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.
7 * Copyright(c) 2004 - 2009 Intel Corporation. All rights reserved.
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.
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
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.
22 * The full GNU General Public License is included in this distribution in
23 * the file called "COPYING".
27 * Copyright(c) 2004-2009 Intel Corporation. All rights reserved.
29 * Redistribution and use in source and binary forms, with or without
30 * modification, are permitted provided that the following conditions are met:
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
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.
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.
56 * Support routines for v3+ hardware
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"
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)
74 /* provide a lookup table for setting the source address in the base or
75 * extended descriptor of an xor or pq descriptor
77 static const u8 xor_idx_to_desc = 0xe0;
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 };
82 static dma_addr_t xor_get_src(struct ioat_raw_descriptor *descs[2], int idx)
84 struct ioat_raw_descriptor *raw = descs[xor_idx_to_desc >> idx & 1];
86 return raw->field[xor_idx_to_field[idx]];
89 static void xor_set_src(struct ioat_raw_descriptor *descs[2],
90 dma_addr_t addr, u32 offset, int idx)
92 struct ioat_raw_descriptor *raw = descs[xor_idx_to_desc >> idx & 1];
94 raw->field[xor_idx_to_field[idx]] = addr + offset;
97 static dma_addr_t pq_get_src(struct ioat_raw_descriptor *descs[2], int idx)
99 struct ioat_raw_descriptor *raw = descs[pq_idx_to_desc >> idx & 1];
101 return raw->field[pq_idx_to_field[idx]];
104 static void pq_set_src(struct ioat_raw_descriptor *descs[2],
105 dma_addr_t addr, u32 offset, u8 coef, int idx)
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];
110 raw->field[pq_idx_to_field[idx]] = addr + offset;
111 pq->coef[idx] = coef;
114 static void ioat3_dma_unmap(struct ioat2_dma_chan *ioat,
115 struct ioat_ring_ent *desc, int idx)
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;
124 switch (desc->hw->ctl_f.op) {
126 if (!desc->hw->ctl_f.null) /* skip 'interrupt' ops */
127 ioat_dma_unmap(chan, flags, len, desc->hw);
130 struct ioat_fill_descriptor *hw = desc->fill;
132 if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP))
133 ioat_unmap(pdev, hw->dst_addr - offset, len,
134 PCI_DMA_FROMDEVICE, flags, 1);
137 case IOAT_OP_XOR_VAL:
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];
147 ext = ioat2_get_ring_ent(ioat, idx + 1);
148 xor_ex = ext->xor_ex;
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);
157 ioat_unmap(pdev, src - offset, len,
158 PCI_DMA_TODEVICE, flags, 0);
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);
169 if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP))
170 ioat_unmap(pdev, xor->dst_addr - offset, len,
171 PCI_DMA_FROMDEVICE, flags, 1);
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];
184 ext = ioat2_get_ring_ent(ioat, idx + 1);
188 /* in the 'continue' case don't unmap the dests as sources */
189 if (dmaf_p_disabled_continue(flags))
191 else if (dmaf_continue(flags))
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);
200 ioat_unmap(pdev, src - offset, len,
201 PCI_DMA_TODEVICE, flags, 0);
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);
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);
227 dev_err(&pdev->dev, "%s: unknown op type: %#x\n",
228 __func__, desc->hw->ctl_f.op);
232 static bool desc_has_ext(struct ioat_ring_ent *desc)
234 struct ioat_dma_descriptor *hw = desc->hw;
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;
240 if (src_cnt_to_sw(xor->ctl_f.src_cnt) > 5)
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;
246 if (src_cnt_to_sw(pq->ctl_f.src_cnt) > 3)
254 * __cleanup - reclaim used descriptors
255 * @ioat: channel (ring) to clean
257 * The difference from the dma_v2.c __cleanup() is that this routine
258 * handles extended descriptors and dma-unmapping raid operations.
260 static void __cleanup(struct ioat2_dma_chan *ioat, dma_addr_t phys_complete)
262 struct ioat_chan_common *chan = &ioat->base;
263 struct ioat_ring_ent *desc;
264 bool seen_current = false;
265 int idx = ioat->tail, i;
268 dev_dbg(to_dev(chan), "%s: head: %#x tail: %#x issued: %#x\n",
269 __func__, ioat->head, ioat->tail, ioat->issued);
271 active = ioat2_ring_active(ioat);
272 for (i = 0; i < active && !seen_current; i++) {
273 struct dma_async_tx_descriptor *tx;
275 smp_read_barrier_depends();
276 prefetch(ioat2_get_ring_ent(ioat, idx + i + 1));
277 desc = ioat2_get_ring_ent(ioat, idx + i);
278 dump_desc_dbg(ioat, desc);
281 dma_cookie_complete(tx);
282 ioat3_dma_unmap(ioat, desc, idx + i);
284 tx->callback(tx->callback_param);
289 if (tx->phys == phys_complete)
292 /* skip extended descriptors */
293 if (desc_has_ext(desc)) {
294 BUG_ON(i + 1 >= active);
298 smp_mb(); /* finish all descriptor reads before incrementing tail */
299 ioat->tail = idx + i;
300 BUG_ON(active && !seen_current); /* no active descs have written a completion? */
301 chan->last_completion = phys_complete;
303 if (active - i == 0) {
304 dev_dbg(to_dev(chan), "%s: cancel completion timeout\n",
306 clear_bit(IOAT_COMPLETION_PENDING, &chan->state);
307 mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
309 /* 5 microsecond delay per pending descriptor */
310 writew(min((5 * (active - i)), IOAT_INTRDELAY_MASK),
311 chan->device->reg_base + IOAT_INTRDELAY_OFFSET);
314 static void ioat3_cleanup(struct ioat2_dma_chan *ioat)
316 struct ioat_chan_common *chan = &ioat->base;
317 dma_addr_t phys_complete;
319 spin_lock_bh(&chan->cleanup_lock);
320 if (ioat_cleanup_preamble(chan, &phys_complete))
321 __cleanup(ioat, phys_complete);
322 spin_unlock_bh(&chan->cleanup_lock);
325 static void ioat3_cleanup_event(unsigned long data)
327 struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
330 writew(IOAT_CHANCTRL_RUN, ioat->base.reg_base + IOAT_CHANCTRL_OFFSET);
333 static void ioat3_restart_channel(struct ioat2_dma_chan *ioat)
335 struct ioat_chan_common *chan = &ioat->base;
336 dma_addr_t phys_complete;
338 ioat2_quiesce(chan, 0);
339 if (ioat_cleanup_preamble(chan, &phys_complete))
340 __cleanup(ioat, phys_complete);
342 __ioat2_restart_chan(ioat);
345 static void ioat3_timer_event(unsigned long data)
347 struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
348 struct ioat_chan_common *chan = &ioat->base;
350 if (test_bit(IOAT_COMPLETION_PENDING, &chan->state)) {
351 dma_addr_t phys_complete;
354 status = ioat_chansts(chan);
356 /* when halted due to errors check for channel
357 * programming errors before advancing the completion state
359 if (is_ioat_halted(status)) {
362 chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET);
363 dev_err(to_dev(chan), "%s: Channel halted (%x)\n",
365 if (test_bit(IOAT_RUN, &chan->state))
366 BUG_ON(is_ioat_bug(chanerr));
367 else /* we never got off the ground */
371 /* if we haven't made progress and we have already
372 * acknowledged a pending completion once, then be more
373 * forceful with a restart
375 spin_lock_bh(&chan->cleanup_lock);
376 if (ioat_cleanup_preamble(chan, &phys_complete))
377 __cleanup(ioat, phys_complete);
378 else if (test_bit(IOAT_COMPLETION_ACK, &chan->state)) {
379 spin_lock_bh(&ioat->prep_lock);
380 ioat3_restart_channel(ioat);
381 spin_unlock_bh(&ioat->prep_lock);
383 set_bit(IOAT_COMPLETION_ACK, &chan->state);
384 mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
386 spin_unlock_bh(&chan->cleanup_lock);
390 /* if the ring is idle, empty, and oversized try to step
393 spin_lock_bh(&chan->cleanup_lock);
394 spin_lock_bh(&ioat->prep_lock);
395 active = ioat2_ring_active(ioat);
396 if (active == 0 && ioat->alloc_order > ioat_get_alloc_order())
397 reshape_ring(ioat, ioat->alloc_order-1);
398 spin_unlock_bh(&ioat->prep_lock);
399 spin_unlock_bh(&chan->cleanup_lock);
401 /* keep shrinking until we get back to our minimum
404 if (ioat->alloc_order > ioat_get_alloc_order())
405 mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
409 static enum dma_status
410 ioat3_tx_status(struct dma_chan *c, dma_cookie_t cookie,
411 struct dma_tx_state *txstate)
413 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
416 ret = dma_cookie_status(c, cookie, txstate);
417 if (ret == DMA_SUCCESS)
422 return dma_cookie_status(c, cookie, txstate);
425 static struct dma_async_tx_descriptor *
426 ioat3_prep_memset_lock(struct dma_chan *c, dma_addr_t dest, int value,
427 size_t len, unsigned long flags)
429 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
430 struct ioat_ring_ent *desc;
431 size_t total_len = len;
432 struct ioat_fill_descriptor *fill;
433 u64 src_data = (0x0101010101010101ULL) * (value & 0xff);
434 int num_descs, idx, i;
436 num_descs = ioat2_xferlen_to_descs(ioat, len);
437 if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs) == 0)
443 size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log);
445 desc = ioat2_get_ring_ent(ioat, idx + i);
448 fill->size = xfer_size;
449 fill->src_data = src_data;
450 fill->dst_addr = dest;
452 fill->ctl_f.op = IOAT_OP_FILL;
456 dump_desc_dbg(ioat, desc);
457 } while (++i < num_descs);
459 desc->txd.flags = flags;
460 desc->len = total_len;
461 fill->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT);
462 fill->ctl_f.fence = !!(flags & DMA_PREP_FENCE);
463 fill->ctl_f.compl_write = 1;
464 dump_desc_dbg(ioat, desc);
466 /* we leave the channel locked to ensure in order submission */
470 static struct dma_async_tx_descriptor *
471 __ioat3_prep_xor_lock(struct dma_chan *c, enum sum_check_flags *result,
472 dma_addr_t dest, dma_addr_t *src, unsigned int src_cnt,
473 size_t len, unsigned long flags)
475 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
476 struct ioat_ring_ent *compl_desc;
477 struct ioat_ring_ent *desc;
478 struct ioat_ring_ent *ext;
479 size_t total_len = len;
480 struct ioat_xor_descriptor *xor;
481 struct ioat_xor_ext_descriptor *xor_ex = NULL;
482 struct ioat_dma_descriptor *hw;
483 int num_descs, with_ext, idx, i;
485 u8 op = result ? IOAT_OP_XOR_VAL : IOAT_OP_XOR;
489 num_descs = ioat2_xferlen_to_descs(ioat, len);
490 /* we need 2x the number of descriptors to cover greater than 5
499 /* completion writes from the raid engine may pass completion
500 * writes from the legacy engine, so we need one extra null
501 * (legacy) descriptor to ensure all completion writes arrive in
504 if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs+1) == 0)
510 struct ioat_raw_descriptor *descs[2];
511 size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log);
514 desc = ioat2_get_ring_ent(ioat, idx + i);
517 /* save a branch by unconditionally retrieving the
518 * extended descriptor xor_set_src() knows to not write
519 * to it in the single descriptor case
521 ext = ioat2_get_ring_ent(ioat, idx + i + 1);
522 xor_ex = ext->xor_ex;
524 descs[0] = (struct ioat_raw_descriptor *) xor;
525 descs[1] = (struct ioat_raw_descriptor *) xor_ex;
526 for (s = 0; s < src_cnt; s++)
527 xor_set_src(descs, src[s], offset, s);
528 xor->size = xfer_size;
529 xor->dst_addr = dest + offset;
532 xor->ctl_f.src_cnt = src_cnt_to_hw(src_cnt);
536 dump_desc_dbg(ioat, desc);
537 } while ((i += 1 + with_ext) < num_descs);
539 /* last xor descriptor carries the unmap parameters and fence bit */
540 desc->txd.flags = flags;
541 desc->len = total_len;
543 desc->result = result;
544 xor->ctl_f.fence = !!(flags & DMA_PREP_FENCE);
546 /* completion descriptor carries interrupt bit */
547 compl_desc = ioat2_get_ring_ent(ioat, idx + i);
548 compl_desc->txd.flags = flags & DMA_PREP_INTERRUPT;
552 hw->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT);
553 hw->ctl_f.compl_write = 1;
554 hw->size = NULL_DESC_BUFFER_SIZE;
555 dump_desc_dbg(ioat, compl_desc);
557 /* we leave the channel locked to ensure in order submission */
558 return &compl_desc->txd;
561 static struct dma_async_tx_descriptor *
562 ioat3_prep_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
563 unsigned int src_cnt, size_t len, unsigned long flags)
565 return __ioat3_prep_xor_lock(chan, NULL, dest, src, src_cnt, len, flags);
568 struct dma_async_tx_descriptor *
569 ioat3_prep_xor_val(struct dma_chan *chan, dma_addr_t *src,
570 unsigned int src_cnt, size_t len,
571 enum sum_check_flags *result, unsigned long flags)
573 /* the cleanup routine only sets bits on validate failure, it
574 * does not clear bits on validate success... so clear it here
578 return __ioat3_prep_xor_lock(chan, result, src[0], &src[1],
579 src_cnt - 1, len, flags);
583 dump_pq_desc_dbg(struct ioat2_dma_chan *ioat, struct ioat_ring_ent *desc, struct ioat_ring_ent *ext)
585 struct device *dev = to_dev(&ioat->base);
586 struct ioat_pq_descriptor *pq = desc->pq;
587 struct ioat_pq_ext_descriptor *pq_ex = ext ? ext->pq_ex : NULL;
588 struct ioat_raw_descriptor *descs[] = { (void *) pq, (void *) pq_ex };
589 int src_cnt = src_cnt_to_sw(pq->ctl_f.src_cnt);
592 dev_dbg(dev, "desc[%d]: (%#llx->%#llx) flags: %#x"
593 " sz: %#x ctl: %#x (op: %d int: %d compl: %d pq: '%s%s' src_cnt: %d)\n",
594 desc_id(desc), (unsigned long long) desc->txd.phys,
595 (unsigned long long) (pq_ex ? pq_ex->next : pq->next),
596 desc->txd.flags, pq->size, pq->ctl, pq->ctl_f.op, pq->ctl_f.int_en,
597 pq->ctl_f.compl_write,
598 pq->ctl_f.p_disable ? "" : "p", pq->ctl_f.q_disable ? "" : "q",
600 for (i = 0; i < src_cnt; i++)
601 dev_dbg(dev, "\tsrc[%d]: %#llx coef: %#x\n", i,
602 (unsigned long long) pq_get_src(descs, i), pq->coef[i]);
603 dev_dbg(dev, "\tP: %#llx\n", pq->p_addr);
604 dev_dbg(dev, "\tQ: %#llx\n", pq->q_addr);
607 static struct dma_async_tx_descriptor *
608 __ioat3_prep_pq_lock(struct dma_chan *c, enum sum_check_flags *result,
609 const dma_addr_t *dst, const dma_addr_t *src,
610 unsigned int src_cnt, const unsigned char *scf,
611 size_t len, unsigned long flags)
613 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
614 struct ioat_chan_common *chan = &ioat->base;
615 struct ioat_ring_ent *compl_desc;
616 struct ioat_ring_ent *desc;
617 struct ioat_ring_ent *ext;
618 size_t total_len = len;
619 struct ioat_pq_descriptor *pq;
620 struct ioat_pq_ext_descriptor *pq_ex = NULL;
621 struct ioat_dma_descriptor *hw;
623 u8 op = result ? IOAT_OP_PQ_VAL : IOAT_OP_PQ;
624 int i, s, idx, with_ext, num_descs;
626 dev_dbg(to_dev(chan), "%s\n", __func__);
627 /* the engine requires at least two sources (we provide
628 * at least 1 implied source in the DMA_PREP_CONTINUE case)
630 BUG_ON(src_cnt + dmaf_continue(flags) < 2);
632 num_descs = ioat2_xferlen_to_descs(ioat, len);
633 /* we need 2x the number of descriptors to cover greater than 3
634 * sources (we need 1 extra source in the q-only continuation
635 * case and 3 extra sources in the p+q continuation case.
637 if (src_cnt + dmaf_p_disabled_continue(flags) > 3 ||
638 (dmaf_continue(flags) && !dmaf_p_disabled_continue(flags))) {
644 /* completion writes from the raid engine may pass completion
645 * writes from the legacy engine, so we need one extra null
646 * (legacy) descriptor to ensure all completion writes arrive in
649 if (likely(num_descs) &&
650 ioat2_check_space_lock(ioat, num_descs+1) == 0)
656 struct ioat_raw_descriptor *descs[2];
657 size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log);
659 desc = ioat2_get_ring_ent(ioat, idx + i);
662 /* save a branch by unconditionally retrieving the
663 * extended descriptor pq_set_src() knows to not write
664 * to it in the single descriptor case
666 ext = ioat2_get_ring_ent(ioat, idx + i + with_ext);
669 descs[0] = (struct ioat_raw_descriptor *) pq;
670 descs[1] = (struct ioat_raw_descriptor *) pq_ex;
672 for (s = 0; s < src_cnt; s++)
673 pq_set_src(descs, src[s], offset, scf[s], s);
675 /* see the comment for dma_maxpq in include/linux/dmaengine.h */
676 if (dmaf_p_disabled_continue(flags))
677 pq_set_src(descs, dst[1], offset, 1, s++);
678 else if (dmaf_continue(flags)) {
679 pq_set_src(descs, dst[0], offset, 0, s++);
680 pq_set_src(descs, dst[1], offset, 1, s++);
681 pq_set_src(descs, dst[1], offset, 0, s++);
683 pq->size = xfer_size;
684 pq->p_addr = dst[0] + offset;
685 pq->q_addr = dst[1] + offset;
688 pq->ctl_f.src_cnt = src_cnt_to_hw(s);
689 pq->ctl_f.p_disable = !!(flags & DMA_PREP_PQ_DISABLE_P);
690 pq->ctl_f.q_disable = !!(flags & DMA_PREP_PQ_DISABLE_Q);
694 } while ((i += 1 + with_ext) < num_descs);
696 /* last pq descriptor carries the unmap parameters and fence bit */
697 desc->txd.flags = flags;
698 desc->len = total_len;
700 desc->result = result;
701 pq->ctl_f.fence = !!(flags & DMA_PREP_FENCE);
702 dump_pq_desc_dbg(ioat, desc, ext);
704 /* completion descriptor carries interrupt bit */
705 compl_desc = ioat2_get_ring_ent(ioat, idx + i);
706 compl_desc->txd.flags = flags & DMA_PREP_INTERRUPT;
710 hw->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT);
711 hw->ctl_f.compl_write = 1;
712 hw->size = NULL_DESC_BUFFER_SIZE;
713 dump_desc_dbg(ioat, compl_desc);
715 /* we leave the channel locked to ensure in order submission */
716 return &compl_desc->txd;
719 static struct dma_async_tx_descriptor *
720 ioat3_prep_pq(struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
721 unsigned int src_cnt, const unsigned char *scf, size_t len,
724 /* specify valid address for disabled result */
725 if (flags & DMA_PREP_PQ_DISABLE_P)
727 if (flags & DMA_PREP_PQ_DISABLE_Q)
730 /* handle the single source multiply case from the raid6
733 if ((flags & DMA_PREP_PQ_DISABLE_P) && src_cnt == 1) {
734 dma_addr_t single_source[2];
735 unsigned char single_source_coef[2];
737 BUG_ON(flags & DMA_PREP_PQ_DISABLE_Q);
738 single_source[0] = src[0];
739 single_source[1] = src[0];
740 single_source_coef[0] = scf[0];
741 single_source_coef[1] = 0;
743 return __ioat3_prep_pq_lock(chan, NULL, dst, single_source, 2,
744 single_source_coef, len, flags);
746 return __ioat3_prep_pq_lock(chan, NULL, dst, src, src_cnt, scf,
750 struct dma_async_tx_descriptor *
751 ioat3_prep_pq_val(struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
752 unsigned int src_cnt, const unsigned char *scf, size_t len,
753 enum sum_check_flags *pqres, unsigned long flags)
755 /* specify valid address for disabled result */
756 if (flags & DMA_PREP_PQ_DISABLE_P)
758 if (flags & DMA_PREP_PQ_DISABLE_Q)
761 /* the cleanup routine only sets bits on validate failure, it
762 * does not clear bits on validate success... so clear it here
766 return __ioat3_prep_pq_lock(chan, pqres, pq, src, src_cnt, scf, len,
770 static struct dma_async_tx_descriptor *
771 ioat3_prep_pqxor(struct dma_chan *chan, dma_addr_t dst, dma_addr_t *src,
772 unsigned int src_cnt, size_t len, unsigned long flags)
774 unsigned char scf[src_cnt];
777 memset(scf, 0, src_cnt);
779 flags |= DMA_PREP_PQ_DISABLE_Q;
780 pq[1] = dst; /* specify valid address for disabled result */
782 return __ioat3_prep_pq_lock(chan, NULL, pq, src, src_cnt, scf, len,
786 struct dma_async_tx_descriptor *
787 ioat3_prep_pqxor_val(struct dma_chan *chan, dma_addr_t *src,
788 unsigned int src_cnt, size_t len,
789 enum sum_check_flags *result, unsigned long flags)
791 unsigned char scf[src_cnt];
794 /* the cleanup routine only sets bits on validate failure, it
795 * does not clear bits on validate success... so clear it here
799 memset(scf, 0, src_cnt);
801 flags |= DMA_PREP_PQ_DISABLE_Q;
802 pq[1] = pq[0]; /* specify valid address for disabled result */
804 return __ioat3_prep_pq_lock(chan, result, pq, &src[1], src_cnt - 1, scf,
808 static struct dma_async_tx_descriptor *
809 ioat3_prep_interrupt_lock(struct dma_chan *c, unsigned long flags)
811 struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
812 struct ioat_ring_ent *desc;
813 struct ioat_dma_descriptor *hw;
815 if (ioat2_check_space_lock(ioat, 1) == 0)
816 desc = ioat2_get_ring_ent(ioat, ioat->head);
823 hw->ctl_f.int_en = 1;
824 hw->ctl_f.fence = !!(flags & DMA_PREP_FENCE);
825 hw->ctl_f.compl_write = 1;
826 hw->size = NULL_DESC_BUFFER_SIZE;
830 desc->txd.flags = flags;
833 dump_desc_dbg(ioat, desc);
835 /* we leave the channel locked to ensure in order submission */
839 static void ioat3_dma_test_callback(void *dma_async_param)
841 struct completion *cmp = dma_async_param;
846 #define IOAT_NUM_SRC_TEST 6 /* must be <= 8 */
847 static int ioat_xor_val_self_test(struct ioatdma_device *device)
851 struct page *xor_srcs[IOAT_NUM_SRC_TEST];
852 struct page *xor_val_srcs[IOAT_NUM_SRC_TEST + 1];
853 dma_addr_t dma_srcs[IOAT_NUM_SRC_TEST + 1];
854 dma_addr_t dma_addr, dest_dma;
855 struct dma_async_tx_descriptor *tx;
856 struct dma_chan *dma_chan;
862 struct completion cmp;
864 struct device *dev = &device->pdev->dev;
865 struct dma_device *dma = &device->common;
867 dev_dbg(dev, "%s\n", __func__);
869 if (!dma_has_cap(DMA_XOR, dma->cap_mask))
872 for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) {
873 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
874 if (!xor_srcs[src_idx]) {
876 __free_page(xor_srcs[src_idx]);
881 dest = alloc_page(GFP_KERNEL);
884 __free_page(xor_srcs[src_idx]);
888 /* Fill in src buffers */
889 for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) {
890 u8 *ptr = page_address(xor_srcs[src_idx]);
891 for (i = 0; i < PAGE_SIZE; i++)
892 ptr[i] = (1 << src_idx);
895 for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++)
896 cmp_byte ^= (u8) (1 << src_idx);
898 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
899 (cmp_byte << 8) | cmp_byte;
901 memset(page_address(dest), 0, PAGE_SIZE);
903 dma_chan = container_of(dma->channels.next, struct dma_chan,
905 if (dma->device_alloc_chan_resources(dma_chan) < 1) {
911 dest_dma = dma_map_page(dev, dest, 0, PAGE_SIZE, DMA_FROM_DEVICE);
912 for (i = 0; i < IOAT_NUM_SRC_TEST; i++)
913 dma_srcs[i] = dma_map_page(dev, xor_srcs[i], 0, PAGE_SIZE,
915 tx = dma->device_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
916 IOAT_NUM_SRC_TEST, PAGE_SIZE,
920 dev_err(dev, "Self-test xor prep failed\n");
926 init_completion(&cmp);
927 tx->callback = ioat3_dma_test_callback;
928 tx->callback_param = &cmp;
929 cookie = tx->tx_submit(tx);
931 dev_err(dev, "Self-test xor setup failed\n");
935 dma->device_issue_pending(dma_chan);
937 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
939 if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
940 dev_err(dev, "Self-test xor timed out\n");
945 dma_sync_single_for_cpu(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
946 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
947 u32 *ptr = page_address(dest);
948 if (ptr[i] != cmp_word) {
949 dev_err(dev, "Self-test xor failed compare\n");
954 dma_sync_single_for_device(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
956 /* skip validate if the capability is not present */
957 if (!dma_has_cap(DMA_XOR_VAL, dma_chan->device->cap_mask))
960 /* validate the sources with the destintation page */
961 for (i = 0; i < IOAT_NUM_SRC_TEST; i++)
962 xor_val_srcs[i] = xor_srcs[i];
963 xor_val_srcs[i] = dest;
967 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++)
968 dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE,
970 tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs,
971 IOAT_NUM_SRC_TEST + 1, PAGE_SIZE,
972 &xor_val_result, DMA_PREP_INTERRUPT);
974 dev_err(dev, "Self-test zero prep failed\n");
980 init_completion(&cmp);
981 tx->callback = ioat3_dma_test_callback;
982 tx->callback_param = &cmp;
983 cookie = tx->tx_submit(tx);
985 dev_err(dev, "Self-test zero setup failed\n");
989 dma->device_issue_pending(dma_chan);
991 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
993 if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
994 dev_err(dev, "Self-test validate timed out\n");
999 if (xor_val_result != 0) {
1000 dev_err(dev, "Self-test validate failed compare\n");
1002 goto free_resources;
1005 /* skip memset if the capability is not present */
1006 if (!dma_has_cap(DMA_MEMSET, dma_chan->device->cap_mask))
1007 goto free_resources;
1010 dma_addr = dma_map_page(dev, dest, 0,
1011 PAGE_SIZE, DMA_FROM_DEVICE);
1012 tx = dma->device_prep_dma_memset(dma_chan, dma_addr, 0, PAGE_SIZE,
1013 DMA_PREP_INTERRUPT);
1015 dev_err(dev, "Self-test memset prep failed\n");
1017 goto free_resources;
1021 init_completion(&cmp);
1022 tx->callback = ioat3_dma_test_callback;
1023 tx->callback_param = &cmp;
1024 cookie = tx->tx_submit(tx);
1026 dev_err(dev, "Self-test memset setup failed\n");
1028 goto free_resources;
1030 dma->device_issue_pending(dma_chan);
1032 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
1034 if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
1035 dev_err(dev, "Self-test memset timed out\n");
1037 goto free_resources;
1040 for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) {
1041 u32 *ptr = page_address(dest);
1043 dev_err(dev, "Self-test memset failed compare\n");
1045 goto free_resources;
1049 /* test for non-zero parity sum */
1051 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++)
1052 dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE,
1054 tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs,
1055 IOAT_NUM_SRC_TEST + 1, PAGE_SIZE,
1056 &xor_val_result, DMA_PREP_INTERRUPT);
1058 dev_err(dev, "Self-test 2nd zero prep failed\n");
1060 goto free_resources;
1064 init_completion(&cmp);
1065 tx->callback = ioat3_dma_test_callback;
1066 tx->callback_param = &cmp;
1067 cookie = tx->tx_submit(tx);
1069 dev_err(dev, "Self-test 2nd zero setup failed\n");
1071 goto free_resources;
1073 dma->device_issue_pending(dma_chan);
1075 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
1077 if (dma->device_tx_status(dma_chan, cookie, NULL) != DMA_SUCCESS) {
1078 dev_err(dev, "Self-test 2nd validate timed out\n");
1080 goto free_resources;
1083 if (xor_val_result != SUM_CHECK_P_RESULT) {
1084 dev_err(dev, "Self-test validate failed compare\n");
1086 goto free_resources;
1090 dma->device_free_chan_resources(dma_chan);
1092 src_idx = IOAT_NUM_SRC_TEST;
1094 __free_page(xor_srcs[src_idx]);
1099 static int ioat3_dma_self_test(struct ioatdma_device *device)
1101 int rc = ioat_dma_self_test(device);
1106 rc = ioat_xor_val_self_test(device);
1113 static int ioat3_reset_hw(struct ioat_chan_common *chan)
1115 /* throw away whatever the channel was doing and get it
1116 * initialized, with ioat3 specific workarounds
1118 struct ioatdma_device *device = chan->device;
1119 struct pci_dev *pdev = device->pdev;
1124 ioat2_quiesce(chan, msecs_to_jiffies(100));
1126 chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET);
1127 writel(chanerr, chan->reg_base + IOAT_CHANERR_OFFSET);
1129 /* -= IOAT ver.3 workarounds =- */
1130 /* Write CHANERRMSK_INT with 3E07h to mask out the errors
1131 * that can cause stability issues for IOAT ver.3, and clear any
1134 pci_write_config_dword(pdev, IOAT_PCI_CHANERRMASK_INT_OFFSET, 0x3e07);
1135 err = pci_read_config_dword(pdev, IOAT_PCI_CHANERR_INT_OFFSET, &chanerr);
1137 dev_err(&pdev->dev, "channel error register unreachable\n");
1140 pci_write_config_dword(pdev, IOAT_PCI_CHANERR_INT_OFFSET, chanerr);
1142 /* Clear DMAUNCERRSTS Cfg-Reg Parity Error status bit
1143 * (workaround for spurious config parity error after restart)
1145 pci_read_config_word(pdev, IOAT_PCI_DEVICE_ID_OFFSET, &dev_id);
1146 if (dev_id == PCI_DEVICE_ID_INTEL_IOAT_TBG0)
1147 pci_write_config_dword(pdev, IOAT_PCI_DMAUNCERRSTS_OFFSET, 0x10);
1149 return ioat2_reset_sync(chan, msecs_to_jiffies(200));
1152 static bool is_jf_ioat(struct pci_dev *pdev)
1154 switch (pdev->device) {
1155 case PCI_DEVICE_ID_INTEL_IOAT_JSF0:
1156 case PCI_DEVICE_ID_INTEL_IOAT_JSF1:
1157 case PCI_DEVICE_ID_INTEL_IOAT_JSF2:
1158 case PCI_DEVICE_ID_INTEL_IOAT_JSF3:
1159 case PCI_DEVICE_ID_INTEL_IOAT_JSF4:
1160 case PCI_DEVICE_ID_INTEL_IOAT_JSF5:
1161 case PCI_DEVICE_ID_INTEL_IOAT_JSF6:
1162 case PCI_DEVICE_ID_INTEL_IOAT_JSF7:
1163 case PCI_DEVICE_ID_INTEL_IOAT_JSF8:
1164 case PCI_DEVICE_ID_INTEL_IOAT_JSF9:
1171 static bool is_snb_ioat(struct pci_dev *pdev)
1173 switch (pdev->device) {
1174 case PCI_DEVICE_ID_INTEL_IOAT_SNB0:
1175 case PCI_DEVICE_ID_INTEL_IOAT_SNB1:
1176 case PCI_DEVICE_ID_INTEL_IOAT_SNB2:
1177 case PCI_DEVICE_ID_INTEL_IOAT_SNB3:
1178 case PCI_DEVICE_ID_INTEL_IOAT_SNB4:
1179 case PCI_DEVICE_ID_INTEL_IOAT_SNB5:
1180 case PCI_DEVICE_ID_INTEL_IOAT_SNB6:
1181 case PCI_DEVICE_ID_INTEL_IOAT_SNB7:
1182 case PCI_DEVICE_ID_INTEL_IOAT_SNB8:
1183 case PCI_DEVICE_ID_INTEL_IOAT_SNB9:
1190 int ioat3_dma_probe(struct ioatdma_device *device, int dca)
1192 struct pci_dev *pdev = device->pdev;
1193 int dca_en = system_has_dca_enabled(pdev);
1194 struct dma_device *dma;
1196 struct ioat_chan_common *chan;
1197 bool is_raid_device = false;
1201 device->enumerate_channels = ioat2_enumerate_channels;
1202 device->reset_hw = ioat3_reset_hw;
1203 device->self_test = ioat3_dma_self_test;
1204 dma = &device->common;
1205 dma->device_prep_dma_memcpy = ioat2_dma_prep_memcpy_lock;
1206 dma->device_issue_pending = ioat2_issue_pending;
1207 dma->device_alloc_chan_resources = ioat2_alloc_chan_resources;
1208 dma->device_free_chan_resources = ioat2_free_chan_resources;
1210 if (is_jf_ioat(pdev) || is_snb_ioat(pdev))
1211 dma->copy_align = 6;
1213 dma_cap_set(DMA_INTERRUPT, dma->cap_mask);
1214 dma->device_prep_dma_interrupt = ioat3_prep_interrupt_lock;
1216 cap = readl(device->reg_base + IOAT_DMA_CAP_OFFSET);
1218 /* dca is incompatible with raid operations */
1219 if (dca_en && (cap & (IOAT_CAP_XOR|IOAT_CAP_PQ)))
1220 cap &= ~(IOAT_CAP_XOR|IOAT_CAP_PQ);
1222 if (cap & IOAT_CAP_XOR) {
1223 is_raid_device = true;
1227 dma_cap_set(DMA_XOR, dma->cap_mask);
1228 dma->device_prep_dma_xor = ioat3_prep_xor;
1230 dma_cap_set(DMA_XOR_VAL, dma->cap_mask);
1231 dma->device_prep_dma_xor_val = ioat3_prep_xor_val;
1233 if (cap & IOAT_CAP_PQ) {
1234 is_raid_device = true;
1235 dma_set_maxpq(dma, 8, 0);
1238 dma_cap_set(DMA_PQ, dma->cap_mask);
1239 dma->device_prep_dma_pq = ioat3_prep_pq;
1241 dma_cap_set(DMA_PQ_VAL, dma->cap_mask);
1242 dma->device_prep_dma_pq_val = ioat3_prep_pq_val;
1244 if (!(cap & IOAT_CAP_XOR)) {
1248 dma_cap_set(DMA_XOR, dma->cap_mask);
1249 dma->device_prep_dma_xor = ioat3_prep_pqxor;
1251 dma_cap_set(DMA_XOR_VAL, dma->cap_mask);
1252 dma->device_prep_dma_xor_val = ioat3_prep_pqxor_val;
1255 if (is_raid_device && (cap & IOAT_CAP_FILL_BLOCK)) {
1256 dma_cap_set(DMA_MEMSET, dma->cap_mask);
1257 dma->device_prep_dma_memset = ioat3_prep_memset_lock;
1261 if (is_raid_device) {
1262 dma->device_tx_status = ioat3_tx_status;
1263 device->cleanup_fn = ioat3_cleanup_event;
1264 device->timer_fn = ioat3_timer_event;
1266 dma->device_tx_status = ioat_dma_tx_status;
1267 device->cleanup_fn = ioat2_cleanup_event;
1268 device->timer_fn = ioat2_timer_event;
1271 #ifdef CONFIG_ASYNC_TX_DISABLE_PQ_VAL_DMA
1272 dma_cap_clear(DMA_PQ_VAL, dma->cap_mask);
1273 dma->device_prep_dma_pq_val = NULL;
1276 #ifdef CONFIG_ASYNC_TX_DISABLE_XOR_VAL_DMA
1277 dma_cap_clear(DMA_XOR_VAL, dma->cap_mask);
1278 dma->device_prep_dma_xor_val = NULL;
1281 err = ioat_probe(device);
1284 ioat_set_tcp_copy_break(262144);
1286 list_for_each_entry(c, &dma->channels, device_node) {
1287 chan = to_chan_common(c);
1288 writel(IOAT_DMA_DCA_ANY_CPU,
1289 chan->reg_base + IOAT_DCACTRL_OFFSET);
1292 err = ioat_register(device);
1296 ioat_kobject_add(device, &ioat2_ktype);
1299 device->dca = ioat3_dca_init(pdev, device->reg_base);