Commit | Line | Data |
---|---|---|
bf40a686 DW |
1 | /* |
2 | * This file is provided under a dual BSD/GPLv2 license. When using or | |
3 | * redistributing this file, you may do so under either license. | |
4 | * | |
5 | * GPL LICENSE SUMMARY | |
6 | * | |
7 | * Copyright(c) 2004 - 2009 Intel Corporation. All rights reserved. | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify it | |
10 | * under the terms and conditions of the GNU General Public License, | |
11 | * version 2, as published by the Free Software Foundation. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
14 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
15 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
16 | * more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License along with | |
19 | * this program; if not, write to the Free Software Foundation, Inc., | |
20 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
21 | * | |
22 | * The full GNU General Public License is included in this distribution in | |
23 | * the file called "COPYING". | |
24 | * | |
25 | * BSD LICENSE | |
26 | * | |
27 | * Copyright(c) 2004-2009 Intel Corporation. All rights reserved. | |
28 | * | |
29 | * Redistribution and use in source and binary forms, with or without | |
30 | * modification, are permitted provided that the following conditions are met: | |
31 | * | |
32 | * * Redistributions of source code must retain the above copyright | |
33 | * notice, this list of conditions and the following disclaimer. | |
34 | * * Redistributions in binary form must reproduce the above copyright | |
35 | * notice, this list of conditions and the following disclaimer in | |
36 | * the documentation and/or other materials provided with the | |
37 | * distribution. | |
38 | * * Neither the name of Intel Corporation nor the names of its | |
39 | * contributors may be used to endorse or promote products derived | |
40 | * from this software without specific prior written permission. | |
41 | * | |
42 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" | |
43 | * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
44 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
45 | * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE | |
46 | * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | |
47 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | |
48 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS | |
49 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN | |
50 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | |
51 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | |
52 | * POSSIBILITY OF SUCH DAMAGE. | |
53 | */ | |
54 | ||
55 | /* | |
56 | * Support routines for v3+ hardware | |
57 | */ | |
58 | ||
59 | #include <linux/pci.h> | |
5a0e3ad6 | 60 | #include <linux/gfp.h> |
bf40a686 DW |
61 | #include <linux/dmaengine.h> |
62 | #include <linux/dma-mapping.h> | |
70c71606 | 63 | #include <linux/prefetch.h> |
949ff5b8 | 64 | #include "../dmaengine.h" |
bf40a686 DW |
65 | #include "registers.h" |
66 | #include "hw.h" | |
67 | #include "dma.h" | |
68 | #include "dma_v2.h" | |
69 | ||
b094ad3b DW |
70 | /* ioat hardware assumes at least two sources for raid operations */ |
71 | #define src_cnt_to_sw(x) ((x) + 2) | |
72 | #define src_cnt_to_hw(x) ((x) - 2) | |
73 | ||
74 | /* provide a lookup table for setting the source address in the base or | |
d69d235b | 75 | * extended descriptor of an xor or pq descriptor |
b094ad3b | 76 | */ |
d0b0c8c7 | 77 | static const u8 xor_idx_to_desc = 0xe0; |
9b487ced AK |
78 | static const u8 xor_idx_to_field[] = { 1, 4, 5, 6, 7, 0, 1, 2 }; |
79 | static const u8 pq_idx_to_desc = 0xf8; | |
80 | static const u8 pq_idx_to_field[] = { 1, 4, 5, 0, 1, 2, 4, 5 }; | |
b094ad3b DW |
81 | |
82 | static dma_addr_t xor_get_src(struct ioat_raw_descriptor *descs[2], int idx) | |
83 | { | |
84 | struct ioat_raw_descriptor *raw = descs[xor_idx_to_desc >> idx & 1]; | |
85 | ||
86 | return raw->field[xor_idx_to_field[idx]]; | |
87 | } | |
88 | ||
89 | static void xor_set_src(struct ioat_raw_descriptor *descs[2], | |
90 | dma_addr_t addr, u32 offset, int idx) | |
91 | { | |
92 | struct ioat_raw_descriptor *raw = descs[xor_idx_to_desc >> idx & 1]; | |
93 | ||
94 | raw->field[xor_idx_to_field[idx]] = addr + offset; | |
95 | } | |
96 | ||
d69d235b DW |
97 | static dma_addr_t pq_get_src(struct ioat_raw_descriptor *descs[2], int idx) |
98 | { | |
99 | struct ioat_raw_descriptor *raw = descs[pq_idx_to_desc >> idx & 1]; | |
100 | ||
101 | return raw->field[pq_idx_to_field[idx]]; | |
102 | } | |
103 | ||
104 | static void pq_set_src(struct ioat_raw_descriptor *descs[2], | |
105 | dma_addr_t addr, u32 offset, u8 coef, int idx) | |
106 | { | |
107 | struct ioat_pq_descriptor *pq = (struct ioat_pq_descriptor *) descs[0]; | |
108 | struct ioat_raw_descriptor *raw = descs[pq_idx_to_desc >> idx & 1]; | |
109 | ||
110 | raw->field[pq_idx_to_field[idx]] = addr + offset; | |
111 | pq->coef[idx] = coef; | |
112 | } | |
113 | ||
bf40a686 | 114 | static void ioat3_dma_unmap(struct ioat2_dma_chan *ioat, |
b094ad3b | 115 | struct ioat_ring_ent *desc, int idx) |
bf40a686 DW |
116 | { |
117 | struct ioat_chan_common *chan = &ioat->base; | |
118 | struct pci_dev *pdev = chan->device->pdev; | |
119 | size_t len = desc->len; | |
120 | size_t offset = len - desc->hw->size; | |
121 | struct dma_async_tx_descriptor *tx = &desc->txd; | |
122 | enum dma_ctrl_flags flags = tx->flags; | |
123 | ||
124 | switch (desc->hw->ctl_f.op) { | |
125 | case IOAT_OP_COPY: | |
58c8649e DW |
126 | if (!desc->hw->ctl_f.null) /* skip 'interrupt' ops */ |
127 | ioat_dma_unmap(chan, flags, len, desc->hw); | |
bf40a686 DW |
128 | break; |
129 | case IOAT_OP_FILL: { | |
130 | struct ioat_fill_descriptor *hw = desc->fill; | |
131 | ||
132 | if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) | |
133 | ioat_unmap(pdev, hw->dst_addr - offset, len, | |
134 | PCI_DMA_FROMDEVICE, flags, 1); | |
135 | break; | |
136 | } | |
b094ad3b DW |
137 | case IOAT_OP_XOR_VAL: |
138 | case IOAT_OP_XOR: { | |
139 | struct ioat_xor_descriptor *xor = desc->xor; | |
140 | struct ioat_ring_ent *ext; | |
141 | struct ioat_xor_ext_descriptor *xor_ex = NULL; | |
142 | int src_cnt = src_cnt_to_sw(xor->ctl_f.src_cnt); | |
143 | struct ioat_raw_descriptor *descs[2]; | |
144 | int i; | |
145 | ||
146 | if (src_cnt > 5) { | |
147 | ext = ioat2_get_ring_ent(ioat, idx + 1); | |
148 | xor_ex = ext->xor_ex; | |
149 | } | |
150 | ||
151 | if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) { | |
152 | descs[0] = (struct ioat_raw_descriptor *) xor; | |
153 | descs[1] = (struct ioat_raw_descriptor *) xor_ex; | |
154 | for (i = 0; i < src_cnt; i++) { | |
155 | dma_addr_t src = xor_get_src(descs, i); | |
156 | ||
157 | ioat_unmap(pdev, src - offset, len, | |
158 | PCI_DMA_TODEVICE, flags, 0); | |
159 | } | |
160 | ||
161 | /* dest is a source in xor validate operations */ | |
162 | if (xor->ctl_f.op == IOAT_OP_XOR_VAL) { | |
163 | ioat_unmap(pdev, xor->dst_addr - offset, len, | |
164 | PCI_DMA_TODEVICE, flags, 1); | |
165 | break; | |
166 | } | |
167 | } | |
168 | ||
169 | if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) | |
170 | ioat_unmap(pdev, xor->dst_addr - offset, len, | |
171 | PCI_DMA_FROMDEVICE, flags, 1); | |
172 | break; | |
173 | } | |
d69d235b DW |
174 | case IOAT_OP_PQ_VAL: |
175 | case IOAT_OP_PQ: { | |
176 | struct ioat_pq_descriptor *pq = desc->pq; | |
177 | struct ioat_ring_ent *ext; | |
178 | struct ioat_pq_ext_descriptor *pq_ex = NULL; | |
179 | int src_cnt = src_cnt_to_sw(pq->ctl_f.src_cnt); | |
180 | struct ioat_raw_descriptor *descs[2]; | |
181 | int i; | |
182 | ||
183 | if (src_cnt > 3) { | |
184 | ext = ioat2_get_ring_ent(ioat, idx + 1); | |
185 | pq_ex = ext->pq_ex; | |
186 | } | |
187 | ||
188 | /* in the 'continue' case don't unmap the dests as sources */ | |
189 | if (dmaf_p_disabled_continue(flags)) | |
190 | src_cnt--; | |
191 | else if (dmaf_continue(flags)) | |
192 | src_cnt -= 3; | |
193 | ||
194 | if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) { | |
195 | descs[0] = (struct ioat_raw_descriptor *) pq; | |
196 | descs[1] = (struct ioat_raw_descriptor *) pq_ex; | |
197 | for (i = 0; i < src_cnt; i++) { | |
198 | dma_addr_t src = pq_get_src(descs, i); | |
199 | ||
200 | ioat_unmap(pdev, src - offset, len, | |
201 | PCI_DMA_TODEVICE, flags, 0); | |
202 | } | |
203 | ||
204 | /* the dests are sources in pq validate operations */ | |
205 | if (pq->ctl_f.op == IOAT_OP_XOR_VAL) { | |
206 | if (!(flags & DMA_PREP_PQ_DISABLE_P)) | |
207 | ioat_unmap(pdev, pq->p_addr - offset, | |
208 | len, PCI_DMA_TODEVICE, flags, 0); | |
209 | if (!(flags & DMA_PREP_PQ_DISABLE_Q)) | |
210 | ioat_unmap(pdev, pq->q_addr - offset, | |
211 | len, PCI_DMA_TODEVICE, flags, 0); | |
212 | break; | |
213 | } | |
214 | } | |
215 | ||
216 | if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) { | |
217 | if (!(flags & DMA_PREP_PQ_DISABLE_P)) | |
218 | ioat_unmap(pdev, pq->p_addr - offset, len, | |
219 | PCI_DMA_BIDIRECTIONAL, flags, 1); | |
220 | if (!(flags & DMA_PREP_PQ_DISABLE_Q)) | |
221 | ioat_unmap(pdev, pq->q_addr - offset, len, | |
222 | PCI_DMA_BIDIRECTIONAL, flags, 1); | |
223 | } | |
224 | break; | |
225 | } | |
bf40a686 DW |
226 | default: |
227 | dev_err(&pdev->dev, "%s: unknown op type: %#x\n", | |
228 | __func__, desc->hw->ctl_f.op); | |
229 | } | |
230 | } | |
231 | ||
b094ad3b DW |
232 | static bool desc_has_ext(struct ioat_ring_ent *desc) |
233 | { | |
234 | struct ioat_dma_descriptor *hw = desc->hw; | |
235 | ||
236 | if (hw->ctl_f.op == IOAT_OP_XOR || | |
237 | hw->ctl_f.op == IOAT_OP_XOR_VAL) { | |
238 | struct ioat_xor_descriptor *xor = desc->xor; | |
bf40a686 | 239 | |
b094ad3b DW |
240 | if (src_cnt_to_sw(xor->ctl_f.src_cnt) > 5) |
241 | return true; | |
d69d235b DW |
242 | } else if (hw->ctl_f.op == IOAT_OP_PQ || |
243 | hw->ctl_f.op == IOAT_OP_PQ_VAL) { | |
244 | struct ioat_pq_descriptor *pq = desc->pq; | |
245 | ||
246 | if (src_cnt_to_sw(pq->ctl_f.src_cnt) > 3) | |
247 | return true; | |
b094ad3b DW |
248 | } |
249 | ||
250 | return false; | |
251 | } | |
252 | ||
253 | /** | |
254 | * __cleanup - reclaim used descriptors | |
255 | * @ioat: channel (ring) to clean | |
256 | * | |
257 | * The difference from the dma_v2.c __cleanup() is that this routine | |
258 | * handles extended descriptors and dma-unmapping raid operations. | |
259 | */ | |
27502935 | 260 | static void __cleanup(struct ioat2_dma_chan *ioat, dma_addr_t phys_complete) |
bf40a686 DW |
261 | { |
262 | struct ioat_chan_common *chan = &ioat->base; | |
263 | struct ioat_ring_ent *desc; | |
264 | bool seen_current = false; | |
074cc476 | 265 | int idx = ioat->tail, i; |
bf40a686 | 266 | u16 active; |
bf40a686 DW |
267 | |
268 | dev_dbg(to_dev(chan), "%s: head: %#x tail: %#x issued: %#x\n", | |
269 | __func__, ioat->head, ioat->tail, ioat->issued); | |
270 | ||
271 | active = ioat2_ring_active(ioat); | |
272 | for (i = 0; i < active && !seen_current; i++) { | |
273 | struct dma_async_tx_descriptor *tx; | |
274 | ||
074cc476 DW |
275 | smp_read_barrier_depends(); |
276 | prefetch(ioat2_get_ring_ent(ioat, idx + i + 1)); | |
277 | desc = ioat2_get_ring_ent(ioat, idx + i); | |
bf40a686 DW |
278 | dump_desc_dbg(ioat, desc); |
279 | tx = &desc->txd; | |
280 | if (tx->cookie) { | |
f7fbce07 | 281 | dma_cookie_complete(tx); |
074cc476 | 282 | ioat3_dma_unmap(ioat, desc, idx + i); |
bf40a686 DW |
283 | if (tx->callback) { |
284 | tx->callback(tx->callback_param); | |
285 | tx->callback = NULL; | |
286 | } | |
287 | } | |
288 | ||
289 | if (tx->phys == phys_complete) | |
290 | seen_current = true; | |
b094ad3b DW |
291 | |
292 | /* skip extended descriptors */ | |
293 | if (desc_has_ext(desc)) { | |
294 | BUG_ON(i + 1 >= active); | |
295 | i++; | |
296 | } | |
bf40a686 | 297 | } |
074cc476 DW |
298 | smp_mb(); /* finish all descriptor reads before incrementing tail */ |
299 | ioat->tail = idx + i; | |
aa75db00 | 300 | BUG_ON(active && !seen_current); /* no active descs have written a completion? */ |
bf40a686 | 301 | chan->last_completion = phys_complete; |
b9cc9869 | 302 | |
074cc476 | 303 | if (active - i == 0) { |
bf40a686 DW |
304 | dev_dbg(to_dev(chan), "%s: cancel completion timeout\n", |
305 | __func__); | |
306 | clear_bit(IOAT_COMPLETION_PENDING, &chan->state); | |
307 | mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT); | |
308 | } | |
b9cc9869 | 309 | /* 5 microsecond delay per pending descriptor */ |
074cc476 | 310 | writew(min((5 * (active - i)), IOAT_INTRDELAY_MASK), |
b9cc9869 | 311 | chan->device->reg_base + IOAT_INTRDELAY_OFFSET); |
bf40a686 DW |
312 | } |
313 | ||
074cc476 | 314 | static void ioat3_cleanup(struct ioat2_dma_chan *ioat) |
bf40a686 DW |
315 | { |
316 | struct ioat_chan_common *chan = &ioat->base; | |
27502935 | 317 | dma_addr_t phys_complete; |
bf40a686 | 318 | |
b9cc9869 | 319 | spin_lock_bh(&chan->cleanup_lock); |
074cc476 DW |
320 | if (ioat_cleanup_preamble(chan, &phys_complete)) |
321 | __cleanup(ioat, phys_complete); | |
b9cc9869 DW |
322 | spin_unlock_bh(&chan->cleanup_lock); |
323 | } | |
324 | ||
aa4d72ae | 325 | static void ioat3_cleanup_event(unsigned long data) |
bf40a686 | 326 | { |
aa4d72ae | 327 | struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data); |
bf40a686 | 328 | |
074cc476 | 329 | ioat3_cleanup(ioat); |
773d9e2d | 330 | writew(IOAT_CHANCTRL_RUN, ioat->base.reg_base + IOAT_CHANCTRL_OFFSET); |
bf40a686 DW |
331 | } |
332 | ||
333 | static void ioat3_restart_channel(struct ioat2_dma_chan *ioat) | |
334 | { | |
335 | struct ioat_chan_common *chan = &ioat->base; | |
27502935 | 336 | dma_addr_t phys_complete; |
bf40a686 | 337 | |
b372ec2d | 338 | ioat2_quiesce(chan, 0); |
bf40a686 DW |
339 | if (ioat_cleanup_preamble(chan, &phys_complete)) |
340 | __cleanup(ioat, phys_complete); | |
341 | ||
342 | __ioat2_restart_chan(ioat); | |
343 | } | |
344 | ||
4dec23d7 | 345 | static void check_active(struct ioat2_dma_chan *ioat) |
bf40a686 | 346 | { |
bf40a686 DW |
347 | struct ioat_chan_common *chan = &ioat->base; |
348 | ||
4dec23d7 DJ |
349 | if (ioat2_ring_active(ioat)) { |
350 | mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT); | |
351 | return; | |
352 | } | |
bf40a686 | 353 | |
4dec23d7 DJ |
354 | if (test_and_clear_bit(IOAT_CHAN_ACTIVE, &chan->state)) |
355 | mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT); | |
356 | else if (ioat->alloc_order > ioat_get_alloc_order()) { | |
bf40a686 DW |
357 | /* if the ring is idle, empty, and oversized try to step |
358 | * down the size | |
359 | */ | |
4dec23d7 | 360 | reshape_ring(ioat, ioat->alloc_order - 1); |
bf40a686 DW |
361 | |
362 | /* keep shrinking until we get back to our minimum | |
363 | * default size | |
364 | */ | |
365 | if (ioat->alloc_order > ioat_get_alloc_order()) | |
366 | mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT); | |
367 | } | |
4dec23d7 DJ |
368 | |
369 | } | |
370 | ||
a20702b8 | 371 | static void ioat3_timer_event(unsigned long data) |
4dec23d7 DJ |
372 | { |
373 | struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data); | |
374 | struct ioat_chan_common *chan = &ioat->base; | |
375 | dma_addr_t phys_complete; | |
376 | u64 status; | |
377 | ||
378 | status = ioat_chansts(chan); | |
379 | ||
380 | /* when halted due to errors check for channel | |
381 | * programming errors before advancing the completion state | |
382 | */ | |
383 | if (is_ioat_halted(status)) { | |
384 | u32 chanerr; | |
385 | ||
386 | chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET); | |
387 | dev_err(to_dev(chan), "%s: Channel halted (%x)\n", | |
388 | __func__, chanerr); | |
389 | if (test_bit(IOAT_RUN, &chan->state)) | |
390 | BUG_ON(is_ioat_bug(chanerr)); | |
391 | else /* we never got off the ground */ | |
392 | return; | |
393 | } | |
394 | ||
395 | /* if we haven't made progress and we have already | |
396 | * acknowledged a pending completion once, then be more | |
397 | * forceful with a restart | |
398 | */ | |
399 | spin_lock_bh(&chan->cleanup_lock); | |
400 | if (ioat_cleanup_preamble(chan, &phys_complete)) | |
401 | __cleanup(ioat, phys_complete); | |
402 | else if (test_bit(IOAT_COMPLETION_ACK, &chan->state)) { | |
403 | spin_lock_bh(&ioat->prep_lock); | |
404 | ioat3_restart_channel(ioat); | |
405 | spin_unlock_bh(&ioat->prep_lock); | |
406 | spin_unlock_bh(&chan->cleanup_lock); | |
407 | return; | |
408 | } else { | |
409 | set_bit(IOAT_COMPLETION_ACK, &chan->state); | |
410 | mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT); | |
411 | } | |
412 | ||
413 | ||
414 | if (ioat2_ring_active(ioat)) | |
415 | mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT); | |
416 | else { | |
417 | spin_lock_bh(&ioat->prep_lock); | |
418 | check_active(ioat); | |
419 | spin_unlock_bh(&ioat->prep_lock); | |
420 | } | |
421 | spin_unlock_bh(&chan->cleanup_lock); | |
bf40a686 DW |
422 | } |
423 | ||
424 | static enum dma_status | |
07934481 LW |
425 | ioat3_tx_status(struct dma_chan *c, dma_cookie_t cookie, |
426 | struct dma_tx_state *txstate) | |
bf40a686 DW |
427 | { |
428 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | |
96a2af41 | 429 | enum dma_status ret; |
bf40a686 | 430 | |
96a2af41 RKAL |
431 | ret = dma_cookie_status(c, cookie, txstate); |
432 | if (ret == DMA_SUCCESS) | |
433 | return ret; | |
bf40a686 | 434 | |
074cc476 | 435 | ioat3_cleanup(ioat); |
bf40a686 | 436 | |
96a2af41 | 437 | return dma_cookie_status(c, cookie, txstate); |
bf40a686 DW |
438 | } |
439 | ||
440 | static struct dma_async_tx_descriptor * | |
441 | ioat3_prep_memset_lock(struct dma_chan *c, dma_addr_t dest, int value, | |
442 | size_t len, unsigned long flags) | |
443 | { | |
444 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | |
445 | struct ioat_ring_ent *desc; | |
446 | size_t total_len = len; | |
447 | struct ioat_fill_descriptor *fill; | |
bf40a686 | 448 | u64 src_data = (0x0101010101010101ULL) * (value & 0xff); |
074cc476 | 449 | int num_descs, idx, i; |
bf40a686 DW |
450 | |
451 | num_descs = ioat2_xferlen_to_descs(ioat, len); | |
074cc476 DW |
452 | if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs) == 0) |
453 | idx = ioat->head; | |
bf40a686 DW |
454 | else |
455 | return NULL; | |
cdef57db DW |
456 | i = 0; |
457 | do { | |
bf40a686 DW |
458 | size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log); |
459 | ||
460 | desc = ioat2_get_ring_ent(ioat, idx + i); | |
461 | fill = desc->fill; | |
462 | ||
463 | fill->size = xfer_size; | |
464 | fill->src_data = src_data; | |
465 | fill->dst_addr = dest; | |
466 | fill->ctl = 0; | |
467 | fill->ctl_f.op = IOAT_OP_FILL; | |
468 | ||
469 | len -= xfer_size; | |
470 | dest += xfer_size; | |
471 | dump_desc_dbg(ioat, desc); | |
cdef57db | 472 | } while (++i < num_descs); |
bf40a686 DW |
473 | |
474 | desc->txd.flags = flags; | |
475 | desc->len = total_len; | |
476 | fill->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT); | |
477 | fill->ctl_f.fence = !!(flags & DMA_PREP_FENCE); | |
478 | fill->ctl_f.compl_write = 1; | |
479 | dump_desc_dbg(ioat, desc); | |
480 | ||
481 | /* we leave the channel locked to ensure in order submission */ | |
482 | return &desc->txd; | |
483 | } | |
484 | ||
b094ad3b DW |
485 | static struct dma_async_tx_descriptor * |
486 | __ioat3_prep_xor_lock(struct dma_chan *c, enum sum_check_flags *result, | |
487 | dma_addr_t dest, dma_addr_t *src, unsigned int src_cnt, | |
488 | size_t len, unsigned long flags) | |
489 | { | |
490 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | |
491 | struct ioat_ring_ent *compl_desc; | |
492 | struct ioat_ring_ent *desc; | |
493 | struct ioat_ring_ent *ext; | |
494 | size_t total_len = len; | |
495 | struct ioat_xor_descriptor *xor; | |
496 | struct ioat_xor_ext_descriptor *xor_ex = NULL; | |
497 | struct ioat_dma_descriptor *hw; | |
074cc476 | 498 | int num_descs, with_ext, idx, i; |
b094ad3b | 499 | u32 offset = 0; |
b094ad3b DW |
500 | u8 op = result ? IOAT_OP_XOR_VAL : IOAT_OP_XOR; |
501 | ||
502 | BUG_ON(src_cnt < 2); | |
503 | ||
504 | num_descs = ioat2_xferlen_to_descs(ioat, len); | |
505 | /* we need 2x the number of descriptors to cover greater than 5 | |
506 | * sources | |
507 | */ | |
508 | if (src_cnt > 5) { | |
509 | with_ext = 1; | |
510 | num_descs *= 2; | |
511 | } else | |
512 | with_ext = 0; | |
513 | ||
514 | /* completion writes from the raid engine may pass completion | |
515 | * writes from the legacy engine, so we need one extra null | |
516 | * (legacy) descriptor to ensure all completion writes arrive in | |
517 | * order. | |
518 | */ | |
074cc476 DW |
519 | if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs+1) == 0) |
520 | idx = ioat->head; | |
b094ad3b DW |
521 | else |
522 | return NULL; | |
cdef57db DW |
523 | i = 0; |
524 | do { | |
b094ad3b DW |
525 | struct ioat_raw_descriptor *descs[2]; |
526 | size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log); | |
527 | int s; | |
528 | ||
529 | desc = ioat2_get_ring_ent(ioat, idx + i); | |
530 | xor = desc->xor; | |
531 | ||
532 | /* save a branch by unconditionally retrieving the | |
533 | * extended descriptor xor_set_src() knows to not write | |
534 | * to it in the single descriptor case | |
535 | */ | |
536 | ext = ioat2_get_ring_ent(ioat, idx + i + 1); | |
537 | xor_ex = ext->xor_ex; | |
538 | ||
539 | descs[0] = (struct ioat_raw_descriptor *) xor; | |
540 | descs[1] = (struct ioat_raw_descriptor *) xor_ex; | |
541 | for (s = 0; s < src_cnt; s++) | |
542 | xor_set_src(descs, src[s], offset, s); | |
543 | xor->size = xfer_size; | |
544 | xor->dst_addr = dest + offset; | |
545 | xor->ctl = 0; | |
546 | xor->ctl_f.op = op; | |
547 | xor->ctl_f.src_cnt = src_cnt_to_hw(src_cnt); | |
548 | ||
549 | len -= xfer_size; | |
550 | offset += xfer_size; | |
551 | dump_desc_dbg(ioat, desc); | |
cdef57db | 552 | } while ((i += 1 + with_ext) < num_descs); |
b094ad3b DW |
553 | |
554 | /* last xor descriptor carries the unmap parameters and fence bit */ | |
555 | desc->txd.flags = flags; | |
556 | desc->len = total_len; | |
557 | if (result) | |
558 | desc->result = result; | |
559 | xor->ctl_f.fence = !!(flags & DMA_PREP_FENCE); | |
560 | ||
561 | /* completion descriptor carries interrupt bit */ | |
562 | compl_desc = ioat2_get_ring_ent(ioat, idx + i); | |
563 | compl_desc->txd.flags = flags & DMA_PREP_INTERRUPT; | |
564 | hw = compl_desc->hw; | |
565 | hw->ctl = 0; | |
566 | hw->ctl_f.null = 1; | |
567 | hw->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT); | |
568 | hw->ctl_f.compl_write = 1; | |
569 | hw->size = NULL_DESC_BUFFER_SIZE; | |
570 | dump_desc_dbg(ioat, compl_desc); | |
571 | ||
572 | /* we leave the channel locked to ensure in order submission */ | |
49954c15 | 573 | return &compl_desc->txd; |
b094ad3b DW |
574 | } |
575 | ||
576 | static struct dma_async_tx_descriptor * | |
577 | ioat3_prep_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src, | |
578 | unsigned int src_cnt, size_t len, unsigned long flags) | |
579 | { | |
580 | return __ioat3_prep_xor_lock(chan, NULL, dest, src, src_cnt, len, flags); | |
581 | } | |
582 | ||
583 | struct dma_async_tx_descriptor * | |
584 | ioat3_prep_xor_val(struct dma_chan *chan, dma_addr_t *src, | |
585 | unsigned int src_cnt, size_t len, | |
586 | enum sum_check_flags *result, unsigned long flags) | |
587 | { | |
588 | /* the cleanup routine only sets bits on validate failure, it | |
589 | * does not clear bits on validate success... so clear it here | |
590 | */ | |
591 | *result = 0; | |
592 | ||
593 | return __ioat3_prep_xor_lock(chan, result, src[0], &src[1], | |
594 | src_cnt - 1, len, flags); | |
595 | } | |
596 | ||
d69d235b DW |
597 | static void |
598 | dump_pq_desc_dbg(struct ioat2_dma_chan *ioat, struct ioat_ring_ent *desc, struct ioat_ring_ent *ext) | |
599 | { | |
600 | struct device *dev = to_dev(&ioat->base); | |
601 | struct ioat_pq_descriptor *pq = desc->pq; | |
602 | struct ioat_pq_ext_descriptor *pq_ex = ext ? ext->pq_ex : NULL; | |
603 | struct ioat_raw_descriptor *descs[] = { (void *) pq, (void *) pq_ex }; | |
604 | int src_cnt = src_cnt_to_sw(pq->ctl_f.src_cnt); | |
605 | int i; | |
606 | ||
607 | dev_dbg(dev, "desc[%d]: (%#llx->%#llx) flags: %#x" | |
608 | " sz: %#x ctl: %#x (op: %d int: %d compl: %d pq: '%s%s' src_cnt: %d)\n", | |
609 | desc_id(desc), (unsigned long long) desc->txd.phys, | |
610 | (unsigned long long) (pq_ex ? pq_ex->next : pq->next), | |
611 | desc->txd.flags, pq->size, pq->ctl, pq->ctl_f.op, pq->ctl_f.int_en, | |
612 | pq->ctl_f.compl_write, | |
613 | pq->ctl_f.p_disable ? "" : "p", pq->ctl_f.q_disable ? "" : "q", | |
614 | pq->ctl_f.src_cnt); | |
615 | for (i = 0; i < src_cnt; i++) | |
616 | dev_dbg(dev, "\tsrc[%d]: %#llx coef: %#x\n", i, | |
617 | (unsigned long long) pq_get_src(descs, i), pq->coef[i]); | |
618 | dev_dbg(dev, "\tP: %#llx\n", pq->p_addr); | |
619 | dev_dbg(dev, "\tQ: %#llx\n", pq->q_addr); | |
620 | } | |
621 | ||
622 | static struct dma_async_tx_descriptor * | |
623 | __ioat3_prep_pq_lock(struct dma_chan *c, enum sum_check_flags *result, | |
624 | const dma_addr_t *dst, const dma_addr_t *src, | |
625 | unsigned int src_cnt, const unsigned char *scf, | |
626 | size_t len, unsigned long flags) | |
627 | { | |
628 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | |
629 | struct ioat_chan_common *chan = &ioat->base; | |
630 | struct ioat_ring_ent *compl_desc; | |
631 | struct ioat_ring_ent *desc; | |
632 | struct ioat_ring_ent *ext; | |
633 | size_t total_len = len; | |
634 | struct ioat_pq_descriptor *pq; | |
635 | struct ioat_pq_ext_descriptor *pq_ex = NULL; | |
636 | struct ioat_dma_descriptor *hw; | |
637 | u32 offset = 0; | |
d69d235b | 638 | u8 op = result ? IOAT_OP_PQ_VAL : IOAT_OP_PQ; |
074cc476 | 639 | int i, s, idx, with_ext, num_descs; |
d69d235b DW |
640 | |
641 | dev_dbg(to_dev(chan), "%s\n", __func__); | |
642 | /* the engine requires at least two sources (we provide | |
643 | * at least 1 implied source in the DMA_PREP_CONTINUE case) | |
644 | */ | |
645 | BUG_ON(src_cnt + dmaf_continue(flags) < 2); | |
646 | ||
647 | num_descs = ioat2_xferlen_to_descs(ioat, len); | |
648 | /* we need 2x the number of descriptors to cover greater than 3 | |
cd78809f DW |
649 | * sources (we need 1 extra source in the q-only continuation |
650 | * case and 3 extra sources in the p+q continuation case. | |
d69d235b | 651 | */ |
cd78809f DW |
652 | if (src_cnt + dmaf_p_disabled_continue(flags) > 3 || |
653 | (dmaf_continue(flags) && !dmaf_p_disabled_continue(flags))) { | |
d69d235b DW |
654 | with_ext = 1; |
655 | num_descs *= 2; | |
656 | } else | |
657 | with_ext = 0; | |
658 | ||
659 | /* completion writes from the raid engine may pass completion | |
660 | * writes from the legacy engine, so we need one extra null | |
661 | * (legacy) descriptor to ensure all completion writes arrive in | |
662 | * order. | |
663 | */ | |
664 | if (likely(num_descs) && | |
074cc476 DW |
665 | ioat2_check_space_lock(ioat, num_descs+1) == 0) |
666 | idx = ioat->head; | |
d69d235b DW |
667 | else |
668 | return NULL; | |
cdef57db DW |
669 | i = 0; |
670 | do { | |
d69d235b DW |
671 | struct ioat_raw_descriptor *descs[2]; |
672 | size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log); | |
673 | ||
674 | desc = ioat2_get_ring_ent(ioat, idx + i); | |
675 | pq = desc->pq; | |
676 | ||
677 | /* save a branch by unconditionally retrieving the | |
678 | * extended descriptor pq_set_src() knows to not write | |
679 | * to it in the single descriptor case | |
680 | */ | |
681 | ext = ioat2_get_ring_ent(ioat, idx + i + with_ext); | |
682 | pq_ex = ext->pq_ex; | |
683 | ||
684 | descs[0] = (struct ioat_raw_descriptor *) pq; | |
685 | descs[1] = (struct ioat_raw_descriptor *) pq_ex; | |
686 | ||
687 | for (s = 0; s < src_cnt; s++) | |
688 | pq_set_src(descs, src[s], offset, scf[s], s); | |
689 | ||
690 | /* see the comment for dma_maxpq in include/linux/dmaengine.h */ | |
691 | if (dmaf_p_disabled_continue(flags)) | |
692 | pq_set_src(descs, dst[1], offset, 1, s++); | |
693 | else if (dmaf_continue(flags)) { | |
694 | pq_set_src(descs, dst[0], offset, 0, s++); | |
695 | pq_set_src(descs, dst[1], offset, 1, s++); | |
696 | pq_set_src(descs, dst[1], offset, 0, s++); | |
697 | } | |
698 | pq->size = xfer_size; | |
699 | pq->p_addr = dst[0] + offset; | |
700 | pq->q_addr = dst[1] + offset; | |
701 | pq->ctl = 0; | |
702 | pq->ctl_f.op = op; | |
703 | pq->ctl_f.src_cnt = src_cnt_to_hw(s); | |
704 | pq->ctl_f.p_disable = !!(flags & DMA_PREP_PQ_DISABLE_P); | |
705 | pq->ctl_f.q_disable = !!(flags & DMA_PREP_PQ_DISABLE_Q); | |
706 | ||
707 | len -= xfer_size; | |
708 | offset += xfer_size; | |
cdef57db | 709 | } while ((i += 1 + with_ext) < num_descs); |
d69d235b DW |
710 | |
711 | /* last pq descriptor carries the unmap parameters and fence bit */ | |
712 | desc->txd.flags = flags; | |
713 | desc->len = total_len; | |
714 | if (result) | |
715 | desc->result = result; | |
716 | pq->ctl_f.fence = !!(flags & DMA_PREP_FENCE); | |
717 | dump_pq_desc_dbg(ioat, desc, ext); | |
718 | ||
719 | /* completion descriptor carries interrupt bit */ | |
720 | compl_desc = ioat2_get_ring_ent(ioat, idx + i); | |
721 | compl_desc->txd.flags = flags & DMA_PREP_INTERRUPT; | |
722 | hw = compl_desc->hw; | |
723 | hw->ctl = 0; | |
724 | hw->ctl_f.null = 1; | |
725 | hw->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT); | |
726 | hw->ctl_f.compl_write = 1; | |
727 | hw->size = NULL_DESC_BUFFER_SIZE; | |
728 | dump_desc_dbg(ioat, compl_desc); | |
729 | ||
730 | /* we leave the channel locked to ensure in order submission */ | |
49954c15 | 731 | return &compl_desc->txd; |
d69d235b DW |
732 | } |
733 | ||
734 | static struct dma_async_tx_descriptor * | |
735 | ioat3_prep_pq(struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src, | |
736 | unsigned int src_cnt, const unsigned char *scf, size_t len, | |
737 | unsigned long flags) | |
738 | { | |
de581b65 DW |
739 | /* specify valid address for disabled result */ |
740 | if (flags & DMA_PREP_PQ_DISABLE_P) | |
741 | dst[0] = dst[1]; | |
742 | if (flags & DMA_PREP_PQ_DISABLE_Q) | |
743 | dst[1] = dst[0]; | |
744 | ||
d69d235b DW |
745 | /* handle the single source multiply case from the raid6 |
746 | * recovery path | |
747 | */ | |
de581b65 | 748 | if ((flags & DMA_PREP_PQ_DISABLE_P) && src_cnt == 1) { |
d69d235b DW |
749 | dma_addr_t single_source[2]; |
750 | unsigned char single_source_coef[2]; | |
751 | ||
752 | BUG_ON(flags & DMA_PREP_PQ_DISABLE_Q); | |
753 | single_source[0] = src[0]; | |
754 | single_source[1] = src[0]; | |
755 | single_source_coef[0] = scf[0]; | |
756 | single_source_coef[1] = 0; | |
757 | ||
758 | return __ioat3_prep_pq_lock(chan, NULL, dst, single_source, 2, | |
759 | single_source_coef, len, flags); | |
760 | } else | |
761 | return __ioat3_prep_pq_lock(chan, NULL, dst, src, src_cnt, scf, | |
762 | len, flags); | |
763 | } | |
764 | ||
765 | struct dma_async_tx_descriptor * | |
766 | ioat3_prep_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src, | |
767 | unsigned int src_cnt, const unsigned char *scf, size_t len, | |
768 | enum sum_check_flags *pqres, unsigned long flags) | |
769 | { | |
de581b65 DW |
770 | /* specify valid address for disabled result */ |
771 | if (flags & DMA_PREP_PQ_DISABLE_P) | |
772 | pq[0] = pq[1]; | |
773 | if (flags & DMA_PREP_PQ_DISABLE_Q) | |
774 | pq[1] = pq[0]; | |
775 | ||
d69d235b DW |
776 | /* the cleanup routine only sets bits on validate failure, it |
777 | * does not clear bits on validate success... so clear it here | |
778 | */ | |
779 | *pqres = 0; | |
780 | ||
781 | return __ioat3_prep_pq_lock(chan, pqres, pq, src, src_cnt, scf, len, | |
782 | flags); | |
783 | } | |
784 | ||
ae786624 DW |
785 | static struct dma_async_tx_descriptor * |
786 | ioat3_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src, | |
787 | unsigned int src_cnt, size_t len, unsigned long flags) | |
788 | { | |
789 | unsigned char scf[src_cnt]; | |
790 | dma_addr_t pq[2]; | |
791 | ||
792 | memset(scf, 0, src_cnt); | |
ae786624 | 793 | pq[0] = dst; |
de581b65 DW |
794 | flags |= DMA_PREP_PQ_DISABLE_Q; |
795 | pq[1] = dst; /* specify valid address for disabled result */ | |
ae786624 DW |
796 | |
797 | return __ioat3_prep_pq_lock(chan, NULL, pq, src, src_cnt, scf, len, | |
798 | flags); | |
799 | } | |
800 | ||
801 | struct dma_async_tx_descriptor * | |
802 | ioat3_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src, | |
803 | unsigned int src_cnt, size_t len, | |
804 | enum sum_check_flags *result, unsigned long flags) | |
805 | { | |
806 | unsigned char scf[src_cnt]; | |
807 | dma_addr_t pq[2]; | |
808 | ||
809 | /* the cleanup routine only sets bits on validate failure, it | |
810 | * does not clear bits on validate success... so clear it here | |
811 | */ | |
812 | *result = 0; | |
813 | ||
814 | memset(scf, 0, src_cnt); | |
ae786624 | 815 | pq[0] = src[0]; |
de581b65 DW |
816 | flags |= DMA_PREP_PQ_DISABLE_Q; |
817 | pq[1] = pq[0]; /* specify valid address for disabled result */ | |
ae786624 DW |
818 | |
819 | return __ioat3_prep_pq_lock(chan, result, pq, &src[1], src_cnt - 1, scf, | |
820 | len, flags); | |
821 | } | |
822 | ||
58c8649e DW |
823 | static struct dma_async_tx_descriptor * |
824 | ioat3_prep_interrupt_lock(struct dma_chan *c, unsigned long flags) | |
825 | { | |
826 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | |
827 | struct ioat_ring_ent *desc; | |
828 | struct ioat_dma_descriptor *hw; | |
58c8649e | 829 | |
074cc476 DW |
830 | if (ioat2_check_space_lock(ioat, 1) == 0) |
831 | desc = ioat2_get_ring_ent(ioat, ioat->head); | |
58c8649e DW |
832 | else |
833 | return NULL; | |
834 | ||
835 | hw = desc->hw; | |
836 | hw->ctl = 0; | |
837 | hw->ctl_f.null = 1; | |
838 | hw->ctl_f.int_en = 1; | |
839 | hw->ctl_f.fence = !!(flags & DMA_PREP_FENCE); | |
840 | hw->ctl_f.compl_write = 1; | |
841 | hw->size = NULL_DESC_BUFFER_SIZE; | |
842 | hw->src_addr = 0; | |
843 | hw->dst_addr = 0; | |
844 | ||
845 | desc->txd.flags = flags; | |
846 | desc->len = 1; | |
847 | ||
848 | dump_desc_dbg(ioat, desc); | |
849 | ||
850 | /* we leave the channel locked to ensure in order submission */ | |
851 | return &desc->txd; | |
852 | } | |
853 | ||
4bf27b8b | 854 | static void ioat3_dma_test_callback(void *dma_async_param) |
9de6fc71 DW |
855 | { |
856 | struct completion *cmp = dma_async_param; | |
857 | ||
858 | complete(cmp); | |
859 | } | |
860 | ||
861 | #define IOAT_NUM_SRC_TEST 6 /* must be <= 8 */ | |
4bf27b8b | 862 | static int ioat_xor_val_self_test(struct ioatdma_device *device) |
9de6fc71 DW |
863 | { |
864 | int i, src_idx; | |
865 | struct page *dest; | |
866 | struct page *xor_srcs[IOAT_NUM_SRC_TEST]; | |
867 | struct page *xor_val_srcs[IOAT_NUM_SRC_TEST + 1]; | |
868 | dma_addr_t dma_srcs[IOAT_NUM_SRC_TEST + 1]; | |
869 | dma_addr_t dma_addr, dest_dma; | |
870 | struct dma_async_tx_descriptor *tx; | |
871 | struct dma_chan *dma_chan; | |
872 | dma_cookie_t cookie; | |
873 | u8 cmp_byte = 0; | |
874 | u32 cmp_word; | |
875 | u32 xor_val_result; | |
876 | int err = 0; | |
877 | struct completion cmp; | |
878 | unsigned long tmo; | |
879 | struct device *dev = &device->pdev->dev; | |
880 | struct dma_device *dma = &device->common; | |
7369f56e | 881 | u8 op = 0; |
9de6fc71 DW |
882 | |
883 | dev_dbg(dev, "%s\n", __func__); | |
884 | ||
885 | if (!dma_has_cap(DMA_XOR, dma->cap_mask)) | |
886 | return 0; | |
887 | ||
888 | for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) { | |
889 | xor_srcs[src_idx] = alloc_page(GFP_KERNEL); | |
890 | if (!xor_srcs[src_idx]) { | |
891 | while (src_idx--) | |
892 | __free_page(xor_srcs[src_idx]); | |
893 | return -ENOMEM; | |
894 | } | |
895 | } | |
896 | ||
897 | dest = alloc_page(GFP_KERNEL); | |
898 | if (!dest) { | |
899 | while (src_idx--) | |
900 | __free_page(xor_srcs[src_idx]); | |
901 | return -ENOMEM; | |
902 | } | |
903 | ||
904 | /* Fill in src buffers */ | |
905 | for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) { | |
906 | u8 *ptr = page_address(xor_srcs[src_idx]); | |
907 | for (i = 0; i < PAGE_SIZE; i++) | |
908 | ptr[i] = (1 << src_idx); | |
909 | } | |
910 | ||
911 | for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) | |
912 | cmp_byte ^= (u8) (1 << src_idx); | |
913 | ||
914 | cmp_word = (cmp_byte << 24) | (cmp_byte << 16) | | |
915 | (cmp_byte << 8) | cmp_byte; | |
916 | ||
917 | memset(page_address(dest), 0, PAGE_SIZE); | |
918 | ||
919 | dma_chan = container_of(dma->channels.next, struct dma_chan, | |
920 | device_node); | |
921 | if (dma->device_alloc_chan_resources(dma_chan) < 1) { | |
922 | err = -ENODEV; | |
923 | goto out; | |
924 | } | |
925 | ||
926 | /* test xor */ | |
7369f56e BZ |
927 | op = IOAT_OP_XOR; |
928 | ||
9de6fc71 DW |
929 | dest_dma = dma_map_page(dev, dest, 0, PAGE_SIZE, DMA_FROM_DEVICE); |
930 | for (i = 0; i < IOAT_NUM_SRC_TEST; i++) | |
931 | dma_srcs[i] = dma_map_page(dev, xor_srcs[i], 0, PAGE_SIZE, | |
932 | DMA_TO_DEVICE); | |
933 | tx = dma->device_prep_dma_xor(dma_chan, dest_dma, dma_srcs, | |
934 | IOAT_NUM_SRC_TEST, PAGE_SIZE, | |
7369f56e BZ |
935 | DMA_PREP_INTERRUPT | |
936 | DMA_COMPL_SKIP_SRC_UNMAP | | |
937 | DMA_COMPL_SKIP_DEST_UNMAP); | |
9de6fc71 DW |
938 | |
939 | if (!tx) { | |
940 | dev_err(dev, "Self-test xor prep failed\n"); | |
941 | err = -ENODEV; | |
7369f56e | 942 | goto dma_unmap; |
9de6fc71 DW |
943 | } |
944 | ||
945 | async_tx_ack(tx); | |
946 | init_completion(&cmp); | |
947 | tx->callback = ioat3_dma_test_callback; | |
948 | tx->callback_param = &cmp; | |
949 | cookie = tx->tx_submit(tx); | |
950 | if (cookie < 0) { | |
951 | dev_err(dev, "Self-test xor setup failed\n"); | |
952 | err = -ENODEV; | |
7369f56e | 953 | goto dma_unmap; |
9de6fc71 DW |
954 | } |
955 | dma->device_issue_pending(dma_chan); | |
956 | ||
957 | tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)); | |
958 | ||
07934481 | 959 | if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_SUCCESS) { |
9de6fc71 DW |
960 | dev_err(dev, "Self-test xor timed out\n"); |
961 | err = -ENODEV; | |
7369f56e | 962 | goto dma_unmap; |
9de6fc71 DW |
963 | } |
964 | ||
7369f56e BZ |
965 | dma_unmap_page(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE); |
966 | for (i = 0; i < IOAT_NUM_SRC_TEST; i++) | |
967 | dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE); | |
968 | ||
9de6fc71 DW |
969 | dma_sync_single_for_cpu(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE); |
970 | for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) { | |
971 | u32 *ptr = page_address(dest); | |
972 | if (ptr[i] != cmp_word) { | |
973 | dev_err(dev, "Self-test xor failed compare\n"); | |
974 | err = -ENODEV; | |
975 | goto free_resources; | |
976 | } | |
977 | } | |
ac498987 | 978 | dma_sync_single_for_device(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE); |
9de6fc71 DW |
979 | |
980 | /* skip validate if the capability is not present */ | |
981 | if (!dma_has_cap(DMA_XOR_VAL, dma_chan->device->cap_mask)) | |
982 | goto free_resources; | |
983 | ||
7369f56e BZ |
984 | op = IOAT_OP_XOR_VAL; |
985 | ||
9de6fc71 DW |
986 | /* validate the sources with the destintation page */ |
987 | for (i = 0; i < IOAT_NUM_SRC_TEST; i++) | |
988 | xor_val_srcs[i] = xor_srcs[i]; | |
989 | xor_val_srcs[i] = dest; | |
990 | ||
991 | xor_val_result = 1; | |
992 | ||
993 | for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) | |
994 | dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE, | |
995 | DMA_TO_DEVICE); | |
996 | tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs, | |
997 | IOAT_NUM_SRC_TEST + 1, PAGE_SIZE, | |
7369f56e BZ |
998 | &xor_val_result, DMA_PREP_INTERRUPT | |
999 | DMA_COMPL_SKIP_SRC_UNMAP | | |
1000 | DMA_COMPL_SKIP_DEST_UNMAP); | |
9de6fc71 DW |
1001 | if (!tx) { |
1002 | dev_err(dev, "Self-test zero prep failed\n"); | |
1003 | err = -ENODEV; | |
7369f56e | 1004 | goto dma_unmap; |
9de6fc71 DW |
1005 | } |
1006 | ||
1007 | async_tx_ack(tx); | |
1008 | init_completion(&cmp); | |
1009 | tx->callback = ioat3_dma_test_callback; | |
1010 | tx->callback_param = &cmp; | |
1011 | cookie = tx->tx_submit(tx); | |
1012 | if (cookie < 0) { | |
1013 | dev_err(dev, "Self-test zero setup failed\n"); | |
1014 | err = -ENODEV; | |
7369f56e | 1015 | goto dma_unmap; |
9de6fc71 DW |
1016 | } |
1017 | dma->device_issue_pending(dma_chan); | |
1018 | ||
1019 | tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)); | |
1020 | ||
07934481 | 1021 | if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_SUCCESS) { |
9de6fc71 DW |
1022 | dev_err(dev, "Self-test validate timed out\n"); |
1023 | err = -ENODEV; | |
7369f56e | 1024 | goto dma_unmap; |
9de6fc71 DW |
1025 | } |
1026 | ||
7369f56e BZ |
1027 | for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) |
1028 | dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE); | |
1029 | ||
9de6fc71 DW |
1030 | if (xor_val_result != 0) { |
1031 | dev_err(dev, "Self-test validate failed compare\n"); | |
1032 | err = -ENODEV; | |
1033 | goto free_resources; | |
1034 | } | |
1035 | ||
1036 | /* skip memset if the capability is not present */ | |
1037 | if (!dma_has_cap(DMA_MEMSET, dma_chan->device->cap_mask)) | |
1038 | goto free_resources; | |
1039 | ||
1040 | /* test memset */ | |
7369f56e BZ |
1041 | op = IOAT_OP_FILL; |
1042 | ||
9de6fc71 DW |
1043 | dma_addr = dma_map_page(dev, dest, 0, |
1044 | PAGE_SIZE, DMA_FROM_DEVICE); | |
1045 | tx = dma->device_prep_dma_memset(dma_chan, dma_addr, 0, PAGE_SIZE, | |
7369f56e BZ |
1046 | DMA_PREP_INTERRUPT | |
1047 | DMA_COMPL_SKIP_SRC_UNMAP | | |
1048 | DMA_COMPL_SKIP_DEST_UNMAP); | |
9de6fc71 DW |
1049 | if (!tx) { |
1050 | dev_err(dev, "Self-test memset prep failed\n"); | |
1051 | err = -ENODEV; | |
7369f56e | 1052 | goto dma_unmap; |
9de6fc71 DW |
1053 | } |
1054 | ||
1055 | async_tx_ack(tx); | |
1056 | init_completion(&cmp); | |
1057 | tx->callback = ioat3_dma_test_callback; | |
1058 | tx->callback_param = &cmp; | |
1059 | cookie = tx->tx_submit(tx); | |
1060 | if (cookie < 0) { | |
1061 | dev_err(dev, "Self-test memset setup failed\n"); | |
1062 | err = -ENODEV; | |
7369f56e | 1063 | goto dma_unmap; |
9de6fc71 DW |
1064 | } |
1065 | dma->device_issue_pending(dma_chan); | |
1066 | ||
1067 | tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)); | |
1068 | ||
07934481 | 1069 | if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_SUCCESS) { |
9de6fc71 DW |
1070 | dev_err(dev, "Self-test memset timed out\n"); |
1071 | err = -ENODEV; | |
7369f56e | 1072 | goto dma_unmap; |
9de6fc71 DW |
1073 | } |
1074 | ||
7369f56e BZ |
1075 | dma_unmap_page(dev, dma_addr, PAGE_SIZE, DMA_FROM_DEVICE); |
1076 | ||
9de6fc71 DW |
1077 | for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) { |
1078 | u32 *ptr = page_address(dest); | |
1079 | if (ptr[i]) { | |
1080 | dev_err(dev, "Self-test memset failed compare\n"); | |
1081 | err = -ENODEV; | |
1082 | goto free_resources; | |
1083 | } | |
1084 | } | |
1085 | ||
1086 | /* test for non-zero parity sum */ | |
7369f56e BZ |
1087 | op = IOAT_OP_XOR_VAL; |
1088 | ||
9de6fc71 DW |
1089 | xor_val_result = 0; |
1090 | for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) | |
1091 | dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE, | |
1092 | DMA_TO_DEVICE); | |
1093 | tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs, | |
1094 | IOAT_NUM_SRC_TEST + 1, PAGE_SIZE, | |
7369f56e BZ |
1095 | &xor_val_result, DMA_PREP_INTERRUPT | |
1096 | DMA_COMPL_SKIP_SRC_UNMAP | | |
1097 | DMA_COMPL_SKIP_DEST_UNMAP); | |
9de6fc71 DW |
1098 | if (!tx) { |
1099 | dev_err(dev, "Self-test 2nd zero prep failed\n"); | |
1100 | err = -ENODEV; | |
7369f56e | 1101 | goto dma_unmap; |
9de6fc71 DW |
1102 | } |
1103 | ||
1104 | async_tx_ack(tx); | |
1105 | init_completion(&cmp); | |
1106 | tx->callback = ioat3_dma_test_callback; | |
1107 | tx->callback_param = &cmp; | |
1108 | cookie = tx->tx_submit(tx); | |
1109 | if (cookie < 0) { | |
1110 | dev_err(dev, "Self-test 2nd zero setup failed\n"); | |
1111 | err = -ENODEV; | |
7369f56e | 1112 | goto dma_unmap; |
9de6fc71 DW |
1113 | } |
1114 | dma->device_issue_pending(dma_chan); | |
1115 | ||
1116 | tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)); | |
1117 | ||
07934481 | 1118 | if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_SUCCESS) { |
9de6fc71 DW |
1119 | dev_err(dev, "Self-test 2nd validate timed out\n"); |
1120 | err = -ENODEV; | |
7369f56e | 1121 | goto dma_unmap; |
9de6fc71 DW |
1122 | } |
1123 | ||
1124 | if (xor_val_result != SUM_CHECK_P_RESULT) { | |
1125 | dev_err(dev, "Self-test validate failed compare\n"); | |
1126 | err = -ENODEV; | |
7369f56e | 1127 | goto dma_unmap; |
9de6fc71 DW |
1128 | } |
1129 | ||
7369f56e BZ |
1130 | for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) |
1131 | dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE); | |
1132 | ||
1133 | goto free_resources; | |
1134 | dma_unmap: | |
1135 | if (op == IOAT_OP_XOR) { | |
1136 | dma_unmap_page(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE); | |
1137 | for (i = 0; i < IOAT_NUM_SRC_TEST; i++) | |
1138 | dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, | |
1139 | DMA_TO_DEVICE); | |
1140 | } else if (op == IOAT_OP_XOR_VAL) { | |
1141 | for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) | |
1142 | dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, | |
1143 | DMA_TO_DEVICE); | |
1144 | } else if (op == IOAT_OP_FILL) | |
1145 | dma_unmap_page(dev, dma_addr, PAGE_SIZE, DMA_FROM_DEVICE); | |
9de6fc71 DW |
1146 | free_resources: |
1147 | dma->device_free_chan_resources(dma_chan); | |
1148 | out: | |
1149 | src_idx = IOAT_NUM_SRC_TEST; | |
1150 | while (src_idx--) | |
1151 | __free_page(xor_srcs[src_idx]); | |
1152 | __free_page(dest); | |
1153 | return err; | |
1154 | } | |
1155 | ||
4bf27b8b | 1156 | static int ioat3_dma_self_test(struct ioatdma_device *device) |
9de6fc71 DW |
1157 | { |
1158 | int rc = ioat_dma_self_test(device); | |
1159 | ||
1160 | if (rc) | |
1161 | return rc; | |
1162 | ||
1163 | rc = ioat_xor_val_self_test(device); | |
1164 | if (rc) | |
1165 | return rc; | |
1166 | ||
1167 | return 0; | |
1168 | } | |
1169 | ||
a6d52d70 DW |
1170 | static int ioat3_reset_hw(struct ioat_chan_common *chan) |
1171 | { | |
1172 | /* throw away whatever the channel was doing and get it | |
1173 | * initialized, with ioat3 specific workarounds | |
1174 | */ | |
1175 | struct ioatdma_device *device = chan->device; | |
1176 | struct pci_dev *pdev = device->pdev; | |
1177 | u32 chanerr; | |
1178 | u16 dev_id; | |
1179 | int err; | |
1180 | ||
1181 | ioat2_quiesce(chan, msecs_to_jiffies(100)); | |
1182 | ||
1183 | chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET); | |
1184 | writel(chanerr, chan->reg_base + IOAT_CHANERR_OFFSET); | |
1185 | ||
6decffd5 | 1186 | /* clear any pending errors */ |
a6d52d70 DW |
1187 | err = pci_read_config_dword(pdev, IOAT_PCI_CHANERR_INT_OFFSET, &chanerr); |
1188 | if (err) { | |
1189 | dev_err(&pdev->dev, "channel error register unreachable\n"); | |
1190 | return err; | |
1191 | } | |
1192 | pci_write_config_dword(pdev, IOAT_PCI_CHANERR_INT_OFFSET, chanerr); | |
1193 | ||
1194 | /* Clear DMAUNCERRSTS Cfg-Reg Parity Error status bit | |
1195 | * (workaround for spurious config parity error after restart) | |
1196 | */ | |
1197 | pci_read_config_word(pdev, IOAT_PCI_DEVICE_ID_OFFSET, &dev_id); | |
1198 | if (dev_id == PCI_DEVICE_ID_INTEL_IOAT_TBG0) | |
1199 | pci_write_config_dword(pdev, IOAT_PCI_DMAUNCERRSTS_OFFSET, 0x10); | |
1200 | ||
1201 | return ioat2_reset_sync(chan, msecs_to_jiffies(200)); | |
1202 | } | |
1203 | ||
f26df1a1 DJ |
1204 | static bool is_jf_ioat(struct pci_dev *pdev) |
1205 | { | |
1206 | switch (pdev->device) { | |
1207 | case PCI_DEVICE_ID_INTEL_IOAT_JSF0: | |
1208 | case PCI_DEVICE_ID_INTEL_IOAT_JSF1: | |
1209 | case PCI_DEVICE_ID_INTEL_IOAT_JSF2: | |
1210 | case PCI_DEVICE_ID_INTEL_IOAT_JSF3: | |
1211 | case PCI_DEVICE_ID_INTEL_IOAT_JSF4: | |
1212 | case PCI_DEVICE_ID_INTEL_IOAT_JSF5: | |
1213 | case PCI_DEVICE_ID_INTEL_IOAT_JSF6: | |
1214 | case PCI_DEVICE_ID_INTEL_IOAT_JSF7: | |
1215 | case PCI_DEVICE_ID_INTEL_IOAT_JSF8: | |
1216 | case PCI_DEVICE_ID_INTEL_IOAT_JSF9: | |
1217 | return true; | |
1218 | default: | |
1219 | return false; | |
1220 | } | |
1221 | } | |
1222 | ||
1223 | static bool is_snb_ioat(struct pci_dev *pdev) | |
1224 | { | |
1225 | switch (pdev->device) { | |
1226 | case PCI_DEVICE_ID_INTEL_IOAT_SNB0: | |
1227 | case PCI_DEVICE_ID_INTEL_IOAT_SNB1: | |
1228 | case PCI_DEVICE_ID_INTEL_IOAT_SNB2: | |
1229 | case PCI_DEVICE_ID_INTEL_IOAT_SNB3: | |
1230 | case PCI_DEVICE_ID_INTEL_IOAT_SNB4: | |
1231 | case PCI_DEVICE_ID_INTEL_IOAT_SNB5: | |
1232 | case PCI_DEVICE_ID_INTEL_IOAT_SNB6: | |
1233 | case PCI_DEVICE_ID_INTEL_IOAT_SNB7: | |
1234 | case PCI_DEVICE_ID_INTEL_IOAT_SNB8: | |
1235 | case PCI_DEVICE_ID_INTEL_IOAT_SNB9: | |
1236 | return true; | |
1237 | default: | |
1238 | return false; | |
1239 | } | |
1240 | } | |
1241 | ||
1a363068 DJ |
1242 | static bool is_ivb_ioat(struct pci_dev *pdev) |
1243 | { | |
1244 | switch (pdev->device) { | |
1245 | case PCI_DEVICE_ID_INTEL_IOAT_IVB0: | |
1246 | case PCI_DEVICE_ID_INTEL_IOAT_IVB1: | |
1247 | case PCI_DEVICE_ID_INTEL_IOAT_IVB2: | |
1248 | case PCI_DEVICE_ID_INTEL_IOAT_IVB3: | |
1249 | case PCI_DEVICE_ID_INTEL_IOAT_IVB4: | |
1250 | case PCI_DEVICE_ID_INTEL_IOAT_IVB5: | |
1251 | case PCI_DEVICE_ID_INTEL_IOAT_IVB6: | |
1252 | case PCI_DEVICE_ID_INTEL_IOAT_IVB7: | |
1253 | case PCI_DEVICE_ID_INTEL_IOAT_IVB8: | |
1254 | case PCI_DEVICE_ID_INTEL_IOAT_IVB9: | |
1255 | return true; | |
1256 | default: | |
1257 | return false; | |
1258 | } | |
1259 | ||
1260 | } | |
1261 | ||
4bf27b8b | 1262 | int ioat3_dma_probe(struct ioatdma_device *device, int dca) |
bf40a686 DW |
1263 | { |
1264 | struct pci_dev *pdev = device->pdev; | |
228c4f5c | 1265 | int dca_en = system_has_dca_enabled(pdev); |
bf40a686 DW |
1266 | struct dma_device *dma; |
1267 | struct dma_chan *c; | |
1268 | struct ioat_chan_common *chan; | |
e3232714 | 1269 | bool is_raid_device = false; |
bf40a686 | 1270 | int err; |
bf40a686 DW |
1271 | u32 cap; |
1272 | ||
1273 | device->enumerate_channels = ioat2_enumerate_channels; | |
a6d52d70 | 1274 | device->reset_hw = ioat3_reset_hw; |
9de6fc71 | 1275 | device->self_test = ioat3_dma_self_test; |
bf40a686 DW |
1276 | dma = &device->common; |
1277 | dma->device_prep_dma_memcpy = ioat2_dma_prep_memcpy_lock; | |
1278 | dma->device_issue_pending = ioat2_issue_pending; | |
1279 | dma->device_alloc_chan_resources = ioat2_alloc_chan_resources; | |
1280 | dma->device_free_chan_resources = ioat2_free_chan_resources; | |
58c8649e | 1281 | |
1a363068 | 1282 | if (is_jf_ioat(pdev) || is_snb_ioat(pdev) || is_ivb_ioat(pdev)) |
f26df1a1 DJ |
1283 | dma->copy_align = 6; |
1284 | ||
58c8649e DW |
1285 | dma_cap_set(DMA_INTERRUPT, dma->cap_mask); |
1286 | dma->device_prep_dma_interrupt = ioat3_prep_interrupt_lock; | |
1287 | ||
bf40a686 | 1288 | cap = readl(device->reg_base + IOAT_DMA_CAP_OFFSET); |
228c4f5c DW |
1289 | |
1290 | /* dca is incompatible with raid operations */ | |
1291 | if (dca_en && (cap & (IOAT_CAP_XOR|IOAT_CAP_PQ))) | |
1292 | cap &= ~(IOAT_CAP_XOR|IOAT_CAP_PQ); | |
1293 | ||
b094ad3b | 1294 | if (cap & IOAT_CAP_XOR) { |
e3232714 | 1295 | is_raid_device = true; |
b094ad3b | 1296 | dma->max_xor = 8; |
2adfc550 | 1297 | dma->xor_align = 6; |
b094ad3b DW |
1298 | |
1299 | dma_cap_set(DMA_XOR, dma->cap_mask); | |
1300 | dma->device_prep_dma_xor = ioat3_prep_xor; | |
1301 | ||
1302 | dma_cap_set(DMA_XOR_VAL, dma->cap_mask); | |
1303 | dma->device_prep_dma_xor_val = ioat3_prep_xor_val; | |
1304 | } | |
d69d235b | 1305 | if (cap & IOAT_CAP_PQ) { |
e3232714 | 1306 | is_raid_device = true; |
d69d235b | 1307 | dma_set_maxpq(dma, 8, 0); |
2adfc550 | 1308 | dma->pq_align = 6; |
d69d235b DW |
1309 | |
1310 | dma_cap_set(DMA_PQ, dma->cap_mask); | |
1311 | dma->device_prep_dma_pq = ioat3_prep_pq; | |
1312 | ||
1313 | dma_cap_set(DMA_PQ_VAL, dma->cap_mask); | |
1314 | dma->device_prep_dma_pq_val = ioat3_prep_pq_val; | |
ae786624 DW |
1315 | |
1316 | if (!(cap & IOAT_CAP_XOR)) { | |
1317 | dma->max_xor = 8; | |
2adfc550 | 1318 | dma->xor_align = 6; |
ae786624 DW |
1319 | |
1320 | dma_cap_set(DMA_XOR, dma->cap_mask); | |
1321 | dma->device_prep_dma_xor = ioat3_prep_pqxor; | |
1322 | ||
1323 | dma_cap_set(DMA_XOR_VAL, dma->cap_mask); | |
1324 | dma->device_prep_dma_xor_val = ioat3_prep_pqxor_val; | |
1325 | } | |
d69d235b | 1326 | } |
e3232714 DW |
1327 | if (is_raid_device && (cap & IOAT_CAP_FILL_BLOCK)) { |
1328 | dma_cap_set(DMA_MEMSET, dma->cap_mask); | |
1329 | dma->device_prep_dma_memset = ioat3_prep_memset_lock; | |
1330 | } | |
1331 | ||
1332 | ||
1333 | if (is_raid_device) { | |
07934481 | 1334 | dma->device_tx_status = ioat3_tx_status; |
aa4d72ae | 1335 | device->cleanup_fn = ioat3_cleanup_event; |
e3232714 DW |
1336 | device->timer_fn = ioat3_timer_event; |
1337 | } else { | |
07934481 | 1338 | dma->device_tx_status = ioat_dma_tx_status; |
aa4d72ae | 1339 | device->cleanup_fn = ioat2_cleanup_event; |
e3232714 DW |
1340 | device->timer_fn = ioat2_timer_event; |
1341 | } | |
bf40a686 | 1342 | |
7b3cc2b1 DW |
1343 | #ifdef CONFIG_ASYNC_TX_DISABLE_PQ_VAL_DMA |
1344 | dma_cap_clear(DMA_PQ_VAL, dma->cap_mask); | |
1345 | dma->device_prep_dma_pq_val = NULL; | |
1346 | #endif | |
1347 | ||
1348 | #ifdef CONFIG_ASYNC_TX_DISABLE_XOR_VAL_DMA | |
1349 | dma_cap_clear(DMA_XOR_VAL, dma->cap_mask); | |
1350 | dma->device_prep_dma_xor_val = NULL; | |
1351 | #endif | |
1352 | ||
bf40a686 DW |
1353 | err = ioat_probe(device); |
1354 | if (err) | |
1355 | return err; | |
1356 | ioat_set_tcp_copy_break(262144); | |
1357 | ||
1358 | list_for_each_entry(c, &dma->channels, device_node) { | |
1359 | chan = to_chan_common(c); | |
1360 | writel(IOAT_DMA_DCA_ANY_CPU, | |
1361 | chan->reg_base + IOAT_DCACTRL_OFFSET); | |
1362 | } | |
1363 | ||
1364 | err = ioat_register(device); | |
1365 | if (err) | |
1366 | return err; | |
5669e31c DW |
1367 | |
1368 | ioat_kobject_add(device, &ioat2_ktype); | |
1369 | ||
bf40a686 DW |
1370 | if (dca) |
1371 | device->dca = ioat3_dca_init(pdev, device->reg_base); | |
1372 | ||
1373 | return 0; | |
1374 | } |