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 */
 86static 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 */
105static 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
135static 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
142static 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
157static void fsl_re_dequeue(unsigned long data)
158{
159	struct fsl_re_chan *re_chan;
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	re_chan = dev_get_drvdata((struct device *)data);
167
168	fsl_re_cleanup_descs(re_chan);
169
170	spin_lock_irqsave(&re_chan->desc_lock, flags);
171	count =	FSL_RE_SLOT_FULL(in_be32(&re_chan->jrregs->oubring_slot_full));
172	while (count--) {
173		found = 0;
174		hwdesc = &re_chan->oub_ring_virt_addr[re_chan->oub_count];
175		list_for_each_entry_safe(desc, _desc, &re_chan->active_q,
176					 node) {
177			/* compare the hw dma addr to find the completed */
178			if (desc->hwdesc.lbea32 == hwdesc->lbea32 &&
179			    desc->hwdesc.addr_low == hwdesc->addr_low) {
180				found = 1;
181				break;
182			}
183		}
184
185		if (found) {
186			fsl_re_desc_done(desc);
187			list_move_tail(&desc->node, &re_chan->ack_q);
188		} else {
189			dev_err(re_chan->dev,
190				"found hwdesc not in sw queue, discard it\n");
191		}
192
193		oub_count = (re_chan->oub_count + 1) & FSL_RE_RING_SIZE_MASK;
194		re_chan->oub_count = oub_count;
195
196		out_be32(&re_chan->jrregs->oubring_job_rmvd,
197			 FSL_RE_RMVD_JOB(1));
198	}
199	spin_unlock_irqrestore(&re_chan->desc_lock, flags);
200}
201
202/* Per Job Ring interrupt handler */
203static irqreturn_t fsl_re_isr(int irq, void *data)
204{
205	struct fsl_re_chan *re_chan;
206	u32 irqstate, status;
207
208	re_chan = dev_get_drvdata((struct device *)data);
209
210	irqstate = in_be32(&re_chan->jrregs->jr_interrupt_status);
211	if (!irqstate)
212		return IRQ_NONE;
213
214	/*
215	 * There's no way in upper layer (read MD layer) to recover from
216	 * error conditions except restart everything. In long term we
217	 * need to do something more than just crashing
218	 */
219	if (irqstate & FSL_RE_ERROR) {
220		status = in_be32(&re_chan->jrregs->jr_status);
221		dev_err(re_chan->dev, "chan error irqstate: %x, status: %x\n",
222			irqstate, status);
223	}
224
225	/* Clear interrupt */
226	out_be32(&re_chan->jrregs->jr_interrupt_status, FSL_RE_CLR_INTR);
227
228	tasklet_schedule(&re_chan->irqtask);
229
230	return IRQ_HANDLED;
231}
232
233static enum dma_status fsl_re_tx_status(struct dma_chan *chan,
234					dma_cookie_t cookie,
235					struct dma_tx_state *txstate)
236{
237	return dma_cookie_status(chan, cookie, txstate);
238}
239
240static void fill_cfd_frame(struct fsl_re_cmpnd_frame *cf, u8 index,
241			   size_t length, dma_addr_t addr, bool final)
242{
243	u32 efrl = length & FSL_RE_CF_LENGTH_MASK;
244
245	efrl |= final << FSL_RE_CF_FINAL_SHIFT;
246	cf[index].efrl32 = efrl;
247	cf[index].addr_high = upper_32_bits(addr);
248	cf[index].addr_low = lower_32_bits(addr);
249}
250
251static struct fsl_re_desc *fsl_re_init_desc(struct fsl_re_chan *re_chan,
252					    struct fsl_re_desc *desc,
253					    void *cf, dma_addr_t paddr)
254{
255	desc->re_chan = re_chan;
256	desc->async_tx.tx_submit = fsl_re_tx_submit;
257	dma_async_tx_descriptor_init(&desc->async_tx, &re_chan->chan);
258	INIT_LIST_HEAD(&desc->node);
259
260	desc->hwdesc.fmt32 = FSL_RE_FRAME_FORMAT << FSL_RE_HWDESC_FMT_SHIFT;
261	desc->hwdesc.lbea32 = upper_32_bits(paddr);
262	desc->hwdesc.addr_low = lower_32_bits(paddr);
263	desc->cf_addr = cf;
264	desc->cf_paddr = paddr;
265
266	desc->cdb_addr = (void *)(cf + FSL_RE_CF_DESC_SIZE);
267	desc->cdb_paddr = paddr + FSL_RE_CF_DESC_SIZE;
268
269	return desc;
270}
271
272static struct fsl_re_desc *fsl_re_chan_alloc_desc(struct fsl_re_chan *re_chan,
273						  unsigned long flags)
274{
275	struct fsl_re_desc *desc = NULL;
276	void *cf;
277	dma_addr_t paddr;
278	unsigned long lock_flag;
279
280	fsl_re_cleanup_descs(re_chan);
281
282	spin_lock_irqsave(&re_chan->desc_lock, lock_flag);
283	if (!list_empty(&re_chan->free_q)) {
284		/* take one desc from free_q */
285		desc = list_first_entry(&re_chan->free_q,
286					struct fsl_re_desc, node);
287		list_del(&desc->node);
288
289		desc->async_tx.flags = flags;
290	}
291	spin_unlock_irqrestore(&re_chan->desc_lock, lock_flag);
292
293	if (!desc) {
294		desc = kzalloc(sizeof(*desc), GFP_NOWAIT);
295		if (!desc)
296			return NULL;
297
298		cf = dma_pool_alloc(re_chan->re_dev->cf_desc_pool, GFP_NOWAIT,
299				    &paddr);
300		if (!cf) {
301			kfree(desc);
302			return NULL;
303		}
304
305		desc = fsl_re_init_desc(re_chan, desc, cf, paddr);
306		desc->async_tx.flags = flags;
307
308		spin_lock_irqsave(&re_chan->desc_lock, lock_flag);
309		re_chan->alloc_count++;
310		spin_unlock_irqrestore(&re_chan->desc_lock, lock_flag);
311	}
312
313	return desc;
314}
315
316static struct dma_async_tx_descriptor *fsl_re_prep_dma_genq(
317		struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
318		unsigned int src_cnt, const unsigned char *scf, size_t len,
319		unsigned long flags)
320{
321	struct fsl_re_chan *re_chan;
322	struct fsl_re_desc *desc;
323	struct fsl_re_xor_cdb *xor;
324	struct fsl_re_cmpnd_frame *cf;
325	u32 cdb;
326	unsigned int i, j;
327	unsigned int save_src_cnt = src_cnt;
328	int cont_q = 0;
329
330	re_chan = container_of(chan, struct fsl_re_chan, chan);
331	if (len > FSL_RE_MAX_DATA_LEN) {
332		dev_err(re_chan->dev, "genq tx length %zu, max length %d\n",
333			len, FSL_RE_MAX_DATA_LEN);
334		return NULL;
335	}
336
337	desc = fsl_re_chan_alloc_desc(re_chan, flags);
338	if (desc <= 0)
339		return NULL;
340
341	if (scf && (flags & DMA_PREP_CONTINUE)) {
342		cont_q = 1;
343		src_cnt += 1;
344	}
345
346	/* Filling xor CDB */
347	cdb = FSL_RE_XOR_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
348	cdb |= (src_cnt - 1) << FSL_RE_CDB_NRCS_SHIFT;
349	cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
350	cdb |= FSL_RE_INTR_ON_ERROR << FSL_RE_CDB_ERROR_SHIFT;
351	cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
352	xor = desc->cdb_addr;
353	xor->cdb32 = cdb;
354
355	if (scf) {
356		/* compute q = src0*coef0^src1*coef1^..., * is GF(8) mult */
357		for (i = 0; i < save_src_cnt; i++)
358			xor->gfm[i] = scf[i];
359		if (cont_q)
360			xor->gfm[i++] = 1;
361	} else {
362		/* compute P, that is XOR all srcs */
363		for (i = 0; i < src_cnt; i++)
364			xor->gfm[i] = 1;
365	}
366
367	/* Filling frame 0 of compound frame descriptor with CDB */
368	cf = desc->cf_addr;
369	fill_cfd_frame(cf, 0, sizeof(*xor), desc->cdb_paddr, 0);
370
371	/* Fill CFD's 1st frame with dest buffer */
372	fill_cfd_frame(cf, 1, len, dest, 0);
373
374	/* Fill CFD's rest of the frames with source buffers */
375	for (i = 2, j = 0; j < save_src_cnt; i++, j++)
376		fill_cfd_frame(cf, i, len, src[j], 0);
377
378	if (cont_q)
379		fill_cfd_frame(cf, i++, len, dest, 0);
380
381	/* Setting the final bit in the last source buffer frame in CFD */
382	cf[i - 1].efrl32 |= 1 << FSL_RE_CF_FINAL_SHIFT;
383
384	return &desc->async_tx;
385}
386
387/*
388 * Prep function for P parity calculation.In RAID Engine terminology,
389 * XOR calculation is called GenQ calculation done through GenQ command
390 */
391static struct dma_async_tx_descriptor *fsl_re_prep_dma_xor(
392		struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
393		unsigned int src_cnt, size_t len, unsigned long flags)
394{
395	/* NULL let genq take all coef as 1 */
396	return fsl_re_prep_dma_genq(chan, dest, src, src_cnt, NULL, len, flags);
397}
398
399/*
400 * Prep function for P/Q parity calculation.In RAID Engine terminology,
401 * P/Q calculation is called GenQQ done through GenQQ command
402 */
403static struct dma_async_tx_descriptor *fsl_re_prep_dma_pq(
404		struct dma_chan *chan, dma_addr_t *dest, dma_addr_t *src,
405		unsigned int src_cnt, const unsigned char *scf, size_t len,
406		unsigned long flags)
407{
408	struct fsl_re_chan *re_chan;
409	struct fsl_re_desc *desc;
410	struct fsl_re_pq_cdb *pq;
411	struct fsl_re_cmpnd_frame *cf;
412	u32 cdb;
413	u8 *p;
414	int gfmq_len, i, j;
415	unsigned int save_src_cnt = src_cnt;
416
417	re_chan = container_of(chan, struct fsl_re_chan, chan);
418	if (len > FSL_RE_MAX_DATA_LEN) {
419		dev_err(re_chan->dev, "pq tx length is %zu, max length is %d\n",
420			len, FSL_RE_MAX_DATA_LEN);
421		return NULL;
422	}
423
424	/*
425	 * RE requires at least 2 sources, if given only one source, we pass the
426	 * second source same as the first one.
427	 * With only one source, generating P is meaningless, only generate Q.
428	 */
429	if (src_cnt == 1) {
430		struct dma_async_tx_descriptor *tx;
431		dma_addr_t dma_src[2];
432		unsigned char coef[2];
433
434		dma_src[0] = *src;
435		coef[0] = *scf;
436		dma_src[1] = *src;
437		coef[1] = 0;
438		tx = fsl_re_prep_dma_genq(chan, dest[1], dma_src, 2, coef, len,
439					  flags);
440		if (tx)
441			desc = to_fsl_re_dma_desc(tx);
442
443		return tx;
444	}
445
446	/*
447	 * During RAID6 array creation, Linux's MD layer gets P and Q
448	 * calculated separately in two steps. But our RAID Engine has
449	 * the capability to calculate both P and Q with a single command
450	 * Hence to merge well with MD layer, we need to provide a hook
451	 * here and call re_jq_prep_dma_genq() function
452	 */
453
454	if (flags & DMA_PREP_PQ_DISABLE_P)
455		return fsl_re_prep_dma_genq(chan, dest[1], src, src_cnt,
456				scf, len, flags);
457
458	if (flags & DMA_PREP_CONTINUE)
459		src_cnt += 3;
460
461	desc = fsl_re_chan_alloc_desc(re_chan, flags);
462	if (desc <= 0)
463		return NULL;
464
465	/* Filling GenQQ CDB */
466	cdb = FSL_RE_PQ_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
467	cdb |= (src_cnt - 1) << FSL_RE_CDB_NRCS_SHIFT;
468	cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
469	cdb |= FSL_RE_BUFFER_OUTPUT << FSL_RE_CDB_BUFFER_SHIFT;
470	cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
471
472	pq = desc->cdb_addr;
473	pq->cdb32 = cdb;
474
475	p = pq->gfm_q1;
476	/* Init gfm_q1[] */
477	for (i = 0; i < src_cnt; i++)
478		p[i] = 1;
479
480	/* Align gfm[] to 32bit */
481	gfmq_len = ALIGN(src_cnt, 4);
482
483	/* Init gfm_q2[] */
484	p += gfmq_len;
485	for (i = 0; i < src_cnt; i++)
486		p[i] = scf[i];
487
488	/* Filling frame 0 of compound frame descriptor with CDB */
489	cf = desc->cf_addr;
490	fill_cfd_frame(cf, 0, sizeof(struct fsl_re_pq_cdb), desc->cdb_paddr, 0);
491
492	/* Fill CFD's 1st & 2nd frame with dest buffers */
493	for (i = 1, j = 0; i < 3; i++, j++)
494		fill_cfd_frame(cf, i, len, dest[j], 0);
495
496	/* Fill CFD's rest of the frames with source buffers */
497	for (i = 3, j = 0; j < save_src_cnt; i++, j++)
498		fill_cfd_frame(cf, i, len, src[j], 0);
499
500	/* PQ computation continuation */
501	if (flags & DMA_PREP_CONTINUE) {
502		if (src_cnt - save_src_cnt == 3) {
503			p[save_src_cnt] = 0;
504			p[save_src_cnt + 1] = 0;
505			p[save_src_cnt + 2] = 1;
506			fill_cfd_frame(cf, i++, len, dest[0], 0);
507			fill_cfd_frame(cf, i++, len, dest[1], 0);
508			fill_cfd_frame(cf, i++, len, dest[1], 0);
509		} else {
510			dev_err(re_chan->dev, "PQ tx continuation error!\n");
511			return NULL;
512		}
513	}
514
515	/* Setting the final bit in the last source buffer frame in CFD */
516	cf[i - 1].efrl32 |= 1 << FSL_RE_CF_FINAL_SHIFT;
517
518	return &desc->async_tx;
519}
520
521/*
522 * Prep function for memcpy. In RAID Engine, memcpy is done through MOVE
523 * command. Logic of this function will need to be modified once multipage
524 * support is added in Linux's MD/ASYNC Layer
525 */
526static struct dma_async_tx_descriptor *fsl_re_prep_dma_memcpy(
527		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
528		size_t len, unsigned long flags)
529{
530	struct fsl_re_chan *re_chan;
531	struct fsl_re_desc *desc;
532	size_t length;
533	struct fsl_re_cmpnd_frame *cf;
534	struct fsl_re_move_cdb *move;
535	u32 cdb;
536
537	re_chan = container_of(chan, struct fsl_re_chan, chan);
538
539	if (len > FSL_RE_MAX_DATA_LEN) {
540		dev_err(re_chan->dev, "cp tx length is %zu, max length is %d\n",
541			len, FSL_RE_MAX_DATA_LEN);
542		return NULL;
543	}
544
545	desc = fsl_re_chan_alloc_desc(re_chan, flags);
546	if (desc <= 0)
547		return NULL;
548
549	/* Filling move CDB */
550	cdb = FSL_RE_MOVE_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
551	cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
552	cdb |= FSL_RE_INTR_ON_ERROR << FSL_RE_CDB_ERROR_SHIFT;
553	cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
554
555	move = desc->cdb_addr;
556	move->cdb32 = cdb;
557
558	/* Filling frame 0 of CFD with move CDB */
559	cf = desc->cf_addr;
560	fill_cfd_frame(cf, 0, sizeof(*move), desc->cdb_paddr, 0);
561
562	length = min_t(size_t, len, FSL_RE_MAX_DATA_LEN);
563
564	/* Fill CFD's 1st frame with dest buffer */
565	fill_cfd_frame(cf, 1, length, dest, 0);
566
567	/* Fill CFD's 2nd frame with src buffer */
568	fill_cfd_frame(cf, 2, length, src, 1);
569
570	return &desc->async_tx;
571}
572
573static int fsl_re_alloc_chan_resources(struct dma_chan *chan)
574{
575	struct fsl_re_chan *re_chan;
576	struct fsl_re_desc *desc;
577	void *cf;
578	dma_addr_t paddr;
579	int i;
580
581	re_chan = container_of(chan, struct fsl_re_chan, chan);
582	for (i = 0; i < FSL_RE_MIN_DESCS; i++) {
583		desc = kzalloc(sizeof(*desc), GFP_KERNEL);
584		if (!desc)
585			break;
586
587		cf = dma_pool_alloc(re_chan->re_dev->cf_desc_pool, GFP_KERNEL,
588				    &paddr);
589		if (!cf) {
590			kfree(desc);
591			break;
592		}
593
594		INIT_LIST_HEAD(&desc->node);
595		fsl_re_init_desc(re_chan, desc, cf, paddr);
596
597		list_add_tail(&desc->node, &re_chan->free_q);
598		re_chan->alloc_count++;
599	}
600	return re_chan->alloc_count;
601}
602
603static void fsl_re_free_chan_resources(struct dma_chan *chan)
604{
605	struct fsl_re_chan *re_chan;
606	struct fsl_re_desc *desc;
607
608	re_chan = container_of(chan, struct fsl_re_chan, chan);
609	while (re_chan->alloc_count--) {
610		desc = list_first_entry(&re_chan->free_q,
611					struct fsl_re_desc,
612					node);
613
614		list_del(&desc->node);
615		dma_pool_free(re_chan->re_dev->cf_desc_pool, desc->cf_addr,
616			      desc->cf_paddr);
617		kfree(desc);
618	}
619
620	if (!list_empty(&re_chan->free_q))
621		dev_err(re_chan->dev, "chan resource cannot be cleaned!\n");
622}
623
624static int fsl_re_chan_probe(struct platform_device *ofdev,
625		      struct device_node *np, u8 q, u32 off)
626{
627	struct device *dev, *chandev;
628	struct fsl_re_drv_private *re_priv;
629	struct fsl_re_chan *chan;
630	struct dma_device *dma_dev;
631	u32 ptr;
632	u32 status;
633	int ret = 0, rc;
634	struct platform_device *chan_ofdev;
635
636	dev = &ofdev->dev;
637	re_priv = dev_get_drvdata(dev);
638	dma_dev = &re_priv->dma_dev;
639
640	chan = devm_kzalloc(dev, sizeof(*chan), GFP_KERNEL);
641	if (!chan)
642		return -ENOMEM;
643
644	/* create platform device for chan node */
645	chan_ofdev = of_platform_device_create(np, NULL, dev);
646	if (!chan_ofdev) {
647		dev_err(dev, "Not able to create ofdev for jr %d\n", q);
648		ret = -EINVAL;
649		goto err_free;
650	}
651
652	/* read reg property from dts */
653	rc = of_property_read_u32(np, "reg", &ptr);
654	if (rc) {
655		dev_err(dev, "Reg property not found in jr %d\n", q);
656		ret = -ENODEV;
657		goto err_free;
658	}
659
660	chan->jrregs = (struct fsl_re_chan_cfg *)((u8 *)re_priv->re_regs +
661			off + ptr);
662
663	/* read irq property from dts */
664	chan->irq = irq_of_parse_and_map(np, 0);
665	if (!chan->irq) {
666		dev_err(dev, "No IRQ defined for JR %d\n", q);
667		ret = -ENODEV;
668		goto err_free;
669	}
670
671	snprintf(chan->name, sizeof(chan->name), "re_jr%02d", q);
672
673	chandev = &chan_ofdev->dev;
674	tasklet_init(&chan->irqtask, fsl_re_dequeue, (unsigned long)chandev);
675
676	ret = request_irq(chan->irq, fsl_re_isr, 0, chan->name, chandev);
677	if (ret) {
678		dev_err(dev, "Unable to register interrupt for JR %d\n", q);
679		ret = -EINVAL;
680		goto err_free;
681	}
682
683	re_priv->re_jrs[q] = chan;
684	chan->chan.device = dma_dev;
685	chan->chan.private = chan;
686	chan->dev = chandev;
687	chan->re_dev = re_priv;
688
689	spin_lock_init(&chan->desc_lock);
690	INIT_LIST_HEAD(&chan->ack_q);
691	INIT_LIST_HEAD(&chan->active_q);
692	INIT_LIST_HEAD(&chan->submit_q);
693	INIT_LIST_HEAD(&chan->free_q);
694
695	chan->inb_ring_virt_addr = dma_pool_alloc(chan->re_dev->hw_desc_pool,
696		GFP_KERNEL, &chan->inb_phys_addr);
697	if (!chan->inb_ring_virt_addr) {
698		dev_err(dev, "No dma memory for inb_ring_virt_addr\n");
699		ret = -ENOMEM;
700		goto err_free;
701	}
702
703	chan->oub_ring_virt_addr = dma_pool_alloc(chan->re_dev->hw_desc_pool,
704		GFP_KERNEL, &chan->oub_phys_addr);
705	if (!chan->oub_ring_virt_addr) {
706		dev_err(dev, "No dma memory for oub_ring_virt_addr\n");
707		ret = -ENOMEM;
708		goto err_free_1;
709	}
710
711	/* Program the Inbound/Outbound ring base addresses and size */
712	out_be32(&chan->jrregs->inbring_base_h,
713		 chan->inb_phys_addr & FSL_RE_ADDR_BIT_MASK);
714	out_be32(&chan->jrregs->oubring_base_h,
715		 chan->oub_phys_addr & FSL_RE_ADDR_BIT_MASK);
716	out_be32(&chan->jrregs->inbring_base_l,
717		 chan->inb_phys_addr >> FSL_RE_ADDR_BIT_SHIFT);
718	out_be32(&chan->jrregs->oubring_base_l,
719		 chan->oub_phys_addr >> FSL_RE_ADDR_BIT_SHIFT);
720	out_be32(&chan->jrregs->inbring_size,
721		 FSL_RE_RING_SIZE << FSL_RE_RING_SIZE_SHIFT);
722	out_be32(&chan->jrregs->oubring_size,
723		 FSL_RE_RING_SIZE << FSL_RE_RING_SIZE_SHIFT);
724
725	/* Read LIODN value from u-boot */
726	status = in_be32(&chan->jrregs->jr_config_1) & FSL_RE_REG_LIODN_MASK;
727
728	/* Program the CFG reg */
729	out_be32(&chan->jrregs->jr_config_1,
730		 FSL_RE_CFG1_CBSI | FSL_RE_CFG1_CBS0 | status);
731
732	dev_set_drvdata(chandev, chan);
733
734	/* Enable RE/CHAN */
735	out_be32(&chan->jrregs->jr_command, FSL_RE_ENABLE);
736
737	return 0;
738
739err_free_1:
740	dma_pool_free(chan->re_dev->hw_desc_pool, chan->inb_ring_virt_addr,
741		      chan->inb_phys_addr);
742err_free:
743	return ret;
744}
745
746/* Probe function for RAID Engine */
747static int fsl_re_probe(struct platform_device *ofdev)
748{
749	struct fsl_re_drv_private *re_priv;
750	struct device_node *np;
751	struct device_node *child;
752	u32 off;
753	u8 ridx = 0;
754	struct dma_device *dma_dev;
755	struct resource *res;
756	int rc;
757	struct device *dev = &ofdev->dev;
758
759	re_priv = devm_kzalloc(dev, sizeof(*re_priv), GFP_KERNEL);
760	if (!re_priv)
761		return -ENOMEM;
762
763	res = platform_get_resource(ofdev, IORESOURCE_MEM, 0);
764	if (!res)
765		return -ENODEV;
766
767	/* IOMAP the entire RAID Engine region */
768	re_priv->re_regs = devm_ioremap(dev, res->start, resource_size(res));
769	if (!re_priv->re_regs)
770		return -EBUSY;
771
772	/* Program the RE mode */
773	out_be32(&re_priv->re_regs->global_config, FSL_RE_NON_DPAA_MODE);
774
775	/* Program Galois Field polynomial */
776	out_be32(&re_priv->re_regs->galois_field_config, FSL_RE_GFM_POLY);
777
778	dev_info(dev, "version %x, mode %x, gfp %x\n",
779		 in_be32(&re_priv->re_regs->re_version_id),
780		 in_be32(&re_priv->re_regs->global_config),
781		 in_be32(&re_priv->re_regs->galois_field_config));
782
783	dma_dev = &re_priv->dma_dev;
784	dma_dev->dev = dev;
785	INIT_LIST_HEAD(&dma_dev->channels);
786	dma_set_mask(dev, DMA_BIT_MASK(40));
787
788	dma_dev->device_alloc_chan_resources = fsl_re_alloc_chan_resources;
789	dma_dev->device_tx_status = fsl_re_tx_status;
790	dma_dev->device_issue_pending = fsl_re_issue_pending;
791
792	dma_dev->max_xor = FSL_RE_MAX_XOR_SRCS;
793	dma_dev->device_prep_dma_xor = fsl_re_prep_dma_xor;
794	dma_cap_set(DMA_XOR, dma_dev->cap_mask);
795
796	dma_dev->max_pq = FSL_RE_MAX_PQ_SRCS;
797	dma_dev->device_prep_dma_pq = fsl_re_prep_dma_pq;
798	dma_cap_set(DMA_PQ, dma_dev->cap_mask);
799
800	dma_dev->device_prep_dma_memcpy = fsl_re_prep_dma_memcpy;
801	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
802
803	dma_dev->device_free_chan_resources = fsl_re_free_chan_resources;
804
805	re_priv->total_chans = 0;
806
807	re_priv->cf_desc_pool = dmam_pool_create("fsl_re_cf_desc_pool", dev,
808					FSL_RE_CF_CDB_SIZE,
809					FSL_RE_CF_CDB_ALIGN, 0);
810
811	if (!re_priv->cf_desc_pool) {
812		dev_err(dev, "No memory for fsl re_cf desc pool\n");
813		return -ENOMEM;
814	}
815
816	re_priv->hw_desc_pool = dmam_pool_create("fsl_re_hw_desc_pool", dev,
817			sizeof(struct fsl_re_hw_desc) * FSL_RE_RING_SIZE,
818			FSL_RE_FRAME_ALIGN, 0);
819	if (!re_priv->hw_desc_pool) {
820		dev_err(dev, "No memory for fsl re_hw desc pool\n");
821		return -ENOMEM;
822	}
823
824	dev_set_drvdata(dev, re_priv);
825
826	/* Parse Device tree to find out the total number of JQs present */
827	for_each_compatible_node(np, NULL, "fsl,raideng-v1.0-job-queue") {
828		rc = of_property_read_u32(np, "reg", &off);
829		if (rc) {
830			dev_err(dev, "Reg property not found in JQ node\n");
831			of_node_put(np);
832			return -ENODEV;
833		}
834		/* Find out the Job Rings present under each JQ */
835		for_each_child_of_node(np, child) {
836			rc = of_device_is_compatible(child,
837					     "fsl,raideng-v1.0-job-ring");
838			if (rc) {
839				fsl_re_chan_probe(ofdev, child, ridx++, off);
840				re_priv->total_chans++;
841			}
842		}
843	}
844
845	dma_async_device_register(dma_dev);
846
847	return 0;
848}
849
850static void fsl_re_remove_chan(struct fsl_re_chan *chan)
851{
852	tasklet_kill(&chan->irqtask);
853
854	dma_pool_free(chan->re_dev->hw_desc_pool, chan->inb_ring_virt_addr,
855		      chan->inb_phys_addr);
856
857	dma_pool_free(chan->re_dev->hw_desc_pool, chan->oub_ring_virt_addr,
858		      chan->oub_phys_addr);
859}
860
861static int fsl_re_remove(struct platform_device *ofdev)
862{
863	struct fsl_re_drv_private *re_priv;
864	struct device *dev;
865	int i;
866
867	dev = &ofdev->dev;
868	re_priv = dev_get_drvdata(dev);
869
870	/* Cleanup chan related memory areas */
871	for (i = 0; i < re_priv->total_chans; i++)
872		fsl_re_remove_chan(re_priv->re_jrs[i]);
873
874	/* Unregister the driver */
875	dma_async_device_unregister(&re_priv->dma_dev);
876
877	return 0;
878}
879
880static const struct of_device_id fsl_re_ids[] = {
881	{ .compatible = "fsl,raideng-v1.0", },
882	{}
883};
884MODULE_DEVICE_TABLE(of, fsl_re_ids);
885
886static struct platform_driver fsl_re_driver = {
887	.driver = {
888		.name = "fsl-raideng",
889		.of_match_table = fsl_re_ids,
890	},
891	.probe = fsl_re_probe,
892	.remove = fsl_re_remove,
893};
894
895module_platform_driver(fsl_re_driver);
896
897MODULE_AUTHOR("Harninder Rai <harninder.rai@freescale.com>");
898MODULE_LICENSE("GPL v2");
899MODULE_DESCRIPTION("Freescale RAID Engine Device Driver");