?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Lynx framebuffers multidomain implementation.
[linux/elbrus.git] / drivers / dma / ioat / dma_v3.c
blobb9b38a1cf92fbdc4147e70af93508f0d9caf6867
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 #include <linux/module.h>
59 #include <linux/pci.h>
60 #include <linux/gfp.h>
61 #include <linux/dmaengine.h>
62 #include <linux/dma-mapping.h>
63 #include <linux/prefetch.h>
64 #include "../dmaengine.h"
65 #include "registers.h"
66 #include "hw.h"
67 #include "dma.h"
68 #include "dma_v2.h"
69
70 extern struct kmem_cache *ioat3_sed_cache;
71
72 /* ioat hardware assumes at least two sources for raid operations */
73 #define src_cnt_to_sw(x) ((x) + 2)
74 #define src_cnt_to_hw(x) ((x) - 2)
75 #define ndest_to_sw(x) ((x) + 1)
76 #define ndest_to_hw(x) ((x) - 1)
77 #define src16_cnt_to_sw(x) ((x) + 9)
78 #define src16_cnt_to_hw(x) ((x) - 9)
79
80 /* provide a lookup table for setting the source address in the base or
81 * extended descriptor of an xor or pq descriptor
82 */
83 static const u8 xor_idx_to_desc = 0xe0;
84 static const u8 xor_idx_to_field[] = { 1, 4, 5, 6, 7, 0, 1, 2 };
85 static const u8 pq_idx_to_desc = 0xf8;
86 static const u8 pq16_idx_to_desc[] = { 0, 0, 1, 1, 1, 1, 1, 1, 1,
87 2, 2, 2, 2, 2, 2, 2 };
88 static const u8 pq_idx_to_field[] = { 1, 4, 5, 0, 1, 2, 4, 5 };
89 static const u8 pq16_idx_to_field[] = { 1, 4, 1, 2, 3, 4, 5, 6, 7,
90 0, 1, 2, 3, 4, 5, 6 };
91
92 static void ioat3_eh(struct ioat2_dma_chan *ioat);
93
94 static void xor_set_src(struct ioat_raw_descriptor *descs[2],
95 dma_addr_t addr, u32 offset, int idx)
96 {
97 struct ioat_raw_descriptor *raw = descs[xor_idx_to_desc >> idx & 1];
98
99 raw->field[xor_idx_to_field[idx]] = addr + offset;
100 }
101
102 static dma_addr_t pq_get_src(struct ioat_raw_descriptor *descs[2], int idx)
103 {
104 struct ioat_raw_descriptor *raw = descs[pq_idx_to_desc >> idx & 1];
105
106 return raw->field[pq_idx_to_field[idx]];
107 }
108
109 static dma_addr_t pq16_get_src(struct ioat_raw_descriptor *desc[3], int idx)
110 {
111 struct ioat_raw_descriptor *raw = desc[pq16_idx_to_desc[idx]];
112
113 return raw->field[pq16_idx_to_field[idx]];
114 }
115
116 static void pq_set_src(struct ioat_raw_descriptor *descs[2],
117 dma_addr_t addr, u32 offset, u8 coef, int idx)
118 {
119 struct ioat_pq_descriptor *pq = (struct ioat_pq_descriptor *) descs[0];
120 struct ioat_raw_descriptor *raw = descs[pq_idx_to_desc >> idx & 1];
121
122 raw->field[pq_idx_to_field[idx]] = addr + offset;
123 pq->coef[idx] = coef;
124 }
125
126 static bool is_jf_ioat(struct pci_dev *pdev)
127 {
128 switch (pdev->device) {
129 case PCI_DEVICE_ID_INTEL_IOAT_JSF0:
130 case PCI_DEVICE_ID_INTEL_IOAT_JSF1:
131 case PCI_DEVICE_ID_INTEL_IOAT_JSF2:
132 case PCI_DEVICE_ID_INTEL_IOAT_JSF3:
133 case PCI_DEVICE_ID_INTEL_IOAT_JSF4:
134 case PCI_DEVICE_ID_INTEL_IOAT_JSF5:
135 case PCI_DEVICE_ID_INTEL_IOAT_JSF6:
136 case PCI_DEVICE_ID_INTEL_IOAT_JSF7:
137 case PCI_DEVICE_ID_INTEL_IOAT_JSF8:
138 case PCI_DEVICE_ID_INTEL_IOAT_JSF9:
139 return true;
140 default:
141 return false;
142 }
143 }
144
145 static bool is_snb_ioat(struct pci_dev *pdev)
146 {
147 switch (pdev->device) {
148 case PCI_DEVICE_ID_INTEL_IOAT_SNB0:
149 case PCI_DEVICE_ID_INTEL_IOAT_SNB1:
150 case PCI_DEVICE_ID_INTEL_IOAT_SNB2:
151 case PCI_DEVICE_ID_INTEL_IOAT_SNB3:
152 case PCI_DEVICE_ID_INTEL_IOAT_SNB4:
153 case PCI_DEVICE_ID_INTEL_IOAT_SNB5:
154 case PCI_DEVICE_ID_INTEL_IOAT_SNB6:
155 case PCI_DEVICE_ID_INTEL_IOAT_SNB7:
156 case PCI_DEVICE_ID_INTEL_IOAT_SNB8:
157 case PCI_DEVICE_ID_INTEL_IOAT_SNB9:
158 return true;
159 default:
160 return false;
161 }
162 }
163
164 static bool is_ivb_ioat(struct pci_dev *pdev)
165 {
166 switch (pdev->device) {
167 case PCI_DEVICE_ID_INTEL_IOAT_IVB0:
168 case PCI_DEVICE_ID_INTEL_IOAT_IVB1:
169 case PCI_DEVICE_ID_INTEL_IOAT_IVB2:
170 case PCI_DEVICE_ID_INTEL_IOAT_IVB3:
171 case PCI_DEVICE_ID_INTEL_IOAT_IVB4:
172 case PCI_DEVICE_ID_INTEL_IOAT_IVB5:
173 case PCI_DEVICE_ID_INTEL_IOAT_IVB6:
174 case PCI_DEVICE_ID_INTEL_IOAT_IVB7:
175 case PCI_DEVICE_ID_INTEL_IOAT_IVB8:
176 case PCI_DEVICE_ID_INTEL_IOAT_IVB9:
177 return true;
178 default:
179 return false;
180 }
181
182 }
183
184 static bool is_hsw_ioat(struct pci_dev *pdev)
185 {
186 switch (pdev->device) {
187 case PCI_DEVICE_ID_INTEL_IOAT_HSW0:
188 case PCI_DEVICE_ID_INTEL_IOAT_HSW1:
189 case PCI_DEVICE_ID_INTEL_IOAT_HSW2:
190 case PCI_DEVICE_ID_INTEL_IOAT_HSW3:
191 case PCI_DEVICE_ID_INTEL_IOAT_HSW4:
192 case PCI_DEVICE_ID_INTEL_IOAT_HSW5:
193 case PCI_DEVICE_ID_INTEL_IOAT_HSW6:
194 case PCI_DEVICE_ID_INTEL_IOAT_HSW7:
195 case PCI_DEVICE_ID_INTEL_IOAT_HSW8:
196 case PCI_DEVICE_ID_INTEL_IOAT_HSW9:
197 return true;
198 default:
199 return false;
200 }
201
202 }
203
204 static bool is_xeon_cb32(struct pci_dev *pdev)
205 {
206 return is_jf_ioat(pdev) || is_snb_ioat(pdev) || is_ivb_ioat(pdev) ||
207 is_hsw_ioat(pdev);
208 }
209
210 static bool is_bwd_ioat(struct pci_dev *pdev)
211 {
212 switch (pdev->device) {
213 case PCI_DEVICE_ID_INTEL_IOAT_BWD0:
214 case PCI_DEVICE_ID_INTEL_IOAT_BWD1:
215 case PCI_DEVICE_ID_INTEL_IOAT_BWD2:
216 case PCI_DEVICE_ID_INTEL_IOAT_BWD3:
217 return true;
218 default:
219 return false;
220 }
221 }
222
223 static bool is_bwd_noraid(struct pci_dev *pdev)
224 {
225 switch (pdev->device) {
226 case PCI_DEVICE_ID_INTEL_IOAT_BWD2:
227 case PCI_DEVICE_ID_INTEL_IOAT_BWD3:
228 return true;
229 default:
230 return false;
231 }
232
233 }
234
235 static void pq16_set_src(struct ioat_raw_descriptor *desc[3],
236 dma_addr_t addr, u32 offset, u8 coef, unsigned idx)
237 {
238 struct ioat_pq_descriptor *pq = (struct ioat_pq_descriptor *)desc[0];
239 struct ioat_pq16a_descriptor *pq16 =
240 (struct ioat_pq16a_descriptor *)desc[1];
241 struct ioat_raw_descriptor *raw = desc[pq16_idx_to_desc[idx]];
242
243 raw->field[pq16_idx_to_field[idx]] = addr + offset;
244
245 if (idx < 8)
246 pq->coef[idx] = coef;
247 else
248 pq16->coef[idx - 8] = coef;
249 }
250
251 static struct ioat_sed_ent *
252 ioat3_alloc_sed(struct ioatdma_device *device, unsigned int hw_pool)
253 {
254 struct ioat_sed_ent *sed;
255 gfp_t flags = __GFP_ZERO | GFP_ATOMIC;
256
257 sed = kmem_cache_alloc(ioat3_sed_cache, flags);
258 if (!sed)
259 return NULL;
260
261 sed->hw_pool = hw_pool;
262 sed->hw = dma_pool_alloc(device->sed_hw_pool[hw_pool],
263 flags, &sed->dma);
264 if (!sed->hw) {
265 kmem_cache_free(ioat3_sed_cache, sed);
266 return NULL;
267 }
268
269 return sed;
270 }
271
272 static void ioat3_free_sed(struct ioatdma_device *device, struct ioat_sed_ent *sed)
273 {
274 if (!sed)
275 return;
276
277 dma_pool_free(device->sed_hw_pool[sed->hw_pool], sed->hw, sed->dma);
278 kmem_cache_free(ioat3_sed_cache, sed);
279 }
280
281 static bool desc_has_ext(struct ioat_ring_ent *desc)
282 {
283 struct ioat_dma_descriptor *hw = desc->hw;
284
285 if (hw->ctl_f.op == IOAT_OP_XOR ||
286 hw->ctl_f.op == IOAT_OP_XOR_VAL) {
287 struct ioat_xor_descriptor *xor = desc->xor;
288
289 if (src_cnt_to_sw(xor->ctl_f.src_cnt) > 5)
290 return true;
291 } else if (hw->ctl_f.op == IOAT_OP_PQ ||
292 hw->ctl_f.op == IOAT_OP_PQ_VAL) {
293 struct ioat_pq_descriptor *pq = desc->pq;
294
295 if (src_cnt_to_sw(pq->ctl_f.src_cnt) > 3)
296 return true;
297 }
298
299 return false;
300 }
301
302 static u64 ioat3_get_current_completion(struct ioat_chan_common *chan)
303 {
304 u64 phys_complete;
305 u64 completion;
306
307 completion = *chan->completion;
308 phys_complete = ioat_chansts_to_addr(completion);
309
310 dev_dbg(to_dev(chan), "%s: phys_complete: %#llx\n", __func__,
311 (unsigned long long) phys_complete);
312
313 return phys_complete;
314 }
315
316 static bool ioat3_cleanup_preamble(struct ioat_chan_common *chan,
317 u64 *phys_complete)
318 {
319 *phys_complete = ioat3_get_current_completion(chan);
320 if (*phys_complete == chan->last_completion)
321 return false;
322
323 clear_bit(IOAT_COMPLETION_ACK, &chan->state);
324 mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
325
326 return true;
327 }
328
329 static void
330 desc_get_errstat(struct ioat2_dma_chan *ioat, struct ioat_ring_ent *desc)
331 {
332 struct ioat_dma_descriptor *hw = desc->hw;
333
334 switch (hw->ctl_f.op) {
335 case IOAT_OP_PQ_VAL:
336 case IOAT_OP_PQ_VAL_16S:
337 {
338 struct ioat_pq_descriptor *pq = desc->pq;
339
340 /* check if there's error written */
341 if (!pq->dwbes_f.wbes)
342 return;
343
344 /* need to set a chanerr var for checking to clear later */
345
346 if (pq->dwbes_f.p_val_err)
347 *desc->result |= SUM_CHECK_P_RESULT;
348
349 if (pq->dwbes_f.q_val_err)
350 *desc->result |= SUM_CHECK_Q_RESULT;
351
352 return;
353 }
354 default:
355 return;
356 }
357 }
358
359 /**
360 * __cleanup - reclaim used descriptors
361 * @ioat: channel (ring) to clean
362 *
363 * The difference from the dma_v2.c __cleanup() is that this routine
364 * handles extended descriptors and dma-unmapping raid operations.
365 */
366 static void __cleanup(struct ioat2_dma_chan *ioat, dma_addr_t phys_complete)
367 {
368 struct ioat_chan_common *chan = &ioat->base;
369 struct ioatdma_device *device = chan->device;
370 struct ioat_ring_ent *desc;
371 bool seen_current = false;
372 int idx = ioat->tail, i;
373 u16 active;
374
375 dev_dbg(to_dev(chan), "%s: head: %#x tail: %#x issued: %#x\n",
376 __func__, ioat->head, ioat->tail, ioat->issued);
377
378 /*
379 * At restart of the channel, the completion address and the
380 * channel status will be 0 due to starting a new chain. Since
381 * it's new chain and the first descriptor "fails", there is
382 * nothing to clean up. We do not want to reap the entire submitted
383 * chain due to this 0 address value and then BUG.
384 */
385 if (!phys_complete)
386 return;
387
388 active = ioat2_ring_active(ioat);
389 for (i = 0; i < active && !seen_current; i++) {
390 struct dma_async_tx_descriptor *tx;
391
392 smp_read_barrier_depends();
393 prefetch(ioat2_get_ring_ent(ioat, idx + i + 1));
394 desc = ioat2_get_ring_ent(ioat, idx + i);
395 dump_desc_dbg(ioat, desc);
396
397 /* set err stat if we are using dwbes */
398 if (device->cap & IOAT_CAP_DWBES)
399 desc_get_errstat(ioat, desc);
400
401 tx = &desc->txd;
402 if (tx->cookie) {
403 dma_cookie_complete(tx);
404 dma_descriptor_unmap(tx);
405 if (tx->callback) {
406 tx->callback(tx->callback_param);
407 tx->callback = NULL;
408 }
409 }
410
411 if (tx->phys == phys_complete)
412 seen_current = true;
413
414 /* skip extended descriptors */
415 if (desc_has_ext(desc)) {
416 BUG_ON(i + 1 >= active);
417 i++;
418 }
419
420 /* cleanup super extended descriptors */
421 if (desc->sed) {
422 ioat3_free_sed(device, desc->sed);
423 desc->sed = NULL;
424 }
425 }
426 smp_mb(); /* finish all descriptor reads before incrementing tail */
427 ioat->tail = idx + i;
428 BUG_ON(active && !seen_current); /* no active descs have written a completion? */
429 chan->last_completion = phys_complete;
430
431 if (active - i == 0) {
432 dev_dbg(to_dev(chan), "%s: cancel completion timeout\n",
433 __func__);
434 clear_bit(IOAT_COMPLETION_PENDING, &chan->state);
435 mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
436 }
437 /* 5 microsecond delay per pending descriptor */
438 writew(min((5 * (active - i)), IOAT_INTRDELAY_MASK),
439 chan->device->reg_base + IOAT_INTRDELAY_OFFSET);
440 }
441
442 static void ioat3_cleanup(struct ioat2_dma_chan *ioat)
443 {
444 struct ioat_chan_common *chan = &ioat->base;
445 u64 phys_complete;
446
447 spin_lock_bh(&chan->cleanup_lock);
448
449 if (ioat3_cleanup_preamble(chan, &phys_complete))
450 __cleanup(ioat, phys_complete);
451
452 if (is_ioat_halted(*chan->completion)) {
453 u32 chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET);
454
455 if (chanerr & IOAT_CHANERR_HANDLE_MASK) {
456 mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
457 ioat3_eh(ioat);
458 }
459 }
460
461 spin_unlock_bh(&chan->cleanup_lock);
462 }
463
464 static void ioat3_cleanup_event(unsigned long data)
465 {
466 struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
467 struct ioat_chan_common *chan = &ioat->base;
468
469 ioat3_cleanup(ioat);
470 if (!test_bit(IOAT_RUN, &chan->state))
471 return;
472 writew(IOAT_CHANCTRL_RUN, ioat->base.reg_base + IOAT_CHANCTRL_OFFSET);
473 }
474
475 static void ioat3_restart_channel(struct ioat2_dma_chan *ioat)
476 {
477 struct ioat_chan_common *chan = &ioat->base;
478 u64 phys_complete;
479
480 ioat2_quiesce(chan, 0);
481 if (ioat3_cleanup_preamble(chan, &phys_complete))
482 __cleanup(ioat, phys_complete);
483
484 __ioat2_restart_chan(ioat);
485 }
486
487 static void ioat3_eh(struct ioat2_dma_chan *ioat)
488 {
489 struct ioat_chan_common *chan = &ioat->base;
490 struct pci_dev *pdev = to_pdev(chan);
491 struct ioat_dma_descriptor *hw;
492 u64 phys_complete;
493 struct ioat_ring_ent *desc;
494 u32 err_handled = 0;
495 u32 chanerr_int;
496 u32 chanerr;
497
498 /* cleanup so tail points to descriptor that caused the error */
499 if (ioat3_cleanup_preamble(chan, &phys_complete))
500 __cleanup(ioat, phys_complete);
501
502 chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET);
503 pci_read_config_dword(pdev, IOAT_PCI_CHANERR_INT_OFFSET, &chanerr_int);
504
505 dev_dbg(to_dev(chan), "%s: error = %x:%x\n",
506 __func__, chanerr, chanerr_int);
507
508 desc = ioat2_get_ring_ent(ioat, ioat->tail);
509 hw = desc->hw;
510 dump_desc_dbg(ioat, desc);
511
512 switch (hw->ctl_f.op) {
513 case IOAT_OP_XOR_VAL:
514 if (chanerr & IOAT_CHANERR_XOR_P_OR_CRC_ERR) {
515 *desc->result |= SUM_CHECK_P_RESULT;
516 err_handled |= IOAT_CHANERR_XOR_P_OR_CRC_ERR;
517 }
518 break;
519 case IOAT_OP_PQ_VAL:
520 case IOAT_OP_PQ_VAL_16S:
521 if (chanerr & IOAT_CHANERR_XOR_P_OR_CRC_ERR) {
522 *desc->result |= SUM_CHECK_P_RESULT;
523 err_handled |= IOAT_CHANERR_XOR_P_OR_CRC_ERR;
524 }
525 if (chanerr & IOAT_CHANERR_XOR_Q_ERR) {
526 *desc->result |= SUM_CHECK_Q_RESULT;
527 err_handled |= IOAT_CHANERR_XOR_Q_ERR;
528 }
529 break;
530 }
531
532 /* fault on unhandled error or spurious halt */
533 if (chanerr ^ err_handled || chanerr == 0) {
534 dev_err(to_dev(chan), "%s: fatal error (%x:%x)\n",
535 __func__, chanerr, err_handled);
536 BUG();
537 }
538
539 writel(chanerr, chan->reg_base + IOAT_CHANERR_OFFSET);
540 pci_write_config_dword(pdev, IOAT_PCI_CHANERR_INT_OFFSET, chanerr_int);
541
542 /* mark faulting descriptor as complete */
543 *chan->completion = desc->txd.phys;
544
545 spin_lock_bh(&ioat->prep_lock);
546 ioat3_restart_channel(ioat);
547 spin_unlock_bh(&ioat->prep_lock);
548 }
549
550 static void check_active(struct ioat2_dma_chan *ioat)
551 {
552 struct ioat_chan_common *chan = &ioat->base;
553
554 if (ioat2_ring_active(ioat)) {
555 mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
556 return;
557 }
558
559 if (test_and_clear_bit(IOAT_CHAN_ACTIVE, &chan->state))
560 mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
561 else if (ioat->alloc_order > ioat_get_alloc_order()) {
562 /* if the ring is idle, empty, and oversized try to step
563 * down the size
564 */
565 reshape_ring(ioat, ioat->alloc_order - 1);
566
567 /* keep shrinking until we get back to our minimum
568 * default size
569 */
570 if (ioat->alloc_order > ioat_get_alloc_order())
571 mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
572 }
573
574 }
575
576 static void ioat3_timer_event(unsigned long data)
577 {
578 struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
579 struct ioat_chan_common *chan = &ioat->base;
580 dma_addr_t phys_complete;
581 u64 status;
582
583 status = ioat_chansts(chan);
584
585 /* when halted due to errors check for channel
586 * programming errors before advancing the completion state
587 */
588 if (is_ioat_halted(status)) {
589 u32 chanerr;
590
591 chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET);
592 dev_err(to_dev(chan), "%s: Channel halted (%x)\n",
593 __func__, chanerr);
594 if (test_bit(IOAT_RUN, &chan->state))
595 BUG_ON(is_ioat_bug(chanerr));
596 else /* we never got off the ground */
597 return;
598 }
599
600 /* if we haven't made progress and we have already
601 * acknowledged a pending completion once, then be more
602 * forceful with a restart
603 */
604 spin_lock_bh(&chan->cleanup_lock);
605 if (ioat_cleanup_preamble(chan, &phys_complete))
606 __cleanup(ioat, phys_complete);
607 else if (test_bit(IOAT_COMPLETION_ACK, &chan->state)) {
608 spin_lock_bh(&ioat->prep_lock);
609 ioat3_restart_channel(ioat);
610 spin_unlock_bh(&ioat->prep_lock);
611 spin_unlock_bh(&chan->cleanup_lock);
612 return;
613 } else {
614 set_bit(IOAT_COMPLETION_ACK, &chan->state);
615 mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
616 }
617
618
619 if (ioat2_ring_active(ioat))
620 mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
621 else {
622 spin_lock_bh(&ioat->prep_lock);
623 check_active(ioat);
624 spin_unlock_bh(&ioat->prep_lock);
625 }
626 spin_unlock_bh(&chan->cleanup_lock);
627 }
628
629 static enum dma_status
630 ioat3_tx_status(struct dma_chan *c, dma_cookie_t cookie,
631 struct dma_tx_state *txstate)
632 {
633 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
634 enum dma_status ret;
635
636 ret = dma_cookie_status(c, cookie, txstate);
637 if (ret == DMA_COMPLETE)
638 return ret;
639
640 ioat3_cleanup(ioat);
641
642 return dma_cookie_status(c, cookie, txstate);
643 }
644
645 static struct dma_async_tx_descriptor *
646 __ioat3_prep_xor_lock(struct dma_chan *c, enum sum_check_flags *result,
647 dma_addr_t dest, dma_addr_t *src, unsigned int src_cnt,
648 size_t len, unsigned long flags)
649 {
650 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
651 struct ioat_ring_ent *compl_desc;
652 struct ioat_ring_ent *desc;
653 struct ioat_ring_ent *ext;
654 size_t total_len = len;
655 struct ioat_xor_descriptor *xor;
656 struct ioat_xor_ext_descriptor *xor_ex = NULL;
657 struct ioat_dma_descriptor *hw;
658 int num_descs, with_ext, idx, i;
659 u32 offset = 0;
660 u8 op = result ? IOAT_OP_XOR_VAL : IOAT_OP_XOR;
661
662 BUG_ON(src_cnt < 2);
663
664 num_descs = ioat2_xferlen_to_descs(ioat, len);
665 /* we need 2x the number of descriptors to cover greater than 5
666 * sources
667 */
668 if (src_cnt > 5) {
669 with_ext = 1;
670 num_descs *= 2;
671 } else
672 with_ext = 0;
673
674 /* completion writes from the raid engine may pass completion
675 * writes from the legacy engine, so we need one extra null
676 * (legacy) descriptor to ensure all completion writes arrive in
677 * order.
678 */
679 if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs+1) == 0)
680 idx = ioat->head;
681 else
682 return NULL;
683 i = 0;
684 do {
685 struct ioat_raw_descriptor *descs[2];
686 size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log);
687 int s;
688
689 desc = ioat2_get_ring_ent(ioat, idx + i);
690 xor = desc->xor;
691
692 /* save a branch by unconditionally retrieving the
693 * extended descriptor xor_set_src() knows to not write
694 * to it in the single descriptor case
695 */
696 ext = ioat2_get_ring_ent(ioat, idx + i + 1);
697 xor_ex = ext->xor_ex;
698
699 descs[0] = (struct ioat_raw_descriptor *) xor;
700 descs[1] = (struct ioat_raw_descriptor *) xor_ex;
701 for (s = 0; s < src_cnt; s++)
702 xor_set_src(descs, src[s], offset, s);
703 xor->size = xfer_size;
704 xor->dst_addr = dest + offset;
705 xor->ctl = 0;
706 xor->ctl_f.op = op;
707 xor->ctl_f.src_cnt = src_cnt_to_hw(src_cnt);
708
709 len -= xfer_size;
710 offset += xfer_size;
711 dump_desc_dbg(ioat, desc);
712 } while ((i += 1 + with_ext) < num_descs);
713
714 /* last xor descriptor carries the unmap parameters and fence bit */
715 desc->txd.flags = flags;
716 desc->len = total_len;
717 if (result)
718 desc->result = result;
719 xor->ctl_f.fence = !!(flags & DMA_PREP_FENCE);
720
721 /* completion descriptor carries interrupt bit */
722 compl_desc = ioat2_get_ring_ent(ioat, idx + i);
723 compl_desc->txd.flags = flags & DMA_PREP_INTERRUPT;
724 hw = compl_desc->hw;
725 hw->ctl = 0;
726 hw->ctl_f.null = 1;
727 hw->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT);
728 hw->ctl_f.compl_write = 1;
729 hw->size = NULL_DESC_BUFFER_SIZE;
730 dump_desc_dbg(ioat, compl_desc);
731
732 /* we leave the channel locked to ensure in order submission */
733 return &compl_desc->txd;
734 }
735
736 static struct dma_async_tx_descriptor *
737 ioat3_prep_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
738 unsigned int src_cnt, size_t len, unsigned long flags)
739 {
740 return __ioat3_prep_xor_lock(chan, NULL, dest, src, src_cnt, len, flags);
741 }
742
743 struct dma_async_tx_descriptor *
744 ioat3_prep_xor_val(struct dma_chan *chan, dma_addr_t *src,
745 unsigned int src_cnt, size_t len,
746 enum sum_check_flags *result, unsigned long flags)
747 {
748 /* the cleanup routine only sets bits on validate failure, it
749 * does not clear bits on validate success... so clear it here
750 */
751 *result = 0;
752
753 return __ioat3_prep_xor_lock(chan, result, src[0], &src[1],
754 src_cnt - 1, len, flags);
755 }
756
757 static void
758 dump_pq_desc_dbg(struct ioat2_dma_chan *ioat, struct ioat_ring_ent *desc, struct ioat_ring_ent *ext)
759 {
760 struct device *dev = to_dev(&ioat->base);
761 struct ioat_pq_descriptor *pq = desc->pq;
762 struct ioat_pq_ext_descriptor *pq_ex = ext ? ext->pq_ex : NULL;
763 struct ioat_raw_descriptor *descs[] = { (void *) pq, (void *) pq_ex };
764 int src_cnt = src_cnt_to_sw(pq->ctl_f.src_cnt);
765 int i;
766
767 dev_dbg(dev, "desc[%d]: (%#llx->%#llx) flags: %#x"
768 " sz: %#10.8x ctl: %#x (op: %#x int: %d compl: %d pq: '%s%s'"
769 " src_cnt: %d)\n",
770 desc_id(desc), (unsigned long long) desc->txd.phys,
771 (unsigned long long) (pq_ex ? pq_ex->next : pq->next),
772 desc->txd.flags, pq->size, pq->ctl, pq->ctl_f.op, pq->ctl_f.int_en,
773 pq->ctl_f.compl_write,
774 pq->ctl_f.p_disable ? "" : "p", pq->ctl_f.q_disable ? "" : "q",
775 pq->ctl_f.src_cnt);
776 for (i = 0; i < src_cnt; i++)
777 dev_dbg(dev, "\tsrc[%d]: %#llx coef: %#x\n", i,
778 (unsigned long long) pq_get_src(descs, i), pq->coef[i]);
779 dev_dbg(dev, "\tP: %#llx\n", pq->p_addr);
780 dev_dbg(dev, "\tQ: %#llx\n", pq->q_addr);
781 dev_dbg(dev, "\tNEXT: %#llx\n", pq->next);
782 }
783
784 static void dump_pq16_desc_dbg(struct ioat2_dma_chan *ioat,
785 struct ioat_ring_ent *desc)
786 {
787 struct device *dev = to_dev(&ioat->base);
788 struct ioat_pq_descriptor *pq = desc->pq;
789 struct ioat_raw_descriptor *descs[] = { (void *)pq,
790 (void *)pq,
791 (void *)pq };
792 int src_cnt = src16_cnt_to_sw(pq->ctl_f.src_cnt);
793 int i;
794
795 if (desc->sed) {
796 descs[1] = (void *)desc->sed->hw;
797 descs[2] = (void *)desc->sed->hw + 64;
798 }
799
800 dev_dbg(dev, "desc[%d]: (%#llx->%#llx) flags: %#x"
801 " sz: %#x ctl: %#x (op: %#x int: %d compl: %d pq: '%s%s'"
802 " src_cnt: %d)\n",
803 desc_id(desc), (unsigned long long) desc->txd.phys,
804 (unsigned long long) pq->next,
805 desc->txd.flags, pq->size, pq->ctl,
806 pq->ctl_f.op, pq->ctl_f.int_en,
807 pq->ctl_f.compl_write,
808 pq->ctl_f.p_disable ? "" : "p", pq->ctl_f.q_disable ? "" : "q",
809 pq->ctl_f.src_cnt);
810 for (i = 0; i < src_cnt; i++) {
811 dev_dbg(dev, "\tsrc[%d]: %#llx coef: %#x\n", i,
812 (unsigned long long) pq16_get_src(descs, i),
813 pq->coef[i]);
814 }
815 dev_dbg(dev, "\tP: %#llx\n", pq->p_addr);
816 dev_dbg(dev, "\tQ: %#llx\n", pq->q_addr);
817 }
818
819 static struct dma_async_tx_descriptor *
820 __ioat3_prep_pq_lock(struct dma_chan *c, enum sum_check_flags *result,
821 const dma_addr_t *dst, const dma_addr_t *src,
822 unsigned int src_cnt, const unsigned char *scf,
823 size_t len, unsigned long flags)
824 {
825 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
826 struct ioat_chan_common *chan = &ioat->base;
827 struct ioatdma_device *device = chan->device;
828 struct ioat_ring_ent *compl_desc;
829 struct ioat_ring_ent *desc;
830 struct ioat_ring_ent *ext;
831 size_t total_len = len;
832 struct ioat_pq_descriptor *pq;
833 struct ioat_pq_ext_descriptor *pq_ex = NULL;
834 struct ioat_dma_descriptor *hw;
835 u32 offset = 0;
836 u8 op = result ? IOAT_OP_PQ_VAL : IOAT_OP_PQ;
837 int i, s, idx, with_ext, num_descs;
838 int cb32 = (device->version < IOAT_VER_3_3) ? 1 : 0;
839
840 dev_dbg(to_dev(chan), "%s\n", __func__);
841 /* the engine requires at least two sources (we provide
842 * at least 1 implied source in the DMA_PREP_CONTINUE case)
843 */
844 BUG_ON(src_cnt + dmaf_continue(flags) < 2);
845
846 num_descs = ioat2_xferlen_to_descs(ioat, len);
847 /* we need 2x the number of descriptors to cover greater than 3
848 * sources (we need 1 extra source in the q-only continuation
849 * case and 3 extra sources in the p+q continuation case.
850 */
851 if (src_cnt + dmaf_p_disabled_continue(flags) > 3 ||
852 (dmaf_continue(flags) && !dmaf_p_disabled_continue(flags))) {
853 with_ext = 1;
854 num_descs *= 2;
855 } else
856 with_ext = 0;
857
858 /* completion writes from the raid engine may pass completion
859 * writes from the legacy engine, so we need one extra null
860 * (legacy) descriptor to ensure all completion writes arrive in
861 * order.
862 */
863 if (likely(num_descs) &&
864 ioat2_check_space_lock(ioat, num_descs + cb32) == 0)
865 idx = ioat->head;
866 else
867 return NULL;
868 i = 0;
869 do {
870 struct ioat_raw_descriptor *descs[2];
871 size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log);
872
873 desc = ioat2_get_ring_ent(ioat, idx + i);
874 pq = desc->pq;
875
876 /* save a branch by unconditionally retrieving the
877 * extended descriptor pq_set_src() knows to not write
878 * to it in the single descriptor case
879 */
880 ext = ioat2_get_ring_ent(ioat, idx + i + with_ext);
881 pq_ex = ext->pq_ex;
882
883 descs[0] = (struct ioat_raw_descriptor *) pq;
884 descs[1] = (struct ioat_raw_descriptor *) pq_ex;
885
886 for (s = 0; s < src_cnt; s++)
887 pq_set_src(descs, src[s], offset, scf[s], s);
888
889 /* see the comment for dma_maxpq in include/linux/dmaengine.h */
890 if (dmaf_p_disabled_continue(flags))
891 pq_set_src(descs, dst[1], offset, 1, s++);
892 else if (dmaf_continue(flags)) {
893 pq_set_src(descs, dst[0], offset, 0, s++);
894 pq_set_src(descs, dst[1], offset, 1, s++);
895 pq_set_src(descs, dst[1], offset, 0, s++);
896 }
897 pq->size = xfer_size;
898 pq->p_addr = dst[0] + offset;
899 pq->q_addr = dst[1] + offset;
900 pq->ctl = 0;
901 pq->ctl_f.op = op;
902 /* we turn on descriptor write back error status */
903 if (device->cap & IOAT_CAP_DWBES)
904 pq->ctl_f.wb_en = result ? 1 : 0;
905 pq->ctl_f.src_cnt = src_cnt_to_hw(s);
906 pq->ctl_f.p_disable = !!(flags & DMA_PREP_PQ_DISABLE_P);
907 pq->ctl_f.q_disable = !!(flags & DMA_PREP_PQ_DISABLE_Q);
908
909 len -= xfer_size;
910 offset += xfer_size;
911 } while ((i += 1 + with_ext) < num_descs);
912
913 /* last pq descriptor carries the unmap parameters and fence bit */
914 desc->txd.flags = flags;
915 desc->len = total_len;
916 if (result)
917 desc->result = result;
918 pq->ctl_f.fence = !!(flags & DMA_PREP_FENCE);
919 dump_pq_desc_dbg(ioat, desc, ext);
920
921 if (!cb32) {
922 pq->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT);
923 pq->ctl_f.compl_write = 1;
924 compl_desc = desc;
925 } else {
926 /* completion descriptor carries interrupt bit */
927 compl_desc = ioat2_get_ring_ent(ioat, idx + i);
928 compl_desc->txd.flags = flags & DMA_PREP_INTERRUPT;
929 hw = compl_desc->hw;
930 hw->ctl = 0;
931 hw->ctl_f.null = 1;
932 hw->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT);
933 hw->ctl_f.compl_write = 1;
934 hw->size = NULL_DESC_BUFFER_SIZE;
935 dump_desc_dbg(ioat, compl_desc);
936 }
937
938
939 /* we leave the channel locked to ensure in order submission */
940 return &compl_desc->txd;
941 }
942
943 static struct dma_async_tx_descriptor *
944 __ioat3_prep_pq16_lock(struct dma_chan *c, enum sum_check_flags *result,
945 const dma_addr_t *dst, const dma_addr_t *src,
946 unsigned int src_cnt, const unsigned char *scf,
947 size_t len, unsigned long flags)
948 {
949 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
950 struct ioat_chan_common *chan = &ioat->base;
951 struct ioatdma_device *device = chan->device;
952 struct ioat_ring_ent *desc;
953 size_t total_len = len;
954 struct ioat_pq_descriptor *pq;
955 u32 offset = 0;
956 u8 op;
957 int i, s, idx, num_descs;
958
959 /* this function is only called with 9-16 sources */
960 op = result ? IOAT_OP_PQ_VAL_16S : IOAT_OP_PQ_16S;
961
962 dev_dbg(to_dev(chan), "%s\n", __func__);
963
964 num_descs = ioat2_xferlen_to_descs(ioat, len);
965
966 /*
967 * 16 source pq is only available on cb3.3 and has no completion
968 * write hw bug.
969 */
970 if (num_descs && ioat2_check_space_lock(ioat, num_descs) == 0)
971 idx = ioat->head;
972 else
973 return NULL;
974
975 i = 0;
976
977 do {
978 struct ioat_raw_descriptor *descs[4];
979 size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log);
980
981 desc = ioat2_get_ring_ent(ioat, idx + i);
982 pq = desc->pq;
983
984 descs[0] = (struct ioat_raw_descriptor *) pq;
985
986 desc->sed = ioat3_alloc_sed(device, (src_cnt-2) >> 3);
987 if (!desc->sed) {
988 dev_err(to_dev(chan),
989 "%s: no free sed entries\n", __func__);
990 return NULL;
991 }
992
993 pq->sed_addr = desc->sed->dma;
994 desc->sed->parent = desc;
995
996 descs[1] = (struct ioat_raw_descriptor *)desc->sed->hw;
997 descs[2] = (void *)descs[1] + 64;
998
999 for (s = 0; s < src_cnt; s++)
1000 pq16_set_src(descs, src[s], offset, scf[s], s);
1001
1002 /* see the comment for dma_maxpq in include/linux/dmaengine.h */
1003 if (dmaf_p_disabled_continue(flags))
1004 pq16_set_src(descs, dst[1], offset, 1, s++);
1005 else if (dmaf_continue(flags)) {
1006 pq16_set_src(descs, dst[0], offset, 0, s++);
1007 pq16_set_src(descs, dst[1], offset, 1, s++);
1008 pq16_set_src(descs, dst[1], offset, 0, s++);
1009 }
1010
1011 pq->size = xfer_size;
1012 pq->p_addr = dst[0] + offset;
1013 pq->q_addr = dst[1] + offset;
1014 pq->ctl = 0;
1015 pq->ctl_f.op = op;
1016 pq->ctl_f.src_cnt = src16_cnt_to_hw(s);
1017 /* we turn on descriptor write back error status */
1018 if (device->cap & IOAT_CAP_DWBES)
1019 pq->ctl_f.wb_en = result ? 1 : 0;
1020 pq->ctl_f.p_disable = !!(flags & DMA_PREP_PQ_DISABLE_P);
1021 pq->ctl_f.q_disable = !!(flags & DMA_PREP_PQ_DISABLE_Q);
1022
1023 len -= xfer_size;
1024 offset += xfer_size;
1025 } while (++i < num_descs);
1026
1027 /* last pq descriptor carries the unmap parameters and fence bit */
1028 desc->txd.flags = flags;
1029 desc->len = total_len;
1030 if (result)
1031 desc->result = result;
1032 pq->ctl_f.fence = !!(flags & DMA_PREP_FENCE);
1033
1034 /* with cb3.3 we should be able to do completion w/o a null desc */
1035 pq->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT);
1036 pq->ctl_f.compl_write = 1;
1037
1038 dump_pq16_desc_dbg(ioat, desc);
1039
1040 /* we leave the channel locked to ensure in order submission */
1041 return &desc->txd;
1042 }
1043
1044 static int src_cnt_flags(unsigned int src_cnt, unsigned long flags)
1045 {
1046 if (dmaf_p_disabled_continue(flags))
1047 return src_cnt + 1;
1048 else if (dmaf_continue(flags))
1049 return src_cnt + 3;
1050 else
1051 return src_cnt;
1052 }
1053
1054 static struct dma_async_tx_descriptor *
1055 ioat3_prep_pq(struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
1056 unsigned int src_cnt, const unsigned char *scf, size_t len,
1057 unsigned long flags)
1058 {
1059 /* specify valid address for disabled result */
1060 if (flags & DMA_PREP_PQ_DISABLE_P)
1061 dst[0] = dst[1];
1062 if (flags & DMA_PREP_PQ_DISABLE_Q)
1063 dst[1] = dst[0];
1064
1065 /* handle the single source multiply case from the raid6
1066 * recovery path
1067 */
1068 if ((flags & DMA_PREP_PQ_DISABLE_P) && src_cnt == 1) {
1069 dma_addr_t single_source[2];
1070 unsigned char single_source_coef[2];
1071
1072 BUG_ON(flags & DMA_PREP_PQ_DISABLE_Q);
1073 single_source[0] = src[0];
1074 single_source[1] = src[0];
1075 single_source_coef[0] = scf[0];
1076 single_source_coef[1] = 0;
1077
1078 return src_cnt_flags(src_cnt, flags) > 8 ?
1079 __ioat3_prep_pq16_lock(chan, NULL, dst, single_source,
1080 2, single_source_coef, len,
1081 flags) :
1082 __ioat3_prep_pq_lock(chan, NULL, dst, single_source, 2,
1083 single_source_coef, len, flags);
1084
1085 } else {
1086 return src_cnt_flags(src_cnt, flags) > 8 ?
1087 __ioat3_prep_pq16_lock(chan, NULL, dst, src, src_cnt,
1088 scf, len, flags) :
1089 __ioat3_prep_pq_lock(chan, NULL, dst, src, src_cnt,
1090 scf, len, flags);
1091 }
1092 }
1093
1094 struct dma_async_tx_descriptor *
1095 ioat3_prep_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
1096 unsigned int src_cnt, const unsigned char *scf, size_t len,
1097 enum sum_check_flags *pqres, unsigned long flags)
1098 {
1099 /* specify valid address for disabled result */
1100 if (flags & DMA_PREP_PQ_DISABLE_P)
1101 pq[0] = pq[1];
1102 if (flags & DMA_PREP_PQ_DISABLE_Q)
1103 pq[1] = pq[0];
1104
1105 /* the cleanup routine only sets bits on validate failure, it
1106 * does not clear bits on validate success... so clear it here
1107 */
1108 *pqres = 0;
1109
1110 return src_cnt_flags(src_cnt, flags) > 8 ?
1111 __ioat3_prep_pq16_lock(chan, pqres, pq, src, src_cnt, scf, len,
1112 flags) :
1113 __ioat3_prep_pq_lock(chan, pqres, pq, src, src_cnt, scf, len,
1114 flags);
1115 }
1116
1117 static struct dma_async_tx_descriptor *
1118 ioat3_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
1119 unsigned int src_cnt, size_t len, unsigned long flags)
1120 {
1121 unsigned char scf[src_cnt];
1122 dma_addr_t pq[2];
1123
1124 memset(scf, 0, src_cnt);
1125 pq[0] = dst;
1126 flags |= DMA_PREP_PQ_DISABLE_Q;
1127 pq[1] = dst; /* specify valid address for disabled result */
1128
1129 return src_cnt_flags(src_cnt, flags) > 8 ?
1130 __ioat3_prep_pq16_lock(chan, NULL, pq, src, src_cnt, scf, len,
1131 flags) :
1132 __ioat3_prep_pq_lock(chan, NULL, pq, src, src_cnt, scf, len,
1133 flags);
1134 }
1135
1136 struct dma_async_tx_descriptor *
1137 ioat3_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
1138 unsigned int src_cnt, size_t len,
1139 enum sum_check_flags *result, unsigned long flags)
1140 {
1141 unsigned char scf[src_cnt];
1142 dma_addr_t pq[2];
1143
1144 /* the cleanup routine only sets bits on validate failure, it
1145 * does not clear bits on validate success... so clear it here
1146 */
1147 *result = 0;
1148
1149 memset(scf, 0, src_cnt);
1150 pq[0] = src[0];
1151 flags |= DMA_PREP_PQ_DISABLE_Q;
1152 pq[1] = pq[0]; /* specify valid address for disabled result */
1153
1154 return src_cnt_flags(src_cnt, flags) > 8 ?
1155 __ioat3_prep_pq16_lock(chan, result, pq, &src[1], src_cnt - 1,
1156 scf, len, flags) :
1157 __ioat3_prep_pq_lock(chan, result, pq, &src[1], src_cnt - 1,
1158 scf, len, flags);
1159 }
1160
1161 static struct dma_async_tx_descriptor *
1162 ioat3_prep_interrupt_lock(struct dma_chan *c, unsigned long flags)
1163 {
1164 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
1165 struct ioat_ring_ent *desc;
1166 struct ioat_dma_descriptor *hw;
1167
1168 if (ioat2_check_space_lock(ioat, 1) == 0)
1169 desc = ioat2_get_ring_ent(ioat, ioat->head);
1170 else
1171 return NULL;
1172
1173 hw = desc->hw;
1174 hw->ctl = 0;
1175 hw->ctl_f.null = 1;
1176 hw->ctl_f.int_en = 1;
1177 hw->ctl_f.fence = !!(flags & DMA_PREP_FENCE);
1178 hw->ctl_f.compl_write = 1;
1179 hw->size = NULL_DESC_BUFFER_SIZE;
1180 hw->src_addr = 0;
1181 hw->dst_addr = 0;
1182
1183 desc->txd.flags = flags;
1184 desc->len = 1;
1185
1186 dump_desc_dbg(ioat, desc);
1187
1188 /* we leave the channel locked to ensure in order submission */
1189 return &desc->txd;
1190 }
1191
1192 static void ioat3_dma_test_callback(void *dma_async_param)
1193 {
1194 struct completion *cmp = dma_async_param;
1195
1196 complete(cmp);
1197 }
1198
1199 #define IOAT_NUM_SRC_TEST 6 /* must be <= 8 */
1200 static int ioat_xor_val_self_test(struct ioatdma_device *device)
1201 {
1202 int i, src_idx;
1203 struct page *dest;
1204 struct page *xor_srcs[IOAT_NUM_SRC_TEST];
1205 struct page *xor_val_srcs[IOAT_NUM_SRC_TEST + 1];
1206 dma_addr_t dma_srcs[IOAT_NUM_SRC_TEST + 1];
1207 dma_addr_t dest_dma;
1208 struct dma_async_tx_descriptor *tx;
1209 struct dma_chan *dma_chan;
1210 dma_cookie_t cookie;
1211 u8 cmp_byte = 0;
1212 u32 cmp_word;
1213 u32 xor_val_result;
1214 int err = 0;
1215 struct completion cmp;
1216 unsigned long tmo;
1217 struct device *dev = &device->pdev->dev;
1218 struct dma_device *dma = &device->common;
1219 u8 op = 0;
1220
1221 dev_dbg(dev, "%s\n", __func__);
1222
1223 if (!dma_has_cap(DMA_XOR, dma->cap_mask))
1224 return 0;
1225
1226 for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) {
1227 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
1228 if (!xor_srcs[src_idx]) {
1229 while (src_idx--)
1230 __free_page(xor_srcs[src_idx]);
1231 return -ENOMEM;
1232 }
1233 }
1234
1235 dest = alloc_page(GFP_KERNEL);
1236 if (!dest) {
1237 while (src_idx--)
1238 __free_page(xor_srcs[src_idx]);
1239 return -ENOMEM;
1240 }
1241
1242 /* Fill in src buffers */
1243 for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) {
1244 u8 *ptr = page_address(xor_srcs[src_idx]);
1245 for (i = 0; i < PAGE_SIZE; i++)
1246 ptr[i] = (1 << src_idx);
1247 }
1248
1249 for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++)
1250 cmp_byte ^= (u8) (1 << src_idx);
1251
1252 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
1253 (cmp_byte << 8) | cmp_byte;
1254
1255 memset(page_address(dest), 0, PAGE_SIZE);
1256
1257 dma_chan = container_of(dma->channels.next, struct dma_chan,
1258 device_node);
1259 if (dma->device_alloc_chan_resources(dma_chan) < 1) {
1260 err = -ENODEV;
1261 goto out;
1262 }
1263
1264 /* test xor */
1265 op = IOAT_OP_XOR;
1266
1267 dest_dma = dma_map_page(dev, dest, 0, PAGE_SIZE, DMA_FROM_DEVICE);
1268 for (i = 0; i < IOAT_NUM_SRC_TEST; i++)
1269 dma_srcs[i] = dma_map_page(dev, xor_srcs[i], 0, PAGE_SIZE,
1270 DMA_TO_DEVICE);
1271 tx = dma->device_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
1272 IOAT_NUM_SRC_TEST, PAGE_SIZE,
1273 DMA_PREP_INTERRUPT);
1274
1275 if (!tx) {
1276 dev_err(dev, "Self-test xor prep failed\n");
1277 err = -ENODEV;
1278 goto dma_unmap;
1279 }
1280
1281 async_tx_ack(tx);
1282 init_completion(&cmp);
1283 tx->callback = ioat3_dma_test_callback;
1284 tx->callback_param = &cmp;
1285 cookie = tx->tx_submit(tx);
1286 if (cookie < 0) {
1287 dev_err(dev, "Self-test xor setup failed\n");
1288 err = -ENODEV;
1289 goto dma_unmap;
1290 }
1291 dma->device_issue_pending(dma_chan);
1292
1293 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
1294
1295 if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_COMPLETE) {
1296 dev_err(dev, "Self-test xor timed out\n");
1297 err = -ENODEV;
1298 goto dma_unmap;
1299 }
1300
1301 dma_unmap_page(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
1302 for (i = 0; i < IOAT_NUM_SRC_TEST; i++)
1303 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE);
1304
1305 dma_sync_single_for_cpu(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
1306 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
1307 u32 *ptr = page_address(dest);
1308 if (ptr[i] != cmp_word) {
1309 dev_err(dev, "Self-test xor failed compare\n");
1310 err = -ENODEV;
1311 goto free_resources;
1312 }
1313 }
1314 dma_sync_single_for_device(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
1315
1316 /* skip validate if the capability is not present */
1317 if (!dma_has_cap(DMA_XOR_VAL, dma_chan->device->cap_mask))
1318 goto free_resources;
1319
1320 op = IOAT_OP_XOR_VAL;
1321
1322 /* validate the sources with the destintation page */
1323 for (i = 0; i < IOAT_NUM_SRC_TEST; i++)
1324 xor_val_srcs[i] = xor_srcs[i];
1325 xor_val_srcs[i] = dest;
1326
1327 xor_val_result = 1;
1328
1329 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++)
1330 dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE,
1331 DMA_TO_DEVICE);
1332 tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs,
1333 IOAT_NUM_SRC_TEST + 1, PAGE_SIZE,
1334 &xor_val_result, DMA_PREP_INTERRUPT);
1335 if (!tx) {
1336 dev_err(dev, "Self-test zero prep failed\n");
1337 err = -ENODEV;
1338 goto dma_unmap;
1339 }
1340
1341 async_tx_ack(tx);
1342 init_completion(&cmp);
1343 tx->callback = ioat3_dma_test_callback;
1344 tx->callback_param = &cmp;
1345 cookie = tx->tx_submit(tx);
1346 if (cookie < 0) {
1347 dev_err(dev, "Self-test zero setup failed\n");
1348 err = -ENODEV;
1349 goto dma_unmap;
1350 }
1351 dma->device_issue_pending(dma_chan);
1352
1353 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
1354
1355 if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_COMPLETE) {
1356 dev_err(dev, "Self-test validate timed out\n");
1357 err = -ENODEV;
1358 goto dma_unmap;
1359 }
1360
1361 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++)
1362 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE);
1363
1364 if (xor_val_result != 0) {
1365 dev_err(dev, "Self-test validate failed compare\n");
1366 err = -ENODEV;
1367 goto free_resources;
1368 }
1369
1370 memset(page_address(dest), 0, PAGE_SIZE);
1371
1372 /* test for non-zero parity sum */
1373 op = IOAT_OP_XOR_VAL;
1374
1375 xor_val_result = 0;
1376 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++)
1377 dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE,
1378 DMA_TO_DEVICE);
1379 tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs,
1380 IOAT_NUM_SRC_TEST + 1, PAGE_SIZE,
1381 &xor_val_result, DMA_PREP_INTERRUPT);
1382 if (!tx) {
1383 dev_err(dev, "Self-test 2nd zero prep failed\n");
1384 err = -ENODEV;
1385 goto dma_unmap;
1386 }
1387
1388 async_tx_ack(tx);
1389 init_completion(&cmp);
1390 tx->callback = ioat3_dma_test_callback;
1391 tx->callback_param = &cmp;
1392 cookie = tx->tx_submit(tx);
1393 if (cookie < 0) {
1394 dev_err(dev, "Self-test 2nd zero setup failed\n");
1395 err = -ENODEV;
1396 goto dma_unmap;
1397 }
1398 dma->device_issue_pending(dma_chan);
1399
1400 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
1401
1402 if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_COMPLETE) {
1403 dev_err(dev, "Self-test 2nd validate timed out\n");
1404 err = -ENODEV;
1405 goto dma_unmap;
1406 }
1407
1408 if (xor_val_result != SUM_CHECK_P_RESULT) {
1409 dev_err(dev, "Self-test validate failed compare\n");
1410 err = -ENODEV;
1411 goto dma_unmap;
1412 }
1413
1414 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++)
1415 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE);
1416
1417 goto free_resources;
1418 dma_unmap:
1419 if (op == IOAT_OP_XOR) {
1420 dma_unmap_page(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
1421 for (i = 0; i < IOAT_NUM_SRC_TEST; i++)
1422 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE,
1423 DMA_TO_DEVICE);
1424 } else if (op == IOAT_OP_XOR_VAL) {
1425 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++)
1426 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE,
1427 DMA_TO_DEVICE);
1428 }
1429 free_resources:
1430 dma->device_free_chan_resources(dma_chan);
1431 out:
1432 src_idx = IOAT_NUM_SRC_TEST;
1433 while (src_idx--)
1434 __free_page(xor_srcs[src_idx]);
1435 __free_page(dest);
1436 return err;
1437 }
1438
1439 static int ioat3_dma_self_test(struct ioatdma_device *device)
1440 {
1441 int rc = ioat_dma_self_test(device);
1442
1443 if (rc)
1444 return rc;
1445
1446 rc = ioat_xor_val_self_test(device);
1447 if (rc)
1448 return rc;
1449
1450 return 0;
1451 }
1452
1453 static int ioat3_irq_reinit(struct ioatdma_device *device)
1454 {
1455 struct pci_dev *pdev = device->pdev;
1456 int irq = pdev->irq, i;
1457
1458 if (!is_bwd_ioat(pdev))
1459 return 0;
1460
1461 switch (device->irq_mode) {
1462 case IOAT_MSIX:
1463 for (i = 0; i < device->common.chancnt; i++) {
1464 struct msix_entry *msix = &device->msix_entries[i];
1465 struct ioat_chan_common *chan;
1466
1467 chan = ioat_chan_by_index(device, i);
1468 devm_free_irq(&pdev->dev, msix->vector, chan);
1469 }
1470
1471 pci_disable_msix(pdev);
1472 break;
1473 case IOAT_MSI:
1474 pci_disable_msi(pdev);
1475 /* fall through */
1476 case IOAT_INTX:
1477 devm_free_irq(&pdev->dev, irq, device);
1478 break;
1479 default:
1480 return 0;
1481 }
1482 device->irq_mode = IOAT_NOIRQ;
1483
1484 return ioat_dma_setup_interrupts(device);
1485 }
1486
1487 static int ioat3_reset_hw(struct ioat_chan_common *chan)
1488 {
1489 /* throw away whatever the channel was doing and get it
1490 * initialized, with ioat3 specific workarounds
1491 */
1492 struct ioatdma_device *device = chan->device;
1493 struct pci_dev *pdev = device->pdev;
1494 u32 chanerr;
1495 u16 dev_id;
1496 int err;
1497
1498 ioat2_quiesce(chan, msecs_to_jiffies(100));
1499
1500 chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET);
1501 writel(chanerr, chan->reg_base + IOAT_CHANERR_OFFSET);
1502
1503 if (device->version < IOAT_VER_3_3) {
1504 /* clear any pending errors */
1505 err = pci_read_config_dword(pdev,
1506 IOAT_PCI_CHANERR_INT_OFFSET, &chanerr);
1507 if (err) {
1508 dev_err(&pdev->dev,
1509 "channel error register unreachable\n");
1510 return err;
1511 }
1512 pci_write_config_dword(pdev,
1513 IOAT_PCI_CHANERR_INT_OFFSET, chanerr);
1514
1515 /* Clear DMAUNCERRSTS Cfg-Reg Parity Error status bit
1516 * (workaround for spurious config parity error after restart)
1517 */
1518 pci_read_config_word(pdev, IOAT_PCI_DEVICE_ID_OFFSET, &dev_id);
1519 if (dev_id == PCI_DEVICE_ID_INTEL_IOAT_TBG0) {
1520 pci_write_config_dword(pdev,
1521 IOAT_PCI_DMAUNCERRSTS_OFFSET,
1522 0x10);
1523 }
1524 }
1525
1526 err = ioat2_reset_sync(chan, msecs_to_jiffies(200));
1527 if (!err)
1528 err = ioat3_irq_reinit(device);
1529
1530 if (err)
1531 dev_err(&pdev->dev, "Failed to reset: %d\n", err);
1532
1533 return err;
1534 }
1535
1536 static void ioat3_intr_quirk(struct ioatdma_device *device)
1537 {
1538 struct dma_device *dma;
1539 struct dma_chan *c;
1540 struct ioat_chan_common *chan;
1541 u32 errmask;
1542
1543 dma = &device->common;
1544
1545 /*
1546 * if we have descriptor write back error status, we mask the
1547 * error interrupts
1548 */
1549 if (device->cap & IOAT_CAP_DWBES) {
1550 list_for_each_entry(c, &dma->channels, device_node) {
1551 chan = to_chan_common(c);
1552 errmask = readl(chan->reg_base +
1553 IOAT_CHANERR_MASK_OFFSET);
1554 errmask |= IOAT_CHANERR_XOR_P_OR_CRC_ERR |
1555 IOAT_CHANERR_XOR_Q_ERR;
1556 writel(errmask, chan->reg_base +
1557 IOAT_CHANERR_MASK_OFFSET);
1558 }
1559 }
1560 }
1561
1562 int ioat3_dma_probe(struct ioatdma_device *device, int dca)
1563 {
1564 struct pci_dev *pdev = device->pdev;
1565 int dca_en = system_has_dca_enabled(pdev);
1566 struct dma_device *dma;
1567 struct dma_chan *c;
1568 struct ioat_chan_common *chan;
1569 bool is_raid_device = false;
1570 int err;
1571
1572 device->enumerate_channels = ioat2_enumerate_channels;
1573 device->reset_hw = ioat3_reset_hw;
1574 device->self_test = ioat3_dma_self_test;
1575 device->intr_quirk = ioat3_intr_quirk;
1576 dma = &device->common;
1577 dma->device_prep_dma_memcpy = ioat2_dma_prep_memcpy_lock;
1578 dma->device_issue_pending = ioat2_issue_pending;
1579 dma->device_alloc_chan_resources = ioat2_alloc_chan_resources;
1580 dma->device_free_chan_resources = ioat2_free_chan_resources;
1581
1582 dma_cap_set(DMA_INTERRUPT, dma->cap_mask);
1583 dma->device_prep_dma_interrupt = ioat3_prep_interrupt_lock;
1584
1585 device->cap = readl(device->reg_base + IOAT_DMA_CAP_OFFSET);
1586
1587 if (is_xeon_cb32(pdev) || is_bwd_noraid(pdev))
1588 device->cap &= ~(IOAT_CAP_XOR | IOAT_CAP_PQ | IOAT_CAP_RAID16SS);
1589
1590 /* dca is incompatible with raid operations */
1591 if (dca_en && (device->cap & (IOAT_CAP_XOR|IOAT_CAP_PQ)))
1592 device->cap &= ~(IOAT_CAP_XOR|IOAT_CAP_PQ);
1593
1594 if (device->cap & IOAT_CAP_XOR) {
1595 is_raid_device = true;
1596 dma->max_xor = 8;
1597
1598 dma_cap_set(DMA_XOR, dma->cap_mask);
1599 dma->device_prep_dma_xor = ioat3_prep_xor;
1600
1601 dma_cap_set(DMA_XOR_VAL, dma->cap_mask);
1602 dma->device_prep_dma_xor_val = ioat3_prep_xor_val;
1603 }
1604
1605 if (device->cap & IOAT_CAP_PQ) {
1606 is_raid_device = true;
1607
1608 dma->device_prep_dma_pq = ioat3_prep_pq;
1609 dma->device_prep_dma_pq_val = ioat3_prep_pq_val;
1610 dma_cap_set(DMA_PQ, dma->cap_mask);
1611 dma_cap_set(DMA_PQ_VAL, dma->cap_mask);
1612
1613 if (device->cap & IOAT_CAP_RAID16SS) {
1614 dma_set_maxpq(dma, 16, 0);
1615 } else {
1616 dma_set_maxpq(dma, 8, 0);
1617 }
1618
1619 if (!(device->cap & IOAT_CAP_XOR)) {
1620 dma->device_prep_dma_xor = ioat3_prep_pqxor;
1621 dma->device_prep_dma_xor_val = ioat3_prep_pqxor_val;
1622 dma_cap_set(DMA_XOR, dma->cap_mask);
1623 dma_cap_set(DMA_XOR_VAL, dma->cap_mask);
1624
1625 if (device->cap & IOAT_CAP_RAID16SS) {
1626 dma->max_xor = 16;
1627 } else {
1628 dma->max_xor = 8;
1629 }
1630 }
1631 }
1632
1633 dma->device_tx_status = ioat3_tx_status;
1634 device->cleanup_fn = ioat3_cleanup_event;
1635 device->timer_fn = ioat3_timer_event;
1636
1637 /* starting with CB3.3 super extended descriptors are supported */
1638 if (device->cap & IOAT_CAP_RAID16SS) {
1639 char pool_name[14];
1640 int i;
1641
1642 for (i = 0; i < MAX_SED_POOLS; i++) {
1643 snprintf(pool_name, 14, "ioat_hw%d_sed", i);
1644
1645 /* allocate SED DMA pool */
1646 device->sed_hw_pool[i] = dmam_pool_create(pool_name,
1647 &pdev->dev,
1648 SED_SIZE * (i + 1), 64, 0);
1649 if (!device->sed_hw_pool[i])
1650 return -ENOMEM;
1651
1652 }
1653 }
1654
1655 err = ioat_probe(device);
1656 if (err)
1657 return err;
1658 ioat_set_tcp_copy_break(262144);
1659
1660 list_for_each_entry(c, &dma->channels, device_node) {
1661 chan = to_chan_common(c);
1662 writel(IOAT_DMA_DCA_ANY_CPU,
1663 chan->reg_base + IOAT_DCACTRL_OFFSET);
1664 }
1665
1666 err = ioat_register(device);
1667 if (err)
1668 return err;
1669
1670 ioat_kobject_add(device, &ioat2_ktype);
1671
1672 if (dca)
1673 device->dca = ioat3_dca_init(pdev, device->reg_base);
1674
1675 return 0;
1676 }
Linux for the MCST Elbrus 2000 architecture