xref: /linux/drivers/dma/pl330.c (revision 7ff836f064e2c814a7504c91a4464eea45d475bd)
1 /*
2  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
3  *		http://www.samsung.com
4  *
5  * Copyright (C) 2010 Samsung Electronics Co. Ltd.
6  *	Jaswinder Singh <jassi.brar@samsung.com>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  */
13 
14 #include <linux/debugfs.h>
15 #include <linux/kernel.h>
16 #include <linux/io.h>
17 #include <linux/init.h>
18 #include <linux/slab.h>
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/delay.h>
22 #include <linux/interrupt.h>
23 #include <linux/dma-mapping.h>
24 #include <linux/dmaengine.h>
25 #include <linux/amba/bus.h>
26 #include <linux/scatterlist.h>
27 #include <linux/of.h>
28 #include <linux/of_dma.h>
29 #include <linux/err.h>
30 #include <linux/pm_runtime.h>
31 #include <linux/bug.h>
32 
33 #include "dmaengine.h"
34 #define PL330_MAX_CHAN		8
35 #define PL330_MAX_IRQS		32
36 #define PL330_MAX_PERI		32
37 #define PL330_MAX_BURST         16
38 
39 #define PL330_QUIRK_BROKEN_NO_FLUSHP BIT(0)
40 
41 enum pl330_cachectrl {
42 	CCTRL0,		/* Noncacheable and nonbufferable */
43 	CCTRL1,		/* Bufferable only */
44 	CCTRL2,		/* Cacheable, but do not allocate */
45 	CCTRL3,		/* Cacheable and bufferable, but do not allocate */
46 	INVALID1,	/* AWCACHE = 0x1000 */
47 	INVALID2,
48 	CCTRL6,		/* Cacheable write-through, allocate on writes only */
49 	CCTRL7,		/* Cacheable write-back, allocate on writes only */
50 };
51 
52 enum pl330_byteswap {
53 	SWAP_NO,
54 	SWAP_2,
55 	SWAP_4,
56 	SWAP_8,
57 	SWAP_16,
58 };
59 
60 /* Register and Bit field Definitions */
61 #define DS			0x0
62 #define DS_ST_STOP		0x0
63 #define DS_ST_EXEC		0x1
64 #define DS_ST_CMISS		0x2
65 #define DS_ST_UPDTPC		0x3
66 #define DS_ST_WFE		0x4
67 #define DS_ST_ATBRR		0x5
68 #define DS_ST_QBUSY		0x6
69 #define DS_ST_WFP		0x7
70 #define DS_ST_KILL		0x8
71 #define DS_ST_CMPLT		0x9
72 #define DS_ST_FLTCMP		0xe
73 #define DS_ST_FAULT		0xf
74 
75 #define DPC			0x4
76 #define INTEN			0x20
77 #define ES			0x24
78 #define INTSTATUS		0x28
79 #define INTCLR			0x2c
80 #define FSM			0x30
81 #define FSC			0x34
82 #define FTM			0x38
83 
84 #define _FTC			0x40
85 #define FTC(n)			(_FTC + (n)*0x4)
86 
87 #define _CS			0x100
88 #define CS(n)			(_CS + (n)*0x8)
89 #define CS_CNS			(1 << 21)
90 
91 #define _CPC			0x104
92 #define CPC(n)			(_CPC + (n)*0x8)
93 
94 #define _SA			0x400
95 #define SA(n)			(_SA + (n)*0x20)
96 
97 #define _DA			0x404
98 #define DA(n)			(_DA + (n)*0x20)
99 
100 #define _CC			0x408
101 #define CC(n)			(_CC + (n)*0x20)
102 
103 #define CC_SRCINC		(1 << 0)
104 #define CC_DSTINC		(1 << 14)
105 #define CC_SRCPRI		(1 << 8)
106 #define CC_DSTPRI		(1 << 22)
107 #define CC_SRCNS		(1 << 9)
108 #define CC_DSTNS		(1 << 23)
109 #define CC_SRCIA		(1 << 10)
110 #define CC_DSTIA		(1 << 24)
111 #define CC_SRCBRSTLEN_SHFT	4
112 #define CC_DSTBRSTLEN_SHFT	18
113 #define CC_SRCBRSTSIZE_SHFT	1
114 #define CC_DSTBRSTSIZE_SHFT	15
115 #define CC_SRCCCTRL_SHFT	11
116 #define CC_SRCCCTRL_MASK	0x7
117 #define CC_DSTCCTRL_SHFT	25
118 #define CC_DRCCCTRL_MASK	0x7
119 #define CC_SWAP_SHFT		28
120 
121 #define _LC0			0x40c
122 #define LC0(n)			(_LC0 + (n)*0x20)
123 
124 #define _LC1			0x410
125 #define LC1(n)			(_LC1 + (n)*0x20)
126 
127 #define DBGSTATUS		0xd00
128 #define DBG_BUSY		(1 << 0)
129 
130 #define DBGCMD			0xd04
131 #define DBGINST0		0xd08
132 #define DBGINST1		0xd0c
133 
134 #define CR0			0xe00
135 #define CR1			0xe04
136 #define CR2			0xe08
137 #define CR3			0xe0c
138 #define CR4			0xe10
139 #define CRD			0xe14
140 
141 #define PERIPH_ID		0xfe0
142 #define PERIPH_REV_SHIFT	20
143 #define PERIPH_REV_MASK		0xf
144 #define PERIPH_REV_R0P0		0
145 #define PERIPH_REV_R1P0		1
146 #define PERIPH_REV_R1P1		2
147 
148 #define CR0_PERIPH_REQ_SET	(1 << 0)
149 #define CR0_BOOT_EN_SET		(1 << 1)
150 #define CR0_BOOT_MAN_NS		(1 << 2)
151 #define CR0_NUM_CHANS_SHIFT	4
152 #define CR0_NUM_CHANS_MASK	0x7
153 #define CR0_NUM_PERIPH_SHIFT	12
154 #define CR0_NUM_PERIPH_MASK	0x1f
155 #define CR0_NUM_EVENTS_SHIFT	17
156 #define CR0_NUM_EVENTS_MASK	0x1f
157 
158 #define CR1_ICACHE_LEN_SHIFT	0
159 #define CR1_ICACHE_LEN_MASK	0x7
160 #define CR1_NUM_ICACHELINES_SHIFT	4
161 #define CR1_NUM_ICACHELINES_MASK	0xf
162 
163 #define CRD_DATA_WIDTH_SHIFT	0
164 #define CRD_DATA_WIDTH_MASK	0x7
165 #define CRD_WR_CAP_SHIFT	4
166 #define CRD_WR_CAP_MASK		0x7
167 #define CRD_WR_Q_DEP_SHIFT	8
168 #define CRD_WR_Q_DEP_MASK	0xf
169 #define CRD_RD_CAP_SHIFT	12
170 #define CRD_RD_CAP_MASK		0x7
171 #define CRD_RD_Q_DEP_SHIFT	16
172 #define CRD_RD_Q_DEP_MASK	0xf
173 #define CRD_DATA_BUFF_SHIFT	20
174 #define CRD_DATA_BUFF_MASK	0x3ff
175 
176 #define PART			0x330
177 #define DESIGNER		0x41
178 #define REVISION		0x0
179 #define INTEG_CFG		0x0
180 #define PERIPH_ID_VAL		((PART << 0) | (DESIGNER << 12))
181 
182 #define PL330_STATE_STOPPED		(1 << 0)
183 #define PL330_STATE_EXECUTING		(1 << 1)
184 #define PL330_STATE_WFE			(1 << 2)
185 #define PL330_STATE_FAULTING		(1 << 3)
186 #define PL330_STATE_COMPLETING		(1 << 4)
187 #define PL330_STATE_WFP			(1 << 5)
188 #define PL330_STATE_KILLING		(1 << 6)
189 #define PL330_STATE_FAULT_COMPLETING	(1 << 7)
190 #define PL330_STATE_CACHEMISS		(1 << 8)
191 #define PL330_STATE_UPDTPC		(1 << 9)
192 #define PL330_STATE_ATBARRIER		(1 << 10)
193 #define PL330_STATE_QUEUEBUSY		(1 << 11)
194 #define PL330_STATE_INVALID		(1 << 15)
195 
196 #define PL330_STABLE_STATES (PL330_STATE_STOPPED | PL330_STATE_EXECUTING \
197 				| PL330_STATE_WFE | PL330_STATE_FAULTING)
198 
199 #define CMD_DMAADDH		0x54
200 #define CMD_DMAEND		0x00
201 #define CMD_DMAFLUSHP		0x35
202 #define CMD_DMAGO		0xa0
203 #define CMD_DMALD		0x04
204 #define CMD_DMALDP		0x25
205 #define CMD_DMALP		0x20
206 #define CMD_DMALPEND		0x28
207 #define CMD_DMAKILL		0x01
208 #define CMD_DMAMOV		0xbc
209 #define CMD_DMANOP		0x18
210 #define CMD_DMARMB		0x12
211 #define CMD_DMASEV		0x34
212 #define CMD_DMAST		0x08
213 #define CMD_DMASTP		0x29
214 #define CMD_DMASTZ		0x0c
215 #define CMD_DMAWFE		0x36
216 #define CMD_DMAWFP		0x30
217 #define CMD_DMAWMB		0x13
218 
219 #define SZ_DMAADDH		3
220 #define SZ_DMAEND		1
221 #define SZ_DMAFLUSHP		2
222 #define SZ_DMALD		1
223 #define SZ_DMALDP		2
224 #define SZ_DMALP		2
225 #define SZ_DMALPEND		2
226 #define SZ_DMAKILL		1
227 #define SZ_DMAMOV		6
228 #define SZ_DMANOP		1
229 #define SZ_DMARMB		1
230 #define SZ_DMASEV		2
231 #define SZ_DMAST		1
232 #define SZ_DMASTP		2
233 #define SZ_DMASTZ		1
234 #define SZ_DMAWFE		2
235 #define SZ_DMAWFP		2
236 #define SZ_DMAWMB		1
237 #define SZ_DMAGO		6
238 
239 #define BRST_LEN(ccr)		((((ccr) >> CC_SRCBRSTLEN_SHFT) & 0xf) + 1)
240 #define BRST_SIZE(ccr)		(1 << (((ccr) >> CC_SRCBRSTSIZE_SHFT) & 0x7))
241 
242 #define BYTE_TO_BURST(b, ccr)	((b) / BRST_SIZE(ccr) / BRST_LEN(ccr))
243 #define BURST_TO_BYTE(c, ccr)	((c) * BRST_SIZE(ccr) * BRST_LEN(ccr))
244 
245 /*
246  * With 256 bytes, we can do more than 2.5MB and 5MB xfers per req
247  * at 1byte/burst for P<->M and M<->M respectively.
248  * For typical scenario, at 1word/burst, 10MB and 20MB xfers per req
249  * should be enough for P<->M and M<->M respectively.
250  */
251 #define MCODE_BUFF_PER_REQ	256
252 
253 /* Use this _only_ to wait on transient states */
254 #define UNTIL(t, s)	while (!(_state(t) & (s))) cpu_relax();
255 
256 #ifdef PL330_DEBUG_MCGEN
257 static unsigned cmd_line;
258 #define PL330_DBGCMD_DUMP(off, x...)	do { \
259 						printk("%x:", cmd_line); \
260 						printk(x); \
261 						cmd_line += off; \
262 					} while (0)
263 #define PL330_DBGMC_START(addr)		(cmd_line = addr)
264 #else
265 #define PL330_DBGCMD_DUMP(off, x...)	do {} while (0)
266 #define PL330_DBGMC_START(addr)		do {} while (0)
267 #endif
268 
269 /* The number of default descriptors */
270 
271 #define NR_DEFAULT_DESC	16
272 
273 /* Delay for runtime PM autosuspend, ms */
274 #define PL330_AUTOSUSPEND_DELAY 20
275 
276 /* Populated by the PL330 core driver for DMA API driver's info */
277 struct pl330_config {
278 	u32	periph_id;
279 #define DMAC_MODE_NS	(1 << 0)
280 	unsigned int	mode;
281 	unsigned int	data_bus_width:10; /* In number of bits */
282 	unsigned int	data_buf_dep:11;
283 	unsigned int	num_chan:4;
284 	unsigned int	num_peri:6;
285 	u32		peri_ns;
286 	unsigned int	num_events:6;
287 	u32		irq_ns;
288 };
289 
290 /**
291  * Request Configuration.
292  * The PL330 core does not modify this and uses the last
293  * working configuration if the request doesn't provide any.
294  *
295  * The Client may want to provide this info only for the
296  * first request and a request with new settings.
297  */
298 struct pl330_reqcfg {
299 	/* Address Incrementing */
300 	unsigned dst_inc:1;
301 	unsigned src_inc:1;
302 
303 	/*
304 	 * For now, the SRC & DST protection levels
305 	 * and burst size/length are assumed same.
306 	 */
307 	bool nonsecure;
308 	bool privileged;
309 	bool insnaccess;
310 	unsigned brst_len:5;
311 	unsigned brst_size:3; /* in power of 2 */
312 
313 	enum pl330_cachectrl dcctl;
314 	enum pl330_cachectrl scctl;
315 	enum pl330_byteswap swap;
316 	struct pl330_config *pcfg;
317 };
318 
319 /*
320  * One cycle of DMAC operation.
321  * There may be more than one xfer in a request.
322  */
323 struct pl330_xfer {
324 	u32 src_addr;
325 	u32 dst_addr;
326 	/* Size to xfer */
327 	u32 bytes;
328 };
329 
330 /* The xfer callbacks are made with one of these arguments. */
331 enum pl330_op_err {
332 	/* The all xfers in the request were success. */
333 	PL330_ERR_NONE,
334 	/* If req aborted due to global error. */
335 	PL330_ERR_ABORT,
336 	/* If req failed due to problem with Channel. */
337 	PL330_ERR_FAIL,
338 };
339 
340 enum dmamov_dst {
341 	SAR = 0,
342 	CCR,
343 	DAR,
344 };
345 
346 enum pl330_dst {
347 	SRC = 0,
348 	DST,
349 };
350 
351 enum pl330_cond {
352 	SINGLE,
353 	BURST,
354 	ALWAYS,
355 };
356 
357 struct dma_pl330_desc;
358 
359 struct _pl330_req {
360 	u32 mc_bus;
361 	void *mc_cpu;
362 	struct dma_pl330_desc *desc;
363 };
364 
365 /* ToBeDone for tasklet */
366 struct _pl330_tbd {
367 	bool reset_dmac;
368 	bool reset_mngr;
369 	u8 reset_chan;
370 };
371 
372 /* A DMAC Thread */
373 struct pl330_thread {
374 	u8 id;
375 	int ev;
376 	/* If the channel is not yet acquired by any client */
377 	bool free;
378 	/* Parent DMAC */
379 	struct pl330_dmac *dmac;
380 	/* Only two at a time */
381 	struct _pl330_req req[2];
382 	/* Index of the last enqueued request */
383 	unsigned lstenq;
384 	/* Index of the last submitted request or -1 if the DMA is stopped */
385 	int req_running;
386 };
387 
388 enum pl330_dmac_state {
389 	UNINIT,
390 	INIT,
391 	DYING,
392 };
393 
394 enum desc_status {
395 	/* In the DMAC pool */
396 	FREE,
397 	/*
398 	 * Allocated to some channel during prep_xxx
399 	 * Also may be sitting on the work_list.
400 	 */
401 	PREP,
402 	/*
403 	 * Sitting on the work_list and already submitted
404 	 * to the PL330 core. Not more than two descriptors
405 	 * of a channel can be BUSY at any time.
406 	 */
407 	BUSY,
408 	/*
409 	 * Sitting on the channel work_list but xfer done
410 	 * by PL330 core
411 	 */
412 	DONE,
413 };
414 
415 struct dma_pl330_chan {
416 	/* Schedule desc completion */
417 	struct tasklet_struct task;
418 
419 	/* DMA-Engine Channel */
420 	struct dma_chan chan;
421 
422 	/* List of submitted descriptors */
423 	struct list_head submitted_list;
424 	/* List of issued descriptors */
425 	struct list_head work_list;
426 	/* List of completed descriptors */
427 	struct list_head completed_list;
428 
429 	/* Pointer to the DMAC that manages this channel,
430 	 * NULL if the channel is available to be acquired.
431 	 * As the parent, this DMAC also provides descriptors
432 	 * to the channel.
433 	 */
434 	struct pl330_dmac *dmac;
435 
436 	/* To protect channel manipulation */
437 	spinlock_t lock;
438 
439 	/*
440 	 * Hardware channel thread of PL330 DMAC. NULL if the channel is
441 	 * available.
442 	 */
443 	struct pl330_thread *thread;
444 
445 	/* For D-to-M and M-to-D channels */
446 	int burst_sz; /* the peripheral fifo width */
447 	int burst_len; /* the number of burst */
448 	phys_addr_t fifo_addr;
449 	/* DMA-mapped view of the FIFO; may differ if an IOMMU is present */
450 	dma_addr_t fifo_dma;
451 	enum dma_data_direction dir;
452 	struct dma_slave_config slave_config;
453 
454 	/* for cyclic capability */
455 	bool cyclic;
456 
457 	/* for runtime pm tracking */
458 	bool active;
459 };
460 
461 struct pl330_dmac {
462 	/* DMA-Engine Device */
463 	struct dma_device ddma;
464 
465 	/* Holds info about sg limitations */
466 	struct device_dma_parameters dma_parms;
467 
468 	/* Pool of descriptors available for the DMAC's channels */
469 	struct list_head desc_pool;
470 	/* To protect desc_pool manipulation */
471 	spinlock_t pool_lock;
472 
473 	/* Size of MicroCode buffers for each channel. */
474 	unsigned mcbufsz;
475 	/* ioremap'ed address of PL330 registers. */
476 	void __iomem	*base;
477 	/* Populated by the PL330 core driver during pl330_add */
478 	struct pl330_config	pcfg;
479 
480 	spinlock_t		lock;
481 	/* Maximum possible events/irqs */
482 	int			events[32];
483 	/* BUS address of MicroCode buffer */
484 	dma_addr_t		mcode_bus;
485 	/* CPU address of MicroCode buffer */
486 	void			*mcode_cpu;
487 	/* List of all Channel threads */
488 	struct pl330_thread	*channels;
489 	/* Pointer to the MANAGER thread */
490 	struct pl330_thread	*manager;
491 	/* To handle bad news in interrupt */
492 	struct tasklet_struct	tasks;
493 	struct _pl330_tbd	dmac_tbd;
494 	/* State of DMAC operation */
495 	enum pl330_dmac_state	state;
496 	/* Holds list of reqs with due callbacks */
497 	struct list_head        req_done;
498 
499 	/* Peripheral channels connected to this DMAC */
500 	unsigned int num_peripherals;
501 	struct dma_pl330_chan *peripherals; /* keep at end */
502 	int quirks;
503 };
504 
505 static struct pl330_of_quirks {
506 	char *quirk;
507 	int id;
508 } of_quirks[] = {
509 	{
510 		.quirk = "arm,pl330-broken-no-flushp",
511 		.id = PL330_QUIRK_BROKEN_NO_FLUSHP,
512 	}
513 };
514 
515 struct dma_pl330_desc {
516 	/* To attach to a queue as child */
517 	struct list_head node;
518 
519 	/* Descriptor for the DMA Engine API */
520 	struct dma_async_tx_descriptor txd;
521 
522 	/* Xfer for PL330 core */
523 	struct pl330_xfer px;
524 
525 	struct pl330_reqcfg rqcfg;
526 
527 	enum desc_status status;
528 
529 	int bytes_requested;
530 	bool last;
531 
532 	/* The channel which currently holds this desc */
533 	struct dma_pl330_chan *pchan;
534 
535 	enum dma_transfer_direction rqtype;
536 	/* Index of peripheral for the xfer. */
537 	unsigned peri:5;
538 	/* Hook to attach to DMAC's list of reqs with due callback */
539 	struct list_head rqd;
540 };
541 
542 struct _xfer_spec {
543 	u32 ccr;
544 	struct dma_pl330_desc *desc;
545 };
546 
547 static int pl330_config_write(struct dma_chan *chan,
548 			struct dma_slave_config *slave_config,
549 			enum dma_transfer_direction direction);
550 
551 static inline bool _queue_full(struct pl330_thread *thrd)
552 {
553 	return thrd->req[0].desc != NULL && thrd->req[1].desc != NULL;
554 }
555 
556 static inline bool is_manager(struct pl330_thread *thrd)
557 {
558 	return thrd->dmac->manager == thrd;
559 }
560 
561 /* If manager of the thread is in Non-Secure mode */
562 static inline bool _manager_ns(struct pl330_thread *thrd)
563 {
564 	return (thrd->dmac->pcfg.mode & DMAC_MODE_NS) ? true : false;
565 }
566 
567 static inline u32 get_revision(u32 periph_id)
568 {
569 	return (periph_id >> PERIPH_REV_SHIFT) & PERIPH_REV_MASK;
570 }
571 
572 static inline u32 _emit_END(unsigned dry_run, u8 buf[])
573 {
574 	if (dry_run)
575 		return SZ_DMAEND;
576 
577 	buf[0] = CMD_DMAEND;
578 
579 	PL330_DBGCMD_DUMP(SZ_DMAEND, "\tDMAEND\n");
580 
581 	return SZ_DMAEND;
582 }
583 
584 static inline u32 _emit_FLUSHP(unsigned dry_run, u8 buf[], u8 peri)
585 {
586 	if (dry_run)
587 		return SZ_DMAFLUSHP;
588 
589 	buf[0] = CMD_DMAFLUSHP;
590 
591 	peri &= 0x1f;
592 	peri <<= 3;
593 	buf[1] = peri;
594 
595 	PL330_DBGCMD_DUMP(SZ_DMAFLUSHP, "\tDMAFLUSHP %u\n", peri >> 3);
596 
597 	return SZ_DMAFLUSHP;
598 }
599 
600 static inline u32 _emit_LD(unsigned dry_run, u8 buf[],	enum pl330_cond cond)
601 {
602 	if (dry_run)
603 		return SZ_DMALD;
604 
605 	buf[0] = CMD_DMALD;
606 
607 	if (cond == SINGLE)
608 		buf[0] |= (0 << 1) | (1 << 0);
609 	else if (cond == BURST)
610 		buf[0] |= (1 << 1) | (1 << 0);
611 
612 	PL330_DBGCMD_DUMP(SZ_DMALD, "\tDMALD%c\n",
613 		cond == SINGLE ? 'S' : (cond == BURST ? 'B' : 'A'));
614 
615 	return SZ_DMALD;
616 }
617 
618 static inline u32 _emit_LDP(unsigned dry_run, u8 buf[],
619 		enum pl330_cond cond, u8 peri)
620 {
621 	if (dry_run)
622 		return SZ_DMALDP;
623 
624 	buf[0] = CMD_DMALDP;
625 
626 	if (cond == BURST)
627 		buf[0] |= (1 << 1);
628 
629 	peri &= 0x1f;
630 	peri <<= 3;
631 	buf[1] = peri;
632 
633 	PL330_DBGCMD_DUMP(SZ_DMALDP, "\tDMALDP%c %u\n",
634 		cond == SINGLE ? 'S' : 'B', peri >> 3);
635 
636 	return SZ_DMALDP;
637 }
638 
639 static inline u32 _emit_LP(unsigned dry_run, u8 buf[],
640 		unsigned loop, u8 cnt)
641 {
642 	if (dry_run)
643 		return SZ_DMALP;
644 
645 	buf[0] = CMD_DMALP;
646 
647 	if (loop)
648 		buf[0] |= (1 << 1);
649 
650 	cnt--; /* DMAC increments by 1 internally */
651 	buf[1] = cnt;
652 
653 	PL330_DBGCMD_DUMP(SZ_DMALP, "\tDMALP_%c %u\n", loop ? '1' : '0', cnt);
654 
655 	return SZ_DMALP;
656 }
657 
658 struct _arg_LPEND {
659 	enum pl330_cond cond;
660 	bool forever;
661 	unsigned loop;
662 	u8 bjump;
663 };
664 
665 static inline u32 _emit_LPEND(unsigned dry_run, u8 buf[],
666 		const struct _arg_LPEND *arg)
667 {
668 	enum pl330_cond cond = arg->cond;
669 	bool forever = arg->forever;
670 	unsigned loop = arg->loop;
671 	u8 bjump = arg->bjump;
672 
673 	if (dry_run)
674 		return SZ_DMALPEND;
675 
676 	buf[0] = CMD_DMALPEND;
677 
678 	if (loop)
679 		buf[0] |= (1 << 2);
680 
681 	if (!forever)
682 		buf[0] |= (1 << 4);
683 
684 	if (cond == SINGLE)
685 		buf[0] |= (0 << 1) | (1 << 0);
686 	else if (cond == BURST)
687 		buf[0] |= (1 << 1) | (1 << 0);
688 
689 	buf[1] = bjump;
690 
691 	PL330_DBGCMD_DUMP(SZ_DMALPEND, "\tDMALP%s%c_%c bjmpto_%x\n",
692 			forever ? "FE" : "END",
693 			cond == SINGLE ? 'S' : (cond == BURST ? 'B' : 'A'),
694 			loop ? '1' : '0',
695 			bjump);
696 
697 	return SZ_DMALPEND;
698 }
699 
700 static inline u32 _emit_KILL(unsigned dry_run, u8 buf[])
701 {
702 	if (dry_run)
703 		return SZ_DMAKILL;
704 
705 	buf[0] = CMD_DMAKILL;
706 
707 	return SZ_DMAKILL;
708 }
709 
710 static inline u32 _emit_MOV(unsigned dry_run, u8 buf[],
711 		enum dmamov_dst dst, u32 val)
712 {
713 	if (dry_run)
714 		return SZ_DMAMOV;
715 
716 	buf[0] = CMD_DMAMOV;
717 	buf[1] = dst;
718 	buf[2] = val;
719 	buf[3] = val >> 8;
720 	buf[4] = val >> 16;
721 	buf[5] = val >> 24;
722 
723 	PL330_DBGCMD_DUMP(SZ_DMAMOV, "\tDMAMOV %s 0x%x\n",
724 		dst == SAR ? "SAR" : (dst == DAR ? "DAR" : "CCR"), val);
725 
726 	return SZ_DMAMOV;
727 }
728 
729 static inline u32 _emit_RMB(unsigned dry_run, u8 buf[])
730 {
731 	if (dry_run)
732 		return SZ_DMARMB;
733 
734 	buf[0] = CMD_DMARMB;
735 
736 	PL330_DBGCMD_DUMP(SZ_DMARMB, "\tDMARMB\n");
737 
738 	return SZ_DMARMB;
739 }
740 
741 static inline u32 _emit_SEV(unsigned dry_run, u8 buf[], u8 ev)
742 {
743 	if (dry_run)
744 		return SZ_DMASEV;
745 
746 	buf[0] = CMD_DMASEV;
747 
748 	ev &= 0x1f;
749 	ev <<= 3;
750 	buf[1] = ev;
751 
752 	PL330_DBGCMD_DUMP(SZ_DMASEV, "\tDMASEV %u\n", ev >> 3);
753 
754 	return SZ_DMASEV;
755 }
756 
757 static inline u32 _emit_ST(unsigned dry_run, u8 buf[], enum pl330_cond cond)
758 {
759 	if (dry_run)
760 		return SZ_DMAST;
761 
762 	buf[0] = CMD_DMAST;
763 
764 	if (cond == SINGLE)
765 		buf[0] |= (0 << 1) | (1 << 0);
766 	else if (cond == BURST)
767 		buf[0] |= (1 << 1) | (1 << 0);
768 
769 	PL330_DBGCMD_DUMP(SZ_DMAST, "\tDMAST%c\n",
770 		cond == SINGLE ? 'S' : (cond == BURST ? 'B' : 'A'));
771 
772 	return SZ_DMAST;
773 }
774 
775 static inline u32 _emit_STP(unsigned dry_run, u8 buf[],
776 		enum pl330_cond cond, u8 peri)
777 {
778 	if (dry_run)
779 		return SZ_DMASTP;
780 
781 	buf[0] = CMD_DMASTP;
782 
783 	if (cond == BURST)
784 		buf[0] |= (1 << 1);
785 
786 	peri &= 0x1f;
787 	peri <<= 3;
788 	buf[1] = peri;
789 
790 	PL330_DBGCMD_DUMP(SZ_DMASTP, "\tDMASTP%c %u\n",
791 		cond == SINGLE ? 'S' : 'B', peri >> 3);
792 
793 	return SZ_DMASTP;
794 }
795 
796 static inline u32 _emit_WFP(unsigned dry_run, u8 buf[],
797 		enum pl330_cond cond, u8 peri)
798 {
799 	if (dry_run)
800 		return SZ_DMAWFP;
801 
802 	buf[0] = CMD_DMAWFP;
803 
804 	if (cond == SINGLE)
805 		buf[0] |= (0 << 1) | (0 << 0);
806 	else if (cond == BURST)
807 		buf[0] |= (1 << 1) | (0 << 0);
808 	else
809 		buf[0] |= (0 << 1) | (1 << 0);
810 
811 	peri &= 0x1f;
812 	peri <<= 3;
813 	buf[1] = peri;
814 
815 	PL330_DBGCMD_DUMP(SZ_DMAWFP, "\tDMAWFP%c %u\n",
816 		cond == SINGLE ? 'S' : (cond == BURST ? 'B' : 'P'), peri >> 3);
817 
818 	return SZ_DMAWFP;
819 }
820 
821 static inline u32 _emit_WMB(unsigned dry_run, u8 buf[])
822 {
823 	if (dry_run)
824 		return SZ_DMAWMB;
825 
826 	buf[0] = CMD_DMAWMB;
827 
828 	PL330_DBGCMD_DUMP(SZ_DMAWMB, "\tDMAWMB\n");
829 
830 	return SZ_DMAWMB;
831 }
832 
833 struct _arg_GO {
834 	u8 chan;
835 	u32 addr;
836 	unsigned ns;
837 };
838 
839 static inline u32 _emit_GO(unsigned dry_run, u8 buf[],
840 		const struct _arg_GO *arg)
841 {
842 	u8 chan = arg->chan;
843 	u32 addr = arg->addr;
844 	unsigned ns = arg->ns;
845 
846 	if (dry_run)
847 		return SZ_DMAGO;
848 
849 	buf[0] = CMD_DMAGO;
850 	buf[0] |= (ns << 1);
851 	buf[1] = chan & 0x7;
852 	buf[2] = addr;
853 	buf[3] = addr >> 8;
854 	buf[4] = addr >> 16;
855 	buf[5] = addr >> 24;
856 
857 	return SZ_DMAGO;
858 }
859 
860 #define msecs_to_loops(t) (loops_per_jiffy / 1000 * HZ * t)
861 
862 /* Returns Time-Out */
863 static bool _until_dmac_idle(struct pl330_thread *thrd)
864 {
865 	void __iomem *regs = thrd->dmac->base;
866 	unsigned long loops = msecs_to_loops(5);
867 
868 	do {
869 		/* Until Manager is Idle */
870 		if (!(readl(regs + DBGSTATUS) & DBG_BUSY))
871 			break;
872 
873 		cpu_relax();
874 	} while (--loops);
875 
876 	if (!loops)
877 		return true;
878 
879 	return false;
880 }
881 
882 static inline void _execute_DBGINSN(struct pl330_thread *thrd,
883 		u8 insn[], bool as_manager)
884 {
885 	void __iomem *regs = thrd->dmac->base;
886 	u32 val;
887 
888 	val = (insn[0] << 16) | (insn[1] << 24);
889 	if (!as_manager) {
890 		val |= (1 << 0);
891 		val |= (thrd->id << 8); /* Channel Number */
892 	}
893 	writel(val, regs + DBGINST0);
894 
895 	val = le32_to_cpu(*((__le32 *)&insn[2]));
896 	writel(val, regs + DBGINST1);
897 
898 	/* If timed out due to halted state-machine */
899 	if (_until_dmac_idle(thrd)) {
900 		dev_err(thrd->dmac->ddma.dev, "DMAC halted!\n");
901 		return;
902 	}
903 
904 	/* Get going */
905 	writel(0, regs + DBGCMD);
906 }
907 
908 static inline u32 _state(struct pl330_thread *thrd)
909 {
910 	void __iomem *regs = thrd->dmac->base;
911 	u32 val;
912 
913 	if (is_manager(thrd))
914 		val = readl(regs + DS) & 0xf;
915 	else
916 		val = readl(regs + CS(thrd->id)) & 0xf;
917 
918 	switch (val) {
919 	case DS_ST_STOP:
920 		return PL330_STATE_STOPPED;
921 	case DS_ST_EXEC:
922 		return PL330_STATE_EXECUTING;
923 	case DS_ST_CMISS:
924 		return PL330_STATE_CACHEMISS;
925 	case DS_ST_UPDTPC:
926 		return PL330_STATE_UPDTPC;
927 	case DS_ST_WFE:
928 		return PL330_STATE_WFE;
929 	case DS_ST_FAULT:
930 		return PL330_STATE_FAULTING;
931 	case DS_ST_ATBRR:
932 		if (is_manager(thrd))
933 			return PL330_STATE_INVALID;
934 		else
935 			return PL330_STATE_ATBARRIER;
936 	case DS_ST_QBUSY:
937 		if (is_manager(thrd))
938 			return PL330_STATE_INVALID;
939 		else
940 			return PL330_STATE_QUEUEBUSY;
941 	case DS_ST_WFP:
942 		if (is_manager(thrd))
943 			return PL330_STATE_INVALID;
944 		else
945 			return PL330_STATE_WFP;
946 	case DS_ST_KILL:
947 		if (is_manager(thrd))
948 			return PL330_STATE_INVALID;
949 		else
950 			return PL330_STATE_KILLING;
951 	case DS_ST_CMPLT:
952 		if (is_manager(thrd))
953 			return PL330_STATE_INVALID;
954 		else
955 			return PL330_STATE_COMPLETING;
956 	case DS_ST_FLTCMP:
957 		if (is_manager(thrd))
958 			return PL330_STATE_INVALID;
959 		else
960 			return PL330_STATE_FAULT_COMPLETING;
961 	default:
962 		return PL330_STATE_INVALID;
963 	}
964 }
965 
966 static void _stop(struct pl330_thread *thrd)
967 {
968 	void __iomem *regs = thrd->dmac->base;
969 	u8 insn[6] = {0, 0, 0, 0, 0, 0};
970 	u32 inten = readl(regs + INTEN);
971 
972 	if (_state(thrd) == PL330_STATE_FAULT_COMPLETING)
973 		UNTIL(thrd, PL330_STATE_FAULTING | PL330_STATE_KILLING);
974 
975 	/* Return if nothing needs to be done */
976 	if (_state(thrd) == PL330_STATE_COMPLETING
977 		  || _state(thrd) == PL330_STATE_KILLING
978 		  || _state(thrd) == PL330_STATE_STOPPED)
979 		return;
980 
981 	_emit_KILL(0, insn);
982 
983 	_execute_DBGINSN(thrd, insn, is_manager(thrd));
984 
985 	/* clear the event */
986 	if (inten & (1 << thrd->ev))
987 		writel(1 << thrd->ev, regs + INTCLR);
988 	/* Stop generating interrupts for SEV */
989 	writel(inten & ~(1 << thrd->ev), regs + INTEN);
990 }
991 
992 /* Start doing req 'idx' of thread 'thrd' */
993 static bool _trigger(struct pl330_thread *thrd)
994 {
995 	void __iomem *regs = thrd->dmac->base;
996 	struct _pl330_req *req;
997 	struct dma_pl330_desc *desc;
998 	struct _arg_GO go;
999 	unsigned ns;
1000 	u8 insn[6] = {0, 0, 0, 0, 0, 0};
1001 	int idx;
1002 
1003 	/* Return if already ACTIVE */
1004 	if (_state(thrd) != PL330_STATE_STOPPED)
1005 		return true;
1006 
1007 	idx = 1 - thrd->lstenq;
1008 	if (thrd->req[idx].desc != NULL) {
1009 		req = &thrd->req[idx];
1010 	} else {
1011 		idx = thrd->lstenq;
1012 		if (thrd->req[idx].desc != NULL)
1013 			req = &thrd->req[idx];
1014 		else
1015 			req = NULL;
1016 	}
1017 
1018 	/* Return if no request */
1019 	if (!req)
1020 		return true;
1021 
1022 	/* Return if req is running */
1023 	if (idx == thrd->req_running)
1024 		return true;
1025 
1026 	desc = req->desc;
1027 
1028 	ns = desc->rqcfg.nonsecure ? 1 : 0;
1029 
1030 	/* See 'Abort Sources' point-4 at Page 2-25 */
1031 	if (_manager_ns(thrd) && !ns)
1032 		dev_info(thrd->dmac->ddma.dev, "%s:%d Recipe for ABORT!\n",
1033 			__func__, __LINE__);
1034 
1035 	go.chan = thrd->id;
1036 	go.addr = req->mc_bus;
1037 	go.ns = ns;
1038 	_emit_GO(0, insn, &go);
1039 
1040 	/* Set to generate interrupts for SEV */
1041 	writel(readl(regs + INTEN) | (1 << thrd->ev), regs + INTEN);
1042 
1043 	/* Only manager can execute GO */
1044 	_execute_DBGINSN(thrd, insn, true);
1045 
1046 	thrd->req_running = idx;
1047 
1048 	return true;
1049 }
1050 
1051 static bool _start(struct pl330_thread *thrd)
1052 {
1053 	switch (_state(thrd)) {
1054 	case PL330_STATE_FAULT_COMPLETING:
1055 		UNTIL(thrd, PL330_STATE_FAULTING | PL330_STATE_KILLING);
1056 
1057 		if (_state(thrd) == PL330_STATE_KILLING)
1058 			UNTIL(thrd, PL330_STATE_STOPPED)
1059 		/* fall through */
1060 
1061 	case PL330_STATE_FAULTING:
1062 		_stop(thrd);
1063 		/* fall through */
1064 
1065 	case PL330_STATE_KILLING:
1066 	case PL330_STATE_COMPLETING:
1067 		UNTIL(thrd, PL330_STATE_STOPPED)
1068 		/* fall through */
1069 
1070 	case PL330_STATE_STOPPED:
1071 		return _trigger(thrd);
1072 
1073 	case PL330_STATE_WFP:
1074 	case PL330_STATE_QUEUEBUSY:
1075 	case PL330_STATE_ATBARRIER:
1076 	case PL330_STATE_UPDTPC:
1077 	case PL330_STATE_CACHEMISS:
1078 	case PL330_STATE_EXECUTING:
1079 		return true;
1080 
1081 	case PL330_STATE_WFE: /* For RESUME, nothing yet */
1082 	default:
1083 		return false;
1084 	}
1085 }
1086 
1087 static inline int _ldst_memtomem(unsigned dry_run, u8 buf[],
1088 		const struct _xfer_spec *pxs, int cyc)
1089 {
1090 	int off = 0;
1091 	struct pl330_config *pcfg = pxs->desc->rqcfg.pcfg;
1092 
1093 	/* check lock-up free version */
1094 	if (get_revision(pcfg->periph_id) >= PERIPH_REV_R1P0) {
1095 		while (cyc--) {
1096 			off += _emit_LD(dry_run, &buf[off], ALWAYS);
1097 			off += _emit_ST(dry_run, &buf[off], ALWAYS);
1098 		}
1099 	} else {
1100 		while (cyc--) {
1101 			off += _emit_LD(dry_run, &buf[off], ALWAYS);
1102 			off += _emit_RMB(dry_run, &buf[off]);
1103 			off += _emit_ST(dry_run, &buf[off], ALWAYS);
1104 			off += _emit_WMB(dry_run, &buf[off]);
1105 		}
1106 	}
1107 
1108 	return off;
1109 }
1110 
1111 static u32 _emit_load(unsigned int dry_run, u8 buf[],
1112 	enum pl330_cond cond, enum dma_transfer_direction direction,
1113 	u8 peri)
1114 {
1115 	int off = 0;
1116 
1117 	switch (direction) {
1118 	case DMA_MEM_TO_MEM:
1119 		/* fall through */
1120 	case DMA_MEM_TO_DEV:
1121 		off += _emit_LD(dry_run, &buf[off], cond);
1122 		break;
1123 
1124 	case DMA_DEV_TO_MEM:
1125 		if (cond == ALWAYS) {
1126 			off += _emit_LDP(dry_run, &buf[off], SINGLE,
1127 				peri);
1128 			off += _emit_LDP(dry_run, &buf[off], BURST,
1129 				peri);
1130 		} else {
1131 			off += _emit_LDP(dry_run, &buf[off], cond,
1132 				peri);
1133 		}
1134 		break;
1135 
1136 	default:
1137 		/* this code should be unreachable */
1138 		WARN_ON(1);
1139 		break;
1140 	}
1141 
1142 	return off;
1143 }
1144 
1145 static inline u32 _emit_store(unsigned int dry_run, u8 buf[],
1146 	enum pl330_cond cond, enum dma_transfer_direction direction,
1147 	u8 peri)
1148 {
1149 	int off = 0;
1150 
1151 	switch (direction) {
1152 	case DMA_MEM_TO_MEM:
1153 		/* fall through */
1154 	case DMA_DEV_TO_MEM:
1155 		off += _emit_ST(dry_run, &buf[off], cond);
1156 		break;
1157 
1158 	case DMA_MEM_TO_DEV:
1159 		if (cond == ALWAYS) {
1160 			off += _emit_STP(dry_run, &buf[off], SINGLE,
1161 				peri);
1162 			off += _emit_STP(dry_run, &buf[off], BURST,
1163 				peri);
1164 		} else {
1165 			off += _emit_STP(dry_run, &buf[off], cond,
1166 				peri);
1167 		}
1168 		break;
1169 
1170 	default:
1171 		/* this code should be unreachable */
1172 		WARN_ON(1);
1173 		break;
1174 	}
1175 
1176 	return off;
1177 }
1178 
1179 static inline int _ldst_peripheral(struct pl330_dmac *pl330,
1180 				 unsigned dry_run, u8 buf[],
1181 				 const struct _xfer_spec *pxs, int cyc,
1182 				 enum pl330_cond cond)
1183 {
1184 	int off = 0;
1185 
1186 	if (pl330->quirks & PL330_QUIRK_BROKEN_NO_FLUSHP)
1187 		cond = BURST;
1188 
1189 	/*
1190 	 * do FLUSHP at beginning to clear any stale dma requests before the
1191 	 * first WFP.
1192 	 */
1193 	if (!(pl330->quirks & PL330_QUIRK_BROKEN_NO_FLUSHP))
1194 		off += _emit_FLUSHP(dry_run, &buf[off], pxs->desc->peri);
1195 	while (cyc--) {
1196 		off += _emit_WFP(dry_run, &buf[off], cond, pxs->desc->peri);
1197 		off += _emit_load(dry_run, &buf[off], cond, pxs->desc->rqtype,
1198 			pxs->desc->peri);
1199 		off += _emit_store(dry_run, &buf[off], cond, pxs->desc->rqtype,
1200 			pxs->desc->peri);
1201 	}
1202 
1203 	return off;
1204 }
1205 
1206 static int _bursts(struct pl330_dmac *pl330, unsigned dry_run, u8 buf[],
1207 		const struct _xfer_spec *pxs, int cyc)
1208 {
1209 	int off = 0;
1210 	enum pl330_cond cond = BRST_LEN(pxs->ccr) > 1 ? BURST : SINGLE;
1211 
1212 	switch (pxs->desc->rqtype) {
1213 	case DMA_MEM_TO_DEV:
1214 		/* fall through */
1215 	case DMA_DEV_TO_MEM:
1216 		off += _ldst_peripheral(pl330, dry_run, &buf[off], pxs, cyc,
1217 			cond);
1218 		break;
1219 
1220 	case DMA_MEM_TO_MEM:
1221 		off += _ldst_memtomem(dry_run, &buf[off], pxs, cyc);
1222 		break;
1223 
1224 	default:
1225 		/* this code should be unreachable */
1226 		WARN_ON(1);
1227 		break;
1228 	}
1229 
1230 	return off;
1231 }
1232 
1233 /*
1234  * transfer dregs with single transfers to peripheral, or a reduced size burst
1235  * for mem-to-mem.
1236  */
1237 static int _dregs(struct pl330_dmac *pl330, unsigned int dry_run, u8 buf[],
1238 		const struct _xfer_spec *pxs, int transfer_length)
1239 {
1240 	int off = 0;
1241 	int dregs_ccr;
1242 
1243 	if (transfer_length == 0)
1244 		return off;
1245 
1246 	switch (pxs->desc->rqtype) {
1247 	case DMA_MEM_TO_DEV:
1248 		/* fall through */
1249 	case DMA_DEV_TO_MEM:
1250 		off += _ldst_peripheral(pl330, dry_run, &buf[off], pxs,
1251 			transfer_length, SINGLE);
1252 		break;
1253 
1254 	case DMA_MEM_TO_MEM:
1255 		dregs_ccr = pxs->ccr;
1256 		dregs_ccr &= ~((0xf << CC_SRCBRSTLEN_SHFT) |
1257 			(0xf << CC_DSTBRSTLEN_SHFT));
1258 		dregs_ccr |= (((transfer_length - 1) & 0xf) <<
1259 			CC_SRCBRSTLEN_SHFT);
1260 		dregs_ccr |= (((transfer_length - 1) & 0xf) <<
1261 			CC_DSTBRSTLEN_SHFT);
1262 		off += _emit_MOV(dry_run, &buf[off], CCR, dregs_ccr);
1263 		off += _ldst_memtomem(dry_run, &buf[off], pxs, 1);
1264 		break;
1265 
1266 	default:
1267 		/* this code should be unreachable */
1268 		WARN_ON(1);
1269 		break;
1270 	}
1271 
1272 	return off;
1273 }
1274 
1275 /* Returns bytes consumed and updates bursts */
1276 static inline int _loop(struct pl330_dmac *pl330, unsigned dry_run, u8 buf[],
1277 		unsigned long *bursts, const struct _xfer_spec *pxs)
1278 {
1279 	int cyc, cycmax, szlp, szlpend, szbrst, off;
1280 	unsigned lcnt0, lcnt1, ljmp0, ljmp1;
1281 	struct _arg_LPEND lpend;
1282 
1283 	if (*bursts == 1)
1284 		return _bursts(pl330, dry_run, buf, pxs, 1);
1285 
1286 	/* Max iterations possible in DMALP is 256 */
1287 	if (*bursts >= 256*256) {
1288 		lcnt1 = 256;
1289 		lcnt0 = 256;
1290 		cyc = *bursts / lcnt1 / lcnt0;
1291 	} else if (*bursts > 256) {
1292 		lcnt1 = 256;
1293 		lcnt0 = *bursts / lcnt1;
1294 		cyc = 1;
1295 	} else {
1296 		lcnt1 = *bursts;
1297 		lcnt0 = 0;
1298 		cyc = 1;
1299 	}
1300 
1301 	szlp = _emit_LP(1, buf, 0, 0);
1302 	szbrst = _bursts(pl330, 1, buf, pxs, 1);
1303 
1304 	lpend.cond = ALWAYS;
1305 	lpend.forever = false;
1306 	lpend.loop = 0;
1307 	lpend.bjump = 0;
1308 	szlpend = _emit_LPEND(1, buf, &lpend);
1309 
1310 	if (lcnt0) {
1311 		szlp *= 2;
1312 		szlpend *= 2;
1313 	}
1314 
1315 	/*
1316 	 * Max bursts that we can unroll due to limit on the
1317 	 * size of backward jump that can be encoded in DMALPEND
1318 	 * which is 8-bits and hence 255
1319 	 */
1320 	cycmax = (255 - (szlp + szlpend)) / szbrst;
1321 
1322 	cyc = (cycmax < cyc) ? cycmax : cyc;
1323 
1324 	off = 0;
1325 
1326 	if (lcnt0) {
1327 		off += _emit_LP(dry_run, &buf[off], 0, lcnt0);
1328 		ljmp0 = off;
1329 	}
1330 
1331 	off += _emit_LP(dry_run, &buf[off], 1, lcnt1);
1332 	ljmp1 = off;
1333 
1334 	off += _bursts(pl330, dry_run, &buf[off], pxs, cyc);
1335 
1336 	lpend.cond = ALWAYS;
1337 	lpend.forever = false;
1338 	lpend.loop = 1;
1339 	lpend.bjump = off - ljmp1;
1340 	off += _emit_LPEND(dry_run, &buf[off], &lpend);
1341 
1342 	if (lcnt0) {
1343 		lpend.cond = ALWAYS;
1344 		lpend.forever = false;
1345 		lpend.loop = 0;
1346 		lpend.bjump = off - ljmp0;
1347 		off += _emit_LPEND(dry_run, &buf[off], &lpend);
1348 	}
1349 
1350 	*bursts = lcnt1 * cyc;
1351 	if (lcnt0)
1352 		*bursts *= lcnt0;
1353 
1354 	return off;
1355 }
1356 
1357 static inline int _setup_loops(struct pl330_dmac *pl330,
1358 			       unsigned dry_run, u8 buf[],
1359 			       const struct _xfer_spec *pxs)
1360 {
1361 	struct pl330_xfer *x = &pxs->desc->px;
1362 	u32 ccr = pxs->ccr;
1363 	unsigned long c, bursts = BYTE_TO_BURST(x->bytes, ccr);
1364 	int num_dregs = (x->bytes - BURST_TO_BYTE(bursts, ccr)) /
1365 		BRST_SIZE(ccr);
1366 	int off = 0;
1367 
1368 	while (bursts) {
1369 		c = bursts;
1370 		off += _loop(pl330, dry_run, &buf[off], &c, pxs);
1371 		bursts -= c;
1372 	}
1373 	off += _dregs(pl330, dry_run, &buf[off], pxs, num_dregs);
1374 
1375 	return off;
1376 }
1377 
1378 static inline int _setup_xfer(struct pl330_dmac *pl330,
1379 			      unsigned dry_run, u8 buf[],
1380 			      const struct _xfer_spec *pxs)
1381 {
1382 	struct pl330_xfer *x = &pxs->desc->px;
1383 	int off = 0;
1384 
1385 	/* DMAMOV SAR, x->src_addr */
1386 	off += _emit_MOV(dry_run, &buf[off], SAR, x->src_addr);
1387 	/* DMAMOV DAR, x->dst_addr */
1388 	off += _emit_MOV(dry_run, &buf[off], DAR, x->dst_addr);
1389 
1390 	/* Setup Loop(s) */
1391 	off += _setup_loops(pl330, dry_run, &buf[off], pxs);
1392 
1393 	return off;
1394 }
1395 
1396 /*
1397  * A req is a sequence of one or more xfer units.
1398  * Returns the number of bytes taken to setup the MC for the req.
1399  */
1400 static int _setup_req(struct pl330_dmac *pl330, unsigned dry_run,
1401 		      struct pl330_thread *thrd, unsigned index,
1402 		      struct _xfer_spec *pxs)
1403 {
1404 	struct _pl330_req *req = &thrd->req[index];
1405 	u8 *buf = req->mc_cpu;
1406 	int off = 0;
1407 
1408 	PL330_DBGMC_START(req->mc_bus);
1409 
1410 	/* DMAMOV CCR, ccr */
1411 	off += _emit_MOV(dry_run, &buf[off], CCR, pxs->ccr);
1412 
1413 	off += _setup_xfer(pl330, dry_run, &buf[off], pxs);
1414 
1415 	/* DMASEV peripheral/event */
1416 	off += _emit_SEV(dry_run, &buf[off], thrd->ev);
1417 	/* DMAEND */
1418 	off += _emit_END(dry_run, &buf[off]);
1419 
1420 	return off;
1421 }
1422 
1423 static inline u32 _prepare_ccr(const struct pl330_reqcfg *rqc)
1424 {
1425 	u32 ccr = 0;
1426 
1427 	if (rqc->src_inc)
1428 		ccr |= CC_SRCINC;
1429 
1430 	if (rqc->dst_inc)
1431 		ccr |= CC_DSTINC;
1432 
1433 	/* We set same protection levels for Src and DST for now */
1434 	if (rqc->privileged)
1435 		ccr |= CC_SRCPRI | CC_DSTPRI;
1436 	if (rqc->nonsecure)
1437 		ccr |= CC_SRCNS | CC_DSTNS;
1438 	if (rqc->insnaccess)
1439 		ccr |= CC_SRCIA | CC_DSTIA;
1440 
1441 	ccr |= (((rqc->brst_len - 1) & 0xf) << CC_SRCBRSTLEN_SHFT);
1442 	ccr |= (((rqc->brst_len - 1) & 0xf) << CC_DSTBRSTLEN_SHFT);
1443 
1444 	ccr |= (rqc->brst_size << CC_SRCBRSTSIZE_SHFT);
1445 	ccr |= (rqc->brst_size << CC_DSTBRSTSIZE_SHFT);
1446 
1447 	ccr |= (rqc->scctl << CC_SRCCCTRL_SHFT);
1448 	ccr |= (rqc->dcctl << CC_DSTCCTRL_SHFT);
1449 
1450 	ccr |= (rqc->swap << CC_SWAP_SHFT);
1451 
1452 	return ccr;
1453 }
1454 
1455 /*
1456  * Submit a list of xfers after which the client wants notification.
1457  * Client is not notified after each xfer unit, just once after all
1458  * xfer units are done or some error occurs.
1459  */
1460 static int pl330_submit_req(struct pl330_thread *thrd,
1461 	struct dma_pl330_desc *desc)
1462 {
1463 	struct pl330_dmac *pl330 = thrd->dmac;
1464 	struct _xfer_spec xs;
1465 	unsigned long flags;
1466 	unsigned idx;
1467 	u32 ccr;
1468 	int ret = 0;
1469 
1470 	switch (desc->rqtype) {
1471 	case DMA_MEM_TO_DEV:
1472 		break;
1473 
1474 	case DMA_DEV_TO_MEM:
1475 		break;
1476 
1477 	case DMA_MEM_TO_MEM:
1478 		break;
1479 
1480 	default:
1481 		return -ENOTSUPP;
1482 	}
1483 
1484 	if (pl330->state == DYING
1485 		|| pl330->dmac_tbd.reset_chan & (1 << thrd->id)) {
1486 		dev_info(thrd->dmac->ddma.dev, "%s:%d\n",
1487 			__func__, __LINE__);
1488 		return -EAGAIN;
1489 	}
1490 
1491 	/* If request for non-existing peripheral */
1492 	if (desc->rqtype != DMA_MEM_TO_MEM &&
1493 	    desc->peri >= pl330->pcfg.num_peri) {
1494 		dev_info(thrd->dmac->ddma.dev,
1495 				"%s:%d Invalid peripheral(%u)!\n",
1496 				__func__, __LINE__, desc->peri);
1497 		return -EINVAL;
1498 	}
1499 
1500 	spin_lock_irqsave(&pl330->lock, flags);
1501 
1502 	if (_queue_full(thrd)) {
1503 		ret = -EAGAIN;
1504 		goto xfer_exit;
1505 	}
1506 
1507 	/* Prefer Secure Channel */
1508 	if (!_manager_ns(thrd))
1509 		desc->rqcfg.nonsecure = 0;
1510 	else
1511 		desc->rqcfg.nonsecure = 1;
1512 
1513 	ccr = _prepare_ccr(&desc->rqcfg);
1514 
1515 	idx = thrd->req[0].desc == NULL ? 0 : 1;
1516 
1517 	xs.ccr = ccr;
1518 	xs.desc = desc;
1519 
1520 	/* First dry run to check if req is acceptable */
1521 	ret = _setup_req(pl330, 1, thrd, idx, &xs);
1522 	if (ret < 0)
1523 		goto xfer_exit;
1524 
1525 	if (ret > pl330->mcbufsz / 2) {
1526 		dev_info(pl330->ddma.dev, "%s:%d Try increasing mcbufsz (%i/%i)\n",
1527 				__func__, __LINE__, ret, pl330->mcbufsz / 2);
1528 		ret = -ENOMEM;
1529 		goto xfer_exit;
1530 	}
1531 
1532 	/* Hook the request */
1533 	thrd->lstenq = idx;
1534 	thrd->req[idx].desc = desc;
1535 	_setup_req(pl330, 0, thrd, idx, &xs);
1536 
1537 	ret = 0;
1538 
1539 xfer_exit:
1540 	spin_unlock_irqrestore(&pl330->lock, flags);
1541 
1542 	return ret;
1543 }
1544 
1545 static void dma_pl330_rqcb(struct dma_pl330_desc *desc, enum pl330_op_err err)
1546 {
1547 	struct dma_pl330_chan *pch;
1548 	unsigned long flags;
1549 
1550 	if (!desc)
1551 		return;
1552 
1553 	pch = desc->pchan;
1554 
1555 	/* If desc aborted */
1556 	if (!pch)
1557 		return;
1558 
1559 	spin_lock_irqsave(&pch->lock, flags);
1560 
1561 	desc->status = DONE;
1562 
1563 	spin_unlock_irqrestore(&pch->lock, flags);
1564 
1565 	tasklet_schedule(&pch->task);
1566 }
1567 
1568 static void pl330_dotask(unsigned long data)
1569 {
1570 	struct pl330_dmac *pl330 = (struct pl330_dmac *) data;
1571 	unsigned long flags;
1572 	int i;
1573 
1574 	spin_lock_irqsave(&pl330->lock, flags);
1575 
1576 	/* The DMAC itself gone nuts */
1577 	if (pl330->dmac_tbd.reset_dmac) {
1578 		pl330->state = DYING;
1579 		/* Reset the manager too */
1580 		pl330->dmac_tbd.reset_mngr = true;
1581 		/* Clear the reset flag */
1582 		pl330->dmac_tbd.reset_dmac = false;
1583 	}
1584 
1585 	if (pl330->dmac_tbd.reset_mngr) {
1586 		_stop(pl330->manager);
1587 		/* Reset all channels */
1588 		pl330->dmac_tbd.reset_chan = (1 << pl330->pcfg.num_chan) - 1;
1589 		/* Clear the reset flag */
1590 		pl330->dmac_tbd.reset_mngr = false;
1591 	}
1592 
1593 	for (i = 0; i < pl330->pcfg.num_chan; i++) {
1594 
1595 		if (pl330->dmac_tbd.reset_chan & (1 << i)) {
1596 			struct pl330_thread *thrd = &pl330->channels[i];
1597 			void __iomem *regs = pl330->base;
1598 			enum pl330_op_err err;
1599 
1600 			_stop(thrd);
1601 
1602 			if (readl(regs + FSC) & (1 << thrd->id))
1603 				err = PL330_ERR_FAIL;
1604 			else
1605 				err = PL330_ERR_ABORT;
1606 
1607 			spin_unlock_irqrestore(&pl330->lock, flags);
1608 			dma_pl330_rqcb(thrd->req[1 - thrd->lstenq].desc, err);
1609 			dma_pl330_rqcb(thrd->req[thrd->lstenq].desc, err);
1610 			spin_lock_irqsave(&pl330->lock, flags);
1611 
1612 			thrd->req[0].desc = NULL;
1613 			thrd->req[1].desc = NULL;
1614 			thrd->req_running = -1;
1615 
1616 			/* Clear the reset flag */
1617 			pl330->dmac_tbd.reset_chan &= ~(1 << i);
1618 		}
1619 	}
1620 
1621 	spin_unlock_irqrestore(&pl330->lock, flags);
1622 
1623 	return;
1624 }
1625 
1626 /* Returns 1 if state was updated, 0 otherwise */
1627 static int pl330_update(struct pl330_dmac *pl330)
1628 {
1629 	struct dma_pl330_desc *descdone;
1630 	unsigned long flags;
1631 	void __iomem *regs;
1632 	u32 val;
1633 	int id, ev, ret = 0;
1634 
1635 	regs = pl330->base;
1636 
1637 	spin_lock_irqsave(&pl330->lock, flags);
1638 
1639 	val = readl(regs + FSM) & 0x1;
1640 	if (val)
1641 		pl330->dmac_tbd.reset_mngr = true;
1642 	else
1643 		pl330->dmac_tbd.reset_mngr = false;
1644 
1645 	val = readl(regs + FSC) & ((1 << pl330->pcfg.num_chan) - 1);
1646 	pl330->dmac_tbd.reset_chan |= val;
1647 	if (val) {
1648 		int i = 0;
1649 		while (i < pl330->pcfg.num_chan) {
1650 			if (val & (1 << i)) {
1651 				dev_info(pl330->ddma.dev,
1652 					"Reset Channel-%d\t CS-%x FTC-%x\n",
1653 						i, readl(regs + CS(i)),
1654 						readl(regs + FTC(i)));
1655 				_stop(&pl330->channels[i]);
1656 			}
1657 			i++;
1658 		}
1659 	}
1660 
1661 	/* Check which event happened i.e, thread notified */
1662 	val = readl(regs + ES);
1663 	if (pl330->pcfg.num_events < 32
1664 			&& val & ~((1 << pl330->pcfg.num_events) - 1)) {
1665 		pl330->dmac_tbd.reset_dmac = true;
1666 		dev_err(pl330->ddma.dev, "%s:%d Unexpected!\n", __func__,
1667 			__LINE__);
1668 		ret = 1;
1669 		goto updt_exit;
1670 	}
1671 
1672 	for (ev = 0; ev < pl330->pcfg.num_events; ev++) {
1673 		if (val & (1 << ev)) { /* Event occurred */
1674 			struct pl330_thread *thrd;
1675 			u32 inten = readl(regs + INTEN);
1676 			int active;
1677 
1678 			/* Clear the event */
1679 			if (inten & (1 << ev))
1680 				writel(1 << ev, regs + INTCLR);
1681 
1682 			ret = 1;
1683 
1684 			id = pl330->events[ev];
1685 
1686 			thrd = &pl330->channels[id];
1687 
1688 			active = thrd->req_running;
1689 			if (active == -1) /* Aborted */
1690 				continue;
1691 
1692 			/* Detach the req */
1693 			descdone = thrd->req[active].desc;
1694 			thrd->req[active].desc = NULL;
1695 
1696 			thrd->req_running = -1;
1697 
1698 			/* Get going again ASAP */
1699 			_start(thrd);
1700 
1701 			/* For now, just make a list of callbacks to be done */
1702 			list_add_tail(&descdone->rqd, &pl330->req_done);
1703 		}
1704 	}
1705 
1706 	/* Now that we are in no hurry, do the callbacks */
1707 	while (!list_empty(&pl330->req_done)) {
1708 		descdone = list_first_entry(&pl330->req_done,
1709 					    struct dma_pl330_desc, rqd);
1710 		list_del(&descdone->rqd);
1711 		spin_unlock_irqrestore(&pl330->lock, flags);
1712 		dma_pl330_rqcb(descdone, PL330_ERR_NONE);
1713 		spin_lock_irqsave(&pl330->lock, flags);
1714 	}
1715 
1716 updt_exit:
1717 	spin_unlock_irqrestore(&pl330->lock, flags);
1718 
1719 	if (pl330->dmac_tbd.reset_dmac
1720 			|| pl330->dmac_tbd.reset_mngr
1721 			|| pl330->dmac_tbd.reset_chan) {
1722 		ret = 1;
1723 		tasklet_schedule(&pl330->tasks);
1724 	}
1725 
1726 	return ret;
1727 }
1728 
1729 /* Reserve an event */
1730 static inline int _alloc_event(struct pl330_thread *thrd)
1731 {
1732 	struct pl330_dmac *pl330 = thrd->dmac;
1733 	int ev;
1734 
1735 	for (ev = 0; ev < pl330->pcfg.num_events; ev++)
1736 		if (pl330->events[ev] == -1) {
1737 			pl330->events[ev] = thrd->id;
1738 			return ev;
1739 		}
1740 
1741 	return -1;
1742 }
1743 
1744 static bool _chan_ns(const struct pl330_dmac *pl330, int i)
1745 {
1746 	return pl330->pcfg.irq_ns & (1 << i);
1747 }
1748 
1749 /* Upon success, returns IdentityToken for the
1750  * allocated channel, NULL otherwise.
1751  */
1752 static struct pl330_thread *pl330_request_channel(struct pl330_dmac *pl330)
1753 {
1754 	struct pl330_thread *thrd = NULL;
1755 	int chans, i;
1756 
1757 	if (pl330->state == DYING)
1758 		return NULL;
1759 
1760 	chans = pl330->pcfg.num_chan;
1761 
1762 	for (i = 0; i < chans; i++) {
1763 		thrd = &pl330->channels[i];
1764 		if ((thrd->free) && (!_manager_ns(thrd) ||
1765 					_chan_ns(pl330, i))) {
1766 			thrd->ev = _alloc_event(thrd);
1767 			if (thrd->ev >= 0) {
1768 				thrd->free = false;
1769 				thrd->lstenq = 1;
1770 				thrd->req[0].desc = NULL;
1771 				thrd->req[1].desc = NULL;
1772 				thrd->req_running = -1;
1773 				break;
1774 			}
1775 		}
1776 		thrd = NULL;
1777 	}
1778 
1779 	return thrd;
1780 }
1781 
1782 /* Release an event */
1783 static inline void _free_event(struct pl330_thread *thrd, int ev)
1784 {
1785 	struct pl330_dmac *pl330 = thrd->dmac;
1786 
1787 	/* If the event is valid and was held by the thread */
1788 	if (ev >= 0 && ev < pl330->pcfg.num_events
1789 			&& pl330->events[ev] == thrd->id)
1790 		pl330->events[ev] = -1;
1791 }
1792 
1793 static void pl330_release_channel(struct pl330_thread *thrd)
1794 {
1795 	if (!thrd || thrd->free)
1796 		return;
1797 
1798 	_stop(thrd);
1799 
1800 	dma_pl330_rqcb(thrd->req[1 - thrd->lstenq].desc, PL330_ERR_ABORT);
1801 	dma_pl330_rqcb(thrd->req[thrd->lstenq].desc, PL330_ERR_ABORT);
1802 
1803 	_free_event(thrd, thrd->ev);
1804 	thrd->free = true;
1805 }
1806 
1807 /* Initialize the structure for PL330 configuration, that can be used
1808  * by the client driver the make best use of the DMAC
1809  */
1810 static void read_dmac_config(struct pl330_dmac *pl330)
1811 {
1812 	void __iomem *regs = pl330->base;
1813 	u32 val;
1814 
1815 	val = readl(regs + CRD) >> CRD_DATA_WIDTH_SHIFT;
1816 	val &= CRD_DATA_WIDTH_MASK;
1817 	pl330->pcfg.data_bus_width = 8 * (1 << val);
1818 
1819 	val = readl(regs + CRD) >> CRD_DATA_BUFF_SHIFT;
1820 	val &= CRD_DATA_BUFF_MASK;
1821 	pl330->pcfg.data_buf_dep = val + 1;
1822 
1823 	val = readl(regs + CR0) >> CR0_NUM_CHANS_SHIFT;
1824 	val &= CR0_NUM_CHANS_MASK;
1825 	val += 1;
1826 	pl330->pcfg.num_chan = val;
1827 
1828 	val = readl(regs + CR0);
1829 	if (val & CR0_PERIPH_REQ_SET) {
1830 		val = (val >> CR0_NUM_PERIPH_SHIFT) & CR0_NUM_PERIPH_MASK;
1831 		val += 1;
1832 		pl330->pcfg.num_peri = val;
1833 		pl330->pcfg.peri_ns = readl(regs + CR4);
1834 	} else {
1835 		pl330->pcfg.num_peri = 0;
1836 	}
1837 
1838 	val = readl(regs + CR0);
1839 	if (val & CR0_BOOT_MAN_NS)
1840 		pl330->pcfg.mode |= DMAC_MODE_NS;
1841 	else
1842 		pl330->pcfg.mode &= ~DMAC_MODE_NS;
1843 
1844 	val = readl(regs + CR0) >> CR0_NUM_EVENTS_SHIFT;
1845 	val &= CR0_NUM_EVENTS_MASK;
1846 	val += 1;
1847 	pl330->pcfg.num_events = val;
1848 
1849 	pl330->pcfg.irq_ns = readl(regs + CR3);
1850 }
1851 
1852 static inline void _reset_thread(struct pl330_thread *thrd)
1853 {
1854 	struct pl330_dmac *pl330 = thrd->dmac;
1855 
1856 	thrd->req[0].mc_cpu = pl330->mcode_cpu
1857 				+ (thrd->id * pl330->mcbufsz);
1858 	thrd->req[0].mc_bus = pl330->mcode_bus
1859 				+ (thrd->id * pl330->mcbufsz);
1860 	thrd->req[0].desc = NULL;
1861 
1862 	thrd->req[1].mc_cpu = thrd->req[0].mc_cpu
1863 				+ pl330->mcbufsz / 2;
1864 	thrd->req[1].mc_bus = thrd->req[0].mc_bus
1865 				+ pl330->mcbufsz / 2;
1866 	thrd->req[1].desc = NULL;
1867 
1868 	thrd->req_running = -1;
1869 }
1870 
1871 static int dmac_alloc_threads(struct pl330_dmac *pl330)
1872 {
1873 	int chans = pl330->pcfg.num_chan;
1874 	struct pl330_thread *thrd;
1875 	int i;
1876 
1877 	/* Allocate 1 Manager and 'chans' Channel threads */
1878 	pl330->channels = kcalloc(1 + chans, sizeof(*thrd),
1879 					GFP_KERNEL);
1880 	if (!pl330->channels)
1881 		return -ENOMEM;
1882 
1883 	/* Init Channel threads */
1884 	for (i = 0; i < chans; i++) {
1885 		thrd = &pl330->channels[i];
1886 		thrd->id = i;
1887 		thrd->dmac = pl330;
1888 		_reset_thread(thrd);
1889 		thrd->free = true;
1890 	}
1891 
1892 	/* MANAGER is indexed at the end */
1893 	thrd = &pl330->channels[chans];
1894 	thrd->id = chans;
1895 	thrd->dmac = pl330;
1896 	thrd->free = false;
1897 	pl330->manager = thrd;
1898 
1899 	return 0;
1900 }
1901 
1902 static int dmac_alloc_resources(struct pl330_dmac *pl330)
1903 {
1904 	int chans = pl330->pcfg.num_chan;
1905 	int ret;
1906 
1907 	/*
1908 	 * Alloc MicroCode buffer for 'chans' Channel threads.
1909 	 * A channel's buffer offset is (Channel_Id * MCODE_BUFF_PERCHAN)
1910 	 */
1911 	pl330->mcode_cpu = dma_alloc_attrs(pl330->ddma.dev,
1912 				chans * pl330->mcbufsz,
1913 				&pl330->mcode_bus, GFP_KERNEL,
1914 				DMA_ATTR_PRIVILEGED);
1915 	if (!pl330->mcode_cpu) {
1916 		dev_err(pl330->ddma.dev, "%s:%d Can't allocate memory!\n",
1917 			__func__, __LINE__);
1918 		return -ENOMEM;
1919 	}
1920 
1921 	ret = dmac_alloc_threads(pl330);
1922 	if (ret) {
1923 		dev_err(pl330->ddma.dev, "%s:%d Can't to create channels for DMAC!\n",
1924 			__func__, __LINE__);
1925 		dma_free_coherent(pl330->ddma.dev,
1926 				chans * pl330->mcbufsz,
1927 				pl330->mcode_cpu, pl330->mcode_bus);
1928 		return ret;
1929 	}
1930 
1931 	return 0;
1932 }
1933 
1934 static int pl330_add(struct pl330_dmac *pl330)
1935 {
1936 	int i, ret;
1937 
1938 	/* Check if we can handle this DMAC */
1939 	if ((pl330->pcfg.periph_id & 0xfffff) != PERIPH_ID_VAL) {
1940 		dev_err(pl330->ddma.dev, "PERIPH_ID 0x%x !\n",
1941 			pl330->pcfg.periph_id);
1942 		return -EINVAL;
1943 	}
1944 
1945 	/* Read the configuration of the DMAC */
1946 	read_dmac_config(pl330);
1947 
1948 	if (pl330->pcfg.num_events == 0) {
1949 		dev_err(pl330->ddma.dev, "%s:%d Can't work without events!\n",
1950 			__func__, __LINE__);
1951 		return -EINVAL;
1952 	}
1953 
1954 	spin_lock_init(&pl330->lock);
1955 
1956 	INIT_LIST_HEAD(&pl330->req_done);
1957 
1958 	/* Use default MC buffer size if not provided */
1959 	if (!pl330->mcbufsz)
1960 		pl330->mcbufsz = MCODE_BUFF_PER_REQ * 2;
1961 
1962 	/* Mark all events as free */
1963 	for (i = 0; i < pl330->pcfg.num_events; i++)
1964 		pl330->events[i] = -1;
1965 
1966 	/* Allocate resources needed by the DMAC */
1967 	ret = dmac_alloc_resources(pl330);
1968 	if (ret) {
1969 		dev_err(pl330->ddma.dev, "Unable to create channels for DMAC\n");
1970 		return ret;
1971 	}
1972 
1973 	tasklet_init(&pl330->tasks, pl330_dotask, (unsigned long) pl330);
1974 
1975 	pl330->state = INIT;
1976 
1977 	return 0;
1978 }
1979 
1980 static int dmac_free_threads(struct pl330_dmac *pl330)
1981 {
1982 	struct pl330_thread *thrd;
1983 	int i;
1984 
1985 	/* Release Channel threads */
1986 	for (i = 0; i < pl330->pcfg.num_chan; i++) {
1987 		thrd = &pl330->channels[i];
1988 		pl330_release_channel(thrd);
1989 	}
1990 
1991 	/* Free memory */
1992 	kfree(pl330->channels);
1993 
1994 	return 0;
1995 }
1996 
1997 static void pl330_del(struct pl330_dmac *pl330)
1998 {
1999 	pl330->state = UNINIT;
2000 
2001 	tasklet_kill(&pl330->tasks);
2002 
2003 	/* Free DMAC resources */
2004 	dmac_free_threads(pl330);
2005 
2006 	dma_free_coherent(pl330->ddma.dev,
2007 		pl330->pcfg.num_chan * pl330->mcbufsz, pl330->mcode_cpu,
2008 		pl330->mcode_bus);
2009 }
2010 
2011 /* forward declaration */
2012 static struct amba_driver pl330_driver;
2013 
2014 static inline struct dma_pl330_chan *
2015 to_pchan(struct dma_chan *ch)
2016 {
2017 	if (!ch)
2018 		return NULL;
2019 
2020 	return container_of(ch, struct dma_pl330_chan, chan);
2021 }
2022 
2023 static inline struct dma_pl330_desc *
2024 to_desc(struct dma_async_tx_descriptor *tx)
2025 {
2026 	return container_of(tx, struct dma_pl330_desc, txd);
2027 }
2028 
2029 static inline void fill_queue(struct dma_pl330_chan *pch)
2030 {
2031 	struct dma_pl330_desc *desc;
2032 	int ret;
2033 
2034 	list_for_each_entry(desc, &pch->work_list, node) {
2035 
2036 		/* If already submitted */
2037 		if (desc->status == BUSY)
2038 			continue;
2039 
2040 		ret = pl330_submit_req(pch->thread, desc);
2041 		if (!ret) {
2042 			desc->status = BUSY;
2043 		} else if (ret == -EAGAIN) {
2044 			/* QFull or DMAC Dying */
2045 			break;
2046 		} else {
2047 			/* Unacceptable request */
2048 			desc->status = DONE;
2049 			dev_err(pch->dmac->ddma.dev, "%s:%d Bad Desc(%d)\n",
2050 					__func__, __LINE__, desc->txd.cookie);
2051 			tasklet_schedule(&pch->task);
2052 		}
2053 	}
2054 }
2055 
2056 static void pl330_tasklet(unsigned long data)
2057 {
2058 	struct dma_pl330_chan *pch = (struct dma_pl330_chan *)data;
2059 	struct dma_pl330_desc *desc, *_dt;
2060 	unsigned long flags;
2061 	bool power_down = false;
2062 
2063 	spin_lock_irqsave(&pch->lock, flags);
2064 
2065 	/* Pick up ripe tomatoes */
2066 	list_for_each_entry_safe(desc, _dt, &pch->work_list, node)
2067 		if (desc->status == DONE) {
2068 			if (!pch->cyclic)
2069 				dma_cookie_complete(&desc->txd);
2070 			list_move_tail(&desc->node, &pch->completed_list);
2071 		}
2072 
2073 	/* Try to submit a req imm. next to the last completed cookie */
2074 	fill_queue(pch);
2075 
2076 	if (list_empty(&pch->work_list)) {
2077 		spin_lock(&pch->thread->dmac->lock);
2078 		_stop(pch->thread);
2079 		spin_unlock(&pch->thread->dmac->lock);
2080 		power_down = true;
2081 		pch->active = false;
2082 	} else {
2083 		/* Make sure the PL330 Channel thread is active */
2084 		spin_lock(&pch->thread->dmac->lock);
2085 		_start(pch->thread);
2086 		spin_unlock(&pch->thread->dmac->lock);
2087 	}
2088 
2089 	while (!list_empty(&pch->completed_list)) {
2090 		struct dmaengine_desc_callback cb;
2091 
2092 		desc = list_first_entry(&pch->completed_list,
2093 					struct dma_pl330_desc, node);
2094 
2095 		dmaengine_desc_get_callback(&desc->txd, &cb);
2096 
2097 		if (pch->cyclic) {
2098 			desc->status = PREP;
2099 			list_move_tail(&desc->node, &pch->work_list);
2100 			if (power_down) {
2101 				pch->active = true;
2102 				spin_lock(&pch->thread->dmac->lock);
2103 				_start(pch->thread);
2104 				spin_unlock(&pch->thread->dmac->lock);
2105 				power_down = false;
2106 			}
2107 		} else {
2108 			desc->status = FREE;
2109 			list_move_tail(&desc->node, &pch->dmac->desc_pool);
2110 		}
2111 
2112 		dma_descriptor_unmap(&desc->txd);
2113 
2114 		if (dmaengine_desc_callback_valid(&cb)) {
2115 			spin_unlock_irqrestore(&pch->lock, flags);
2116 			dmaengine_desc_callback_invoke(&cb, NULL);
2117 			spin_lock_irqsave(&pch->lock, flags);
2118 		}
2119 	}
2120 	spin_unlock_irqrestore(&pch->lock, flags);
2121 
2122 	/* If work list empty, power down */
2123 	if (power_down) {
2124 		pm_runtime_mark_last_busy(pch->dmac->ddma.dev);
2125 		pm_runtime_put_autosuspend(pch->dmac->ddma.dev);
2126 	}
2127 }
2128 
2129 static struct dma_chan *of_dma_pl330_xlate(struct of_phandle_args *dma_spec,
2130 						struct of_dma *ofdma)
2131 {
2132 	int count = dma_spec->args_count;
2133 	struct pl330_dmac *pl330 = ofdma->of_dma_data;
2134 	unsigned int chan_id;
2135 
2136 	if (!pl330)
2137 		return NULL;
2138 
2139 	if (count != 1)
2140 		return NULL;
2141 
2142 	chan_id = dma_spec->args[0];
2143 	if (chan_id >= pl330->num_peripherals)
2144 		return NULL;
2145 
2146 	return dma_get_slave_channel(&pl330->peripherals[chan_id].chan);
2147 }
2148 
2149 static int pl330_alloc_chan_resources(struct dma_chan *chan)
2150 {
2151 	struct dma_pl330_chan *pch = to_pchan(chan);
2152 	struct pl330_dmac *pl330 = pch->dmac;
2153 	unsigned long flags;
2154 
2155 	spin_lock_irqsave(&pl330->lock, flags);
2156 
2157 	dma_cookie_init(chan);
2158 	pch->cyclic = false;
2159 
2160 	pch->thread = pl330_request_channel(pl330);
2161 	if (!pch->thread) {
2162 		spin_unlock_irqrestore(&pl330->lock, flags);
2163 		return -ENOMEM;
2164 	}
2165 
2166 	tasklet_init(&pch->task, pl330_tasklet, (unsigned long) pch);
2167 
2168 	spin_unlock_irqrestore(&pl330->lock, flags);
2169 
2170 	return 1;
2171 }
2172 
2173 /*
2174  * We need the data direction between the DMAC (the dma-mapping "device") and
2175  * the FIFO (the dmaengine "dev"), from the FIFO's point of view. Confusing!
2176  */
2177 static enum dma_data_direction
2178 pl330_dma_slave_map_dir(enum dma_transfer_direction dir)
2179 {
2180 	switch (dir) {
2181 	case DMA_MEM_TO_DEV:
2182 		return DMA_FROM_DEVICE;
2183 	case DMA_DEV_TO_MEM:
2184 		return DMA_TO_DEVICE;
2185 	case DMA_DEV_TO_DEV:
2186 		return DMA_BIDIRECTIONAL;
2187 	default:
2188 		return DMA_NONE;
2189 	}
2190 }
2191 
2192 static void pl330_unprep_slave_fifo(struct dma_pl330_chan *pch)
2193 {
2194 	if (pch->dir != DMA_NONE)
2195 		dma_unmap_resource(pch->chan.device->dev, pch->fifo_dma,
2196 				   1 << pch->burst_sz, pch->dir, 0);
2197 	pch->dir = DMA_NONE;
2198 }
2199 
2200 
2201 static bool pl330_prep_slave_fifo(struct dma_pl330_chan *pch,
2202 				  enum dma_transfer_direction dir)
2203 {
2204 	struct device *dev = pch->chan.device->dev;
2205 	enum dma_data_direction dma_dir = pl330_dma_slave_map_dir(dir);
2206 
2207 	/* Already mapped for this config? */
2208 	if (pch->dir == dma_dir)
2209 		return true;
2210 
2211 	pl330_unprep_slave_fifo(pch);
2212 	pch->fifo_dma = dma_map_resource(dev, pch->fifo_addr,
2213 					 1 << pch->burst_sz, dma_dir, 0);
2214 	if (dma_mapping_error(dev, pch->fifo_dma))
2215 		return false;
2216 
2217 	pch->dir = dma_dir;
2218 	return true;
2219 }
2220 
2221 static int fixup_burst_len(int max_burst_len, int quirks)
2222 {
2223 	if (quirks & PL330_QUIRK_BROKEN_NO_FLUSHP)
2224 		return 1;
2225 	else if (max_burst_len > PL330_MAX_BURST)
2226 		return PL330_MAX_BURST;
2227 	else if (max_burst_len < 1)
2228 		return 1;
2229 	else
2230 		return max_burst_len;
2231 }
2232 
2233 static int pl330_config_write(struct dma_chan *chan,
2234 			struct dma_slave_config *slave_config,
2235 			enum dma_transfer_direction direction)
2236 {
2237 	struct dma_pl330_chan *pch = to_pchan(chan);
2238 
2239 	pl330_unprep_slave_fifo(pch);
2240 	if (direction == DMA_MEM_TO_DEV) {
2241 		if (slave_config->dst_addr)
2242 			pch->fifo_addr = slave_config->dst_addr;
2243 		if (slave_config->dst_addr_width)
2244 			pch->burst_sz = __ffs(slave_config->dst_addr_width);
2245 		pch->burst_len = fixup_burst_len(slave_config->dst_maxburst,
2246 			pch->dmac->quirks);
2247 	} else if (direction == DMA_DEV_TO_MEM) {
2248 		if (slave_config->src_addr)
2249 			pch->fifo_addr = slave_config->src_addr;
2250 		if (slave_config->src_addr_width)
2251 			pch->burst_sz = __ffs(slave_config->src_addr_width);
2252 		pch->burst_len = fixup_burst_len(slave_config->src_maxburst,
2253 			pch->dmac->quirks);
2254 	}
2255 
2256 	return 0;
2257 }
2258 
2259 static int pl330_config(struct dma_chan *chan,
2260 			struct dma_slave_config *slave_config)
2261 {
2262 	struct dma_pl330_chan *pch = to_pchan(chan);
2263 
2264 	memcpy(&pch->slave_config, slave_config, sizeof(*slave_config));
2265 
2266 	return 0;
2267 }
2268 
2269 static int pl330_terminate_all(struct dma_chan *chan)
2270 {
2271 	struct dma_pl330_chan *pch = to_pchan(chan);
2272 	struct dma_pl330_desc *desc;
2273 	unsigned long flags;
2274 	struct pl330_dmac *pl330 = pch->dmac;
2275 	bool power_down = false;
2276 
2277 	pm_runtime_get_sync(pl330->ddma.dev);
2278 	spin_lock_irqsave(&pch->lock, flags);
2279 
2280 	spin_lock(&pl330->lock);
2281 	_stop(pch->thread);
2282 	pch->thread->req[0].desc = NULL;
2283 	pch->thread->req[1].desc = NULL;
2284 	pch->thread->req_running = -1;
2285 	spin_unlock(&pl330->lock);
2286 
2287 	power_down = pch->active;
2288 	pch->active = false;
2289 
2290 	/* Mark all desc done */
2291 	list_for_each_entry(desc, &pch->submitted_list, node) {
2292 		desc->status = FREE;
2293 		dma_cookie_complete(&desc->txd);
2294 	}
2295 
2296 	list_for_each_entry(desc, &pch->work_list , node) {
2297 		desc->status = FREE;
2298 		dma_cookie_complete(&desc->txd);
2299 	}
2300 
2301 	list_splice_tail_init(&pch->submitted_list, &pl330->desc_pool);
2302 	list_splice_tail_init(&pch->work_list, &pl330->desc_pool);
2303 	list_splice_tail_init(&pch->completed_list, &pl330->desc_pool);
2304 	spin_unlock_irqrestore(&pch->lock, flags);
2305 	pm_runtime_mark_last_busy(pl330->ddma.dev);
2306 	if (power_down)
2307 		pm_runtime_put_autosuspend(pl330->ddma.dev);
2308 	pm_runtime_put_autosuspend(pl330->ddma.dev);
2309 
2310 	return 0;
2311 }
2312 
2313 /*
2314  * We don't support DMA_RESUME command because of hardware
2315  * limitations, so after pausing the channel we cannot restore
2316  * it to active state. We have to terminate channel and setup
2317  * DMA transfer again. This pause feature was implemented to
2318  * allow safely read residue before channel termination.
2319  */
2320 static int pl330_pause(struct dma_chan *chan)
2321 {
2322 	struct dma_pl330_chan *pch = to_pchan(chan);
2323 	struct pl330_dmac *pl330 = pch->dmac;
2324 	unsigned long flags;
2325 
2326 	pm_runtime_get_sync(pl330->ddma.dev);
2327 	spin_lock_irqsave(&pch->lock, flags);
2328 
2329 	spin_lock(&pl330->lock);
2330 	_stop(pch->thread);
2331 	spin_unlock(&pl330->lock);
2332 
2333 	spin_unlock_irqrestore(&pch->lock, flags);
2334 	pm_runtime_mark_last_busy(pl330->ddma.dev);
2335 	pm_runtime_put_autosuspend(pl330->ddma.dev);
2336 
2337 	return 0;
2338 }
2339 
2340 static void pl330_free_chan_resources(struct dma_chan *chan)
2341 {
2342 	struct dma_pl330_chan *pch = to_pchan(chan);
2343 	struct pl330_dmac *pl330 = pch->dmac;
2344 	unsigned long flags;
2345 
2346 	tasklet_kill(&pch->task);
2347 
2348 	pm_runtime_get_sync(pch->dmac->ddma.dev);
2349 	spin_lock_irqsave(&pl330->lock, flags);
2350 
2351 	pl330_release_channel(pch->thread);
2352 	pch->thread = NULL;
2353 
2354 	if (pch->cyclic)
2355 		list_splice_tail_init(&pch->work_list, &pch->dmac->desc_pool);
2356 
2357 	spin_unlock_irqrestore(&pl330->lock, flags);
2358 	pm_runtime_mark_last_busy(pch->dmac->ddma.dev);
2359 	pm_runtime_put_autosuspend(pch->dmac->ddma.dev);
2360 	pl330_unprep_slave_fifo(pch);
2361 }
2362 
2363 static int pl330_get_current_xferred_count(struct dma_pl330_chan *pch,
2364 					   struct dma_pl330_desc *desc)
2365 {
2366 	struct pl330_thread *thrd = pch->thread;
2367 	struct pl330_dmac *pl330 = pch->dmac;
2368 	void __iomem *regs = thrd->dmac->base;
2369 	u32 val, addr;
2370 
2371 	pm_runtime_get_sync(pl330->ddma.dev);
2372 	val = addr = 0;
2373 	if (desc->rqcfg.src_inc) {
2374 		val = readl(regs + SA(thrd->id));
2375 		addr = desc->px.src_addr;
2376 	} else {
2377 		val = readl(regs + DA(thrd->id));
2378 		addr = desc->px.dst_addr;
2379 	}
2380 	pm_runtime_mark_last_busy(pch->dmac->ddma.dev);
2381 	pm_runtime_put_autosuspend(pl330->ddma.dev);
2382 
2383 	/* If DMAMOV hasn't finished yet, SAR/DAR can be zero */
2384 	if (!val)
2385 		return 0;
2386 
2387 	return val - addr;
2388 }
2389 
2390 static enum dma_status
2391 pl330_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
2392 		 struct dma_tx_state *txstate)
2393 {
2394 	enum dma_status ret;
2395 	unsigned long flags;
2396 	struct dma_pl330_desc *desc, *running = NULL, *last_enq = NULL;
2397 	struct dma_pl330_chan *pch = to_pchan(chan);
2398 	unsigned int transferred, residual = 0;
2399 
2400 	ret = dma_cookie_status(chan, cookie, txstate);
2401 
2402 	if (!txstate)
2403 		return ret;
2404 
2405 	if (ret == DMA_COMPLETE)
2406 		goto out;
2407 
2408 	spin_lock_irqsave(&pch->lock, flags);
2409 	spin_lock(&pch->thread->dmac->lock);
2410 
2411 	if (pch->thread->req_running != -1)
2412 		running = pch->thread->req[pch->thread->req_running].desc;
2413 
2414 	last_enq = pch->thread->req[pch->thread->lstenq].desc;
2415 
2416 	/* Check in pending list */
2417 	list_for_each_entry(desc, &pch->work_list, node) {
2418 		if (desc->status == DONE)
2419 			transferred = desc->bytes_requested;
2420 		else if (running && desc == running)
2421 			transferred =
2422 				pl330_get_current_xferred_count(pch, desc);
2423 		else if (desc->status == BUSY)
2424 			/*
2425 			 * Busy but not running means either just enqueued,
2426 			 * or finished and not yet marked done
2427 			 */
2428 			if (desc == last_enq)
2429 				transferred = 0;
2430 			else
2431 				transferred = desc->bytes_requested;
2432 		else
2433 			transferred = 0;
2434 		residual += desc->bytes_requested - transferred;
2435 		if (desc->txd.cookie == cookie) {
2436 			switch (desc->status) {
2437 			case DONE:
2438 				ret = DMA_COMPLETE;
2439 				break;
2440 			case PREP:
2441 			case BUSY:
2442 				ret = DMA_IN_PROGRESS;
2443 				break;
2444 			default:
2445 				WARN_ON(1);
2446 			}
2447 			break;
2448 		}
2449 		if (desc->last)
2450 			residual = 0;
2451 	}
2452 	spin_unlock(&pch->thread->dmac->lock);
2453 	spin_unlock_irqrestore(&pch->lock, flags);
2454 
2455 out:
2456 	dma_set_residue(txstate, residual);
2457 
2458 	return ret;
2459 }
2460 
2461 static void pl330_issue_pending(struct dma_chan *chan)
2462 {
2463 	struct dma_pl330_chan *pch = to_pchan(chan);
2464 	unsigned long flags;
2465 
2466 	spin_lock_irqsave(&pch->lock, flags);
2467 	if (list_empty(&pch->work_list)) {
2468 		/*
2469 		 * Warn on nothing pending. Empty submitted_list may
2470 		 * break our pm_runtime usage counter as it is
2471 		 * updated on work_list emptiness status.
2472 		 */
2473 		WARN_ON(list_empty(&pch->submitted_list));
2474 		pch->active = true;
2475 		pm_runtime_get_sync(pch->dmac->ddma.dev);
2476 	}
2477 	list_splice_tail_init(&pch->submitted_list, &pch->work_list);
2478 	spin_unlock_irqrestore(&pch->lock, flags);
2479 
2480 	pl330_tasklet((unsigned long)pch);
2481 }
2482 
2483 /*
2484  * We returned the last one of the circular list of descriptor(s)
2485  * from prep_xxx, so the argument to submit corresponds to the last
2486  * descriptor of the list.
2487  */
2488 static dma_cookie_t pl330_tx_submit(struct dma_async_tx_descriptor *tx)
2489 {
2490 	struct dma_pl330_desc *desc, *last = to_desc(tx);
2491 	struct dma_pl330_chan *pch = to_pchan(tx->chan);
2492 	dma_cookie_t cookie;
2493 	unsigned long flags;
2494 
2495 	spin_lock_irqsave(&pch->lock, flags);
2496 
2497 	/* Assign cookies to all nodes */
2498 	while (!list_empty(&last->node)) {
2499 		desc = list_entry(last->node.next, struct dma_pl330_desc, node);
2500 		if (pch->cyclic) {
2501 			desc->txd.callback = last->txd.callback;
2502 			desc->txd.callback_param = last->txd.callback_param;
2503 		}
2504 		desc->last = false;
2505 
2506 		dma_cookie_assign(&desc->txd);
2507 
2508 		list_move_tail(&desc->node, &pch->submitted_list);
2509 	}
2510 
2511 	last->last = true;
2512 	cookie = dma_cookie_assign(&last->txd);
2513 	list_add_tail(&last->node, &pch->submitted_list);
2514 	spin_unlock_irqrestore(&pch->lock, flags);
2515 
2516 	return cookie;
2517 }
2518 
2519 static inline void _init_desc(struct dma_pl330_desc *desc)
2520 {
2521 	desc->rqcfg.swap = SWAP_NO;
2522 	desc->rqcfg.scctl = CCTRL0;
2523 	desc->rqcfg.dcctl = CCTRL0;
2524 	desc->txd.tx_submit = pl330_tx_submit;
2525 
2526 	INIT_LIST_HEAD(&desc->node);
2527 }
2528 
2529 /* Returns the number of descriptors added to the DMAC pool */
2530 static int add_desc(struct list_head *pool, spinlock_t *lock,
2531 		    gfp_t flg, int count)
2532 {
2533 	struct dma_pl330_desc *desc;
2534 	unsigned long flags;
2535 	int i;
2536 
2537 	desc = kcalloc(count, sizeof(*desc), flg);
2538 	if (!desc)
2539 		return 0;
2540 
2541 	spin_lock_irqsave(lock, flags);
2542 
2543 	for (i = 0; i < count; i++) {
2544 		_init_desc(&desc[i]);
2545 		list_add_tail(&desc[i].node, pool);
2546 	}
2547 
2548 	spin_unlock_irqrestore(lock, flags);
2549 
2550 	return count;
2551 }
2552 
2553 static struct dma_pl330_desc *pluck_desc(struct list_head *pool,
2554 					 spinlock_t *lock)
2555 {
2556 	struct dma_pl330_desc *desc = NULL;
2557 	unsigned long flags;
2558 
2559 	spin_lock_irqsave(lock, flags);
2560 
2561 	if (!list_empty(pool)) {
2562 		desc = list_entry(pool->next,
2563 				struct dma_pl330_desc, node);
2564 
2565 		list_del_init(&desc->node);
2566 
2567 		desc->status = PREP;
2568 		desc->txd.callback = NULL;
2569 	}
2570 
2571 	spin_unlock_irqrestore(lock, flags);
2572 
2573 	return desc;
2574 }
2575 
2576 static struct dma_pl330_desc *pl330_get_desc(struct dma_pl330_chan *pch)
2577 {
2578 	struct pl330_dmac *pl330 = pch->dmac;
2579 	u8 *peri_id = pch->chan.private;
2580 	struct dma_pl330_desc *desc;
2581 
2582 	/* Pluck one desc from the pool of DMAC */
2583 	desc = pluck_desc(&pl330->desc_pool, &pl330->pool_lock);
2584 
2585 	/* If the DMAC pool is empty, alloc new */
2586 	if (!desc) {
2587 		DEFINE_SPINLOCK(lock);
2588 		LIST_HEAD(pool);
2589 
2590 		if (!add_desc(&pool, &lock, GFP_ATOMIC, 1))
2591 			return NULL;
2592 
2593 		desc = pluck_desc(&pool, &lock);
2594 		WARN_ON(!desc || !list_empty(&pool));
2595 	}
2596 
2597 	/* Initialize the descriptor */
2598 	desc->pchan = pch;
2599 	desc->txd.cookie = 0;
2600 	async_tx_ack(&desc->txd);
2601 
2602 	desc->peri = peri_id ? pch->chan.chan_id : 0;
2603 	desc->rqcfg.pcfg = &pch->dmac->pcfg;
2604 
2605 	dma_async_tx_descriptor_init(&desc->txd, &pch->chan);
2606 
2607 	return desc;
2608 }
2609 
2610 static inline void fill_px(struct pl330_xfer *px,
2611 		dma_addr_t dst, dma_addr_t src, size_t len)
2612 {
2613 	px->bytes = len;
2614 	px->dst_addr = dst;
2615 	px->src_addr = src;
2616 }
2617 
2618 static struct dma_pl330_desc *
2619 __pl330_prep_dma_memcpy(struct dma_pl330_chan *pch, dma_addr_t dst,
2620 		dma_addr_t src, size_t len)
2621 {
2622 	struct dma_pl330_desc *desc = pl330_get_desc(pch);
2623 
2624 	if (!desc) {
2625 		dev_err(pch->dmac->ddma.dev, "%s:%d Unable to fetch desc\n",
2626 			__func__, __LINE__);
2627 		return NULL;
2628 	}
2629 
2630 	/*
2631 	 * Ideally we should lookout for reqs bigger than
2632 	 * those that can be programmed with 256 bytes of
2633 	 * MC buffer, but considering a req size is seldom
2634 	 * going to be word-unaligned and more than 200MB,
2635 	 * we take it easy.
2636 	 * Also, should the limit is reached we'd rather
2637 	 * have the platform increase MC buffer size than
2638 	 * complicating this API driver.
2639 	 */
2640 	fill_px(&desc->px, dst, src, len);
2641 
2642 	return desc;
2643 }
2644 
2645 /* Call after fixing burst size */
2646 static inline int get_burst_len(struct dma_pl330_desc *desc, size_t len)
2647 {
2648 	struct dma_pl330_chan *pch = desc->pchan;
2649 	struct pl330_dmac *pl330 = pch->dmac;
2650 	int burst_len;
2651 
2652 	burst_len = pl330->pcfg.data_bus_width / 8;
2653 	burst_len *= pl330->pcfg.data_buf_dep / pl330->pcfg.num_chan;
2654 	burst_len >>= desc->rqcfg.brst_size;
2655 
2656 	/* src/dst_burst_len can't be more than 16 */
2657 	if (burst_len > PL330_MAX_BURST)
2658 		burst_len = PL330_MAX_BURST;
2659 
2660 	return burst_len;
2661 }
2662 
2663 static struct dma_async_tx_descriptor *pl330_prep_dma_cyclic(
2664 		struct dma_chan *chan, dma_addr_t dma_addr, size_t len,
2665 		size_t period_len, enum dma_transfer_direction direction,
2666 		unsigned long flags)
2667 {
2668 	struct dma_pl330_desc *desc = NULL, *first = NULL;
2669 	struct dma_pl330_chan *pch = to_pchan(chan);
2670 	struct pl330_dmac *pl330 = pch->dmac;
2671 	unsigned int i;
2672 	dma_addr_t dst;
2673 	dma_addr_t src;
2674 
2675 	if (len % period_len != 0)
2676 		return NULL;
2677 
2678 	if (!is_slave_direction(direction)) {
2679 		dev_err(pch->dmac->ddma.dev, "%s:%d Invalid dma direction\n",
2680 		__func__, __LINE__);
2681 		return NULL;
2682 	}
2683 
2684 	pl330_config_write(chan, &pch->slave_config, direction);
2685 
2686 	if (!pl330_prep_slave_fifo(pch, direction))
2687 		return NULL;
2688 
2689 	for (i = 0; i < len / period_len; i++) {
2690 		desc = pl330_get_desc(pch);
2691 		if (!desc) {
2692 			dev_err(pch->dmac->ddma.dev, "%s:%d Unable to fetch desc\n",
2693 				__func__, __LINE__);
2694 
2695 			if (!first)
2696 				return NULL;
2697 
2698 			spin_lock_irqsave(&pl330->pool_lock, flags);
2699 
2700 			while (!list_empty(&first->node)) {
2701 				desc = list_entry(first->node.next,
2702 						struct dma_pl330_desc, node);
2703 				list_move_tail(&desc->node, &pl330->desc_pool);
2704 			}
2705 
2706 			list_move_tail(&first->node, &pl330->desc_pool);
2707 
2708 			spin_unlock_irqrestore(&pl330->pool_lock, flags);
2709 
2710 			return NULL;
2711 		}
2712 
2713 		switch (direction) {
2714 		case DMA_MEM_TO_DEV:
2715 			desc->rqcfg.src_inc = 1;
2716 			desc->rqcfg.dst_inc = 0;
2717 			src = dma_addr;
2718 			dst = pch->fifo_dma;
2719 			break;
2720 		case DMA_DEV_TO_MEM:
2721 			desc->rqcfg.src_inc = 0;
2722 			desc->rqcfg.dst_inc = 1;
2723 			src = pch->fifo_dma;
2724 			dst = dma_addr;
2725 			break;
2726 		default:
2727 			break;
2728 		}
2729 
2730 		desc->rqtype = direction;
2731 		desc->rqcfg.brst_size = pch->burst_sz;
2732 		desc->rqcfg.brst_len = pch->burst_len;
2733 		desc->bytes_requested = period_len;
2734 		fill_px(&desc->px, dst, src, period_len);
2735 
2736 		if (!first)
2737 			first = desc;
2738 		else
2739 			list_add_tail(&desc->node, &first->node);
2740 
2741 		dma_addr += period_len;
2742 	}
2743 
2744 	if (!desc)
2745 		return NULL;
2746 
2747 	pch->cyclic = true;
2748 	desc->txd.flags = flags;
2749 
2750 	return &desc->txd;
2751 }
2752 
2753 static struct dma_async_tx_descriptor *
2754 pl330_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dst,
2755 		dma_addr_t src, size_t len, unsigned long flags)
2756 {
2757 	struct dma_pl330_desc *desc;
2758 	struct dma_pl330_chan *pch = to_pchan(chan);
2759 	struct pl330_dmac *pl330;
2760 	int burst;
2761 
2762 	if (unlikely(!pch || !len))
2763 		return NULL;
2764 
2765 	pl330 = pch->dmac;
2766 
2767 	desc = __pl330_prep_dma_memcpy(pch, dst, src, len);
2768 	if (!desc)
2769 		return NULL;
2770 
2771 	desc->rqcfg.src_inc = 1;
2772 	desc->rqcfg.dst_inc = 1;
2773 	desc->rqtype = DMA_MEM_TO_MEM;
2774 
2775 	/* Select max possible burst size */
2776 	burst = pl330->pcfg.data_bus_width / 8;
2777 
2778 	/*
2779 	 * Make sure we use a burst size that aligns with all the memcpy
2780 	 * parameters because our DMA programming algorithm doesn't cope with
2781 	 * transfers which straddle an entry in the DMA device's MFIFO.
2782 	 */
2783 	while ((src | dst | len) & (burst - 1))
2784 		burst /= 2;
2785 
2786 	desc->rqcfg.brst_size = 0;
2787 	while (burst != (1 << desc->rqcfg.brst_size))
2788 		desc->rqcfg.brst_size++;
2789 
2790 	/*
2791 	 * If burst size is smaller than bus width then make sure we only
2792 	 * transfer one at a time to avoid a burst stradling an MFIFO entry.
2793 	 */
2794 	if (desc->rqcfg.brst_size * 8 < pl330->pcfg.data_bus_width)
2795 		desc->rqcfg.brst_len = 1;
2796 
2797 	desc->rqcfg.brst_len = get_burst_len(desc, len);
2798 	desc->bytes_requested = len;
2799 
2800 	desc->txd.flags = flags;
2801 
2802 	return &desc->txd;
2803 }
2804 
2805 static void __pl330_giveback_desc(struct pl330_dmac *pl330,
2806 				  struct dma_pl330_desc *first)
2807 {
2808 	unsigned long flags;
2809 	struct dma_pl330_desc *desc;
2810 
2811 	if (!first)
2812 		return;
2813 
2814 	spin_lock_irqsave(&pl330->pool_lock, flags);
2815 
2816 	while (!list_empty(&first->node)) {
2817 		desc = list_entry(first->node.next,
2818 				struct dma_pl330_desc, node);
2819 		list_move_tail(&desc->node, &pl330->desc_pool);
2820 	}
2821 
2822 	list_move_tail(&first->node, &pl330->desc_pool);
2823 
2824 	spin_unlock_irqrestore(&pl330->pool_lock, flags);
2825 }
2826 
2827 static struct dma_async_tx_descriptor *
2828 pl330_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2829 		unsigned int sg_len, enum dma_transfer_direction direction,
2830 		unsigned long flg, void *context)
2831 {
2832 	struct dma_pl330_desc *first, *desc = NULL;
2833 	struct dma_pl330_chan *pch = to_pchan(chan);
2834 	struct scatterlist *sg;
2835 	int i;
2836 
2837 	if (unlikely(!pch || !sgl || !sg_len))
2838 		return NULL;
2839 
2840 	pl330_config_write(chan, &pch->slave_config, direction);
2841 
2842 	if (!pl330_prep_slave_fifo(pch, direction))
2843 		return NULL;
2844 
2845 	first = NULL;
2846 
2847 	for_each_sg(sgl, sg, sg_len, i) {
2848 
2849 		desc = pl330_get_desc(pch);
2850 		if (!desc) {
2851 			struct pl330_dmac *pl330 = pch->dmac;
2852 
2853 			dev_err(pch->dmac->ddma.dev,
2854 				"%s:%d Unable to fetch desc\n",
2855 				__func__, __LINE__);
2856 			__pl330_giveback_desc(pl330, first);
2857 
2858 			return NULL;
2859 		}
2860 
2861 		if (!first)
2862 			first = desc;
2863 		else
2864 			list_add_tail(&desc->node, &first->node);
2865 
2866 		if (direction == DMA_MEM_TO_DEV) {
2867 			desc->rqcfg.src_inc = 1;
2868 			desc->rqcfg.dst_inc = 0;
2869 			fill_px(&desc->px, pch->fifo_dma, sg_dma_address(sg),
2870 				sg_dma_len(sg));
2871 		} else {
2872 			desc->rqcfg.src_inc = 0;
2873 			desc->rqcfg.dst_inc = 1;
2874 			fill_px(&desc->px, sg_dma_address(sg), pch->fifo_dma,
2875 				sg_dma_len(sg));
2876 		}
2877 
2878 		desc->rqcfg.brst_size = pch->burst_sz;
2879 		desc->rqcfg.brst_len = pch->burst_len;
2880 		desc->rqtype = direction;
2881 		desc->bytes_requested = sg_dma_len(sg);
2882 	}
2883 
2884 	/* Return the last desc in the chain */
2885 	desc->txd.flags = flg;
2886 	return &desc->txd;
2887 }
2888 
2889 static irqreturn_t pl330_irq_handler(int irq, void *data)
2890 {
2891 	if (pl330_update(data))
2892 		return IRQ_HANDLED;
2893 	else
2894 		return IRQ_NONE;
2895 }
2896 
2897 #define PL330_DMA_BUSWIDTHS \
2898 	BIT(DMA_SLAVE_BUSWIDTH_UNDEFINED) | \
2899 	BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
2900 	BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
2901 	BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) | \
2902 	BIT(DMA_SLAVE_BUSWIDTH_8_BYTES)
2903 
2904 #ifdef CONFIG_DEBUG_FS
2905 static int pl330_debugfs_show(struct seq_file *s, void *data)
2906 {
2907 	struct pl330_dmac *pl330 = s->private;
2908 	int chans, pchs, ch, pr;
2909 
2910 	chans = pl330->pcfg.num_chan;
2911 	pchs = pl330->num_peripherals;
2912 
2913 	seq_puts(s, "PL330 physical channels:\n");
2914 	seq_puts(s, "THREAD:\t\tCHANNEL:\n");
2915 	seq_puts(s, "--------\t-----\n");
2916 	for (ch = 0; ch < chans; ch++) {
2917 		struct pl330_thread *thrd = &pl330->channels[ch];
2918 		int found = -1;
2919 
2920 		for (pr = 0; pr < pchs; pr++) {
2921 			struct dma_pl330_chan *pch = &pl330->peripherals[pr];
2922 
2923 			if (!pch->thread || thrd->id != pch->thread->id)
2924 				continue;
2925 
2926 			found = pr;
2927 		}
2928 
2929 		seq_printf(s, "%d\t\t", thrd->id);
2930 		if (found == -1)
2931 			seq_puts(s, "--\n");
2932 		else
2933 			seq_printf(s, "%d\n", found);
2934 	}
2935 
2936 	return 0;
2937 }
2938 
2939 DEFINE_SHOW_ATTRIBUTE(pl330_debugfs);
2940 
2941 static inline void init_pl330_debugfs(struct pl330_dmac *pl330)
2942 {
2943 	debugfs_create_file(dev_name(pl330->ddma.dev),
2944 			    S_IFREG | 0444, NULL, pl330,
2945 			    &pl330_debugfs_fops);
2946 }
2947 #else
2948 static inline void init_pl330_debugfs(struct pl330_dmac *pl330)
2949 {
2950 }
2951 #endif
2952 
2953 /*
2954  * Runtime PM callbacks are provided by amba/bus.c driver.
2955  *
2956  * It is assumed here that IRQ safe runtime PM is chosen in probe and amba
2957  * bus driver will only disable/enable the clock in runtime PM callbacks.
2958  */
2959 static int __maybe_unused pl330_suspend(struct device *dev)
2960 {
2961 	struct amba_device *pcdev = to_amba_device(dev);
2962 
2963 	pm_runtime_disable(dev);
2964 
2965 	if (!pm_runtime_status_suspended(dev)) {
2966 		/* amba did not disable the clock */
2967 		amba_pclk_disable(pcdev);
2968 	}
2969 	amba_pclk_unprepare(pcdev);
2970 
2971 	return 0;
2972 }
2973 
2974 static int __maybe_unused pl330_resume(struct device *dev)
2975 {
2976 	struct amba_device *pcdev = to_amba_device(dev);
2977 	int ret;
2978 
2979 	ret = amba_pclk_prepare(pcdev);
2980 	if (ret)
2981 		return ret;
2982 
2983 	if (!pm_runtime_status_suspended(dev))
2984 		ret = amba_pclk_enable(pcdev);
2985 
2986 	pm_runtime_enable(dev);
2987 
2988 	return ret;
2989 }
2990 
2991 static SIMPLE_DEV_PM_OPS(pl330_pm, pl330_suspend, pl330_resume);
2992 
2993 static int
2994 pl330_probe(struct amba_device *adev, const struct amba_id *id)
2995 {
2996 	struct pl330_config *pcfg;
2997 	struct pl330_dmac *pl330;
2998 	struct dma_pl330_chan *pch, *_p;
2999 	struct dma_device *pd;
3000 	struct resource *res;
3001 	int i, ret, irq;
3002 	int num_chan;
3003 	struct device_node *np = adev->dev.of_node;
3004 
3005 	ret = dma_set_mask_and_coherent(&adev->dev, DMA_BIT_MASK(32));
3006 	if (ret)
3007 		return ret;
3008 
3009 	/* Allocate a new DMAC and its Channels */
3010 	pl330 = devm_kzalloc(&adev->dev, sizeof(*pl330), GFP_KERNEL);
3011 	if (!pl330)
3012 		return -ENOMEM;
3013 
3014 	pd = &pl330->ddma;
3015 	pd->dev = &adev->dev;
3016 
3017 	pl330->mcbufsz = 0;
3018 
3019 	/* get quirk */
3020 	for (i = 0; i < ARRAY_SIZE(of_quirks); i++)
3021 		if (of_property_read_bool(np, of_quirks[i].quirk))
3022 			pl330->quirks |= of_quirks[i].id;
3023 
3024 	res = &adev->res;
3025 	pl330->base = devm_ioremap_resource(&adev->dev, res);
3026 	if (IS_ERR(pl330->base))
3027 		return PTR_ERR(pl330->base);
3028 
3029 	amba_set_drvdata(adev, pl330);
3030 
3031 	for (i = 0; i < AMBA_NR_IRQS; i++) {
3032 		irq = adev->irq[i];
3033 		if (irq) {
3034 			ret = devm_request_irq(&adev->dev, irq,
3035 					       pl330_irq_handler, 0,
3036 					       dev_name(&adev->dev), pl330);
3037 			if (ret)
3038 				return ret;
3039 		} else {
3040 			break;
3041 		}
3042 	}
3043 
3044 	pcfg = &pl330->pcfg;
3045 
3046 	pcfg->periph_id = adev->periphid;
3047 	ret = pl330_add(pl330);
3048 	if (ret)
3049 		return ret;
3050 
3051 	INIT_LIST_HEAD(&pl330->desc_pool);
3052 	spin_lock_init(&pl330->pool_lock);
3053 
3054 	/* Create a descriptor pool of default size */
3055 	if (!add_desc(&pl330->desc_pool, &pl330->pool_lock,
3056 		      GFP_KERNEL, NR_DEFAULT_DESC))
3057 		dev_warn(&adev->dev, "unable to allocate desc\n");
3058 
3059 	INIT_LIST_HEAD(&pd->channels);
3060 
3061 	/* Initialize channel parameters */
3062 	num_chan = max_t(int, pcfg->num_peri, pcfg->num_chan);
3063 
3064 	pl330->num_peripherals = num_chan;
3065 
3066 	pl330->peripherals = kcalloc(num_chan, sizeof(*pch), GFP_KERNEL);
3067 	if (!pl330->peripherals) {
3068 		ret = -ENOMEM;
3069 		goto probe_err2;
3070 	}
3071 
3072 	for (i = 0; i < num_chan; i++) {
3073 		pch = &pl330->peripherals[i];
3074 
3075 		pch->chan.private = adev->dev.of_node;
3076 		INIT_LIST_HEAD(&pch->submitted_list);
3077 		INIT_LIST_HEAD(&pch->work_list);
3078 		INIT_LIST_HEAD(&pch->completed_list);
3079 		spin_lock_init(&pch->lock);
3080 		pch->thread = NULL;
3081 		pch->chan.device = pd;
3082 		pch->dmac = pl330;
3083 		pch->dir = DMA_NONE;
3084 
3085 		/* Add the channel to the DMAC list */
3086 		list_add_tail(&pch->chan.device_node, &pd->channels);
3087 	}
3088 
3089 	dma_cap_set(DMA_MEMCPY, pd->cap_mask);
3090 	if (pcfg->num_peri) {
3091 		dma_cap_set(DMA_SLAVE, pd->cap_mask);
3092 		dma_cap_set(DMA_CYCLIC, pd->cap_mask);
3093 		dma_cap_set(DMA_PRIVATE, pd->cap_mask);
3094 	}
3095 
3096 	pd->device_alloc_chan_resources = pl330_alloc_chan_resources;
3097 	pd->device_free_chan_resources = pl330_free_chan_resources;
3098 	pd->device_prep_dma_memcpy = pl330_prep_dma_memcpy;
3099 	pd->device_prep_dma_cyclic = pl330_prep_dma_cyclic;
3100 	pd->device_tx_status = pl330_tx_status;
3101 	pd->device_prep_slave_sg = pl330_prep_slave_sg;
3102 	pd->device_config = pl330_config;
3103 	pd->device_pause = pl330_pause;
3104 	pd->device_terminate_all = pl330_terminate_all;
3105 	pd->device_issue_pending = pl330_issue_pending;
3106 	pd->src_addr_widths = PL330_DMA_BUSWIDTHS;
3107 	pd->dst_addr_widths = PL330_DMA_BUSWIDTHS;
3108 	pd->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
3109 	pd->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
3110 	pd->max_burst = ((pl330->quirks & PL330_QUIRK_BROKEN_NO_FLUSHP) ?
3111 			 1 : PL330_MAX_BURST);
3112 
3113 	ret = dma_async_device_register(pd);
3114 	if (ret) {
3115 		dev_err(&adev->dev, "unable to register DMAC\n");
3116 		goto probe_err3;
3117 	}
3118 
3119 	if (adev->dev.of_node) {
3120 		ret = of_dma_controller_register(adev->dev.of_node,
3121 					 of_dma_pl330_xlate, pl330);
3122 		if (ret) {
3123 			dev_err(&adev->dev,
3124 			"unable to register DMA to the generic DT DMA helpers\n");
3125 		}
3126 	}
3127 
3128 	adev->dev.dma_parms = &pl330->dma_parms;
3129 
3130 	/*
3131 	 * This is the limit for transfers with a buswidth of 1, larger
3132 	 * buswidths will have larger limits.
3133 	 */
3134 	ret = dma_set_max_seg_size(&adev->dev, 1900800);
3135 	if (ret)
3136 		dev_err(&adev->dev, "unable to set the seg size\n");
3137 
3138 
3139 	init_pl330_debugfs(pl330);
3140 	dev_info(&adev->dev,
3141 		"Loaded driver for PL330 DMAC-%x\n", adev->periphid);
3142 	dev_info(&adev->dev,
3143 		"\tDBUFF-%ux%ubytes Num_Chans-%u Num_Peri-%u Num_Events-%u\n",
3144 		pcfg->data_buf_dep, pcfg->data_bus_width / 8, pcfg->num_chan,
3145 		pcfg->num_peri, pcfg->num_events);
3146 
3147 	pm_runtime_irq_safe(&adev->dev);
3148 	pm_runtime_use_autosuspend(&adev->dev);
3149 	pm_runtime_set_autosuspend_delay(&adev->dev, PL330_AUTOSUSPEND_DELAY);
3150 	pm_runtime_mark_last_busy(&adev->dev);
3151 	pm_runtime_put_autosuspend(&adev->dev);
3152 
3153 	return 0;
3154 probe_err3:
3155 	/* Idle the DMAC */
3156 	list_for_each_entry_safe(pch, _p, &pl330->ddma.channels,
3157 			chan.device_node) {
3158 
3159 		/* Remove the channel */
3160 		list_del(&pch->chan.device_node);
3161 
3162 		/* Flush the channel */
3163 		if (pch->thread) {
3164 			pl330_terminate_all(&pch->chan);
3165 			pl330_free_chan_resources(&pch->chan);
3166 		}
3167 	}
3168 probe_err2:
3169 	pl330_del(pl330);
3170 
3171 	return ret;
3172 }
3173 
3174 static int pl330_remove(struct amba_device *adev)
3175 {
3176 	struct pl330_dmac *pl330 = amba_get_drvdata(adev);
3177 	struct dma_pl330_chan *pch, *_p;
3178 	int i, irq;
3179 
3180 	pm_runtime_get_noresume(pl330->ddma.dev);
3181 
3182 	if (adev->dev.of_node)
3183 		of_dma_controller_free(adev->dev.of_node);
3184 
3185 	for (i = 0; i < AMBA_NR_IRQS; i++) {
3186 		irq = adev->irq[i];
3187 		if (irq)
3188 			devm_free_irq(&adev->dev, irq, pl330);
3189 	}
3190 
3191 	dma_async_device_unregister(&pl330->ddma);
3192 
3193 	/* Idle the DMAC */
3194 	list_for_each_entry_safe(pch, _p, &pl330->ddma.channels,
3195 			chan.device_node) {
3196 
3197 		/* Remove the channel */
3198 		list_del(&pch->chan.device_node);
3199 
3200 		/* Flush the channel */
3201 		if (pch->thread) {
3202 			pl330_terminate_all(&pch->chan);
3203 			pl330_free_chan_resources(&pch->chan);
3204 		}
3205 	}
3206 
3207 	pl330_del(pl330);
3208 
3209 	return 0;
3210 }
3211 
3212 static const struct amba_id pl330_ids[] = {
3213 	{
3214 		.id	= 0x00041330,
3215 		.mask	= 0x000fffff,
3216 	},
3217 	{ 0, 0 },
3218 };
3219 
3220 MODULE_DEVICE_TABLE(amba, pl330_ids);
3221 
3222 static struct amba_driver pl330_driver = {
3223 	.drv = {
3224 		.owner = THIS_MODULE,
3225 		.name = "dma-pl330",
3226 		.pm = &pl330_pm,
3227 	},
3228 	.id_table = pl330_ids,
3229 	.probe = pl330_probe,
3230 	.remove = pl330_remove,
3231 };
3232 
3233 module_amba_driver(pl330_driver);
3234 
3235 MODULE_AUTHOR("Jaswinder Singh <jassisinghbrar@gmail.com>");
3236 MODULE_DESCRIPTION("API Driver for PL330 DMAC");
3237 MODULE_LICENSE("GPL");
3238