xref: /openbmc/linux/drivers/dma/bcm-sba-raid.c (revision 2127c01b)
1 /*
2  * Copyright (C) 2017 Broadcom
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License as
6  * published by the Free Software Foundation version 2.
7  *
8  * This program is distributed "as is" WITHOUT ANY WARRANTY of any
9  * kind, whether express or implied; without even the implied warranty
10  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  */
13 
14 /*
15  * Broadcom SBA RAID Driver
16  *
17  * The Broadcom stream buffer accelerator (SBA) provides offloading
18  * capabilities for RAID operations. The SBA offload engine is accessible
19  * via Broadcom SoC specific ring manager. Two or more offload engines
20  * can share same Broadcom SoC specific ring manager due to this Broadcom
21  * SoC specific ring manager driver is implemented as a mailbox controller
22  * driver and offload engine drivers are implemented as mallbox clients.
23  *
24  * Typically, Broadcom SoC specific ring manager will implement larger
25  * number of hardware rings over one or more SBA hardware devices. By
26  * design, the internal buffer size of SBA hardware device is limited
27  * but all offload operations supported by SBA can be broken down into
28  * multiple small size requests and executed parallely on multiple SBA
29  * hardware devices for achieving high through-put.
30  *
31  * The Broadcom SBA RAID driver does not require any register programming
32  * except submitting request to SBA hardware device via mailbox channels.
33  * This driver implements a DMA device with one DMA channel using a single
34  * mailbox channel provided by Broadcom SoC specific ring manager driver.
35  * For having more SBA DMA channels, we can create more SBA device nodes
36  * in Broadcom SoC specific DTS based on number of hardware rings supported
37  * by Broadcom SoC ring manager.
38  */
39 
40 #include <linux/bitops.h>
41 #include <linux/debugfs.h>
42 #include <linux/dma-mapping.h>
43 #include <linux/dmaengine.h>
44 #include <linux/list.h>
45 #include <linux/mailbox_client.h>
46 #include <linux/mailbox/brcm-message.h>
47 #include <linux/module.h>
48 #include <linux/of_device.h>
49 #include <linux/slab.h>
50 #include <linux/raid/pq.h>
51 
52 #include "dmaengine.h"
53 
54 /* ====== Driver macros and defines ===== */
55 
56 #define SBA_TYPE_SHIFT					48
57 #define SBA_TYPE_MASK					GENMASK(1, 0)
58 #define SBA_TYPE_A					0x0
59 #define SBA_TYPE_B					0x2
60 #define SBA_TYPE_C					0x3
61 #define SBA_USER_DEF_SHIFT				32
62 #define SBA_USER_DEF_MASK				GENMASK(15, 0)
63 #define SBA_R_MDATA_SHIFT				24
64 #define SBA_R_MDATA_MASK				GENMASK(7, 0)
65 #define SBA_C_MDATA_MS_SHIFT				18
66 #define SBA_C_MDATA_MS_MASK				GENMASK(1, 0)
67 #define SBA_INT_SHIFT					17
68 #define SBA_INT_MASK					BIT(0)
69 #define SBA_RESP_SHIFT					16
70 #define SBA_RESP_MASK					BIT(0)
71 #define SBA_C_MDATA_SHIFT				8
72 #define SBA_C_MDATA_MASK				GENMASK(7, 0)
73 #define SBA_C_MDATA_BNUMx_SHIFT(__bnum)			(2 * (__bnum))
74 #define SBA_C_MDATA_BNUMx_MASK				GENMASK(1, 0)
75 #define SBA_C_MDATA_DNUM_SHIFT				5
76 #define SBA_C_MDATA_DNUM_MASK				GENMASK(4, 0)
77 #define SBA_C_MDATA_LS(__v)				((__v) & 0xff)
78 #define SBA_C_MDATA_MS(__v)				(((__v) >> 8) & 0x3)
79 #define SBA_CMD_SHIFT					0
80 #define SBA_CMD_MASK					GENMASK(3, 0)
81 #define SBA_CMD_ZERO_BUFFER				0x4
82 #define SBA_CMD_ZERO_ALL_BUFFERS			0x8
83 #define SBA_CMD_LOAD_BUFFER				0x9
84 #define SBA_CMD_XOR					0xa
85 #define SBA_CMD_GALOIS_XOR				0xb
86 #define SBA_CMD_WRITE_BUFFER				0xc
87 #define SBA_CMD_GALOIS					0xe
88 
89 #define SBA_MAX_REQ_PER_MBOX_CHANNEL			8192
90 #define SBA_MAX_MSG_SEND_PER_MBOX_CHANNEL		8
91 
92 /* Driver helper macros */
93 #define to_sba_request(tx)		\
94 	container_of(tx, struct sba_request, tx)
95 #define to_sba_device(dchan)		\
96 	container_of(dchan, struct sba_device, dma_chan)
97 
98 /* ===== Driver data structures ===== */
99 
100 enum sba_request_flags {
101 	SBA_REQUEST_STATE_FREE		= 0x001,
102 	SBA_REQUEST_STATE_ALLOCED	= 0x002,
103 	SBA_REQUEST_STATE_PENDING	= 0x004,
104 	SBA_REQUEST_STATE_ACTIVE	= 0x008,
105 	SBA_REQUEST_STATE_ABORTED	= 0x010,
106 	SBA_REQUEST_STATE_MASK		= 0x0ff,
107 	SBA_REQUEST_FENCE		= 0x100,
108 };
109 
110 struct sba_request {
111 	/* Global state */
112 	struct list_head node;
113 	struct sba_device *sba;
114 	u32 flags;
115 	/* Chained requests management */
116 	struct sba_request *first;
117 	struct list_head next;
118 	atomic_t next_pending_count;
119 	/* BRCM message data */
120 	struct brcm_message msg;
121 	struct dma_async_tx_descriptor tx;
122 	/* SBA commands */
123 	struct brcm_sba_command cmds[0];
124 };
125 
126 enum sba_version {
127 	SBA_VER_1 = 0,
128 	SBA_VER_2
129 };
130 
131 struct sba_device {
132 	/* Underlying device */
133 	struct device *dev;
134 	/* DT configuration parameters */
135 	enum sba_version ver;
136 	/* Derived configuration parameters */
137 	u32 max_req;
138 	u32 hw_buf_size;
139 	u32 hw_resp_size;
140 	u32 max_pq_coefs;
141 	u32 max_pq_srcs;
142 	u32 max_cmd_per_req;
143 	u32 max_xor_srcs;
144 	u32 max_resp_pool_size;
145 	u32 max_cmds_pool_size;
146 	/* Maibox client and Mailbox channels */
147 	struct mbox_client client;
148 	struct mbox_chan *mchan;
149 	struct device *mbox_dev;
150 	/* DMA device and DMA channel */
151 	struct dma_device dma_dev;
152 	struct dma_chan dma_chan;
153 	/* DMA channel resources */
154 	void *resp_base;
155 	dma_addr_t resp_dma_base;
156 	void *cmds_base;
157 	dma_addr_t cmds_dma_base;
158 	spinlock_t reqs_lock;
159 	bool reqs_fence;
160 	struct list_head reqs_alloc_list;
161 	struct list_head reqs_pending_list;
162 	struct list_head reqs_active_list;
163 	struct list_head reqs_aborted_list;
164 	struct list_head reqs_free_list;
165 	/* DebugFS directory entries */
166 	struct dentry *root;
167 };
168 
169 /* ====== Command helper routines ===== */
170 
171 static inline u64 __pure sba_cmd_enc(u64 cmd, u32 val, u32 shift, u32 mask)
172 {
173 	cmd &= ~((u64)mask << shift);
174 	cmd |= ((u64)(val & mask) << shift);
175 	return cmd;
176 }
177 
178 static inline u32 __pure sba_cmd_load_c_mdata(u32 b0)
179 {
180 	return b0 & SBA_C_MDATA_BNUMx_MASK;
181 }
182 
183 static inline u32 __pure sba_cmd_write_c_mdata(u32 b0)
184 {
185 	return b0 & SBA_C_MDATA_BNUMx_MASK;
186 }
187 
188 static inline u32 __pure sba_cmd_xor_c_mdata(u32 b1, u32 b0)
189 {
190 	return (b0 & SBA_C_MDATA_BNUMx_MASK) |
191 	       ((b1 & SBA_C_MDATA_BNUMx_MASK) << SBA_C_MDATA_BNUMx_SHIFT(1));
192 }
193 
194 static inline u32 __pure sba_cmd_pq_c_mdata(u32 d, u32 b1, u32 b0)
195 {
196 	return (b0 & SBA_C_MDATA_BNUMx_MASK) |
197 	       ((b1 & SBA_C_MDATA_BNUMx_MASK) << SBA_C_MDATA_BNUMx_SHIFT(1)) |
198 	       ((d & SBA_C_MDATA_DNUM_MASK) << SBA_C_MDATA_DNUM_SHIFT);
199 }
200 
201 /* ====== General helper routines ===== */
202 
203 static struct sba_request *sba_alloc_request(struct sba_device *sba)
204 {
205 	bool found = false;
206 	unsigned long flags;
207 	struct sba_request *req = NULL;
208 
209 	spin_lock_irqsave(&sba->reqs_lock, flags);
210 	list_for_each_entry(req, &sba->reqs_free_list, node) {
211 		if (async_tx_test_ack(&req->tx)) {
212 			list_move_tail(&req->node, &sba->reqs_alloc_list);
213 			found = true;
214 			break;
215 		}
216 	}
217 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
218 
219 	if (!found) {
220 		/*
221 		 * We have no more free requests so, we peek
222 		 * mailbox channels hoping few active requests
223 		 * would have completed which will create more
224 		 * room for new requests.
225 		 */
226 		mbox_client_peek_data(sba->mchan);
227 		return NULL;
228 	}
229 
230 	req->flags = SBA_REQUEST_STATE_ALLOCED;
231 	req->first = req;
232 	INIT_LIST_HEAD(&req->next);
233 	atomic_set(&req->next_pending_count, 1);
234 
235 	dma_async_tx_descriptor_init(&req->tx, &sba->dma_chan);
236 	async_tx_ack(&req->tx);
237 
238 	return req;
239 }
240 
241 /* Note: Must be called with sba->reqs_lock held */
242 static void _sba_pending_request(struct sba_device *sba,
243 				 struct sba_request *req)
244 {
245 	lockdep_assert_held(&sba->reqs_lock);
246 	req->flags &= ~SBA_REQUEST_STATE_MASK;
247 	req->flags |= SBA_REQUEST_STATE_PENDING;
248 	list_move_tail(&req->node, &sba->reqs_pending_list);
249 	if (list_empty(&sba->reqs_active_list))
250 		sba->reqs_fence = false;
251 }
252 
253 /* Note: Must be called with sba->reqs_lock held */
254 static bool _sba_active_request(struct sba_device *sba,
255 				struct sba_request *req)
256 {
257 	lockdep_assert_held(&sba->reqs_lock);
258 	if (list_empty(&sba->reqs_active_list))
259 		sba->reqs_fence = false;
260 	if (sba->reqs_fence)
261 		return false;
262 	req->flags &= ~SBA_REQUEST_STATE_MASK;
263 	req->flags |= SBA_REQUEST_STATE_ACTIVE;
264 	list_move_tail(&req->node, &sba->reqs_active_list);
265 	if (req->flags & SBA_REQUEST_FENCE)
266 		sba->reqs_fence = true;
267 	return true;
268 }
269 
270 /* Note: Must be called with sba->reqs_lock held */
271 static void _sba_abort_request(struct sba_device *sba,
272 			       struct sba_request *req)
273 {
274 	lockdep_assert_held(&sba->reqs_lock);
275 	req->flags &= ~SBA_REQUEST_STATE_MASK;
276 	req->flags |= SBA_REQUEST_STATE_ABORTED;
277 	list_move_tail(&req->node, &sba->reqs_aborted_list);
278 	if (list_empty(&sba->reqs_active_list))
279 		sba->reqs_fence = false;
280 }
281 
282 /* Note: Must be called with sba->reqs_lock held */
283 static void _sba_free_request(struct sba_device *sba,
284 			      struct sba_request *req)
285 {
286 	lockdep_assert_held(&sba->reqs_lock);
287 	req->flags &= ~SBA_REQUEST_STATE_MASK;
288 	req->flags |= SBA_REQUEST_STATE_FREE;
289 	list_move_tail(&req->node, &sba->reqs_free_list);
290 	if (list_empty(&sba->reqs_active_list))
291 		sba->reqs_fence = false;
292 }
293 
294 static void sba_free_chained_requests(struct sba_request *req)
295 {
296 	unsigned long flags;
297 	struct sba_request *nreq;
298 	struct sba_device *sba = req->sba;
299 
300 	spin_lock_irqsave(&sba->reqs_lock, flags);
301 
302 	_sba_free_request(sba, req);
303 	list_for_each_entry(nreq, &req->next, next)
304 		_sba_free_request(sba, nreq);
305 
306 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
307 }
308 
309 static void sba_chain_request(struct sba_request *first,
310 			      struct sba_request *req)
311 {
312 	unsigned long flags;
313 	struct sba_device *sba = req->sba;
314 
315 	spin_lock_irqsave(&sba->reqs_lock, flags);
316 
317 	list_add_tail(&req->next, &first->next);
318 	req->first = first;
319 	atomic_inc(&first->next_pending_count);
320 
321 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
322 }
323 
324 static void sba_cleanup_nonpending_requests(struct sba_device *sba)
325 {
326 	unsigned long flags;
327 	struct sba_request *req, *req1;
328 
329 	spin_lock_irqsave(&sba->reqs_lock, flags);
330 
331 	/* Freeup all alloced request */
332 	list_for_each_entry_safe(req, req1, &sba->reqs_alloc_list, node)
333 		_sba_free_request(sba, req);
334 
335 	/* Set all active requests as aborted */
336 	list_for_each_entry_safe(req, req1, &sba->reqs_active_list, node)
337 		_sba_abort_request(sba, req);
338 
339 	/*
340 	 * Note: We expect that aborted request will be eventually
341 	 * freed by sba_receive_message()
342 	 */
343 
344 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
345 }
346 
347 static void sba_cleanup_pending_requests(struct sba_device *sba)
348 {
349 	unsigned long flags;
350 	struct sba_request *req, *req1;
351 
352 	spin_lock_irqsave(&sba->reqs_lock, flags);
353 
354 	/* Freeup all pending request */
355 	list_for_each_entry_safe(req, req1, &sba->reqs_pending_list, node)
356 		_sba_free_request(sba, req);
357 
358 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
359 }
360 
361 static int sba_send_mbox_request(struct sba_device *sba,
362 				 struct sba_request *req)
363 {
364 	int ret = 0;
365 
366 	/* Send message for the request */
367 	req->msg.error = 0;
368 	ret = mbox_send_message(sba->mchan, &req->msg);
369 	if (ret < 0) {
370 		dev_err(sba->dev, "send message failed with error %d", ret);
371 		return ret;
372 	}
373 
374 	/* Check error returned by mailbox controller */
375 	ret = req->msg.error;
376 	if (ret < 0) {
377 		dev_err(sba->dev, "message error %d", ret);
378 	}
379 
380 	/* Signal txdone for mailbox channel */
381 	mbox_client_txdone(sba->mchan, ret);
382 
383 	return ret;
384 }
385 
386 /* Note: Must be called with sba->reqs_lock held */
387 static void _sba_process_pending_requests(struct sba_device *sba)
388 {
389 	int ret;
390 	u32 count;
391 	struct sba_request *req;
392 
393 	/* Process few pending requests */
394 	count = SBA_MAX_MSG_SEND_PER_MBOX_CHANNEL;
395 	while (!list_empty(&sba->reqs_pending_list) && count) {
396 		/* Get the first pending request */
397 		req = list_first_entry(&sba->reqs_pending_list,
398 				       struct sba_request, node);
399 
400 		/* Try to make request active */
401 		if (!_sba_active_request(sba, req))
402 			break;
403 
404 		/* Send request to mailbox channel */
405 		ret = sba_send_mbox_request(sba, req);
406 		if (ret < 0) {
407 			_sba_pending_request(sba, req);
408 			break;
409 		}
410 
411 		count--;
412 	}
413 }
414 
415 static void sba_process_received_request(struct sba_device *sba,
416 					 struct sba_request *req)
417 {
418 	unsigned long flags;
419 	struct dma_async_tx_descriptor *tx;
420 	struct sba_request *nreq, *first = req->first;
421 
422 	/* Process only after all chained requests are received */
423 	if (!atomic_dec_return(&first->next_pending_count)) {
424 		tx = &first->tx;
425 
426 		WARN_ON(tx->cookie < 0);
427 		if (tx->cookie > 0) {
428 			spin_lock_irqsave(&sba->reqs_lock, flags);
429 			dma_cookie_complete(tx);
430 			spin_unlock_irqrestore(&sba->reqs_lock, flags);
431 			dmaengine_desc_get_callback_invoke(tx, NULL);
432 			dma_descriptor_unmap(tx);
433 			tx->callback = NULL;
434 			tx->callback_result = NULL;
435 		}
436 
437 		dma_run_dependencies(tx);
438 
439 		spin_lock_irqsave(&sba->reqs_lock, flags);
440 
441 		/* Free all requests chained to first request */
442 		list_for_each_entry(nreq, &first->next, next)
443 			_sba_free_request(sba, nreq);
444 		INIT_LIST_HEAD(&first->next);
445 
446 		/* Free the first request */
447 		_sba_free_request(sba, first);
448 
449 		/* Process pending requests */
450 		_sba_process_pending_requests(sba);
451 
452 		spin_unlock_irqrestore(&sba->reqs_lock, flags);
453 	}
454 }
455 
456 static void sba_write_stats_in_seqfile(struct sba_device *sba,
457 				       struct seq_file *file)
458 {
459 	unsigned long flags;
460 	struct sba_request *req;
461 	u32 free_count = 0, alloced_count = 0;
462 	u32 pending_count = 0, active_count = 0, aborted_count = 0;
463 
464 	spin_lock_irqsave(&sba->reqs_lock, flags);
465 
466 	list_for_each_entry(req, &sba->reqs_free_list, node)
467 		if (async_tx_test_ack(&req->tx))
468 			free_count++;
469 
470 	list_for_each_entry(req, &sba->reqs_alloc_list, node)
471 		alloced_count++;
472 
473 	list_for_each_entry(req, &sba->reqs_pending_list, node)
474 		pending_count++;
475 
476 	list_for_each_entry(req, &sba->reqs_active_list, node)
477 		active_count++;
478 
479 	list_for_each_entry(req, &sba->reqs_aborted_list, node)
480 		aborted_count++;
481 
482 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
483 
484 	seq_printf(file, "maximum requests   = %d\n", sba->max_req);
485 	seq_printf(file, "free requests      = %d\n", free_count);
486 	seq_printf(file, "alloced requests   = %d\n", alloced_count);
487 	seq_printf(file, "pending requests   = %d\n", pending_count);
488 	seq_printf(file, "active requests    = %d\n", active_count);
489 	seq_printf(file, "aborted requests   = %d\n", aborted_count);
490 }
491 
492 /* ====== DMAENGINE callbacks ===== */
493 
494 static void sba_free_chan_resources(struct dma_chan *dchan)
495 {
496 	/*
497 	 * Channel resources are pre-alloced so we just free-up
498 	 * whatever we can so that we can re-use pre-alloced
499 	 * channel resources next time.
500 	 */
501 	sba_cleanup_nonpending_requests(to_sba_device(dchan));
502 }
503 
504 static int sba_device_terminate_all(struct dma_chan *dchan)
505 {
506 	/* Cleanup all pending requests */
507 	sba_cleanup_pending_requests(to_sba_device(dchan));
508 
509 	return 0;
510 }
511 
512 static void sba_issue_pending(struct dma_chan *dchan)
513 {
514 	unsigned long flags;
515 	struct sba_device *sba = to_sba_device(dchan);
516 
517 	/* Process pending requests */
518 	spin_lock_irqsave(&sba->reqs_lock, flags);
519 	_sba_process_pending_requests(sba);
520 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
521 }
522 
523 static dma_cookie_t sba_tx_submit(struct dma_async_tx_descriptor *tx)
524 {
525 	unsigned long flags;
526 	dma_cookie_t cookie;
527 	struct sba_device *sba;
528 	struct sba_request *req, *nreq;
529 
530 	if (unlikely(!tx))
531 		return -EINVAL;
532 
533 	sba = to_sba_device(tx->chan);
534 	req = to_sba_request(tx);
535 
536 	/* Assign cookie and mark all chained requests pending */
537 	spin_lock_irqsave(&sba->reqs_lock, flags);
538 	cookie = dma_cookie_assign(tx);
539 	_sba_pending_request(sba, req);
540 	list_for_each_entry(nreq, &req->next, next)
541 		_sba_pending_request(sba, nreq);
542 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
543 
544 	return cookie;
545 }
546 
547 static enum dma_status sba_tx_status(struct dma_chan *dchan,
548 				     dma_cookie_t cookie,
549 				     struct dma_tx_state *txstate)
550 {
551 	enum dma_status ret;
552 	struct sba_device *sba = to_sba_device(dchan);
553 
554 	ret = dma_cookie_status(dchan, cookie, txstate);
555 	if (ret == DMA_COMPLETE)
556 		return ret;
557 
558 	mbox_client_peek_data(sba->mchan);
559 
560 	return dma_cookie_status(dchan, cookie, txstate);
561 }
562 
563 static void sba_fillup_interrupt_msg(struct sba_request *req,
564 				     struct brcm_sba_command *cmds,
565 				     struct brcm_message *msg)
566 {
567 	u64 cmd;
568 	u32 c_mdata;
569 	dma_addr_t resp_dma = req->tx.phys;
570 	struct brcm_sba_command *cmdsp = cmds;
571 
572 	/* Type-B command to load dummy data into buf0 */
573 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
574 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
575 	cmd = sba_cmd_enc(cmd, req->sba->hw_resp_size,
576 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
577 	c_mdata = sba_cmd_load_c_mdata(0);
578 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
579 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
580 	cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
581 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
582 	cmdsp->cmd = cmd;
583 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
584 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
585 	cmdsp->data = resp_dma;
586 	cmdsp->data_len = req->sba->hw_resp_size;
587 	cmdsp++;
588 
589 	/* Type-A command to write buf0 to dummy location */
590 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
591 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
592 	cmd = sba_cmd_enc(cmd, req->sba->hw_resp_size,
593 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
594 	cmd = sba_cmd_enc(cmd, 0x1,
595 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
596 	c_mdata = sba_cmd_write_c_mdata(0);
597 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
598 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
599 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
600 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
601 	cmdsp->cmd = cmd;
602 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
603 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
604 	if (req->sba->hw_resp_size) {
605 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
606 		cmdsp->resp = resp_dma;
607 		cmdsp->resp_len = req->sba->hw_resp_size;
608 	}
609 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
610 	cmdsp->data = resp_dma;
611 	cmdsp->data_len = req->sba->hw_resp_size;
612 	cmdsp++;
613 
614 	/* Fillup brcm_message */
615 	msg->type = BRCM_MESSAGE_SBA;
616 	msg->sba.cmds = cmds;
617 	msg->sba.cmds_count = cmdsp - cmds;
618 	msg->ctx = req;
619 	msg->error = 0;
620 }
621 
622 static struct dma_async_tx_descriptor *
623 sba_prep_dma_interrupt(struct dma_chan *dchan, unsigned long flags)
624 {
625 	struct sba_request *req = NULL;
626 	struct sba_device *sba = to_sba_device(dchan);
627 
628 	/* Alloc new request */
629 	req = sba_alloc_request(sba);
630 	if (!req)
631 		return NULL;
632 
633 	/*
634 	 * Force fence so that no requests are submitted
635 	 * until DMA callback for this request is invoked.
636 	 */
637 	req->flags |= SBA_REQUEST_FENCE;
638 
639 	/* Fillup request message */
640 	sba_fillup_interrupt_msg(req, req->cmds, &req->msg);
641 
642 	/* Init async_tx descriptor */
643 	req->tx.flags = flags;
644 	req->tx.cookie = -EBUSY;
645 
646 	return &req->tx;
647 }
648 
649 static void sba_fillup_memcpy_msg(struct sba_request *req,
650 				  struct brcm_sba_command *cmds,
651 				  struct brcm_message *msg,
652 				  dma_addr_t msg_offset, size_t msg_len,
653 				  dma_addr_t dst, dma_addr_t src)
654 {
655 	u64 cmd;
656 	u32 c_mdata;
657 	dma_addr_t resp_dma = req->tx.phys;
658 	struct brcm_sba_command *cmdsp = cmds;
659 
660 	/* Type-B command to load data into buf0 */
661 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
662 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
663 	cmd = sba_cmd_enc(cmd, msg_len,
664 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
665 	c_mdata = sba_cmd_load_c_mdata(0);
666 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
667 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
668 	cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
669 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
670 	cmdsp->cmd = cmd;
671 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
672 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
673 	cmdsp->data = src + msg_offset;
674 	cmdsp->data_len = msg_len;
675 	cmdsp++;
676 
677 	/* Type-A command to write buf0 */
678 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
679 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
680 	cmd = sba_cmd_enc(cmd, msg_len,
681 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
682 	cmd = sba_cmd_enc(cmd, 0x1,
683 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
684 	c_mdata = sba_cmd_write_c_mdata(0);
685 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
686 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
687 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
688 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
689 	cmdsp->cmd = cmd;
690 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
691 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
692 	if (req->sba->hw_resp_size) {
693 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
694 		cmdsp->resp = resp_dma;
695 		cmdsp->resp_len = req->sba->hw_resp_size;
696 	}
697 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
698 	cmdsp->data = dst + msg_offset;
699 	cmdsp->data_len = msg_len;
700 	cmdsp++;
701 
702 	/* Fillup brcm_message */
703 	msg->type = BRCM_MESSAGE_SBA;
704 	msg->sba.cmds = cmds;
705 	msg->sba.cmds_count = cmdsp - cmds;
706 	msg->ctx = req;
707 	msg->error = 0;
708 }
709 
710 static struct sba_request *
711 sba_prep_dma_memcpy_req(struct sba_device *sba,
712 			dma_addr_t off, dma_addr_t dst, dma_addr_t src,
713 			size_t len, unsigned long flags)
714 {
715 	struct sba_request *req = NULL;
716 
717 	/* Alloc new request */
718 	req = sba_alloc_request(sba);
719 	if (!req)
720 		return NULL;
721 	if (flags & DMA_PREP_FENCE)
722 		req->flags |= SBA_REQUEST_FENCE;
723 
724 	/* Fillup request message */
725 	sba_fillup_memcpy_msg(req, req->cmds, &req->msg,
726 			      off, len, dst, src);
727 
728 	/* Init async_tx descriptor */
729 	req->tx.flags = flags;
730 	req->tx.cookie = -EBUSY;
731 
732 	return req;
733 }
734 
735 static struct dma_async_tx_descriptor *
736 sba_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst, dma_addr_t src,
737 		    size_t len, unsigned long flags)
738 {
739 	size_t req_len;
740 	dma_addr_t off = 0;
741 	struct sba_device *sba = to_sba_device(dchan);
742 	struct sba_request *first = NULL, *req;
743 
744 	/* Create chained requests where each request is upto hw_buf_size */
745 	while (len) {
746 		req_len = (len < sba->hw_buf_size) ? len : sba->hw_buf_size;
747 
748 		req = sba_prep_dma_memcpy_req(sba, off, dst, src,
749 					      req_len, flags);
750 		if (!req) {
751 			if (first)
752 				sba_free_chained_requests(first);
753 			return NULL;
754 		}
755 
756 		if (first)
757 			sba_chain_request(first, req);
758 		else
759 			first = req;
760 
761 		off += req_len;
762 		len -= req_len;
763 	}
764 
765 	return (first) ? &first->tx : NULL;
766 }
767 
768 static void sba_fillup_xor_msg(struct sba_request *req,
769 				struct brcm_sba_command *cmds,
770 				struct brcm_message *msg,
771 				dma_addr_t msg_offset, size_t msg_len,
772 				dma_addr_t dst, dma_addr_t *src, u32 src_cnt)
773 {
774 	u64 cmd;
775 	u32 c_mdata;
776 	unsigned int i;
777 	dma_addr_t resp_dma = req->tx.phys;
778 	struct brcm_sba_command *cmdsp = cmds;
779 
780 	/* Type-B command to load data into buf0 */
781 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
782 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
783 	cmd = sba_cmd_enc(cmd, msg_len,
784 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
785 	c_mdata = sba_cmd_load_c_mdata(0);
786 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
787 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
788 	cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
789 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
790 	cmdsp->cmd = cmd;
791 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
792 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
793 	cmdsp->data = src[0] + msg_offset;
794 	cmdsp->data_len = msg_len;
795 	cmdsp++;
796 
797 	/* Type-B commands to xor data with buf0 and put it back in buf0 */
798 	for (i = 1; i < src_cnt; i++) {
799 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
800 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
801 		cmd = sba_cmd_enc(cmd, msg_len,
802 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
803 		c_mdata = sba_cmd_xor_c_mdata(0, 0);
804 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
805 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
806 		cmd = sba_cmd_enc(cmd, SBA_CMD_XOR,
807 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
808 		cmdsp->cmd = cmd;
809 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
810 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
811 		cmdsp->data = src[i] + msg_offset;
812 		cmdsp->data_len = msg_len;
813 		cmdsp++;
814 	}
815 
816 	/* Type-A command to write buf0 */
817 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
818 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
819 	cmd = sba_cmd_enc(cmd, msg_len,
820 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
821 	cmd = sba_cmd_enc(cmd, 0x1,
822 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
823 	c_mdata = sba_cmd_write_c_mdata(0);
824 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
825 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
826 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
827 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
828 	cmdsp->cmd = cmd;
829 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
830 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
831 	if (req->sba->hw_resp_size) {
832 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
833 		cmdsp->resp = resp_dma;
834 		cmdsp->resp_len = req->sba->hw_resp_size;
835 	}
836 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
837 	cmdsp->data = dst + msg_offset;
838 	cmdsp->data_len = msg_len;
839 	cmdsp++;
840 
841 	/* Fillup brcm_message */
842 	msg->type = BRCM_MESSAGE_SBA;
843 	msg->sba.cmds = cmds;
844 	msg->sba.cmds_count = cmdsp - cmds;
845 	msg->ctx = req;
846 	msg->error = 0;
847 }
848 
849 static struct sba_request *
850 sba_prep_dma_xor_req(struct sba_device *sba,
851 		     dma_addr_t off, dma_addr_t dst, dma_addr_t *src,
852 		     u32 src_cnt, size_t len, unsigned long flags)
853 {
854 	struct sba_request *req = NULL;
855 
856 	/* Alloc new request */
857 	req = sba_alloc_request(sba);
858 	if (!req)
859 		return NULL;
860 	if (flags & DMA_PREP_FENCE)
861 		req->flags |= SBA_REQUEST_FENCE;
862 
863 	/* Fillup request message */
864 	sba_fillup_xor_msg(req, req->cmds, &req->msg,
865 			   off, len, dst, src, src_cnt);
866 
867 	/* Init async_tx descriptor */
868 	req->tx.flags = flags;
869 	req->tx.cookie = -EBUSY;
870 
871 	return req;
872 }
873 
874 static struct dma_async_tx_descriptor *
875 sba_prep_dma_xor(struct dma_chan *dchan, dma_addr_t dst, dma_addr_t *src,
876 		 u32 src_cnt, size_t len, unsigned long flags)
877 {
878 	size_t req_len;
879 	dma_addr_t off = 0;
880 	struct sba_device *sba = to_sba_device(dchan);
881 	struct sba_request *first = NULL, *req;
882 
883 	/* Sanity checks */
884 	if (unlikely(src_cnt > sba->max_xor_srcs))
885 		return NULL;
886 
887 	/* Create chained requests where each request is upto hw_buf_size */
888 	while (len) {
889 		req_len = (len < sba->hw_buf_size) ? len : sba->hw_buf_size;
890 
891 		req = sba_prep_dma_xor_req(sba, off, dst, src, src_cnt,
892 					   req_len, flags);
893 		if (!req) {
894 			if (first)
895 				sba_free_chained_requests(first);
896 			return NULL;
897 		}
898 
899 		if (first)
900 			sba_chain_request(first, req);
901 		else
902 			first = req;
903 
904 		off += req_len;
905 		len -= req_len;
906 	}
907 
908 	return (first) ? &first->tx : NULL;
909 }
910 
911 static void sba_fillup_pq_msg(struct sba_request *req,
912 				bool pq_continue,
913 				struct brcm_sba_command *cmds,
914 				struct brcm_message *msg,
915 				dma_addr_t msg_offset, size_t msg_len,
916 				dma_addr_t *dst_p, dma_addr_t *dst_q,
917 				const u8 *scf, dma_addr_t *src, u32 src_cnt)
918 {
919 	u64 cmd;
920 	u32 c_mdata;
921 	unsigned int i;
922 	dma_addr_t resp_dma = req->tx.phys;
923 	struct brcm_sba_command *cmdsp = cmds;
924 
925 	if (pq_continue) {
926 		/* Type-B command to load old P into buf0 */
927 		if (dst_p) {
928 			cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
929 				SBA_TYPE_SHIFT, SBA_TYPE_MASK);
930 			cmd = sba_cmd_enc(cmd, msg_len,
931 				SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
932 			c_mdata = sba_cmd_load_c_mdata(0);
933 			cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
934 				SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
935 			cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
936 				SBA_CMD_SHIFT, SBA_CMD_MASK);
937 			cmdsp->cmd = cmd;
938 			*cmdsp->cmd_dma = cpu_to_le64(cmd);
939 			cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
940 			cmdsp->data = *dst_p + msg_offset;
941 			cmdsp->data_len = msg_len;
942 			cmdsp++;
943 		}
944 
945 		/* Type-B command to load old Q into buf1 */
946 		if (dst_q) {
947 			cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
948 				SBA_TYPE_SHIFT, SBA_TYPE_MASK);
949 			cmd = sba_cmd_enc(cmd, msg_len,
950 				SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
951 			c_mdata = sba_cmd_load_c_mdata(1);
952 			cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
953 				SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
954 			cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
955 				SBA_CMD_SHIFT, SBA_CMD_MASK);
956 			cmdsp->cmd = cmd;
957 			*cmdsp->cmd_dma = cpu_to_le64(cmd);
958 			cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
959 			cmdsp->data = *dst_q + msg_offset;
960 			cmdsp->data_len = msg_len;
961 			cmdsp++;
962 		}
963 	} else {
964 		/* Type-A command to zero all buffers */
965 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
966 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
967 		cmd = sba_cmd_enc(cmd, msg_len,
968 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
969 		cmd = sba_cmd_enc(cmd, SBA_CMD_ZERO_ALL_BUFFERS,
970 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
971 		cmdsp->cmd = cmd;
972 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
973 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
974 		cmdsp++;
975 	}
976 
977 	/* Type-B commands for generate P onto buf0 and Q onto buf1 */
978 	for (i = 0; i < src_cnt; i++) {
979 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
980 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
981 		cmd = sba_cmd_enc(cmd, msg_len,
982 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
983 		c_mdata = sba_cmd_pq_c_mdata(raid6_gflog[scf[i]], 1, 0);
984 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
985 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
986 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_MS(c_mdata),
987 				  SBA_C_MDATA_MS_SHIFT, SBA_C_MDATA_MS_MASK);
988 		cmd = sba_cmd_enc(cmd, SBA_CMD_GALOIS_XOR,
989 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
990 		cmdsp->cmd = cmd;
991 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
992 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
993 		cmdsp->data = src[i] + msg_offset;
994 		cmdsp->data_len = msg_len;
995 		cmdsp++;
996 	}
997 
998 	/* Type-A command to write buf0 */
999 	if (dst_p) {
1000 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1001 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1002 		cmd = sba_cmd_enc(cmd, msg_len,
1003 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1004 		cmd = sba_cmd_enc(cmd, 0x1,
1005 				  SBA_RESP_SHIFT, SBA_RESP_MASK);
1006 		c_mdata = sba_cmd_write_c_mdata(0);
1007 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1008 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1009 		cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1010 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1011 		cmdsp->cmd = cmd;
1012 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1013 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1014 		if (req->sba->hw_resp_size) {
1015 			cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1016 			cmdsp->resp = resp_dma;
1017 			cmdsp->resp_len = req->sba->hw_resp_size;
1018 		}
1019 		cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1020 		cmdsp->data = *dst_p + msg_offset;
1021 		cmdsp->data_len = msg_len;
1022 		cmdsp++;
1023 	}
1024 
1025 	/* Type-A command to write buf1 */
1026 	if (dst_q) {
1027 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1028 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1029 		cmd = sba_cmd_enc(cmd, msg_len,
1030 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1031 		cmd = sba_cmd_enc(cmd, 0x1,
1032 				  SBA_RESP_SHIFT, SBA_RESP_MASK);
1033 		c_mdata = sba_cmd_write_c_mdata(1);
1034 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1035 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1036 		cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1037 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1038 		cmdsp->cmd = cmd;
1039 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1040 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1041 		if (req->sba->hw_resp_size) {
1042 			cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1043 			cmdsp->resp = resp_dma;
1044 			cmdsp->resp_len = req->sba->hw_resp_size;
1045 		}
1046 		cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1047 		cmdsp->data = *dst_q + msg_offset;
1048 		cmdsp->data_len = msg_len;
1049 		cmdsp++;
1050 	}
1051 
1052 	/* Fillup brcm_message */
1053 	msg->type = BRCM_MESSAGE_SBA;
1054 	msg->sba.cmds = cmds;
1055 	msg->sba.cmds_count = cmdsp - cmds;
1056 	msg->ctx = req;
1057 	msg->error = 0;
1058 }
1059 
1060 static struct sba_request *
1061 sba_prep_dma_pq_req(struct sba_device *sba, dma_addr_t off,
1062 		    dma_addr_t *dst_p, dma_addr_t *dst_q, dma_addr_t *src,
1063 		    u32 src_cnt, const u8 *scf, size_t len, unsigned long flags)
1064 {
1065 	struct sba_request *req = NULL;
1066 
1067 	/* Alloc new request */
1068 	req = sba_alloc_request(sba);
1069 	if (!req)
1070 		return NULL;
1071 	if (flags & DMA_PREP_FENCE)
1072 		req->flags |= SBA_REQUEST_FENCE;
1073 
1074 	/* Fillup request messages */
1075 	sba_fillup_pq_msg(req, dmaf_continue(flags),
1076 			  req->cmds, &req->msg,
1077 			  off, len, dst_p, dst_q, scf, src, src_cnt);
1078 
1079 	/* Init async_tx descriptor */
1080 	req->tx.flags = flags;
1081 	req->tx.cookie = -EBUSY;
1082 
1083 	return req;
1084 }
1085 
1086 static void sba_fillup_pq_single_msg(struct sba_request *req,
1087 				bool pq_continue,
1088 				struct brcm_sba_command *cmds,
1089 				struct brcm_message *msg,
1090 				dma_addr_t msg_offset, size_t msg_len,
1091 				dma_addr_t *dst_p, dma_addr_t *dst_q,
1092 				dma_addr_t src, u8 scf)
1093 {
1094 	u64 cmd;
1095 	u32 c_mdata;
1096 	u8 pos, dpos = raid6_gflog[scf];
1097 	dma_addr_t resp_dma = req->tx.phys;
1098 	struct brcm_sba_command *cmdsp = cmds;
1099 
1100 	if (!dst_p)
1101 		goto skip_p;
1102 
1103 	if (pq_continue) {
1104 		/* Type-B command to load old P into buf0 */
1105 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1106 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1107 		cmd = sba_cmd_enc(cmd, msg_len,
1108 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1109 		c_mdata = sba_cmd_load_c_mdata(0);
1110 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1111 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1112 		cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
1113 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1114 		cmdsp->cmd = cmd;
1115 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1116 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1117 		cmdsp->data = *dst_p + msg_offset;
1118 		cmdsp->data_len = msg_len;
1119 		cmdsp++;
1120 
1121 		/*
1122 		 * Type-B commands to xor data with buf0 and put it
1123 		 * back in buf0
1124 		 */
1125 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1126 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1127 		cmd = sba_cmd_enc(cmd, msg_len,
1128 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1129 		c_mdata = sba_cmd_xor_c_mdata(0, 0);
1130 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1131 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1132 		cmd = sba_cmd_enc(cmd, SBA_CMD_XOR,
1133 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1134 		cmdsp->cmd = cmd;
1135 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1136 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1137 		cmdsp->data = src + msg_offset;
1138 		cmdsp->data_len = msg_len;
1139 		cmdsp++;
1140 	} else {
1141 		/* Type-B command to load old P into buf0 */
1142 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1143 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1144 		cmd = sba_cmd_enc(cmd, msg_len,
1145 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1146 		c_mdata = sba_cmd_load_c_mdata(0);
1147 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1148 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1149 		cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
1150 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1151 		cmdsp->cmd = cmd;
1152 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1153 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1154 		cmdsp->data = src + msg_offset;
1155 		cmdsp->data_len = msg_len;
1156 		cmdsp++;
1157 	}
1158 
1159 	/* Type-A command to write buf0 */
1160 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1161 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1162 	cmd = sba_cmd_enc(cmd, msg_len,
1163 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1164 	cmd = sba_cmd_enc(cmd, 0x1,
1165 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
1166 	c_mdata = sba_cmd_write_c_mdata(0);
1167 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1168 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1169 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1170 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1171 	cmdsp->cmd = cmd;
1172 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1173 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1174 	if (req->sba->hw_resp_size) {
1175 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1176 		cmdsp->resp = resp_dma;
1177 		cmdsp->resp_len = req->sba->hw_resp_size;
1178 	}
1179 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1180 	cmdsp->data = *dst_p + msg_offset;
1181 	cmdsp->data_len = msg_len;
1182 	cmdsp++;
1183 
1184 skip_p:
1185 	if (!dst_q)
1186 		goto skip_q;
1187 
1188 	/* Type-A command to zero all buffers */
1189 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1190 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1191 	cmd = sba_cmd_enc(cmd, msg_len,
1192 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1193 	cmd = sba_cmd_enc(cmd, SBA_CMD_ZERO_ALL_BUFFERS,
1194 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1195 	cmdsp->cmd = cmd;
1196 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1197 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1198 	cmdsp++;
1199 
1200 	if (dpos == 255)
1201 		goto skip_q_computation;
1202 	pos = (dpos < req->sba->max_pq_coefs) ?
1203 		dpos : (req->sba->max_pq_coefs - 1);
1204 
1205 	/*
1206 	 * Type-B command to generate initial Q from data
1207 	 * and store output into buf0
1208 	 */
1209 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1210 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1211 	cmd = sba_cmd_enc(cmd, msg_len,
1212 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1213 	c_mdata = sba_cmd_pq_c_mdata(pos, 0, 0);
1214 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1215 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1216 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_MS(c_mdata),
1217 			  SBA_C_MDATA_MS_SHIFT, SBA_C_MDATA_MS_MASK);
1218 	cmd = sba_cmd_enc(cmd, SBA_CMD_GALOIS,
1219 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1220 	cmdsp->cmd = cmd;
1221 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1222 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1223 	cmdsp->data = src + msg_offset;
1224 	cmdsp->data_len = msg_len;
1225 	cmdsp++;
1226 
1227 	dpos -= pos;
1228 
1229 	/* Multiple Type-A command to generate final Q */
1230 	while (dpos) {
1231 		pos = (dpos < req->sba->max_pq_coefs) ?
1232 			dpos : (req->sba->max_pq_coefs - 1);
1233 
1234 		/*
1235 		 * Type-A command to generate Q with buf0 and
1236 		 * buf1 store result in buf0
1237 		 */
1238 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1239 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1240 		cmd = sba_cmd_enc(cmd, msg_len,
1241 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1242 		c_mdata = sba_cmd_pq_c_mdata(pos, 0, 1);
1243 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1244 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1245 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_MS(c_mdata),
1246 				  SBA_C_MDATA_MS_SHIFT, SBA_C_MDATA_MS_MASK);
1247 		cmd = sba_cmd_enc(cmd, SBA_CMD_GALOIS,
1248 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1249 		cmdsp->cmd = cmd;
1250 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1251 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1252 		cmdsp++;
1253 
1254 		dpos -= pos;
1255 	}
1256 
1257 skip_q_computation:
1258 	if (pq_continue) {
1259 		/*
1260 		 * Type-B command to XOR previous output with
1261 		 * buf0 and write it into buf0
1262 		 */
1263 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1264 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1265 		cmd = sba_cmd_enc(cmd, msg_len,
1266 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1267 		c_mdata = sba_cmd_xor_c_mdata(0, 0);
1268 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1269 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1270 		cmd = sba_cmd_enc(cmd, SBA_CMD_XOR,
1271 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1272 		cmdsp->cmd = cmd;
1273 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1274 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1275 		cmdsp->data = *dst_q + msg_offset;
1276 		cmdsp->data_len = msg_len;
1277 		cmdsp++;
1278 	}
1279 
1280 	/* Type-A command to write buf0 */
1281 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1282 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1283 	cmd = sba_cmd_enc(cmd, msg_len,
1284 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1285 	cmd = sba_cmd_enc(cmd, 0x1,
1286 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
1287 	c_mdata = sba_cmd_write_c_mdata(0);
1288 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1289 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1290 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1291 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1292 	cmdsp->cmd = cmd;
1293 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1294 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1295 	if (req->sba->hw_resp_size) {
1296 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1297 		cmdsp->resp = resp_dma;
1298 		cmdsp->resp_len = req->sba->hw_resp_size;
1299 	}
1300 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1301 	cmdsp->data = *dst_q + msg_offset;
1302 	cmdsp->data_len = msg_len;
1303 	cmdsp++;
1304 
1305 skip_q:
1306 	/* Fillup brcm_message */
1307 	msg->type = BRCM_MESSAGE_SBA;
1308 	msg->sba.cmds = cmds;
1309 	msg->sba.cmds_count = cmdsp - cmds;
1310 	msg->ctx = req;
1311 	msg->error = 0;
1312 }
1313 
1314 static struct sba_request *
1315 sba_prep_dma_pq_single_req(struct sba_device *sba, dma_addr_t off,
1316 			   dma_addr_t *dst_p, dma_addr_t *dst_q,
1317 			   dma_addr_t src, u8 scf, size_t len,
1318 			   unsigned long flags)
1319 {
1320 	struct sba_request *req = NULL;
1321 
1322 	/* Alloc new request */
1323 	req = sba_alloc_request(sba);
1324 	if (!req)
1325 		return NULL;
1326 	if (flags & DMA_PREP_FENCE)
1327 		req->flags |= SBA_REQUEST_FENCE;
1328 
1329 	/* Fillup request messages */
1330 	sba_fillup_pq_single_msg(req,  dmaf_continue(flags),
1331 				 req->cmds, &req->msg, off, len,
1332 				 dst_p, dst_q, src, scf);
1333 
1334 	/* Init async_tx descriptor */
1335 	req->tx.flags = flags;
1336 	req->tx.cookie = -EBUSY;
1337 
1338 	return req;
1339 }
1340 
1341 static struct dma_async_tx_descriptor *
1342 sba_prep_dma_pq(struct dma_chan *dchan, dma_addr_t *dst, dma_addr_t *src,
1343 		u32 src_cnt, const u8 *scf, size_t len, unsigned long flags)
1344 {
1345 	u32 i, dst_q_index;
1346 	size_t req_len;
1347 	bool slow = false;
1348 	dma_addr_t off = 0;
1349 	dma_addr_t *dst_p = NULL, *dst_q = NULL;
1350 	struct sba_device *sba = to_sba_device(dchan);
1351 	struct sba_request *first = NULL, *req;
1352 
1353 	/* Sanity checks */
1354 	if (unlikely(src_cnt > sba->max_pq_srcs))
1355 		return NULL;
1356 	for (i = 0; i < src_cnt; i++)
1357 		if (sba->max_pq_coefs <= raid6_gflog[scf[i]])
1358 			slow = true;
1359 
1360 	/* Figure-out P and Q destination addresses */
1361 	if (!(flags & DMA_PREP_PQ_DISABLE_P))
1362 		dst_p = &dst[0];
1363 	if (!(flags & DMA_PREP_PQ_DISABLE_Q))
1364 		dst_q = &dst[1];
1365 
1366 	/* Create chained requests where each request is upto hw_buf_size */
1367 	while (len) {
1368 		req_len = (len < sba->hw_buf_size) ? len : sba->hw_buf_size;
1369 
1370 		if (slow) {
1371 			dst_q_index = src_cnt;
1372 
1373 			if (dst_q) {
1374 				for (i = 0; i < src_cnt; i++) {
1375 					if (*dst_q == src[i]) {
1376 						dst_q_index = i;
1377 						break;
1378 					}
1379 				}
1380 			}
1381 
1382 			if (dst_q_index < src_cnt) {
1383 				i = dst_q_index;
1384 				req = sba_prep_dma_pq_single_req(sba,
1385 					off, dst_p, dst_q, src[i], scf[i],
1386 					req_len, flags | DMA_PREP_FENCE);
1387 				if (!req)
1388 					goto fail;
1389 
1390 				if (first)
1391 					sba_chain_request(first, req);
1392 				else
1393 					first = req;
1394 
1395 				flags |= DMA_PREP_CONTINUE;
1396 			}
1397 
1398 			for (i = 0; i < src_cnt; i++) {
1399 				if (dst_q_index == i)
1400 					continue;
1401 
1402 				req = sba_prep_dma_pq_single_req(sba,
1403 					off, dst_p, dst_q, src[i], scf[i],
1404 					req_len, flags | DMA_PREP_FENCE);
1405 				if (!req)
1406 					goto fail;
1407 
1408 				if (first)
1409 					sba_chain_request(first, req);
1410 				else
1411 					first = req;
1412 
1413 				flags |= DMA_PREP_CONTINUE;
1414 			}
1415 		} else {
1416 			req = sba_prep_dma_pq_req(sba, off,
1417 						  dst_p, dst_q, src, src_cnt,
1418 						  scf, req_len, flags);
1419 			if (!req)
1420 				goto fail;
1421 
1422 			if (first)
1423 				sba_chain_request(first, req);
1424 			else
1425 				first = req;
1426 		}
1427 
1428 		off += req_len;
1429 		len -= req_len;
1430 	}
1431 
1432 	return (first) ? &first->tx : NULL;
1433 
1434 fail:
1435 	if (first)
1436 		sba_free_chained_requests(first);
1437 	return NULL;
1438 }
1439 
1440 /* ====== Mailbox callbacks ===== */
1441 
1442 static void sba_receive_message(struct mbox_client *cl, void *msg)
1443 {
1444 	struct brcm_message *m = msg;
1445 	struct sba_request *req = m->ctx;
1446 	struct sba_device *sba = req->sba;
1447 
1448 	/* Error count if message has error */
1449 	if (m->error < 0)
1450 		dev_err(sba->dev, "%s got message with error %d",
1451 			dma_chan_name(&sba->dma_chan), m->error);
1452 
1453 	/* Process received request */
1454 	sba_process_received_request(sba, req);
1455 }
1456 
1457 /* ====== Debugfs callbacks ====== */
1458 
1459 static int sba_debugfs_stats_show(struct seq_file *file, void *offset)
1460 {
1461 	struct sba_device *sba = dev_get_drvdata(file->private);
1462 
1463 	/* Write stats in file */
1464 	sba_write_stats_in_seqfile(sba, file);
1465 
1466 	return 0;
1467 }
1468 
1469 /* ====== Platform driver routines ===== */
1470 
1471 static int sba_prealloc_channel_resources(struct sba_device *sba)
1472 {
1473 	int i, j, ret = 0;
1474 	struct sba_request *req = NULL;
1475 
1476 	sba->resp_base = dma_alloc_coherent(sba->mbox_dev,
1477 					    sba->max_resp_pool_size,
1478 					    &sba->resp_dma_base, GFP_KERNEL);
1479 	if (!sba->resp_base)
1480 		return -ENOMEM;
1481 
1482 	sba->cmds_base = dma_alloc_coherent(sba->mbox_dev,
1483 					    sba->max_cmds_pool_size,
1484 					    &sba->cmds_dma_base, GFP_KERNEL);
1485 	if (!sba->cmds_base) {
1486 		ret = -ENOMEM;
1487 		goto fail_free_resp_pool;
1488 	}
1489 
1490 	spin_lock_init(&sba->reqs_lock);
1491 	sba->reqs_fence = false;
1492 	INIT_LIST_HEAD(&sba->reqs_alloc_list);
1493 	INIT_LIST_HEAD(&sba->reqs_pending_list);
1494 	INIT_LIST_HEAD(&sba->reqs_active_list);
1495 	INIT_LIST_HEAD(&sba->reqs_aborted_list);
1496 	INIT_LIST_HEAD(&sba->reqs_free_list);
1497 
1498 	for (i = 0; i < sba->max_req; i++) {
1499 		req = devm_kzalloc(sba->dev,
1500 				   struct_size(req, cmds, sba->max_cmd_per_req),
1501 				   GFP_KERNEL);
1502 		if (!req) {
1503 			ret = -ENOMEM;
1504 			goto fail_free_cmds_pool;
1505 		}
1506 		INIT_LIST_HEAD(&req->node);
1507 		req->sba = sba;
1508 		req->flags = SBA_REQUEST_STATE_FREE;
1509 		INIT_LIST_HEAD(&req->next);
1510 		atomic_set(&req->next_pending_count, 0);
1511 		for (j = 0; j < sba->max_cmd_per_req; j++) {
1512 			req->cmds[j].cmd = 0;
1513 			req->cmds[j].cmd_dma = sba->cmds_base +
1514 				(i * sba->max_cmd_per_req + j) * sizeof(u64);
1515 			req->cmds[j].cmd_dma_addr = sba->cmds_dma_base +
1516 				(i * sba->max_cmd_per_req + j) * sizeof(u64);
1517 			req->cmds[j].flags = 0;
1518 		}
1519 		memset(&req->msg, 0, sizeof(req->msg));
1520 		dma_async_tx_descriptor_init(&req->tx, &sba->dma_chan);
1521 		async_tx_ack(&req->tx);
1522 		req->tx.tx_submit = sba_tx_submit;
1523 		req->tx.phys = sba->resp_dma_base + i * sba->hw_resp_size;
1524 		list_add_tail(&req->node, &sba->reqs_free_list);
1525 	}
1526 
1527 	return 0;
1528 
1529 fail_free_cmds_pool:
1530 	dma_free_coherent(sba->mbox_dev,
1531 			  sba->max_cmds_pool_size,
1532 			  sba->cmds_base, sba->cmds_dma_base);
1533 fail_free_resp_pool:
1534 	dma_free_coherent(sba->mbox_dev,
1535 			  sba->max_resp_pool_size,
1536 			  sba->resp_base, sba->resp_dma_base);
1537 	return ret;
1538 }
1539 
1540 static void sba_freeup_channel_resources(struct sba_device *sba)
1541 {
1542 	dmaengine_terminate_all(&sba->dma_chan);
1543 	dma_free_coherent(sba->mbox_dev, sba->max_cmds_pool_size,
1544 			  sba->cmds_base, sba->cmds_dma_base);
1545 	dma_free_coherent(sba->mbox_dev, sba->max_resp_pool_size,
1546 			  sba->resp_base, sba->resp_dma_base);
1547 	sba->resp_base = NULL;
1548 	sba->resp_dma_base = 0;
1549 }
1550 
1551 static int sba_async_register(struct sba_device *sba)
1552 {
1553 	int ret;
1554 	struct dma_device *dma_dev = &sba->dma_dev;
1555 
1556 	/* Initialize DMA channel cookie */
1557 	sba->dma_chan.device = dma_dev;
1558 	dma_cookie_init(&sba->dma_chan);
1559 
1560 	/* Initialize DMA device capability mask */
1561 	dma_cap_zero(dma_dev->cap_mask);
1562 	dma_cap_set(DMA_INTERRUPT, dma_dev->cap_mask);
1563 	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
1564 	dma_cap_set(DMA_XOR, dma_dev->cap_mask);
1565 	dma_cap_set(DMA_PQ, dma_dev->cap_mask);
1566 
1567 	/*
1568 	 * Set mailbox channel device as the base device of
1569 	 * our dma_device because the actual memory accesses
1570 	 * will be done by mailbox controller
1571 	 */
1572 	dma_dev->dev = sba->mbox_dev;
1573 
1574 	/* Set base prep routines */
1575 	dma_dev->device_free_chan_resources = sba_free_chan_resources;
1576 	dma_dev->device_terminate_all = sba_device_terminate_all;
1577 	dma_dev->device_issue_pending = sba_issue_pending;
1578 	dma_dev->device_tx_status = sba_tx_status;
1579 
1580 	/* Set interrupt routine */
1581 	if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1582 		dma_dev->device_prep_dma_interrupt = sba_prep_dma_interrupt;
1583 
1584 	/* Set memcpy routine */
1585 	if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1586 		dma_dev->device_prep_dma_memcpy = sba_prep_dma_memcpy;
1587 
1588 	/* Set xor routine and capability */
1589 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1590 		dma_dev->device_prep_dma_xor = sba_prep_dma_xor;
1591 		dma_dev->max_xor = sba->max_xor_srcs;
1592 	}
1593 
1594 	/* Set pq routine and capability */
1595 	if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1596 		dma_dev->device_prep_dma_pq = sba_prep_dma_pq;
1597 		dma_set_maxpq(dma_dev, sba->max_pq_srcs, 0);
1598 	}
1599 
1600 	/* Initialize DMA device channel list */
1601 	INIT_LIST_HEAD(&dma_dev->channels);
1602 	list_add_tail(&sba->dma_chan.device_node, &dma_dev->channels);
1603 
1604 	/* Register with Linux async DMA framework*/
1605 	ret = dma_async_device_register(dma_dev);
1606 	if (ret) {
1607 		dev_err(sba->dev, "async device register error %d", ret);
1608 		return ret;
1609 	}
1610 
1611 	dev_info(sba->dev, "%s capabilities: %s%s%s%s\n",
1612 	dma_chan_name(&sba->dma_chan),
1613 	dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "interrupt " : "",
1614 	dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "memcpy " : "",
1615 	dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1616 	dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "pq " : "");
1617 
1618 	return 0;
1619 }
1620 
1621 static int sba_probe(struct platform_device *pdev)
1622 {
1623 	int ret = 0;
1624 	struct sba_device *sba;
1625 	struct platform_device *mbox_pdev;
1626 	struct of_phandle_args args;
1627 
1628 	/* Allocate main SBA struct */
1629 	sba = devm_kzalloc(&pdev->dev, sizeof(*sba), GFP_KERNEL);
1630 	if (!sba)
1631 		return -ENOMEM;
1632 
1633 	sba->dev = &pdev->dev;
1634 	platform_set_drvdata(pdev, sba);
1635 
1636 	/* Number of mailbox channels should be atleast 1 */
1637 	ret = of_count_phandle_with_args(pdev->dev.of_node,
1638 					 "mboxes", "#mbox-cells");
1639 	if (ret <= 0)
1640 		return -ENODEV;
1641 
1642 	/* Determine SBA version from DT compatible string */
1643 	if (of_device_is_compatible(sba->dev->of_node, "brcm,iproc-sba"))
1644 		sba->ver = SBA_VER_1;
1645 	else if (of_device_is_compatible(sba->dev->of_node,
1646 					 "brcm,iproc-sba-v2"))
1647 		sba->ver = SBA_VER_2;
1648 	else
1649 		return -ENODEV;
1650 
1651 	/* Derived Configuration parameters */
1652 	switch (sba->ver) {
1653 	case SBA_VER_1:
1654 		sba->hw_buf_size = 4096;
1655 		sba->hw_resp_size = 8;
1656 		sba->max_pq_coefs = 6;
1657 		sba->max_pq_srcs = 6;
1658 		break;
1659 	case SBA_VER_2:
1660 		sba->hw_buf_size = 4096;
1661 		sba->hw_resp_size = 8;
1662 		sba->max_pq_coefs = 30;
1663 		/*
1664 		 * We can support max_pq_srcs == max_pq_coefs because
1665 		 * we are limited by number of SBA commands that we can
1666 		 * fit in one message for underlying ring manager HW.
1667 		 */
1668 		sba->max_pq_srcs = 12;
1669 		break;
1670 	default:
1671 		return -EINVAL;
1672 	}
1673 	sba->max_req = SBA_MAX_REQ_PER_MBOX_CHANNEL;
1674 	sba->max_cmd_per_req = sba->max_pq_srcs + 3;
1675 	sba->max_xor_srcs = sba->max_cmd_per_req - 1;
1676 	sba->max_resp_pool_size = sba->max_req * sba->hw_resp_size;
1677 	sba->max_cmds_pool_size = sba->max_req *
1678 				  sba->max_cmd_per_req * sizeof(u64);
1679 
1680 	/* Setup mailbox client */
1681 	sba->client.dev			= &pdev->dev;
1682 	sba->client.rx_callback		= sba_receive_message;
1683 	sba->client.tx_block		= false;
1684 	sba->client.knows_txdone	= true;
1685 	sba->client.tx_tout		= 0;
1686 
1687 	/* Request mailbox channel */
1688 	sba->mchan = mbox_request_channel(&sba->client, 0);
1689 	if (IS_ERR(sba->mchan)) {
1690 		ret = PTR_ERR(sba->mchan);
1691 		goto fail_free_mchan;
1692 	}
1693 
1694 	/* Find-out underlying mailbox device */
1695 	ret = of_parse_phandle_with_args(pdev->dev.of_node,
1696 					 "mboxes", "#mbox-cells", 0, &args);
1697 	if (ret)
1698 		goto fail_free_mchan;
1699 	mbox_pdev = of_find_device_by_node(args.np);
1700 	of_node_put(args.np);
1701 	if (!mbox_pdev) {
1702 		ret = -ENODEV;
1703 		goto fail_free_mchan;
1704 	}
1705 	sba->mbox_dev = &mbox_pdev->dev;
1706 
1707 	/* Prealloc channel resource */
1708 	ret = sba_prealloc_channel_resources(sba);
1709 	if (ret)
1710 		goto fail_free_mchan;
1711 
1712 	/* Check availability of debugfs */
1713 	if (!debugfs_initialized())
1714 		goto skip_debugfs;
1715 
1716 	/* Create debugfs root entry */
1717 	sba->root = debugfs_create_dir(dev_name(sba->dev), NULL);
1718 
1719 	/* Create debugfs stats entry */
1720 	debugfs_create_devm_seqfile(sba->dev, "stats", sba->root,
1721 				    sba_debugfs_stats_show);
1722 
1723 skip_debugfs:
1724 
1725 	/* Register DMA device with Linux async framework */
1726 	ret = sba_async_register(sba);
1727 	if (ret)
1728 		goto fail_free_resources;
1729 
1730 	/* Print device info */
1731 	dev_info(sba->dev, "%s using SBAv%d mailbox channel from %s",
1732 		 dma_chan_name(&sba->dma_chan), sba->ver+1,
1733 		 dev_name(sba->mbox_dev));
1734 
1735 	return 0;
1736 
1737 fail_free_resources:
1738 	debugfs_remove_recursive(sba->root);
1739 	sba_freeup_channel_resources(sba);
1740 fail_free_mchan:
1741 	mbox_free_channel(sba->mchan);
1742 	return ret;
1743 }
1744 
1745 static int sba_remove(struct platform_device *pdev)
1746 {
1747 	struct sba_device *sba = platform_get_drvdata(pdev);
1748 
1749 	dma_async_device_unregister(&sba->dma_dev);
1750 
1751 	debugfs_remove_recursive(sba->root);
1752 
1753 	sba_freeup_channel_resources(sba);
1754 
1755 	mbox_free_channel(sba->mchan);
1756 
1757 	return 0;
1758 }
1759 
1760 static const struct of_device_id sba_of_match[] = {
1761 	{ .compatible = "brcm,iproc-sba", },
1762 	{ .compatible = "brcm,iproc-sba-v2", },
1763 	{},
1764 };
1765 MODULE_DEVICE_TABLE(of, sba_of_match);
1766 
1767 static struct platform_driver sba_driver = {
1768 	.probe = sba_probe,
1769 	.remove = sba_remove,
1770 	.driver = {
1771 		.name = "bcm-sba-raid",
1772 		.of_match_table = sba_of_match,
1773 	},
1774 };
1775 module_platform_driver(sba_driver);
1776 
1777 MODULE_DESCRIPTION("Broadcom SBA RAID driver");
1778 MODULE_AUTHOR("Anup Patel <anup.patel@broadcom.com>");
1779 MODULE_LICENSE("GPL v2");
1780