xref: /openbmc/linux/drivers/dma/bcm-sba-raid.c (revision 7559e757)
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_device.h>
39 #include <linux/slab.h>
40 #include <linux/raid/pq.h>
41 
42 #include "dmaengine.h"
43 
44 /* ====== Driver macros and defines ===== */
45 
46 #define SBA_TYPE_SHIFT					48
47 #define SBA_TYPE_MASK					GENMASK(1, 0)
48 #define SBA_TYPE_A					0x0
49 #define SBA_TYPE_B					0x2
50 #define SBA_TYPE_C					0x3
51 #define SBA_USER_DEF_SHIFT				32
52 #define SBA_USER_DEF_MASK				GENMASK(15, 0)
53 #define SBA_R_MDATA_SHIFT				24
54 #define SBA_R_MDATA_MASK				GENMASK(7, 0)
55 #define SBA_C_MDATA_MS_SHIFT				18
56 #define SBA_C_MDATA_MS_MASK				GENMASK(1, 0)
57 #define SBA_INT_SHIFT					17
58 #define SBA_INT_MASK					BIT(0)
59 #define SBA_RESP_SHIFT					16
60 #define SBA_RESP_MASK					BIT(0)
61 #define SBA_C_MDATA_SHIFT				8
62 #define SBA_C_MDATA_MASK				GENMASK(7, 0)
63 #define SBA_C_MDATA_BNUMx_SHIFT(__bnum)			(2 * (__bnum))
64 #define SBA_C_MDATA_BNUMx_MASK				GENMASK(1, 0)
65 #define SBA_C_MDATA_DNUM_SHIFT				5
66 #define SBA_C_MDATA_DNUM_MASK				GENMASK(4, 0)
67 #define SBA_C_MDATA_LS(__v)				((__v) & 0xff)
68 #define SBA_C_MDATA_MS(__v)				(((__v) >> 8) & 0x3)
69 #define SBA_CMD_SHIFT					0
70 #define SBA_CMD_MASK					GENMASK(3, 0)
71 #define SBA_CMD_ZERO_BUFFER				0x4
72 #define SBA_CMD_ZERO_ALL_BUFFERS			0x8
73 #define SBA_CMD_LOAD_BUFFER				0x9
74 #define SBA_CMD_XOR					0xa
75 #define SBA_CMD_GALOIS_XOR				0xb
76 #define SBA_CMD_WRITE_BUFFER				0xc
77 #define SBA_CMD_GALOIS					0xe
78 
79 #define SBA_MAX_REQ_PER_MBOX_CHANNEL			8192
80 #define SBA_MAX_MSG_SEND_PER_MBOX_CHANNEL		8
81 
82 /* Driver helper macros */
83 #define to_sba_request(tx)		\
84 	container_of(tx, struct sba_request, tx)
85 #define to_sba_device(dchan)		\
86 	container_of(dchan, struct sba_device, dma_chan)
87 
88 /* ===== Driver data structures ===== */
89 
90 enum sba_request_flags {
91 	SBA_REQUEST_STATE_FREE		= 0x001,
92 	SBA_REQUEST_STATE_ALLOCED	= 0x002,
93 	SBA_REQUEST_STATE_PENDING	= 0x004,
94 	SBA_REQUEST_STATE_ACTIVE	= 0x008,
95 	SBA_REQUEST_STATE_ABORTED	= 0x010,
96 	SBA_REQUEST_STATE_MASK		= 0x0ff,
97 	SBA_REQUEST_FENCE		= 0x100,
98 };
99 
100 struct sba_request {
101 	/* Global state */
102 	struct list_head node;
103 	struct sba_device *sba;
104 	u32 flags;
105 	/* Chained requests management */
106 	struct sba_request *first;
107 	struct list_head next;
108 	atomic_t next_pending_count;
109 	/* BRCM message data */
110 	struct brcm_message msg;
111 	struct dma_async_tx_descriptor tx;
112 	/* SBA commands */
113 	struct brcm_sba_command cmds[];
114 };
115 
116 enum sba_version {
117 	SBA_VER_1 = 0,
118 	SBA_VER_2
119 };
120 
121 struct sba_device {
122 	/* Underlying device */
123 	struct device *dev;
124 	/* DT configuration parameters */
125 	enum sba_version ver;
126 	/* Derived configuration parameters */
127 	u32 max_req;
128 	u32 hw_buf_size;
129 	u32 hw_resp_size;
130 	u32 max_pq_coefs;
131 	u32 max_pq_srcs;
132 	u32 max_cmd_per_req;
133 	u32 max_xor_srcs;
134 	u32 max_resp_pool_size;
135 	u32 max_cmds_pool_size;
136 	/* Maibox client and Mailbox channels */
137 	struct mbox_client client;
138 	struct mbox_chan *mchan;
139 	struct device *mbox_dev;
140 	/* DMA device and DMA channel */
141 	struct dma_device dma_dev;
142 	struct dma_chan dma_chan;
143 	/* DMA channel resources */
144 	void *resp_base;
145 	dma_addr_t resp_dma_base;
146 	void *cmds_base;
147 	dma_addr_t cmds_dma_base;
148 	spinlock_t reqs_lock;
149 	bool reqs_fence;
150 	struct list_head reqs_alloc_list;
151 	struct list_head reqs_pending_list;
152 	struct list_head reqs_active_list;
153 	struct list_head reqs_aborted_list;
154 	struct list_head reqs_free_list;
155 	/* DebugFS directory entries */
156 	struct dentry *root;
157 };
158 
159 /* ====== Command helper routines ===== */
160 
161 static inline u64 __pure sba_cmd_enc(u64 cmd, u32 val, u32 shift, u32 mask)
162 {
163 	cmd &= ~((u64)mask << shift);
164 	cmd |= ((u64)(val & mask) << shift);
165 	return cmd;
166 }
167 
168 static inline u32 __pure sba_cmd_load_c_mdata(u32 b0)
169 {
170 	return b0 & SBA_C_MDATA_BNUMx_MASK;
171 }
172 
173 static inline u32 __pure sba_cmd_write_c_mdata(u32 b0)
174 {
175 	return b0 & SBA_C_MDATA_BNUMx_MASK;
176 }
177 
178 static inline u32 __pure sba_cmd_xor_c_mdata(u32 b1, u32 b0)
179 {
180 	return (b0 & SBA_C_MDATA_BNUMx_MASK) |
181 	       ((b1 & SBA_C_MDATA_BNUMx_MASK) << SBA_C_MDATA_BNUMx_SHIFT(1));
182 }
183 
184 static inline u32 __pure sba_cmd_pq_c_mdata(u32 d, u32 b1, u32 b0)
185 {
186 	return (b0 & SBA_C_MDATA_BNUMx_MASK) |
187 	       ((b1 & SBA_C_MDATA_BNUMx_MASK) << SBA_C_MDATA_BNUMx_SHIFT(1)) |
188 	       ((d & SBA_C_MDATA_DNUM_MASK) << SBA_C_MDATA_DNUM_SHIFT);
189 }
190 
191 /* ====== General helper routines ===== */
192 
193 static struct sba_request *sba_alloc_request(struct sba_device *sba)
194 {
195 	bool found = false;
196 	unsigned long flags;
197 	struct sba_request *req = NULL;
198 
199 	spin_lock_irqsave(&sba->reqs_lock, flags);
200 	list_for_each_entry(req, &sba->reqs_free_list, node) {
201 		if (async_tx_test_ack(&req->tx)) {
202 			list_move_tail(&req->node, &sba->reqs_alloc_list);
203 			found = true;
204 			break;
205 		}
206 	}
207 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
208 
209 	if (!found) {
210 		/*
211 		 * We have no more free requests so, we peek
212 		 * mailbox channels hoping few active requests
213 		 * would have completed which will create more
214 		 * room for new requests.
215 		 */
216 		mbox_client_peek_data(sba->mchan);
217 		return NULL;
218 	}
219 
220 	req->flags = SBA_REQUEST_STATE_ALLOCED;
221 	req->first = req;
222 	INIT_LIST_HEAD(&req->next);
223 	atomic_set(&req->next_pending_count, 1);
224 
225 	dma_async_tx_descriptor_init(&req->tx, &sba->dma_chan);
226 	async_tx_ack(&req->tx);
227 
228 	return req;
229 }
230 
231 /* Note: Must be called with sba->reqs_lock held */
232 static void _sba_pending_request(struct sba_device *sba,
233 				 struct sba_request *req)
234 {
235 	lockdep_assert_held(&sba->reqs_lock);
236 	req->flags &= ~SBA_REQUEST_STATE_MASK;
237 	req->flags |= SBA_REQUEST_STATE_PENDING;
238 	list_move_tail(&req->node, &sba->reqs_pending_list);
239 	if (list_empty(&sba->reqs_active_list))
240 		sba->reqs_fence = false;
241 }
242 
243 /* Note: Must be called with sba->reqs_lock held */
244 static bool _sba_active_request(struct sba_device *sba,
245 				struct sba_request *req)
246 {
247 	lockdep_assert_held(&sba->reqs_lock);
248 	if (list_empty(&sba->reqs_active_list))
249 		sba->reqs_fence = false;
250 	if (sba->reqs_fence)
251 		return false;
252 	req->flags &= ~SBA_REQUEST_STATE_MASK;
253 	req->flags |= SBA_REQUEST_STATE_ACTIVE;
254 	list_move_tail(&req->node, &sba->reqs_active_list);
255 	if (req->flags & SBA_REQUEST_FENCE)
256 		sba->reqs_fence = true;
257 	return true;
258 }
259 
260 /* Note: Must be called with sba->reqs_lock held */
261 static void _sba_abort_request(struct sba_device *sba,
262 			       struct sba_request *req)
263 {
264 	lockdep_assert_held(&sba->reqs_lock);
265 	req->flags &= ~SBA_REQUEST_STATE_MASK;
266 	req->flags |= SBA_REQUEST_STATE_ABORTED;
267 	list_move_tail(&req->node, &sba->reqs_aborted_list);
268 	if (list_empty(&sba->reqs_active_list))
269 		sba->reqs_fence = false;
270 }
271 
272 /* Note: Must be called with sba->reqs_lock held */
273 static void _sba_free_request(struct sba_device *sba,
274 			      struct sba_request *req)
275 {
276 	lockdep_assert_held(&sba->reqs_lock);
277 	req->flags &= ~SBA_REQUEST_STATE_MASK;
278 	req->flags |= SBA_REQUEST_STATE_FREE;
279 	list_move_tail(&req->node, &sba->reqs_free_list);
280 	if (list_empty(&sba->reqs_active_list))
281 		sba->reqs_fence = false;
282 }
283 
284 static void sba_free_chained_requests(struct sba_request *req)
285 {
286 	unsigned long flags;
287 	struct sba_request *nreq;
288 	struct sba_device *sba = req->sba;
289 
290 	spin_lock_irqsave(&sba->reqs_lock, flags);
291 
292 	_sba_free_request(sba, req);
293 	list_for_each_entry(nreq, &req->next, next)
294 		_sba_free_request(sba, nreq);
295 
296 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
297 }
298 
299 static void sba_chain_request(struct sba_request *first,
300 			      struct sba_request *req)
301 {
302 	unsigned long flags;
303 	struct sba_device *sba = req->sba;
304 
305 	spin_lock_irqsave(&sba->reqs_lock, flags);
306 
307 	list_add_tail(&req->next, &first->next);
308 	req->first = first;
309 	atomic_inc(&first->next_pending_count);
310 
311 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
312 }
313 
314 static void sba_cleanup_nonpending_requests(struct sba_device *sba)
315 {
316 	unsigned long flags;
317 	struct sba_request *req, *req1;
318 
319 	spin_lock_irqsave(&sba->reqs_lock, flags);
320 
321 	/* Freeup all alloced request */
322 	list_for_each_entry_safe(req, req1, &sba->reqs_alloc_list, node)
323 		_sba_free_request(sba, req);
324 
325 	/* Set all active requests as aborted */
326 	list_for_each_entry_safe(req, req1, &sba->reqs_active_list, node)
327 		_sba_abort_request(sba, req);
328 
329 	/*
330 	 * Note: We expect that aborted request will be eventually
331 	 * freed by sba_receive_message()
332 	 */
333 
334 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
335 }
336 
337 static void sba_cleanup_pending_requests(struct sba_device *sba)
338 {
339 	unsigned long flags;
340 	struct sba_request *req, *req1;
341 
342 	spin_lock_irqsave(&sba->reqs_lock, flags);
343 
344 	/* Freeup all pending request */
345 	list_for_each_entry_safe(req, req1, &sba->reqs_pending_list, node)
346 		_sba_free_request(sba, req);
347 
348 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
349 }
350 
351 static int sba_send_mbox_request(struct sba_device *sba,
352 				 struct sba_request *req)
353 {
354 	int ret = 0;
355 
356 	/* Send message for the request */
357 	req->msg.error = 0;
358 	ret = mbox_send_message(sba->mchan, &req->msg);
359 	if (ret < 0) {
360 		dev_err(sba->dev, "send message failed with error %d", ret);
361 		return ret;
362 	}
363 
364 	/* Check error returned by mailbox controller */
365 	ret = req->msg.error;
366 	if (ret < 0) {
367 		dev_err(sba->dev, "message error %d", ret);
368 	}
369 
370 	/* Signal txdone for mailbox channel */
371 	mbox_client_txdone(sba->mchan, ret);
372 
373 	return ret;
374 }
375 
376 /* Note: Must be called with sba->reqs_lock held */
377 static void _sba_process_pending_requests(struct sba_device *sba)
378 {
379 	int ret;
380 	u32 count;
381 	struct sba_request *req;
382 
383 	/* Process few pending requests */
384 	count = SBA_MAX_MSG_SEND_PER_MBOX_CHANNEL;
385 	while (!list_empty(&sba->reqs_pending_list) && count) {
386 		/* Get the first pending request */
387 		req = list_first_entry(&sba->reqs_pending_list,
388 				       struct sba_request, node);
389 
390 		/* Try to make request active */
391 		if (!_sba_active_request(sba, req))
392 			break;
393 
394 		/* Send request to mailbox channel */
395 		ret = sba_send_mbox_request(sba, req);
396 		if (ret < 0) {
397 			_sba_pending_request(sba, req);
398 			break;
399 		}
400 
401 		count--;
402 	}
403 }
404 
405 static void sba_process_received_request(struct sba_device *sba,
406 					 struct sba_request *req)
407 {
408 	unsigned long flags;
409 	struct dma_async_tx_descriptor *tx;
410 	struct sba_request *nreq, *first = req->first;
411 
412 	/* Process only after all chained requests are received */
413 	if (!atomic_dec_return(&first->next_pending_count)) {
414 		tx = &first->tx;
415 
416 		WARN_ON(tx->cookie < 0);
417 		if (tx->cookie > 0) {
418 			spin_lock_irqsave(&sba->reqs_lock, flags);
419 			dma_cookie_complete(tx);
420 			spin_unlock_irqrestore(&sba->reqs_lock, flags);
421 			dmaengine_desc_get_callback_invoke(tx, NULL);
422 			dma_descriptor_unmap(tx);
423 			tx->callback = NULL;
424 			tx->callback_result = NULL;
425 		}
426 
427 		dma_run_dependencies(tx);
428 
429 		spin_lock_irqsave(&sba->reqs_lock, flags);
430 
431 		/* Free all requests chained to first request */
432 		list_for_each_entry(nreq, &first->next, next)
433 			_sba_free_request(sba, nreq);
434 		INIT_LIST_HEAD(&first->next);
435 
436 		/* Free the first request */
437 		_sba_free_request(sba, first);
438 
439 		/* Process pending requests */
440 		_sba_process_pending_requests(sba);
441 
442 		spin_unlock_irqrestore(&sba->reqs_lock, flags);
443 	}
444 }
445 
446 static void sba_write_stats_in_seqfile(struct sba_device *sba,
447 				       struct seq_file *file)
448 {
449 	unsigned long flags;
450 	struct sba_request *req;
451 	u32 free_count = 0, alloced_count = 0;
452 	u32 pending_count = 0, active_count = 0, aborted_count = 0;
453 
454 	spin_lock_irqsave(&sba->reqs_lock, flags);
455 
456 	list_for_each_entry(req, &sba->reqs_free_list, node)
457 		if (async_tx_test_ack(&req->tx))
458 			free_count++;
459 
460 	list_for_each_entry(req, &sba->reqs_alloc_list, node)
461 		alloced_count++;
462 
463 	list_for_each_entry(req, &sba->reqs_pending_list, node)
464 		pending_count++;
465 
466 	list_for_each_entry(req, &sba->reqs_active_list, node)
467 		active_count++;
468 
469 	list_for_each_entry(req, &sba->reqs_aborted_list, node)
470 		aborted_count++;
471 
472 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
473 
474 	seq_printf(file, "maximum requests   = %d\n", sba->max_req);
475 	seq_printf(file, "free requests      = %d\n", free_count);
476 	seq_printf(file, "alloced requests   = %d\n", alloced_count);
477 	seq_printf(file, "pending requests   = %d\n", pending_count);
478 	seq_printf(file, "active requests    = %d\n", active_count);
479 	seq_printf(file, "aborted requests   = %d\n", aborted_count);
480 }
481 
482 /* ====== DMAENGINE callbacks ===== */
483 
484 static void sba_free_chan_resources(struct dma_chan *dchan)
485 {
486 	/*
487 	 * Channel resources are pre-alloced so we just free-up
488 	 * whatever we can so that we can re-use pre-alloced
489 	 * channel resources next time.
490 	 */
491 	sba_cleanup_nonpending_requests(to_sba_device(dchan));
492 }
493 
494 static int sba_device_terminate_all(struct dma_chan *dchan)
495 {
496 	/* Cleanup all pending requests */
497 	sba_cleanup_pending_requests(to_sba_device(dchan));
498 
499 	return 0;
500 }
501 
502 static void sba_issue_pending(struct dma_chan *dchan)
503 {
504 	unsigned long flags;
505 	struct sba_device *sba = to_sba_device(dchan);
506 
507 	/* Process pending requests */
508 	spin_lock_irqsave(&sba->reqs_lock, flags);
509 	_sba_process_pending_requests(sba);
510 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
511 }
512 
513 static dma_cookie_t sba_tx_submit(struct dma_async_tx_descriptor *tx)
514 {
515 	unsigned long flags;
516 	dma_cookie_t cookie;
517 	struct sba_device *sba;
518 	struct sba_request *req, *nreq;
519 
520 	if (unlikely(!tx))
521 		return -EINVAL;
522 
523 	sba = to_sba_device(tx->chan);
524 	req = to_sba_request(tx);
525 
526 	/* Assign cookie and mark all chained requests pending */
527 	spin_lock_irqsave(&sba->reqs_lock, flags);
528 	cookie = dma_cookie_assign(tx);
529 	_sba_pending_request(sba, req);
530 	list_for_each_entry(nreq, &req->next, next)
531 		_sba_pending_request(sba, nreq);
532 	spin_unlock_irqrestore(&sba->reqs_lock, flags);
533 
534 	return cookie;
535 }
536 
537 static enum dma_status sba_tx_status(struct dma_chan *dchan,
538 				     dma_cookie_t cookie,
539 				     struct dma_tx_state *txstate)
540 {
541 	enum dma_status ret;
542 	struct sba_device *sba = to_sba_device(dchan);
543 
544 	ret = dma_cookie_status(dchan, cookie, txstate);
545 	if (ret == DMA_COMPLETE)
546 		return ret;
547 
548 	mbox_client_peek_data(sba->mchan);
549 
550 	return dma_cookie_status(dchan, cookie, txstate);
551 }
552 
553 static void sba_fillup_interrupt_msg(struct sba_request *req,
554 				     struct brcm_sba_command *cmds,
555 				     struct brcm_message *msg)
556 {
557 	u64 cmd;
558 	u32 c_mdata;
559 	dma_addr_t resp_dma = req->tx.phys;
560 	struct brcm_sba_command *cmdsp = cmds;
561 
562 	/* Type-B command to load dummy data into buf0 */
563 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
564 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
565 	cmd = sba_cmd_enc(cmd, req->sba->hw_resp_size,
566 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
567 	c_mdata = sba_cmd_load_c_mdata(0);
568 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
569 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
570 	cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
571 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
572 	cmdsp->cmd = cmd;
573 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
574 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
575 	cmdsp->data = resp_dma;
576 	cmdsp->data_len = req->sba->hw_resp_size;
577 	cmdsp++;
578 
579 	/* Type-A command to write buf0 to dummy location */
580 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
581 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
582 	cmd = sba_cmd_enc(cmd, req->sba->hw_resp_size,
583 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
584 	cmd = sba_cmd_enc(cmd, 0x1,
585 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
586 	c_mdata = sba_cmd_write_c_mdata(0);
587 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
588 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
589 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
590 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
591 	cmdsp->cmd = cmd;
592 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
593 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
594 	if (req->sba->hw_resp_size) {
595 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
596 		cmdsp->resp = resp_dma;
597 		cmdsp->resp_len = req->sba->hw_resp_size;
598 	}
599 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
600 	cmdsp->data = resp_dma;
601 	cmdsp->data_len = req->sba->hw_resp_size;
602 	cmdsp++;
603 
604 	/* Fillup brcm_message */
605 	msg->type = BRCM_MESSAGE_SBA;
606 	msg->sba.cmds = cmds;
607 	msg->sba.cmds_count = cmdsp - cmds;
608 	msg->ctx = req;
609 	msg->error = 0;
610 }
611 
612 static struct dma_async_tx_descriptor *
613 sba_prep_dma_interrupt(struct dma_chan *dchan, unsigned long flags)
614 {
615 	struct sba_request *req = NULL;
616 	struct sba_device *sba = to_sba_device(dchan);
617 
618 	/* Alloc new request */
619 	req = sba_alloc_request(sba);
620 	if (!req)
621 		return NULL;
622 
623 	/*
624 	 * Force fence so that no requests are submitted
625 	 * until DMA callback for this request is invoked.
626 	 */
627 	req->flags |= SBA_REQUEST_FENCE;
628 
629 	/* Fillup request message */
630 	sba_fillup_interrupt_msg(req, req->cmds, &req->msg);
631 
632 	/* Init async_tx descriptor */
633 	req->tx.flags = flags;
634 	req->tx.cookie = -EBUSY;
635 
636 	return &req->tx;
637 }
638 
639 static void sba_fillup_memcpy_msg(struct sba_request *req,
640 				  struct brcm_sba_command *cmds,
641 				  struct brcm_message *msg,
642 				  dma_addr_t msg_offset, size_t msg_len,
643 				  dma_addr_t dst, dma_addr_t src)
644 {
645 	u64 cmd;
646 	u32 c_mdata;
647 	dma_addr_t resp_dma = req->tx.phys;
648 	struct brcm_sba_command *cmdsp = cmds;
649 
650 	/* Type-B command to load data into buf0 */
651 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
652 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
653 	cmd = sba_cmd_enc(cmd, msg_len,
654 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
655 	c_mdata = sba_cmd_load_c_mdata(0);
656 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
657 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
658 	cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
659 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
660 	cmdsp->cmd = cmd;
661 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
662 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
663 	cmdsp->data = src + msg_offset;
664 	cmdsp->data_len = msg_len;
665 	cmdsp++;
666 
667 	/* Type-A command to write buf0 */
668 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
669 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
670 	cmd = sba_cmd_enc(cmd, msg_len,
671 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
672 	cmd = sba_cmd_enc(cmd, 0x1,
673 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
674 	c_mdata = sba_cmd_write_c_mdata(0);
675 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
676 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
677 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
678 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
679 	cmdsp->cmd = cmd;
680 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
681 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
682 	if (req->sba->hw_resp_size) {
683 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
684 		cmdsp->resp = resp_dma;
685 		cmdsp->resp_len = req->sba->hw_resp_size;
686 	}
687 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
688 	cmdsp->data = dst + msg_offset;
689 	cmdsp->data_len = msg_len;
690 	cmdsp++;
691 
692 	/* Fillup brcm_message */
693 	msg->type = BRCM_MESSAGE_SBA;
694 	msg->sba.cmds = cmds;
695 	msg->sba.cmds_count = cmdsp - cmds;
696 	msg->ctx = req;
697 	msg->error = 0;
698 }
699 
700 static struct sba_request *
701 sba_prep_dma_memcpy_req(struct sba_device *sba,
702 			dma_addr_t off, dma_addr_t dst, dma_addr_t src,
703 			size_t len, unsigned long flags)
704 {
705 	struct sba_request *req = NULL;
706 
707 	/* Alloc new request */
708 	req = sba_alloc_request(sba);
709 	if (!req)
710 		return NULL;
711 	if (flags & DMA_PREP_FENCE)
712 		req->flags |= SBA_REQUEST_FENCE;
713 
714 	/* Fillup request message */
715 	sba_fillup_memcpy_msg(req, req->cmds, &req->msg,
716 			      off, len, dst, src);
717 
718 	/* Init async_tx descriptor */
719 	req->tx.flags = flags;
720 	req->tx.cookie = -EBUSY;
721 
722 	return req;
723 }
724 
725 static struct dma_async_tx_descriptor *
726 sba_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst, dma_addr_t src,
727 		    size_t len, unsigned long flags)
728 {
729 	size_t req_len;
730 	dma_addr_t off = 0;
731 	struct sba_device *sba = to_sba_device(dchan);
732 	struct sba_request *first = NULL, *req;
733 
734 	/* Create chained requests where each request is upto hw_buf_size */
735 	while (len) {
736 		req_len = (len < sba->hw_buf_size) ? len : sba->hw_buf_size;
737 
738 		req = sba_prep_dma_memcpy_req(sba, off, dst, src,
739 					      req_len, flags);
740 		if (!req) {
741 			if (first)
742 				sba_free_chained_requests(first);
743 			return NULL;
744 		}
745 
746 		if (first)
747 			sba_chain_request(first, req);
748 		else
749 			first = req;
750 
751 		off += req_len;
752 		len -= req_len;
753 	}
754 
755 	return (first) ? &first->tx : NULL;
756 }
757 
758 static void sba_fillup_xor_msg(struct sba_request *req,
759 				struct brcm_sba_command *cmds,
760 				struct brcm_message *msg,
761 				dma_addr_t msg_offset, size_t msg_len,
762 				dma_addr_t dst, dma_addr_t *src, u32 src_cnt)
763 {
764 	u64 cmd;
765 	u32 c_mdata;
766 	unsigned int i;
767 	dma_addr_t resp_dma = req->tx.phys;
768 	struct brcm_sba_command *cmdsp = cmds;
769 
770 	/* Type-B command to load data into buf0 */
771 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
772 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
773 	cmd = sba_cmd_enc(cmd, msg_len,
774 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
775 	c_mdata = sba_cmd_load_c_mdata(0);
776 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
777 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
778 	cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
779 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
780 	cmdsp->cmd = cmd;
781 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
782 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
783 	cmdsp->data = src[0] + msg_offset;
784 	cmdsp->data_len = msg_len;
785 	cmdsp++;
786 
787 	/* Type-B commands to xor data with buf0 and put it back in buf0 */
788 	for (i = 1; i < src_cnt; i++) {
789 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
790 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
791 		cmd = sba_cmd_enc(cmd, msg_len,
792 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
793 		c_mdata = sba_cmd_xor_c_mdata(0, 0);
794 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
795 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
796 		cmd = sba_cmd_enc(cmd, SBA_CMD_XOR,
797 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
798 		cmdsp->cmd = cmd;
799 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
800 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
801 		cmdsp->data = src[i] + msg_offset;
802 		cmdsp->data_len = msg_len;
803 		cmdsp++;
804 	}
805 
806 	/* Type-A command to write buf0 */
807 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
808 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
809 	cmd = sba_cmd_enc(cmd, msg_len,
810 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
811 	cmd = sba_cmd_enc(cmd, 0x1,
812 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
813 	c_mdata = sba_cmd_write_c_mdata(0);
814 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
815 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
816 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
817 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
818 	cmdsp->cmd = cmd;
819 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
820 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
821 	if (req->sba->hw_resp_size) {
822 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
823 		cmdsp->resp = resp_dma;
824 		cmdsp->resp_len = req->sba->hw_resp_size;
825 	}
826 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
827 	cmdsp->data = dst + msg_offset;
828 	cmdsp->data_len = msg_len;
829 	cmdsp++;
830 
831 	/* Fillup brcm_message */
832 	msg->type = BRCM_MESSAGE_SBA;
833 	msg->sba.cmds = cmds;
834 	msg->sba.cmds_count = cmdsp - cmds;
835 	msg->ctx = req;
836 	msg->error = 0;
837 }
838 
839 static struct sba_request *
840 sba_prep_dma_xor_req(struct sba_device *sba,
841 		     dma_addr_t off, dma_addr_t dst, dma_addr_t *src,
842 		     u32 src_cnt, size_t len, unsigned long flags)
843 {
844 	struct sba_request *req = NULL;
845 
846 	/* Alloc new request */
847 	req = sba_alloc_request(sba);
848 	if (!req)
849 		return NULL;
850 	if (flags & DMA_PREP_FENCE)
851 		req->flags |= SBA_REQUEST_FENCE;
852 
853 	/* Fillup request message */
854 	sba_fillup_xor_msg(req, req->cmds, &req->msg,
855 			   off, len, dst, src, src_cnt);
856 
857 	/* Init async_tx descriptor */
858 	req->tx.flags = flags;
859 	req->tx.cookie = -EBUSY;
860 
861 	return req;
862 }
863 
864 static struct dma_async_tx_descriptor *
865 sba_prep_dma_xor(struct dma_chan *dchan, dma_addr_t dst, dma_addr_t *src,
866 		 u32 src_cnt, size_t len, unsigned long flags)
867 {
868 	size_t req_len;
869 	dma_addr_t off = 0;
870 	struct sba_device *sba = to_sba_device(dchan);
871 	struct sba_request *first = NULL, *req;
872 
873 	/* Sanity checks */
874 	if (unlikely(src_cnt > sba->max_xor_srcs))
875 		return NULL;
876 
877 	/* Create chained requests where each request is upto hw_buf_size */
878 	while (len) {
879 		req_len = (len < sba->hw_buf_size) ? len : sba->hw_buf_size;
880 
881 		req = sba_prep_dma_xor_req(sba, off, dst, src, src_cnt,
882 					   req_len, flags);
883 		if (!req) {
884 			if (first)
885 				sba_free_chained_requests(first);
886 			return NULL;
887 		}
888 
889 		if (first)
890 			sba_chain_request(first, req);
891 		else
892 			first = req;
893 
894 		off += req_len;
895 		len -= req_len;
896 	}
897 
898 	return (first) ? &first->tx : NULL;
899 }
900 
901 static void sba_fillup_pq_msg(struct sba_request *req,
902 				bool pq_continue,
903 				struct brcm_sba_command *cmds,
904 				struct brcm_message *msg,
905 				dma_addr_t msg_offset, size_t msg_len,
906 				dma_addr_t *dst_p, dma_addr_t *dst_q,
907 				const u8 *scf, dma_addr_t *src, u32 src_cnt)
908 {
909 	u64 cmd;
910 	u32 c_mdata;
911 	unsigned int i;
912 	dma_addr_t resp_dma = req->tx.phys;
913 	struct brcm_sba_command *cmdsp = cmds;
914 
915 	if (pq_continue) {
916 		/* Type-B command to load old P into buf0 */
917 		if (dst_p) {
918 			cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
919 				SBA_TYPE_SHIFT, SBA_TYPE_MASK);
920 			cmd = sba_cmd_enc(cmd, msg_len,
921 				SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
922 			c_mdata = sba_cmd_load_c_mdata(0);
923 			cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
924 				SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
925 			cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
926 				SBA_CMD_SHIFT, SBA_CMD_MASK);
927 			cmdsp->cmd = cmd;
928 			*cmdsp->cmd_dma = cpu_to_le64(cmd);
929 			cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
930 			cmdsp->data = *dst_p + msg_offset;
931 			cmdsp->data_len = msg_len;
932 			cmdsp++;
933 		}
934 
935 		/* Type-B command to load old Q into buf1 */
936 		if (dst_q) {
937 			cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
938 				SBA_TYPE_SHIFT, SBA_TYPE_MASK);
939 			cmd = sba_cmd_enc(cmd, msg_len,
940 				SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
941 			c_mdata = sba_cmd_load_c_mdata(1);
942 			cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
943 				SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
944 			cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
945 				SBA_CMD_SHIFT, SBA_CMD_MASK);
946 			cmdsp->cmd = cmd;
947 			*cmdsp->cmd_dma = cpu_to_le64(cmd);
948 			cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
949 			cmdsp->data = *dst_q + msg_offset;
950 			cmdsp->data_len = msg_len;
951 			cmdsp++;
952 		}
953 	} else {
954 		/* Type-A command to zero all buffers */
955 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
956 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
957 		cmd = sba_cmd_enc(cmd, msg_len,
958 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
959 		cmd = sba_cmd_enc(cmd, SBA_CMD_ZERO_ALL_BUFFERS,
960 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
961 		cmdsp->cmd = cmd;
962 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
963 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
964 		cmdsp++;
965 	}
966 
967 	/* Type-B commands for generate P onto buf0 and Q onto buf1 */
968 	for (i = 0; i < src_cnt; i++) {
969 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
970 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
971 		cmd = sba_cmd_enc(cmd, msg_len,
972 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
973 		c_mdata = sba_cmd_pq_c_mdata(raid6_gflog[scf[i]], 1, 0);
974 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
975 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
976 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_MS(c_mdata),
977 				  SBA_C_MDATA_MS_SHIFT, SBA_C_MDATA_MS_MASK);
978 		cmd = sba_cmd_enc(cmd, SBA_CMD_GALOIS_XOR,
979 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
980 		cmdsp->cmd = cmd;
981 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
982 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
983 		cmdsp->data = src[i] + msg_offset;
984 		cmdsp->data_len = msg_len;
985 		cmdsp++;
986 	}
987 
988 	/* Type-A command to write buf0 */
989 	if (dst_p) {
990 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
991 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
992 		cmd = sba_cmd_enc(cmd, msg_len,
993 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
994 		cmd = sba_cmd_enc(cmd, 0x1,
995 				  SBA_RESP_SHIFT, SBA_RESP_MASK);
996 		c_mdata = sba_cmd_write_c_mdata(0);
997 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
998 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
999 		cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1000 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1001 		cmdsp->cmd = cmd;
1002 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1003 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1004 		if (req->sba->hw_resp_size) {
1005 			cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1006 			cmdsp->resp = resp_dma;
1007 			cmdsp->resp_len = req->sba->hw_resp_size;
1008 		}
1009 		cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1010 		cmdsp->data = *dst_p + msg_offset;
1011 		cmdsp->data_len = msg_len;
1012 		cmdsp++;
1013 	}
1014 
1015 	/* Type-A command to write buf1 */
1016 	if (dst_q) {
1017 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1018 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1019 		cmd = sba_cmd_enc(cmd, msg_len,
1020 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1021 		cmd = sba_cmd_enc(cmd, 0x1,
1022 				  SBA_RESP_SHIFT, SBA_RESP_MASK);
1023 		c_mdata = sba_cmd_write_c_mdata(1);
1024 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1025 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1026 		cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1027 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1028 		cmdsp->cmd = cmd;
1029 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1030 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1031 		if (req->sba->hw_resp_size) {
1032 			cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1033 			cmdsp->resp = resp_dma;
1034 			cmdsp->resp_len = req->sba->hw_resp_size;
1035 		}
1036 		cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1037 		cmdsp->data = *dst_q + msg_offset;
1038 		cmdsp->data_len = msg_len;
1039 		cmdsp++;
1040 	}
1041 
1042 	/* Fillup brcm_message */
1043 	msg->type = BRCM_MESSAGE_SBA;
1044 	msg->sba.cmds = cmds;
1045 	msg->sba.cmds_count = cmdsp - cmds;
1046 	msg->ctx = req;
1047 	msg->error = 0;
1048 }
1049 
1050 static struct sba_request *
1051 sba_prep_dma_pq_req(struct sba_device *sba, dma_addr_t off,
1052 		    dma_addr_t *dst_p, dma_addr_t *dst_q, dma_addr_t *src,
1053 		    u32 src_cnt, const u8 *scf, size_t len, unsigned long flags)
1054 {
1055 	struct sba_request *req = NULL;
1056 
1057 	/* Alloc new request */
1058 	req = sba_alloc_request(sba);
1059 	if (!req)
1060 		return NULL;
1061 	if (flags & DMA_PREP_FENCE)
1062 		req->flags |= SBA_REQUEST_FENCE;
1063 
1064 	/* Fillup request messages */
1065 	sba_fillup_pq_msg(req, dmaf_continue(flags),
1066 			  req->cmds, &req->msg,
1067 			  off, len, dst_p, dst_q, scf, src, src_cnt);
1068 
1069 	/* Init async_tx descriptor */
1070 	req->tx.flags = flags;
1071 	req->tx.cookie = -EBUSY;
1072 
1073 	return req;
1074 }
1075 
1076 static void sba_fillup_pq_single_msg(struct sba_request *req,
1077 				bool pq_continue,
1078 				struct brcm_sba_command *cmds,
1079 				struct brcm_message *msg,
1080 				dma_addr_t msg_offset, size_t msg_len,
1081 				dma_addr_t *dst_p, dma_addr_t *dst_q,
1082 				dma_addr_t src, u8 scf)
1083 {
1084 	u64 cmd;
1085 	u32 c_mdata;
1086 	u8 pos, dpos = raid6_gflog[scf];
1087 	dma_addr_t resp_dma = req->tx.phys;
1088 	struct brcm_sba_command *cmdsp = cmds;
1089 
1090 	if (!dst_p)
1091 		goto skip_p;
1092 
1093 	if (pq_continue) {
1094 		/* Type-B command to load old P into buf0 */
1095 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1096 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1097 		cmd = sba_cmd_enc(cmd, msg_len,
1098 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1099 		c_mdata = sba_cmd_load_c_mdata(0);
1100 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1101 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1102 		cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
1103 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1104 		cmdsp->cmd = cmd;
1105 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1106 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1107 		cmdsp->data = *dst_p + msg_offset;
1108 		cmdsp->data_len = msg_len;
1109 		cmdsp++;
1110 
1111 		/*
1112 		 * Type-B commands to xor data with buf0 and put it
1113 		 * back in buf0
1114 		 */
1115 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1116 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1117 		cmd = sba_cmd_enc(cmd, msg_len,
1118 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1119 		c_mdata = sba_cmd_xor_c_mdata(0, 0);
1120 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1121 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1122 		cmd = sba_cmd_enc(cmd, SBA_CMD_XOR,
1123 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1124 		cmdsp->cmd = cmd;
1125 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1126 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1127 		cmdsp->data = src + msg_offset;
1128 		cmdsp->data_len = msg_len;
1129 		cmdsp++;
1130 	} else {
1131 		/* Type-B command to load old P into buf0 */
1132 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1133 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1134 		cmd = sba_cmd_enc(cmd, msg_len,
1135 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1136 		c_mdata = sba_cmd_load_c_mdata(0);
1137 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1138 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1139 		cmd = sba_cmd_enc(cmd, SBA_CMD_LOAD_BUFFER,
1140 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1141 		cmdsp->cmd = cmd;
1142 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1143 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1144 		cmdsp->data = src + msg_offset;
1145 		cmdsp->data_len = msg_len;
1146 		cmdsp++;
1147 	}
1148 
1149 	/* Type-A command to write buf0 */
1150 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1151 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1152 	cmd = sba_cmd_enc(cmd, msg_len,
1153 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1154 	cmd = sba_cmd_enc(cmd, 0x1,
1155 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
1156 	c_mdata = sba_cmd_write_c_mdata(0);
1157 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1158 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1159 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1160 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1161 	cmdsp->cmd = cmd;
1162 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1163 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1164 	if (req->sba->hw_resp_size) {
1165 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1166 		cmdsp->resp = resp_dma;
1167 		cmdsp->resp_len = req->sba->hw_resp_size;
1168 	}
1169 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1170 	cmdsp->data = *dst_p + msg_offset;
1171 	cmdsp->data_len = msg_len;
1172 	cmdsp++;
1173 
1174 skip_p:
1175 	if (!dst_q)
1176 		goto skip_q;
1177 
1178 	/* Type-A command to zero all buffers */
1179 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1180 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1181 	cmd = sba_cmd_enc(cmd, msg_len,
1182 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1183 	cmd = sba_cmd_enc(cmd, SBA_CMD_ZERO_ALL_BUFFERS,
1184 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1185 	cmdsp->cmd = cmd;
1186 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1187 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1188 	cmdsp++;
1189 
1190 	if (dpos == 255)
1191 		goto skip_q_computation;
1192 	pos = (dpos < req->sba->max_pq_coefs) ?
1193 		dpos : (req->sba->max_pq_coefs - 1);
1194 
1195 	/*
1196 	 * Type-B command to generate initial Q from data
1197 	 * and store output into buf0
1198 	 */
1199 	cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1200 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1201 	cmd = sba_cmd_enc(cmd, msg_len,
1202 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1203 	c_mdata = sba_cmd_pq_c_mdata(pos, 0, 0);
1204 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1205 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1206 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_MS(c_mdata),
1207 			  SBA_C_MDATA_MS_SHIFT, SBA_C_MDATA_MS_MASK);
1208 	cmd = sba_cmd_enc(cmd, SBA_CMD_GALOIS,
1209 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1210 	cmdsp->cmd = cmd;
1211 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1212 	cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1213 	cmdsp->data = src + msg_offset;
1214 	cmdsp->data_len = msg_len;
1215 	cmdsp++;
1216 
1217 	dpos -= pos;
1218 
1219 	/* Multiple Type-A command to generate final Q */
1220 	while (dpos) {
1221 		pos = (dpos < req->sba->max_pq_coefs) ?
1222 			dpos : (req->sba->max_pq_coefs - 1);
1223 
1224 		/*
1225 		 * Type-A command to generate Q with buf0 and
1226 		 * buf1 store result in buf0
1227 		 */
1228 		cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1229 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1230 		cmd = sba_cmd_enc(cmd, msg_len,
1231 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1232 		c_mdata = sba_cmd_pq_c_mdata(pos, 0, 1);
1233 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1234 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1235 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_MS(c_mdata),
1236 				  SBA_C_MDATA_MS_SHIFT, SBA_C_MDATA_MS_MASK);
1237 		cmd = sba_cmd_enc(cmd, SBA_CMD_GALOIS,
1238 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1239 		cmdsp->cmd = cmd;
1240 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1241 		cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1242 		cmdsp++;
1243 
1244 		dpos -= pos;
1245 	}
1246 
1247 skip_q_computation:
1248 	if (pq_continue) {
1249 		/*
1250 		 * Type-B command to XOR previous output with
1251 		 * buf0 and write it into buf0
1252 		 */
1253 		cmd = sba_cmd_enc(0x0, SBA_TYPE_B,
1254 				  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1255 		cmd = sba_cmd_enc(cmd, msg_len,
1256 				  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1257 		c_mdata = sba_cmd_xor_c_mdata(0, 0);
1258 		cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1259 				  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1260 		cmd = sba_cmd_enc(cmd, SBA_CMD_XOR,
1261 				  SBA_CMD_SHIFT, SBA_CMD_MASK);
1262 		cmdsp->cmd = cmd;
1263 		*cmdsp->cmd_dma = cpu_to_le64(cmd);
1264 		cmdsp->flags = BRCM_SBA_CMD_TYPE_B;
1265 		cmdsp->data = *dst_q + msg_offset;
1266 		cmdsp->data_len = msg_len;
1267 		cmdsp++;
1268 	}
1269 
1270 	/* Type-A command to write buf0 */
1271 	cmd = sba_cmd_enc(0x0, SBA_TYPE_A,
1272 			  SBA_TYPE_SHIFT, SBA_TYPE_MASK);
1273 	cmd = sba_cmd_enc(cmd, msg_len,
1274 			  SBA_USER_DEF_SHIFT, SBA_USER_DEF_MASK);
1275 	cmd = sba_cmd_enc(cmd, 0x1,
1276 			  SBA_RESP_SHIFT, SBA_RESP_MASK);
1277 	c_mdata = sba_cmd_write_c_mdata(0);
1278 	cmd = sba_cmd_enc(cmd, SBA_C_MDATA_LS(c_mdata),
1279 			  SBA_C_MDATA_SHIFT, SBA_C_MDATA_MASK);
1280 	cmd = sba_cmd_enc(cmd, SBA_CMD_WRITE_BUFFER,
1281 			  SBA_CMD_SHIFT, SBA_CMD_MASK);
1282 	cmdsp->cmd = cmd;
1283 	*cmdsp->cmd_dma = cpu_to_le64(cmd);
1284 	cmdsp->flags = BRCM_SBA_CMD_TYPE_A;
1285 	if (req->sba->hw_resp_size) {
1286 		cmdsp->flags |= BRCM_SBA_CMD_HAS_RESP;
1287 		cmdsp->resp = resp_dma;
1288 		cmdsp->resp_len = req->sba->hw_resp_size;
1289 	}
1290 	cmdsp->flags |= BRCM_SBA_CMD_HAS_OUTPUT;
1291 	cmdsp->data = *dst_q + msg_offset;
1292 	cmdsp->data_len = msg_len;
1293 	cmdsp++;
1294 
1295 skip_q:
1296 	/* Fillup brcm_message */
1297 	msg->type = BRCM_MESSAGE_SBA;
1298 	msg->sba.cmds = cmds;
1299 	msg->sba.cmds_count = cmdsp - cmds;
1300 	msg->ctx = req;
1301 	msg->error = 0;
1302 }
1303 
1304 static struct sba_request *
1305 sba_prep_dma_pq_single_req(struct sba_device *sba, dma_addr_t off,
1306 			   dma_addr_t *dst_p, dma_addr_t *dst_q,
1307 			   dma_addr_t src, u8 scf, size_t len,
1308 			   unsigned long flags)
1309 {
1310 	struct sba_request *req = NULL;
1311 
1312 	/* Alloc new request */
1313 	req = sba_alloc_request(sba);
1314 	if (!req)
1315 		return NULL;
1316 	if (flags & DMA_PREP_FENCE)
1317 		req->flags |= SBA_REQUEST_FENCE;
1318 
1319 	/* Fillup request messages */
1320 	sba_fillup_pq_single_msg(req,  dmaf_continue(flags),
1321 				 req->cmds, &req->msg, off, len,
1322 				 dst_p, dst_q, src, scf);
1323 
1324 	/* Init async_tx descriptor */
1325 	req->tx.flags = flags;
1326 	req->tx.cookie = -EBUSY;
1327 
1328 	return req;
1329 }
1330 
1331 static struct dma_async_tx_descriptor *
1332 sba_prep_dma_pq(struct dma_chan *dchan, dma_addr_t *dst, dma_addr_t *src,
1333 		u32 src_cnt, const u8 *scf, size_t len, unsigned long flags)
1334 {
1335 	u32 i, dst_q_index;
1336 	size_t req_len;
1337 	bool slow = false;
1338 	dma_addr_t off = 0;
1339 	dma_addr_t *dst_p = NULL, *dst_q = NULL;
1340 	struct sba_device *sba = to_sba_device(dchan);
1341 	struct sba_request *first = NULL, *req;
1342 
1343 	/* Sanity checks */
1344 	if (unlikely(src_cnt > sba->max_pq_srcs))
1345 		return NULL;
1346 	for (i = 0; i < src_cnt; i++)
1347 		if (sba->max_pq_coefs <= raid6_gflog[scf[i]])
1348 			slow = true;
1349 
1350 	/* Figure-out P and Q destination addresses */
1351 	if (!(flags & DMA_PREP_PQ_DISABLE_P))
1352 		dst_p = &dst[0];
1353 	if (!(flags & DMA_PREP_PQ_DISABLE_Q))
1354 		dst_q = &dst[1];
1355 
1356 	/* Create chained requests where each request is upto hw_buf_size */
1357 	while (len) {
1358 		req_len = (len < sba->hw_buf_size) ? len : sba->hw_buf_size;
1359 
1360 		if (slow) {
1361 			dst_q_index = src_cnt;
1362 
1363 			if (dst_q) {
1364 				for (i = 0; i < src_cnt; i++) {
1365 					if (*dst_q == src[i]) {
1366 						dst_q_index = i;
1367 						break;
1368 					}
1369 				}
1370 			}
1371 
1372 			if (dst_q_index < src_cnt) {
1373 				i = dst_q_index;
1374 				req = sba_prep_dma_pq_single_req(sba,
1375 					off, dst_p, dst_q, src[i], scf[i],
1376 					req_len, flags | DMA_PREP_FENCE);
1377 				if (!req)
1378 					goto fail;
1379 
1380 				if (first)
1381 					sba_chain_request(first, req);
1382 				else
1383 					first = req;
1384 
1385 				flags |= DMA_PREP_CONTINUE;
1386 			}
1387 
1388 			for (i = 0; i < src_cnt; i++) {
1389 				if (dst_q_index == i)
1390 					continue;
1391 
1392 				req = sba_prep_dma_pq_single_req(sba,
1393 					off, dst_p, dst_q, src[i], scf[i],
1394 					req_len, flags | DMA_PREP_FENCE);
1395 				if (!req)
1396 					goto fail;
1397 
1398 				if (first)
1399 					sba_chain_request(first, req);
1400 				else
1401 					first = req;
1402 
1403 				flags |= DMA_PREP_CONTINUE;
1404 			}
1405 		} else {
1406 			req = sba_prep_dma_pq_req(sba, off,
1407 						  dst_p, dst_q, src, src_cnt,
1408 						  scf, req_len, flags);
1409 			if (!req)
1410 				goto fail;
1411 
1412 			if (first)
1413 				sba_chain_request(first, req);
1414 			else
1415 				first = req;
1416 		}
1417 
1418 		off += req_len;
1419 		len -= req_len;
1420 	}
1421 
1422 	return (first) ? &first->tx : NULL;
1423 
1424 fail:
1425 	if (first)
1426 		sba_free_chained_requests(first);
1427 	return NULL;
1428 }
1429 
1430 /* ====== Mailbox callbacks ===== */
1431 
1432 static void sba_receive_message(struct mbox_client *cl, void *msg)
1433 {
1434 	struct brcm_message *m = msg;
1435 	struct sba_request *req = m->ctx;
1436 	struct sba_device *sba = req->sba;
1437 
1438 	/* Error count if message has error */
1439 	if (m->error < 0)
1440 		dev_err(sba->dev, "%s got message with error %d",
1441 			dma_chan_name(&sba->dma_chan), m->error);
1442 
1443 	/* Process received request */
1444 	sba_process_received_request(sba, req);
1445 }
1446 
1447 /* ====== Debugfs callbacks ====== */
1448 
1449 static int sba_debugfs_stats_show(struct seq_file *file, void *offset)
1450 {
1451 	struct sba_device *sba = dev_get_drvdata(file->private);
1452 
1453 	/* Write stats in file */
1454 	sba_write_stats_in_seqfile(sba, file);
1455 
1456 	return 0;
1457 }
1458 
1459 /* ====== Platform driver routines ===== */
1460 
1461 static int sba_prealloc_channel_resources(struct sba_device *sba)
1462 {
1463 	int i, j, ret = 0;
1464 	struct sba_request *req = NULL;
1465 
1466 	sba->resp_base = dma_alloc_coherent(sba->mbox_dev,
1467 					    sba->max_resp_pool_size,
1468 					    &sba->resp_dma_base, GFP_KERNEL);
1469 	if (!sba->resp_base)
1470 		return -ENOMEM;
1471 
1472 	sba->cmds_base = dma_alloc_coherent(sba->mbox_dev,
1473 					    sba->max_cmds_pool_size,
1474 					    &sba->cmds_dma_base, GFP_KERNEL);
1475 	if (!sba->cmds_base) {
1476 		ret = -ENOMEM;
1477 		goto fail_free_resp_pool;
1478 	}
1479 
1480 	spin_lock_init(&sba->reqs_lock);
1481 	sba->reqs_fence = false;
1482 	INIT_LIST_HEAD(&sba->reqs_alloc_list);
1483 	INIT_LIST_HEAD(&sba->reqs_pending_list);
1484 	INIT_LIST_HEAD(&sba->reqs_active_list);
1485 	INIT_LIST_HEAD(&sba->reqs_aborted_list);
1486 	INIT_LIST_HEAD(&sba->reqs_free_list);
1487 
1488 	for (i = 0; i < sba->max_req; i++) {
1489 		req = devm_kzalloc(sba->dev,
1490 				   struct_size(req, cmds, sba->max_cmd_per_req),
1491 				   GFP_KERNEL);
1492 		if (!req) {
1493 			ret = -ENOMEM;
1494 			goto fail_free_cmds_pool;
1495 		}
1496 		INIT_LIST_HEAD(&req->node);
1497 		req->sba = sba;
1498 		req->flags = SBA_REQUEST_STATE_FREE;
1499 		INIT_LIST_HEAD(&req->next);
1500 		atomic_set(&req->next_pending_count, 0);
1501 		for (j = 0; j < sba->max_cmd_per_req; j++) {
1502 			req->cmds[j].cmd = 0;
1503 			req->cmds[j].cmd_dma = sba->cmds_base +
1504 				(i * sba->max_cmd_per_req + j) * sizeof(u64);
1505 			req->cmds[j].cmd_dma_addr = sba->cmds_dma_base +
1506 				(i * sba->max_cmd_per_req + j) * sizeof(u64);
1507 			req->cmds[j].flags = 0;
1508 		}
1509 		memset(&req->msg, 0, sizeof(req->msg));
1510 		dma_async_tx_descriptor_init(&req->tx, &sba->dma_chan);
1511 		async_tx_ack(&req->tx);
1512 		req->tx.tx_submit = sba_tx_submit;
1513 		req->tx.phys = sba->resp_dma_base + i * sba->hw_resp_size;
1514 		list_add_tail(&req->node, &sba->reqs_free_list);
1515 	}
1516 
1517 	return 0;
1518 
1519 fail_free_cmds_pool:
1520 	dma_free_coherent(sba->mbox_dev,
1521 			  sba->max_cmds_pool_size,
1522 			  sba->cmds_base, sba->cmds_dma_base);
1523 fail_free_resp_pool:
1524 	dma_free_coherent(sba->mbox_dev,
1525 			  sba->max_resp_pool_size,
1526 			  sba->resp_base, sba->resp_dma_base);
1527 	return ret;
1528 }
1529 
1530 static void sba_freeup_channel_resources(struct sba_device *sba)
1531 {
1532 	dmaengine_terminate_all(&sba->dma_chan);
1533 	dma_free_coherent(sba->mbox_dev, sba->max_cmds_pool_size,
1534 			  sba->cmds_base, sba->cmds_dma_base);
1535 	dma_free_coherent(sba->mbox_dev, sba->max_resp_pool_size,
1536 			  sba->resp_base, sba->resp_dma_base);
1537 	sba->resp_base = NULL;
1538 	sba->resp_dma_base = 0;
1539 }
1540 
1541 static int sba_async_register(struct sba_device *sba)
1542 {
1543 	int ret;
1544 	struct dma_device *dma_dev = &sba->dma_dev;
1545 
1546 	/* Initialize DMA channel cookie */
1547 	sba->dma_chan.device = dma_dev;
1548 	dma_cookie_init(&sba->dma_chan);
1549 
1550 	/* Initialize DMA device capability mask */
1551 	dma_cap_zero(dma_dev->cap_mask);
1552 	dma_cap_set(DMA_INTERRUPT, dma_dev->cap_mask);
1553 	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
1554 	dma_cap_set(DMA_XOR, dma_dev->cap_mask);
1555 	dma_cap_set(DMA_PQ, dma_dev->cap_mask);
1556 
1557 	/*
1558 	 * Set mailbox channel device as the base device of
1559 	 * our dma_device because the actual memory accesses
1560 	 * will be done by mailbox controller
1561 	 */
1562 	dma_dev->dev = sba->mbox_dev;
1563 
1564 	/* Set base prep routines */
1565 	dma_dev->device_free_chan_resources = sba_free_chan_resources;
1566 	dma_dev->device_terminate_all = sba_device_terminate_all;
1567 	dma_dev->device_issue_pending = sba_issue_pending;
1568 	dma_dev->device_tx_status = sba_tx_status;
1569 
1570 	/* Set interrupt routine */
1571 	if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1572 		dma_dev->device_prep_dma_interrupt = sba_prep_dma_interrupt;
1573 
1574 	/* Set memcpy routine */
1575 	if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1576 		dma_dev->device_prep_dma_memcpy = sba_prep_dma_memcpy;
1577 
1578 	/* Set xor routine and capability */
1579 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1580 		dma_dev->device_prep_dma_xor = sba_prep_dma_xor;
1581 		dma_dev->max_xor = sba->max_xor_srcs;
1582 	}
1583 
1584 	/* Set pq routine and capability */
1585 	if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1586 		dma_dev->device_prep_dma_pq = sba_prep_dma_pq;
1587 		dma_set_maxpq(dma_dev, sba->max_pq_srcs, 0);
1588 	}
1589 
1590 	/* Initialize DMA device channel list */
1591 	INIT_LIST_HEAD(&dma_dev->channels);
1592 	list_add_tail(&sba->dma_chan.device_node, &dma_dev->channels);
1593 
1594 	/* Register with Linux async DMA framework*/
1595 	ret = dma_async_device_register(dma_dev);
1596 	if (ret) {
1597 		dev_err(sba->dev, "async device register error %d", ret);
1598 		return ret;
1599 	}
1600 
1601 	dev_info(sba->dev, "%s capabilities: %s%s%s%s\n",
1602 	dma_chan_name(&sba->dma_chan),
1603 	dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "interrupt " : "",
1604 	dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "memcpy " : "",
1605 	dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1606 	dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "pq " : "");
1607 
1608 	return 0;
1609 }
1610 
1611 static int sba_probe(struct platform_device *pdev)
1612 {
1613 	int ret = 0;
1614 	struct sba_device *sba;
1615 	struct platform_device *mbox_pdev;
1616 	struct of_phandle_args args;
1617 
1618 	/* Allocate main SBA struct */
1619 	sba = devm_kzalloc(&pdev->dev, sizeof(*sba), GFP_KERNEL);
1620 	if (!sba)
1621 		return -ENOMEM;
1622 
1623 	sba->dev = &pdev->dev;
1624 	platform_set_drvdata(pdev, sba);
1625 
1626 	/* Number of mailbox channels should be atleast 1 */
1627 	ret = of_count_phandle_with_args(pdev->dev.of_node,
1628 					 "mboxes", "#mbox-cells");
1629 	if (ret <= 0)
1630 		return -ENODEV;
1631 
1632 	/* Determine SBA version from DT compatible string */
1633 	if (of_device_is_compatible(sba->dev->of_node, "brcm,iproc-sba"))
1634 		sba->ver = SBA_VER_1;
1635 	else if (of_device_is_compatible(sba->dev->of_node,
1636 					 "brcm,iproc-sba-v2"))
1637 		sba->ver = SBA_VER_2;
1638 	else
1639 		return -ENODEV;
1640 
1641 	/* Derived Configuration parameters */
1642 	switch (sba->ver) {
1643 	case SBA_VER_1:
1644 		sba->hw_buf_size = 4096;
1645 		sba->hw_resp_size = 8;
1646 		sba->max_pq_coefs = 6;
1647 		sba->max_pq_srcs = 6;
1648 		break;
1649 	case SBA_VER_2:
1650 		sba->hw_buf_size = 4096;
1651 		sba->hw_resp_size = 8;
1652 		sba->max_pq_coefs = 30;
1653 		/*
1654 		 * We can support max_pq_srcs == max_pq_coefs because
1655 		 * we are limited by number of SBA commands that we can
1656 		 * fit in one message for underlying ring manager HW.
1657 		 */
1658 		sba->max_pq_srcs = 12;
1659 		break;
1660 	default:
1661 		return -EINVAL;
1662 	}
1663 	sba->max_req = SBA_MAX_REQ_PER_MBOX_CHANNEL;
1664 	sba->max_cmd_per_req = sba->max_pq_srcs + 3;
1665 	sba->max_xor_srcs = sba->max_cmd_per_req - 1;
1666 	sba->max_resp_pool_size = sba->max_req * sba->hw_resp_size;
1667 	sba->max_cmds_pool_size = sba->max_req *
1668 				  sba->max_cmd_per_req * sizeof(u64);
1669 
1670 	/* Setup mailbox client */
1671 	sba->client.dev			= &pdev->dev;
1672 	sba->client.rx_callback		= sba_receive_message;
1673 	sba->client.tx_block		= false;
1674 	sba->client.knows_txdone	= true;
1675 	sba->client.tx_tout		= 0;
1676 
1677 	/* Request mailbox channel */
1678 	sba->mchan = mbox_request_channel(&sba->client, 0);
1679 	if (IS_ERR(sba->mchan)) {
1680 		ret = PTR_ERR(sba->mchan);
1681 		goto fail_free_mchan;
1682 	}
1683 
1684 	/* Find-out underlying mailbox device */
1685 	ret = of_parse_phandle_with_args(pdev->dev.of_node,
1686 					 "mboxes", "#mbox-cells", 0, &args);
1687 	if (ret)
1688 		goto fail_free_mchan;
1689 	mbox_pdev = of_find_device_by_node(args.np);
1690 	of_node_put(args.np);
1691 	if (!mbox_pdev) {
1692 		ret = -ENODEV;
1693 		goto fail_free_mchan;
1694 	}
1695 	sba->mbox_dev = &mbox_pdev->dev;
1696 
1697 	/* Prealloc channel resource */
1698 	ret = sba_prealloc_channel_resources(sba);
1699 	if (ret)
1700 		goto fail_free_mchan;
1701 
1702 	/* Check availability of debugfs */
1703 	if (!debugfs_initialized())
1704 		goto skip_debugfs;
1705 
1706 	/* Create debugfs root entry */
1707 	sba->root = debugfs_create_dir(dev_name(sba->dev), NULL);
1708 
1709 	/* Create debugfs stats entry */
1710 	debugfs_create_devm_seqfile(sba->dev, "stats", sba->root,
1711 				    sba_debugfs_stats_show);
1712 
1713 skip_debugfs:
1714 
1715 	/* Register DMA device with Linux async framework */
1716 	ret = sba_async_register(sba);
1717 	if (ret)
1718 		goto fail_free_resources;
1719 
1720 	/* Print device info */
1721 	dev_info(sba->dev, "%s using SBAv%d mailbox channel from %s",
1722 		 dma_chan_name(&sba->dma_chan), sba->ver+1,
1723 		 dev_name(sba->mbox_dev));
1724 
1725 	return 0;
1726 
1727 fail_free_resources:
1728 	debugfs_remove_recursive(sba->root);
1729 	sba_freeup_channel_resources(sba);
1730 fail_free_mchan:
1731 	mbox_free_channel(sba->mchan);
1732 	return ret;
1733 }
1734 
1735 static int sba_remove(struct platform_device *pdev)
1736 {
1737 	struct sba_device *sba = platform_get_drvdata(pdev);
1738 
1739 	dma_async_device_unregister(&sba->dma_dev);
1740 
1741 	debugfs_remove_recursive(sba->root);
1742 
1743 	sba_freeup_channel_resources(sba);
1744 
1745 	mbox_free_channel(sba->mchan);
1746 
1747 	return 0;
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 = 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