diff options
Diffstat (limited to 'drivers/dma/ioatdma.c')
-rw-r--r-- | drivers/dma/ioatdma.c | 839 |
1 files changed, 839 insertions, 0 deletions
diff --git a/drivers/dma/ioatdma.c b/drivers/dma/ioatdma.c new file mode 100644 index 000000000000..11d48b97b36e --- /dev/null +++ b/drivers/dma/ioatdma.c | |||
@@ -0,0 +1,839 @@ | |||
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 "ioatdma.h" | ||
34 | #include "ioatdma_io.h" | ||
35 | #include "ioatdma_registers.h" | ||
36 | #include "ioatdma_hw.h" | ||
37 | |||
38 | #define to_ioat_chan(chan) container_of(chan, struct ioat_dma_chan, common) | ||
39 | #define to_ioat_device(dev) container_of(dev, struct ioat_device, common) | ||
40 | #define to_ioat_desc(lh) container_of(lh, struct ioat_desc_sw, node) | ||
41 | |||
42 | /* internal functions */ | ||
43 | static int __devinit ioat_probe(struct pci_dev *pdev, const struct pci_device_id *ent); | ||
44 | static void __devexit ioat_remove(struct pci_dev *pdev); | ||
45 | |||
46 | static int enumerate_dma_channels(struct ioat_device *device) | ||
47 | { | ||
48 | u8 xfercap_scale; | ||
49 | u32 xfercap; | ||
50 | int i; | ||
51 | struct ioat_dma_chan *ioat_chan; | ||
52 | |||
53 | device->common.chancnt = ioatdma_read8(device, IOAT_CHANCNT_OFFSET); | ||
54 | xfercap_scale = ioatdma_read8(device, IOAT_XFERCAP_OFFSET); | ||
55 | xfercap = (xfercap_scale == 0 ? -1 : (1UL << xfercap_scale)); | ||
56 | |||
57 | for (i = 0; i < device->common.chancnt; i++) { | ||
58 | ioat_chan = kzalloc(sizeof(*ioat_chan), GFP_KERNEL); | ||
59 | if (!ioat_chan) { | ||
60 | device->common.chancnt = i; | ||
61 | break; | ||
62 | } | ||
63 | |||
64 | ioat_chan->device = device; | ||
65 | ioat_chan->reg_base = device->reg_base + (0x80 * (i + 1)); | ||
66 | ioat_chan->xfercap = xfercap; | ||
67 | spin_lock_init(&ioat_chan->cleanup_lock); | ||
68 | spin_lock_init(&ioat_chan->desc_lock); | ||
69 | INIT_LIST_HEAD(&ioat_chan->free_desc); | ||
70 | INIT_LIST_HEAD(&ioat_chan->used_desc); | ||
71 | /* This should be made common somewhere in dmaengine.c */ | ||
72 | ioat_chan->common.device = &device->common; | ||
73 | ioat_chan->common.client = NULL; | ||
74 | list_add_tail(&ioat_chan->common.device_node, | ||
75 | &device->common.channels); | ||
76 | } | ||
77 | return device->common.chancnt; | ||
78 | } | ||
79 | |||
80 | static struct ioat_desc_sw *ioat_dma_alloc_descriptor( | ||
81 | struct ioat_dma_chan *ioat_chan, | ||
82 | int flags) | ||
83 | { | ||
84 | struct ioat_dma_descriptor *desc; | ||
85 | struct ioat_desc_sw *desc_sw; | ||
86 | struct ioat_device *ioat_device; | ||
87 | dma_addr_t phys; | ||
88 | |||
89 | ioat_device = to_ioat_device(ioat_chan->common.device); | ||
90 | desc = pci_pool_alloc(ioat_device->dma_pool, flags, &phys); | ||
91 | if (unlikely(!desc)) | ||
92 | return NULL; | ||
93 | |||
94 | desc_sw = kzalloc(sizeof(*desc_sw), flags); | ||
95 | if (unlikely(!desc_sw)) { | ||
96 | pci_pool_free(ioat_device->dma_pool, desc, phys); | ||
97 | return NULL; | ||
98 | } | ||
99 | |||
100 | memset(desc, 0, sizeof(*desc)); | ||
101 | desc_sw->hw = desc; | ||
102 | desc_sw->phys = phys; | ||
103 | |||
104 | return desc_sw; | ||
105 | } | ||
106 | |||
107 | #define INITIAL_IOAT_DESC_COUNT 128 | ||
108 | |||
109 | static void ioat_start_null_desc(struct ioat_dma_chan *ioat_chan); | ||
110 | |||
111 | /* returns the actual number of allocated descriptors */ | ||
112 | static int ioat_dma_alloc_chan_resources(struct dma_chan *chan) | ||
113 | { | ||
114 | struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan); | ||
115 | struct ioat_desc_sw *desc = NULL; | ||
116 | u16 chanctrl; | ||
117 | u32 chanerr; | ||
118 | int i; | ||
119 | LIST_HEAD(tmp_list); | ||
120 | |||
121 | /* | ||
122 | * In-use bit automatically set by reading chanctrl | ||
123 | * If 0, we got it, if 1, someone else did | ||
124 | */ | ||
125 | chanctrl = ioatdma_chan_read16(ioat_chan, IOAT_CHANCTRL_OFFSET); | ||
126 | if (chanctrl & IOAT_CHANCTRL_CHANNEL_IN_USE) | ||
127 | return -EBUSY; | ||
128 | |||
129 | /* Setup register to interrupt and write completion status on error */ | ||
130 | chanctrl = IOAT_CHANCTRL_CHANNEL_IN_USE | | ||
131 | IOAT_CHANCTRL_ERR_INT_EN | | ||
132 | IOAT_CHANCTRL_ANY_ERR_ABORT_EN | | ||
133 | IOAT_CHANCTRL_ERR_COMPLETION_EN; | ||
134 | ioatdma_chan_write16(ioat_chan, IOAT_CHANCTRL_OFFSET, chanctrl); | ||
135 | |||
136 | chanerr = ioatdma_chan_read32(ioat_chan, IOAT_CHANERR_OFFSET); | ||
137 | if (chanerr) { | ||
138 | printk("IOAT: CHANERR = %x, clearing\n", chanerr); | ||
139 | ioatdma_chan_write32(ioat_chan, IOAT_CHANERR_OFFSET, chanerr); | ||
140 | } | ||
141 | |||
142 | /* Allocate descriptors */ | ||
143 | for (i = 0; i < INITIAL_IOAT_DESC_COUNT; i++) { | ||
144 | desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_KERNEL); | ||
145 | if (!desc) { | ||
146 | printk(KERN_ERR "IOAT: Only %d initial descriptors\n", i); | ||
147 | break; | ||
148 | } | ||
149 | list_add_tail(&desc->node, &tmp_list); | ||
150 | } | ||
151 | spin_lock_bh(&ioat_chan->desc_lock); | ||
152 | list_splice(&tmp_list, &ioat_chan->free_desc); | ||
153 | spin_unlock_bh(&ioat_chan->desc_lock); | ||
154 | |||
155 | /* allocate a completion writeback area */ | ||
156 | /* doing 2 32bit writes to mmio since 1 64b write doesn't work */ | ||
157 | ioat_chan->completion_virt = | ||
158 | pci_pool_alloc(ioat_chan->device->completion_pool, | ||
159 | GFP_KERNEL, | ||
160 | &ioat_chan->completion_addr); | ||
161 | memset(ioat_chan->completion_virt, 0, | ||
162 | sizeof(*ioat_chan->completion_virt)); | ||
163 | ioatdma_chan_write32(ioat_chan, IOAT_CHANCMP_OFFSET_LOW, | ||
164 | ((u64) ioat_chan->completion_addr) & 0x00000000FFFFFFFF); | ||
165 | ioatdma_chan_write32(ioat_chan, IOAT_CHANCMP_OFFSET_HIGH, | ||
166 | ((u64) ioat_chan->completion_addr) >> 32); | ||
167 | |||
168 | ioat_start_null_desc(ioat_chan); | ||
169 | return i; | ||
170 | } | ||
171 | |||
172 | static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat_chan); | ||
173 | |||
174 | static void ioat_dma_free_chan_resources(struct dma_chan *chan) | ||
175 | { | ||
176 | struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan); | ||
177 | struct ioat_device *ioat_device = to_ioat_device(chan->device); | ||
178 | struct ioat_desc_sw *desc, *_desc; | ||
179 | u16 chanctrl; | ||
180 | int in_use_descs = 0; | ||
181 | |||
182 | ioat_dma_memcpy_cleanup(ioat_chan); | ||
183 | |||
184 | ioatdma_chan_write8(ioat_chan, IOAT_CHANCMD_OFFSET, IOAT_CHANCMD_RESET); | ||
185 | |||
186 | spin_lock_bh(&ioat_chan->desc_lock); | ||
187 | list_for_each_entry_safe(desc, _desc, &ioat_chan->used_desc, node) { | ||
188 | in_use_descs++; | ||
189 | list_del(&desc->node); | ||
190 | pci_pool_free(ioat_device->dma_pool, desc->hw, desc->phys); | ||
191 | kfree(desc); | ||
192 | } | ||
193 | list_for_each_entry_safe(desc, _desc, &ioat_chan->free_desc, node) { | ||
194 | list_del(&desc->node); | ||
195 | pci_pool_free(ioat_device->dma_pool, desc->hw, desc->phys); | ||
196 | kfree(desc); | ||
197 | } | ||
198 | spin_unlock_bh(&ioat_chan->desc_lock); | ||
199 | |||
200 | pci_pool_free(ioat_device->completion_pool, | ||
201 | ioat_chan->completion_virt, | ||
202 | ioat_chan->completion_addr); | ||
203 | |||
204 | /* one is ok since we left it on there on purpose */ | ||
205 | if (in_use_descs > 1) | ||
206 | printk(KERN_ERR "IOAT: Freeing %d in use descriptors!\n", | ||
207 | in_use_descs - 1); | ||
208 | |||
209 | ioat_chan->last_completion = ioat_chan->completion_addr = 0; | ||
210 | |||
211 | /* Tell hw the chan is free */ | ||
212 | chanctrl = ioatdma_chan_read16(ioat_chan, IOAT_CHANCTRL_OFFSET); | ||
213 | chanctrl &= ~IOAT_CHANCTRL_CHANNEL_IN_USE; | ||
214 | ioatdma_chan_write16(ioat_chan, IOAT_CHANCTRL_OFFSET, chanctrl); | ||
215 | } | ||
216 | |||
217 | /** | ||
218 | * do_ioat_dma_memcpy - actual function that initiates a IOAT DMA transaction | ||
219 | * @chan: IOAT DMA channel handle | ||
220 | * @dest: DMA destination address | ||
221 | * @src: DMA source address | ||
222 | * @len: transaction length in bytes | ||
223 | */ | ||
224 | |||
225 | static dma_cookie_t do_ioat_dma_memcpy(struct ioat_dma_chan *ioat_chan, | ||
226 | dma_addr_t dest, | ||
227 | dma_addr_t src, | ||
228 | size_t len) | ||
229 | { | ||
230 | struct ioat_desc_sw *first; | ||
231 | struct ioat_desc_sw *prev; | ||
232 | struct ioat_desc_sw *new; | ||
233 | dma_cookie_t cookie; | ||
234 | LIST_HEAD(new_chain); | ||
235 | u32 copy; | ||
236 | size_t orig_len; | ||
237 | dma_addr_t orig_src, orig_dst; | ||
238 | unsigned int desc_count = 0; | ||
239 | unsigned int append = 0; | ||
240 | |||
241 | if (!ioat_chan || !dest || !src) | ||
242 | return -EFAULT; | ||
243 | |||
244 | if (!len) | ||
245 | return ioat_chan->common.cookie; | ||
246 | |||
247 | orig_len = len; | ||
248 | orig_src = src; | ||
249 | orig_dst = dest; | ||
250 | |||
251 | first = NULL; | ||
252 | prev = NULL; | ||
253 | |||
254 | spin_lock_bh(&ioat_chan->desc_lock); | ||
255 | |||
256 | while (len) { | ||
257 | if (!list_empty(&ioat_chan->free_desc)) { | ||
258 | new = to_ioat_desc(ioat_chan->free_desc.next); | ||
259 | list_del(&new->node); | ||
260 | } else { | ||
261 | /* try to get another desc */ | ||
262 | new = ioat_dma_alloc_descriptor(ioat_chan, GFP_ATOMIC); | ||
263 | /* will this ever happen? */ | ||
264 | /* TODO add upper limit on these */ | ||
265 | BUG_ON(!new); | ||
266 | } | ||
267 | |||
268 | copy = min((u32) len, ioat_chan->xfercap); | ||
269 | |||
270 | new->hw->size = copy; | ||
271 | new->hw->ctl = 0; | ||
272 | new->hw->src_addr = src; | ||
273 | new->hw->dst_addr = dest; | ||
274 | new->cookie = 0; | ||
275 | |||
276 | /* chain together the physical address list for the HW */ | ||
277 | if (!first) | ||
278 | first = new; | ||
279 | else | ||
280 | prev->hw->next = (u64) new->phys; | ||
281 | |||
282 | prev = new; | ||
283 | |||
284 | len -= copy; | ||
285 | dest += copy; | ||
286 | src += copy; | ||
287 | |||
288 | list_add_tail(&new->node, &new_chain); | ||
289 | desc_count++; | ||
290 | } | ||
291 | new->hw->ctl = IOAT_DMA_DESCRIPTOR_CTL_CP_STS; | ||
292 | new->hw->next = 0; | ||
293 | |||
294 | /* cookie incr and addition to used_list must be atomic */ | ||
295 | |||
296 | cookie = ioat_chan->common.cookie; | ||
297 | cookie++; | ||
298 | if (cookie < 0) | ||
299 | cookie = 1; | ||
300 | ioat_chan->common.cookie = new->cookie = cookie; | ||
301 | |||
302 | pci_unmap_addr_set(new, src, orig_src); | ||
303 | pci_unmap_addr_set(new, dst, orig_dst); | ||
304 | pci_unmap_len_set(new, src_len, orig_len); | ||
305 | pci_unmap_len_set(new, dst_len, orig_len); | ||
306 | |||
307 | /* write address into NextDescriptor field of last desc in chain */ | ||
308 | to_ioat_desc(ioat_chan->used_desc.prev)->hw->next = first->phys; | ||
309 | list_splice_init(&new_chain, ioat_chan->used_desc.prev); | ||
310 | |||
311 | ioat_chan->pending += desc_count; | ||
312 | if (ioat_chan->pending >= 20) { | ||
313 | append = 1; | ||
314 | ioat_chan->pending = 0; | ||
315 | } | ||
316 | |||
317 | spin_unlock_bh(&ioat_chan->desc_lock); | ||
318 | |||
319 | if (append) | ||
320 | ioatdma_chan_write8(ioat_chan, | ||
321 | IOAT_CHANCMD_OFFSET, | ||
322 | IOAT_CHANCMD_APPEND); | ||
323 | return cookie; | ||
324 | } | ||
325 | |||
326 | /** | ||
327 | * ioat_dma_memcpy_buf_to_buf - wrapper that takes src & dest bufs | ||
328 | * @chan: IOAT DMA channel handle | ||
329 | * @dest: DMA destination address | ||
330 | * @src: DMA source address | ||
331 | * @len: transaction length in bytes | ||
332 | */ | ||
333 | |||
334 | static dma_cookie_t ioat_dma_memcpy_buf_to_buf(struct dma_chan *chan, | ||
335 | void *dest, | ||
336 | void *src, | ||
337 | size_t len) | ||
338 | { | ||
339 | dma_addr_t dest_addr; | ||
340 | dma_addr_t src_addr; | ||
341 | struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan); | ||
342 | |||
343 | dest_addr = pci_map_single(ioat_chan->device->pdev, | ||
344 | dest, len, PCI_DMA_FROMDEVICE); | ||
345 | src_addr = pci_map_single(ioat_chan->device->pdev, | ||
346 | src, len, PCI_DMA_TODEVICE); | ||
347 | |||
348 | return do_ioat_dma_memcpy(ioat_chan, dest_addr, src_addr, len); | ||
349 | } | ||
350 | |||
351 | /** | ||
352 | * ioat_dma_memcpy_buf_to_pg - wrapper, copying from a buf to a page | ||
353 | * @chan: IOAT DMA channel handle | ||
354 | * @page: pointer to the page to copy to | ||
355 | * @offset: offset into that page | ||
356 | * @src: DMA source address | ||
357 | * @len: transaction length in bytes | ||
358 | */ | ||
359 | |||
360 | static dma_cookie_t ioat_dma_memcpy_buf_to_pg(struct dma_chan *chan, | ||
361 | struct page *page, | ||
362 | unsigned int offset, | ||
363 | void *src, | ||
364 | size_t len) | ||
365 | { | ||
366 | dma_addr_t dest_addr; | ||
367 | dma_addr_t src_addr; | ||
368 | struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan); | ||
369 | |||
370 | dest_addr = pci_map_page(ioat_chan->device->pdev, | ||
371 | page, offset, len, PCI_DMA_FROMDEVICE); | ||
372 | src_addr = pci_map_single(ioat_chan->device->pdev, | ||
373 | src, len, PCI_DMA_TODEVICE); | ||
374 | |||
375 | return do_ioat_dma_memcpy(ioat_chan, dest_addr, src_addr, len); | ||
376 | } | ||
377 | |||
378 | /** | ||
379 | * ioat_dma_memcpy_pg_to_pg - wrapper, copying between two pages | ||
380 | * @chan: IOAT DMA channel handle | ||
381 | * @dest_pg: pointer to the page to copy to | ||
382 | * @dest_off: offset into that page | ||
383 | * @src_pg: pointer to the page to copy from | ||
384 | * @src_off: offset into that page | ||
385 | * @len: transaction length in bytes. This is guaranteed to not make a copy | ||
386 | * across a page boundary. | ||
387 | */ | ||
388 | |||
389 | static dma_cookie_t ioat_dma_memcpy_pg_to_pg(struct dma_chan *chan, | ||
390 | struct page *dest_pg, | ||
391 | unsigned int dest_off, | ||
392 | struct page *src_pg, | ||
393 | unsigned int src_off, | ||
394 | size_t len) | ||
395 | { | ||
396 | dma_addr_t dest_addr; | ||
397 | dma_addr_t src_addr; | ||
398 | struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan); | ||
399 | |||
400 | dest_addr = pci_map_page(ioat_chan->device->pdev, | ||
401 | dest_pg, dest_off, len, PCI_DMA_FROMDEVICE); | ||
402 | src_addr = pci_map_page(ioat_chan->device->pdev, | ||
403 | src_pg, src_off, len, PCI_DMA_TODEVICE); | ||
404 | |||
405 | return do_ioat_dma_memcpy(ioat_chan, dest_addr, src_addr, len); | ||
406 | } | ||
407 | |||
408 | /** | ||
409 | * ioat_dma_memcpy_issue_pending - push potentially unrecognoized appended descriptors to hw | ||
410 | * @chan: DMA channel handle | ||
411 | */ | ||
412 | |||
413 | static void ioat_dma_memcpy_issue_pending(struct dma_chan *chan) | ||
414 | { | ||
415 | struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan); | ||
416 | |||
417 | if (ioat_chan->pending != 0) { | ||
418 | ioat_chan->pending = 0; | ||
419 | ioatdma_chan_write8(ioat_chan, | ||
420 | IOAT_CHANCMD_OFFSET, | ||
421 | IOAT_CHANCMD_APPEND); | ||
422 | } | ||
423 | } | ||
424 | |||
425 | static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *chan) | ||
426 | { | ||
427 | unsigned long phys_complete; | ||
428 | struct ioat_desc_sw *desc, *_desc; | ||
429 | dma_cookie_t cookie = 0; | ||
430 | |||
431 | prefetch(chan->completion_virt); | ||
432 | |||
433 | if (!spin_trylock(&chan->cleanup_lock)) | ||
434 | return; | ||
435 | |||
436 | /* The completion writeback can happen at any time, | ||
437 | so reads by the driver need to be atomic operations | ||
438 | The descriptor physical addresses are limited to 32-bits | ||
439 | when the CPU can only do a 32-bit mov */ | ||
440 | |||
441 | #if (BITS_PER_LONG == 64) | ||
442 | phys_complete = | ||
443 | chan->completion_virt->full & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR; | ||
444 | #else | ||
445 | phys_complete = chan->completion_virt->low & IOAT_LOW_COMPLETION_MASK; | ||
446 | #endif | ||
447 | |||
448 | if ((chan->completion_virt->full & IOAT_CHANSTS_DMA_TRANSFER_STATUS) == | ||
449 | IOAT_CHANSTS_DMA_TRANSFER_STATUS_HALTED) { | ||
450 | printk("IOAT: Channel halted, chanerr = %x\n", | ||
451 | ioatdma_chan_read32(chan, IOAT_CHANERR_OFFSET)); | ||
452 | |||
453 | /* TODO do something to salvage the situation */ | ||
454 | } | ||
455 | |||
456 | if (phys_complete == chan->last_completion) { | ||
457 | spin_unlock(&chan->cleanup_lock); | ||
458 | return; | ||
459 | } | ||
460 | |||
461 | spin_lock_bh(&chan->desc_lock); | ||
462 | list_for_each_entry_safe(desc, _desc, &chan->used_desc, node) { | ||
463 | |||
464 | /* | ||
465 | * Incoming DMA requests may use multiple descriptors, due to | ||
466 | * exceeding xfercap, perhaps. If so, only the last one will | ||
467 | * have a cookie, and require unmapping. | ||
468 | */ | ||
469 | if (desc->cookie) { | ||
470 | cookie = desc->cookie; | ||
471 | |||
472 | /* yes we are unmapping both _page and _single alloc'd | ||
473 | regions with unmap_page. Is this *really* that bad? | ||
474 | */ | ||
475 | pci_unmap_page(chan->device->pdev, | ||
476 | pci_unmap_addr(desc, dst), | ||
477 | pci_unmap_len(desc, dst_len), | ||
478 | PCI_DMA_FROMDEVICE); | ||
479 | pci_unmap_page(chan->device->pdev, | ||
480 | pci_unmap_addr(desc, src), | ||
481 | pci_unmap_len(desc, src_len), | ||
482 | PCI_DMA_TODEVICE); | ||
483 | } | ||
484 | |||
485 | if (desc->phys != phys_complete) { | ||
486 | /* a completed entry, but not the last, so cleanup */ | ||
487 | list_del(&desc->node); | ||
488 | list_add_tail(&desc->node, &chan->free_desc); | ||
489 | } else { | ||
490 | /* last used desc. Do not remove, so we can append from | ||
491 | it, but don't look at it next time, either */ | ||
492 | desc->cookie = 0; | ||
493 | |||
494 | /* TODO check status bits? */ | ||
495 | break; | ||
496 | } | ||
497 | } | ||
498 | |||
499 | spin_unlock_bh(&chan->desc_lock); | ||
500 | |||
501 | chan->last_completion = phys_complete; | ||
502 | if (cookie != 0) | ||
503 | chan->completed_cookie = cookie; | ||
504 | |||
505 | spin_unlock(&chan->cleanup_lock); | ||
506 | } | ||
507 | |||
508 | /** | ||
509 | * ioat_dma_is_complete - poll the status of a IOAT DMA transaction | ||
510 | * @chan: IOAT DMA channel handle | ||
511 | * @cookie: DMA transaction identifier | ||
512 | */ | ||
513 | |||
514 | static enum dma_status ioat_dma_is_complete(struct dma_chan *chan, | ||
515 | dma_cookie_t cookie, | ||
516 | dma_cookie_t *done, | ||
517 | dma_cookie_t *used) | ||
518 | { | ||
519 | struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan); | ||
520 | dma_cookie_t last_used; | ||
521 | dma_cookie_t last_complete; | ||
522 | enum dma_status ret; | ||
523 | |||
524 | last_used = chan->cookie; | ||
525 | last_complete = ioat_chan->completed_cookie; | ||
526 | |||
527 | if (done) | ||
528 | *done= last_complete; | ||
529 | if (used) | ||
530 | *used = last_used; | ||
531 | |||
532 | ret = dma_async_is_complete(cookie, last_complete, last_used); | ||
533 | if (ret == DMA_SUCCESS) | ||
534 | return ret; | ||
535 | |||
536 | ioat_dma_memcpy_cleanup(ioat_chan); | ||
537 | |||
538 | last_used = chan->cookie; | ||
539 | last_complete = ioat_chan->completed_cookie; | ||
540 | |||
541 | if (done) | ||
542 | *done= last_complete; | ||
543 | if (used) | ||
544 | *used = last_used; | ||
545 | |||
546 | return dma_async_is_complete(cookie, last_complete, last_used); | ||
547 | } | ||
548 | |||
549 | /* PCI API */ | ||
550 | |||
551 | static struct pci_device_id ioat_pci_tbl[] = { | ||
552 | { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT) }, | ||
553 | { 0, } | ||
554 | }; | ||
555 | |||
556 | static struct pci_driver ioat_pci_drv = { | ||
557 | .name = "ioatdma", | ||
558 | .id_table = ioat_pci_tbl, | ||
559 | .probe = ioat_probe, | ||
560 | .remove = __devexit_p(ioat_remove), | ||
561 | }; | ||
562 | |||
563 | static irqreturn_t ioat_do_interrupt(int irq, void *data, struct pt_regs *regs) | ||
564 | { | ||
565 | struct ioat_device *instance = data; | ||
566 | unsigned long attnstatus; | ||
567 | u8 intrctrl; | ||
568 | |||
569 | intrctrl = ioatdma_read8(instance, IOAT_INTRCTRL_OFFSET); | ||
570 | |||
571 | if (!(intrctrl & IOAT_INTRCTRL_MASTER_INT_EN)) | ||
572 | return IRQ_NONE; | ||
573 | |||
574 | if (!(intrctrl & IOAT_INTRCTRL_INT_STATUS)) { | ||
575 | ioatdma_write8(instance, IOAT_INTRCTRL_OFFSET, intrctrl); | ||
576 | return IRQ_NONE; | ||
577 | } | ||
578 | |||
579 | attnstatus = ioatdma_read32(instance, IOAT_ATTNSTATUS_OFFSET); | ||
580 | |||
581 | printk(KERN_ERR "ioatdma error: interrupt! status %lx\n", attnstatus); | ||
582 | |||
583 | ioatdma_write8(instance, IOAT_INTRCTRL_OFFSET, intrctrl); | ||
584 | return IRQ_HANDLED; | ||
585 | } | ||
586 | |||
587 | static void ioat_start_null_desc(struct ioat_dma_chan *ioat_chan) | ||
588 | { | ||
589 | struct ioat_desc_sw *desc; | ||
590 | |||
591 | spin_lock_bh(&ioat_chan->desc_lock); | ||
592 | |||
593 | if (!list_empty(&ioat_chan->free_desc)) { | ||
594 | desc = to_ioat_desc(ioat_chan->free_desc.next); | ||
595 | list_del(&desc->node); | ||
596 | } else { | ||
597 | /* try to get another desc */ | ||
598 | spin_unlock_bh(&ioat_chan->desc_lock); | ||
599 | desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_KERNEL); | ||
600 | spin_lock_bh(&ioat_chan->desc_lock); | ||
601 | /* will this ever happen? */ | ||
602 | BUG_ON(!desc); | ||
603 | } | ||
604 | |||
605 | desc->hw->ctl = IOAT_DMA_DESCRIPTOR_NUL; | ||
606 | desc->hw->next = 0; | ||
607 | |||
608 | list_add_tail(&desc->node, &ioat_chan->used_desc); | ||
609 | spin_unlock_bh(&ioat_chan->desc_lock); | ||
610 | |||
611 | #if (BITS_PER_LONG == 64) | ||
612 | ioatdma_chan_write64(ioat_chan, IOAT_CHAINADDR_OFFSET, desc->phys); | ||
613 | #else | ||
614 | ioatdma_chan_write32(ioat_chan, | ||
615 | IOAT_CHAINADDR_OFFSET_LOW, | ||
616 | (u32) desc->phys); | ||
617 | ioatdma_chan_write32(ioat_chan, IOAT_CHAINADDR_OFFSET_HIGH, 0); | ||
618 | #endif | ||
619 | ioatdma_chan_write8(ioat_chan, IOAT_CHANCMD_OFFSET, IOAT_CHANCMD_START); | ||
620 | } | ||
621 | |||
622 | /* | ||
623 | * Perform a IOAT transaction to verify the HW works. | ||
624 | */ | ||
625 | #define IOAT_TEST_SIZE 2000 | ||
626 | |||
627 | static int ioat_self_test(struct ioat_device *device) | ||
628 | { | ||
629 | int i; | ||
630 | u8 *src; | ||
631 | u8 *dest; | ||
632 | struct dma_chan *dma_chan; | ||
633 | dma_cookie_t cookie; | ||
634 | int err = 0; | ||
635 | |||
636 | src = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, SLAB_KERNEL); | ||
637 | if (!src) | ||
638 | return -ENOMEM; | ||
639 | dest = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, SLAB_KERNEL); | ||
640 | if (!dest) { | ||
641 | kfree(src); | ||
642 | return -ENOMEM; | ||
643 | } | ||
644 | |||
645 | /* Fill in src buffer */ | ||
646 | for (i = 0; i < IOAT_TEST_SIZE; i++) | ||
647 | src[i] = (u8)i; | ||
648 | |||
649 | /* Start copy, using first DMA channel */ | ||
650 | dma_chan = container_of(device->common.channels.next, | ||
651 | struct dma_chan, | ||
652 | device_node); | ||
653 | if (ioat_dma_alloc_chan_resources(dma_chan) < 1) { | ||
654 | err = -ENODEV; | ||
655 | goto out; | ||
656 | } | ||
657 | |||
658 | cookie = ioat_dma_memcpy_buf_to_buf(dma_chan, dest, src, IOAT_TEST_SIZE); | ||
659 | ioat_dma_memcpy_issue_pending(dma_chan); | ||
660 | msleep(1); | ||
661 | |||
662 | if (ioat_dma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) { | ||
663 | printk(KERN_ERR "ioatdma: Self-test copy timed out, disabling\n"); | ||
664 | err = -ENODEV; | ||
665 | goto free_resources; | ||
666 | } | ||
667 | if (memcmp(src, dest, IOAT_TEST_SIZE)) { | ||
668 | printk(KERN_ERR "ioatdma: Self-test copy failed compare, disabling\n"); | ||
669 | err = -ENODEV; | ||
670 | goto free_resources; | ||
671 | } | ||
672 | |||
673 | free_resources: | ||
674 | ioat_dma_free_chan_resources(dma_chan); | ||
675 | out: | ||
676 | kfree(src); | ||
677 | kfree(dest); | ||
678 | return err; | ||
679 | } | ||
680 | |||
681 | static int __devinit ioat_probe(struct pci_dev *pdev, | ||
682 | const struct pci_device_id *ent) | ||
683 | { | ||
684 | int err; | ||
685 | unsigned long mmio_start, mmio_len; | ||
686 | void *reg_base; | ||
687 | struct ioat_device *device; | ||
688 | |||
689 | err = pci_enable_device(pdev); | ||
690 | if (err) | ||
691 | goto err_enable_device; | ||
692 | |||
693 | err = pci_set_dma_mask(pdev, DMA_64BIT_MASK); | ||
694 | if (err) | ||
695 | err = pci_set_dma_mask(pdev, DMA_32BIT_MASK); | ||
696 | if (err) | ||
697 | goto err_set_dma_mask; | ||
698 | |||
699 | err = pci_request_regions(pdev, ioat_pci_drv.name); | ||
700 | if (err) | ||
701 | goto err_request_regions; | ||
702 | |||
703 | mmio_start = pci_resource_start(pdev, 0); | ||
704 | mmio_len = pci_resource_len(pdev, 0); | ||
705 | |||
706 | reg_base = ioremap(mmio_start, mmio_len); | ||
707 | if (!reg_base) { | ||
708 | err = -ENOMEM; | ||
709 | goto err_ioremap; | ||
710 | } | ||
711 | |||
712 | device = kzalloc(sizeof(*device), GFP_KERNEL); | ||
713 | if (!device) { | ||
714 | err = -ENOMEM; | ||
715 | goto err_kzalloc; | ||
716 | } | ||
717 | |||
718 | /* DMA coherent memory pool for DMA descriptor allocations */ | ||
719 | device->dma_pool = pci_pool_create("dma_desc_pool", pdev, | ||
720 | sizeof(struct ioat_dma_descriptor), 64, 0); | ||
721 | if (!device->dma_pool) { | ||
722 | err = -ENOMEM; | ||
723 | goto err_dma_pool; | ||
724 | } | ||
725 | |||
726 | device->completion_pool = pci_pool_create("completion_pool", pdev, sizeof(u64), SMP_CACHE_BYTES, SMP_CACHE_BYTES); | ||
727 | if (!device->completion_pool) { | ||
728 | err = -ENOMEM; | ||
729 | goto err_completion_pool; | ||
730 | } | ||
731 | |||
732 | device->pdev = pdev; | ||
733 | pci_set_drvdata(pdev, device); | ||
734 | #ifdef CONFIG_PCI_MSI | ||
735 | if (pci_enable_msi(pdev) == 0) { | ||
736 | device->msi = 1; | ||
737 | } else { | ||
738 | device->msi = 0; | ||
739 | } | ||
740 | #endif | ||
741 | err = request_irq(pdev->irq, &ioat_do_interrupt, SA_SHIRQ, "ioat", | ||
742 | device); | ||
743 | if (err) | ||
744 | goto err_irq; | ||
745 | |||
746 | device->reg_base = reg_base; | ||
747 | |||
748 | ioatdma_write8(device, IOAT_INTRCTRL_OFFSET, IOAT_INTRCTRL_MASTER_INT_EN); | ||
749 | pci_set_master(pdev); | ||
750 | |||
751 | INIT_LIST_HEAD(&device->common.channels); | ||
752 | enumerate_dma_channels(device); | ||
753 | |||
754 | device->common.device_alloc_chan_resources = ioat_dma_alloc_chan_resources; | ||
755 | device->common.device_free_chan_resources = ioat_dma_free_chan_resources; | ||
756 | device->common.device_memcpy_buf_to_buf = ioat_dma_memcpy_buf_to_buf; | ||
757 | device->common.device_memcpy_buf_to_pg = ioat_dma_memcpy_buf_to_pg; | ||
758 | device->common.device_memcpy_pg_to_pg = ioat_dma_memcpy_pg_to_pg; | ||
759 | device->common.device_memcpy_complete = ioat_dma_is_complete; | ||
760 | device->common.device_memcpy_issue_pending = ioat_dma_memcpy_issue_pending; | ||
761 | printk(KERN_INFO "Intel(R) I/OAT DMA Engine found, %d channels\n", | ||
762 | device->common.chancnt); | ||
763 | |||
764 | err = ioat_self_test(device); | ||
765 | if (err) | ||
766 | goto err_self_test; | ||
767 | |||
768 | dma_async_device_register(&device->common); | ||
769 | |||
770 | return 0; | ||
771 | |||
772 | err_self_test: | ||
773 | err_irq: | ||
774 | pci_pool_destroy(device->completion_pool); | ||
775 | err_completion_pool: | ||
776 | pci_pool_destroy(device->dma_pool); | ||
777 | err_dma_pool: | ||
778 | kfree(device); | ||
779 | err_kzalloc: | ||
780 | iounmap(reg_base); | ||
781 | err_ioremap: | ||
782 | pci_release_regions(pdev); | ||
783 | err_request_regions: | ||
784 | err_set_dma_mask: | ||
785 | pci_disable_device(pdev); | ||
786 | err_enable_device: | ||
787 | return err; | ||
788 | } | ||
789 | |||
790 | static void __devexit ioat_remove(struct pci_dev *pdev) | ||
791 | { | ||
792 | struct ioat_device *device; | ||
793 | struct dma_chan *chan, *_chan; | ||
794 | struct ioat_dma_chan *ioat_chan; | ||
795 | |||
796 | device = pci_get_drvdata(pdev); | ||
797 | dma_async_device_unregister(&device->common); | ||
798 | |||
799 | free_irq(device->pdev->irq, device); | ||
800 | #ifdef CONFIG_PCI_MSI | ||
801 | if (device->msi) | ||
802 | pci_disable_msi(device->pdev); | ||
803 | #endif | ||
804 | pci_pool_destroy(device->dma_pool); | ||
805 | pci_pool_destroy(device->completion_pool); | ||
806 | iounmap(device->reg_base); | ||
807 | pci_release_regions(pdev); | ||
808 | pci_disable_device(pdev); | ||
809 | list_for_each_entry_safe(chan, _chan, &device->common.channels, device_node) { | ||
810 | ioat_chan = to_ioat_chan(chan); | ||
811 | list_del(&chan->device_node); | ||
812 | kfree(ioat_chan); | ||
813 | } | ||
814 | kfree(device); | ||
815 | } | ||
816 | |||
817 | /* MODULE API */ | ||
818 | MODULE_VERSION("1.7"); | ||
819 | MODULE_LICENSE("GPL"); | ||
820 | MODULE_AUTHOR("Intel Corporation"); | ||
821 | |||
822 | static int __init ioat_init_module(void) | ||
823 | { | ||
824 | /* it's currently unsafe to unload this module */ | ||
825 | /* if forced, worst case is that rmmod hangs */ | ||
826 | if (THIS_MODULE != NULL) | ||
827 | THIS_MODULE->unsafe = 1; | ||
828 | |||
829 | return pci_module_init(&ioat_pci_drv); | ||
830 | } | ||
831 | |||
832 | module_init(ioat_init_module); | ||
833 | |||
834 | static void __exit ioat_exit_module(void) | ||
835 | { | ||
836 | pci_unregister_driver(&ioat_pci_drv); | ||
837 | } | ||
838 | |||
839 | module_exit(ioat_exit_module); | ||