1/*
2 * Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the Free
6 * Software Foundation; either version 2 of the License, or (at your option)
7 * any later version.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 59
16 * Temple Place - Suite 330, Boston, MA  02111-1307, USA.
17 *
18 * The full GNU General Public License is included in this distribution in the
19 * file called COPYING.
20 */
21
22/*
23 * This driver supports an Intel I/OAT DMA engine, which does asynchronous
24 * copy operations.
25 */
26
27#include <linux/init.h>
28#include <linux/module.h>
29#include <linux/pci.h>
30#include <linux/interrupt.h>
31#include <linux/dmaengine.h>
32#include <linux/delay.h>
33#include <linux/dma-mapping.h>
34#include "ioatdma.h"
35#include "ioatdma_io.h"
36#include "ioatdma_registers.h"
37#include "ioatdma_hw.h"
38
39#define to_ioat_chan(chan) container_of(chan, struct ioat_dma_chan, common)
40#define to_ioat_device(dev) container_of(dev, struct ioat_device, common)
41#define to_ioat_desc(lh) container_of(lh, struct ioat_desc_sw, node)
42
43/* internal functions */
44static int __devinit ioat_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
45static void __devexit ioat_remove(struct pci_dev *pdev);
46
47static int enumerate_dma_channels(struct ioat_device *device)
48{
49	u8 xfercap_scale;
50	u32 xfercap;
51	int i;
52	struct ioat_dma_chan *ioat_chan;
53
54	device->common.chancnt = ioatdma_read8(device, IOAT_CHANCNT_OFFSET);
55	xfercap_scale = ioatdma_read8(device, IOAT_XFERCAP_OFFSET);
56	xfercap = (xfercap_scale == 0 ? -1 : (1UL << xfercap_scale));
57
58	for (i = 0; i < device->common.chancnt; i++) {
59		ioat_chan = kzalloc(sizeof(*ioat_chan), GFP_KERNEL);
60		if (!ioat_chan) {
61			device->common.chancnt = i;
62			break;
63		}
64
65		ioat_chan->device = device;
66		ioat_chan->reg_base = device->reg_base + (0x80 * (i + 1));
67		ioat_chan->xfercap = xfercap;
68		spin_lock_init(&ioat_chan->cleanup_lock);
69		spin_lock_init(&ioat_chan->desc_lock);
70		INIT_LIST_HEAD(&ioat_chan->free_desc);
71		INIT_LIST_HEAD(&ioat_chan->used_desc);
72		/* This should be made common somewhere in dmaengine.c */
73		ioat_chan->common.device = &device->common;
74		ioat_chan->common.client = NULL;
75		list_add_tail(&ioat_chan->common.device_node,
76		              &device->common.channels);
77	}
78	return device->common.chancnt;
79}
80
81static struct ioat_desc_sw *ioat_dma_alloc_descriptor(
82	struct ioat_dma_chan *ioat_chan,
83	gfp_t flags)
84{
85	struct ioat_dma_descriptor *desc;
86	struct ioat_desc_sw *desc_sw;
87	struct ioat_device *ioat_device;
88	dma_addr_t phys;
89
90	ioat_device = to_ioat_device(ioat_chan->common.device);
91	desc = pci_pool_alloc(ioat_device->dma_pool, flags, &phys);
92	if (unlikely(!desc))
93		return NULL;
94
95	desc_sw = kzalloc(sizeof(*desc_sw), flags);
96	if (unlikely(!desc_sw)) {
97		pci_pool_free(ioat_device->dma_pool, desc, phys);
98		return NULL;
99	}
100
101	memset(desc, 0, sizeof(*desc));
102	desc_sw->hw = desc;
103	desc_sw->phys = phys;
104
105	return desc_sw;
106}
107
108#define INITIAL_IOAT_DESC_COUNT 128
109
110static void ioat_start_null_desc(struct ioat_dma_chan *ioat_chan);
111
112/* returns the actual number of allocated descriptors */
113static int ioat_dma_alloc_chan_resources(struct dma_chan *chan)
114{
115	struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
116	struct ioat_desc_sw *desc = NULL;
117	u16 chanctrl;
118	u32 chanerr;
119	int i;
120	LIST_HEAD(tmp_list);
121
122	/*
123	 * In-use bit automatically set by reading chanctrl
124	 * If 0, we got it, if 1, someone else did
125	 */
126	chanctrl = ioatdma_chan_read16(ioat_chan, IOAT_CHANCTRL_OFFSET);
127	if (chanctrl & IOAT_CHANCTRL_CHANNEL_IN_USE)
128		return -EBUSY;
129
130        /* Setup register to interrupt and write completion status on error */
131	chanctrl = IOAT_CHANCTRL_CHANNEL_IN_USE |
132		IOAT_CHANCTRL_ERR_INT_EN |
133		IOAT_CHANCTRL_ANY_ERR_ABORT_EN |
134		IOAT_CHANCTRL_ERR_COMPLETION_EN;
135        ioatdma_chan_write16(ioat_chan, IOAT_CHANCTRL_OFFSET, chanctrl);
136
137	chanerr = ioatdma_chan_read32(ioat_chan, IOAT_CHANERR_OFFSET);
138	if (chanerr) {
139		printk("IOAT: CHANERR = %x, clearing\n", chanerr);
140		ioatdma_chan_write32(ioat_chan, IOAT_CHANERR_OFFSET, chanerr);
141	}
142
143	/* Allocate descriptors */
144	for (i = 0; i < INITIAL_IOAT_DESC_COUNT; i++) {
145		desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_KERNEL);
146		if (!desc) {
147			printk(KERN_ERR "IOAT: Only %d initial descriptors\n", i);
148			break;
149		}
150		list_add_tail(&desc->node, &tmp_list);
151	}
152	spin_lock_bh(&ioat_chan->desc_lock);
153	list_splice(&tmp_list, &ioat_chan->free_desc);
154	spin_unlock_bh(&ioat_chan->desc_lock);
155
156	/* allocate a completion writeback area */
157	/* doing 2 32bit writes to mmio since 1 64b write doesn't work */
158	ioat_chan->completion_virt =
159		pci_pool_alloc(ioat_chan->device->completion_pool,
160		               GFP_KERNEL,
161		               &ioat_chan->completion_addr);
162	memset(ioat_chan->completion_virt, 0,
163	       sizeof(*ioat_chan->completion_virt));
164	ioatdma_chan_write32(ioat_chan, IOAT_CHANCMP_OFFSET_LOW,
165	               ((u64) ioat_chan->completion_addr) & 0x00000000FFFFFFFF);
166	ioatdma_chan_write32(ioat_chan, IOAT_CHANCMP_OFFSET_HIGH,
167	               ((u64) ioat_chan->completion_addr) >> 32);
168
169	ioat_start_null_desc(ioat_chan);
170	return i;
171}
172
173static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat_chan);
174
175static void ioat_dma_free_chan_resources(struct dma_chan *chan)
176{
177	struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
178	struct ioat_device *ioat_device = to_ioat_device(chan->device);
179	struct ioat_desc_sw *desc, *_desc;
180	u16 chanctrl;
181	int in_use_descs = 0;
182
183	ioat_dma_memcpy_cleanup(ioat_chan);
184
185	ioatdma_chan_write8(ioat_chan, IOAT_CHANCMD_OFFSET, IOAT_CHANCMD_RESET);
186
187	spin_lock_bh(&ioat_chan->desc_lock);
188	list_for_each_entry_safe(desc, _desc, &ioat_chan->used_desc, node) {
189		in_use_descs++;
190		list_del(&desc->node);
191		pci_pool_free(ioat_device->dma_pool, desc->hw, desc->phys);
192		kfree(desc);
193	}
194	list_for_each_entry_safe(desc, _desc, &ioat_chan->free_desc, node) {
195		list_del(&desc->node);
196		pci_pool_free(ioat_device->dma_pool, desc->hw, desc->phys);
197		kfree(desc);
198	}
199	spin_unlock_bh(&ioat_chan->desc_lock);
200
201	pci_pool_free(ioat_device->completion_pool,
202	              ioat_chan->completion_virt,
203	              ioat_chan->completion_addr);
204
205	/* one is ok since we left it on there on purpose */
206	if (in_use_descs > 1)
207		printk(KERN_ERR "IOAT: Freeing %d in use descriptors!\n",
208			in_use_descs - 1);
209
210	ioat_chan->last_completion = ioat_chan->completion_addr = 0;
211
212	/* Tell hw the chan is free */
213	chanctrl = ioatdma_chan_read16(ioat_chan, IOAT_CHANCTRL_OFFSET);
214	chanctrl &= ~IOAT_CHANCTRL_CHANNEL_IN_USE;
215	ioatdma_chan_write16(ioat_chan, IOAT_CHANCTRL_OFFSET, chanctrl);
216}
217
218/**
219 * do_ioat_dma_memcpy - actual function that initiates a IOAT DMA transaction
220 * @ioat_chan: IOAT DMA channel handle
221 * @dest: DMA destination address
222 * @src: DMA source address
223 * @len: transaction length in bytes
224 */
225
226static dma_cookie_t do_ioat_dma_memcpy(struct ioat_dma_chan *ioat_chan,
227                                       dma_addr_t dest,
228                                       dma_addr_t src,
229                                       size_t len)
230{
231	struct ioat_desc_sw *first;
232	struct ioat_desc_sw *prev;
233	struct ioat_desc_sw *new;
234	dma_cookie_t cookie;
235	LIST_HEAD(new_chain);
236	u32 copy;
237	size_t orig_len;
238	dma_addr_t orig_src, orig_dst;
239	unsigned int desc_count = 0;
240	unsigned int append = 0;
241
242	if (!ioat_chan || !dest || !src)
243		return -EFAULT;
244
245	if (!len)
246		return ioat_chan->common.cookie;
247
248	orig_len = len;
249	orig_src = src;
250	orig_dst = dest;
251
252	first = NULL;
253	prev = NULL;
254
255	spin_lock_bh(&ioat_chan->desc_lock);
256
257	while (len) {
258		if (!list_empty(&ioat_chan->free_desc)) {
259			new = to_ioat_desc(ioat_chan->free_desc.next);
260			list_del(&new->node);
261		} else {
262			/* try to get another desc */
263			new = ioat_dma_alloc_descriptor(ioat_chan, GFP_ATOMIC);
264			/* will this ever happen? */
265			/* TODO add upper limit on these */
266			BUG_ON(!new);
267		}
268
269		copy = min((u32) len, ioat_chan->xfercap);
270
271		new->hw->size = copy;
272		new->hw->ctl = 0;
273		new->hw->src_addr = src;
274		new->hw->dst_addr = dest;
275		new->cookie = 0;
276
277		/* chain together the physical address list for the HW */
278		if (!first)
279			first = new;
280		else
281			prev->hw->next = (u64) new->phys;
282
283		prev = new;
284
285		len  -= copy;
286		dest += copy;
287		src  += copy;
288
289		list_add_tail(&new->node, &new_chain);
290		desc_count++;
291	}
292	new->hw->ctl = IOAT_DMA_DESCRIPTOR_CTL_CP_STS;
293	new->hw->next = 0;
294
295	/* cookie incr and addition to used_list must be atomic */
296
297	cookie = ioat_chan->common.cookie;
298	cookie++;
299	if (cookie < 0)
300		cookie = 1;
301	ioat_chan->common.cookie = new->cookie = cookie;
302
303	pci_unmap_addr_set(new, src, orig_src);
304	pci_unmap_addr_set(new, dst, orig_dst);
305	pci_unmap_len_set(new, src_len, orig_len);
306	pci_unmap_len_set(new, dst_len, orig_len);
307
308	/* write address into NextDescriptor field of last desc in chain */
309	to_ioat_desc(ioat_chan->used_desc.prev)->hw->next = first->phys;
310	list_splice_init(&new_chain, ioat_chan->used_desc.prev);
311
312	ioat_chan->pending += desc_count;
313	if (ioat_chan->pending >= 20) {
314		append = 1;
315		ioat_chan->pending = 0;
316	}
317
318	spin_unlock_bh(&ioat_chan->desc_lock);
319
320	if (append)
321		ioatdma_chan_write8(ioat_chan,
322		                    IOAT_CHANCMD_OFFSET,
323		                    IOAT_CHANCMD_APPEND);
324	return cookie;
325}
326
327/**
328 * ioat_dma_memcpy_buf_to_buf - wrapper that takes src & dest bufs
329 * @chan: IOAT DMA channel handle
330 * @dest: DMA destination address
331 * @src: DMA source address
332 * @len: transaction length in bytes
333 */
334
335static dma_cookie_t ioat_dma_memcpy_buf_to_buf(struct dma_chan *chan,
336                                               void *dest,
337                                               void *src,
338                                               size_t len)
339{
340	dma_addr_t dest_addr;
341	dma_addr_t src_addr;
342	struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
343
344	dest_addr = pci_map_single(ioat_chan->device->pdev,
345		dest, len, PCI_DMA_FROMDEVICE);
346	src_addr = pci_map_single(ioat_chan->device->pdev,
347		src, len, PCI_DMA_TODEVICE);
348
349	return do_ioat_dma_memcpy(ioat_chan, dest_addr, src_addr, len);
350}
351
352/**
353 * ioat_dma_memcpy_buf_to_pg - wrapper, copying from a buf to a page
354 * @chan: IOAT DMA channel handle
355 * @page: pointer to the page to copy to
356 * @offset: offset into that page
357 * @src: DMA source address
358 * @len: transaction length in bytes
359 */
360
361static dma_cookie_t ioat_dma_memcpy_buf_to_pg(struct dma_chan *chan,
362                                              struct page *page,
363                                              unsigned int offset,
364                                              void *src,
365                                              size_t len)
366{
367	dma_addr_t dest_addr;
368	dma_addr_t src_addr;
369	struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
370
371	dest_addr = pci_map_page(ioat_chan->device->pdev,
372		page, offset, len, PCI_DMA_FROMDEVICE);
373	src_addr = pci_map_single(ioat_chan->device->pdev,
374		src, len, PCI_DMA_TODEVICE);
375
376	return do_ioat_dma_memcpy(ioat_chan, dest_addr, src_addr, len);
377}
378
379/**
380 * ioat_dma_memcpy_pg_to_pg - wrapper, copying between two pages
381 * @chan: IOAT DMA channel handle
382 * @dest_pg: pointer to the page to copy to
383 * @dest_off: offset into that page
384 * @src_pg: pointer to the page to copy from
385 * @src_off: offset into that page
386 * @len: transaction length in bytes. This is guaranteed not to make a copy
387 *	 across a page boundary.
388 */
389
390static dma_cookie_t ioat_dma_memcpy_pg_to_pg(struct dma_chan *chan,
391                                             struct page *dest_pg,
392                                             unsigned int dest_off,
393                                             struct page *src_pg,
394                                             unsigned int src_off,
395                                             size_t len)
396{
397	dma_addr_t dest_addr;
398	dma_addr_t src_addr;
399	struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
400
401	dest_addr = pci_map_page(ioat_chan->device->pdev,
402		dest_pg, dest_off, len, PCI_DMA_FROMDEVICE);
403	src_addr = pci_map_page(ioat_chan->device->pdev,
404		src_pg, src_off, len, PCI_DMA_TODEVICE);
405
406	return do_ioat_dma_memcpy(ioat_chan, dest_addr, src_addr, len);
407}
408
409/**
410 * ioat_dma_memcpy_issue_pending - push potentially unrecognized appended descriptors to hw
411 * @chan: DMA channel handle
412 */
413
414static void ioat_dma_memcpy_issue_pending(struct dma_chan *chan)
415{
416	struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
417
418	if (ioat_chan->pending != 0) {
419		ioat_chan->pending = 0;
420		ioatdma_chan_write8(ioat_chan,
421		                    IOAT_CHANCMD_OFFSET,
422		                    IOAT_CHANCMD_APPEND);
423	}
424}
425
426static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *chan)
427{
428	unsigned long phys_complete;
429	struct ioat_desc_sw *desc, *_desc;
430	dma_cookie_t cookie = 0;
431
432	prefetch(chan->completion_virt);
433
434	if (!spin_trylock(&chan->cleanup_lock))
435		return;
436
437	/* The completion writeback can happen at any time,
438	   so reads by the driver need to be atomic operations
439	   The descriptor physical addresses are limited to 32-bits
440	   when the CPU can only do a 32-bit mov */
441
442#if (BITS_PER_LONG == 64)
443	phys_complete =
444	chan->completion_virt->full & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR;
445#else
446	phys_complete = chan->completion_virt->low & IOAT_LOW_COMPLETION_MASK;
447#endif
448
449	if ((chan->completion_virt->full & IOAT_CHANSTS_DMA_TRANSFER_STATUS) ==
450		IOAT_CHANSTS_DMA_TRANSFER_STATUS_HALTED) {
451		printk("IOAT: Channel halted, chanerr = %x\n",
452			ioatdma_chan_read32(chan, IOAT_CHANERR_OFFSET));
453
454		/* TODO do something to salvage the situation */
455	}
456
457	if (phys_complete == chan->last_completion) {
458		spin_unlock(&chan->cleanup_lock);
459		return;
460	}
461
462	spin_lock_bh(&chan->desc_lock);
463	list_for_each_entry_safe(desc, _desc, &chan->used_desc, node) {
464
465		/*
466		 * Incoming DMA requests may use multiple descriptors, due to
467		 * exceeding xfercap, perhaps. If so, only the last one will
468		 * have a cookie, and require unmapping.
469		 */
470		if (desc->cookie) {
471			cookie = desc->cookie;
472
473			/* yes we are unmapping both _page and _single alloc'd
474			   regions with unmap_page. Is this *really* that bad?
475			*/
476			pci_unmap_page(chan->device->pdev,
477					pci_unmap_addr(desc, dst),
478					pci_unmap_len(desc, dst_len),
479					PCI_DMA_FROMDEVICE);
480			pci_unmap_page(chan->device->pdev,
481					pci_unmap_addr(desc, src),
482					pci_unmap_len(desc, src_len),
483					PCI_DMA_TODEVICE);
484		}
485
486		if (desc->phys != phys_complete) {
487			/* a completed entry, but not the last, so cleanup */
488			list_del(&desc->node);
489			list_add_tail(&desc->node, &chan->free_desc);
490		} else {
491			/* last used desc. Do not remove, so we can append from
492			   it, but don't look at it next time, either */
493			desc->cookie = 0;
494
495			/* TODO check status bits? */
496			break;
497		}
498	}
499
500	spin_unlock_bh(&chan->desc_lock);
501
502	chan->last_completion = phys_complete;
503	if (cookie != 0)
504		chan->completed_cookie = cookie;
505
506	spin_unlock(&chan->cleanup_lock);
507}
508
509/**
510 * ioat_dma_is_complete - poll the status of a IOAT DMA transaction
511 * @chan: IOAT DMA channel handle
512 * @cookie: DMA transaction identifier
513 * @done: if not %NULL, updated with last completed transaction
514 * @used: if not %NULL, updated with last used transaction
515 */
516
517static enum dma_status ioat_dma_is_complete(struct dma_chan *chan,
518                                            dma_cookie_t cookie,
519                                            dma_cookie_t *done,
520                                            dma_cookie_t *used)
521{
522	struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
523	dma_cookie_t last_used;
524	dma_cookie_t last_complete;
525	enum dma_status ret;
526
527	last_used = chan->cookie;
528	last_complete = ioat_chan->completed_cookie;
529
530	if (done)
531		*done= last_complete;
532	if (used)
533		*used = last_used;
534
535	ret = dma_async_is_complete(cookie, last_complete, last_used);
536	if (ret == DMA_SUCCESS)
537		return ret;
538
539	ioat_dma_memcpy_cleanup(ioat_chan);
540
541	last_used = chan->cookie;
542	last_complete = ioat_chan->completed_cookie;
543
544	if (done)
545		*done= last_complete;
546	if (used)
547		*used = last_used;
548
549	return dma_async_is_complete(cookie, last_complete, last_used);
550}
551
552/* PCI API */
553
554static struct pci_device_id ioat_pci_tbl[] = {
555	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT) },
556	{ 0, }
557};
558
559static struct pci_driver ioat_pci_driver = {
560	.name 	= "ioatdma",
561	.id_table = ioat_pci_tbl,
562	.probe	= ioat_probe,
563	.remove	= __devexit_p(ioat_remove),
564};
565
566static irqreturn_t ioat_do_interrupt(int irq, void *data)
567{
568	struct ioat_device *instance = data;
569	unsigned long attnstatus;
570	u8 intrctrl;
571
572	intrctrl = ioatdma_read8(instance, IOAT_INTRCTRL_OFFSET);
573
574	if (!(intrctrl & IOAT_INTRCTRL_MASTER_INT_EN))
575		return IRQ_NONE;
576
577	if (!(intrctrl & IOAT_INTRCTRL_INT_STATUS)) {
578		ioatdma_write8(instance, IOAT_INTRCTRL_OFFSET, intrctrl);
579		return IRQ_NONE;
580	}
581
582	attnstatus = ioatdma_read32(instance, IOAT_ATTNSTATUS_OFFSET);
583
584	printk(KERN_ERR "ioatdma error: interrupt! status %lx\n", attnstatus);
585
586	ioatdma_write8(instance, IOAT_INTRCTRL_OFFSET, intrctrl);
587	return IRQ_HANDLED;
588}
589
590static void ioat_start_null_desc(struct ioat_dma_chan *ioat_chan)
591{
592	struct ioat_desc_sw *desc;
593
594	spin_lock_bh(&ioat_chan->desc_lock);
595
596	if (!list_empty(&ioat_chan->free_desc)) {
597		desc = to_ioat_desc(ioat_chan->free_desc.next);
598		list_del(&desc->node);
599	} else {
600		/* try to get another desc */
601		spin_unlock_bh(&ioat_chan->desc_lock);
602		desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_KERNEL);
603		spin_lock_bh(&ioat_chan->desc_lock);
604		/* will this ever happen? */
605		BUG_ON(!desc);
606	}
607
608	desc->hw->ctl = IOAT_DMA_DESCRIPTOR_NUL;
609	desc->hw->next = 0;
610
611	list_add_tail(&desc->node, &ioat_chan->used_desc);
612	spin_unlock_bh(&ioat_chan->desc_lock);
613
614#if (BITS_PER_LONG == 64)
615	ioatdma_chan_write64(ioat_chan, IOAT_CHAINADDR_OFFSET, desc->phys);
616#else
617	ioatdma_chan_write32(ioat_chan,
618	                     IOAT_CHAINADDR_OFFSET_LOW,
619	                     (u32) desc->phys);
620	ioatdma_chan_write32(ioat_chan, IOAT_CHAINADDR_OFFSET_HIGH, 0);
621#endif
622	ioatdma_chan_write8(ioat_chan, IOAT_CHANCMD_OFFSET, IOAT_CHANCMD_START);
623}
624
625/*
626 * Perform a IOAT transaction to verify the HW works.
627 */
628#define IOAT_TEST_SIZE 2000
629
630static int ioat_self_test(struct ioat_device *device)
631{
632	int i;
633	u8 *src;
634	u8 *dest;
635	struct dma_chan *dma_chan;
636	dma_cookie_t cookie;
637	int err = 0;
638
639	src = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, GFP_KERNEL);
640	if (!src)
641		return -ENOMEM;
642	dest = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, GFP_KERNEL);
643	if (!dest) {
644		kfree(src);
645		return -ENOMEM;
646	}
647
648	/* Fill in src buffer */
649	for (i = 0; i < IOAT_TEST_SIZE; i++)
650		src[i] = (u8)i;
651
652	/* Start copy, using first DMA channel */
653	dma_chan = container_of(device->common.channels.next,
654	                        struct dma_chan,
655	                        device_node);
656	if (ioat_dma_alloc_chan_resources(dma_chan) < 1) {
657		err = -ENODEV;
658		goto out;
659	}
660
661	cookie = ioat_dma_memcpy_buf_to_buf(dma_chan, dest, src, IOAT_TEST_SIZE);
662	ioat_dma_memcpy_issue_pending(dma_chan);
663	msleep(1);
664
665	if (ioat_dma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
666		printk(KERN_ERR "ioatdma: Self-test copy timed out, disabling\n");
667		err = -ENODEV;
668		goto free_resources;
669	}
670	if (memcmp(src, dest, IOAT_TEST_SIZE)) {
671		printk(KERN_ERR "ioatdma: Self-test copy failed compare, disabling\n");
672		err = -ENODEV;
673		goto free_resources;
674	}
675
676free_resources:
677	ioat_dma_free_chan_resources(dma_chan);
678out:
679	kfree(src);
680	kfree(dest);
681	return err;
682}
683
684static int __devinit ioat_probe(struct pci_dev *pdev,
685                                const struct pci_device_id *ent)
686{
687	int err;
688	unsigned long mmio_start, mmio_len;
689	void __iomem *reg_base;
690	struct ioat_device *device;
691
692	err = pci_enable_device(pdev);
693	if (err)
694		goto err_enable_device;
695
696	err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
697	if (err)
698		err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
699	if (err)
700		goto err_set_dma_mask;
701
702	err = pci_request_regions(pdev, ioat_pci_driver.name);
703	if (err)
704		goto err_request_regions;
705
706	mmio_start = pci_resource_start(pdev, 0);
707	mmio_len = pci_resource_len(pdev, 0);
708
709	reg_base = ioremap(mmio_start, mmio_len);
710	if (!reg_base) {
711		err = -ENOMEM;
712		goto err_ioremap;
713	}
714
715	device = kzalloc(sizeof(*device), GFP_KERNEL);
716	if (!device) {
717		err = -ENOMEM;
718		goto err_kzalloc;
719	}
720
721	/* DMA coherent memory pool for DMA descriptor allocations */
722	device->dma_pool = pci_pool_create("dma_desc_pool", pdev,
723		sizeof(struct ioat_dma_descriptor), 64, 0);
724	if (!device->dma_pool) {
725		err = -ENOMEM;
726		goto err_dma_pool;
727	}
728
729	device->completion_pool = pci_pool_create("completion_pool", pdev, sizeof(u64), SMP_CACHE_BYTES, SMP_CACHE_BYTES);
730	if (!device->completion_pool) {
731		err = -ENOMEM;
732		goto err_completion_pool;
733	}
734
735	device->pdev = pdev;
736	pci_set_drvdata(pdev, device);
737#ifdef CONFIG_PCI_MSI
738	if (pci_enable_msi(pdev) == 0) {
739		device->msi = 1;
740	} else {
741		device->msi = 0;
742	}
743#endif
744	err = request_irq(pdev->irq, &ioat_do_interrupt, IRQF_SHARED, "ioat",
745		device);
746	if (err)
747		goto err_irq;
748
749	device->reg_base = reg_base;
750
751	ioatdma_write8(device, IOAT_INTRCTRL_OFFSET, IOAT_INTRCTRL_MASTER_INT_EN);
752	pci_set_master(pdev);
753
754	INIT_LIST_HEAD(&device->common.channels);
755	enumerate_dma_channels(device);
756
757	device->common.device_alloc_chan_resources = ioat_dma_alloc_chan_resources;
758	device->common.device_free_chan_resources = ioat_dma_free_chan_resources;
759	device->common.device_memcpy_buf_to_buf = ioat_dma_memcpy_buf_to_buf;
760	device->common.device_memcpy_buf_to_pg = ioat_dma_memcpy_buf_to_pg;
761	device->common.device_memcpy_pg_to_pg = ioat_dma_memcpy_pg_to_pg;
762	device->common.device_memcpy_complete = ioat_dma_is_complete;
763	device->common.device_memcpy_issue_pending = ioat_dma_memcpy_issue_pending;
764	printk(KERN_INFO "Intel(R) I/OAT DMA Engine found, %d channels\n",
765		device->common.chancnt);
766
767	err = ioat_self_test(device);
768	if (err)
769		goto err_self_test;
770
771	dma_async_device_register(&device->common);
772
773	return 0;
774
775err_self_test:
776err_irq:
777	pci_pool_destroy(device->completion_pool);
778err_completion_pool:
779	pci_pool_destroy(device->dma_pool);
780err_dma_pool:
781	kfree(device);
782err_kzalloc:
783	iounmap(reg_base);
784err_ioremap:
785	pci_release_regions(pdev);
786err_request_regions:
787err_set_dma_mask:
788	pci_disable_device(pdev);
789err_enable_device:
790	return err;
791}
792
793static void __devexit ioat_remove(struct pci_dev *pdev)
794{
795	struct ioat_device *device;
796	struct dma_chan *chan, *_chan;
797	struct ioat_dma_chan *ioat_chan;
798
799	device = pci_get_drvdata(pdev);
800	dma_async_device_unregister(&device->common);
801
802	free_irq(device->pdev->irq, device);
803#ifdef CONFIG_PCI_MSI
804	if (device->msi)
805		pci_disable_msi(device->pdev);
806#endif
807	pci_pool_destroy(device->dma_pool);
808	pci_pool_destroy(device->completion_pool);
809	iounmap(device->reg_base);
810	pci_release_regions(pdev);
811	pci_disable_device(pdev);
812	list_for_each_entry_safe(chan, _chan, &device->common.channels, device_node) {
813		ioat_chan = to_ioat_chan(chan);
814		list_del(&chan->device_node);
815		kfree(ioat_chan);
816	}
817	kfree(device);
818}
819
820/* MODULE API */
821MODULE_VERSION("1.7");
822MODULE_LICENSE("GPL");
823MODULE_AUTHOR("Intel Corporation");
824
825static int __init ioat_init_module(void)
826{
827	/* it's currently unsafe to unload this module */
828	/* if forced, worst case is that rmmod hangs */
829	__unsafe(THIS_MODULE);
830
831	return pci_register_driver(&ioat_pci_driver);
832}
833
834module_init(ioat_init_module);
835
836static void __exit ioat_exit_module(void)
837{
838	pci_unregister_driver(&ioat_pci_driver);
839}
840
841module_exit(ioat_exit_module);
842