &= ~DMAC_MODE_NS;
1839 
1840 	val = readl(regs + CR0) >> CR0_NUM_EVENTS_SHIFT;
1841 	val &= CR0_NUM_EVENTS_MASK;
1842 	val += 1;
1843 	pl330->pcfg.num_events = val;
1844 
1845 	pl330->pcfg.irq_ns = readl(regs + CR3);
1846 }
1847 
1848 static inline void _reset_thread(struct pl330_thread *thrd)
1849 {
1850 	struct pl330_dmac *pl330 = thrd->dmac;
1851 
1852 	thrd->req[0].mc_cpu = pl330->mcode_cpu
1853 				+ (thrd->id * pl330->mcbufsz);
1854 	thrd->req[0].mc_bus = pl330->mcode_bus
1855 				+ (thrd->id * pl330->mcbufsz);
1856 	thrd->req[0].desc = NULL;
1857 
1858 	thrd->req[1].mc_cpu = thrd->req[0].mc_cpu
1859 				+ pl330->mcbufsz / 2;
1860 	thrd->req[1].mc_bus = thrd->req[0].mc_bus
1861 				+ pl330->mcbufsz / 2;
1862 	thrd->req[1].desc = NULL;
1863 
1864 	thrd->req_running = -1;
1865 }
1866 
1867 static int dmac_alloc_threads(struct pl330_dmac *pl330)
1868 {
1869 	int chans = pl330->pcfg.num_chan;
1870 	struct pl330_thread *thrd;
1871 	int i;
1872 
1873 	/* Allocate 1 Manager and 'chans' Channel threads */
1874 	pl330->channels = kcalloc(1 + chans, sizeof(*thrd),
1875 					GFP_KERNEL);
1876 	if (!pl330->channels)
1877 		return -ENOMEM;
1878 
1879 	/* Init Channel threads */
1880 	for (i = 0; i < chans; i++) {
1881 		thrd = &pl330->channels[i];
1882 		thrd->id = i;
1883 		thrd->dmac = pl330;
1884 		_reset_thread(thrd);
1885 		thrd->free = true;
1886 	}
1887 
1888 	/* MANAGER is indexed at the end */
1889 	thrd = &pl330->channels[chans];
1890 	thrd->id = chans;
1891 	thrd->dmac = pl330;
1892 	thrd->free = false;
1893 	pl330->manager = thrd;
1894 
1895 	return 0;
1896 }
1897 
1898 static int dmac_alloc_resources(struct pl330_dmac *pl330)
1899 {
1900 	int chans = pl330->pcfg.num_chan;
1901 	int ret;
1902 
1903 	/*
1904 	 * Alloc MicroCode buffer for 'chans' Channel threads.
1905 	 * A channel's buffer offset is (Channel_Id * MCODE_BUFF_PERCHAN)
1906 	 */
1907 	pl330->mcode_cpu = dma_alloc_attrs(pl330->ddma.dev,
1908 				chans * pl330->mcbufsz,
1909 				&pl330->mcode_bus, GFP_KERNEL,
1910 				DMA_ATTR_PRIVILEGED);
1911 	if (!pl330->mcode_cpu) {
1912 		dev_err(pl330->ddma.dev, "%s:%d Can't allocate memory!\n",
1913 			__func__, __LINE__);
1914 		return -ENOMEM;
1915 	}
1916 
1917 	ret = dmac_alloc_threads(pl330);
1918 	if (ret) {
1919 		dev_err(pl330->ddma.dev, "%s:%d Can't to create channels for DMAC!\n",
1920 			__func__, __LINE__);
1921 		dma_free_coherent(pl330->ddma.dev,
1922 				chans * pl330->mcbufsz,
1923 				pl330->mcode_cpu, pl330->mcode_bus);
1924 		return ret;
1925 	}
1926 
1927 	return 0;
1928 }
1929 
1930 static int pl330_add(struct pl330_dmac *pl330)
1931 {
1932 	int i, ret;
1933 
1934 	/* Check if we can handle this DMAC */
1935 	if ((pl330->pcfg.periph_id & 0xfffff) != PERIPH_ID_VAL) {
1936 		dev_err(pl330->ddma.dev, "PERIPH_ID 0x%x !\n",
1937 			pl330->pcfg.periph_id);
1938 		return -EINVAL;
1939 	}
1940 
1941 	/* Read the configuration of the DMAC */
1942 	read_dmac_config(pl330);
1943 
1944 	if (pl330->pcfg.num_events == 0) {
1945 		dev_err(pl330->ddma.dev, "%s:%d Can't work without events!\n",
1946 			__func__, __LINE__);
1947 		return -EINVAL;
1948 	}
1949 
1950 	spin_lock_init(&pl330->lock);
1951 
1952 	INIT_LIST_HEAD(&pl330->req_done);
1953 
1954 	/* Use default MC buffer size if not provided */
1955 	if (!pl330->mcbufsz)
1956 		pl330->mcbufsz = MCODE_BUFF_PER_REQ * 2;
1957 
1958 	/* Mark all events as free */
1959 	for (i = 0; i < pl330->pcfg.num_events; i++)
1960 		pl330->events[i] = -1;
1961 
1962 	/* Allocate resources needed by the DMAC */
1963 	ret = dmac_alloc_resources(pl330);
1964 	if (ret) {
1965 		dev_err(pl330->ddma.dev, "Unable to create channels for DMAC\n");
1966 		return ret;
1967 	}
1968 
1969 	tasklet_init(&pl330->tasks, pl330_dotask, (unsigned long) pl330);
1970 
1971 	pl330->state = INIT;
1972 
1973 	return 0;
1974 }
1975 
1976 static int dmac_free_threads(struct pl330_dmac *pl330)
1977 {
1978 	struct pl330_thread *thrd;
1979 	int i;
1980 
1981 	/* Release Channel threads */
1982 	for (i = 0; i < pl330->pcfg.num_chan; i++) {
1983 		thrd = &pl330->channels[i];
1984 		pl330_release_channel(thrd);
1985 	}
1986 
1987 	/* Free memory */
1988 	kfree(pl330->channels);
1989 
1990 	return 0;
1991 }
1992 
1993 static void pl330_del(struct pl330_dmac *pl330)
1994 {
1995 	pl330->state = UNINIT;
1996 
1997 	tasklet_kill(&pl330->tasks);
1998 
1999 	/* Free DMAC resources */
2000 	dmac_free_threads(pl330);
2001 
2002 	dma_free_coherent(pl330->ddma.dev,
2003 		pl330->pcfg.num_chan * pl330->mcbufsz, pl330->mcode_cpu,
2004 		pl330->mcode_bus);
2005 }
2006 
2007 /* forward declaration */
2008 static struct amba_driver pl330_driver;
2009 
2010 static inline struct dma_pl330_chan *
2011 to_pchan(struct dma_chan *ch)
2012 {
2013 	if (!ch)
2014 		return NULL;
2015 
2016 	return container_of(ch, struct dma_pl330_chan, chan);
2017 }
2018 
2019 static inline struct dma_pl330_desc *
2020 to_desc(struct dma_async_tx_descriptor *tx)
2021 {
2022 	return container_of(tx, struct dma_pl330_desc, txd);
2023 }
2024 
2025 static inline void fill_queue(struct dma_pl330_chan *pch)
2026 {
2027 	struct dma_pl330_desc *desc;
2028 	int ret;
2029 
2030 	list_for_each_entry(desc, &pch->work_list, node) {
2031 
2032 		/* If already submitted */
2033 		if (desc->status == BUSY)
2034 			continue;
2035 
2036 		ret = pl330_submit_req(pch->thread, desc);
2037 		if (!ret) {
2038 			desc->status = BUSY;
2039 		} else if (ret == -EAGAIN) {
2040 			/* QFull or DMAC Dying */
2041 			break;
2042 		} else {
2043 			/* Unacceptable request */
2044 			desc->status = DONE;
2045 			dev_err(pch->dmac->ddma.dev, "%s:%d Bad Desc(%d)\n",
2046 					__func__, __LINE__, desc->txd.cookie);
2047 			tasklet_schedule(&pch->task);
2048 		}
2049 	}
2050 }
2051 
2052 static void pl330_tasklet(unsigned long data)
2053 {
2054 	struct dma_pl330_chan *pch = (struct dma_pl330_chan *)data;
2055 	struct dma_pl330_desc *desc, *_dt;
2056 	unsigned long flags;
2057 	bool power_down = false;
2058 
2059 	spin_lock_irqsave(&pch->lock, flags);
2060 
2061 	/* Pick up ripe tomatoes */
2062 	list_for_each_entry_safe(desc, _dt, &pch->work_list, node)
2063 		if (desc->status == DONE) {
2064 			if (!pch->cyclic)
2065 				dma_cookie_complete(&desc->txd);
2066 			list_move_tail(&desc->node, &pch->completed_list);
2067 		}
2068 
2069 	/* Try to submit a req imm. next to the last completed cookie */
2070 	fill_queue(pch);
2071 
2072 	if (list_empty(&pch->work_list)) {
2073 		spin_lock(&pch->thread->dmac->lock);
2074 		_stop(pch->thread);
2075 		spin_unlock(&pch->thread->dmac->lock);
2076 		power_down = true;
2077 		pch->active = false;
2078 	} else {
2079 		/* Make sure the PL330 Channel thread is active */
2080 		spin_lock(&pch->thread->dmac->lock);
2081 		_start(pch->thread);
2082 		spin_unlock(&pch->thread->dmac->lock);
2083 	}
2084 
2085 	while (!list_empty(&pch->completed_list)) {
2086 		struct dmaengine_desc_callback cb;
2087 
2088 		desc = list_first_entry(&pch->completed_list,
2089 					struct dma_pl330_desc, node);
2090 
2091 		dmaengine_desc_get_callback(&desc->txd, &cb);
2092 
2093 		if (pch->cyclic) {
2094 			desc->status = PREP;
2095 			list_move_tail(&desc->node, &pch->work_list);
2096 			if (power_down) {
2097 				pch->active = true;
2098 				spin_lock(&pch->thread->dmac->lock);
2099 				_start(pch->thread);
2100 				spin_unlock(&pch->thread->dmac->lock);
2101 				power_down = false;
2102 			}
2103 		} else {
2104 			desc->status = FREE;
2105 			list_move_tail(&desc->node, &pch->dmac->desc_pool);
2106 		}
2107 
2108 		dma_descriptor_unmap(&desc->txd);
2109 
2110 		if (dmaengine_desc_callback_valid(&cb)) {
2111 			spin_unlock_irqrestore(&pch->lock, flags);
2112 			dmaengine_desc_callback_invoke(&cb, NULL);
2113 			spin_lock_irqsave(&pch->lock, flags);
2114 		}
2115 	}
2116 	spin_unlock_irqrestore(&pch->lock, flags);
2117 
2118 	/* If work list empty, power down */
2119 	if (power_down) {
2120 		pm_runtime_mark_last_busy(pch->dmac->ddma.dev);
2121 		pm_runtime_put_autosuspend(pch->dmac->ddma.dev);
2122 	}
2123 }
2124 
2125 static struct dma_chan *of_dma_pl330_xlate(struct of_phandle_args *dma_spec,
2126 						struct of_dma *ofdma)
2127 {
2128 	int count = dma_spec->args_count;
2129 	struct pl330_dmac *pl330 = ofdma->of_dma_data;
2130 	unsigned int chan_id;
2131 
2132 	if (!pl330)
2133 		return NULL;
2134 
2135 	if (count != 1)
2136 		return NULL;
2137 
2138 	chan_id = dma_spec->args[0];
2139 	if (chan_id >= pl330->num_peripherals)
2140 		return NULL;
2141 
2142 	return dma_get_slave_channel(&pl330->peripherals[chan_id].chan);
2143 }
2144 
2145 static int pl330_alloc_chan_resources(struct dma_chan *chan)
2146 {
2147 	struct dma_pl330_chan *pch = to_pchan(chan);
2148 	struct pl330_dmac *pl330 = pch->dmac;
2149 	unsigned long flags;
2150 
2151 	spin_lock_irqsave(&pl330->lock, flags);
2152 
2153 	dma_cookie_init(chan);
2154 	pch->cyclic = false;
2155 
2156 	pch->thread = pl330_request_channel(pl330);
2157 	if (!pch->thread) {
2158 		spin_unlock_irqrestore(&pl330->lock, flags);
2159 		return -ENOMEM;
2160 	}
2161 
2162 	tasklet_init(&pch->task, pl330_tasklet, (unsigned long) pch);
2163 
2164 	spin_unlock_irqrestore(&pl330->lock, flags);
2165 
2166 	return 1;
2167 }
2168 
2169 /*
2170  * We need the data direction between the DMAC (the dma-mapping "device") and
2171  * the FIFO (the dmaengine "dev"), from the FIFO's point of view. Confusing!
2172  */
2173 static enum dma_data_direction
2174 pl330_dma_slave_map_dir(enum dma_transfer_direction dir)
2175 {
2176 	switch (dir) {
2177 	case DMA_MEM_TO_DEV:
2178 		return DMA_FROM_DEVICE;
2179 	case DMA_DEV_TO_MEM:
2180 		return DMA_TO_DEVICE;
2181 	case DMA_DEV_TO_DEV:
2182 		return DMA_BIDIRECTIONAL;
2183 	default:
2184 		return DMA_NONE;
2185 	}
2186 }
2187 
2188 static void pl330_unprep_slave_fifo(struct dma_pl330_chan *pch)
2189 {
2190 	if (pch->dir != DMA_NONE)
2191 		dma_unmap_resource(pch->chan.device->dev, pch->fifo_dma,
2192 				   1 << pch->burst_sz, pch->dir, 0);
2193 	pch->dir = DMA_NONE;
2194 }
2195 
2196 
2197 static bool pl330_prep_slave_fifo(struct dma_pl330_chan *pch,
2198 				  enum dma_transfer_direction dir)
2199 {
2200 	struct device *dev = pch->chan.device->dev;
2201 	enum dma_data_direction dma_dir = pl330_dma_slave_map_dir(dir);
2202 
2203 	/* Already mapped for this config? */
2204 	if (pch->dir == dma_dir)
2205 		return true;
2206 
2207 	pl330_unprep_slave_fifo(pch);
2208 	pch->fifo_dma = dma_map_resource(dev, pch->fifo_addr,
2209 					 1 << pch->burst_sz, dma_dir, 0);
2210 	if (dma_mapping_error(dev, pch->fifo_dma))
2211 		return false;
2212 
2213 	pch->dir = dma_dir;
2214 	return true;
2215 }
2216 
2217 static int fixup_burst_len(int max_burst_len, int quirks)
2218 {
2219 	if (quirks & PL330_QUIRK_BROKEN_NO_FLUSHP)
2220 		return 1;
2221 	else if (max_burst_len > PL330_MAX_BURST)
2222 		return PL330_MAX_BURST;
2223 	else if (max_burst_len < 1)
2224 		return 1;
2225 	else
2226 		return max_burst_len;
2227 }
2228 
2229 static int pl330_config_write(struct dma_chan *chan,
2230 			struct dma_slave_config *slave_config,
2231 			enum dma_transfer_direction direction)
2232 {
2233 	struct dma_pl330_chan *pch = to_pchan(chan);
2234 
2235 	pl330_unprep_slave_fifo(pch);
2236 	if (direction == DMA_MEM_TO_DEV) {
2237 		if (slave_config->dst_addr)
2238 			pch->fifo_addr = slave_config->dst_addr;
2239 		if (slave_config->dst_addr_width)
2240 			pch->burst_sz = __ffs(slave_config->dst_addr_width);
2241 		pch->burst_len = fixup_burst_len(slave_config->dst_maxburst,
2242 			pch->dmac->quirks);
2243 	} else if (direction == DMA_DEV_TO_MEM) {
2244 		if (slave_config->src_addr)
2245 			pch->fifo_addr = slave_config->src_addr;
2246 		if (slave_config->src_addr_width)
2247 			pch->burst_sz = __ffs(slave_config->src_addr_width);
2248 		pch->burst_len = fixup_burst_len(slave_config->src_maxburst,
2249 			pch->dmac->quirks);
2250 	}
2251 
2252 	return 0;
2253 }
2254 
2255 static int pl330_config(struct dma_chan *chan,
2256 			struct dma_slave_config *slave_config)
2257 {
2258 	struct dma_pl330_chan *pch = to_pchan(chan);
2259 
2260 	memcpy(&pch->slave_config, slave_config, sizeof(*slave_config));
2261 
2262 	return 0;
2263 }
2264 
2265 static int pl330_terminate_all(struct dma_chan *chan)
2266 {
2267 	struct dma_pl330_chan *pch = to_pchan(chan);
2268 	struct dma_pl330_desc *desc;
2269 	unsigned long flags;
2270 	struct pl330_dmac *pl330 = pch->dmac;
2271 	bool power_down = false;
2272 
2273 	pm_runtime_get_sync(pl330->ddma.dev);
2274 	spin_lock_irqsave(&pch->lock, flags);
2275 
2276 	spin_lock(&pl330->lock);
2277 	_stop(pch->thread);
2278 	pch->thread->req[0].desc = NULL;
2279 	pch->thread->req[1].desc = NULL;
2280 	pch->thread->req_running = -1;
2281 	spin_unlock(&pl330->lock);
2282 
2283 	power_down = pch->active;
2284 	pch->active = false;
2285 
2286 	/* Mark all desc done */
2287 	list_for_each_entry(desc, &pch->submitted_list, node) {
2288 		desc->status = FREE;
2289 		dma_cookie_complete(&desc->txd);
2290 	}
2291 
2292 	list_for_each_entry(desc, &pch->work_list , node) {
2293 		desc->status = FREE;
2294 		dma_cookie_complete(&desc->txd);
2295 	}
2296 
2297 	list_splice_tail_init(&pch->submitted_list, &pl330->desc_pool);
2298 	list_splice_tail_init(&pch->work_list, &pl330->desc_pool);
2299 	list_splice_tail_init(&pch->completed_list, &pl330->desc_pool);
2300 	spin_unlock_irqrestore(&pch->lock, flags);
2301 	pm_runtime_mark_last_busy(pl330->ddma.dev);
2302 	if (power_down)
2303 		pm_runtime_put_autosuspend(pl330->ddma.dev);
2304 	pm_runtime_put_autosuspend(pl330->ddma.dev);
2305 
2306 	return 0;
2307 }
2308 
2309 /*
2310  * We don't support DMA_RESUME command because of hardware
2311  * limitations, so after pausing the channel we cannot restore
2312  * it to active state. We have to terminate channel and setup
2313  * DMA transfer again. This pause feature was implemented to
2314  * allow safely read residue before channel termination.
2315  */
2316 static int pl330_pause(struct dma_chan *chan)
2317 {
2318 	struct dma_pl330_chan *pch = to_pchan(chan);
2319 	struct pl330_dmac *pl330 = pch->dmac;
2320 	unsigned long flags;
2321 
2322 	pm_runtime_get_sync(pl330->ddma.dev);
2323 	spin_lock_irqsave(&pch->lock, flags);
2324 
2325 	spin_lock(&pl330->lock);
2326 	_stop(pch->thread);
2327 	spin_unlock(&pl330->lock);
2328 
2329 	spin_unlock_irqrestore(&pch->lock, flags);
2330 	pm_runtime_mark_last_busy(pl330->ddma.dev);
2331 	pm_runtime_put_autosuspend(pl330->ddma.dev);
2332 
2333 	return 0;
2334 }
2335 
2336 static void pl330_free_chan_resources(struct dma_chan *chan)
2337 {
2338 	struct dma_pl330_chan *pch = to_pchan(chan);
2339 	struct pl330_dmac *pl330 = pch->dmac;
2340 	unsigned long flags;
2341 
2342 	tasklet_kill(&pch->task);
2343 
2344 	pm_runtime_get_sync(pch->dmac->ddma.dev);
2345 	spin_lock_irqsave(&pl330->lock, flags);
2346 
2347 	pl330_release_channel(pch->thread);
2348 	pch->thread = NULL;
2349 
2350 	if (pch->cyclic)
2351 		list_splice_tail_init(&pch->work_list, &pch->dmac->desc_pool);
2352 
2353 	spin_unlock_irqrestore(&pl330->lock, flags);
2354 	pm_runtime_mark_last_busy(pch->dmac->ddma.dev);
2355 	pm_runtime_put_autosuspend(pch->dmac->ddma.dev);
2356 	pl330_unprep_slave_fifo(pch);
2357 }
2358 
2359 static int pl330_get_current_xferred_count(struct dma_pl330_chan *pch,
2360 					   struct dma_pl330_desc *desc)
2361 {
2362 	struct pl330_thread *thrd = pch->thread;
2363 	struct pl330_dmac *pl330 = pch->dmac;
2364 	void __iomem *regs = thrd->dmac->base;
2365 	u32 val, addr;
2366 
2367 	pm_runtime_get_sync(pl330->ddma.dev);
2368 	val = addr = 0;
2369 	if (desc->rqcfg.src_inc) {
2370 		val = readl(regs + SA(thrd->id));
2371 		addr = desc->px.src_addr;
2372 	} else {
2373 		val = readl(regs + DA(thrd->id));
2374 		addr = desc->px.dst_addr;
2375 	}
2376 	pm_runtime_mark_last_busy(pch->dmac->ddma.dev);
2377 	pm_runtime_put_autosuspend(pl330->ddma.dev);
2378 
2379 	/* If DMAMOV hasn't finished yet, SAR/DAR can be zero */
2380 	if (!val)
2381 		return 0;
2382 
2383 	return val - addr;
2384 }
2385 
2386 static enum dma_status
2387 pl330_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
2388 		 struct dma_tx_state *txstate)
2389 {
2390 	enum dma_status ret;
2391 	unsigned long flags;
2392 	struct dma_pl330_desc *desc, *running = NULL, *last_enq = NULL;
2393 	struct dma_pl330_chan *pch = to_pchan(chan);
2394 	unsigned int transferred, residual = 0;
2395 
2396 	ret = dma_cookie_status(chan, cookie, txstate);
2397 
2398 	if (!txstate)
2399 		return ret;
2400 
2401 	if (ret == DMA_COMPLETE)
2402 		goto out;
2403 
2404 	spin_lock_irqsave(&pch->lock, flags);
2405 	spin_lock(&pch->thread->dmac->lock);
2406 
2407 	if (pch->thread->req_running != -1)
2408 		running = pch->thread->req[pch->thread->req_running].desc;
2409 
2410 	last_enq = pch->thread->req[pch->thread->lstenq].desc;
2411 
2412 	/* Check in pending list */
2413 	list_for_each_entry(desc, &pch->work_list, node) {
2414 		if (desc->status == DONE)
2415 			transferred = desc->bytes_requested;
2416 		else if (running && desc == running)
2417 			transferred =
2418 				pl330_get_current_xferred_count(pch, desc);
2419 		else if (desc->status == BUSY)
2420 			/*
2421 			 * Busy but not running means either just enqueued,
2422 			 * or finished and not yet marked done
2423 			 */
2424 			if (desc == last_enq)
2425 				transferred = 0;
2426 			else
2427 				transferred = desc->bytes_requested;
2428 		else
2429 			transferred = 0;
2430 		residual += desc->bytes_requested - transferred;
2431 		if (desc->txd.cookie == cookie) {
2432 			switch (desc->status) {
2433 			case DONE:
2434 				ret = DMA_COMPLETE;
2435 				break;
2436 			case PREP:
2437 			case BUSY:
2438 				ret = DMA_IN_PROGRESS;
2439 				break;
2440 			default:
2441 				WARN_ON(1);
2442 			}
2443 			break;
2444 		}
2445 		if (desc->last)
2446 			residual = 0;
2447 	}
2448 	spin_unlock(&pch->thread->dmac->lock);
2449 	spin_unlock_irqrestore(&pch->lock, flags);
2450 
2451 out:
2452 	dma_set_residue(txstate, residual);
2453 
2454 	return ret;
2455 }
2456 
2457 static void pl330_issue_pending(struct dma_chan *chan)
2458 {
2459 	struct dma_pl330_chan *pch = to_pchan(chan);
2460 	unsigned long flags;
2461 
2462 	spin_lock_irqsave(&pch->lock, flags);
2463 	if (list_empty(&pch->work_list)) {
2464 		/*
2465 		 * Warn on nothing pending. Empty submitted_list may
2466 		 * break our pm_runtime usage counter as it is
2467 		 * updated on work_list emptiness status.
2468 		 */
2469 		WARN_ON(list_empty(&pch->submitted_list));
2470 		pch->active = true;
2471 		pm_runtime_get_sync(pch->dmac->ddma.dev);
2472 	}
2473 	list_splice_tail_init(&pch->submitted_list, &pch->work_list);
2474 	spin_unlock_irqrestore(&pch->lock, flags);
2475 
2476 	pl330_tasklet((unsigned long)pch);
2477 }
2478 
2479 /*
2480  * We returned the last one of the circular list of descriptor(s)
2481  * from prep_xxx, so the argument to submit corresponds to the last
2482  * descriptor of the list.
2483  */
2484 static dma_cookie_t pl330_tx_submit(struct dma_async_tx_descriptor *tx)
2485 {
2486 	struct dma_pl330_desc *desc, *last = to_desc(tx);
2487 	struct dma_pl330_chan *pch = to_pchan(tx->chan);
2488 	dma_cookie_t cookie;
2489 	unsigned long flags;
2490 
2491 	spin_lock_irqsave(&pch->lock, flags);
2492 
2493 	/* Assign cookies to all nodes */
2494 	while (!list_empty(&last->node)) {
2495 		desc = list_entry(last->node.next, struct dma_pl330_desc, node);
2496 		if (pch->cyclic) {
2497 			desc->txd.callback = last->txd.callback;
2498 			desc->txd.callback_param = last->txd.callback_param;
2499 		}
2500 		desc->last = false;
2501 
2502 		dma_cookie_assign(&desc->txd);
2503 
2504 		list_move_tail(&desc->node, &pch->submitted_list);
2505 	}
2506 
2507 	last->last = true;
2508 	cookie = dma_cookie_assign(&last->txd);
2509 	list_add_tail(&last->node, &pch->submitted_list);
2510 	spin_unlock_irqrestore(&pch->lock, flags);
2511 
2512 	return cookie;
2513 }
2514 
2515 static inline void _init_desc(struct dma_pl330_desc *desc)
2516 {
2517 	desc->rqcfg.swap = SWAP_NO;
2518 	desc->rqcfg.scctl = CCTRL0;
2519 	desc->rqcfg.dcctl = CCTRL0;
2520 	desc->txd.tx_submit = pl330_tx_submit;
2521 
2522 	INIT_LIST_HEAD(&desc->node);
2523 }
2524 
2525 /* Returns the number of descriptors added to the DMAC pool */
2526 static int add_desc(struct list_head *pool, spinlock_t *lock,
2527 		    gfp_t flg, int count)
2528 {
2529 	struct dma_pl330_desc *desc;
2530 	unsigned long flags;
2531 	int i;
2532 
2533 	desc = kcalloc(count, sizeof(*desc), flg);
2534 	if (!desc)
2535 		return 0;
2536 
2537 	spin_lock_irqsave(lock, flags);
2538 
2539 	for (i = 0; i < count; i++) {
2540 		_init_desc(&desc[i]);
2541 		list_add_tail(&desc[i].node, pool);
2542 	}
2543 
2544 	spin_unlock_irqrestore(lock, flags);
2545 
2546 	return count;
2547 }
2548 
2549 static struct dma_pl330_desc *pluck_desc(struct list_head *pool,
2550 					 spinlock_t *lock)
2551 {
2552 	struct dma_pl330_desc *desc = NULL;
2553 	unsigned long flags;
2554 
2555 	spin_lock_irqsave(lock, flags);
2556 
2557 	if (!list_empty(pool)) {
2558 		desc = list_entry(pool->next,
2559 				struct dma_pl330_desc, node);
2560 
2561 		list_del_init(&desc->node);
2562 
2563 		desc->status = PREP;
2564 		desc->txd.callback = NULL;
2565 	}
2566 
2567 	spin_unlock_irqrestore(lock, flags);
2568 
2569 	return desc;
2570 }
2571 
2572 static struct dma_pl330_desc *pl330_get_desc(struct dma_pl330_chan *pch)
2573 {
2574 	struct pl330_dmac *pl330 = pch->dmac;
2575 	u8 *peri_id = pch->chan.private;
2576 	struct dma_pl330_desc *desc;
2577 
2578 	/* Pluck one desc from the pool of DMAC */
2579 	desc = pluck_desc(&pl330->desc_pool, &pl330->pool_lock);
2580 
2581 	/* If the DMAC pool is empty, alloc new */
2582 	if (!desc) {
2583 		DEFINE_SPINLOCK(lock);
2584 		LIST_HEAD(pool);
2585 
2586 		if (!add_desc(&pool, &lock, GFP_ATOMIC, 1))
2587 			return NULL;
2588 
2589 		desc = pluck_desc(&pool, &lock);
2590 		WARN_ON(!desc || !list_empty(&pool));
2591 	}
2592 
2593 	/* Initialize the descriptor */
2594 	desc->pchan = pch;
2595 	desc->txd.cookie = 0;
2596 	async_tx_ack(&desc->txd);
2597 
2598 	desc->peri = peri_id ? pch->chan.chan_id : 0;
2599 	desc->rqcfg.pcfg = &pch->dmac->pcfg;
2600 
2601 	dma_async_tx_descriptor_init(&desc->txd, &pch->chan);
2602 
2603 	return desc;
2604 }
2605 
2606 static inline void fill_px(struct pl330_xfer *px,
2607 		dma_addr_t dst, dma_addr_t src, size_t len)
2608 {
2609 	px->bytes = len;
2610 	px->dst_addr = dst;
2611 	px->src_addr = src;
2612 }
2613 
2614 static struct dma_pl330_desc *
2615 __pl330_prep_dma_memcpy(struct dma_pl330_chan *pch, dma_addr_t dst,
2616 		dma_addr_t src, size_t len)
2617 {
2618 	struct dma_pl330_desc *desc = pl330_get_desc(pch);
2619 
2620 	if (!desc) {
2621 		dev_err(pch->dmac->ddma.dev, "%s:%d Unable to fetch desc\n",
2622 			__func__, __LINE__);
2623 		return NULL;
2624 	}
2625 
2626 	/*
2627 	 * Ideally we should lookout for reqs bigger than
2628 	 * those that can be programmed with 256 bytes of
2629 	 * MC buffer, but considering a req size is seldom
2630 	 * going to be word-unaligned and more than 200MB,
2631 	 * we take it easy.
2632 	 * Also, should the limit is reached we'd rather
2633 	 * have the platform increase MC buffer size than
2634 	 * complicating this API driver.
2635 	 */
2636 	fill_px(&desc->px, dst, src, len);
2637 
2638 	return desc;
2639 }
2640 
2641 /* Call after fixing burst size */
2642 static inline int get_burst_len(struct dma_pl330_desc *desc, size_t len)
2643 {
2644 	struct dma_pl330_chan *pch = desc->pchan;
2645 	struct pl330_dmac *pl330 = pch->dmac;
2646 	int burst_len;
2647 
2648 	burst_len = pl330->pcfg.data_bus_width / 8;
2649 	burst_len *= pl330->pcfg.data_buf_dep / pl330->pcfg.num_chan;
2650 	burst_len >>= desc->rqcfg.brst_size;
2651 
2652 	/* src/dst_burst_len can't be more than 16 */
2653 	if (burst_len > PL330_MAX_BURST)
2654 		burst_len = PL330_MAX_BURST;
2655 
2656 	return burst_len;
2657 }
2658 
2659 static struct dma_async_tx_descriptor *pl330_prep_dma_cyclic(
2660 		struct dma_chan *chan, dma_addr_t dma_addr, size_t len,
2661 		size_t period_len, enum dma_transfer_direction direction,
2662 		unsigned long flags)
2663 {
2664 	struct dma_pl330_desc *desc = NULL, *first = NULL;
2665 	struct dma_pl330_chan *pch = to_pchan(chan);
2666 	struct pl330_dmac *pl330 = pch->dmac;
2667 	unsigned int i;
2668 	dma_addr_t dst;
2669 	dma_addr_t src;
2670 
2671 	if (len % period_len != 0)
2672 		return NULL;
2673 
2674 	if (!is_slave_direction(direction)) {
2675 		dev_err(pch->dmac->ddma.dev, "%s:%d Invalid dma direction\n",
2676 		__func__, __LINE__);
2677 		return NULL;
2678 	}
2679 
2680 	pl330_config_write(chan, &pch->slave_config, direction);
2681 
2682 	if (!pl330_prep_slave_fifo(pch, direction))
2683 		return NULL;
2684 
2685 	for (i = 0; i < len / period_len; i++) {
2686 		desc = pl330_get_desc(pch);
2687 		if (!desc) {
2688 			dev_err(pch->dmac->ddma.dev, "%s:%d Unable to fetch desc\n",
2689 				__func__, __LINE__);
2690 
2691 			if (!first)
2692 				return NULL;
2693 
2694 			spin_lock_irqsave(&pl330->pool_lock, flags);
2695 
2696 			while (!list_empty(&first->node)) {
2697 				desc = list_entry(first->node.next,
2698 						struct dma_pl330_desc, node);
2699 				list_move_tail(&desc->node, &pl330->desc_pool);
2700 			}
2701 
2702 			list_move_tail(&first->node, &pl330->desc_pool);
2703 
2704 			spin_unlock_irqrestore(&pl330->pool_lock, flags);
2705 
2706 			return NULL;
2707 		}
2708 
2709 		switch (direction) {
2710 		case DMA_MEM_TO_DEV:
2711 			desc->rqcfg.src_inc = 1;
2712 			desc->rqcfg.dst_inc = 0;
2713 			src = dma_addr;
2714 			dst = pch->fifo_dma;
2715 			break;
2716 		case DMA_DEV_TO_MEM:
2717 			desc->rqcfg.src_inc = 0;
2718 			desc->rqcfg.dst_inc = 1;
2719 			src = pch->fifo_dma;
2720 			dst = dma_addr;
2721 			break;
2722 		default:
2723 			break;
2724 		}
2725 
2726 		desc->rqtype = direction;
2727 		desc->rqcfg.brst_size = pch->burst_sz;
2728 		desc->rqcfg.brst_len = pch->burst_len;
2729 		desc->bytes_requested = period_len;
2730 		fill_px(&desc->px, dst, src, period_len);
2731 
2732 		if (!first)
2733 			first = desc;
2734 		else
2735 			list_add_tail(&desc->node, &first->node);
2736 
2737 		dma_addr += period_len;
2738 	}
2739 
2740 	if (!desc)
2741 		return NULL;
2742 
2743 	pch->cyclic = true;
2744 	desc->txd.flags = flags;
2745 
2746 	return &desc->txd;
2747 }
2748 
2749 static struct dma_async_tx_descriptor *
2750 pl330_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dst,
2751 		dma_addr_t src, size_t len, unsigned long flags)
2752 {
2753 	struct dma_pl330_desc *desc;
2754 	struct dma_pl330_chan *pch = to_pchan(chan);
2755 	struct pl330_dmac *pl330;
2756 	int burst;
2757 
2758 	if (unlikely(!pch || !len))
2759 		return NULL;
2760 
2761 	pl330 = pch->dmac;
2762 
2763 	desc = __pl330_prep_dma_memcpy(pch, dst, src, len);
2764 	if (!desc)
2765 		return NULL;
2766 
2767 	desc->rqcfg.src_inc = 1;
2768 	desc->rqcfg.dst_inc = 1;
2769 	desc->rqtype = DMA_MEM_TO_MEM;
2770 
2771 	/* Select max possible burst size */
2772 	burst = pl330->pcfg.data_bus_width / 8;
2773 
2774 	/*
2775 	 * Make sure we use a burst size that aligns with all the memcpy
2776 	 * parameters because our DMA programming algorithm doesn't cope with
2777 	 * transfers which straddle an entry in the DMA device's MFIFO.
2778 	 */
2779 	while ((src | dst | len) & (burst - 1))
2780 		burst /= 2;
2781 
2782 	desc->rqcfg.brst_size = 0;
2783 	while (burst != (1 << desc->rqcfg.brst_size))
2784 		desc->rqcfg.brst_size++;
2785 
2786 	/*
2787 	 * If burst size is smaller than bus width then make sure we only
2788 	 * transfer one at a time to avoid a burst stradling an MFIFO entry.
2789 	 */
2790 	if (desc->rqcfg.brst_size * 8 < pl330->pcfg.data_bus_width)
2791 		desc->rqcfg.brst_len = 1;
2792 
2793 	desc->rqcfg.brst_len = get_burst_len(desc, len);
2794 	desc->bytes_requested = len;
2795 
2796 	desc->txd.flags = flags;
2797 
2798 	return &desc->txd;
2799 }
2800 
2801 static void __pl330_giveback_desc(struct pl330_dmac *pl330,
2802 				  struct dma_pl330_desc *first)
2803 {
2804 	unsigned long flags;
2805 	struct dma_pl330_desc *desc;
2806 
2807 	if (!first)
2808 		return;
2809 
2810 	spin_lock_irqsave(&pl330->pool_lock, flags);
2811 
2812 	while (!list_empty(&first->node)) {
2813 		desc = list_entry(first->node.next,
2814 				struct dma_pl330_desc, node);
2815 		list_move_tail(&desc->node, &pl330->desc_pool);
2816 	}
2817 
2818 	list_move_tail(&first->node, &pl330->desc_pool);
2819 
2820 	spin_unlock_irqrestore(&pl330->pool_lock, flags);
2821 }
2822 
2823 static struct dma_async_tx_descriptor *
2824 pl330_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2825 		unsigned int sg_len, enum dma_transfer_direction direction,
2826 		unsigned long flg, void *context)
2827 {
2828 	struct dma_pl330_desc *first, *desc = NULL;
2829 	struct dma_pl330_chan *pch = to_pchan(chan);
2830 	struct scatterlist *sg;
2831 	int i;
2832 
2833 	if (unlikely(!pch || !sgl || !sg_len))
2834 		return NULL;
2835 
2836 	pl330_config_write(chan, &pch->slave_config, direction);
2837 
2838 	if (!pl330_prep_slave_fifo(pch, direction))
2839 		return NULL;
2840 
2841 	first = NULL;
2842 
2843 	for_each_sg(sgl, sg, sg_len, i) {
2844 
2845 		desc = pl330_get_desc(pch);
2846 		if (!desc) {
2847 			struct pl330_dmac *pl330 = pch->dmac;
2848 
2849 			dev_err(pch->dmac->ddma.dev,
2850 				"%s:%d Unable to fetch desc\n",
2851 				__func__, __LINE__);
2852 			__pl330_giveback_desc(pl330, first);
2853 
2854 			return NULL;
2855 		}
2856 
2857 		if (!first)
2858 			first = desc;
2859 		else
2860 			list_add_tail(&desc->node, &first->node);
2861 
2862 		if (direction == DMA_MEM_TO_DEV) {
2863 			desc->rqcfg.src_inc = 1;
2864 			desc->rqcfg.dst_inc = 0;
2865 			fill_px(&desc->px, pch->fifo_dma, sg_dma_address(sg),
2866 				sg_dma_len(sg));
2867 		} else {
2868 			desc->rqcfg.src_inc = 0;
2869 			desc->rqcfg.dst_inc = 1;
2870 			fill_px(&desc->px, sg_dma_address(sg), pch->fifo_dma,
2871 				sg_dma_len(sg));
2872 		}
2873 
2874 		desc->rqcfg.brst_size = pch->burst_sz;
2875 		desc->rqcfg.brst_len = pch->burst_len;
2876 		desc->rqtype = direction;
2877 		desc->bytes_requested = sg_dma_len(sg);
2878 	}
2879 
2880 	/* Return the last desc in the chain */
2881 	desc->txd.flags = flg;
2882 	return &desc->txd;
2883 }
2884 
2885 static irqreturn_t pl330_irq_handler(int irq, void *data)
2886 {
2887 	if (pl330_update(data))
2888 		return IRQ_HANDLED;
2889 	else
2890 		return IRQ_NONE;
2891 }
2892 
2893 #define PL330_DMA_BUSWIDTHS \
2894 	BIT(DMA_SLAVE_BUSWIDTH_UNDEFINED) | \
2895 	BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
2896 	BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
2897 	BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) | \
2898 	BIT(DMA_SLAVE_BUSWIDTH_8_BYTES)
2899 
2900 #ifdef CONFIG_DEBUG_FS
2901 static int pl330_debugfs_show(struct seq_file *s, void *data)
2902 {
2903 	struct pl330_dmac *pl330 = s->private;
2904 	int chans, pchs, ch, pr;
2905 
2906 	chans = pl330->pcfg.num_chan;
2907 	pchs = pl330->num_peripherals;
2908 
2909 	seq_puts(s, "PL330 physical channels:\n");
2910 	seq_puts(s, "THREAD:\t\tCHANNEL:\n");
2911 	seq_puts(s, "--------\t-----\n");
2912 	for (ch = 0; ch < chans; ch++) {
2913 		struct pl330_thread *thrd = &pl330->channels[ch];
2914 		int found = -1;
2915 
2916 		for (pr = 0; pr < pchs; pr++) {
2917 			struct dma_pl330_chan *pch = &pl330->peripherals[pr];
2918 
2919 			if (!pch->thread || thrd->id != pch->thread->id)
2920 				continue;
2921 
2922 			found = pr;
2923 		}
2924 
2925 		seq_printf(s, "%d\t\t", thrd->id);
2926 		if (found == -1)
2927 			seq_puts(s, "--\n");
2928 		else
2929 			seq_printf(s, "%d\n", found);
2930 	}
2931 
2932 	return 0;
2933 }
2934 
2935 DEFINE_SHOW_ATTRIBUTE(pl330_debugfs);
2936 
2937 static inline void init_pl330_debugfs(struct pl330_dmac *pl330)
2938 {
2939 	debugfs_create_file(dev_name(pl330->ddma.dev),
2940 			    S_IFREG | 0444, NULL, pl330,
2941 			    &pl330_debugfs_fops);
2942 }
2943 #else
2944 static inline void init_pl330_debugfs(struct pl330_dmac *pl330)
2945 {
2946 }
2947 #endif
2948 
2949 /*
2950  * Runtime PM callbacks are provided by amba/bus.c driver.
2951  *
2952  * It is assumed here that IRQ safe runtime PM is chosen in probe and amba
2953  * bus driver will only disable/enable the clock in runtime PM callbacks.
2954  */
2955 static int __maybe_unused pl330_suspend(struct device *dev)
2956 {
2957 	struct amba_device *pcdev = to_amba_device(dev);
2958 
2959 	pm_runtime_disable(dev);
2960 
2961 	if (!pm_runtime_status_suspended(dev)) {
2962 		/* amba did not disable the clock */
2963 		amba_pclk_disable(pcdev);
2964 	}
2965 	amba_pclk_unprepare(pcdev);
2966 
2967 	return 0;
2968 }
2969 
2970 static int __maybe_unused pl330_resume(struct device *dev)
2971 {
2972 	struct amba_device *pcdev = to_amba_device(dev);
2973 	int ret;
2974 
2975 	ret = amba_pclk_prepare(pcdev);
2976 	if (ret)
2977 		return ret;
2978 
2979 	if (!pm_runtime_status_suspended(dev))
2980 		ret = amba_pclk_enable(pcdev);
2981 
2982 	pm_runtime_enable(dev);
2983 
2984 	return ret;
2985 }
2986 
2987 static SIMPLE_DEV_PM_OPS(pl330_pm, pl330_suspend, pl330_resume);
2988 
2989 static int
2990 pl330_probe(struct amba_device *adev, const struct amba_id *id)
2991 {
2992 	struct pl330_config *pcfg;
2993 	struct pl330_dmac *pl330;
2994 	struct dma_pl330_chan *pch, *_p;
2995 	struct dma_device *pd;
2996 	struct resource *res;
2997 	int i, ret, irq;
2998 	int num_chan;
2999 	struct device_node *np = adev->dev.of_node;
3000 
3001 	ret = dma_set_mask_and_coherent(&adev->dev, DMA_BIT_MASK(32));
3002 	if (ret)
3003 		return ret;
3004 
3005 	/* Allocate a new DMAC and its Channels */
3006 	pl330 = devm_kzalloc(&adev->dev, sizeof(*pl330), GFP_KERNEL);
3007 	if (!pl330)
3008 		return -ENOMEM;
3009 
3010 	pd = &pl330->ddma;
3011 	pd->dev = &adev->dev;
3012 
3013 	pl330->mcbufsz = 0;
3014 
3015 	/* get quirk */
3016 	for (i = 0; i < ARRAY_SIZE(of_quirks); i++)
3017 		if (of_property_read_bool(np, of_quirks[i].quirk))
3018 			pl330->quirks |= of_quirks[i].id;
3019 
3020 	res = &adev->res;
3021 	pl330->base = devm_ioremap_resource(&adev->dev, res);
3022 	if (IS_ERR(pl330->base))
3023 		return PTR_ERR(pl330->base);
3024 
3025 	amba_set_drvdata(adev, pl330);
3026 
3027 	for (i = 0; i < AMBA_NR_IRQS; i++) {
3028 		irq = adev->irq[i];
3029 		if (irq) {
3030 			ret = devm_request_irq(&adev->dev, irq,
3031 					       pl330_irq_handler, 0,
3032 					       dev_name(&adev->dev), pl330);
3033 			if (ret)
3034 				return ret;
3035 		} else {
3036 			break;
3037 		}
3038 	}
3039 
3040 	pcfg = &pl330->pcfg;
3041 
3042 	pcfg->periph_id = adev->periphid;
3043 	ret = pl330_add(pl330);
3044 	if (ret)
3045 		return ret;
3046 
3047 	INIT_LIST_HEAD(&pl330->desc_pool);
3048 	spin_lock_init(&pl330->pool_lock);
3049 
3050 	/* Create a descriptor pool of default size */
3051 	if (!add_desc(&pl330->desc_pool, &pl330->pool_lock,
3052 		      GFP_KERNEL, NR_DEFAULT_DESC))
3053 		dev_warn(&adev->dev, "unable to allocate desc\n");
3054 
3055 	INIT_LIST_HEAD(&pd->channels);
3056 
3057 	/* Initialize channel parameters */
3058 	num_chan = max_t(int, pcfg->num_peri, pcfg->num_chan);
3059 
3060 	pl330->num_peripherals = num_chan;
3061 
3062 	pl330->peripherals = kcalloc(num_chan, sizeof(*pch), GFP_KERNEL);
3063 	if (!pl330->peripherals) {
3064 		ret = -ENOMEM;
3065 		goto probe_err2;
3066 	}
3067 
3068 	for (i = 0; i < num_chan; i++) {
3069 		pch = &pl330->peripherals[i];
3070 
3071 		pch->chan.private = adev->dev.of_node;
3072 		INIT_LIST_HEAD(&pch->submitted_list);
3073 		INIT_LIST_HEAD(&pch->work_list);
3074 		INIT_LIST_HEAD(&pch->completed_list);
3075 		spin_lock_init(&pch->lock);
3076 		pch->thread = NULL;
3077 		pch->chan.device = pd;
3078 		pch->dmac = pl330;
3079 		pch->dir = DMA_NONE;
3080 
3081 		/* Add the channel to the DMAC list */
3082 		list_add_tail(&pch->chan.device_node, &pd->channels);
3083 	}
3084 
3085 	dma_cap_set(DMA_MEMCPY, pd->cap_mask);
3086 	if (pcfg->num_peri) {
3087 		dma_cap_set(DMA_SLAVE, pd->cap_mask);
3088 		dma_cap_set(DMA_CYCLIC, pd->cap_mask);
3089 		dma_cap_set(DMA_PRIVATE, pd->cap_mask);
3090 	}
3091 
3092 	pd->device_alloc_chan_resources = pl330_alloc_chan_resources;
3093 	pd->device_free_chan_resources = pl330_free_chan_resources;
3094 	pd->device_prep_dma_memcpy = pl330_prep_dma_memcpy;
3095 	pd->device_prep_dma_cyclic = pl330_prep_dma_cyclic;
3096 	pd->device_tx_status = pl330_tx_status;
3097 	pd->device_prep_slave_sg = pl330_prep_slave_sg;
3098 	pd->device_config = pl330_config;
3099 	pd->device_pause = pl330_pause;
3100 	pd->device_terminate_all = pl330_terminate_all;
3101 	pd->device_issue_pending = pl330_issue_pending;
3102 	pd->src_addr_widths = PL330_DMA_BUSWIDTHS;
3103 	pd->dst_addr_widths = PL330_DMA_BUSWIDTHS;
3104 	pd->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
3105 	pd->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
3106 	pd->max_burst = ((pl330->quirks & PL330_QUIRK_BROKEN_NO_FLUSHP) ?
3107 			 1 : PL330_MAX_BURST);
3108 
3109 	ret = dma_async_device_register(pd);
3110 	if (ret) {
3111 		dev_err(&adev->dev, "unable to register DMAC\n");
3112 		goto probe_err3;
3113 	}
3114 
3115 	if (adev->dev.of_node) {
3116 		ret = of_dma_controller_register(adev->dev.of_node,
3117 					 of_dma_pl330_xlate, pl330);
3118 		if (ret) {
3119 			dev_err(&adev->dev,
3120 			"unable to register DMA to the generic DT DMA helpers\n");
3121 		}
3122 	}
3123 
3124 	adev->dev.dma_parms = &pl330->dma_parms;
3125 
3126 	/*
3127 	 * This is the limit for transfers with a buswidth of 1, larger
3128 	 * buswidths will have larger limits.
3129 	 */
3130 	ret = dma_set_max_seg_size(&adev->dev, 1900800);
3131 	if (ret)
3132 		dev_err(&adev->dev, "unable to set the seg size\n");
3133 
3134 
3135 	init_pl330_debugfs(pl330);
3136 	dev_info(&adev->dev,
3137 		"Loaded driver for PL330 DMAC-%x\n", adev->periphid);
3138 	dev_info(&adev->dev,
3139 		"\tDBUFF-%ux%ubytes Num_Chans-%u Num_Peri-%u Num_Events-%u\n",
3140 		pcfg->data_buf_dep, pcfg->data_bus_width / 8, pcfg->num_chan,
3141 		pcfg->num_peri, pcfg->num_events);
3142 
3143 	pm_runtime_irq_safe(&adev->dev);
3144 	pm_runtime_use_autosuspend(&adev->dev);
3145 	pm_runtime_set_autosuspend_delay(&adev->dev, PL330_AUTOSUSPEND_DELAY);
3146 	pm_runtime_mark_last_busy(&adev->dev);
3147 	pm_runtime_put_autosuspend(&adev->dev);
3148 
3149 	return 0;
3150 probe_err3:
3151 	/* Idle the DMAC */
3152 	list_for_each_entry_safe(pch, _p, &pl330->ddma.channels,
3153 			chan.device_node) {
3154 
3155 		/* Remove the channel */
3156 		list_del(&pch->chan.device_node);
3157 
3158 		/* Flush the channel */
3159 		if (pch->thread) {
3160 			pl330_terminate_all(&pch->chan);
3161 			pl330_free_chan_resources(&pch->chan);
3162 		}
3163 	}
3164 probe_err2:
3165 	pl330_del(pl330);
3166 
3167 	return ret;
3168 }
3169 
3170 static int pl330_remove(struct amba_device *adev)
3171 {
3172 	struct pl330_dmac *pl330 = amba_get_drvdata(adev);
3173 	struct dma_pl330_chan *pch, *_p;
3174 	int i, irq;
3175 
3176 	pm_runtime_get_noresume(pl330->ddma.dev);
3177 
3178 	if (adev->dev.of_node)
3179 		of_dma_controller_free(adev->dev.of_node);
3180 
3181 	for (i = 0; i < AMBA_NR_IRQS; i++) {
3182 		irq = adev->irq[i];
3183 		if (irq)
3184 			devm_free_irq(&adev->dev, irq, pl330);
3185 	}
3186 
3187 	dma_async_device_unregister(&pl330->ddma);
3188 
3189 	/* Idle the DMAC */
3190 	list_for_each_entry_safe(pch, _p, &pl330->ddma.channels,
3191 			chan.device_node) {
3192 
3193 		/* Remove the channel */
3194 		list_del(&pch->chan.device_node);
3195 
3196 		/* Flush the channel */
3197 		if (pch->thread) {
3198 			pl330_terminate_all(&pch->chan);
3199 			pl330_free_chan_resources(&pch->chan);
3200 		}
3201 	}
3202 
3203 	pl330_del(pl330);
3204 
3205 	return 0;
3206 }
3207 
3208 static const struct amba_id pl330_ids[] = {
3209 	{
3210 		.id	= 0x00041330,
3211 		.mask	= 0x000fffff,
3212 	},
3213 	{ 0, 0 },
3214 };
3215 
3216 MODULE_DEVICE_TABLE(amba, pl330_ids);
3217 
3218 static struct amba_driver pl330_driver = {
3219 	.drv = {
3220 		.owner = THIS_MODULE,
3221 		.name = "dma-pl330",
3222 		.pm = &pl330_pm,
3223 	},
3224 	.id_table = pl330_ids,
3225 	.probe = pl330_probe,
3226 	.remove = pl330_remove,
3227 };
3228 
3229 module_amba_driver(pl330_driver);
3230 
3231 MODULE_AUTHOR("Jaswinder Singh <jassisinghbrar@gmail.com>");
3232 MODULE_DESCRIPTION("API Driver for PL330 DMAC");
3233 MODULE_LICENSE("GPL");
3234