xref: /linux/drivers/dma/bcm-sba-raid.c (revision 3d0fe49454652117522f60bfbefb978ba0e5300b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 // Copyright (C) 2017 Broadcom
3 
4 /*
5  * Broadcom SBA RAID Driver
6  *
7  * The Broadcom stream buffer accelerator (SBA) provides offloading
8  * capabilities for RAID operations. The SBA offload engine is accessible
9  * via Broadcom SoC specific ring manager. Two or more offload engines
10  * can share same Broadcom SoC specific ring manager due to this Broadcom
11  * SoC specific ring manager driver is implemented as a mailbox controller
12  * driver and offload engine drivers are implemented as mallbox clients.
13  *
14  * Typically, Broadcom SoC specific ring manager will implement larger
15  * number of hardware rings over one or more SBA hardware devices. By
16  * design, the internal buffer size of SBA hardware device is limited
17  * but all offload operations supported by SBA can be broken down into
18  * multiple small size requests and executed parallely on multiple SBA
19  * hardware devices for achieving high through-put.
20  *
21  * The Broadcom SBA RAID driver does not require any register programming
22  * except submitting request to SBA hardware device via mailbox channels.
23  * This driver implements a DMA device with one DMA channel using a single
24  * mailbox channel provided by Broadcom SoC specific ring manager driver.
25  * For having more SBA DMA channels, we can create more SBA device nodes
26  * in Broadcom SoC specific DTS based on number of hardware rings supported
27  * by Broadcom SoC ring manager.
28  */
29 
30 #include <linux/bitops.h>
31 #include <linux/debugfs.h>
32 #include <linux/dma-mapping.h>
33 #include <linux/dmaengine.h>
34 #include <linux/list.h>
35 #include <linux/mailbox_client.h>
36 #include <linux/mailbox/brcm-message.h>
37 #include <linux/module.h>
38 #include <linux/of.h>
39 #include <linux/of_platform.h>
40 #include <linux/platform_device.h>
41 #include <linux/slab.h>
42 #include <linux/raid/pq.h>
43 
44 #include "dmaengine.h"
45 
46 /* ====== Driver macros and defines ===== */
47 
48 #define SBA_TYPE_SHIFT					48
49 #define SBA_TYPE_MASK					GENMASK(1, 0)
50 #define SBA_TYPE_A					0x0
51 #define SBA_TYPE_B					0x2
52 #define SBA_TYPE_C					0x3
53 #define SBA_USER_DEF_SHIFT				32
54 #define SBA_USER_DEF_MASK				GENMASK(15, 0)
55 #define SBA_R_MDATA_SHIFT				24
56 #define SBA_R_MDATA_MASK				GENMASK(7, 0)
57 #define SBA_C_MDATA_MS_SHIFT				18
58 #define SBA_C_MDATA_MS_MASK				GENMASK(1, 0)
59 #define SBA_INT_SHIFT					17
60 #define SBA_INT_MASK					BIT(0)
61 #define SBA_RESP_SHIFT					16
62 #define SBA_RESP_MASK					BIT(0)
63 #define SBA_C_MDATA_SHIFT				8
64 #define SBA_C_MDATA_MASK				GENMASK(7, 0)
65 #define SBA_C_MDATA_BNUMx_SHIFT(__bnum)			(2 * (__bnum))
66 #define SBA_C_MDATA_BNUMx_MASK				GENMASK(1, 0)
67 #define SBA_C_MDATA_DNUM_SHIFT				5
68 #define SBA_C_MDATA_DNUM_MASK				GENMASK(4, 0)
69 #define SBA_C_MDATA_LS(__v)				((__v) & 0xff)
70 #define SBA_C_MDATA_MS(__v)				(((__v) >> 8) & 0x3)
71 #define SBA_CMD_SHIFT					0
72 #define SBA_CMD_MASK					GENMASK(3, 0)
73 #define SBA_CMD_ZERO_BUFFER				0x4
74 #define SBA_CMD_ZERO_ALL_BUFFERS			0x8
75 #define SBA_CMD_LOAD_BUFFER				0x9
76 #define SBA_CMD_XOR					0xa
77 #define SBA_CMD_GALOIS_XOR				0xb
78 #define SBA_CMD_WRITE_BUFFER				0xc
79 #define SBA_CMD_GALOIS					0xe
80 
81 #define SBA_MAX_REQ_PER_MBOX_CHANNEL			8192
82 #define SBA_MAX_MSG_SEND_PER_MBOX_CHANNEL		8
83 
84 /* Driver helper macros */
85 #define to_sba_request(tx)		\
86 	container_of(tx, struct sba_request, tx)
87 #define to_sba_device(dchan)		\
88 	container_of(dchan, struct sba_device, dma_chan)
89 
90 /* ===== Driver data structures ===== */
91 
92 enum sba_request_flags {
93 	SBA_REQUEST_STATE_FREE		= 0x001,
94 	SBA_REQUEST_STATE_ALLOCED	= 0x002,
95 	SBA_REQUEST_STATE_PENDING	= 0x004,
96 	SBA_REQUEST_STATE_ACTIVE	= 0x008,
97 	SBA_REQUEST_STATE_ABORTED	= 0x010,
98 	SBA_REQUEST_STATE_MASK		= 0x0ff,
99 	SBA_REQUEST_FENCE		= 0x100,
100 };
101 
102 struct sba_request {
103 	/* Global state */
104 	struct list_head node;
105 	struct sba_device *sba;
106 	u32 flags;
107 	/* Chained requests management */
108 	struct sba_request *first;
109 	struct list_head next;
110 	atomic_t next_pending_count;
111 	/* BRCM message data */
112 	struct brcm_message msg;
113 	struct dma_async_tx_descriptor tx;
114 	/* SBA commands */
115 	struct brcm_sba_command cmds[];
116 };
117 
118 enum sba_version {
119 	SBA_VER_1 = 0,
120 	SBA_VER_2
121 };
122 
123 struct sba_device {
124 	/* Underlying device */
125 	struct device *dev;
126 	/* DT configuration parameters */
127 	enum sba_version ver;
128 	/* Derived configuration parameters */
129 	u32 max_req;
130 	u32 hw_buf_size;
131 	u32 hw_resp_size;
132 	u32 max_pq_coefs;
133 	u32 max_pq_srcs;
134 	u32 max_cmd_per_req;
135 	u32 max_xor_srcs;
136 	u32 max_resp_pool_size;
137 	u32 max_cmds_pool_size;
138 	/* Maibox client and Mailbox channels */
139 	struct mbox_client client;
140 	struct mbox_chan *mchan;
141 	struct device *mbox_dev;
142 	/* DMA device and DMA channel */
143 	struct dma_device dma_dev;
144 	struct dma_chan dma_chan;
145 	/* DMA channel resources */
146 	void *resp_base;
147 	dma_addr_t resp_dma_base;
148 	void *cmds_base;
149 	dma_addr_t cmds_dma_base;
150 	spinlock_t reqs_lock;
151 	bool reqs_fence;
152 	struct list_head reqs_alloc_list;
153 	struct list_head reqs_pending_list;
154 	struct list_head reqs_active_list;
155 	struct list_head reqs_aborted_list;
156 	struct list_head reqs_free_list;
157 	/* DebugFS directory entries */
158 	struct dentry *root;
159 };
160 
161 /* ====== Command helper routines ===== */
162 
163 static inline u64 __pure sba_cmd_enc(u64 cmd, u32 val, u32 shift, u32 mask)
164 {
165 	cmd &= ~((u64)mask << shift);
166 	cmd |= ((u64)(val & mask) << shift);
167 	return cmd;
168 }
169 
170 static inline u32 __pure sba_cmd_load_c_mdata(u32 b0)
171 {
172 	return b0 & SBA_C_MDATA_BNUMx_MASK;
173 }
174 
175 static inline u32 __pure sba_cmd_write_c_mdata(u32 b0)
176 {
177 	return b0 & SBA_C_MDATA_BNUMx_MASK;
178 }
179 
180 static inline u32 __pure sba_cmd_xor_c_mdata(u32 b1, u32 b0)
181 {
182 	return (b0 & SBA_C_MDATA_BNUMx_MASK) |
183 	       ((b1 & SBA_C_MDATA_BNUMx_MASK) << SBA_C_MDATA_BNUMx_SHIFT(1));
184 }
185 
186 static inline u32 __pure sba_cmd_pq_c_mdata(u32 d, u32 b1, u32 b0)
187 {
188 	return (b0 & SBA_C_MDATA_BNUMx_MASK) |
189 	       ((b1 & SBA_C_MDATA_BNUMx_MASK) << SBA_C_MDATA_BNUMx_SHIFT(1)) |
190 	       ((d & SBA_C_MDATA_DNUM_MASK) << SBA_C_MDATA_DNUM_SHIFT);
191 }
192 
193 /* ====== General helper routines ===== */
194 
195 static struct sba_request *sba_alloc_request(struct sba_device *sba)
196 {
197 	bool found = false;
198 	unsigned long flags;
199 	struct sba_request *req = NULL;
200 
201 	spin_lock_irqsave(&sba->reqs_lock, flags);
202 	list_for_each_entry(req, &sba->reqs_free_list, node) {
203 		if (async_tx_test_ack(&req->tx)) {
204 			list_move_tail(&req->node, &sba->reqs_alloc_list);
205 			found = true;
206 			break;
207 		}
208 	}
209 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
210 
211 	if (!found) {
212 		/*
213 		 * We have no more free requests so, we peek
214 		 * mailbox channels hoping few active requests
215 		 * would have completed which will create more
216 		 * room for new requests.
217 		 */
218 		mbox_client_peek_data(sba->mchan);
219 		return NULL;
220 	}
221 
222 	req->flags = SBA_REQUEST_STATE_ALLOCED;
223 	req->first = req;
224 	INIT_LIST_HEAD(&req->next);
225 	atomic_set(&req->next_pending_count, 1);
226 
227 	dma_async_tx_descriptor_init(&req->tx, &sba->dma_chan);
228 	async_tx_ack(&req->tx);
229 
230 	return req;
231 }
232 
233 /* Note: Must be called with sba->reqs_lock held */
234 static void _sba_pending_request(struct sba_device *sba,
235 				 struct sba_request *req)
236 {
237 	lockdep_assert_held(&sba->reqs_lock);
238 	req->flags &= ~SBA_REQUEST_STATE_MASK;
239 	req->flags |= SBA_REQUEST_STATE_PENDING;
240 	list_move_tail(&req->node, &sba->reqs_pending_list);
241 	if (list_empty(&sba->reqs_active_list))
242 		sba->reqs_fence = false;
243 }
244 
245 /* Note: Must be called with sba->reqs_lock held */
246 static bool _sba_active_request(struct sba_device *sba,
247 				struct sba_request *req)
248 {
249 	lockdep_assert_held(&sba->reqs_lock);
250 	if (list_empty(&sba->reqs_active_list))
251 		sba->reqs_fence = false;
252 	if (sba->reqs_fence)
253 		return false;
254 	req->flags &= ~SBA_REQUEST_STATE_MASK;
255 	req->flags |= SBA_REQUEST_STATE_ACTIVE;
256 	list_move_tail(&req->node, &sba->reqs_active_list);
257 	if (req->flags & SBA_REQUEST_FENCE)
258 		sba->reqs_fence = true;
259 	return true;
260 }
261 
262 /* Note: Must be called with sba->reqs_lock held */
263 static void _sba_abort_request(struct sba_device *sba,
264 			       struct sba_request *req)
265 {
266 	lockdep_assert_held(&sba->reqs_lock);
267 	req->flags &= ~SBA_REQUEST_STATE_MASK;
268 	req->flags |= SBA_REQUEST_STATE_ABORTED;
269 	list_move_tail(&req->node, &sba->reqs_aborted_list);
270 	if (list_empty(&sba->reqs_active_list))
271 		sba->reqs_fence = false;
272 }
273 
274 /* Note: Must be called with sba->reqs_lock held */
275 static void _sba_free_request(struct sba_device *sba,
276 			      struct sba_request *req)
277 {
278 	lockdep_assert_held(&sba->reqs_lock);
279 	req->flags &= ~SBA_REQUEST_STATE_MASK;
280 	req->flags |= SBA_REQUEST_STATE_FREE;
281 	list_move_tail(&req->node, &sba->reqs_free_list);
282 	if (list_empty(&sba->reqs_active_list))
283 		sba->reqs_fence = false;
284 }
285 
286 static void sba_free_chained_requests(struct sba_request *req)
287 {
288 	unsigned long flags;
289 	struct sba_request *nreq;
290 	struct sba_device *sba = req->sba;
291 
292 	spin_lock_irqsave(&sba->reqs_lock, flags);
293 
294 	_sba_free_request(sba, req);
295 	list_for_each_entry(nreq, &req->next, next)
296 		_sba_free_request(sba, nreq);
297 
298 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
299 }
300 
301 static void sba_chain_request(struct sba_request *first,
302 			      struct sba_request *req)
303 {
304 	unsigned long flags;
305 	struct sba_device *sba = req->sba;
306 
307 	spin_lock_irqsave(&sba->reqs_lock, flags);
308 
309 	list_add_tail(&req->next, &first->next);
310 	req->first = first;
311 	atomic_inc(&first->next_pending_count);
312 
313 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
314 }
315 
316 static void sba_cleanup_nonpending_requests(struct sba_device *sba)
317 {
318 	unsigned long flags;
319 	struct sba_request *req, *req1;
320 
321 	spin_lock_irqsave(&sba->reqs_lock, flags);
322 
323 	/* Freeup all alloced request */
324 	list_for_each_entry_safe(req, req1, &sba->reqs_alloc_list, node)
325 		_sba_free_request(sba, req);
326 
327 	/* Set all active requests as aborted */
328 	list_for_each_entry_safe(req, req1, &sba->reqs_active_list, node)
329 		_sba_abort_request(sba, req);
330 
331 	/*
332 	 * Note: We expect that aborted request will be eventually
333 	 * freed by sba_receive_message()
334 	 */
335 
336 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
337 }
338 
339 static void sba_cleanup_pending_requests(struct sba_device *sba)
340 {
341 	unsigned long flags;
342 	struct sba_request *req, *req1;
343 
344 	spin_lock_irqsave(&sba->reqs_lock, flags);
345 
346 	/* Freeup all pending request */
347 	list_for_each_entry_safe(req, req1, &sba->reqs_pending_list, node)
348 		_sba_free_request(sba, req);
349 
350 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
351 }
352 
353 static int sba_send_mbox_request(struct sba_device *sba,
354 				 struct sba_request *req)
355 {
356 	int ret = 0;
357 
358 	/* Send message for the request */
359 	req->msg.error = 0;
360 	ret = mbox_send_message(sba->mchan, &req->msg);
361 	if (ret < 0) {
362 		dev_err(sba->dev, "send message failed with error %d", ret);
363 		return ret;
364 	}
365 
366 	/* Check error returned by mailbox controller */
367 	ret = req->msg.error;
368 	if (ret < 0) {
369 		dev_err(sba->dev, "message error %d", ret);
370 	}
371 
372 	/* Signal txdone for mailbox channel */
373 	mbox_client_txdone(sba->mchan, ret);
374 
375 	return ret;
376 }
377 
378 /* Note: Must be called with sba->reqs_lock held */
379 static void _sba_process_pending_requests(struct sba_device *sba)
380 {
381 	int ret;
382 	u32 count;
383 	struct sba_request *req;
384 
385 	/* Process few pending requests */
386 	count = SBA_MAX_MSG_SEND_PER_MBOX_CHANNEL;
387 	while (!list_empty(&sba->reqs_pending_list) && count) {
388 		/* Get the first pending request */
389 		req = list_first_entry(&sba->reqs_pending_list,
390 				       struct sba_request, node);
391 
392 		/* Try to make request active */
393 		if (!_sba_active_request(sba, req))
394 			break;
395 
396 		/* Send request to mailbox channel */
397 		ret = sba_send_mbox_request(sba, req);
398 		if (ret < 0) {
399 			_sba_pending_request(sba, req);
400 			break;
401 		}
402 
403 		count--;
404 	}
405 }
406 
407 static void sba_process_received_request(struct sba_device *sba,
408 					 struct sba_request *req)
409 {
410 	unsigned long flags;
411 	struct dma_async_tx_descriptor *tx;
412 	struct sba_request *nreq, *first = req->first;
413 
414 	/* Process only after all chained requests are received */
415 	if (!atomic_dec_return(&first->next_pending_count)) {
416 		tx = &first->tx;
417 
418 		WARN_ON(tx->cookie < 0);
419 		if (tx->cookie > 0) {
420 			spin_lock_irqsave(&sba->reqs_lock, flags);
421 			dma_cookie_complete(tx);
422 			spin_unlock_irqrestore(&sba->reqs_lock, flags);
423 			dmaengine_desc_get_callback_invoke(tx, NULL);
424 			dma_descriptor_unmap(tx);
425 			tx->callback = NULL;
426 			tx->callback_result = NULL;
427 		}
428 
429 		dma_run_dependencies(tx);
430 
431 		spin_lock_irqsave(&sba->reqs_lock, flags);
432 
433 		/* Free all requests chained to first request */
434 		list_for_each_entry(nreq, &first->next, next)
435 			_sba_free_request(sba, nreq);
436 		INIT_LIST_HEAD(&first->next);
437 
438 		/* Free the first request */
439 		_sba_free_request(sba, first);
440 
441 		/* Process pending requests */
442 		_sba_process_pending_requests(sba);
443 
444 		spin_unlock_irqrestore(&sba->reqs_lock, flags);
445 	}
446 }
447 
448 static void sba_write_stats_in_seqfile(struct sba_device *sba,
449 				       struct seq_file *file)
450 {
451 	unsigned long flags;
452 	struct sba_request *req;
453 	u32 free_count = 0, alloced_count = 0;
454 	u32 pending_count = 0, active_count = 0, aborted_count = 0;
455 
456 	spin_lock_irqsave(&sba->reqs_lock, flags);
457 
458 	list_for_each_entry(req, &sba->reqs_free_list, node)
459 		if (async_tx_test_ack(&req->tx))
460 			free_count++;
461 
462 	list_for_each_entry(req, &sba->reqs_alloc_list, node)
463 		alloced_count++;
464 
465 	list_for_each_entry(req, &sba->reqs_pending_list, node)
466 		pending_count++;
467 
468 	list_for_each_entry(req, &sba->reqs_active_list, node)
469 		active_count++;
470 
471 	list_for_each_entry(req, &sba->reqs_aborted_list, node)
472 		aborted_count++;
473 
474 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
475 
476 	seq_printf(file, "maximum requests   = %d\n", sba->max_req);
477 	seq_printf(file, "free requests      = %d\n", free_count);
478 	seq_printf(file, "alloced requests   = %d\n", alloced_count);
479 	seq_printf(file, "pending requests   = %d\n", pending_count);
480 	seq_printf(file, "active requests    = %d\n", active_count);
481 	seq_printf(file, "aborted requests   = %d\n", aborted_count);
482 }
483 
484 /* ====== DMAENGINE callbacks ===== */
485 
486 static void sba_free_chan_resources(struct dma_chan *dchan)
487 {
488 	/*
489 	 * Channel resources are pre-alloced so we just free-up
490 	 * whatever we can so that we can re-use pre-alloced
491 	 * channel resources next time.
492 	 */
493 	sba_cleanup_nonpending_requests(to_sba_device(dchan));
494 }
495 
496 static int sba_device_terminate_all(struct dma_chan *dchan)
497 {
498 	/* Cleanup all pending requests */
499 	sba_cleanup_pending_requests(to_sba_device(dchan));
500 
501 	return 0;
502 }
503 
504 static void sba_issue_pending(struct dma_chan *dchan)
505 {
506 	unsigned long flags;
507 	struct sba_device *sba = to_sba_device(dchan);
508 
509 	/* Process pending requests */
510 	spin_lock_irqsave(&sba->reqs_lock, flags);
511 	_sba_process_pending_requests(sba);
512 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
513 }
514 
515 static dma_cookie_t sba_tx_submit(struct dma_async_tx_descriptor *tx)
516 {
517 	unsigned long flags;
518 	dma_cookie_t cookie;
519 	struct sba_device *sba;
520 	struct sba_request *req, *nreq;
521 
522 	if (unlikely(!tx))
523 		return -EINVAL;
524 
525 	sba = to_sba_device(tx->chan);
526 	req = to_sba_request(tx);
527 
528 	/* Assign cookie and mark all chained requests pending */
529 	spin_lock_irqsave(&sba->reqs_lock, flags);
530 	cookie = dma_cookie_assign(tx);
531 	_sba_pending_request(sba, req);
532 	list_for_each_entry(nreq, &req->next, next)
533 		_sba_pending_request(sba, nreq);
534 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
535 
536 	return cookie;
537 }
538 
539 static enum dma_status sba_tx_status(struct dma_chan *dchan,
540 				     dma_cookie_t cookie,
541 				     struct dma_tx_state *txstate)
542 {
543 	enum dma_status ret;
544 	struct sba_device *sba = to_sba_device(dchan);
545 
546 	ret = dma_cookie_status(dchan, cookie, txstate);
547 	if (ret == DMA_COMPLETE)
548 		return ret;
549 
550 	mbox_client_peek_data(sba->mchan);
551 
552 	return dma_cookie_status(dchan, cookie, txstate);
553 }
554 
555 static void sba_fillup_interrupt_msg(struct sba_request *req,
556 				     struct brcm_sba_command *cmds,
557 				     struct brcm_message *msg)
558 {
559 	u64 cmd;
560 	u32 c_mdata;
561 	dma_addr_t resp_dma = req->tx.phys;
562 	struct brcm_sba_command *cmdsp = cmds;
563 
564 	/* Type-B command to load dummy data into buf0 */
565 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
566 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
567 	cmd = sba_cmd_enc(cmd, req->sba->hw_resp_size,
568 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
569 	c_mdata = sba_cmd_load_c_mdata(0);
570 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
571 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
572 	cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
573 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
574 	cmdsp->cmd = cmd;
575 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
576 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
577 	cmdsp->data = resp_dma;
578 	cmdsp->data_len = req->sba->hw_resp_size;
579 	cmdsp++;
580 
581 	/* Type-A command to write buf0 to dummy location */
582 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
583 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
584 	cmd = sba_cmd_enc(cmd, req->sba->hw_resp_size,
585 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
586 	cmd = sba_cmd_enc(cmd, 0x1,
587 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
588 	c_mdata = sba_cmd_write_c_mdata(0);
589 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
590 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
591 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
592 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
593 	cmdsp->cmd = cmd;
594 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
595 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
596 	if (req->sba->hw_resp_size) {
597 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
598 		cmdsp->resp = resp_dma;
599 		cmdsp->resp_len = req->sba->hw_resp_size;
600 	}
601 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
602 	cmdsp->data = resp_dma;
603 	cmdsp->data_len = req->sba->hw_resp_size;
604 	cmdsp++;
605 
606 	/* Fillup brcm_message */
607 	msg->type = BRCM_MESSAGE_SBA;
608 	msg->sba.cmds = cmds;
609 	msg->sba.cmds_count = cmdsp - cmds;
610 	msg->ctx = req;
611 	msg->error = 0;
612 }
613 
614 static struct dma_async_tx_descriptor *
615 sba_prep_dma_interrupt(struct dma_chan *dchan, unsigned long flags)
616 {
617 	struct sba_request *req = NULL;
618 	struct sba_device *sba = to_sba_device(dchan);
619 
620 	/* Alloc new request */
621 	req = sba_alloc_request(sba);
622 	if (!req)
623 		return NULL;
624 
625 	/*
626 	 * Force fence so that no requests are submitted
627 	 * until DMA callback for this request is invoked.
628 	 */
629 	req->flags |= SBA_REQUEST_FENCE;
630 
631 	/* Fillup request message */
632 	sba_fillup_interrupt_msg(req, req->cmds, &req->msg);
633 
634 	/* Init async_tx descriptor */
635 	req->tx.flags = flags;
636 	req->tx.cookie = -EBUSY;
637 
638 	return &req->tx;
639 }
640 
641 static void sba_fillup_memcpy_msg(struct sba_request *req,
642 				  struct brcm_sba_command *cmds,
643 				  struct brcm_message *msg,
644 				  dma_addr_t msg_offset, size_t msg_len,
645 				  dma_addr_t dst, dma_addr_t src)
646 {
647 	u64 cmd;
648 	u32 c_mdata;
649 	dma_addr_t resp_dma = req->tx.phys;
650 	struct brcm_sba_command *cmdsp = cmds;
651 
652 	/* Type-B command to load data into buf0 */
653 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
654 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
655 	cmd = sba_cmd_enc(cmd, msg_len,
656 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
657 	c_mdata = sba_cmd_load_c_mdata(0);
658 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
659 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
660 	cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
661 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
662 	cmdsp->cmd = cmd;
663 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
664 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
665 	cmdsp->data = src + msg_offset;
666 	cmdsp->data_len = msg_len;
667 	cmdsp++;
668 
669 	/* Type-A command to write buf0 */
670 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
671 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
672 	cmd = sba_cmd_enc(cmd, msg_len,
673 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
674 	cmd = sba_cmd_enc(cmd, 0x1,
675 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
676 	c_mdata = sba_cmd_write_c_mdata(0);
677 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
678 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
679 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
680 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
681 	cmdsp->cmd = cmd;
682 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
683 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
684 	if (req->sba->hw_resp_size) {
685 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
686 		cmdsp->resp = resp_dma;
687 		cmdsp->resp_len = req->sba->hw_resp_size;
688 	}
689 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
690 	cmdsp->data = dst + msg_offset;
691 	cmdsp->data_len = msg_len;
692 	cmdsp++;
693 
694 	/* Fillup brcm_message */
695 	msg->type = BRCM_MESSAGE_SBA;
696 	msg->sba.cmds = cmds;
697 	msg->sba.cmds_count = cmdsp - cmds;
698 	msg->ctx = req;
699 	msg->error = 0;
700 }
701 
702 static struct sba_request *
703 sba_prep_dma_memcpy_req(struct sba_device *sba,
704 			dma_addr_t off, dma_addr_t dst, dma_addr_t src,
705 			size_t len, unsigned long flags)
706 {
707 	struct sba_request *req = NULL;
708 
709 	/* Alloc new request */
710 	req = sba_alloc_request(sba);
711 	if (!req)
712 		return NULL;
713 	if (flags & DMA_PREP_FENCE)
714 		req->flags |= SBA_REQUEST_FENCE;
715 
716 	/* Fillup request message */
717 	sba_fillup_memcpy_msg(req, req->cmds, &req->msg,
718 			      off, len, dst, src);
719 
720 	/* Init async_tx descriptor */
721 	req->tx.flags = flags;
722 	req->tx.cookie = -EBUSY;
723 
724 	return req;
725 }
726 
727 static struct dma_async_tx_descriptor *
728 sba_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst, dma_addr_t src,
729 		    size_t len, unsigned long flags)
730 {
731 	size_t req_len;
732 	dma_addr_t off = 0;
733 	struct sba_device *sba = to_sba_device(dchan);
734 	struct sba_request *first = NULL, *req;
735 
736 	/* Create chained requests where each request is upto hw_buf_size */
737 	while (len) {
738 		req_len = (len < sba->hw_buf_size) ? len : sba->hw_buf_size;
739 
740 		req = sba_prep_dma_memcpy_req(sba, off, dst, src,
741 					      req_len, flags);
742 		if (!req) {
743 			if (first)
744 				sba_free_chained_requests(first);
745 			return NULL;
746 		}
747 
748 		if (first)
749 			sba_chain_request(first, req);
750 		else
751 			first = req;
752 
753 		off += req_len;
754 		len -= req_len;
755 	}
756 
757 	return (first) ? &first->tx : NULL;
758 }
759 
760 static void sba_fillup_xor_msg(struct sba_request *req,
761 				struct brcm_sba_command *cmds,
762 				struct brcm_message *msg,
763 				dma_addr_t msg_offset, size_t msg_len,
764 				dma_addr_t dst, dma_addr_t *src, u32 src_cnt)
765 {
766 	u64 cmd;
767 	u32 c_mdata;
768 	unsigned int i;
769 	dma_addr_t resp_dma = req->tx.phys;
770 	struct brcm_sba_command *cmdsp = cmds;
771 
772 	/* Type-B command to load data into buf0 */
773 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
774 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
775 	cmd = sba_cmd_enc(cmd, msg_len,
776 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
777 	c_mdata = sba_cmd_load_c_mdata(0);
778 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
779 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
780 	cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
781 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
782 	cmdsp->cmd = cmd;
783 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
784 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
785 	cmdsp->data = src[0] + msg_offset;
786 	cmdsp->data_len = msg_len;
787 	cmdsp++;
788 
789 	/* Type-B commands to xor data with buf0 and put it back in buf0 */
790 	for (i = 1; i < src_cnt; i++) {
791 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
792 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
793 		cmd = sba_cmd_enc(cmd, msg_len,
794 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
795 		c_mdata = sba_cmd_xor_c_mdata(0, 0);
796 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
797 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
798 		cmd = sba_cmd_enc(cmd, SBA_CMD_XOR,
799 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
800 		cmdsp->cmd = cmd;
801 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
802 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
803 		cmdsp->data = src[i] + msg_offset;
804 		cmdsp->data_len = msg_len;
805 		cmdsp++;
806 	}
807 
808 	/* Type-A command to write buf0 */
809 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
810 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
811 	cmd = sba_cmd_enc(cmd, msg_len,
812 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
813 	cmd = sba_cmd_enc(cmd, 0x1,
814 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
815 	c_mdata = sba_cmd_write_c_mdata(0);
816 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
817 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
818 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
819 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
820 	cmdsp->cmd = cmd;
821 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
822 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
823 	if (req->sba->hw_resp_size) {
824 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
825 		cmdsp->resp = resp_dma;
826 		cmdsp->resp_len = req->sba->hw_resp_size;
827 	}
828 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
829 	cmdsp->data = dst + msg_offset;
830 	cmdsp->data_len = msg_len;
831 	cmdsp++;
832 
833 	/* Fillup brcm_message */
834 	msg->type = BRCM_MESSAGE_SBA;
835 	msg->sba.cmds = cmds;
836 	msg->sba.cmds_count = cmdsp - cmds;
837 	msg->ctx = req;
838 	msg->error = 0;
839 }
840 
841 static struct sba_request *
842 sba_prep_dma_xor_req(struct sba_device *sba,
843 		     dma_addr_t off, dma_addr_t dst, dma_addr_t *src,
844 		     u32 src_cnt, size_t len, unsigned long flags)
845 {
846 	struct sba_request *req = NULL;
847 
848 	/* Alloc new request */
849 	req = sba_alloc_request(sba);
850 	if (!req)
851 		return NULL;
852 	if (flags & DMA_PREP_FENCE)
853 		req->flags |= SBA_REQUEST_FENCE;
854 
855 	/* Fillup request message */
856 	sba_fillup_xor_msg(req, req->cmds, &req->msg,
857 			   off, len, dst, src, src_cnt);
858 
859 	/* Init async_tx descriptor */
860 	req->tx.flags = flags;
861 	req->tx.cookie = -EBUSY;
862 
863 	return req;
864 }
865 
866 static struct dma_async_tx_descriptor *
867 sba_prep_dma_xor(struct dma_chan *dchan, dma_addr_t dst, dma_addr_t *src,
868 		 u32 src_cnt, size_t len, unsigned long flags)
869 {
870 	size_t req_len;
871 	dma_addr_t off = 0;
872 	struct sba_device *sba = to_sba_device(dchan);
873 	struct sba_request *first = NULL, *req;
874 
875 	/* Sanity checks */
876 	if (unlikely(src_cnt > sba->max_xor_srcs))
877 		return NULL;
878 
879 	/* Create chained requests where each request is upto hw_buf_size */
880 	while (len) {
881 		req_len = (len < sba->hw_buf_size) ? len : sba->hw_buf_size;
882 
883 		req = sba_prep_dma_xor_req(sba, off, dst, src, src_cnt,
884 					   req_len, flags);
885 		if (!req) {
886 			if (first)
887 				sba_free_chained_requests(first);
888 			return NULL;
889 		}
890 
891 		if (first)
892 			sba_chain_request(first, req);
893 		else
894 			first = req;
895 
896 		off += req_len;
897 		len -= req_len;
898 	}
899 
900 	return (first) ? &first->tx : NULL;
901 }
902 
903 static void sba_fillup_pq_msg(struct sba_request *req,
904 				bool pq_continue,
905 				struct brcm_sba_command *cmds,
906 				struct brcm_message *msg,
907 				dma_addr_t msg_offset, size_t msg_len,
908 				dma_addr_t *dst_p, dma_addr_t *dst_q,
909 				const u8 *scf, dma_addr_t *src, u32 src_cnt)
910 {
911 	u64 cmd;
912 	u32 c_mdata;
913 	unsigned int i;
914 	dma_addr_t resp_dma = req->tx.phys;
915 	struct brcm_sba_command *cmdsp = cmds;
916 
917 	if (pq_continue) {
918 		/* Type-B command to load old P into buf0 */
919 		if (dst_p) {
920 			cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
921 				SBA_TYPE_SHIFT, SBA_TYPE_MASK);
922 			cmd = sba_cmd_enc(cmd, msg_len,
923 				SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
924 			c_mdata = sba_cmd_load_c_mdata(0);
925 			cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
926 				SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
927 			cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
928 				SBA_CMD_SHIFT, SBA_CMD_MASK);
929 			cmdsp->cmd = cmd;
930 			*cmdsp->cmd_dma = cpu_to_le64(cmd);
931 			cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
932 			cmdsp->data = *dst_p + msg_offset;
933 			cmdsp->data_len = msg_len;
934 			cmdsp++;
935 		}
936 
937 		/* Type-B command to load old Q into buf1 */
938 		if (dst_q) {
939 			cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
940 				SBA_TYPE_SHIFT, SBA_TYPE_MASK);
941 			cmd = sba_cmd_enc(cmd, msg_len,
942 				SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
943 			c_mdata = sba_cmd_load_c_mdata(1);
944 			cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
945 				SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
946 			cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
947 				SBA_CMD_SHIFT, SBA_CMD_MASK);
948 			cmdsp->cmd = cmd;
949 			*cmdsp->cmd_dma = cpu_to_le64(cmd);
950 			cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
951 			cmdsp->data = *dst_q + msg_offset;
952 			cmdsp->data_len = msg_len;
953 			cmdsp++;
954 		}
955 	} else {
956 		/* Type-A command to zero all buffers */
957 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
958 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
959 		cmd = sba_cmd_enc(cmd, msg_len,
960 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
961 		cmd = sba_cmd_enc(cmd, SBA_CMD_ZERO_ALL_BUFFERS,
962 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
963 		cmdsp->cmd = cmd;
964 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
965 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
966 		cmdsp++;
967 	}
968 
969 	/* Type-B commands for generate P onto buf0 and Q onto buf1 */
970 	for (i = 0; i < src_cnt; i++) {
971 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
972 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
973 		cmd = sba_cmd_enc(cmd, msg_len,
974 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
975 		c_mdata = sba_cmd_pq_c_mdata(raid6_gflog[scf[i]], 1, 0);
976 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
977 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
978 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_MS(c_mdata),
979 				  SBA_C_MDATA_MS_SHIFT, SBA_C_MDATA_MS_MASK);
980 		cmd = sba_cmd_enc(cmd, SBA_CMD_GALOIS_XOR,
981 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
982 		cmdsp->cmd = cmd;
983 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
984 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
985 		cmdsp->data = src[i] + msg_offset;
986 		cmdsp->data_len = msg_len;
987 		cmdsp++;
988 	}
989 
990 	/* Type-A command to write buf0 */
991 	if (dst_p) {
992 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
993 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
994 		cmd = sba_cmd_enc(cmd, msg_len,
995 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
996 		cmd = sba_cmd_enc(cmd, 0x1,
997 				  SBA_RESP_SHIFT, SBA_RESP_MASK);
998 		c_mdata = sba_cmd_write_c_mdata(0);
999 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1000 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1001 		cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1002 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1003 		cmdsp->cmd = cmd;
1004 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1005 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1006 		if (req->sba->hw_resp_size) {
1007 			cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1008 			cmdsp->resp = resp_dma;
1009 			cmdsp->resp_len = req->sba->hw_resp_size;
1010 		}
1011 		cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1012 		cmdsp->data = *dst_p + msg_offset;
1013 		cmdsp->data_len = msg_len;
1014 		cmdsp++;
1015 	}
1016 
1017 	/* Type-A command to write buf1 */
1018 	if (dst_q) {
1019 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1020 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1021 		cmd = sba_cmd_enc(cmd, msg_len,
1022 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1023 		cmd = sba_cmd_enc(cmd, 0x1,
1024 				  SBA_RESP_SHIFT, SBA_RESP_MASK);
1025 		c_mdata = sba_cmd_write_c_mdata(1);
1026 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1027 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1028 		cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1029 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1030 		cmdsp->cmd = cmd;
1031 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1032 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1033 		if (req->sba->hw_resp_size) {
1034 			cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1035 			cmdsp->resp = resp_dma;
1036 			cmdsp->resp_len = req->sba->hw_resp_size;
1037 		}
1038 		cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1039 		cmdsp->data = *dst_q + msg_offset;
1040 		cmdsp->data_len = msg_len;
1041 		cmdsp++;
1042 	}
1043 
1044 	/* Fillup brcm_message */
1045 	msg->type = BRCM_MESSAGE_SBA;
1046 	msg->sba.cmds = cmds;
1047 	msg->sba.cmds_count = cmdsp - cmds;
1048 	msg->ctx = req;
1049 	msg->error = 0;
1050 }
1051 
1052 static struct sba_request *
1053 sba_prep_dma_pq_req(struct sba_device *sba, dma_addr_t off,
1054 		    dma_addr_t *dst_p, dma_addr_t *dst_q, dma_addr_t *src,
1055 		    u32 src_cnt, const u8 *scf, size_t len, unsigned long flags)
1056 {
1057 	struct sba_request *req = NULL;
1058 
1059 	/* Alloc new request */
1060 	req = sba_alloc_request(sba);
1061 	if (!req)
1062 		return NULL;
1063 	if (flags & DMA_PREP_FENCE)
1064 		req->flags |= SBA_REQUEST_FENCE;
1065 
1066 	/* Fillup request messages */
1067 	sba_fillup_pq_msg(req, dmaf_continue(flags),
1068 			  req->cmds, &req->msg,
1069 			  off, len, dst_p, dst_q, scf, src, src_cnt);
1070 
1071 	/* Init async_tx descriptor */
1072 	req->tx.flags = flags;
1073 	req->tx.cookie = -EBUSY;
1074 
1075 	return req;
1076 }
1077 
1078 static void sba_fillup_pq_single_msg(struct sba_request *req,
1079 				bool pq_continue,
1080 				struct brcm_sba_command *cmds,
1081 				struct brcm_message *msg,
1082 				dma_addr_t msg_offset, size_t msg_len,
1083 				dma_addr_t *dst_p, dma_addr_t *dst_q,
1084 				dma_addr_t src, u8 scf)
1085 {
1086 	u64 cmd;
1087 	u32 c_mdata;
1088 	u8 pos, dpos = raid6_gflog[scf];
1089 	dma_addr_t resp_dma = req->tx.phys;
1090 	struct brcm_sba_command *cmdsp = cmds;
1091 
1092 	if (!dst_p)
1093 		goto skip_p;
1094 
1095 	if (pq_continue) {
1096 		/* Type-B command to load old P into buf0 */
1097 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1098 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1099 		cmd = sba_cmd_enc(cmd, msg_len,
1100 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1101 		c_mdata = sba_cmd_load_c_mdata(0);
1102 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1103 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1104 		cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
1105 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1106 		cmdsp->cmd = cmd;
1107 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1108 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1109 		cmdsp->data = *dst_p + msg_offset;
1110 		cmdsp->data_len = msg_len;
1111 		cmdsp++;
1112 
1113 		/*
1114 		 * Type-B commands to xor data with buf0 and put it
1115 		 * back in buf0
1116 		 */
1117 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1118 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1119 		cmd = sba_cmd_enc(cmd, msg_len,
1120 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1121 		c_mdata = sba_cmd_xor_c_mdata(0, 0);
1122 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1123 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1124 		cmd = sba_cmd_enc(cmd, SBA_CMD_XOR,
1125 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1126 		cmdsp->cmd = cmd;
1127 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1128 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1129 		cmdsp->data = src + msg_offset;
1130 		cmdsp->data_len = msg_len;
1131 		cmdsp++;
1132 	} else {
1133 		/* Type-B command to load old P into buf0 */
1134 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1135 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1136 		cmd = sba_cmd_enc(cmd, msg_len,
1137 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1138 		c_mdata = sba_cmd_load_c_mdata(0);
1139 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1140 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1141 		cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
1142 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1143 		cmdsp->cmd = cmd;
1144 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1145 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1146 		cmdsp->data = src + msg_offset;
1147 		cmdsp->data_len = msg_len;
1148 		cmdsp++;
1149 	}
1150 
1151 	/* Type-A command to write buf0 */
1152 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1153 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1154 	cmd = sba_cmd_enc(cmd, msg_len,
1155 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1156 	cmd = sba_cmd_enc(cmd, 0x1,
1157 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
1158 	c_mdata = sba_cmd_write_c_mdata(0);
1159 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1160 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1161 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1162 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1163 	cmdsp->cmd = cmd;
1164 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1165 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1166 	if (req->sba->hw_resp_size) {
1167 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1168 		cmdsp->resp = resp_dma;
1169 		cmdsp->resp_len = req->sba->hw_resp_size;
1170 	}
1171 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1172 	cmdsp->data = *dst_p + msg_offset;
1173 	cmdsp->data_len = msg_len;
1174 	cmdsp++;
1175 
1176 skip_p:
1177 	if (!dst_q)
1178 		goto skip_q;
1179 
1180 	/* Type-A command to zero all buffers */
1181 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1182 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1183 	cmd = sba_cmd_enc(cmd, msg_len,
1184 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1185 	cmd = sba_cmd_enc(cmd, SBA_CMD_ZERO_ALL_BUFFERS,
1186 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1187 	cmdsp->cmd = cmd;
1188 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1189 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1190 	cmdsp++;
1191 
1192 	if (dpos == 255)
1193 		goto skip_q_computation;
1194 	pos = (dpos < req->sba->max_pq_coefs) ?
1195 		dpos : (req->sba->max_pq_coefs - 1);
1196 
1197 	/*
1198 	 * Type-B command to generate initial Q from data
1199 	 * and store output into buf0
1200 	 */
1201 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1202 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1203 	cmd = sba_cmd_enc(cmd, msg_len,
1204 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1205 	c_mdata = sba_cmd_pq_c_mdata(pos, 0, 0);
1206 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1207 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1208 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_MS(c_mdata),
1209 			  SBA_C_MDATA_MS_SHIFT, SBA_C_MDATA_MS_MASK);
1210 	cmd = sba_cmd_enc(cmd, SBA_CMD_GALOIS,
1211 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1212 	cmdsp->cmd = cmd;
1213 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1214 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1215 	cmdsp->data = src + msg_offset;
1216 	cmdsp->data_len = msg_len;
1217 	cmdsp++;
1218 
1219 	dpos -= pos;
1220 
1221 	/* Multiple Type-A command to generate final Q */
1222 	while (dpos) {
1223 		pos = (dpos < req->sba->max_pq_coefs) ?
1224 			dpos : (req->sba->max_pq_coefs - 1);
1225 
1226 		/*
1227 		 * Type-A command to generate Q with buf0 and
1228 		 * buf1 store result in buf0
1229 		 */
1230 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1231 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1232 		cmd = sba_cmd_enc(cmd, msg_len,
1233 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1234 		c_mdata = sba_cmd_pq_c_mdata(pos, 0, 1);
1235 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1236 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1237 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_MS(c_mdata),
1238 				  SBA_C_MDATA_MS_SHIFT, SBA_C_MDATA_MS_MASK);
1239 		cmd = sba_cmd_enc(cmd, SBA_CMD_GALOIS,
1240 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1241 		cmdsp->cmd = cmd;
1242 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1243 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1244 		cmdsp++;
1245 
1246 		dpos -= pos;
1247 	}
1248 
1249 skip_q_computation:
1250 	if (pq_continue) {
1251 		/*
1252 		 * Type-B command to XOR previous output with
1253 		 * buf0 and write it into buf0
1254 		 */
1255 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1256 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1257 		cmd = sba_cmd_enc(cmd, msg_len,
1258 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1259 		c_mdata = sba_cmd_xor_c_mdata(0, 0);
1260 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1261 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1262 		cmd = sba_cmd_enc(cmd, SBA_CMD_XOR,
1263 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1264 		cmdsp->cmd = cmd;
1265 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1266 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1267 		cmdsp->data = *dst_q + msg_offset;
1268 		cmdsp->data_len = msg_len;
1269 		cmdsp++;
1270 	}
1271 
1272 	/* Type-A command to write buf0 */
1273 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1274 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1275 	cmd = sba_cmd_enc(cmd, msg_len,
1276 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1277 	cmd = sba_cmd_enc(cmd, 0x1,
1278 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
1279 	c_mdata = sba_cmd_write_c_mdata(0);
1280 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1281 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1282 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1283 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1284 	cmdsp->cmd = cmd;
1285 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1286 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1287 	if (req->sba->hw_resp_size) {
1288 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1289 		cmdsp->resp = resp_dma;
1290 		cmdsp->resp_len = req->sba->hw_resp_size;
1291 	}
1292 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1293 	cmdsp->data = *dst_q + msg_offset;
1294 	cmdsp->data_len = msg_len;
1295 	cmdsp++;
1296 
1297 skip_q:
1298 	/* Fillup brcm_message */
1299 	msg->type = BRCM_MESSAGE_SBA;
1300 	msg->sba.cmds = cmds;
1301 	msg->sba.cmds_count = cmdsp - cmds;
1302 	msg->ctx = req;
1303 	msg->error = 0;
1304 }
1305 
1306 static struct sba_request *
1307 sba_prep_dma_pq_single_req(struct sba_device *sba, dma_addr_t off,
1308 			   dma_addr_t *dst_p, dma_addr_t *dst_q,
1309 			   dma_addr_t src, u8 scf, size_t len,
1310 			   unsigned long flags)
1311 {
1312 	struct sba_request *req = NULL;
1313 
1314 	/* Alloc new request */
1315 	req = sba_alloc_request(sba);
1316 	if (!req)
1317 		return NULL;
1318 	if (flags & DMA_PREP_FENCE)
1319 		req->flags |= SBA_REQUEST_FENCE;
1320 
1321 	/* Fillup request messages */
1322 	sba_fillup_pq_single_msg(req,  dmaf_continue(flags),
1323 				 req->cmds, &req->msg, off, len,
1324 				 dst_p, dst_q, src, scf);
1325 
1326 	/* Init async_tx descriptor */
1327 	req->tx.flags = flags;
1328 	req->tx.cookie = -EBUSY;
1329 
1330 	return req;
1331 }
1332 
1333 static struct dma_async_tx_descriptor *
1334 sba_prep_dma_pq(struct dma_chan *dchan, dma_addr_t *dst, dma_addr_t *src,
1335 		u32 src_cnt, const u8 *scf, size_t len, unsigned long flags)
1336 {
1337 	u32 i, dst_q_index;
1338 	size_t req_len;
1339 	bool slow = false;
1340 	dma_addr_t off = 0;
1341 	dma_addr_t *dst_p = NULL, *dst_q = NULL;
1342 	struct sba_device *sba = to_sba_device(dchan);
1343 	struct sba_request *first = NULL, *req;
1344 
1345 	/* Sanity checks */
1346 	if (unlikely(src_cnt > sba->max_pq_srcs))
1347 		return NULL;
1348 	for (i = 0; i < src_cnt; i++)
1349 		if (sba->max_pq_coefs <= raid6_gflog[scf[i]])
1350 			slow = true;
1351 
1352 	/* Figure-out P and Q destination addresses */
1353 	if (!(flags & DMA_PREP_PQ_DISABLE_P))
1354 		dst_p = &dst[0];
1355 	if (!(flags & DMA_PREP_PQ_DISABLE_Q))
1356 		dst_q = &dst[1];
1357 
1358 	/* Create chained requests where each request is upto hw_buf_size */
1359 	while (len) {
1360 		req_len = (len < sba->hw_buf_size) ? len : sba->hw_buf_size;
1361 
1362 		if (slow) {
1363 			dst_q_index = src_cnt;
1364 
1365 			if (dst_q) {
1366 				for (i = 0; i < src_cnt; i++) {
1367 					if (*dst_q == src[i]) {
1368 						dst_q_index = i;
1369 						break;
1370 					}
1371 				}
1372 			}
1373 
1374 			if (dst_q_index < src_cnt) {
1375 				i = dst_q_index;
1376 				req = sba_prep_dma_pq_single_req(sba,
1377 					off, dst_p, dst_q, src[i], scf[i],
1378 					req_len, flags | DMA_PREP_FENCE);
1379 				if (!req)
1380 					goto fail;
1381 
1382 				if (first)
1383 					sba_chain_request(first, req);
1384 				else
1385 					first = req;
1386 
1387 				flags |= DMA_PREP_CONTINUE;
1388 			}
1389 
1390 			for (i = 0; i < src_cnt; i++) {
1391 				if (dst_q_index == i)
1392 					continue;
1393 
1394 				req = sba_prep_dma_pq_single_req(sba,
1395 					off, dst_p, dst_q, src[i], scf[i],
1396 					req_len, flags | DMA_PREP_FENCE);
1397 				if (!req)
1398 					goto fail;
1399 
1400 				if (first)
1401 					sba_chain_request(first, req);
1402 				else
1403 					first = req;
1404 
1405 				flags |= DMA_PREP_CONTINUE;
1406 			}
1407 		} else {
1408 			req = sba_prep_dma_pq_req(sba, off,
1409 						  dst_p, dst_q, src, src_cnt,
1410 						  scf, req_len, flags);
1411 			if (!req)
1412 				goto fail;
1413 
1414 			if (first)
1415 				sba_chain_request(first, req);
1416 			else
1417 				first = req;
1418 		}
1419 
1420 		off += req_len;
1421 		len -= req_len;
1422 	}
1423 
1424 	return (first) ? &first->tx : NULL;
1425 
1426 fail:
1427 	if (first)
1428 		sba_free_chained_requests(first);
1429 	return NULL;
1430 }
1431 
1432 /* ====== Mailbox callbacks ===== */
1433 
1434 static void sba_receive_message(struct mbox_client *cl, void *msg)
1435 {
1436 	struct brcm_message *m = msg;
1437 	struct sba_request *req = m->ctx;
1438 	struct sba_device *sba = req->sba;
1439 
1440 	/* Error count if message has error */
1441 	if (m->error < 0)
1442 		dev_err(sba->dev, "%s got message with error %d",
1443 			dma_chan_name(&sba->dma_chan), m->error);
1444 
1445 	/* Process received request */
1446 	sba_process_received_request(sba, req);
1447 }
1448 
1449 /* ====== Debugfs callbacks ====== */
1450 
1451 static int sba_debugfs_stats_show(struct seq_file *file, void *offset)
1452 {
1453 	struct sba_device *sba = dev_get_drvdata(file->private);
1454 
1455 	/* Write stats in file */
1456 	sba_write_stats_in_seqfile(sba, file);
1457 
1458 	return 0;
1459 }
1460 
1461 /* ====== Platform driver routines ===== */
1462 
1463 static int sba_prealloc_channel_resources(struct sba_device *sba)
1464 {
1465 	int i, j, ret = 0;
1466 	struct sba_request *req = NULL;
1467 
1468 	sba->resp_base = dma_alloc_coherent(sba->mbox_dev,
1469 					    sba->max_resp_pool_size,
1470 					    &sba->resp_dma_base, GFP_KERNEL);
1471 	if (!sba->resp_base)
1472 		return -ENOMEM;
1473 
1474 	sba->cmds_base = dma_alloc_coherent(sba->mbox_dev,
1475 					    sba->max_cmds_pool_size,
1476 					    &sba->cmds_dma_base, GFP_KERNEL);
1477 	if (!sba->cmds_base) {
1478 		ret = -ENOMEM;
1479 		goto fail_free_resp_pool;
1480 	}
1481 
1482 	spin_lock_init(&sba->reqs_lock);
1483 	sba->reqs_fence = false;
1484 	INIT_LIST_HEAD(&sba->reqs_alloc_list);
1485 	INIT_LIST_HEAD(&sba->reqs_pending_list);
1486 	INIT_LIST_HEAD(&sba->reqs_active_list);
1487 	INIT_LIST_HEAD(&sba->reqs_aborted_list);
1488 	INIT_LIST_HEAD(&sba->reqs_free_list);
1489 
1490 	for (i = 0; i < sba->max_req; i++) {
1491 		req = devm_kzalloc(sba->dev,
1492 				   struct_size(req, cmds, sba->max_cmd_per_req),
1493 				   GFP_KERNEL);
1494 		if (!req) {
1495 			ret = -ENOMEM;
1496 			goto fail_free_cmds_pool;
1497 		}
1498 		INIT_LIST_HEAD(&req->node);
1499 		req->sba = sba;
1500 		req->flags = SBA_REQUEST_STATE_FREE;
1501 		INIT_LIST_HEAD(&req->next);
1502 		atomic_set(&req->next_pending_count, 0);
1503 		for (j = 0; j < sba->max_cmd_per_req; j++) {
1504 			req->cmds[j].cmd = 0;
1505 			req->cmds[j].cmd_dma = sba->cmds_base +
1506 				(i * sba->max_cmd_per_req + j) * sizeof(u64);
1507 			req->cmds[j].cmd_dma_addr = sba->cmds_dma_base +
1508 				(i * sba->max_cmd_per_req + j) * sizeof(u64);
1509 			req->cmds[j].flags = 0;
1510 		}
1511 		memset(&req->msg, 0, sizeof(req->msg));
1512 		dma_async_tx_descriptor_init(&req->tx, &sba->dma_chan);
1513 		async_tx_ack(&req->tx);
1514 		req->tx.tx_submit = sba_tx_submit;
1515 		req->tx.phys = sba->resp_dma_base + i * sba->hw_resp_size;
1516 		list_add_tail(&req->node, &sba->reqs_free_list);
1517 	}
1518 
1519 	return 0;
1520 
1521 fail_free_cmds_pool:
1522 	dma_free_coherent(sba->mbox_dev,
1523 			  sba->max_cmds_pool_size,
1524 			  sba->cmds_base, sba->cmds_dma_base);
1525 fail_free_resp_pool:
1526 	dma_free_coherent(sba->mbox_dev,
1527 			  sba->max_resp_pool_size,
1528 			  sba->resp_base, sba->resp_dma_base);
1529 	return ret;
1530 }
1531 
1532 static void sba_freeup_channel_resources(struct sba_device *sba)
1533 {
1534 	dmaengine_terminate_all(&sba->dma_chan);
1535 	dma_free_coherent(sba->mbox_dev, sba->max_cmds_pool_size,
1536 			  sba->cmds_base, sba->cmds_dma_base);
1537 	dma_free_coherent(sba->mbox_dev, sba->max_resp_pool_size,
1538 			  sba->resp_base, sba->resp_dma_base);
1539 	sba->resp_base = NULL;
1540 	sba->resp_dma_base = 0;
1541 }
1542 
1543 static int sba_async_register(struct sba_device *sba)
1544 {
1545 	int ret;
1546 	struct dma_device *dma_dev = &sba->dma_dev;
1547 
1548 	/* Initialize DMA channel cookie */
1549 	sba->dma_chan.device = dma_dev;
1550 	dma_cookie_init(&sba->dma_chan);
1551 
1552 	/* Initialize DMA device capability mask */
1553 	dma_cap_zero(dma_dev->cap_mask);
1554 	dma_cap_set(DMA_INTERRUPT, dma_dev->cap_mask);
1555 	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
1556 	dma_cap_set(DMA_XOR, dma_dev->cap_mask);
1557 	dma_cap_set(DMA_PQ, dma_dev->cap_mask);
1558 
1559 	/*
1560 	 * Set mailbox channel device as the base device of
1561 	 * our dma_device because the actual memory accesses
1562 	 * will be done by mailbox controller
1563 	 */
1564 	dma_dev->dev = sba->mbox_dev;
1565 
1566 	/* Set base prep routines */
1567 	dma_dev->device_free_chan_resources = sba_free_chan_resources;
1568 	dma_dev->device_terminate_all = sba_device_terminate_all;
1569 	dma_dev->device_issue_pending = sba_issue_pending;
1570 	dma_dev->device_tx_status = sba_tx_status;
1571 
1572 	/* Set interrupt routine */
1573 	if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1574 		dma_dev->device_prep_dma_interrupt = sba_prep_dma_interrupt;
1575 
1576 	/* Set memcpy routine */
1577 	if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1578 		dma_dev->device_prep_dma_memcpy = sba_prep_dma_memcpy;
1579 
1580 	/* Set xor routine and capability */
1581 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1582 		dma_dev->device_prep_dma_xor = sba_prep_dma_xor;
1583 		dma_dev->max_xor = sba->max_xor_srcs;
1584 	}
1585 
1586 	/* Set pq routine and capability */
1587 	if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1588 		dma_dev->device_prep_dma_pq = sba_prep_dma_pq;
1589 		dma_set_maxpq(dma_dev, sba->max_pq_srcs, 0);
1590 	}
1591 
1592 	/* Initialize DMA device channel list */
1593 	INIT_LIST_HEAD(&dma_dev->channels);
1594 	list_add_tail(&sba->dma_chan.device_node, &dma_dev->channels);
1595 
1596 	/* Register with Linux async DMA framework*/
1597 	ret = dma_async_device_register(dma_dev);
1598 	if (ret) {
1599 		dev_err(sba->dev, "async device register error %d", ret);
1600 		return ret;
1601 	}
1602 
1603 	dev_info(sba->dev, "%s capabilities: %s%s%s%s\n",
1604 	dma_chan_name(&sba->dma_chan),
1605 	dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "interrupt " : "",
1606 	dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "memcpy " : "",
1607 	dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1608 	dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "pq " : "");
1609 
1610 	return 0;
1611 }
1612 
1613 static int sba_probe(struct platform_device *pdev)
1614 {
1615 	int ret = 0;
1616 	struct sba_device *sba;
1617 	struct platform_device *mbox_pdev;
1618 	struct of_phandle_args args;
1619 
1620 	/* Allocate main SBA struct */
1621 	sba = devm_kzalloc(&pdev->dev, sizeof(*sba), GFP_KERNEL);
1622 	if (!sba)
1623 		return -ENOMEM;
1624 
1625 	sba->dev = &pdev->dev;
1626 	platform_set_drvdata(pdev, sba);
1627 
1628 	/* Number of mailbox channels should be atleast 1 */
1629 	ret = of_count_phandle_with_args(pdev->dev.of_node,
1630 					 "mboxes", "#mbox-cells");
1631 	if (ret <= 0)
1632 		return -ENODEV;
1633 
1634 	/* Determine SBA version from DT compatible string */
1635 	if (of_device_is_compatible(sba->dev->of_node, "brcm,iproc-sba"))
1636 		sba->ver = SBA_VER_1;
1637 	else if (of_device_is_compatible(sba->dev->of_node,
1638 					 "brcm,iproc-sba-v2"))
1639 		sba->ver = SBA_VER_2;
1640 	else
1641 		return -ENODEV;
1642 
1643 	/* Derived Configuration parameters */
1644 	switch (sba->ver) {
1645 	case SBA_VER_1:
1646 		sba->hw_buf_size = 4096;
1647 		sba->hw_resp_size = 8;
1648 		sba->max_pq_coefs = 6;
1649 		sba->max_pq_srcs = 6;
1650 		break;
1651 	case SBA_VER_2:
1652 		sba->hw_buf_size = 4096;
1653 		sba->hw_resp_size = 8;
1654 		sba->max_pq_coefs = 30;
1655 		/*
1656 		 * We can support max_pq_srcs == max_pq_coefs because
1657 		 * we are limited by number of SBA commands that we can
1658 		 * fit in one message for underlying ring manager HW.
1659 		 */
1660 		sba->max_pq_srcs = 12;
1661 		break;
1662 	default:
1663 		return -EINVAL;
1664 	}
1665 	sba->max_req = SBA_MAX_REQ_PER_MBOX_CHANNEL;
1666 	sba->max_cmd_per_req = sba->max_pq_srcs + 3;
1667 	sba->max_xor_srcs = sba->max_cmd_per_req - 1;
1668 	sba->max_resp_pool_size = sba->max_req * sba->hw_resp_size;
1669 	sba->max_cmds_pool_size = sba->max_req *
1670 				  sba->max_cmd_per_req * sizeof(u64);
1671 
1672 	/* Setup mailbox client */
1673 	sba->client.dev			= &pdev->dev;
1674 	sba->client.rx_callback		= sba_receive_message;
1675 	sba->client.tx_block		= false;
1676 	sba->client.knows_txdone	= true;
1677 	sba->client.tx_tout		= 0;
1678 
1679 	/* Request mailbox channel */
1680 	sba->mchan = mbox_request_channel(&sba->client, 0);
1681 	if (IS_ERR(sba->mchan)) {
1682 		ret = PTR_ERR(sba->mchan);
1683 		goto fail_free_mchan;
1684 	}
1685 
1686 	/* Find-out underlying mailbox device */
1687 	ret = of_parse_phandle_with_args(pdev->dev.of_node,
1688 					 "mboxes", "#mbox-cells", 0, &args);
1689 	if (ret)
1690 		goto fail_free_mchan;
1691 	mbox_pdev = of_find_device_by_node(args.np);
1692 	of_node_put(args.np);
1693 	if (!mbox_pdev) {
1694 		ret = -ENODEV;
1695 		goto fail_free_mchan;
1696 	}
1697 	sba->mbox_dev = &mbox_pdev->dev;
1698 
1699 	/* Prealloc channel resource */
1700 	ret = sba_prealloc_channel_resources(sba);
1701 	if (ret)
1702 		goto fail_free_mchan;
1703 
1704 	/* Check availability of debugfs */
1705 	if (!debugfs_initialized())
1706 		goto skip_debugfs;
1707 
1708 	/* Create debugfs root entry */
1709 	sba->root = debugfs_create_dir(dev_name(sba->dev), NULL);
1710 
1711 	/* Create debugfs stats entry */
1712 	debugfs_create_devm_seqfile(sba->dev, "stats", sba->root,
1713 				    sba_debugfs_stats_show);
1714 
1715 skip_debugfs:
1716 
1717 	/* Register DMA device with Linux async framework */
1718 	ret = sba_async_register(sba);
1719 	if (ret)
1720 		goto fail_free_resources;
1721 
1722 	/* Print device info */
1723 	dev_info(sba->dev, "%s using SBAv%d mailbox channel from %s",
1724 		 dma_chan_name(&sba->dma_chan), sba->ver+1,
1725 		 dev_name(sba->mbox_dev));
1726 
1727 	return 0;
1728 
1729 fail_free_resources:
1730 	debugfs_remove_recursive(sba->root);
1731 	sba_freeup_channel_resources(sba);
1732 fail_free_mchan:
1733 	mbox_free_channel(sba->mchan);
1734 	return ret;
1735 }
1736 
1737 static void sba_remove(struct platform_device *pdev)
1738 {
1739 	struct sba_device *sba = platform_get_drvdata(pdev);
1740 
1741 	dma_async_device_unregister(&sba->dma_dev);
1742 
1743 	debugfs_remove_recursive(sba->root);
1744 
1745 	sba_freeup_channel_resources(sba);
1746 
1747 	mbox_free_channel(sba->mchan);
1748 }
1749 
1750 static const struct of_device_id sba_of_match[] = {
1751 	{ .compatible = "brcm,iproc-sba", },
1752 	{ .compatible = "brcm,iproc-sba-v2", },
1753 	{},
1754 };
1755 MODULE_DEVICE_TABLE(of, sba_of_match);
1756 
1757 static struct platform_driver sba_driver = {
1758 	.probe = sba_probe,
1759 	.remove_new = sba_remove,
1760 	.driver = {
1761 		.name = "bcm-sba-raid",
1762 		.of_match_table = sba_of_match,
1763 	},
1764 };
1765 module_platform_driver(sba_driver);
1766 
1767 MODULE_DESCRIPTION("Broadcom SBA RAID driver");
1768 MODULE_AUTHOR("Anup Patel <anup.patel@broadcom.com>");
1769 MODULE_LICENSE("GPL v2");
1770