xref: /linux/drivers/dma/fsl_raid.c (revision 37b24b50)
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.h>
64 #include <linux/of_irq.h>
65 #include <linux/of_platform.h>
66 #include <linux/platform_device.h>
67 #include <linux/dma-mapping.h>
68 #include <linux/dmapool.h>
69 #include <linux/dmaengine.h>
70 #include <linux/io.h>
71 #include <linux/spinlock.h>
72 #include <linux/slab.h>
73 
74 #include "dmaengine.h"
75 #include "fsl_raid.h"
76 
77 #define FSL_RE_MAX_XOR_SRCS	16
78 #define FSL_RE_MAX_PQ_SRCS	16
79 #define FSL_RE_MIN_DESCS	256
80 #define FSL_RE_MAX_DESCS	(4 * FSL_RE_MIN_DESCS)
81 #define FSL_RE_FRAME_FORMAT	0x1
82 #define FSL_RE_MAX_DATA_LEN	(1024*1024)
83 
84 #define to_fsl_re_dma_desc(tx) container_of(tx, struct fsl_re_desc, async_tx)
85 
86 /* Add descriptors into per chan software queue - submit_q */
fsl_re_tx_submit(struct dma_async_tx_descriptor * tx)87 static dma_cookie_t fsl_re_tx_submit(struct dma_async_tx_descriptor *tx)
88 {
89 	struct fsl_re_desc *desc;
90 	struct fsl_re_chan *re_chan;
91 	dma_cookie_t cookie;
92 	unsigned long flags;
93 
94 	desc = to_fsl_re_dma_desc(tx);
95 	re_chan = container_of(tx->chan, struct fsl_re_chan, chan);
96 
97 	spin_lock_irqsave(&re_chan->desc_lock, flags);
98 	cookie = dma_cookie_assign(tx);
99 	list_add_tail(&desc->node, &re_chan->submit_q);
100 	spin_unlock_irqrestore(&re_chan->desc_lock, flags);
101 
102 	return cookie;
103 }
104 
105 /* Copy descriptor from per chan software queue into hardware job ring */
fsl_re_issue_pending(struct dma_chan * chan)106 static void fsl_re_issue_pending(struct dma_chan *chan)
107 {
108 	struct fsl_re_chan *re_chan;
109 	int avail;
110 	struct fsl_re_desc *desc, *_desc;
111 	unsigned long flags;
112 
113 	re_chan = container_of(chan, struct fsl_re_chan, chan);
114 
115 	spin_lock_irqsave(&re_chan->desc_lock, flags);
116 	avail = FSL_RE_SLOT_AVAIL(
117 		in_be32(&re_chan->jrregs->inbring_slot_avail));
118 
119 	list_for_each_entry_safe(desc, _desc, &re_chan->submit_q, node) {
120 		if (!avail)
121 			break;
122 
123 		list_move_tail(&desc->node, &re_chan->active_q);
124 
125 		memcpy(&re_chan->inb_ring_virt_addr[re_chan->inb_count],
126 		       &desc->hwdesc, sizeof(struct fsl_re_hw_desc));
127 
128 		re_chan->inb_count = (re_chan->inb_count + 1) &
129 						FSL_RE_RING_SIZE_MASK;
130 		out_be32(&re_chan->jrregs->inbring_add_job, FSL_RE_ADD_JOB(1));
131 		avail--;
132 	}
133 	spin_unlock_irqrestore(&re_chan->desc_lock, flags);
134 }
135 
fsl_re_desc_done(struct fsl_re_desc * desc)136 static void fsl_re_desc_done(struct fsl_re_desc *desc)
137 {
138 	dma_cookie_complete(&desc->async_tx);
139 	dma_descriptor_unmap(&desc->async_tx);
140 	dmaengine_desc_get_callback_invoke(&desc->async_tx, NULL);
141 }
142 
fsl_re_cleanup_descs(struct fsl_re_chan * re_chan)143 static void fsl_re_cleanup_descs(struct fsl_re_chan *re_chan)
144 {
145 	struct fsl_re_desc *desc, *_desc;
146 	unsigned long flags;
147 
148 	spin_lock_irqsave(&re_chan->desc_lock, flags);
149 	list_for_each_entry_safe(desc, _desc, &re_chan->ack_q, node) {
150 		if (async_tx_test_ack(&desc->async_tx))
151 			list_move_tail(&desc->node, &re_chan->free_q);
152 	}
153 	spin_unlock_irqrestore(&re_chan->desc_lock, flags);
154 
155 	fsl_re_issue_pending(&re_chan->chan);
156 }
157 
fsl_re_dequeue(struct tasklet_struct * t)158 static void fsl_re_dequeue(struct tasklet_struct *t)
159 {
160 	struct fsl_re_chan *re_chan = from_tasklet(re_chan, t, irqtask);
161 	struct fsl_re_desc *desc, *_desc;
162 	struct fsl_re_hw_desc *hwdesc;
163 	unsigned long flags;
164 	unsigned int count, oub_count;
165 	int found;
166 
167 	fsl_re_cleanup_descs(re_chan);
168 
169 	spin_lock_irqsave(&re_chan->desc_lock, flags);
170 	count =	FSL_RE_SLOT_FULL(in_be32(&re_chan->jrregs->oubring_slot_full));
171 	while (count--) {
172 		found = 0;
173 		hwdesc = &re_chan->oub_ring_virt_addr[re_chan->oub_count];
174 		list_for_each_entry_safe(desc, _desc, &re_chan->active_q,
175 					 node) {
176 			/* compare the hw dma addr to find the completed */
177 			if (desc->hwdesc.lbea32 == hwdesc->lbea32 &&
178 			    desc->hwdesc.addr_low == hwdesc->addr_low) {
179 				found = 1;
180 				break;
181 			}
182 		}
183 
184 		if (found) {
185 			fsl_re_desc_done(desc);
186 			list_move_tail(&desc->node, &re_chan->ack_q);
187 		} else {
188 			dev_err(re_chan->dev,
189 				"found hwdesc not in sw queue, discard it\n");
190 		}
191 
192 		oub_count = (re_chan->oub_count + 1) & FSL_RE_RING_SIZE_MASK;
193 		re_chan->oub_count = oub_count;
194 
195 		out_be32(&re_chan->jrregs->oubring_job_rmvd,
196 			 FSL_RE_RMVD_JOB(1));
197 	}
198 	spin_unlock_irqrestore(&re_chan->desc_lock, flags);
199 }
200 
201 /* Per Job Ring interrupt handler */
fsl_re_isr(int irq,void * data)202 static irqreturn_t fsl_re_isr(int irq, void *data)
203 {
204 	struct fsl_re_chan *re_chan;
205 	u32 irqstate, status;
206 
207 	re_chan = dev_get_drvdata((struct device *)data);
208 
209 	irqstate = in_be32(&re_chan->jrregs->jr_interrupt_status);
210 	if (!irqstate)
211 		return IRQ_NONE;
212 
213 	/*
214 	 * There's no way in upper layer (read MD layer) to recover from
215 	 * error conditions except restart everything. In long term we
216 	 * need to do something more than just crashing
217 	 */
218 	if (irqstate & FSL_RE_ERROR) {
219 		status = in_be32(&re_chan->jrregs->jr_status);
220 		dev_err(re_chan->dev, "chan error irqstate: %x, status: %x\n",
221 			irqstate, status);
222 	}
223 
224 	/* Clear interrupt */
225 	out_be32(&re_chan->jrregs->jr_interrupt_status, FSL_RE_CLR_INTR);
226 
227 	tasklet_schedule(&re_chan->irqtask);
228 
229 	return IRQ_HANDLED;
230 }
231 
fsl_re_tx_status(struct dma_chan * chan,dma_cookie_t cookie,struct dma_tx_state * txstate)232 static enum dma_status fsl_re_tx_status(struct dma_chan *chan,
233 					dma_cookie_t cookie,
234 					struct dma_tx_state *txstate)
235 {
236 	return dma_cookie_status(chan, cookie, txstate);
237 }
238 
fill_cfd_frame(struct fsl_re_cmpnd_frame * cf,u8 index,size_t length,dma_addr_t addr,bool final)239 static void fill_cfd_frame(struct fsl_re_cmpnd_frame *cf, u8 index,
240 			   size_t length, dma_addr_t addr, bool final)
241 {
242 	u32 efrl = length & FSL_RE_CF_LENGTH_MASK;
243 
244 	efrl |= final << FSL_RE_CF_FINAL_SHIFT;
245 	cf[index].efrl32 = efrl;
246 	cf[index].addr_high = upper_32_bits(addr);
247 	cf[index].addr_low = lower_32_bits(addr);
248 }
249 
fsl_re_init_desc(struct fsl_re_chan * re_chan,struct fsl_re_desc * desc,void * cf,dma_addr_t paddr)250 static struct fsl_re_desc *fsl_re_init_desc(struct fsl_re_chan *re_chan,
251 					    struct fsl_re_desc *desc,
252 					    void *cf, dma_addr_t paddr)
253 {
254 	desc->re_chan = re_chan;
255 	desc->async_tx.tx_submit = fsl_re_tx_submit;
256 	dma_async_tx_descriptor_init(&desc->async_tx, &re_chan->chan);
257 	INIT_LIST_HEAD(&desc->node);
258 
259 	desc->hwdesc.fmt32 = FSL_RE_FRAME_FORMAT << FSL_RE_HWDESC_FMT_SHIFT;
260 	desc->hwdesc.lbea32 = upper_32_bits(paddr);
261 	desc->hwdesc.addr_low = lower_32_bits(paddr);
262 	desc->cf_addr = cf;
263 	desc->cf_paddr = paddr;
264 
265 	desc->cdb_addr = (void *)(cf + FSL_RE_CF_DESC_SIZE);
266 	desc->cdb_paddr = paddr + FSL_RE_CF_DESC_SIZE;
267 
268 	return desc;
269 }
270 
fsl_re_chan_alloc_desc(struct fsl_re_chan * re_chan,unsigned long flags)271 static struct fsl_re_desc *fsl_re_chan_alloc_desc(struct fsl_re_chan *re_chan,
272 						  unsigned long flags)
273 {
274 	struct fsl_re_desc *desc = NULL;
275 	void *cf;
276 	dma_addr_t paddr;
277 	unsigned long lock_flag;
278 
279 	fsl_re_cleanup_descs(re_chan);
280 
281 	spin_lock_irqsave(&re_chan->desc_lock, lock_flag);
282 	if (!list_empty(&re_chan->free_q)) {
283 		/* take one desc from free_q */
284 		desc = list_first_entry(&re_chan->free_q,
285 					struct fsl_re_desc, node);
286 		list_del(&desc->node);
287 
288 		desc->async_tx.flags = flags;
289 	}
290 	spin_unlock_irqrestore(&re_chan->desc_lock, lock_flag);
291 
292 	if (!desc) {
293 		desc = kzalloc(sizeof(*desc), GFP_NOWAIT);
294 		if (!desc)
295 			return NULL;
296 
297 		cf = dma_pool_alloc(re_chan->re_dev->cf_desc_pool, GFP_NOWAIT,
298 				    &paddr);
299 		if (!cf) {
300 			kfree(desc);
301 			return NULL;
302 		}
303 
304 		desc = fsl_re_init_desc(re_chan, desc, cf, paddr);
305 		desc->async_tx.flags = flags;
306 
307 		spin_lock_irqsave(&re_chan->desc_lock, lock_flag);
308 		re_chan->alloc_count++;
309 		spin_unlock_irqrestore(&re_chan->desc_lock, lock_flag);
310 	}
311 
312 	return desc;
313 }
314 
fsl_re_prep_dma_genq(struct dma_chan * chan,dma_addr_t dest,dma_addr_t * src,unsigned int src_cnt,const unsigned char * scf,size_t len,unsigned long flags)315 static struct dma_async_tx_descriptor *fsl_re_prep_dma_genq(
316 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
317 		unsigned int src_cnt, const unsigned char *scf, size_t len,
318 		unsigned long flags)
319 {
320 	struct fsl_re_chan *re_chan;
321 	struct fsl_re_desc *desc;
322 	struct fsl_re_xor_cdb *xor;
323 	struct fsl_re_cmpnd_frame *cf;
324 	u32 cdb;
325 	unsigned int i, j;
326 	unsigned int save_src_cnt = src_cnt;
327 	int cont_q = 0;
328 
329 	re_chan = container_of(chan, struct fsl_re_chan, chan);
330 	if (len > FSL_RE_MAX_DATA_LEN) {
331 		dev_err(re_chan->dev, "genq tx length %zu, max length %d\n",
332 			len, FSL_RE_MAX_DATA_LEN);
333 		return NULL;
334 	}
335 
336 	desc = fsl_re_chan_alloc_desc(re_chan, flags);
337 	if (desc <= 0)
338 		return NULL;
339 
340 	if (scf && (flags & DMA_PREP_CONTINUE)) {
341 		cont_q = 1;
342 		src_cnt += 1;
343 	}
344 
345 	/* Filling xor CDB */
346 	cdb = FSL_RE_XOR_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
347 	cdb |= (src_cnt - 1) << FSL_RE_CDB_NRCS_SHIFT;
348 	cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
349 	cdb |= FSL_RE_INTR_ON_ERROR << FSL_RE_CDB_ERROR_SHIFT;
350 	cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
351 	xor = desc->cdb_addr;
352 	xor->cdb32 = cdb;
353 
354 	if (scf) {
355 		/* compute q = src0*coef0^src1*coef1^..., * is GF(8) mult */
356 		for (i = 0; i < save_src_cnt; i++)
357 			xor->gfm[i] = scf[i];
358 		if (cont_q)
359 			xor->gfm[i++] = 1;
360 	} else {
361 		/* compute P, that is XOR all srcs */
362 		for (i = 0; i < src_cnt; i++)
363 			xor->gfm[i] = 1;
364 	}
365 
366 	/* Filling frame 0 of compound frame descriptor with CDB */
367 	cf = desc->cf_addr;
368 	fill_cfd_frame(cf, 0, sizeof(*xor), desc->cdb_paddr, 0);
369 
370 	/* Fill CFD's 1st frame with dest buffer */
371 	fill_cfd_frame(cf, 1, len, dest, 0);
372 
373 	/* Fill CFD's rest of the frames with source buffers */
374 	for (i = 2, j = 0; j < save_src_cnt; i++, j++)
375 		fill_cfd_frame(cf, i, len, src[j], 0);
376 
377 	if (cont_q)
378 		fill_cfd_frame(cf, i++, len, dest, 0);
379 
380 	/* Setting the final bit in the last source buffer frame in CFD */
381 	cf[i - 1].efrl32 |= 1 << FSL_RE_CF_FINAL_SHIFT;
382 
383 	return &desc->async_tx;
384 }
385 
386 /*
387  * Prep function for P parity calculation.In RAID Engine terminology,
388  * XOR calculation is called GenQ calculation done through GenQ command
389  */
fsl_re_prep_dma_xor(struct dma_chan * chan,dma_addr_t dest,dma_addr_t * src,unsigned int src_cnt,size_t len,unsigned long flags)390 static struct dma_async_tx_descriptor *fsl_re_prep_dma_xor(
391 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
392 		unsigned int src_cnt, size_t len, unsigned long flags)
393 {
394 	/* NULL let genq take all coef as 1 */
395 	return fsl_re_prep_dma_genq(chan, dest, src, src_cnt, NULL, len, flags);
396 }
397 
398 /*
399  * Prep function for P/Q parity calculation.In RAID Engine terminology,
400  * P/Q calculation is called GenQQ done through GenQQ command
401  */
fsl_re_prep_dma_pq(struct dma_chan * chan,dma_addr_t * dest,dma_addr_t * src,unsigned int src_cnt,const unsigned char * scf,size_t len,unsigned long flags)402 static struct dma_async_tx_descriptor *fsl_re_prep_dma_pq(
403 		struct dma_chan *chan, dma_addr_t *dest, dma_addr_t *src,
404 		unsigned int src_cnt, const unsigned char *scf, size_t len,
405 		unsigned long flags)
406 {
407 	struct fsl_re_chan *re_chan;
408 	struct fsl_re_desc *desc;
409 	struct fsl_re_pq_cdb *pq;
410 	struct fsl_re_cmpnd_frame *cf;
411 	u32 cdb;
412 	u8 *p;
413 	int gfmq_len, i, j;
414 	unsigned int save_src_cnt = src_cnt;
415 
416 	re_chan = container_of(chan, struct fsl_re_chan, chan);
417 	if (len > FSL_RE_MAX_DATA_LEN) {
418 		dev_err(re_chan->dev, "pq tx length is %zu, max length is %d\n",
419 			len, FSL_RE_MAX_DATA_LEN);
420 		return NULL;
421 	}
422 
423 	/*
424 	 * RE requires at least 2 sources, if given only one source, we pass the
425 	 * second source same as the first one.
426 	 * With only one source, generating P is meaningless, only generate Q.
427 	 */
428 	if (src_cnt == 1) {
429 		struct dma_async_tx_descriptor *tx;
430 		dma_addr_t dma_src[2];
431 		unsigned char coef[2];
432 
433 		dma_src[0] = *src;
434 		coef[0] = *scf;
435 		dma_src[1] = *src;
436 		coef[1] = 0;
437 		tx = fsl_re_prep_dma_genq(chan, dest[1], dma_src, 2, coef, len,
438 					  flags);
439 		if (tx)
440 			desc = to_fsl_re_dma_desc(tx);
441 
442 		return tx;
443 	}
444 
445 	/*
446 	 * During RAID6 array creation, Linux's MD layer gets P and Q
447 	 * calculated separately in two steps. But our RAID Engine has
448 	 * the capability to calculate both P and Q with a single command
449 	 * Hence to merge well with MD layer, we need to provide a hook
450 	 * here and call re_jq_prep_dma_genq() function
451 	 */
452 
453 	if (flags & DMA_PREP_PQ_DISABLE_P)
454 		return fsl_re_prep_dma_genq(chan, dest[1], src, src_cnt,
455 				scf, len, flags);
456 
457 	if (flags & DMA_PREP_CONTINUE)
458 		src_cnt += 3;
459 
460 	desc = fsl_re_chan_alloc_desc(re_chan, flags);
461 	if (desc <= 0)
462 		return NULL;
463 
464 	/* Filling GenQQ CDB */
465 	cdb = FSL_RE_PQ_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
466 	cdb |= (src_cnt - 1) << FSL_RE_CDB_NRCS_SHIFT;
467 	cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
468 	cdb |= FSL_RE_BUFFER_OUTPUT << FSL_RE_CDB_BUFFER_SHIFT;
469 	cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
470 
471 	pq = desc->cdb_addr;
472 	pq->cdb32 = cdb;
473 
474 	p = pq->gfm_q1;
475 	/* Init gfm_q1[] */
476 	for (i = 0; i < src_cnt; i++)
477 		p[i] = 1;
478 
479 	/* Align gfm[] to 32bit */
480 	gfmq_len = ALIGN(src_cnt, 4);
481 
482 	/* Init gfm_q2[] */
483 	p += gfmq_len;
484 	for (i = 0; i < src_cnt; i++)
485 		p[i] = scf[i];
486 
487 	/* Filling frame 0 of compound frame descriptor with CDB */
488 	cf = desc->cf_addr;
489 	fill_cfd_frame(cf, 0, sizeof(struct fsl_re_pq_cdb), desc->cdb_paddr, 0);
490 
491 	/* Fill CFD's 1st & 2nd frame with dest buffers */
492 	for (i = 1, j = 0; i < 3; i++, j++)
493 		fill_cfd_frame(cf, i, len, dest[j], 0);
494 
495 	/* Fill CFD's rest of the frames with source buffers */
496 	for (i = 3, j = 0; j < save_src_cnt; i++, j++)
497 		fill_cfd_frame(cf, i, len, src[j], 0);
498 
499 	/* PQ computation continuation */
500 	if (flags & DMA_PREP_CONTINUE) {
501 		if (src_cnt - save_src_cnt == 3) {
502 			p[save_src_cnt] = 0;
503 			p[save_src_cnt + 1] = 0;
504 			p[save_src_cnt + 2] = 1;
505 			fill_cfd_frame(cf, i++, len, dest[0], 0);
506 			fill_cfd_frame(cf, i++, len, dest[1], 0);
507 			fill_cfd_frame(cf, i++, len, dest[1], 0);
508 		} else {
509 			dev_err(re_chan->dev, "PQ tx continuation error!\n");
510 			return NULL;
511 		}
512 	}
513 
514 	/* Setting the final bit in the last source buffer frame in CFD */
515 	cf[i - 1].efrl32 |= 1 << FSL_RE_CF_FINAL_SHIFT;
516 
517 	return &desc->async_tx;
518 }
519 
520 /*
521  * Prep function for memcpy. In RAID Engine, memcpy is done through MOVE
522  * command. Logic of this function will need to be modified once multipage
523  * support is added in Linux's MD/ASYNC Layer
524  */
fsl_re_prep_dma_memcpy(struct dma_chan * chan,dma_addr_t dest,dma_addr_t src,size_t len,unsigned long flags)525 static struct dma_async_tx_descriptor *fsl_re_prep_dma_memcpy(
526 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
527 		size_t len, unsigned long flags)
528 {
529 	struct fsl_re_chan *re_chan;
530 	struct fsl_re_desc *desc;
531 	size_t length;
532 	struct fsl_re_cmpnd_frame *cf;
533 	struct fsl_re_move_cdb *move;
534 	u32 cdb;
535 
536 	re_chan = container_of(chan, struct fsl_re_chan, chan);
537 
538 	if (len > FSL_RE_MAX_DATA_LEN) {
539 		dev_err(re_chan->dev, "cp tx length is %zu, max length is %d\n",
540 			len, FSL_RE_MAX_DATA_LEN);
541 		return NULL;
542 	}
543 
544 	desc = fsl_re_chan_alloc_desc(re_chan, flags);
545 	if (desc <= 0)
546 		return NULL;
547 
548 	/* Filling move CDB */
549 	cdb = FSL_RE_MOVE_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
550 	cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
551 	cdb |= FSL_RE_INTR_ON_ERROR << FSL_RE_CDB_ERROR_SHIFT;
552 	cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
553 
554 	move = desc->cdb_addr;
555 	move->cdb32 = cdb;
556 
557 	/* Filling frame 0 of CFD with move CDB */
558 	cf = desc->cf_addr;
559 	fill_cfd_frame(cf, 0, sizeof(*move), desc->cdb_paddr, 0);
560 
561 	length = min_t(size_t, len, FSL_RE_MAX_DATA_LEN);
562 
563 	/* Fill CFD's 1st frame with dest buffer */
564 	fill_cfd_frame(cf, 1, length, dest, 0);
565 
566 	/* Fill CFD's 2nd frame with src buffer */
567 	fill_cfd_frame(cf, 2, length, src, 1);
568 
569 	return &desc->async_tx;
570 }
571 
fsl_re_alloc_chan_resources(struct dma_chan * chan)572 static int fsl_re_alloc_chan_resources(struct dma_chan *chan)
573 {
574 	struct fsl_re_chan *re_chan;
575 	struct fsl_re_desc *desc;
576 	void *cf;
577 	dma_addr_t paddr;
578 	int i;
579 
580 	re_chan = container_of(chan, struct fsl_re_chan, chan);
581 	for (i = 0; i < FSL_RE_MIN_DESCS; i++) {
582 		desc = kzalloc(sizeof(*desc), GFP_KERNEL);
583 		if (!desc)
584 			break;
585 
586 		cf = dma_pool_alloc(re_chan->re_dev->cf_desc_pool, GFP_KERNEL,
587 				    &paddr);
588 		if (!cf) {
589 			kfree(desc);
590 			break;
591 		}
592 
593 		INIT_LIST_HEAD(&desc->node);
594 		fsl_re_init_desc(re_chan, desc, cf, paddr);
595 
596 		list_add_tail(&desc->node, &re_chan->free_q);
597 		re_chan->alloc_count++;
598 	}
599 	return re_chan->alloc_count;
600 }
601 
fsl_re_free_chan_resources(struct dma_chan * chan)602 static void fsl_re_free_chan_resources(struct dma_chan *chan)
603 {
604 	struct fsl_re_chan *re_chan;
605 	struct fsl_re_desc *desc;
606 
607 	re_chan = container_of(chan, struct fsl_re_chan, chan);
608 	while (re_chan->alloc_count--) {
609 		desc = list_first_entry(&re_chan->free_q,
610 					struct fsl_re_desc,
611 					node);
612 
613 		list_del(&desc->node);
614 		dma_pool_free(re_chan->re_dev->cf_desc_pool, desc->cf_addr,
615 			      desc->cf_paddr);
616 		kfree(desc);
617 	}
618 
619 	if (!list_empty(&re_chan->free_q))
620 		dev_err(re_chan->dev, "chan resource cannot be cleaned!\n");
621 }
622 
fsl_re_chan_probe(struct platform_device * ofdev,struct device_node * np,u8 q,u32 off)623 static int fsl_re_chan_probe(struct platform_device *ofdev,
624 		      struct device_node *np, u8 q, u32 off)
625 {
626 	struct device *dev, *chandev;
627 	struct fsl_re_drv_private *re_priv;
628 	struct fsl_re_chan *chan;
629 	struct dma_device *dma_dev;
630 	u32 ptr;
631 	u32 status;
632 	int ret = 0, rc;
633 	struct platform_device *chan_ofdev;
634 
635 	dev = &ofdev->dev;
636 	re_priv = dev_get_drvdata(dev);
637 	dma_dev = &re_priv->dma_dev;
638 
639 	chan = devm_kzalloc(dev, sizeof(*chan), GFP_KERNEL);
640 	if (!chan)
641 		return -ENOMEM;
642 
643 	/* create platform device for chan node */
644 	chan_ofdev = of_platform_device_create(np, NULL, dev);
645 	if (!chan_ofdev) {
646 		dev_err(dev, "Not able to create ofdev for jr %d\n", q);
647 		ret = -EINVAL;
648 		goto err_free;
649 	}
650 
651 	/* read reg property from dts */
652 	rc = of_property_read_u32(np, "reg", &ptr);
653 	if (rc) {
654 		dev_err(dev, "Reg property not found in jr %d\n", q);
655 		ret = -ENODEV;
656 		goto err_free;
657 	}
658 
659 	chan->jrregs = (struct fsl_re_chan_cfg *)((u8 *)re_priv->re_regs +
660 			off + ptr);
661 
662 	/* read irq property from dts */
663 	chan->irq = irq_of_parse_and_map(np, 0);
664 	if (!chan->irq) {
665 		dev_err(dev, "No IRQ defined for JR %d\n", q);
666 		ret = -ENODEV;
667 		goto err_free;
668 	}
669 
670 	snprintf(chan->name, sizeof(chan->name), "re_jr%02d", q);
671 
672 	chandev = &chan_ofdev->dev;
673 	tasklet_setup(&chan->irqtask, fsl_re_dequeue);
674 
675 	ret = request_irq(chan->irq, fsl_re_isr, 0, chan->name, chandev);
676 	if (ret) {
677 		dev_err(dev, "Unable to register interrupt for JR %d\n", q);
678 		ret = -EINVAL;
679 		goto err_free;
680 	}
681 
682 	re_priv->re_jrs[q] = chan;
683 	chan->chan.device = dma_dev;
684 	chan->chan.private = chan;
685 	chan->dev = chandev;
686 	chan->re_dev = re_priv;
687 
688 	spin_lock_init(&chan->desc_lock);
689 	INIT_LIST_HEAD(&chan->ack_q);
690 	INIT_LIST_HEAD(&chan->active_q);
691 	INIT_LIST_HEAD(&chan->submit_q);
692 	INIT_LIST_HEAD(&chan->free_q);
693 
694 	chan->inb_ring_virt_addr = dma_pool_alloc(chan->re_dev->hw_desc_pool,
695 		GFP_KERNEL, &chan->inb_phys_addr);
696 	if (!chan->inb_ring_virt_addr) {
697 		dev_err(dev, "No dma memory for inb_ring_virt_addr\n");
698 		ret = -ENOMEM;
699 		goto err_free;
700 	}
701 
702 	chan->oub_ring_virt_addr = dma_pool_alloc(chan->re_dev->hw_desc_pool,
703 		GFP_KERNEL, &chan->oub_phys_addr);
704 	if (!chan->oub_ring_virt_addr) {
705 		dev_err(dev, "No dma memory for oub_ring_virt_addr\n");
706 		ret = -ENOMEM;
707 		goto err_free_1;
708 	}
709 
710 	/* Program the Inbound/Outbound ring base addresses and size */
711 	out_be32(&chan->jrregs->inbring_base_h,
712 		 chan->inb_phys_addr & FSL_RE_ADDR_BIT_MASK);
713 	out_be32(&chan->jrregs->oubring_base_h,
714 		 chan->oub_phys_addr & FSL_RE_ADDR_BIT_MASK);
715 	out_be32(&chan->jrregs->inbring_base_l,
716 		 chan->inb_phys_addr >> FSL_RE_ADDR_BIT_SHIFT);
717 	out_be32(&chan->jrregs->oubring_base_l,
718 		 chan->oub_phys_addr >> FSL_RE_ADDR_BIT_SHIFT);
719 	out_be32(&chan->jrregs->inbring_size,
720 		 FSL_RE_RING_SIZE << FSL_RE_RING_SIZE_SHIFT);
721 	out_be32(&chan->jrregs->oubring_size,
722 		 FSL_RE_RING_SIZE << FSL_RE_RING_SIZE_SHIFT);
723 
724 	/* Read LIODN value from u-boot */
725 	status = in_be32(&chan->jrregs->jr_config_1) & FSL_RE_REG_LIODN_MASK;
726 
727 	/* Program the CFG reg */
728 	out_be32(&chan->jrregs->jr_config_1,
729 		 FSL_RE_CFG1_CBSI | FSL_RE_CFG1_CBS0 | status);
730 
731 	dev_set_drvdata(chandev, chan);
732 
733 	/* Enable RE/CHAN */
734 	out_be32(&chan->jrregs->jr_command, FSL_RE_ENABLE);
735 
736 	return 0;
737 
738 err_free_1:
739 	dma_pool_free(chan->re_dev->hw_desc_pool, chan->inb_ring_virt_addr,
740 		      chan->inb_phys_addr);
741 err_free:
742 	return ret;
743 }
744 
745 /* Probe function for RAID Engine */
fsl_re_probe(struct platform_device * ofdev)746 static int fsl_re_probe(struct platform_device *ofdev)
747 {
748 	struct fsl_re_drv_private *re_priv;
749 	struct device_node *np;
750 	struct device_node *child;
751 	u32 off;
752 	u8 ridx = 0;
753 	struct dma_device *dma_dev;
754 	struct resource *res;
755 	int rc;
756 	struct device *dev = &ofdev->dev;
757 
758 	re_priv = devm_kzalloc(dev, sizeof(*re_priv), GFP_KERNEL);
759 	if (!re_priv)
760 		return -ENOMEM;
761 
762 	res = platform_get_resource(ofdev, IORESOURCE_MEM, 0);
763 	if (!res)
764 		return -ENODEV;
765 
766 	/* IOMAP the entire RAID Engine region */
767 	re_priv->re_regs = devm_ioremap(dev, res->start, resource_size(res));
768 	if (!re_priv->re_regs)
769 		return -EBUSY;
770 
771 	/* Program the RE mode */
772 	out_be32(&re_priv->re_regs->global_config, FSL_RE_NON_DPAA_MODE);
773 
774 	/* Program Galois Field polynomial */
775 	out_be32(&re_priv->re_regs->galois_field_config, FSL_RE_GFM_POLY);
776 
777 	dev_info(dev, "version %x, mode %x, gfp %x\n",
778 		 in_be32(&re_priv->re_regs->re_version_id),
779 		 in_be32(&re_priv->re_regs->global_config),
780 		 in_be32(&re_priv->re_regs->galois_field_config));
781 
782 	dma_dev = &re_priv->dma_dev;
783 	dma_dev->dev = dev;
784 	INIT_LIST_HEAD(&dma_dev->channels);
785 	dma_set_mask(dev, DMA_BIT_MASK(40));
786 
787 	dma_dev->device_alloc_chan_resources = fsl_re_alloc_chan_resources;
788 	dma_dev->device_tx_status = fsl_re_tx_status;
789 	dma_dev->device_issue_pending = fsl_re_issue_pending;
790 
791 	dma_dev->max_xor = FSL_RE_MAX_XOR_SRCS;
792 	dma_dev->device_prep_dma_xor = fsl_re_prep_dma_xor;
793 	dma_cap_set(DMA_XOR, dma_dev->cap_mask);
794 
795 	dma_dev->max_pq = FSL_RE_MAX_PQ_SRCS;
796 	dma_dev->device_prep_dma_pq = fsl_re_prep_dma_pq;
797 	dma_cap_set(DMA_PQ, dma_dev->cap_mask);
798 
799 	dma_dev->device_prep_dma_memcpy = fsl_re_prep_dma_memcpy;
800 	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
801 
802 	dma_dev->device_free_chan_resources = fsl_re_free_chan_resources;
803 
804 	re_priv->total_chans = 0;
805 
806 	re_priv->cf_desc_pool = dmam_pool_create("fsl_re_cf_desc_pool", dev,
807 					FSL_RE_CF_CDB_SIZE,
808 					FSL_RE_CF_CDB_ALIGN, 0);
809 
810 	if (!re_priv->cf_desc_pool) {
811 		dev_err(dev, "No memory for fsl re_cf desc pool\n");
812 		return -ENOMEM;
813 	}
814 
815 	re_priv->hw_desc_pool = dmam_pool_create("fsl_re_hw_desc_pool", dev,
816 			sizeof(struct fsl_re_hw_desc) * FSL_RE_RING_SIZE,
817 			FSL_RE_FRAME_ALIGN, 0);
818 	if (!re_priv->hw_desc_pool) {
819 		dev_err(dev, "No memory for fsl re_hw desc pool\n");
820 		return -ENOMEM;
821 	}
822 
823 	dev_set_drvdata(dev, re_priv);
824 
825 	/* Parse Device tree to find out the total number of JQs present */
826 	for_each_compatible_node(np, NULL, "fsl,raideng-v1.0-job-queue") {
827 		rc = of_property_read_u32(np, "reg", &off);
828 		if (rc) {
829 			dev_err(dev, "Reg property not found in JQ node\n");
830 			of_node_put(np);
831 			return -ENODEV;
832 		}
833 		/* Find out the Job Rings present under each JQ */
834 		for_each_child_of_node(np, child) {
835 			rc = of_device_is_compatible(child,
836 					     "fsl,raideng-v1.0-job-ring");
837 			if (rc) {
838 				fsl_re_chan_probe(ofdev, child, ridx++, off);
839 				re_priv->total_chans++;
840 			}
841 		}
842 	}
843 
844 	dma_async_device_register(dma_dev);
845 
846 	return 0;
847 }
848 
fsl_re_remove_chan(struct fsl_re_chan * chan)849 static void fsl_re_remove_chan(struct fsl_re_chan *chan)
850 {
851 	tasklet_kill(&chan->irqtask);
852 
853 	dma_pool_free(chan->re_dev->hw_desc_pool, chan->inb_ring_virt_addr,
854 		      chan->inb_phys_addr);
855 
856 	dma_pool_free(chan->re_dev->hw_desc_pool, chan->oub_ring_virt_addr,
857 		      chan->oub_phys_addr);
858 }
859 
fsl_re_remove(struct platform_device * ofdev)860 static void fsl_re_remove(struct platform_device *ofdev)
861 {
862 	struct fsl_re_drv_private *re_priv;
863 	struct device *dev;
864 	int i;
865 
866 	dev = &ofdev->dev;
867 	re_priv = dev_get_drvdata(dev);
868 
869 	/* Cleanup chan related memory areas */
870 	for (i = 0; i < re_priv->total_chans; i++)
871 		fsl_re_remove_chan(re_priv->re_jrs[i]);
872 
873 	/* Unregister the driver */
874 	dma_async_device_unregister(&re_priv->dma_dev);
875 }
876 
877 static const struct of_device_id fsl_re_ids[] = {
878 	{ .compatible = "fsl,raideng-v1.0", },
879 	{}
880 };
881 MODULE_DEVICE_TABLE(of, fsl_re_ids);
882 
883 static struct platform_driver fsl_re_driver = {
884 	.driver = {
885 		.name = "fsl-raideng",
886 		.of_match_table = fsl_re_ids,
887 	},
888 	.probe = fsl_re_probe,
889 	.remove_new = fsl_re_remove,
890 };
891 
892 module_platform_driver(fsl_re_driver);
893 
894 MODULE_AUTHOR("Harninder Rai <harninder.rai@freescale.com>");
895 MODULE_LICENSE("GPL v2");
896 MODULE_DESCRIPTION("Freescale RAID Engine Device Driver");
897