xref: /openbmc/linux/drivers/dma/fsl_raid.c (revision 63705da3)
1 /*
2  * drivers/dma/fsl_raid.c
3  *
4  * Freescale RAID Engine device driver
5  *
6  * Author:
7  *	Harninder Rai <harninder.rai@freescale.com>
8  *	Naveen Burmi <naveenburmi@freescale.com>
9  *
10  * Rewrite:
11  *	Xuelin Shi <xuelin.shi@freescale.com>
12  *
13  * Copyright (c) 2010-2014 Freescale Semiconductor, Inc.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions are met:
17  *     * Redistributions of source code must retain the above copyright
18  *       notice, this list of conditions and the following disclaimer.
19  *     * Redistributions in binary form must reproduce the above copyright
20  *       notice, this list of conditions and the following disclaimer in the
21  *       documentation and/or other materials provided with the distribution.
22  *     * Neither the name of Freescale Semiconductor nor the
23  *       names of its contributors may be used to endorse or promote products
24  *       derived from this software without specific prior written permission.
25  *
26  * ALTERNATIVELY, this software may be distributed under the terms of the
27  * GNU General Public License ("GPL") as published by the Free Software
28  * Foundation, either version 2 of that License or (at your option) any
29  * later version.
30  *
31  * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
32  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
33  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
34  * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
35  * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
36  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
37  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
38  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
39  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
40  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
41  *
42  * Theory of operation:
43  *
44  * General capabilities:
45  *	RAID Engine (RE) block is capable of offloading XOR, memcpy and P/Q
46  *	calculations required in RAID5 and RAID6 operations. RE driver
47  *	registers with Linux's ASYNC layer as dma driver. RE hardware
48  *	maintains strict ordering of the requests through chained
49  *	command queueing.
50  *
51  * Data flow:
52  *	Software RAID layer of Linux (MD layer) maintains RAID partitions,
53  *	strips, stripes etc. It sends requests to the underlying ASYNC layer
54  *	which further passes it to RE driver. ASYNC layer decides which request
55  *	goes to which job ring of RE hardware. For every request processed by
56  *	RAID Engine, driver gets an interrupt unless coalescing is set. The
57  *	per job ring interrupt handler checks the status register for errors,
58  *	clears the interrupt and leave the post interrupt processing to the irq
59  *	thread.
60  */
61 #include <linux/interrupt.h>
62 #include <linux/module.h>
63 #include <linux/of_irq.h>
64 #include <linux/of_address.h>
65 #include <linux/of_platform.h>
66 #include <linux/dma-mapping.h>
67 #include <linux/dmapool.h>
68 #include <linux/dmaengine.h>
69 #include <linux/io.h>
70 #include <linux/spinlock.h>
71 #include <linux/slab.h>
72 
73 #include "dmaengine.h"
74 #include "fsl_raid.h"
75 
76 #define FSL_RE_MAX_XOR_SRCS	16
77 #define FSL_RE_MAX_PQ_SRCS	16
78 #define FSL_RE_MIN_DESCS	256
79 #define FSL_RE_MAX_DESCS	(4 * FSL_RE_MIN_DESCS)
80 #define FSL_RE_FRAME_FORMAT	0x1
81 #define FSL_RE_MAX_DATA_LEN	(1024*1024)
82 
83 #define to_fsl_re_dma_desc(tx) container_of(tx, struct fsl_re_desc, async_tx)
84 
85 /* Add descriptors into per chan software queue - submit_q */
86 static dma_cookie_t fsl_re_tx_submit(struct dma_async_tx_descriptor *tx)
87 {
88 	struct fsl_re_desc *desc;
89 	struct fsl_re_chan *re_chan;
90 	dma_cookie_t cookie;
91 	unsigned long flags;
92 
93 	desc = to_fsl_re_dma_desc(tx);
94 	re_chan = container_of(tx->chan, struct fsl_re_chan, chan);
95 
96 	spin_lock_irqsave(&re_chan->desc_lock, flags);
97 	cookie = dma_cookie_assign(tx);
98 	list_add_tail(&desc->node, &re_chan->submit_q);
99 	spin_unlock_irqrestore(&re_chan->desc_lock, flags);
100 
101 	return cookie;
102 }
103 
104 /* Copy descriptor from per chan software queue into hardware job ring */
105 static void fsl_re_issue_pending(struct dma_chan *chan)
106 {
107 	struct fsl_re_chan *re_chan;
108 	int avail;
109 	struct fsl_re_desc *desc, *_desc;
110 	unsigned long flags;
111 
112 	re_chan = container_of(chan, struct fsl_re_chan, chan);
113 
114 	spin_lock_irqsave(&re_chan->desc_lock, flags);
115 	avail = FSL_RE_SLOT_AVAIL(
116 		in_be32(&re_chan->jrregs->inbring_slot_avail));
117 
118 	list_for_each_entry_safe(desc, _desc, &re_chan->submit_q, node) {
119 		if (!avail)
120 			break;
121 
122 		list_move_tail(&desc->node, &re_chan->active_q);
123 
124 		memcpy(&re_chan->inb_ring_virt_addr[re_chan->inb_count],
125 		       &desc->hwdesc, sizeof(struct fsl_re_hw_desc));
126 
127 		re_chan->inb_count = (re_chan->inb_count + 1) &
128 						FSL_RE_RING_SIZE_MASK;
129 		out_be32(&re_chan->jrregs->inbring_add_job, FSL_RE_ADD_JOB(1));
130 		avail--;
131 	}
132 	spin_unlock_irqrestore(&re_chan->desc_lock, flags);
133 }
134 
135 static void fsl_re_desc_done(struct fsl_re_desc *desc)
136 {
137 	dma_cookie_complete(&desc->async_tx);
138 	dma_descriptor_unmap(&desc->async_tx);
139 	dmaengine_desc_get_callback_invoke(&desc->async_tx, NULL);
140 }
141 
142 static void fsl_re_cleanup_descs(struct fsl_re_chan *re_chan)
143 {
144 	struct fsl_re_desc *desc, *_desc;
145 	unsigned long flags;
146 
147 	spin_lock_irqsave(&re_chan->desc_lock, flags);
148 	list_for_each_entry_safe(desc, _desc, &re_chan->ack_q, node) {
149 		if (async_tx_test_ack(&desc->async_tx))
150 			list_move_tail(&desc->node, &re_chan->free_q);
151 	}
152 	spin_unlock_irqrestore(&re_chan->desc_lock, flags);
153 
154 	fsl_re_issue_pending(&re_chan->chan);
155 }
156 
157 static void fsl_re_dequeue(struct tasklet_struct *t)
158 {
159 	struct fsl_re_chan *re_chan = from_tasklet(re_chan, t, irqtask);
160 	struct fsl_re_desc *desc, *_desc;
161 	struct fsl_re_hw_desc *hwdesc;
162 	unsigned long flags;
163 	unsigned int count, oub_count;
164 	int found;
165 
166 	fsl_re_cleanup_descs(re_chan);
167 
168 	spin_lock_irqsave(&re_chan->desc_lock, flags);
169 	count =	FSL_RE_SLOT_FULL(in_be32(&re_chan->jrregs->oubring_slot_full));
170 	while (count--) {
171 		found = 0;
172 		hwdesc = &re_chan->oub_ring_virt_addr[re_chan->oub_count];
173 		list_for_each_entry_safe(desc, _desc, &re_chan->active_q,
174 					 node) {
175 			/* compare the hw dma addr to find the completed */
176 			if (desc->hwdesc.lbea32 == hwdesc->lbea32 &&
177 			    desc->hwdesc.addr_low == hwdesc->addr_low) {
178 				found = 1;
179 				break;
180 			}
181 		}
182 
183 		if (found) {
184 			fsl_re_desc_done(desc);
185 			list_move_tail(&desc->node, &re_chan->ack_q);
186 		} else {
187 			dev_err(re_chan->dev,
188 				"found hwdesc not in sw queue, discard it\n");
189 		}
190 
191 		oub_count = (re_chan->oub_count + 1) & FSL_RE_RING_SIZE_MASK;
192 		re_chan->oub_count = oub_count;
193 
194 		out_be32(&re_chan->jrregs->oubring_job_rmvd,
195 			 FSL_RE_RMVD_JOB(1));
196 	}
197 	spin_unlock_irqrestore(&re_chan->desc_lock, flags);
198 }
199 
200 /* Per Job Ring interrupt handler */
201 static irqreturn_t fsl_re_isr(int irq, void *data)
202 {
203 	struct fsl_re_chan *re_chan;
204 	u32 irqstate, status;
205 
206 	re_chan = dev_get_drvdata((struct device *)data);
207 
208 	irqstate = in_be32(&re_chan->jrregs->jr_interrupt_status);
209 	if (!irqstate)
210 		return IRQ_NONE;
211 
212 	/*
213 	 * There's no way in upper layer (read MD layer) to recover from
214 	 * error conditions except restart everything. In long term we
215 	 * need to do something more than just crashing
216 	 */
217 	if (irqstate & FSL_RE_ERROR) {
218 		status = in_be32(&re_chan->jrregs->jr_status);
219 		dev_err(re_chan->dev, "chan error irqstate: %x, status: %x\n",
220 			irqstate, status);
221 	}
222 
223 	/* Clear interrupt */
224 	out_be32(&re_chan->jrregs->jr_interrupt_status, FSL_RE_CLR_INTR);
225 
226 	tasklet_schedule(&re_chan->irqtask);
227 
228 	return IRQ_HANDLED;
229 }
230 
231 static enum dma_status fsl_re_tx_status(struct dma_chan *chan,
232 					dma_cookie_t cookie,
233 					struct dma_tx_state *txstate)
234 {
235 	return dma_cookie_status(chan, cookie, txstate);
236 }
237 
238 static void fill_cfd_frame(struct fsl_re_cmpnd_frame *cf, u8 index,
239 			   size_t length, dma_addr_t addr, bool final)
240 {
241 	u32 efrl = length & FSL_RE_CF_LENGTH_MASK;
242 
243 	efrl |= final << FSL_RE_CF_FINAL_SHIFT;
244 	cf[index].efrl32 = efrl;
245 	cf[index].addr_high = upper_32_bits(addr);
246 	cf[index].addr_low = lower_32_bits(addr);
247 }
248 
249 static struct fsl_re_desc *fsl_re_init_desc(struct fsl_re_chan *re_chan,
250 					    struct fsl_re_desc *desc,
251 					    void *cf, dma_addr_t paddr)
252 {
253 	desc->re_chan = re_chan;
254 	desc->async_tx.tx_submit = fsl_re_tx_submit;
255 	dma_async_tx_descriptor_init(&desc->async_tx, &re_chan->chan);
256 	INIT_LIST_HEAD(&desc->node);
257 
258 	desc->hwdesc.fmt32 = FSL_RE_FRAME_FORMAT << FSL_RE_HWDESC_FMT_SHIFT;
259 	desc->hwdesc.lbea32 = upper_32_bits(paddr);
260 	desc->hwdesc.addr_low = lower_32_bits(paddr);
261 	desc->cf_addr = cf;
262 	desc->cf_paddr = paddr;
263 
264 	desc->cdb_addr = (void *)(cf + FSL_RE_CF_DESC_SIZE);
265 	desc->cdb_paddr = paddr + FSL_RE_CF_DESC_SIZE;
266 
267 	return desc;
268 }
269 
270 static struct fsl_re_desc *fsl_re_chan_alloc_desc(struct fsl_re_chan *re_chan,
271 						  unsigned long flags)
272 {
273 	struct fsl_re_desc *desc = NULL;
274 	void *cf;
275 	dma_addr_t paddr;
276 	unsigned long lock_flag;
277 
278 	fsl_re_cleanup_descs(re_chan);
279 
280 	spin_lock_irqsave(&re_chan->desc_lock, lock_flag);
281 	if (!list_empty(&re_chan->free_q)) {
282 		/* take one desc from free_q */
283 		desc = list_first_entry(&re_chan->free_q,
284 					struct fsl_re_desc, node);
285 		list_del(&desc->node);
286 
287 		desc->async_tx.flags = flags;
288 	}
289 	spin_unlock_irqrestore(&re_chan->desc_lock, lock_flag);
290 
291 	if (!desc) {
292 		desc = kzalloc(sizeof(*desc), GFP_NOWAIT);
293 		if (!desc)
294 			return NULL;
295 
296 		cf = dma_pool_alloc(re_chan->re_dev->cf_desc_pool, GFP_NOWAIT,
297 				    &paddr);
298 		if (!cf) {
299 			kfree(desc);
300 			return NULL;
301 		}
302 
303 		desc = fsl_re_init_desc(re_chan, desc, cf, paddr);
304 		desc->async_tx.flags = flags;
305 
306 		spin_lock_irqsave(&re_chan->desc_lock, lock_flag);
307 		re_chan->alloc_count++;
308 		spin_unlock_irqrestore(&re_chan->desc_lock, lock_flag);
309 	}
310 
311 	return desc;
312 }
313 
314 static struct dma_async_tx_descriptor *fsl_re_prep_dma_genq(
315 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
316 		unsigned int src_cnt, const unsigned char *scf, size_t len,
317 		unsigned long flags)
318 {
319 	struct fsl_re_chan *re_chan;
320 	struct fsl_re_desc *desc;
321 	struct fsl_re_xor_cdb *xor;
322 	struct fsl_re_cmpnd_frame *cf;
323 	u32 cdb;
324 	unsigned int i, j;
325 	unsigned int save_src_cnt = src_cnt;
326 	int cont_q = 0;
327 
328 	re_chan = container_of(chan, struct fsl_re_chan, chan);
329 	if (len > FSL_RE_MAX_DATA_LEN) {
330 		dev_err(re_chan->dev, "genq tx length %zu, max length %d\n",
331 			len, FSL_RE_MAX_DATA_LEN);
332 		return NULL;
333 	}
334 
335 	desc = fsl_re_chan_alloc_desc(re_chan, flags);
336 	if (desc <= 0)
337 		return NULL;
338 
339 	if (scf && (flags & DMA_PREP_CONTINUE)) {
340 		cont_q = 1;
341 		src_cnt += 1;
342 	}
343 
344 	/* Filling xor CDB */
345 	cdb = FSL_RE_XOR_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
346 	cdb |= (src_cnt - 1) << FSL_RE_CDB_NRCS_SHIFT;
347 	cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
348 	cdb |= FSL_RE_INTR_ON_ERROR << FSL_RE_CDB_ERROR_SHIFT;
349 	cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
350 	xor = desc->cdb_addr;
351 	xor->cdb32 = cdb;
352 
353 	if (scf) {
354 		/* compute q = src0*coef0^src1*coef1^..., * is GF(8) mult */
355 		for (i = 0; i < save_src_cnt; i++)
356 			xor->gfm[i] = scf[i];
357 		if (cont_q)
358 			xor->gfm[i++] = 1;
359 	} else {
360 		/* compute P, that is XOR all srcs */
361 		for (i = 0; i < src_cnt; i++)
362 			xor->gfm[i] = 1;
363 	}
364 
365 	/* Filling frame 0 of compound frame descriptor with CDB */
366 	cf = desc->cf_addr;
367 	fill_cfd_frame(cf, 0, sizeof(*xor), desc->cdb_paddr, 0);
368 
369 	/* Fill CFD's 1st frame with dest buffer */
370 	fill_cfd_frame(cf, 1, len, dest, 0);
371 
372 	/* Fill CFD's rest of the frames with source buffers */
373 	for (i = 2, j = 0; j < save_src_cnt; i++, j++)
374 		fill_cfd_frame(cf, i, len, src[j], 0);
375 
376 	if (cont_q)
377 		fill_cfd_frame(cf, i++, len, dest, 0);
378 
379 	/* Setting the final bit in the last source buffer frame in CFD */
380 	cf[i - 1].efrl32 |= 1 << FSL_RE_CF_FINAL_SHIFT;
381 
382 	return &desc->async_tx;
383 }
384 
385 /*
386  * Prep function for P parity calculation.In RAID Engine terminology,
387  * XOR calculation is called GenQ calculation done through GenQ command
388  */
389 static struct dma_async_tx_descriptor *fsl_re_prep_dma_xor(
390 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
391 		unsigned int src_cnt, size_t len, unsigned long flags)
392 {
393 	/* NULL let genq take all coef as 1 */
394 	return fsl_re_prep_dma_genq(chan, dest, src, src_cnt, NULL, len, flags);
395 }
396 
397 /*
398  * Prep function for P/Q parity calculation.In RAID Engine terminology,
399  * P/Q calculation is called GenQQ done through GenQQ command
400  */
401 static struct dma_async_tx_descriptor *fsl_re_prep_dma_pq(
402 		struct dma_chan *chan, dma_addr_t *dest, dma_addr_t *src,
403 		unsigned int src_cnt, const unsigned char *scf, size_t len,
404 		unsigned long flags)
405 {
406 	struct fsl_re_chan *re_chan;
407 	struct fsl_re_desc *desc;
408 	struct fsl_re_pq_cdb *pq;
409 	struct fsl_re_cmpnd_frame *cf;
410 	u32 cdb;
411 	u8 *p;
412 	int gfmq_len, i, j;
413 	unsigned int save_src_cnt = src_cnt;
414 
415 	re_chan = container_of(chan, struct fsl_re_chan, chan);
416 	if (len > FSL_RE_MAX_DATA_LEN) {
417 		dev_err(re_chan->dev, "pq tx length is %zu, max length is %d\n",
418 			len, FSL_RE_MAX_DATA_LEN);
419 		return NULL;
420 	}
421 
422 	/*
423 	 * RE requires at least 2 sources, if given only one source, we pass the
424 	 * second source same as the first one.
425 	 * With only one source, generating P is meaningless, only generate Q.
426 	 */
427 	if (src_cnt == 1) {
428 		struct dma_async_tx_descriptor *tx;
429 		dma_addr_t dma_src[2];
430 		unsigned char coef[2];
431 
432 		dma_src[0] = *src;
433 		coef[0] = *scf;
434 		dma_src[1] = *src;
435 		coef[1] = 0;
436 		tx = fsl_re_prep_dma_genq(chan, dest[1], dma_src, 2, coef, len,
437 					  flags);
438 		if (tx)
439 			desc = to_fsl_re_dma_desc(tx);
440 
441 		return tx;
442 	}
443 
444 	/*
445 	 * During RAID6 array creation, Linux's MD layer gets P and Q
446 	 * calculated separately in two steps. But our RAID Engine has
447 	 * the capability to calculate both P and Q with a single command
448 	 * Hence to merge well with MD layer, we need to provide a hook
449 	 * here and call re_jq_prep_dma_genq() function
450 	 */
451 
452 	if (flags & DMA_PREP_PQ_DISABLE_P)
453 		return fsl_re_prep_dma_genq(chan, dest[1], src, src_cnt,
454 				scf, len, flags);
455 
456 	if (flags & DMA_PREP_CONTINUE)
457 		src_cnt += 3;
458 
459 	desc = fsl_re_chan_alloc_desc(re_chan, flags);
460 	if (desc <= 0)
461 		return NULL;
462 
463 	/* Filling GenQQ CDB */
464 	cdb = FSL_RE_PQ_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
465 	cdb |= (src_cnt - 1) << FSL_RE_CDB_NRCS_SHIFT;
466 	cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
467 	cdb |= FSL_RE_BUFFER_OUTPUT << FSL_RE_CDB_BUFFER_SHIFT;
468 	cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
469 
470 	pq = desc->cdb_addr;
471 	pq->cdb32 = cdb;
472 
473 	p = pq->gfm_q1;
474 	/* Init gfm_q1[] */
475 	for (i = 0; i < src_cnt; i++)
476 		p[i] = 1;
477 
478 	/* Align gfm[] to 32bit */
479 	gfmq_len = ALIGN(src_cnt, 4);
480 
481 	/* Init gfm_q2[] */
482 	p += gfmq_len;
483 	for (i = 0; i < src_cnt; i++)
484 		p[i] = scf[i];
485 
486 	/* Filling frame 0 of compound frame descriptor with CDB */
487 	cf = desc->cf_addr;
488 	fill_cfd_frame(cf, 0, sizeof(struct fsl_re_pq_cdb), desc->cdb_paddr, 0);
489 
490 	/* Fill CFD's 1st & 2nd frame with dest buffers */
491 	for (i = 1, j = 0; i < 3; i++, j++)
492 		fill_cfd_frame(cf, i, len, dest[j], 0);
493 
494 	/* Fill CFD's rest of the frames with source buffers */
495 	for (i = 3, j = 0; j < save_src_cnt; i++, j++)
496 		fill_cfd_frame(cf, i, len, src[j], 0);
497 
498 	/* PQ computation continuation */
499 	if (flags & DMA_PREP_CONTINUE) {
500 		if (src_cnt - save_src_cnt == 3) {
501 			p[save_src_cnt] = 0;
502 			p[save_src_cnt + 1] = 0;
503 			p[save_src_cnt + 2] = 1;
504 			fill_cfd_frame(cf, i++, len, dest[0], 0);
505 			fill_cfd_frame(cf, i++, len, dest[1], 0);
506 			fill_cfd_frame(cf, i++, len, dest[1], 0);
507 		} else {
508 			dev_err(re_chan->dev, "PQ tx continuation error!\n");
509 			return NULL;
510 		}
511 	}
512 
513 	/* Setting the final bit in the last source buffer frame in CFD */
514 	cf[i - 1].efrl32 |= 1 << FSL_RE_CF_FINAL_SHIFT;
515 
516 	return &desc->async_tx;
517 }
518 
519 /*
520  * Prep function for memcpy. In RAID Engine, memcpy is done through MOVE
521  * command. Logic of this function will need to be modified once multipage
522  * support is added in Linux's MD/ASYNC Layer
523  */
524 static struct dma_async_tx_descriptor *fsl_re_prep_dma_memcpy(
525 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
526 		size_t len, unsigned long flags)
527 {
528 	struct fsl_re_chan *re_chan;
529 	struct fsl_re_desc *desc;
530 	size_t length;
531 	struct fsl_re_cmpnd_frame *cf;
532 	struct fsl_re_move_cdb *move;
533 	u32 cdb;
534 
535 	re_chan = container_of(chan, struct fsl_re_chan, chan);
536 
537 	if (len > FSL_RE_MAX_DATA_LEN) {
538 		dev_err(re_chan->dev, "cp tx length is %zu, max length is %d\n",
539 			len, FSL_RE_MAX_DATA_LEN);
540 		return NULL;
541 	}
542 
543 	desc = fsl_re_chan_alloc_desc(re_chan, flags);
544 	if (desc <= 0)
545 		return NULL;
546 
547 	/* Filling move CDB */
548 	cdb = FSL_RE_MOVE_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
549 	cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
550 	cdb |= FSL_RE_INTR_ON_ERROR << FSL_RE_CDB_ERROR_SHIFT;
551 	cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
552 
553 	move = desc->cdb_addr;
554 	move->cdb32 = cdb;
555 
556 	/* Filling frame 0 of CFD with move CDB */
557 	cf = desc->cf_addr;
558 	fill_cfd_frame(cf, 0, sizeof(*move), desc->cdb_paddr, 0);
559 
560 	length = min_t(size_t, len, FSL_RE_MAX_DATA_LEN);
561 
562 	/* Fill CFD's 1st frame with dest buffer */
563 	fill_cfd_frame(cf, 1, length, dest, 0);
564 
565 	/* Fill CFD's 2nd frame with src buffer */
566 	fill_cfd_frame(cf, 2, length, src, 1);
567 
568 	return &desc->async_tx;
569 }
570 
571 static int fsl_re_alloc_chan_resources(struct dma_chan *chan)
572 {
573 	struct fsl_re_chan *re_chan;
574 	struct fsl_re_desc *desc;
575 	void *cf;
576 	dma_addr_t paddr;
577 	int i;
578 
579 	re_chan = container_of(chan, struct fsl_re_chan, chan);
580 	for (i = 0; i < FSL_RE_MIN_DESCS; i++) {
581 		desc = kzalloc(sizeof(*desc), GFP_KERNEL);
582 		if (!desc)
583 			break;
584 
585 		cf = dma_pool_alloc(re_chan->re_dev->cf_desc_pool, GFP_KERNEL,
586 				    &paddr);
587 		if (!cf) {
588 			kfree(desc);
589 			break;
590 		}
591 
592 		INIT_LIST_HEAD(&desc->node);
593 		fsl_re_init_desc(re_chan, desc, cf, paddr);
594 
595 		list_add_tail(&desc->node, &re_chan->free_q);
596 		re_chan->alloc_count++;
597 	}
598 	return re_chan->alloc_count;
599 }
600 
601 static void fsl_re_free_chan_resources(struct dma_chan *chan)
602 {
603 	struct fsl_re_chan *re_chan;
604 	struct fsl_re_desc *desc;
605 
606 	re_chan = container_of(chan, struct fsl_re_chan, chan);
607 	while (re_chan->alloc_count--) {
608 		desc = list_first_entry(&re_chan->free_q,
609 					struct fsl_re_desc,
610 					node);
611 
612 		list_del(&desc->node);
613 		dma_pool_free(re_chan->re_dev->cf_desc_pool, desc->cf_addr,
614 			      desc->cf_paddr);
615 		kfree(desc);
616 	}
617 
618 	if (!list_empty(&re_chan->free_q))
619 		dev_err(re_chan->dev, "chan resource cannot be cleaned!\n");
620 }
621 
622 static int fsl_re_chan_probe(struct platform_device *ofdev,
623 		      struct device_node *np, u8 q, u32 off)
624 {
625 	struct device *dev, *chandev;
626 	struct fsl_re_drv_private *re_priv;
627 	struct fsl_re_chan *chan;
628 	struct dma_device *dma_dev;
629 	u32 ptr;
630 	u32 status;
631 	int ret = 0, rc;
632 	struct platform_device *chan_ofdev;
633 
634 	dev = &ofdev->dev;
635 	re_priv = dev_get_drvdata(dev);
636 	dma_dev = &re_priv->dma_dev;
637 
638 	chan = devm_kzalloc(dev, sizeof(*chan), GFP_KERNEL);
639 	if (!chan)
640 		return -ENOMEM;
641 
642 	/* create platform device for chan node */
643 	chan_ofdev = of_platform_device_create(np, NULL, dev);
644 	if (!chan_ofdev) {
645 		dev_err(dev, "Not able to create ofdev for jr %d\n", q);
646 		ret = -EINVAL;
647 		goto err_free;
648 	}
649 
650 	/* read reg property from dts */
651 	rc = of_property_read_u32(np, "reg", &ptr);
652 	if (rc) {
653 		dev_err(dev, "Reg property not found in jr %d\n", q);
654 		ret = -ENODEV;
655 		goto err_free;
656 	}
657 
658 	chan->jrregs = (struct fsl_re_chan_cfg *)((u8 *)re_priv->re_regs +
659 			off + ptr);
660 
661 	/* read irq property from dts */
662 	chan->irq = irq_of_parse_and_map(np, 0);
663 	if (!chan->irq) {
664 		dev_err(dev, "No IRQ defined for JR %d\n", q);
665 		ret = -ENODEV;
666 		goto err_free;
667 	}
668 
669 	snprintf(chan->name, sizeof(chan->name), "re_jr%02d", q);
670 
671 	chandev = &chan_ofdev->dev;
672 	tasklet_setup(&chan->irqtask, fsl_re_dequeue);
673 
674 	ret = request_irq(chan->irq, fsl_re_isr, 0, chan->name, chandev);
675 	if (ret) {
676 		dev_err(dev, "Unable to register interrupt for JR %d\n", q);
677 		ret = -EINVAL;
678 		goto err_free;
679 	}
680 
681 	re_priv->re_jrs[q] = chan;
682 	chan->chan.device = dma_dev;
683 	chan->chan.private = chan;
684 	chan->dev = chandev;
685 	chan->re_dev = re_priv;
686 
687 	spin_lock_init(&chan->desc_lock);
688 	INIT_LIST_HEAD(&chan->ack_q);
689 	INIT_LIST_HEAD(&chan->active_q);
690 	INIT_LIST_HEAD(&chan->submit_q);
691 	INIT_LIST_HEAD(&chan->free_q);
692 
693 	chan->inb_ring_virt_addr = dma_pool_alloc(chan->re_dev->hw_desc_pool,
694 		GFP_KERNEL, &chan->inb_phys_addr);
695 	if (!chan->inb_ring_virt_addr) {
696 		dev_err(dev, "No dma memory for inb_ring_virt_addr\n");
697 		ret = -ENOMEM;
698 		goto err_free;
699 	}
700 
701 	chan->oub_ring_virt_addr = dma_pool_alloc(chan->re_dev->hw_desc_pool,
702 		GFP_KERNEL, &chan->oub_phys_addr);
703 	if (!chan->oub_ring_virt_addr) {
704 		dev_err(dev, "No dma memory for oub_ring_virt_addr\n");
705 		ret = -ENOMEM;
706 		goto err_free_1;
707 	}
708 
709 	/* Program the Inbound/Outbound ring base addresses and size */
710 	out_be32(&chan->jrregs->inbring_base_h,
711 		 chan->inb_phys_addr & FSL_RE_ADDR_BIT_MASK);
712 	out_be32(&chan->jrregs->oubring_base_h,
713 		 chan->oub_phys_addr & FSL_RE_ADDR_BIT_MASK);
714 	out_be32(&chan->jrregs->inbring_base_l,
715 		 chan->inb_phys_addr >> FSL_RE_ADDR_BIT_SHIFT);
716 	out_be32(&chan->jrregs->oubring_base_l,
717 		 chan->oub_phys_addr >> FSL_RE_ADDR_BIT_SHIFT);
718 	out_be32(&chan->jrregs->inbring_size,
719 		 FSL_RE_RING_SIZE << FSL_RE_RING_SIZE_SHIFT);
720 	out_be32(&chan->jrregs->oubring_size,
721 		 FSL_RE_RING_SIZE << FSL_RE_RING_SIZE_SHIFT);
722 
723 	/* Read LIODN value from u-boot */
724 	status = in_be32(&chan->jrregs->jr_config_1) & FSL_RE_REG_LIODN_MASK;
725 
726 	/* Program the CFG reg */
727 	out_be32(&chan->jrregs->jr_config_1,
728 		 FSL_RE_CFG1_CBSI | FSL_RE_CFG1_CBS0 | status);
729 
730 	dev_set_drvdata(chandev, chan);
731 
732 	/* Enable RE/CHAN */
733 	out_be32(&chan->jrregs->jr_command, FSL_RE_ENABLE);
734 
735 	return 0;
736 
737 err_free_1:
738 	dma_pool_free(chan->re_dev->hw_desc_pool, chan->inb_ring_virt_addr,
739 		      chan->inb_phys_addr);
740 err_free:
741 	return ret;
742 }
743 
744 /* Probe function for RAID Engine */
745 static int fsl_re_probe(struct platform_device *ofdev)
746 {
747 	struct fsl_re_drv_private *re_priv;
748 	struct device_node *np;
749 	struct device_node *child;
750 	u32 off;
751 	u8 ridx = 0;
752 	struct dma_device *dma_dev;
753 	struct resource *res;
754 	int rc;
755 	struct device *dev = &ofdev->dev;
756 
757 	re_priv = devm_kzalloc(dev, sizeof(*re_priv), GFP_KERNEL);
758 	if (!re_priv)
759 		return -ENOMEM;
760 
761 	res = platform_get_resource(ofdev, IORESOURCE_MEM, 0);
762 	if (!res)
763 		return -ENODEV;
764 
765 	/* IOMAP the entire RAID Engine region */
766 	re_priv->re_regs = devm_ioremap(dev, res->start, resource_size(res));
767 	if (!re_priv->re_regs)
768 		return -EBUSY;
769 
770 	/* Program the RE mode */
771 	out_be32(&re_priv->re_regs->global_config, FSL_RE_NON_DPAA_MODE);
772 
773 	/* Program Galois Field polynomial */
774 	out_be32(&re_priv->re_regs->galois_field_config, FSL_RE_GFM_POLY);
775 
776 	dev_info(dev, "version %x, mode %x, gfp %x\n",
777 		 in_be32(&re_priv->re_regs->re_version_id),
778 		 in_be32(&re_priv->re_regs->global_config),
779 		 in_be32(&re_priv->re_regs->galois_field_config));
780 
781 	dma_dev = &re_priv->dma_dev;
782 	dma_dev->dev = dev;
783 	INIT_LIST_HEAD(&dma_dev->channels);
784 	dma_set_mask(dev, DMA_BIT_MASK(40));
785 
786 	dma_dev->device_alloc_chan_resources = fsl_re_alloc_chan_resources;
787 	dma_dev->device_tx_status = fsl_re_tx_status;
788 	dma_dev->device_issue_pending = fsl_re_issue_pending;
789 
790 	dma_dev->max_xor = FSL_RE_MAX_XOR_SRCS;
791 	dma_dev->device_prep_dma_xor = fsl_re_prep_dma_xor;
792 	dma_cap_set(DMA_XOR, dma_dev->cap_mask);
793 
794 	dma_dev->max_pq = FSL_RE_MAX_PQ_SRCS;
795 	dma_dev->device_prep_dma_pq = fsl_re_prep_dma_pq;
796 	dma_cap_set(DMA_PQ, dma_dev->cap_mask);
797 
798 	dma_dev->device_prep_dma_memcpy = fsl_re_prep_dma_memcpy;
799 	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
800 
801 	dma_dev->device_free_chan_resources = fsl_re_free_chan_resources;
802 
803 	re_priv->total_chans = 0;
804 
805 	re_priv->cf_desc_pool = dmam_pool_create("fsl_re_cf_desc_pool", dev,
806 					FSL_RE_CF_CDB_SIZE,
807 					FSL_RE_CF_CDB_ALIGN, 0);
808 
809 	if (!re_priv->cf_desc_pool) {
810 		dev_err(dev, "No memory for fsl re_cf desc pool\n");
811 		return -ENOMEM;
812 	}
813 
814 	re_priv->hw_desc_pool = dmam_pool_create("fsl_re_hw_desc_pool", dev,
815 			sizeof(struct fsl_re_hw_desc) * FSL_RE_RING_SIZE,
816 			FSL_RE_FRAME_ALIGN, 0);
817 	if (!re_priv->hw_desc_pool) {
818 		dev_err(dev, "No memory for fsl re_hw desc pool\n");
819 		return -ENOMEM;
820 	}
821 
822 	dev_set_drvdata(dev, re_priv);
823 
824 	/* Parse Device tree to find out the total number of JQs present */
825 	for_each_compatible_node(np, NULL, "fsl,raideng-v1.0-job-queue") {
826 		rc = of_property_read_u32(np, "reg", &off);
827 		if (rc) {
828 			dev_err(dev, "Reg property not found in JQ node\n");
829 			of_node_put(np);
830 			return -ENODEV;
831 		}
832 		/* Find out the Job Rings present under each JQ */
833 		for_each_child_of_node(np, child) {
834 			rc = of_device_is_compatible(child,
835 					     "fsl,raideng-v1.0-job-ring");
836 			if (rc) {
837 				fsl_re_chan_probe(ofdev, child, ridx++, off);
838 				re_priv->total_chans++;
839 			}
840 		}
841 	}
842 
843 	dma_async_device_register(dma_dev);
844 
845 	return 0;
846 }
847 
848 static void fsl_re_remove_chan(struct fsl_re_chan *chan)
849 {
850 	tasklet_kill(&chan->irqtask);
851 
852 	dma_pool_free(chan->re_dev->hw_desc_pool, chan->inb_ring_virt_addr,
853 		      chan->inb_phys_addr);
854 
855 	dma_pool_free(chan->re_dev->hw_desc_pool, chan->oub_ring_virt_addr,
856 		      chan->oub_phys_addr);
857 }
858 
859 static int fsl_re_remove(struct platform_device *ofdev)
860 {
861 	struct fsl_re_drv_private *re_priv;
862 	struct device *dev;
863 	int i;
864 
865 	dev = &ofdev->dev;
866 	re_priv = dev_get_drvdata(dev);
867 
868 	/* Cleanup chan related memory areas */
869 	for (i = 0; i < re_priv->total_chans; i++)
870 		fsl_re_remove_chan(re_priv->re_jrs[i]);
871 
872 	/* Unregister the driver */
873 	dma_async_device_unregister(&re_priv->dma_dev);
874 
875 	return 0;
876 }
877 
878 static const struct of_device_id fsl_re_ids[] = {
879 	{ .compatible = "fsl,raideng-v1.0", },
880 	{}
881 };
882 MODULE_DEVICE_TABLE(of, fsl_re_ids);
883 
884 static struct platform_driver fsl_re_driver = {
885 	.driver = {
886 		.name = "fsl-raideng",
887 		.of_match_table = fsl_re_ids,
888 	},
889 	.probe = fsl_re_probe,
890 	.remove = fsl_re_remove,
891 };
892 
893 module_platform_driver(fsl_re_driver);
894 
895 MODULE_AUTHOR("Harninder Rai <harninder.rai@freescale.com>");
896 MODULE_LICENSE("GPL v2");
897 MODULE_DESCRIPTION("Freescale RAID Engine Device Driver");
898