1 /*
2 * Copyright (C) 2011
3 * Boaz Harrosh <ooo@electrozaur.com>
4 *
5 * This file is part of the objects raid engine (ore).
6 *
7 * It is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as published
9 * by the Free Software Foundation.
10 *
11 * You should have received a copy of the GNU General Public License
12 * along with "ore". If not, write to the Free Software Foundation, Inc:
13 * "Free Software Foundation <info@fsf.org>"
14 */
15
16 #include <linux/gfp.h>
17 #include <linux/async_tx.h>
18
19 #include "ore_raid.h"
20
21 #undef ORE_DBGMSG2
22 #define ORE_DBGMSG2 ORE_DBGMSG
23
_raid_page_alloc(void)24 static struct page *_raid_page_alloc(void)
25 {
26 return alloc_page(GFP_KERNEL);
27 }
28
_raid_page_free(struct page * p)29 static void _raid_page_free(struct page *p)
30 {
31 __free_page(p);
32 }
33
34 /* This struct is forward declare in ore_io_state, but is private to here.
35 * It is put on ios->sp2d for RAID5/6 writes only. See _gen_xor_unit.
36 *
37 * __stripe_pages_2d is a 2d array of pages, and it is also a corner turn.
38 * Ascending page index access is sp2d(p-minor, c-major). But storage is
39 * sp2d[p-minor][c-major], so it can be properlly presented to the async-xor
40 * API.
41 */
42 struct __stripe_pages_2d {
43 /* Cache some hot path repeated calculations */
44 unsigned parity;
45 unsigned data_devs;
46 unsigned pages_in_unit;
47
48 bool needed ;
49
50 /* Array size is pages_in_unit (layout->stripe_unit / PAGE_SIZE) */
51 struct __1_page_stripe {
52 bool alloc;
53 unsigned write_count;
54 struct async_submit_ctl submit;
55 struct dma_async_tx_descriptor *tx;
56
57 /* The size of this array is data_devs + parity */
58 struct page **pages;
59 struct page **scribble;
60 /* bool array, size of this array is data_devs */
61 char *page_is_read;
62 } _1p_stripes[];
63 };
64
65 /* This can get bigger then a page. So support multiple page allocations
66 * _sp2d_free should be called even if _sp2d_alloc fails (by returning
67 * none-zero).
68 */
_sp2d_alloc(unsigned pages_in_unit,unsigned group_width,unsigned parity,struct __stripe_pages_2d ** psp2d)69 static int _sp2d_alloc(unsigned pages_in_unit, unsigned group_width,
70 unsigned parity, struct __stripe_pages_2d **psp2d)
71 {
72 struct __stripe_pages_2d *sp2d;
73 unsigned data_devs = group_width - parity;
74 struct _alloc_all_bytes {
75 struct __alloc_stripe_pages_2d {
76 struct __stripe_pages_2d sp2d;
77 struct __1_page_stripe _1p_stripes[pages_in_unit];
78 } __asp2d;
79 struct __alloc_1p_arrays {
80 struct page *pages[group_width];
81 struct page *scribble[group_width];
82 char page_is_read[data_devs];
83 } __a1pa[pages_in_unit];
84 } *_aab;
85 struct __alloc_1p_arrays *__a1pa;
86 struct __alloc_1p_arrays *__a1pa_end;
87 const unsigned sizeof__a1pa = sizeof(_aab->__a1pa[0]);
88 unsigned num_a1pa, alloc_size, i;
89
90 /* FIXME: check these numbers in ore_verify_layout */
91 BUG_ON(sizeof(_aab->__asp2d) > PAGE_SIZE);
92 BUG_ON(sizeof__a1pa > PAGE_SIZE);
93
94 if (sizeof(*_aab) > PAGE_SIZE) {
95 num_a1pa = (PAGE_SIZE - sizeof(_aab->__asp2d)) / sizeof__a1pa;
96 alloc_size = sizeof(_aab->__asp2d) + sizeof__a1pa * num_a1pa;
97 } else {
98 num_a1pa = pages_in_unit;
99 alloc_size = sizeof(*_aab);
100 }
101
102 _aab = kzalloc(alloc_size, GFP_KERNEL);
103 if (unlikely(!_aab)) {
104 ORE_DBGMSG("!! Failed to alloc sp2d size=%d\n", alloc_size);
105 return -ENOMEM;
106 }
107
108 sp2d = &_aab->__asp2d.sp2d;
109 *psp2d = sp2d; /* From here Just call _sp2d_free */
110
111 __a1pa = _aab->__a1pa;
112 __a1pa_end = __a1pa + num_a1pa;
113
114 for (i = 0; i < pages_in_unit; ++i) {
115 if (unlikely(__a1pa >= __a1pa_end)) {
116 num_a1pa = min_t(unsigned, PAGE_SIZE / sizeof__a1pa,
117 pages_in_unit - i);
118
119 __a1pa = kcalloc(num_a1pa, sizeof__a1pa, GFP_KERNEL);
120 if (unlikely(!__a1pa)) {
121 ORE_DBGMSG("!! Failed to _alloc_1p_arrays=%d\n",
122 num_a1pa);
123 return -ENOMEM;
124 }
125 __a1pa_end = __a1pa + num_a1pa;
126 /* First *pages is marked for kfree of the buffer */
127 sp2d->_1p_stripes[i].alloc = true;
128 }
129
130 sp2d->_1p_stripes[i].pages = __a1pa->pages;
131 sp2d->_1p_stripes[i].scribble = __a1pa->scribble ;
132 sp2d->_1p_stripes[i].page_is_read = __a1pa->page_is_read;
133 ++__a1pa;
134 }
135
136 sp2d->parity = parity;
137 sp2d->data_devs = data_devs;
138 sp2d->pages_in_unit = pages_in_unit;
139 return 0;
140 }
141
_sp2d_reset(struct __stripe_pages_2d * sp2d,const struct _ore_r4w_op * r4w,void * priv)142 static void _sp2d_reset(struct __stripe_pages_2d *sp2d,
143 const struct _ore_r4w_op *r4w, void *priv)
144 {
145 unsigned data_devs = sp2d->data_devs;
146 unsigned group_width = data_devs + sp2d->parity;
147 int p, c;
148
149 if (!sp2d->needed)
150 return;
151
152 for (c = data_devs - 1; c >= 0; --c)
153 for (p = sp2d->pages_in_unit - 1; p >= 0; --p) {
154 struct __1_page_stripe *_1ps = &sp2d->_1p_stripes[p];
155
156 if (_1ps->page_is_read[c]) {
157 struct page *page = _1ps->pages[c];
158
159 r4w->put_page(priv, page);
160 _1ps->page_is_read[c] = false;
161 }
162 }
163
164 for (p = 0; p < sp2d->pages_in_unit; p++) {
165 struct __1_page_stripe *_1ps = &sp2d->_1p_stripes[p];
166
167 memset(_1ps->pages, 0, group_width * sizeof(*_1ps->pages));
168 _1ps->write_count = 0;
169 _1ps->tx = NULL;
170 }
171
172 sp2d->needed = false;
173 }
174
_sp2d_free(struct __stripe_pages_2d * sp2d)175 static void _sp2d_free(struct __stripe_pages_2d *sp2d)
176 {
177 unsigned i;
178
179 if (!sp2d)
180 return;
181
182 for (i = 0; i < sp2d->pages_in_unit; ++i) {
183 if (sp2d->_1p_stripes[i].alloc)
184 kfree(sp2d->_1p_stripes[i].pages);
185 }
186
187 kfree(sp2d);
188 }
189
_sp2d_min_pg(struct __stripe_pages_2d * sp2d)190 static unsigned _sp2d_min_pg(struct __stripe_pages_2d *sp2d)
191 {
192 unsigned p;
193
194 for (p = 0; p < sp2d->pages_in_unit; p++) {
195 struct __1_page_stripe *_1ps = &sp2d->_1p_stripes[p];
196
197 if (_1ps->write_count)
198 return p;
199 }
200
201 return ~0;
202 }
203
_sp2d_max_pg(struct __stripe_pages_2d * sp2d)204 static unsigned _sp2d_max_pg(struct __stripe_pages_2d *sp2d)
205 {
206 int p;
207
208 for (p = sp2d->pages_in_unit - 1; p >= 0; --p) {
209 struct __1_page_stripe *_1ps = &sp2d->_1p_stripes[p];
210
211 if (_1ps->write_count)
212 return p;
213 }
214
215 return ~0;
216 }
217
_gen_xor_unit(struct __stripe_pages_2d * sp2d)218 static void _gen_xor_unit(struct __stripe_pages_2d *sp2d)
219 {
220 unsigned p;
221 unsigned tx_flags = ASYNC_TX_ACK;
222
223 if (sp2d->parity == 1)
224 tx_flags |= ASYNC_TX_XOR_ZERO_DST;
225
226 for (p = 0; p < sp2d->pages_in_unit; p++) {
227 struct __1_page_stripe *_1ps = &sp2d->_1p_stripes[p];
228
229 if (!_1ps->write_count)
230 continue;
231
232 init_async_submit(&_1ps->submit, tx_flags,
233 NULL, NULL, NULL, (addr_conv_t *)_1ps->scribble);
234
235 if (sp2d->parity == 1)
236 _1ps->tx = async_xor(_1ps->pages[sp2d->data_devs],
237 _1ps->pages, 0, sp2d->data_devs,
238 PAGE_SIZE, &_1ps->submit);
239 else /* parity == 2 */
240 _1ps->tx = async_gen_syndrome(_1ps->pages, 0,
241 sp2d->data_devs + sp2d->parity,
242 PAGE_SIZE, &_1ps->submit);
243 }
244
245 for (p = 0; p < sp2d->pages_in_unit; p++) {
246 struct __1_page_stripe *_1ps = &sp2d->_1p_stripes[p];
247 /* NOTE: We wait for HW synchronously (I don't have such HW
248 * to test with.) Is parallelism needed with today's multi
249 * cores?
250 */
251 async_tx_issue_pending(_1ps->tx);
252 }
253 }
254
_ore_add_stripe_page(struct __stripe_pages_2d * sp2d,struct ore_striping_info * si,struct page * page)255 void _ore_add_stripe_page(struct __stripe_pages_2d *sp2d,
256 struct ore_striping_info *si, struct page *page)
257 {
258 struct __1_page_stripe *_1ps;
259
260 sp2d->needed = true;
261
262 _1ps = &sp2d->_1p_stripes[si->cur_pg];
263 _1ps->pages[si->cur_comp] = page;
264 ++_1ps->write_count;
265
266 si->cur_pg = (si->cur_pg + 1) % sp2d->pages_in_unit;
267 /* si->cur_comp is advanced outside at main loop */
268 }
269
_ore_add_sg_seg(struct ore_per_dev_state * per_dev,unsigned cur_len,bool not_last)270 void _ore_add_sg_seg(struct ore_per_dev_state *per_dev, unsigned cur_len,
271 bool not_last)
272 {
273 struct osd_sg_entry *sge;
274
275 ORE_DBGMSG("dev=%d cur_len=0x%x not_last=%d cur_sg=%d "
276 "offset=0x%llx length=0x%x last_sgs_total=0x%x\n",
277 per_dev->dev, cur_len, not_last, per_dev->cur_sg,
278 _LLU(per_dev->offset), per_dev->length,
279 per_dev->last_sgs_total);
280
281 if (!per_dev->cur_sg) {
282 sge = per_dev->sglist;
283
284 /* First time we prepare two entries */
285 if (per_dev->length) {
286 ++per_dev->cur_sg;
287 sge->offset = per_dev->offset;
288 sge->len = per_dev->length;
289 } else {
290 /* Here the parity is the first unit of this object.
291 * This happens every time we reach a parity device on
292 * the same stripe as the per_dev->offset. We need to
293 * just skip this unit.
294 */
295 per_dev->offset += cur_len;
296 return;
297 }
298 } else {
299 /* finalize the last one */
300 sge = &per_dev->sglist[per_dev->cur_sg - 1];
301 sge->len = per_dev->length - per_dev->last_sgs_total;
302 }
303
304 if (not_last) {
305 /* Partly prepare the next one */
306 struct osd_sg_entry *next_sge = sge + 1;
307
308 ++per_dev->cur_sg;
309 next_sge->offset = sge->offset + sge->len + cur_len;
310 /* Save cur len so we know how mutch was added next time */
311 per_dev->last_sgs_total = per_dev->length;
312 next_sge->len = 0;
313 } else if (!sge->len) {
314 /* Optimize for when the last unit is a parity */
315 --per_dev->cur_sg;
316 }
317 }
318
_alloc_read_4_write(struct ore_io_state * ios)319 static int _alloc_read_4_write(struct ore_io_state *ios)
320 {
321 struct ore_layout *layout = ios->layout;
322 int ret;
323 /* We want to only read those pages not in cache so worst case
324 * is a stripe populated with every other page
325 */
326 unsigned sgs_per_dev = ios->sp2d->pages_in_unit + 2;
327
328 ret = _ore_get_io_state(layout, ios->oc,
329 layout->group_width * layout->mirrors_p1,
330 sgs_per_dev, 0, &ios->ios_read_4_write);
331 return ret;
332 }
333
334 /* @si contains info of the to-be-inserted page. Update of @si should be
335 * maintained by caller. Specificaly si->dev, si->obj_offset, ...
336 */
_add_to_r4w(struct ore_io_state * ios,struct ore_striping_info * si,struct page * page,unsigned pg_len)337 static int _add_to_r4w(struct ore_io_state *ios, struct ore_striping_info *si,
338 struct page *page, unsigned pg_len)
339 {
340 struct request_queue *q;
341 struct ore_per_dev_state *per_dev;
342 struct ore_io_state *read_ios;
343 unsigned first_dev = si->dev - (si->dev %
344 (ios->layout->group_width * ios->layout->mirrors_p1));
345 unsigned comp = si->dev - first_dev;
346 unsigned added_len;
347
348 if (!ios->ios_read_4_write) {
349 int ret = _alloc_read_4_write(ios);
350
351 if (unlikely(ret))
352 return ret;
353 }
354
355 read_ios = ios->ios_read_4_write;
356 read_ios->numdevs = ios->layout->group_width * ios->layout->mirrors_p1;
357
358 per_dev = &read_ios->per_dev[comp];
359 if (!per_dev->length) {
360 per_dev->bio = bio_kmalloc(GFP_KERNEL,
361 ios->sp2d->pages_in_unit);
362 if (unlikely(!per_dev->bio)) {
363 ORE_DBGMSG("Failed to allocate BIO size=%u\n",
364 ios->sp2d->pages_in_unit);
365 return -ENOMEM;
366 }
367 per_dev->offset = si->obj_offset;
368 per_dev->dev = si->dev;
369 } else if (si->obj_offset != (per_dev->offset + per_dev->length)) {
370 u64 gap = si->obj_offset - (per_dev->offset + per_dev->length);
371
372 _ore_add_sg_seg(per_dev, gap, true);
373 }
374 q = osd_request_queue(ore_comp_dev(read_ios->oc, per_dev->dev));
375 added_len = bio_add_pc_page(q, per_dev->bio, page, pg_len,
376 si->obj_offset % PAGE_SIZE);
377 if (unlikely(added_len != pg_len)) {
378 ORE_DBGMSG("Failed to bio_add_pc_page bi_vcnt=%d\n",
379 per_dev->bio->bi_vcnt);
380 return -ENOMEM;
381 }
382
383 per_dev->length += pg_len;
384 return 0;
385 }
386
387 /* read the beginning of an unaligned first page */
_add_to_r4w_first_page(struct ore_io_state * ios,struct page * page)388 static int _add_to_r4w_first_page(struct ore_io_state *ios, struct page *page)
389 {
390 struct ore_striping_info si;
391 unsigned pg_len;
392
393 ore_calc_stripe_info(ios->layout, ios->offset, 0, &si);
394
395 pg_len = si.obj_offset % PAGE_SIZE;
396 si.obj_offset -= pg_len;
397
398 ORE_DBGMSG("offset=0x%llx len=0x%x index=0x%lx dev=%x\n",
399 _LLU(si.obj_offset), pg_len, page->index, si.dev);
400
401 return _add_to_r4w(ios, &si, page, pg_len);
402 }
403
404 /* read the end of an incomplete last page */
_add_to_r4w_last_page(struct ore_io_state * ios,u64 * offset)405 static int _add_to_r4w_last_page(struct ore_io_state *ios, u64 *offset)
406 {
407 struct ore_striping_info si;
408 struct page *page;
409 unsigned pg_len, p, c;
410
411 ore_calc_stripe_info(ios->layout, *offset, 0, &si);
412
413 p = si.cur_pg;
414 c = si.cur_comp;
415 page = ios->sp2d->_1p_stripes[p].pages[c];
416
417 pg_len = PAGE_SIZE - (si.unit_off % PAGE_SIZE);
418 *offset += pg_len;
419
420 ORE_DBGMSG("p=%d, c=%d next-offset=0x%llx len=0x%x dev=%x par_dev=%d\n",
421 p, c, _LLU(*offset), pg_len, si.dev, si.par_dev);
422
423 BUG_ON(!page);
424
425 return _add_to_r4w(ios, &si, page, pg_len);
426 }
427
_mark_read4write_pages_uptodate(struct ore_io_state * ios,int ret)428 static void _mark_read4write_pages_uptodate(struct ore_io_state *ios, int ret)
429 {
430 struct bio_vec *bv;
431 unsigned i, d;
432
433 /* loop on all devices all pages */
434 for (d = 0; d < ios->numdevs; d++) {
435 struct bio *bio = ios->per_dev[d].bio;
436
437 if (!bio)
438 continue;
439
440 bio_for_each_segment_all(bv, bio, i) {
441 struct page *page = bv->bv_page;
442
443 SetPageUptodate(page);
444 if (PageError(page))
445 ClearPageError(page);
446 }
447 }
448 }
449
450 /* read_4_write is hacked to read the start of the first stripe and/or
451 * the end of the last stripe. If needed, with an sg-gap at each device/page.
452 * It is assumed to be called after the to_be_written pages of the first stripe
453 * are populating ios->sp2d[][]
454 *
455 * NOTE: We call ios->r4w->lock_fn for all pages needed for parity calculations
456 * These pages are held at sp2d[p].pages[c] but with
457 * sp2d[p].page_is_read[c] = true. At _sp2d_reset these pages are
458 * ios->r4w->lock_fn(). The ios->r4w->lock_fn might signal that the page is
459 * @uptodate=true, so we don't need to read it, only unlock, after IO.
460 *
461 * TODO: The read_4_write should calc a need_to_read_pages_count, if bigger then
462 * to-be-written count, we should consider the xor-in-place mode.
463 * need_to_read_pages_count is the actual number of pages not present in cache.
464 * maybe "devs_in_group - ios->sp2d[p].write_count" is a good enough
465 * approximation? In this mode the read pages are put in the empty places of
466 * ios->sp2d[p][*], xor is calculated the same way. These pages are
467 * allocated/freed and don't go through cache
468 */
_read_4_write_first_stripe(struct ore_io_state * ios)469 static int _read_4_write_first_stripe(struct ore_io_state *ios)
470 {
471 struct ore_striping_info read_si;
472 struct __stripe_pages_2d *sp2d = ios->sp2d;
473 u64 offset = ios->si.first_stripe_start;
474 unsigned c, p, min_p = sp2d->pages_in_unit, max_p = -1;
475
476 if (offset == ios->offset) /* Go to start collect $200 */
477 goto read_last_stripe;
478
479 min_p = _sp2d_min_pg(sp2d);
480 max_p = _sp2d_max_pg(sp2d);
481
482 ORE_DBGMSG("stripe_start=0x%llx ios->offset=0x%llx min_p=%d max_p=%d\n",
483 offset, ios->offset, min_p, max_p);
484
485 for (c = 0; ; c++) {
486 ore_calc_stripe_info(ios->layout, offset, 0, &read_si);
487 read_si.obj_offset += min_p * PAGE_SIZE;
488 offset += min_p * PAGE_SIZE;
489 for (p = min_p; p <= max_p; p++) {
490 struct __1_page_stripe *_1ps = &sp2d->_1p_stripes[p];
491 struct page **pp = &_1ps->pages[c];
492 bool uptodate;
493
494 if (*pp) {
495 if (ios->offset % PAGE_SIZE)
496 /* Read the remainder of the page */
497 _add_to_r4w_first_page(ios, *pp);
498 /* to-be-written pages start here */
499 goto read_last_stripe;
500 }
501
502 *pp = ios->r4w->get_page(ios->private, offset,
503 &uptodate);
504 if (unlikely(!*pp))
505 return -ENOMEM;
506
507 if (!uptodate)
508 _add_to_r4w(ios, &read_si, *pp, PAGE_SIZE);
509
510 /* Mark read-pages to be cache_released */
511 _1ps->page_is_read[c] = true;
512 read_si.obj_offset += PAGE_SIZE;
513 offset += PAGE_SIZE;
514 }
515 offset += (sp2d->pages_in_unit - p) * PAGE_SIZE;
516 }
517
518 read_last_stripe:
519 return 0;
520 }
521
_read_4_write_last_stripe(struct ore_io_state * ios)522 static int _read_4_write_last_stripe(struct ore_io_state *ios)
523 {
524 struct ore_striping_info read_si;
525 struct __stripe_pages_2d *sp2d = ios->sp2d;
526 u64 offset;
527 u64 last_stripe_end;
528 unsigned bytes_in_stripe = ios->si.bytes_in_stripe;
529 unsigned c, p, min_p = sp2d->pages_in_unit, max_p = -1;
530
531 offset = ios->offset + ios->length;
532 if (offset % PAGE_SIZE)
533 _add_to_r4w_last_page(ios, &offset);
534 /* offset will be aligned to next page */
535
536 last_stripe_end = div_u64(offset + bytes_in_stripe - 1, bytes_in_stripe)
537 * bytes_in_stripe;
538 if (offset == last_stripe_end) /* Optimize for the aligned case */
539 goto read_it;
540
541 ore_calc_stripe_info(ios->layout, offset, 0, &read_si);
542 p = read_si.cur_pg;
543 c = read_si.cur_comp;
544
545 if (min_p == sp2d->pages_in_unit) {
546 /* Didn't do it yet */
547 min_p = _sp2d_min_pg(sp2d);
548 max_p = _sp2d_max_pg(sp2d);
549 }
550
551 ORE_DBGMSG("offset=0x%llx stripe_end=0x%llx min_p=%d max_p=%d\n",
552 offset, last_stripe_end, min_p, max_p);
553
554 while (offset < last_stripe_end) {
555 struct __1_page_stripe *_1ps = &sp2d->_1p_stripes[p];
556
557 if ((min_p <= p) && (p <= max_p)) {
558 struct page *page;
559 bool uptodate;
560
561 BUG_ON(_1ps->pages[c]);
562 page = ios->r4w->get_page(ios->private, offset,
563 &uptodate);
564 if (unlikely(!page))
565 return -ENOMEM;
566
567 _1ps->pages[c] = page;
568 /* Mark read-pages to be cache_released */
569 _1ps->page_is_read[c] = true;
570 if (!uptodate)
571 _add_to_r4w(ios, &read_si, page, PAGE_SIZE);
572 }
573
574 offset += PAGE_SIZE;
575 if (p == (sp2d->pages_in_unit - 1)) {
576 ++c;
577 p = 0;
578 ore_calc_stripe_info(ios->layout, offset, 0, &read_si);
579 } else {
580 read_si.obj_offset += PAGE_SIZE;
581 ++p;
582 }
583 }
584
585 read_it:
586 return 0;
587 }
588
_read_4_write_execute(struct ore_io_state * ios)589 static int _read_4_write_execute(struct ore_io_state *ios)
590 {
591 struct ore_io_state *ios_read;
592 unsigned i;
593 int ret;
594
595 ios_read = ios->ios_read_4_write;
596 if (!ios_read)
597 return 0;
598
599 /* FIXME: Ugly to signal _sbi_read_mirror that we have bio(s). Change
600 * to check for per_dev->bio
601 */
602 ios_read->pages = ios->pages;
603
604 /* Now read these devices */
605 for (i = 0; i < ios_read->numdevs; i += ios_read->layout->mirrors_p1) {
606 ret = _ore_read_mirror(ios_read, i);
607 if (unlikely(ret))
608 return ret;
609 }
610
611 ret = ore_io_execute(ios_read); /* Synchronus execution */
612 if (unlikely(ret)) {
613 ORE_DBGMSG("!! ore_io_execute => %d\n", ret);
614 return ret;
615 }
616
617 _mark_read4write_pages_uptodate(ios_read, ret);
618 ore_put_io_state(ios_read);
619 ios->ios_read_4_write = NULL; /* Might need a reuse at last stripe */
620 return 0;
621 }
622
623 /* In writes @cur_len means length left. .i.e cur_len==0 is the last parity U */
_ore_add_parity_unit(struct ore_io_state * ios,struct ore_striping_info * si,struct ore_per_dev_state * per_dev,unsigned cur_len,bool do_xor)624 int _ore_add_parity_unit(struct ore_io_state *ios,
625 struct ore_striping_info *si,
626 struct ore_per_dev_state *per_dev,
627 unsigned cur_len, bool do_xor)
628 {
629 if (ios->reading) {
630 if (per_dev->cur_sg >= ios->sgs_per_dev) {
631 ORE_DBGMSG("cur_sg(%d) >= sgs_per_dev(%d)\n" ,
632 per_dev->cur_sg, ios->sgs_per_dev);
633 return -ENOMEM;
634 }
635 _ore_add_sg_seg(per_dev, cur_len, true);
636 } else {
637 struct __stripe_pages_2d *sp2d = ios->sp2d;
638 struct page **pages = ios->parity_pages + ios->cur_par_page;
639 unsigned num_pages;
640 unsigned array_start = 0;
641 unsigned i;
642 int ret;
643
644 si->cur_pg = _sp2d_min_pg(sp2d);
645 num_pages = _sp2d_max_pg(sp2d) + 1 - si->cur_pg;
646
647 if (!per_dev->length) {
648 per_dev->offset += si->cur_pg * PAGE_SIZE;
649 /* If first stripe, Read in all read4write pages
650 * (if needed) before we calculate the first parity.
651 */
652 if (do_xor)
653 _read_4_write_first_stripe(ios);
654 }
655 if (!cur_len && do_xor)
656 /* If last stripe r4w pages of last stripe */
657 _read_4_write_last_stripe(ios);
658 _read_4_write_execute(ios);
659
660 for (i = 0; i < num_pages; i++) {
661 pages[i] = _raid_page_alloc();
662 if (unlikely(!pages[i]))
663 return -ENOMEM;
664
665 ++(ios->cur_par_page);
666 }
667
668 BUG_ON(si->cur_comp < sp2d->data_devs);
669 BUG_ON(si->cur_pg + num_pages > sp2d->pages_in_unit);
670
671 ret = _ore_add_stripe_unit(ios, &array_start, 0, pages,
672 per_dev, num_pages * PAGE_SIZE);
673 if (unlikely(ret))
674 return ret;
675
676 if (do_xor) {
677 _gen_xor_unit(sp2d);
678 _sp2d_reset(sp2d, ios->r4w, ios->private);
679 }
680 }
681 return 0;
682 }
683
_ore_post_alloc_raid_stuff(struct ore_io_state * ios)684 int _ore_post_alloc_raid_stuff(struct ore_io_state *ios)
685 {
686 if (ios->parity_pages) {
687 struct ore_layout *layout = ios->layout;
688 unsigned pages_in_unit = layout->stripe_unit / PAGE_SIZE;
689
690 if (_sp2d_alloc(pages_in_unit, layout->group_width,
691 layout->parity, &ios->sp2d)) {
692 return -ENOMEM;
693 }
694 }
695 return 0;
696 }
697
_ore_free_raid_stuff(struct ore_io_state * ios)698 void _ore_free_raid_stuff(struct ore_io_state *ios)
699 {
700 if (ios->sp2d) { /* writing and raid */
701 unsigned i;
702
703 for (i = 0; i < ios->cur_par_page; i++) {
704 struct page *page = ios->parity_pages[i];
705
706 if (page)
707 _raid_page_free(page);
708 }
709 if (ios->extra_part_alloc)
710 kfree(ios->parity_pages);
711 /* If IO returned an error pages might need unlocking */
712 _sp2d_reset(ios->sp2d, ios->r4w, ios->private);
713 _sp2d_free(ios->sp2d);
714 } else {
715 /* Will only be set if raid reading && sglist is big */
716 if (ios->extra_part_alloc)
717 kfree(ios->per_dev[0].sglist);
718 }
719 if (ios->ios_read_4_write)
720 ore_put_io_state(ios->ios_read_4_write);
721 }
722