diff options
Diffstat (limited to 'drivers/dma/ioatdma.c')
-rw-r--r-- | drivers/dma/ioatdma.c | 840 |
1 files changed, 840 insertions, 0 deletions
diff --git a/drivers/dma/ioatdma.c b/drivers/dma/ioatdma.c new file mode 100644 index 000000000000..0fdf7fbd6495 --- /dev/null +++ b/drivers/dma/ioatdma.c | |||
@@ -0,0 +1,840 @@ | |||
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 */ | ||
44 | static int __devinit ioat_probe(struct pci_dev *pdev, const struct pci_device_id *ent); | ||
45 | static void __devexit ioat_remove(struct pci_dev *pdev); | ||
46 | |||
47 | static 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 | |||
81 | static struct ioat_desc_sw *ioat_dma_alloc_descriptor( | ||
82 | struct ioat_dma_chan *ioat_chan, | ||
83 | int 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 | |||
110 | static void ioat_start_null_desc(struct ioat_dma_chan *ioat_chan); | ||
111 | |||
112 | /* returns the actual number of allocated descriptors */ | ||
113 | static 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 | |||
173 | static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat_chan); | ||
174 | |||
175 | static 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 | * @chan: IOAT DMA channel handle | ||
221 | * @dest: DMA destination address | ||
222 | * @src: DMA source address | ||
223 | * @len: transaction length in bytes | ||
224 | */ | ||
225 | |||
226 | static 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 | |||
335 | static 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 | |||
361 | static 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 to not make a copy | ||
387 | * across a page boundary. | ||
388 | */ | ||
389 | |||
390 | static 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 unrecognoized appended descriptors to hw | ||
411 | * @chan: DMA channel handle | ||
412 | */ | ||
413 | |||
414 | static 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 | |||
426 | static 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 | */ | ||
514 | |||
515 | static enum dma_status ioat_dma_is_complete(struct dma_chan *chan, | ||
516 | dma_cookie_t cookie, | ||
517 | dma_cookie_t *done, | ||
518 | dma_cookie_t *used) | ||
519 | { | ||
520 | struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan); | ||
521 | dma_cookie_t last_used; | ||
522 | dma_cookie_t last_complete; | ||
523 | enum dma_status ret; | ||
524 | |||
525 | last_used = chan->cookie; | ||
526 | last_complete = ioat_chan->completed_cookie; | ||
527 | |||
528 | if (done) | ||
529 | *done= last_complete; | ||
530 | if (used) | ||
531 | *used = last_used; | ||
532 | |||
533 | ret = dma_async_is_complete(cookie, last_complete, last_used); | ||
534 | if (ret == DMA_SUCCESS) | ||
535 | return ret; | ||
536 | |||
537 | ioat_dma_memcpy_cleanup(ioat_chan); | ||
538 | |||
539 | last_used = chan->cookie; | ||
540 | last_complete = ioat_chan->completed_cookie; | ||
541 | |||
542 | if (done) | ||
543 | *done= last_complete; | ||
544 | if (used) | ||
545 | *used = last_used; | ||
546 | |||
547 | return dma_async_is_complete(cookie, last_complete, last_used); | ||
548 | } | ||
549 | |||
550 | /* PCI API */ | ||
551 | |||
552 | static struct pci_device_id ioat_pci_tbl[] = { | ||
553 | { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT) }, | ||
554 | { 0, } | ||
555 | }; | ||
556 | |||
557 | static struct pci_driver ioat_pci_drv = { | ||
558 | .name = "ioatdma", | ||
559 | .id_table = ioat_pci_tbl, | ||
560 | .probe = ioat_probe, | ||
561 | .remove = __devexit_p(ioat_remove), | ||
562 | }; | ||
563 | |||
564 | static irqreturn_t ioat_do_interrupt(int irq, void *data, struct pt_regs *regs) | ||
565 | { | ||
566 | struct ioat_device *instance = data; | ||
567 | unsigned long attnstatus; | ||
568 | u8 intrctrl; | ||
569 | |||
570 | intrctrl = ioatdma_read8(instance, IOAT_INTRCTRL_OFFSET); | ||
571 | |||
572 | if (!(intrctrl & IOAT_INTRCTRL_MASTER_INT_EN)) | ||
573 | return IRQ_NONE; | ||
574 | |||
575 | if (!(intrctrl & IOAT_INTRCTRL_INT_STATUS)) { | ||
576 | ioatdma_write8(instance, IOAT_INTRCTRL_OFFSET, intrctrl); | ||
577 | return IRQ_NONE; | ||
578 | } | ||
579 | |||
580 | attnstatus = ioatdma_read32(instance, IOAT_ATTNSTATUS_OFFSET); | ||
581 | |||
582 | printk(KERN_ERR "ioatdma error: interrupt! status %lx\n", attnstatus); | ||
583 | |||
584 | ioatdma_write8(instance, IOAT_INTRCTRL_OFFSET, intrctrl); | ||
585 | return IRQ_HANDLED; | ||
586 | } | ||
587 | |||
588 | static void ioat_start_null_desc(struct ioat_dma_chan *ioat_chan) | ||
589 | { | ||
590 | struct ioat_desc_sw *desc; | ||
591 | |||
592 | spin_lock_bh(&ioat_chan->desc_lock); | ||
593 | |||
594 | if (!list_empty(&ioat_chan->free_desc)) { | ||
595 | desc = to_ioat_desc(ioat_chan->free_desc.next); | ||
596 | list_del(&desc->node); | ||
597 | } else { | ||
598 | /* try to get another desc */ | ||
599 | spin_unlock_bh(&ioat_chan->desc_lock); | ||
600 | desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_KERNEL); | ||
601 | spin_lock_bh(&ioat_chan->desc_lock); | ||
602 | /* will this ever happen? */ | ||
603 | BUG_ON(!desc); | ||
604 | } | ||
605 | |||
606 | desc->hw->ctl = IOAT_DMA_DESCRIPTOR_NUL; | ||
607 | desc->hw->next = 0; | ||
608 | |||
609 | list_add_tail(&desc->node, &ioat_chan->used_desc); | ||
610 | spin_unlock_bh(&ioat_chan->desc_lock); | ||
611 | |||
612 | #if (BITS_PER_LONG == 64) | ||
613 | ioatdma_chan_write64(ioat_chan, IOAT_CHAINADDR_OFFSET, desc->phys); | ||
614 | #else | ||
615 | ioatdma_chan_write32(ioat_chan, | ||
616 | IOAT_CHAINADDR_OFFSET_LOW, | ||
617 | (u32) desc->phys); | ||
618 | ioatdma_chan_write32(ioat_chan, IOAT_CHAINADDR_OFFSET_HIGH, 0); | ||
619 | #endif | ||
620 | ioatdma_chan_write8(ioat_chan, IOAT_CHANCMD_OFFSET, IOAT_CHANCMD_START); | ||
621 | } | ||
622 | |||
623 | /* | ||
624 | * Perform a IOAT transaction to verify the HW works. | ||
625 | */ | ||
626 | #define IOAT_TEST_SIZE 2000 | ||
627 | |||
628 | static int ioat_self_test(struct ioat_device *device) | ||
629 | { | ||
630 | int i; | ||
631 | u8 *src; | ||
632 | u8 *dest; | ||
633 | struct dma_chan *dma_chan; | ||
634 | dma_cookie_t cookie; | ||
635 | int err = 0; | ||
636 | |||
637 | src = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, SLAB_KERNEL); | ||
638 | if (!src) | ||
639 | return -ENOMEM; | ||
640 | dest = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, SLAB_KERNEL); | ||
641 | if (!dest) { | ||
642 | kfree(src); | ||
643 | return -ENOMEM; | ||
644 | } | ||
645 | |||
646 | /* Fill in src buffer */ | ||
647 | for (i = 0; i < IOAT_TEST_SIZE; i++) | ||
648 | src[i] = (u8)i; | ||
649 | |||
650 | /* Start copy, using first DMA channel */ | ||
651 | dma_chan = container_of(device->common.channels.next, | ||
652 | struct dma_chan, | ||
653 | device_node); | ||
654 | if (ioat_dma_alloc_chan_resources(dma_chan) < 1) { | ||
655 | err = -ENODEV; | ||
656 | goto out; | ||
657 | } | ||
658 | |||
659 | cookie = ioat_dma_memcpy_buf_to_buf(dma_chan, dest, src, IOAT_TEST_SIZE); | ||
660 | ioat_dma_memcpy_issue_pending(dma_chan); | ||
661 | msleep(1); | ||
662 | |||
663 | if (ioat_dma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) { | ||
664 | printk(KERN_ERR "ioatdma: Self-test copy timed out, disabling\n"); | ||
665 | err = -ENODEV; | ||
666 | goto free_resources; | ||
667 | } | ||
668 | if (memcmp(src, dest, IOAT_TEST_SIZE)) { | ||
669 | printk(KERN_ERR "ioatdma: Self-test copy failed compare, disabling\n"); | ||
670 | err = -ENODEV; | ||
671 | goto free_resources; | ||
672 | } | ||
673 | |||
674 | free_resources: | ||
675 | ioat_dma_free_chan_resources(dma_chan); | ||
676 | out: | ||
677 | kfree(src); | ||
678 | kfree(dest); | ||
679 | return err; | ||
680 | } | ||
681 | |||
682 | static int __devinit ioat_probe(struct pci_dev *pdev, | ||
683 | const struct pci_device_id *ent) | ||
684 | { | ||
685 | int err; | ||
686 | unsigned long mmio_start, mmio_len; | ||
687 | void *reg_base; | ||
688 | struct ioat_device *device; | ||
689 | |||
690 | err = pci_enable_device(pdev); | ||
691 | if (err) | ||
692 | goto err_enable_device; | ||
693 | |||
694 | err = pci_set_dma_mask(pdev, DMA_64BIT_MASK); | ||
695 | if (err) | ||
696 | err = pci_set_dma_mask(pdev, DMA_32BIT_MASK); | ||
697 | if (err) | ||
698 | goto err_set_dma_mask; | ||
699 | |||
700 | err = pci_request_regions(pdev, ioat_pci_drv.name); | ||
701 | if (err) | ||
702 | goto err_request_regions; | ||
703 | |||
704 | mmio_start = pci_resource_start(pdev, 0); | ||
705 | mmio_len = pci_resource_len(pdev, 0); | ||
706 | |||
707 | reg_base = ioremap(mmio_start, mmio_len); | ||
708 | if (!reg_base) { | ||
709 | err = -ENOMEM; | ||
710 | goto err_ioremap; | ||
711 | } | ||
712 | |||
713 | device = kzalloc(sizeof(*device), GFP_KERNEL); | ||
714 | if (!device) { | ||
715 | err = -ENOMEM; | ||
716 | goto err_kzalloc; | ||
717 | } | ||
718 | |||
719 | /* DMA coherent memory pool for DMA descriptor allocations */ | ||
720 | device->dma_pool = pci_pool_create("dma_desc_pool", pdev, | ||
721 | sizeof(struct ioat_dma_descriptor), 64, 0); | ||
722 | if (!device->dma_pool) { | ||
723 | err = -ENOMEM; | ||
724 | goto err_dma_pool; | ||
725 | } | ||
726 | |||
727 | device->completion_pool = pci_pool_create("completion_pool", pdev, sizeof(u64), SMP_CACHE_BYTES, SMP_CACHE_BYTES); | ||
728 | if (!device->completion_pool) { | ||
729 | err = -ENOMEM; | ||
730 | goto err_completion_pool; | ||
731 | } | ||
732 | |||
733 | device->pdev = pdev; | ||
734 | pci_set_drvdata(pdev, device); | ||
735 | #ifdef CONFIG_PCI_MSI | ||
736 | if (pci_enable_msi(pdev) == 0) { | ||
737 | device->msi = 1; | ||
738 | } else { | ||
739 | device->msi = 0; | ||
740 | } | ||
741 | #endif | ||
742 | err = request_irq(pdev->irq, &ioat_do_interrupt, SA_SHIRQ, "ioat", | ||
743 | device); | ||
744 | if (err) | ||
745 | goto err_irq; | ||
746 | |||
747 | device->reg_base = reg_base; | ||
748 | |||
749 | ioatdma_write8(device, IOAT_INTRCTRL_OFFSET, IOAT_INTRCTRL_MASTER_INT_EN); | ||
750 | pci_set_master(pdev); | ||
751 | |||
752 | INIT_LIST_HEAD(&device->common.channels); | ||
753 | enumerate_dma_channels(device); | ||
754 | |||
755 | device->common.device_alloc_chan_resources = ioat_dma_alloc_chan_resources; | ||
756 | device->common.device_free_chan_resources = ioat_dma_free_chan_resources; | ||
757 | device->common.device_memcpy_buf_to_buf = ioat_dma_memcpy_buf_to_buf; | ||
758 | device->common.device_memcpy_buf_to_pg = ioat_dma_memcpy_buf_to_pg; | ||
759 | device->common.device_memcpy_pg_to_pg = ioat_dma_memcpy_pg_to_pg; | ||
760 | device->common.device_memcpy_complete = ioat_dma_is_complete; | ||
761 | device->common.device_memcpy_issue_pending = ioat_dma_memcpy_issue_pending; | ||
762 | printk(KERN_INFO "Intel(R) I/OAT DMA Engine found, %d channels\n", | ||
763 | device->common.chancnt); | ||
764 | |||
765 | err = ioat_self_test(device); | ||
766 | if (err) | ||
767 | goto err_self_test; | ||
768 | |||
769 | dma_async_device_register(&device->common); | ||
770 | |||
771 | return 0; | ||
772 | |||
773 | err_self_test: | ||
774 | err_irq: | ||
775 | pci_pool_destroy(device->completion_pool); | ||
776 | err_completion_pool: | ||
777 | pci_pool_destroy(device->dma_pool); | ||
778 | err_dma_pool: | ||
779 | kfree(device); | ||
780 | err_kzalloc: | ||
781 | iounmap(reg_base); | ||
782 | err_ioremap: | ||
783 | pci_release_regions(pdev); | ||
784 | err_request_regions: | ||
785 | err_set_dma_mask: | ||
786 | pci_disable_device(pdev); | ||
787 | err_enable_device: | ||
788 | return err; | ||
789 | } | ||
790 | |||
791 | static void __devexit ioat_remove(struct pci_dev *pdev) | ||
792 | { | ||
793 | struct ioat_device *device; | ||
794 | struct dma_chan *chan, *_chan; | ||
795 | struct ioat_dma_chan *ioat_chan; | ||
796 | |||
797 | device = pci_get_drvdata(pdev); | ||
798 | dma_async_device_unregister(&device->common); | ||
799 | |||
800 | free_irq(device->pdev->irq, device); | ||
801 | #ifdef CONFIG_PCI_MSI | ||
802 | if (device->msi) | ||
803 | pci_disable_msi(device->pdev); | ||
804 | #endif | ||
805 | pci_pool_destroy(device->dma_pool); | ||
806 | pci_pool_destroy(device->completion_pool); | ||
807 | iounmap(device->reg_base); | ||
808 | pci_release_regions(pdev); | ||
809 | pci_disable_device(pdev); | ||
810 | list_for_each_entry_safe(chan, _chan, &device->common.channels, device_node) { | ||
811 | ioat_chan = to_ioat_chan(chan); | ||
812 | list_del(&chan->device_node); | ||
813 | kfree(ioat_chan); | ||
814 | } | ||
815 | kfree(device); | ||
816 | } | ||
817 | |||
818 | /* MODULE API */ | ||
819 | MODULE_VERSION("1.7"); | ||
820 | MODULE_LICENSE("GPL"); | ||
821 | MODULE_AUTHOR("Intel Corporation"); | ||
822 | |||
823 | static int __init ioat_init_module(void) | ||
824 | { | ||
825 | /* it's currently unsafe to unload this module */ | ||
826 | /* if forced, worst case is that rmmod hangs */ | ||
827 | if (THIS_MODULE != NULL) | ||
828 | THIS_MODULE->unsafe = 1; | ||
829 | |||
830 | return pci_module_init(&ioat_pci_drv); | ||
831 | } | ||
832 | |||
833 | module_init(ioat_init_module); | ||
834 | |||
835 | static void __exit ioat_exit_module(void) | ||
836 | { | ||
837 | pci_unregister_driver(&ioat_pci_drv); | ||
838 | } | ||
839 | |||
840 | module_exit(ioat_exit_module); | ||