1 /*
2  * DMA Engine support for Tsi721 PCIExpress-to-SRIO bridge
3  *
4  * Copyright (c) 2011-2014 Integrated Device Technology, Inc.
5  * Alexandre Bounine <alexandre.bounine@idt.com>
6  *
7  * This program is free software; you can redistribute it and/or modify it
8  * under the terms of the GNU General Public License as published by the Free
9  * Software Foundation; either version 2 of the License, or (at your option)
10  * any later version.
11  *
12  * This program is distributed in the hope that it will be useful, but WITHOUT
13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
15  * more details.
16  *
17  * The full GNU General Public License is included in this distribution in the
18  * file called COPYING.
19  */
20 
21 #include <linux/io.h>
22 #include <linux/errno.h>
23 #include <linux/init.h>
24 #include <linux/ioport.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/pci.h>
28 #include <linux/rio.h>
29 #include <linux/rio_drv.h>
30 #include <linux/dma-mapping.h>
31 #include <linux/interrupt.h>
32 #include <linux/kfifo.h>
33 #include <linux/sched.h>
34 #include <linux/delay.h>
35 #include "../../dma/dmaengine.h"
36 
37 #include "tsi721.h"
38 
39 #ifdef CONFIG_PCI_MSI
40 static irqreturn_t tsi721_bdma_msix(int irq, void *ptr);
41 #endif
42 static int tsi721_submit_sg(struct tsi721_tx_desc *desc);
43 
44 static unsigned int dma_desc_per_channel = 128;
45 module_param(dma_desc_per_channel, uint, S_IRUGO);
46 MODULE_PARM_DESC(dma_desc_per_channel,
47 		 "Number of DMA descriptors per channel (default: 128)");
48 
49 static unsigned int dma_txqueue_sz = 16;
50 module_param(dma_txqueue_sz, uint, S_IRUGO);
51 MODULE_PARM_DESC(dma_txqueue_sz,
52 		 "DMA Transactions Queue Size (default: 16)");
53 
54 static u8 dma_sel = 0x7f;
55 module_param(dma_sel, byte, S_IRUGO);
56 MODULE_PARM_DESC(dma_sel,
57 		 "DMA Channel Selection Mask (default: 0x7f = all)");
58 
59 static inline struct tsi721_bdma_chan *to_tsi721_chan(struct dma_chan *chan)
60 {
61 	return container_of(chan, struct tsi721_bdma_chan, dchan);
62 }
63 
64 static inline struct tsi721_device *to_tsi721(struct dma_device *ddev)
65 {
66 	return container_of(ddev, struct rio_mport, dma)->priv;
67 }
68 
69 static inline
70 struct tsi721_tx_desc *to_tsi721_desc(struct dma_async_tx_descriptor *txd)
71 {
72 	return container_of(txd, struct tsi721_tx_desc, txd);
73 }
74 
75 static int tsi721_bdma_ch_init(struct tsi721_bdma_chan *bdma_chan, int bd_num)
76 {
77 	struct tsi721_dma_desc *bd_ptr;
78 	struct device *dev = bdma_chan->dchan.device->dev;
79 	u64		*sts_ptr;
80 	dma_addr_t	bd_phys;
81 	dma_addr_t	sts_phys;
82 	int		sts_size;
83 #ifdef CONFIG_PCI_MSI
84 	struct tsi721_device *priv = to_tsi721(bdma_chan->dchan.device);
85 #endif
86 
87 	tsi_debug(DMA, &bdma_chan->dchan.dev->device, "DMAC%d", bdma_chan->id);
88 
89 	/*
90 	 * Allocate space for DMA descriptors
91 	 * (add an extra element for link descriptor)
92 	 */
93 	bd_ptr = dma_alloc_coherent(dev,
94 				    (bd_num + 1) * sizeof(struct tsi721_dma_desc),
95 				    &bd_phys, GFP_ATOMIC);
96 	if (!bd_ptr)
97 		return -ENOMEM;
98 
99 	bdma_chan->bd_num = bd_num;
100 	bdma_chan->bd_phys = bd_phys;
101 	bdma_chan->bd_base = bd_ptr;
102 
103 	tsi_debug(DMA, &bdma_chan->dchan.dev->device,
104 		  "DMAC%d descriptors @ %p (phys = %pad)",
105 		  bdma_chan->id, bd_ptr, &bd_phys);
106 
107 	/* Allocate space for descriptor status FIFO */
108 	sts_size = ((bd_num + 1) >= TSI721_DMA_MINSTSSZ) ?
109 					(bd_num + 1) : TSI721_DMA_MINSTSSZ;
110 	sts_size = roundup_pow_of_two(sts_size);
111 	sts_ptr = dma_alloc_coherent(dev,
112 				     sts_size * sizeof(struct tsi721_dma_sts),
113 				     &sts_phys, GFP_ATOMIC);
114 	if (!sts_ptr) {
115 		/* Free space allocated for DMA descriptors */
116 		dma_free_coherent(dev,
117 				  (bd_num + 1) * sizeof(struct tsi721_dma_desc),
118 				  bd_ptr, bd_phys);
119 		bdma_chan->bd_base = NULL;
120 		return -ENOMEM;
121 	}
122 
123 	bdma_chan->sts_phys = sts_phys;
124 	bdma_chan->sts_base = sts_ptr;
125 	bdma_chan->sts_size = sts_size;
126 
127 	tsi_debug(DMA, &bdma_chan->dchan.dev->device,
128 		"DMAC%d desc status FIFO @ %p (phys = %pad) size=0x%x",
129 		bdma_chan->id, sts_ptr, &sts_phys, sts_size);
130 
131 	/* Initialize DMA descriptors ring using added link descriptor */
132 	bd_ptr[bd_num].type_id = cpu_to_le32(DTYPE3 << 29);
133 	bd_ptr[bd_num].next_lo = cpu_to_le32((u64)bd_phys &
134 						 TSI721_DMAC_DPTRL_MASK);
135 	bd_ptr[bd_num].next_hi = cpu_to_le32((u64)bd_phys >> 32);
136 
137 	/* Setup DMA descriptor pointers */
138 	iowrite32(((u64)bd_phys >> 32),
139 		bdma_chan->regs + TSI721_DMAC_DPTRH);
140 	iowrite32(((u64)bd_phys & TSI721_DMAC_DPTRL_MASK),
141 		bdma_chan->regs + TSI721_DMAC_DPTRL);
142 
143 	/* Setup descriptor status FIFO */
144 	iowrite32(((u64)sts_phys >> 32),
145 		bdma_chan->regs + TSI721_DMAC_DSBH);
146 	iowrite32(((u64)sts_phys & TSI721_DMAC_DSBL_MASK),
147 		bdma_chan->regs + TSI721_DMAC_DSBL);
148 	iowrite32(TSI721_DMAC_DSSZ_SIZE(sts_size),
149 		bdma_chan->regs + TSI721_DMAC_DSSZ);
150 
151 	/* Clear interrupt bits */
152 	iowrite32(TSI721_DMAC_INT_ALL,
153 		bdma_chan->regs + TSI721_DMAC_INT);
154 
155 	ioread32(bdma_chan->regs + TSI721_DMAC_INT);
156 
157 #ifdef CONFIG_PCI_MSI
158 	/* Request interrupt service if we are in MSI-X mode */
159 	if (priv->flags & TSI721_USING_MSIX) {
160 		int rc, idx;
161 
162 		idx = TSI721_VECT_DMA0_DONE + bdma_chan->id;
163 
164 		rc = request_irq(priv->msix[idx].vector, tsi721_bdma_msix, 0,
165 				 priv->msix[idx].irq_name, (void *)bdma_chan);
166 
167 		if (rc) {
168 			tsi_debug(DMA, &bdma_chan->dchan.dev->device,
169 				  "Unable to get MSI-X for DMAC%d-DONE",
170 				  bdma_chan->id);
171 			goto err_out;
172 		}
173 
174 		idx = TSI721_VECT_DMA0_INT + bdma_chan->id;
175 
176 		rc = request_irq(priv->msix[idx].vector, tsi721_bdma_msix, 0,
177 				priv->msix[idx].irq_name, (void *)bdma_chan);
178 
179 		if (rc)	{
180 			tsi_debug(DMA, &bdma_chan->dchan.dev->device,
181 				  "Unable to get MSI-X for DMAC%d-INT",
182 				  bdma_chan->id);
183 			free_irq(
184 				priv->msix[TSI721_VECT_DMA0_DONE +
185 					    bdma_chan->id].vector,
186 				(void *)bdma_chan);
187 		}
188 
189 err_out:
190 		if (rc) {
191 			/* Free space allocated for DMA descriptors */
192 			dma_free_coherent(dev,
193 				(bd_num + 1) * sizeof(struct tsi721_dma_desc),
194 				bd_ptr, bd_phys);
195 			bdma_chan->bd_base = NULL;
196 
197 			/* Free space allocated for status descriptors */
198 			dma_free_coherent(dev,
199 				sts_size * sizeof(struct tsi721_dma_sts),
200 				sts_ptr, sts_phys);
201 			bdma_chan->sts_base = NULL;
202 
203 			return -EIO;
204 		}
205 	}
206 #endif /* CONFIG_PCI_MSI */
207 
208 	/* Toggle DMA channel initialization */
209 	iowrite32(TSI721_DMAC_CTL_INIT,	bdma_chan->regs + TSI721_DMAC_CTL);
210 	ioread32(bdma_chan->regs + TSI721_DMAC_CTL);
211 	bdma_chan->wr_count = bdma_chan->wr_count_next = 0;
212 	bdma_chan->sts_rdptr = 0;
213 	udelay(10);
214 
215 	return 0;
216 }
217 
218 static int tsi721_bdma_ch_free(struct tsi721_bdma_chan *bdma_chan)
219 {
220 	u32 ch_stat;
221 #ifdef CONFIG_PCI_MSI
222 	struct tsi721_device *priv = to_tsi721(bdma_chan->dchan.device);
223 #endif
224 
225 	if (!bdma_chan->bd_base)
226 		return 0;
227 
228 	/* Check if DMA channel still running */
229 	ch_stat = ioread32(bdma_chan->regs + TSI721_DMAC_STS);
230 	if (ch_stat & TSI721_DMAC_STS_RUN)
231 		return -EFAULT;
232 
233 	/* Put DMA channel into init state */
234 	iowrite32(TSI721_DMAC_CTL_INIT,	bdma_chan->regs + TSI721_DMAC_CTL);
235 
236 #ifdef CONFIG_PCI_MSI
237 	if (priv->flags & TSI721_USING_MSIX) {
238 		free_irq(priv->msix[TSI721_VECT_DMA0_DONE +
239 				    bdma_chan->id].vector, (void *)bdma_chan);
240 		free_irq(priv->msix[TSI721_VECT_DMA0_INT +
241 				    bdma_chan->id].vector, (void *)bdma_chan);
242 	}
243 #endif /* CONFIG_PCI_MSI */
244 
245 	/* Free space allocated for DMA descriptors */
246 	dma_free_coherent(bdma_chan->dchan.device->dev,
247 		(bdma_chan->bd_num + 1) * sizeof(struct tsi721_dma_desc),
248 		bdma_chan->bd_base, bdma_chan->bd_phys);
249 	bdma_chan->bd_base = NULL;
250 
251 	/* Free space allocated for status FIFO */
252 	dma_free_coherent(bdma_chan->dchan.device->dev,
253 		bdma_chan->sts_size * sizeof(struct tsi721_dma_sts),
254 		bdma_chan->sts_base, bdma_chan->sts_phys);
255 	bdma_chan->sts_base = NULL;
256 	return 0;
257 }
258 
259 static void
260 tsi721_bdma_interrupt_enable(struct tsi721_bdma_chan *bdma_chan, int enable)
261 {
262 	if (enable) {
263 		/* Clear pending BDMA channel interrupts */
264 		iowrite32(TSI721_DMAC_INT_ALL,
265 			bdma_chan->regs + TSI721_DMAC_INT);
266 		ioread32(bdma_chan->regs + TSI721_DMAC_INT);
267 		/* Enable BDMA channel interrupts */
268 		iowrite32(TSI721_DMAC_INT_ALL,
269 			bdma_chan->regs + TSI721_DMAC_INTE);
270 	} else {
271 		/* Disable BDMA channel interrupts */
272 		iowrite32(0, bdma_chan->regs + TSI721_DMAC_INTE);
273 		/* Clear pending BDMA channel interrupts */
274 		iowrite32(TSI721_DMAC_INT_ALL,
275 			bdma_chan->regs + TSI721_DMAC_INT);
276 	}
277 
278 }
279 
280 static bool tsi721_dma_is_idle(struct tsi721_bdma_chan *bdma_chan)
281 {
282 	u32 sts;
283 
284 	sts = ioread32(bdma_chan->regs + TSI721_DMAC_STS);
285 	return ((sts & TSI721_DMAC_STS_RUN) == 0);
286 }
287 
288 void tsi721_bdma_handler(struct tsi721_bdma_chan *bdma_chan)
289 {
290 	/* Disable BDMA channel interrupts */
291 	iowrite32(0, bdma_chan->regs + TSI721_DMAC_INTE);
292 	if (bdma_chan->active)
293 		tasklet_hi_schedule(&bdma_chan->tasklet);
294 }
295 
296 #ifdef CONFIG_PCI_MSI
297 /**
298  * tsi721_omsg_msix - MSI-X interrupt handler for BDMA channels
299  * @irq: Linux interrupt number
300  * @ptr: Pointer to interrupt-specific data (BDMA channel structure)
301  *
302  * Handles BDMA channel interrupts signaled using MSI-X.
303  */
304 static irqreturn_t tsi721_bdma_msix(int irq, void *ptr)
305 {
306 	struct tsi721_bdma_chan *bdma_chan = ptr;
307 
308 	if (bdma_chan->active)
309 		tasklet_hi_schedule(&bdma_chan->tasklet);
310 	return IRQ_HANDLED;
311 }
312 #endif /* CONFIG_PCI_MSI */
313 
314 /* Must be called with the spinlock held */
315 static void tsi721_start_dma(struct tsi721_bdma_chan *bdma_chan)
316 {
317 	if (!tsi721_dma_is_idle(bdma_chan)) {
318 		tsi_err(&bdma_chan->dchan.dev->device,
319 			"DMAC%d Attempt to start non-idle channel",
320 			bdma_chan->id);
321 		return;
322 	}
323 
324 	if (bdma_chan->wr_count == bdma_chan->wr_count_next) {
325 		tsi_err(&bdma_chan->dchan.dev->device,
326 			"DMAC%d Attempt to start DMA with no BDs ready %d",
327 			bdma_chan->id, task_pid_nr(current));
328 		return;
329 	}
330 
331 	tsi_debug(DMA, &bdma_chan->dchan.dev->device, "DMAC%d (wrc=%d) %d",
332 		  bdma_chan->id, bdma_chan->wr_count_next,
333 		  task_pid_nr(current));
334 
335 	iowrite32(bdma_chan->wr_count_next,
336 		bdma_chan->regs + TSI721_DMAC_DWRCNT);
337 	ioread32(bdma_chan->regs + TSI721_DMAC_DWRCNT);
338 
339 	bdma_chan->wr_count = bdma_chan->wr_count_next;
340 }
341 
342 static int
343 tsi721_desc_fill_init(struct tsi721_tx_desc *desc,
344 		      struct tsi721_dma_desc *bd_ptr,
345 		      struct scatterlist *sg, u32 sys_size)
346 {
347 	u64 rio_addr;
348 
349 	if (!bd_ptr)
350 		return -EINVAL;
351 
352 	/* Initialize DMA descriptor */
353 	bd_ptr->type_id = cpu_to_le32((DTYPE1 << 29) |
354 				      (desc->rtype << 19) | desc->destid);
355 	bd_ptr->bcount = cpu_to_le32(((desc->rio_addr & 0x3) << 30) |
356 				     (sys_size << 26));
357 	rio_addr = (desc->rio_addr >> 2) |
358 				((u64)(desc->rio_addr_u & 0x3) << 62);
359 	bd_ptr->raddr_lo = cpu_to_le32(rio_addr & 0xffffffff);
360 	bd_ptr->raddr_hi = cpu_to_le32(rio_addr >> 32);
361 	bd_ptr->t1.bufptr_lo = cpu_to_le32(
362 					(u64)sg_dma_address(sg) & 0xffffffff);
363 	bd_ptr->t1.bufptr_hi = cpu_to_le32((u64)sg_dma_address(sg) >> 32);
364 	bd_ptr->t1.s_dist = 0;
365 	bd_ptr->t1.s_size = 0;
366 
367 	return 0;
368 }
369 
370 static int
371 tsi721_desc_fill_end(struct tsi721_dma_desc *bd_ptr, u32 bcount, bool interrupt)
372 {
373 	if (!bd_ptr)
374 		return -EINVAL;
375 
376 	/* Update DMA descriptor */
377 	if (interrupt)
378 		bd_ptr->type_id |= cpu_to_le32(TSI721_DMAD_IOF);
379 	bd_ptr->bcount |= cpu_to_le32(bcount & TSI721_DMAD_BCOUNT1);
380 
381 	return 0;
382 }
383 
384 static void tsi721_dma_tx_err(struct tsi721_bdma_chan *bdma_chan,
385 			      struct tsi721_tx_desc *desc)
386 {
387 	struct dma_async_tx_descriptor *txd = &desc->txd;
388 	dma_async_tx_callback callback = txd->callback;
389 	void *param = txd->callback_param;
390 
391 	list_move(&desc->desc_node, &bdma_chan->free_list);
392 
393 	if (callback)
394 		callback(param);
395 }
396 
397 static void tsi721_clr_stat(struct tsi721_bdma_chan *bdma_chan)
398 {
399 	u32 srd_ptr;
400 	u64 *sts_ptr;
401 	int i, j;
402 
403 	/* Check and clear descriptor status FIFO entries */
404 	srd_ptr = bdma_chan->sts_rdptr;
405 	sts_ptr = bdma_chan->sts_base;
406 	j = srd_ptr * 8;
407 	while (sts_ptr[j]) {
408 		for (i = 0; i < 8 && sts_ptr[j]; i++, j++)
409 			sts_ptr[j] = 0;
410 
411 		++srd_ptr;
412 		srd_ptr %= bdma_chan->sts_size;
413 		j = srd_ptr * 8;
414 	}
415 
416 	iowrite32(srd_ptr, bdma_chan->regs + TSI721_DMAC_DSRP);
417 	bdma_chan->sts_rdptr = srd_ptr;
418 }
419 
420 /* Must be called with the channel spinlock held */
421 static int tsi721_submit_sg(struct tsi721_tx_desc *desc)
422 {
423 	struct dma_chan *dchan = desc->txd.chan;
424 	struct tsi721_bdma_chan *bdma_chan = to_tsi721_chan(dchan);
425 	u32 sys_size;
426 	u64 rio_addr;
427 	dma_addr_t next_addr;
428 	u32 bcount;
429 	struct scatterlist *sg;
430 	unsigned int i;
431 	int err = 0;
432 	struct tsi721_dma_desc *bd_ptr = NULL;
433 	u32 idx, rd_idx;
434 	u32 add_count = 0;
435 	struct device *ch_dev = &dchan->dev->device;
436 
437 	if (!tsi721_dma_is_idle(bdma_chan)) {
438 		tsi_err(ch_dev, "DMAC%d ERR: Attempt to use non-idle channel",
439 			bdma_chan->id);
440 		return -EIO;
441 	}
442 
443 	/*
444 	 * Fill DMA channel's hardware buffer descriptors.
445 	 * (NOTE: RapidIO destination address is limited to 64 bits for now)
446 	 */
447 	rio_addr = desc->rio_addr;
448 	next_addr = -1;
449 	bcount = 0;
450 	sys_size = dma_to_mport(dchan->device)->sys_size;
451 
452 	rd_idx = ioread32(bdma_chan->regs + TSI721_DMAC_DRDCNT);
453 	rd_idx %= (bdma_chan->bd_num + 1);
454 
455 	idx = bdma_chan->wr_count_next % (bdma_chan->bd_num + 1);
456 	if (idx == bdma_chan->bd_num) {
457 		/* wrap around link descriptor */
458 		idx = 0;
459 		add_count++;
460 	}
461 
462 	tsi_debug(DMA, ch_dev, "DMAC%d BD ring status: rdi=%d wri=%d",
463 		  bdma_chan->id, rd_idx, idx);
464 
465 	for_each_sg(desc->sg, sg, desc->sg_len, i) {
466 
467 		tsi_debug(DMAV, ch_dev, "DMAC%d sg%d/%d addr: 0x%llx len: %d",
468 			bdma_chan->id, i, desc->sg_len,
469 			(unsigned long long)sg_dma_address(sg), sg_dma_len(sg));
470 
471 		if (sg_dma_len(sg) > TSI721_BDMA_MAX_BCOUNT) {
472 			tsi_err(ch_dev, "DMAC%d SG entry %d is too large",
473 				bdma_chan->id, i);
474 			err = -EINVAL;
475 			break;
476 		}
477 
478 		/*
479 		 * If this sg entry forms contiguous block with previous one,
480 		 * try to merge it into existing DMA descriptor
481 		 */
482 		if (next_addr == sg_dma_address(sg) &&
483 		    bcount + sg_dma_len(sg) <= TSI721_BDMA_MAX_BCOUNT) {
484 			/* Adjust byte count of the descriptor */
485 			bcount += sg_dma_len(sg);
486 			goto entry_done;
487 		} else if (next_addr != -1) {
488 			/* Finalize descriptor using total byte count value */
489 			tsi721_desc_fill_end(bd_ptr, bcount, 0);
490 			tsi_debug(DMAV, ch_dev,	"DMAC%d prev desc final len: %d",
491 				  bdma_chan->id, bcount);
492 		}
493 
494 		desc->rio_addr = rio_addr;
495 
496 		if (i && idx == rd_idx) {
497 			tsi_debug(DMAV, ch_dev,
498 				  "DMAC%d HW descriptor ring is full @ %d",
499 				  bdma_chan->id, i);
500 			desc->sg = sg;
501 			desc->sg_len -= i;
502 			break;
503 		}
504 
505 		bd_ptr = &((struct tsi721_dma_desc *)bdma_chan->bd_base)[idx];
506 		err = tsi721_desc_fill_init(desc, bd_ptr, sg, sys_size);
507 		if (err) {
508 			tsi_err(ch_dev, "Failed to build desc: err=%d", err);
509 			break;
510 		}
511 
512 		tsi_debug(DMAV, ch_dev, "DMAC%d bd_ptr = %p did=%d raddr=0x%llx",
513 			  bdma_chan->id, bd_ptr, desc->destid, desc->rio_addr);
514 
515 		next_addr = sg_dma_address(sg);
516 		bcount = sg_dma_len(sg);
517 
518 		add_count++;
519 		if (++idx == bdma_chan->bd_num) {
520 			/* wrap around link descriptor */
521 			idx = 0;
522 			add_count++;
523 		}
524 
525 entry_done:
526 		if (sg_is_last(sg)) {
527 			tsi721_desc_fill_end(bd_ptr, bcount, 0);
528 			tsi_debug(DMAV, ch_dev,
529 				  "DMAC%d last desc final len: %d",
530 				  bdma_chan->id, bcount);
531 			desc->sg_len = 0;
532 		} else {
533 			rio_addr += sg_dma_len(sg);
534 			next_addr += sg_dma_len(sg);
535 		}
536 	}
537 
538 	if (!err)
539 		bdma_chan->wr_count_next += add_count;
540 
541 	return err;
542 }
543 
544 static void tsi721_advance_work(struct tsi721_bdma_chan *bdma_chan,
545 				struct tsi721_tx_desc *desc)
546 {
547 	int err;
548 
549 	tsi_debug(DMA, &bdma_chan->dchan.dev->device, "DMAC%d", bdma_chan->id);
550 
551 	if (!tsi721_dma_is_idle(bdma_chan))
552 		return;
553 
554 	/*
555 	 * If there is no data transfer in progress, fetch new descriptor from
556 	 * the pending queue.
557 	*/
558 	if (!desc && !bdma_chan->active_tx && !list_empty(&bdma_chan->queue)) {
559 		desc = list_first_entry(&bdma_chan->queue,
560 					struct tsi721_tx_desc, desc_node);
561 		list_del_init((&desc->desc_node));
562 		bdma_chan->active_tx = desc;
563 	}
564 
565 	if (desc) {
566 		err = tsi721_submit_sg(desc);
567 		if (!err)
568 			tsi721_start_dma(bdma_chan);
569 		else {
570 			tsi721_dma_tx_err(bdma_chan, desc);
571 			tsi_debug(DMA, &bdma_chan->dchan.dev->device,
572 				"DMAC%d ERR: tsi721_submit_sg failed with err=%d",
573 				bdma_chan->id, err);
574 		}
575 	}
576 
577 	tsi_debug(DMA, &bdma_chan->dchan.dev->device, "DMAC%d Exit",
578 		  bdma_chan->id);
579 }
580 
581 static void tsi721_dma_tasklet(unsigned long data)
582 {
583 	struct tsi721_bdma_chan *bdma_chan = (struct tsi721_bdma_chan *)data;
584 	u32 dmac_int, dmac_sts;
585 
586 	dmac_int = ioread32(bdma_chan->regs + TSI721_DMAC_INT);
587 	tsi_debug(DMA, &bdma_chan->dchan.dev->device, "DMAC%d_INT = 0x%x",
588 		  bdma_chan->id, dmac_int);
589 	/* Clear channel interrupts */
590 	iowrite32(dmac_int, bdma_chan->regs + TSI721_DMAC_INT);
591 
592 	if (dmac_int & TSI721_DMAC_INT_ERR) {
593 		int i = 10000;
594 		struct tsi721_tx_desc *desc;
595 
596 		desc = bdma_chan->active_tx;
597 		dmac_sts = ioread32(bdma_chan->regs + TSI721_DMAC_STS);
598 		tsi_err(&bdma_chan->dchan.dev->device,
599 			"DMAC%d_STS = 0x%x did=%d raddr=0x%llx",
600 			bdma_chan->id, dmac_sts, desc->destid, desc->rio_addr);
601 
602 		/* Re-initialize DMA channel if possible */
603 
604 		if ((dmac_sts & TSI721_DMAC_STS_ABORT) == 0)
605 			goto err_out;
606 
607 		tsi721_clr_stat(bdma_chan);
608 
609 		spin_lock(&bdma_chan->lock);
610 
611 		/* Put DMA channel into init state */
612 		iowrite32(TSI721_DMAC_CTL_INIT,
613 			  bdma_chan->regs + TSI721_DMAC_CTL);
614 		do {
615 			udelay(1);
616 			dmac_sts = ioread32(bdma_chan->regs + TSI721_DMAC_STS);
617 			i--;
618 		} while ((dmac_sts & TSI721_DMAC_STS_ABORT) && i);
619 
620 		if (dmac_sts & TSI721_DMAC_STS_ABORT) {
621 			tsi_err(&bdma_chan->dchan.dev->device,
622 				"Failed to re-initiate DMAC%d",	bdma_chan->id);
623 			spin_unlock(&bdma_chan->lock);
624 			goto err_out;
625 		}
626 
627 		/* Setup DMA descriptor pointers */
628 		iowrite32(((u64)bdma_chan->bd_phys >> 32),
629 			bdma_chan->regs + TSI721_DMAC_DPTRH);
630 		iowrite32(((u64)bdma_chan->bd_phys & TSI721_DMAC_DPTRL_MASK),
631 			bdma_chan->regs + TSI721_DMAC_DPTRL);
632 
633 		/* Setup descriptor status FIFO */
634 		iowrite32(((u64)bdma_chan->sts_phys >> 32),
635 			bdma_chan->regs + TSI721_DMAC_DSBH);
636 		iowrite32(((u64)bdma_chan->sts_phys & TSI721_DMAC_DSBL_MASK),
637 			bdma_chan->regs + TSI721_DMAC_DSBL);
638 		iowrite32(TSI721_DMAC_DSSZ_SIZE(bdma_chan->sts_size),
639 			bdma_chan->regs + TSI721_DMAC_DSSZ);
640 
641 		/* Clear interrupt bits */
642 		iowrite32(TSI721_DMAC_INT_ALL,
643 			bdma_chan->regs + TSI721_DMAC_INT);
644 
645 		ioread32(bdma_chan->regs + TSI721_DMAC_INT);
646 
647 		bdma_chan->wr_count = bdma_chan->wr_count_next = 0;
648 		bdma_chan->sts_rdptr = 0;
649 		udelay(10);
650 
651 		desc = bdma_chan->active_tx;
652 		desc->status = DMA_ERROR;
653 		dma_cookie_complete(&desc->txd);
654 		list_add(&desc->desc_node, &bdma_chan->free_list);
655 		bdma_chan->active_tx = NULL;
656 		if (bdma_chan->active)
657 			tsi721_advance_work(bdma_chan, NULL);
658 		spin_unlock(&bdma_chan->lock);
659 	}
660 
661 	if (dmac_int & TSI721_DMAC_INT_STFULL) {
662 		tsi_err(&bdma_chan->dchan.dev->device,
663 			"DMAC%d descriptor status FIFO is full",
664 			bdma_chan->id);
665 	}
666 
667 	if (dmac_int & (TSI721_DMAC_INT_DONE | TSI721_DMAC_INT_IOFDONE)) {
668 		struct tsi721_tx_desc *desc;
669 
670 		tsi721_clr_stat(bdma_chan);
671 		spin_lock(&bdma_chan->lock);
672 		desc = bdma_chan->active_tx;
673 
674 		if (desc->sg_len == 0) {
675 			dma_async_tx_callback callback = NULL;
676 			void *param = NULL;
677 
678 			desc->status = DMA_COMPLETE;
679 			dma_cookie_complete(&desc->txd);
680 			if (desc->txd.flags & DMA_PREP_INTERRUPT) {
681 				callback = desc->txd.callback;
682 				param = desc->txd.callback_param;
683 			}
684 			list_add(&desc->desc_node, &bdma_chan->free_list);
685 			bdma_chan->active_tx = NULL;
686 			if (bdma_chan->active)
687 				tsi721_advance_work(bdma_chan, NULL);
688 			spin_unlock(&bdma_chan->lock);
689 			if (callback)
690 				callback(param);
691 		} else {
692 			if (bdma_chan->active)
693 				tsi721_advance_work(bdma_chan,
694 						    bdma_chan->active_tx);
695 			spin_unlock(&bdma_chan->lock);
696 		}
697 	}
698 err_out:
699 	/* Re-Enable BDMA channel interrupts */
700 	iowrite32(TSI721_DMAC_INT_ALL, bdma_chan->regs + TSI721_DMAC_INTE);
701 }
702 
703 static dma_cookie_t tsi721_tx_submit(struct dma_async_tx_descriptor *txd)
704 {
705 	struct tsi721_tx_desc *desc = to_tsi721_desc(txd);
706 	struct tsi721_bdma_chan *bdma_chan = to_tsi721_chan(txd->chan);
707 	dma_cookie_t cookie;
708 
709 	/* Check if the descriptor is detached from any lists */
710 	if (!list_empty(&desc->desc_node)) {
711 		tsi_err(&bdma_chan->dchan.dev->device,
712 			"DMAC%d wrong state of descriptor %p",
713 			bdma_chan->id, txd);
714 		return -EIO;
715 	}
716 
717 	spin_lock_bh(&bdma_chan->lock);
718 
719 	if (!bdma_chan->active) {
720 		spin_unlock_bh(&bdma_chan->lock);
721 		return -ENODEV;
722 	}
723 
724 	cookie = dma_cookie_assign(txd);
725 	desc->status = DMA_IN_PROGRESS;
726 	list_add_tail(&desc->desc_node, &bdma_chan->queue);
727 	tsi721_advance_work(bdma_chan, NULL);
728 
729 	spin_unlock_bh(&bdma_chan->lock);
730 	return cookie;
731 }
732 
733 static int tsi721_alloc_chan_resources(struct dma_chan *dchan)
734 {
735 	struct tsi721_bdma_chan *bdma_chan = to_tsi721_chan(dchan);
736 	struct tsi721_tx_desc *desc;
737 	int i;
738 
739 	tsi_debug(DMA, &dchan->dev->device, "DMAC%d", bdma_chan->id);
740 
741 	if (bdma_chan->bd_base)
742 		return dma_txqueue_sz;
743 
744 	/* Initialize BDMA channel */
745 	if (tsi721_bdma_ch_init(bdma_chan, dma_desc_per_channel)) {
746 		tsi_err(&dchan->dev->device, "Unable to initialize DMAC%d",
747 			bdma_chan->id);
748 		return -ENODEV;
749 	}
750 
751 	/* Allocate queue of transaction descriptors */
752 	desc = kcalloc(dma_txqueue_sz, sizeof(struct tsi721_tx_desc),
753 			GFP_ATOMIC);
754 	if (!desc) {
755 		tsi721_bdma_ch_free(bdma_chan);
756 		return -ENOMEM;
757 	}
758 
759 	bdma_chan->tx_desc = desc;
760 
761 	for (i = 0; i < dma_txqueue_sz; i++) {
762 		dma_async_tx_descriptor_init(&desc[i].txd, dchan);
763 		desc[i].txd.tx_submit = tsi721_tx_submit;
764 		desc[i].txd.flags = DMA_CTRL_ACK;
765 		list_add(&desc[i].desc_node, &bdma_chan->free_list);
766 	}
767 
768 	dma_cookie_init(dchan);
769 
770 	bdma_chan->active = true;
771 	tsi721_bdma_interrupt_enable(bdma_chan, 1);
772 
773 	return dma_txqueue_sz;
774 }
775 
776 static void tsi721_sync_dma_irq(struct tsi721_bdma_chan *bdma_chan)
777 {
778 	struct tsi721_device *priv = to_tsi721(bdma_chan->dchan.device);
779 
780 #ifdef CONFIG_PCI_MSI
781 	if (priv->flags & TSI721_USING_MSIX) {
782 		synchronize_irq(priv->msix[TSI721_VECT_DMA0_DONE +
783 					   bdma_chan->id].vector);
784 		synchronize_irq(priv->msix[TSI721_VECT_DMA0_INT +
785 					   bdma_chan->id].vector);
786 	} else
787 #endif
788 	synchronize_irq(priv->pdev->irq);
789 }
790 
791 static void tsi721_free_chan_resources(struct dma_chan *dchan)
792 {
793 	struct tsi721_bdma_chan *bdma_chan = to_tsi721_chan(dchan);
794 
795 	tsi_debug(DMA, &dchan->dev->device, "DMAC%d", bdma_chan->id);
796 
797 	if (!bdma_chan->bd_base)
798 		return;
799 
800 	tsi721_bdma_interrupt_enable(bdma_chan, 0);
801 	bdma_chan->active = false;
802 	tsi721_sync_dma_irq(bdma_chan);
803 	tasklet_kill(&bdma_chan->tasklet);
804 	INIT_LIST_HEAD(&bdma_chan->free_list);
805 	kfree(bdma_chan->tx_desc);
806 	tsi721_bdma_ch_free(bdma_chan);
807 }
808 
809 static
810 enum dma_status tsi721_tx_status(struct dma_chan *dchan, dma_cookie_t cookie,
811 				 struct dma_tx_state *txstate)
812 {
813 	struct tsi721_bdma_chan *bdma_chan = to_tsi721_chan(dchan);
814 	enum dma_status	status;
815 
816 	spin_lock_bh(&bdma_chan->lock);
817 	status = dma_cookie_status(dchan, cookie, txstate);
818 	spin_unlock_bh(&bdma_chan->lock);
819 	return status;
820 }
821 
822 static void tsi721_issue_pending(struct dma_chan *dchan)
823 {
824 	struct tsi721_bdma_chan *bdma_chan = to_tsi721_chan(dchan);
825 
826 	tsi_debug(DMA, &dchan->dev->device, "DMAC%d", bdma_chan->id);
827 
828 	spin_lock_bh(&bdma_chan->lock);
829 	if (tsi721_dma_is_idle(bdma_chan) && bdma_chan->active) {
830 		tsi721_advance_work(bdma_chan, NULL);
831 	}
832 	spin_unlock_bh(&bdma_chan->lock);
833 }
834 
835 static
836 struct dma_async_tx_descriptor *tsi721_prep_rio_sg(struct dma_chan *dchan,
837 			struct scatterlist *sgl, unsigned int sg_len,
838 			enum dma_transfer_direction dir, unsigned long flags,
839 			void *tinfo)
840 {
841 	struct tsi721_bdma_chan *bdma_chan = to_tsi721_chan(dchan);
842 	struct tsi721_tx_desc *desc;
843 	struct rio_dma_ext *rext = tinfo;
844 	enum dma_rtype rtype;
845 	struct dma_async_tx_descriptor *txd = NULL;
846 
847 	if (!sgl || !sg_len) {
848 		tsi_err(&dchan->dev->device, "DMAC%d No SG list",
849 			bdma_chan->id);
850 		return ERR_PTR(-EINVAL);
851 	}
852 
853 	tsi_debug(DMA, &dchan->dev->device, "DMAC%d %s", bdma_chan->id,
854 		  (dir == DMA_DEV_TO_MEM)?"READ":"WRITE");
855 
856 	if (dir == DMA_DEV_TO_MEM)
857 		rtype = NREAD;
858 	else if (dir == DMA_MEM_TO_DEV) {
859 		switch (rext->wr_type) {
860 		case RDW_ALL_NWRITE:
861 			rtype = ALL_NWRITE;
862 			break;
863 		case RDW_ALL_NWRITE_R:
864 			rtype = ALL_NWRITE_R;
865 			break;
866 		case RDW_LAST_NWRITE_R:
867 		default:
868 			rtype = LAST_NWRITE_R;
869 			break;
870 		}
871 	} else {
872 		tsi_err(&dchan->dev->device,
873 			"DMAC%d Unsupported DMA direction option",
874 			bdma_chan->id);
875 		return ERR_PTR(-EINVAL);
876 	}
877 
878 	spin_lock_bh(&bdma_chan->lock);
879 
880 	if (!list_empty(&bdma_chan->free_list)) {
881 		desc = list_first_entry(&bdma_chan->free_list,
882 				struct tsi721_tx_desc, desc_node);
883 		list_del_init(&desc->desc_node);
884 		desc->destid = rext->destid;
885 		desc->rio_addr = rext->rio_addr;
886 		desc->rio_addr_u = 0;
887 		desc->rtype = rtype;
888 		desc->sg_len	= sg_len;
889 		desc->sg	= sgl;
890 		txd		= &desc->txd;
891 		txd->flags	= flags;
892 	}
893 
894 	spin_unlock_bh(&bdma_chan->lock);
895 
896 	if (!txd) {
897 		tsi_debug(DMA, &dchan->dev->device,
898 			  "DMAC%d free TXD is not available", bdma_chan->id);
899 		return ERR_PTR(-EBUSY);
900 	}
901 
902 	return txd;
903 }
904 
905 static int tsi721_terminate_all(struct dma_chan *dchan)
906 {
907 	struct tsi721_bdma_chan *bdma_chan = to_tsi721_chan(dchan);
908 	struct tsi721_tx_desc *desc, *_d;
909 	LIST_HEAD(list);
910 
911 	tsi_debug(DMA, &dchan->dev->device, "DMAC%d", bdma_chan->id);
912 
913 	spin_lock_bh(&bdma_chan->lock);
914 
915 	bdma_chan->active = false;
916 
917 	while (!tsi721_dma_is_idle(bdma_chan)) {
918 
919 		udelay(5);
920 #if (0)
921 		/* make sure to stop the transfer */
922 		iowrite32(TSI721_DMAC_CTL_SUSP,
923 			  bdma_chan->regs + TSI721_DMAC_CTL);
924 
925 		/* Wait until DMA channel stops */
926 		do {
927 			dmac_int = ioread32(bdma_chan->regs + TSI721_DMAC_INT);
928 		} while ((dmac_int & TSI721_DMAC_INT_SUSP) == 0);
929 #endif
930 	}
931 
932 	if (bdma_chan->active_tx)
933 		list_add(&bdma_chan->active_tx->desc_node, &list);
934 	list_splice_init(&bdma_chan->queue, &list);
935 
936 	list_for_each_entry_safe(desc, _d, &list, desc_node)
937 		tsi721_dma_tx_err(bdma_chan, desc);
938 
939 	spin_unlock_bh(&bdma_chan->lock);
940 
941 	return 0;
942 }
943 
944 static void tsi721_dma_stop(struct tsi721_bdma_chan *bdma_chan)
945 {
946 	if (!bdma_chan->active)
947 		return;
948 	spin_lock_bh(&bdma_chan->lock);
949 	if (!tsi721_dma_is_idle(bdma_chan)) {
950 		int timeout = 100000;
951 
952 		/* stop the transfer in progress */
953 		iowrite32(TSI721_DMAC_CTL_SUSP,
954 			  bdma_chan->regs + TSI721_DMAC_CTL);
955 
956 		/* Wait until DMA channel stops */
957 		while (!tsi721_dma_is_idle(bdma_chan) && --timeout)
958 			udelay(1);
959 	}
960 
961 	spin_unlock_bh(&bdma_chan->lock);
962 }
963 
964 void tsi721_dma_stop_all(struct tsi721_device *priv)
965 {
966 	int i;
967 
968 	for (i = 0; i < TSI721_DMA_MAXCH; i++) {
969 		if ((i != TSI721_DMACH_MAINT) && (dma_sel & (1 << i)))
970 			tsi721_dma_stop(&priv->bdma[i]);
971 	}
972 }
973 
974 int tsi721_register_dma(struct tsi721_device *priv)
975 {
976 	int i;
977 	int nr_channels = 0;
978 	int err;
979 	struct rio_mport *mport = &priv->mport;
980 
981 	INIT_LIST_HEAD(&mport->dma.channels);
982 
983 	for (i = 0; i < TSI721_DMA_MAXCH; i++) {
984 		struct tsi721_bdma_chan *bdma_chan = &priv->bdma[i];
985 
986 		if ((i == TSI721_DMACH_MAINT) || (dma_sel & (1 << i)) == 0)
987 			continue;
988 
989 		bdma_chan->regs = priv->regs + TSI721_DMAC_BASE(i);
990 
991 		bdma_chan->dchan.device = &mport->dma;
992 		bdma_chan->dchan.cookie = 1;
993 		bdma_chan->dchan.chan_id = i;
994 		bdma_chan->id = i;
995 		bdma_chan->active = false;
996 
997 		spin_lock_init(&bdma_chan->lock);
998 
999 		bdma_chan->active_tx = NULL;
1000 		INIT_LIST_HEAD(&bdma_chan->queue);
1001 		INIT_LIST_HEAD(&bdma_chan->free_list);
1002 
1003 		tasklet_init(&bdma_chan->tasklet, tsi721_dma_tasklet,
1004 			     (unsigned long)bdma_chan);
1005 		list_add_tail(&bdma_chan->dchan.device_node,
1006 			      &mport->dma.channels);
1007 		nr_channels++;
1008 	}
1009 
1010 	mport->dma.chancnt = nr_channels;
1011 	dma_cap_zero(mport->dma.cap_mask);
1012 	dma_cap_set(DMA_PRIVATE, mport->dma.cap_mask);
1013 	dma_cap_set(DMA_SLAVE, mport->dma.cap_mask);
1014 
1015 	mport->dma.dev = &priv->pdev->dev;
1016 	mport->dma.device_alloc_chan_resources = tsi721_alloc_chan_resources;
1017 	mport->dma.device_free_chan_resources = tsi721_free_chan_resources;
1018 	mport->dma.device_tx_status = tsi721_tx_status;
1019 	mport->dma.device_issue_pending = tsi721_issue_pending;
1020 	mport->dma.device_prep_slave_sg = tsi721_prep_rio_sg;
1021 	mport->dma.device_terminate_all = tsi721_terminate_all;
1022 
1023 	err = dma_async_device_register(&mport->dma);
1024 	if (err)
1025 		tsi_err(&priv->pdev->dev, "Failed to register DMA device");
1026 
1027 	return err;
1028 }
1029 
1030 void tsi721_unregister_dma(struct tsi721_device *priv)
1031 {
1032 	struct rio_mport *mport = &priv->mport;
1033 	struct dma_chan *chan, *_c;
1034 	struct tsi721_bdma_chan *bdma_chan;
1035 
1036 	tsi721_dma_stop_all(priv);
1037 	dma_async_device_unregister(&mport->dma);
1038 
1039 	list_for_each_entry_safe(chan, _c, &mport->dma.channels,
1040 					device_node) {
1041 		bdma_chan = to_tsi721_chan(chan);
1042 		if (bdma_chan->active) {
1043 			tsi721_bdma_interrupt_enable(bdma_chan, 0);
1044 			bdma_chan->active = false;
1045 			tsi721_sync_dma_irq(bdma_chan);
1046 			tasklet_kill(&bdma_chan->tasklet);
1047 			INIT_LIST_HEAD(&bdma_chan->free_list);
1048 			kfree(bdma_chan->tx_desc);
1049 			tsi721_bdma_ch_free(bdma_chan);
1050 		}
1051 
1052 		list_del(&chan->device_node);
1053 	}
1054 }
1055