This source file includes following definitions.
- persistent_gnt_timeout
- get_free_page
- put_free_pages
- shrink_free_pagepool
- add_persistent_gnt
- get_persistent_gnt
- put_persistent_gnt
- free_persistent_gnts
- xen_blkbk_unmap_purged_grants
- purge_persistent_gnt
- alloc_req
- free_req
- xen_vbd_translate
- xen_vbd_resize
- blkif_notify_work
- xen_blkif_be_int
- print_stats
- xen_blkif_schedule
- xen_blkbk_free_caches
- xen_blkbk_unmap_prepare
- xen_blkbk_unmap_and_respond_callback
- xen_blkbk_unmap_and_respond
- xen_blkbk_unmap
- xen_blkbk_map
- xen_blkbk_map_seg
- xen_blkbk_parse_indirect
- dispatch_discard_io
- dispatch_other_io
- xen_blk_drain_io
- __end_block_io_op
- end_block_io_op
- __do_block_io_op
- do_block_io_op
- dispatch_rw_block_io
- make_response
- xen_blkif_init
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37 #define pr_fmt(fmt) "xen-blkback: " fmt
38
39 #include <linux/spinlock.h>
40 #include <linux/kthread.h>
41 #include <linux/list.h>
42 #include <linux/delay.h>
43 #include <linux/freezer.h>
44 #include <linux/bitmap.h>
45
46 #include <xen/events.h>
47 #include <xen/page.h>
48 #include <xen/xen.h>
49 #include <asm/xen/hypervisor.h>
50 #include <asm/xen/hypercall.h>
51 #include <xen/balloon.h>
52 #include <xen/grant_table.h>
53 #include "common.h"
54
55
56
57
58
59
60
61
62
63
64
65 static int xen_blkif_max_buffer_pages = 1024;
66 module_param_named(max_buffer_pages, xen_blkif_max_buffer_pages, int, 0644);
67 MODULE_PARM_DESC(max_buffer_pages,
68 "Maximum number of free pages to keep in each block backend buffer");
69
70
71
72
73
74
75
76
77
78
79
80
81 static int xen_blkif_max_pgrants = 1056;
82 module_param_named(max_persistent_grants, xen_blkif_max_pgrants, int, 0644);
83 MODULE_PARM_DESC(max_persistent_grants,
84 "Maximum number of grants to map persistently");
85
86
87
88
89
90
91 static unsigned int xen_blkif_pgrant_timeout = 60;
92 module_param_named(persistent_grant_unused_seconds, xen_blkif_pgrant_timeout,
93 uint, 0644);
94 MODULE_PARM_DESC(persistent_grant_unused_seconds,
95 "Time in seconds an unused persistent grant is allowed to "
96 "remain allocated. Default is 60, 0 means unlimited.");
97
98
99
100
101
102 unsigned int xenblk_max_queues;
103 module_param_named(max_queues, xenblk_max_queues, uint, 0644);
104 MODULE_PARM_DESC(max_queues,
105 "Maximum number of hardware queues per virtual disk." \
106 "By default it is the number of online CPUs.");
107
108
109
110
111
112 unsigned int xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
113 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
114 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
115
116
117
118
119
120 #define LRU_INTERVAL 100
121
122
123
124
125
126
127 #define LRU_PERCENT_CLEAN 5
128
129
130 static unsigned int log_stats;
131 module_param(log_stats, int, 0644);
132
133 #define BLKBACK_INVALID_HANDLE (~0)
134
135
136 #define NUM_BATCH_FREE_PAGES 10
137
138 static inline bool persistent_gnt_timeout(struct persistent_gnt *persistent_gnt)
139 {
140 return xen_blkif_pgrant_timeout &&
141 (jiffies - persistent_gnt->last_used >=
142 HZ * xen_blkif_pgrant_timeout);
143 }
144
145 static inline int get_free_page(struct xen_blkif_ring *ring, struct page **page)
146 {
147 unsigned long flags;
148
149 spin_lock_irqsave(&ring->free_pages_lock, flags);
150 if (list_empty(&ring->free_pages)) {
151 BUG_ON(ring->free_pages_num != 0);
152 spin_unlock_irqrestore(&ring->free_pages_lock, flags);
153 return gnttab_alloc_pages(1, page);
154 }
155 BUG_ON(ring->free_pages_num == 0);
156 page[0] = list_first_entry(&ring->free_pages, struct page, lru);
157 list_del(&page[0]->lru);
158 ring->free_pages_num--;
159 spin_unlock_irqrestore(&ring->free_pages_lock, flags);
160
161 return 0;
162 }
163
164 static inline void put_free_pages(struct xen_blkif_ring *ring, struct page **page,
165 int num)
166 {
167 unsigned long flags;
168 int i;
169
170 spin_lock_irqsave(&ring->free_pages_lock, flags);
171 for (i = 0; i < num; i++)
172 list_add(&page[i]->lru, &ring->free_pages);
173 ring->free_pages_num += num;
174 spin_unlock_irqrestore(&ring->free_pages_lock, flags);
175 }
176
177 static inline void shrink_free_pagepool(struct xen_blkif_ring *ring, int num)
178 {
179
180 struct page *page[NUM_BATCH_FREE_PAGES];
181 unsigned int num_pages = 0;
182 unsigned long flags;
183
184 spin_lock_irqsave(&ring->free_pages_lock, flags);
185 while (ring->free_pages_num > num) {
186 BUG_ON(list_empty(&ring->free_pages));
187 page[num_pages] = list_first_entry(&ring->free_pages,
188 struct page, lru);
189 list_del(&page[num_pages]->lru);
190 ring->free_pages_num--;
191 if (++num_pages == NUM_BATCH_FREE_PAGES) {
192 spin_unlock_irqrestore(&ring->free_pages_lock, flags);
193 gnttab_free_pages(num_pages, page);
194 spin_lock_irqsave(&ring->free_pages_lock, flags);
195 num_pages = 0;
196 }
197 }
198 spin_unlock_irqrestore(&ring->free_pages_lock, flags);
199 if (num_pages != 0)
200 gnttab_free_pages(num_pages, page);
201 }
202
203 #define vaddr(page) ((unsigned long)pfn_to_kaddr(page_to_pfn(page)))
204
205 static int do_block_io_op(struct xen_blkif_ring *ring);
206 static int dispatch_rw_block_io(struct xen_blkif_ring *ring,
207 struct blkif_request *req,
208 struct pending_req *pending_req);
209 static void make_response(struct xen_blkif_ring *ring, u64 id,
210 unsigned short op, int st);
211
212 #define foreach_grant_safe(pos, n, rbtree, node) \
213 for ((pos) = container_of(rb_first((rbtree)), typeof(*(pos)), node), \
214 (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL; \
215 &(pos)->node != NULL; \
216 (pos) = container_of(n, typeof(*(pos)), node), \
217 (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL)
218
219
220
221
222
223
224
225
226
227
228
229
230 static int add_persistent_gnt(struct xen_blkif_ring *ring,
231 struct persistent_gnt *persistent_gnt)
232 {
233 struct rb_node **new = NULL, *parent = NULL;
234 struct persistent_gnt *this;
235 struct xen_blkif *blkif = ring->blkif;
236
237 if (ring->persistent_gnt_c >= xen_blkif_max_pgrants) {
238 if (!blkif->vbd.overflow_max_grants)
239 blkif->vbd.overflow_max_grants = 1;
240 return -EBUSY;
241 }
242
243 new = &ring->persistent_gnts.rb_node;
244 while (*new) {
245 this = container_of(*new, struct persistent_gnt, node);
246
247 parent = *new;
248 if (persistent_gnt->gnt < this->gnt)
249 new = &((*new)->rb_left);
250 else if (persistent_gnt->gnt > this->gnt)
251 new = &((*new)->rb_right);
252 else {
253 pr_alert_ratelimited("trying to add a gref that's already in the tree\n");
254 return -EINVAL;
255 }
256 }
257
258 persistent_gnt->active = true;
259
260 rb_link_node(&(persistent_gnt->node), parent, new);
261 rb_insert_color(&(persistent_gnt->node), &ring->persistent_gnts);
262 ring->persistent_gnt_c++;
263 atomic_inc(&ring->persistent_gnt_in_use);
264 return 0;
265 }
266
267 static struct persistent_gnt *get_persistent_gnt(struct xen_blkif_ring *ring,
268 grant_ref_t gref)
269 {
270 struct persistent_gnt *data;
271 struct rb_node *node = NULL;
272
273 node = ring->persistent_gnts.rb_node;
274 while (node) {
275 data = container_of(node, struct persistent_gnt, node);
276
277 if (gref < data->gnt)
278 node = node->rb_left;
279 else if (gref > data->gnt)
280 node = node->rb_right;
281 else {
282 if (data->active) {
283 pr_alert_ratelimited("requesting a grant already in use\n");
284 return NULL;
285 }
286 data->active = true;
287 atomic_inc(&ring->persistent_gnt_in_use);
288 return data;
289 }
290 }
291 return NULL;
292 }
293
294 static void put_persistent_gnt(struct xen_blkif_ring *ring,
295 struct persistent_gnt *persistent_gnt)
296 {
297 if (!persistent_gnt->active)
298 pr_alert_ratelimited("freeing a grant already unused\n");
299 persistent_gnt->last_used = jiffies;
300 persistent_gnt->active = false;
301 atomic_dec(&ring->persistent_gnt_in_use);
302 }
303
304 static void free_persistent_gnts(struct xen_blkif_ring *ring, struct rb_root *root,
305 unsigned int num)
306 {
307 struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
308 struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
309 struct persistent_gnt *persistent_gnt;
310 struct rb_node *n;
311 int segs_to_unmap = 0;
312 struct gntab_unmap_queue_data unmap_data;
313
314 unmap_data.pages = pages;
315 unmap_data.unmap_ops = unmap;
316 unmap_data.kunmap_ops = NULL;
317
318 foreach_grant_safe(persistent_gnt, n, root, node) {
319 BUG_ON(persistent_gnt->handle ==
320 BLKBACK_INVALID_HANDLE);
321 gnttab_set_unmap_op(&unmap[segs_to_unmap],
322 (unsigned long) pfn_to_kaddr(page_to_pfn(
323 persistent_gnt->page)),
324 GNTMAP_host_map,
325 persistent_gnt->handle);
326
327 pages[segs_to_unmap] = persistent_gnt->page;
328
329 if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST ||
330 !rb_next(&persistent_gnt->node)) {
331
332 unmap_data.count = segs_to_unmap;
333 BUG_ON(gnttab_unmap_refs_sync(&unmap_data));
334
335 put_free_pages(ring, pages, segs_to_unmap);
336 segs_to_unmap = 0;
337 }
338
339 rb_erase(&persistent_gnt->node, root);
340 kfree(persistent_gnt);
341 num--;
342 }
343 BUG_ON(num != 0);
344 }
345
346 void xen_blkbk_unmap_purged_grants(struct work_struct *work)
347 {
348 struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
349 struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
350 struct persistent_gnt *persistent_gnt;
351 int segs_to_unmap = 0;
352 struct xen_blkif_ring *ring = container_of(work, typeof(*ring), persistent_purge_work);
353 struct gntab_unmap_queue_data unmap_data;
354
355 unmap_data.pages = pages;
356 unmap_data.unmap_ops = unmap;
357 unmap_data.kunmap_ops = NULL;
358
359 while(!list_empty(&ring->persistent_purge_list)) {
360 persistent_gnt = list_first_entry(&ring->persistent_purge_list,
361 struct persistent_gnt,
362 remove_node);
363 list_del(&persistent_gnt->remove_node);
364
365 gnttab_set_unmap_op(&unmap[segs_to_unmap],
366 vaddr(persistent_gnt->page),
367 GNTMAP_host_map,
368 persistent_gnt->handle);
369
370 pages[segs_to_unmap] = persistent_gnt->page;
371
372 if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
373 unmap_data.count = segs_to_unmap;
374 BUG_ON(gnttab_unmap_refs_sync(&unmap_data));
375 put_free_pages(ring, pages, segs_to_unmap);
376 segs_to_unmap = 0;
377 }
378 kfree(persistent_gnt);
379 }
380 if (segs_to_unmap > 0) {
381 unmap_data.count = segs_to_unmap;
382 BUG_ON(gnttab_unmap_refs_sync(&unmap_data));
383 put_free_pages(ring, pages, segs_to_unmap);
384 }
385 }
386
387 static void purge_persistent_gnt(struct xen_blkif_ring *ring)
388 {
389 struct persistent_gnt *persistent_gnt;
390 struct rb_node *n;
391 unsigned int num_clean, total;
392 bool scan_used = false;
393 struct rb_root *root;
394
395 if (work_busy(&ring->persistent_purge_work)) {
396 pr_alert_ratelimited("Scheduled work from previous purge is still busy, cannot purge list\n");
397 goto out;
398 }
399
400 if (ring->persistent_gnt_c < xen_blkif_max_pgrants ||
401 (ring->persistent_gnt_c == xen_blkif_max_pgrants &&
402 !ring->blkif->vbd.overflow_max_grants)) {
403 num_clean = 0;
404 } else {
405 num_clean = (xen_blkif_max_pgrants / 100) * LRU_PERCENT_CLEAN;
406 num_clean = ring->persistent_gnt_c - xen_blkif_max_pgrants +
407 num_clean;
408 num_clean = min(ring->persistent_gnt_c, num_clean);
409 pr_debug("Going to purge at least %u persistent grants\n",
410 num_clean);
411 }
412
413
414
415
416
417
418
419
420
421
422 total = 0;
423
424 BUG_ON(!list_empty(&ring->persistent_purge_list));
425 root = &ring->persistent_gnts;
426 purge_list:
427 foreach_grant_safe(persistent_gnt, n, root, node) {
428 BUG_ON(persistent_gnt->handle ==
429 BLKBACK_INVALID_HANDLE);
430
431 if (persistent_gnt->active)
432 continue;
433 if (!scan_used && !persistent_gnt_timeout(persistent_gnt))
434 continue;
435 if (scan_used && total >= num_clean)
436 continue;
437
438 rb_erase(&persistent_gnt->node, root);
439 list_add(&persistent_gnt->remove_node,
440 &ring->persistent_purge_list);
441 total++;
442 }
443
444
445
446
447
448 if (!scan_used && total < num_clean) {
449 pr_debug("Still missing %u purged frames\n", num_clean - total);
450 scan_used = true;
451 goto purge_list;
452 }
453
454 if (total) {
455 ring->persistent_gnt_c -= total;
456 ring->blkif->vbd.overflow_max_grants = 0;
457
458
459 schedule_work(&ring->persistent_purge_work);
460 pr_debug("Purged %u/%u\n", num_clean, total);
461 }
462
463 out:
464 return;
465 }
466
467
468
469
470 static struct pending_req *alloc_req(struct xen_blkif_ring *ring)
471 {
472 struct pending_req *req = NULL;
473 unsigned long flags;
474
475 spin_lock_irqsave(&ring->pending_free_lock, flags);
476 if (!list_empty(&ring->pending_free)) {
477 req = list_entry(ring->pending_free.next, struct pending_req,
478 free_list);
479 list_del(&req->free_list);
480 }
481 spin_unlock_irqrestore(&ring->pending_free_lock, flags);
482 return req;
483 }
484
485
486
487
488
489 static void free_req(struct xen_blkif_ring *ring, struct pending_req *req)
490 {
491 unsigned long flags;
492 int was_empty;
493
494 spin_lock_irqsave(&ring->pending_free_lock, flags);
495 was_empty = list_empty(&ring->pending_free);
496 list_add(&req->free_list, &ring->pending_free);
497 spin_unlock_irqrestore(&ring->pending_free_lock, flags);
498 if (was_empty)
499 wake_up(&ring->pending_free_wq);
500 }
501
502
503
504
505 static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif,
506 int operation)
507 {
508 struct xen_vbd *vbd = &blkif->vbd;
509 int rc = -EACCES;
510
511 if ((operation != REQ_OP_READ) && vbd->readonly)
512 goto out;
513
514 if (likely(req->nr_sects)) {
515 blkif_sector_t end = req->sector_number + req->nr_sects;
516
517 if (unlikely(end < req->sector_number))
518 goto out;
519 if (unlikely(end > vbd_sz(vbd)))
520 goto out;
521 }
522
523 req->dev = vbd->pdevice;
524 req->bdev = vbd->bdev;
525 rc = 0;
526
527 out:
528 return rc;
529 }
530
531 static void xen_vbd_resize(struct xen_blkif *blkif)
532 {
533 struct xen_vbd *vbd = &blkif->vbd;
534 struct xenbus_transaction xbt;
535 int err;
536 struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be);
537 unsigned long long new_size = vbd_sz(vbd);
538
539 pr_info("VBD Resize: Domid: %d, Device: (%d, %d)\n",
540 blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice));
541 pr_info("VBD Resize: new size %llu\n", new_size);
542 vbd->size = new_size;
543 again:
544 err = xenbus_transaction_start(&xbt);
545 if (err) {
546 pr_warn("Error starting transaction\n");
547 return;
548 }
549 err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
550 (unsigned long long)vbd_sz(vbd));
551 if (err) {
552 pr_warn("Error writing new size\n");
553 goto abort;
554 }
555
556
557
558
559
560 err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state);
561 if (err) {
562 pr_warn("Error writing the state\n");
563 goto abort;
564 }
565
566 err = xenbus_transaction_end(xbt, 0);
567 if (err == -EAGAIN)
568 goto again;
569 if (err)
570 pr_warn("Error ending transaction\n");
571 return;
572 abort:
573 xenbus_transaction_end(xbt, 1);
574 }
575
576
577
578
579 static void blkif_notify_work(struct xen_blkif_ring *ring)
580 {
581 ring->waiting_reqs = 1;
582 wake_up(&ring->wq);
583 }
584
585 irqreturn_t xen_blkif_be_int(int irq, void *dev_id)
586 {
587 blkif_notify_work(dev_id);
588 return IRQ_HANDLED;
589 }
590
591
592
593
594
595 static void print_stats(struct xen_blkif_ring *ring)
596 {
597 pr_info("(%s): oo %3llu | rd %4llu | wr %4llu | f %4llu"
598 " | ds %4llu | pg: %4u/%4d\n",
599 current->comm, ring->st_oo_req,
600 ring->st_rd_req, ring->st_wr_req,
601 ring->st_f_req, ring->st_ds_req,
602 ring->persistent_gnt_c,
603 xen_blkif_max_pgrants);
604 ring->st_print = jiffies + msecs_to_jiffies(10 * 1000);
605 ring->st_rd_req = 0;
606 ring->st_wr_req = 0;
607 ring->st_oo_req = 0;
608 ring->st_ds_req = 0;
609 }
610
611 int xen_blkif_schedule(void *arg)
612 {
613 struct xen_blkif_ring *ring = arg;
614 struct xen_blkif *blkif = ring->blkif;
615 struct xen_vbd *vbd = &blkif->vbd;
616 unsigned long timeout;
617 int ret;
618
619 set_freezable();
620 while (!kthread_should_stop()) {
621 if (try_to_freeze())
622 continue;
623 if (unlikely(vbd->size != vbd_sz(vbd)))
624 xen_vbd_resize(blkif);
625
626 timeout = msecs_to_jiffies(LRU_INTERVAL);
627
628 timeout = wait_event_interruptible_timeout(
629 ring->wq,
630 ring->waiting_reqs || kthread_should_stop(),
631 timeout);
632 if (timeout == 0)
633 goto purge_gnt_list;
634 timeout = wait_event_interruptible_timeout(
635 ring->pending_free_wq,
636 !list_empty(&ring->pending_free) ||
637 kthread_should_stop(),
638 timeout);
639 if (timeout == 0)
640 goto purge_gnt_list;
641
642 ring->waiting_reqs = 0;
643 smp_mb();
644
645 ret = do_block_io_op(ring);
646 if (ret > 0)
647 ring->waiting_reqs = 1;
648 if (ret == -EACCES)
649 wait_event_interruptible(ring->shutdown_wq,
650 kthread_should_stop());
651
652 purge_gnt_list:
653 if (blkif->vbd.feature_gnt_persistent &&
654 time_after(jiffies, ring->next_lru)) {
655 purge_persistent_gnt(ring);
656 ring->next_lru = jiffies + msecs_to_jiffies(LRU_INTERVAL);
657 }
658
659
660 shrink_free_pagepool(ring, xen_blkif_max_buffer_pages);
661
662 if (log_stats && time_after(jiffies, ring->st_print))
663 print_stats(ring);
664 }
665
666
667 flush_work(&ring->persistent_purge_work);
668
669 if (log_stats)
670 print_stats(ring);
671
672 ring->xenblkd = NULL;
673
674 return 0;
675 }
676
677
678
679
680 void xen_blkbk_free_caches(struct xen_blkif_ring *ring)
681 {
682
683 if (!RB_EMPTY_ROOT(&ring->persistent_gnts))
684 free_persistent_gnts(ring, &ring->persistent_gnts,
685 ring->persistent_gnt_c);
686
687 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
688 ring->persistent_gnt_c = 0;
689
690
691 shrink_free_pagepool(ring, 0 );
692 }
693
694 static unsigned int xen_blkbk_unmap_prepare(
695 struct xen_blkif_ring *ring,
696 struct grant_page **pages,
697 unsigned int num,
698 struct gnttab_unmap_grant_ref *unmap_ops,
699 struct page **unmap_pages)
700 {
701 unsigned int i, invcount = 0;
702
703 for (i = 0; i < num; i++) {
704 if (pages[i]->persistent_gnt != NULL) {
705 put_persistent_gnt(ring, pages[i]->persistent_gnt);
706 continue;
707 }
708 if (pages[i]->handle == BLKBACK_INVALID_HANDLE)
709 continue;
710 unmap_pages[invcount] = pages[i]->page;
711 gnttab_set_unmap_op(&unmap_ops[invcount], vaddr(pages[i]->page),
712 GNTMAP_host_map, pages[i]->handle);
713 pages[i]->handle = BLKBACK_INVALID_HANDLE;
714 invcount++;
715 }
716
717 return invcount;
718 }
719
720 static void xen_blkbk_unmap_and_respond_callback(int result, struct gntab_unmap_queue_data *data)
721 {
722 struct pending_req *pending_req = (struct pending_req *)(data->data);
723 struct xen_blkif_ring *ring = pending_req->ring;
724 struct xen_blkif *blkif = ring->blkif;
725
726
727
728 BUG_ON(result);
729
730 put_free_pages(ring, data->pages, data->count);
731 make_response(ring, pending_req->id,
732 pending_req->operation, pending_req->status);
733 free_req(ring, pending_req);
734
735
736
737
738
739
740
741
742
743
744
745
746 if (atomic_dec_and_test(&ring->inflight) && atomic_read(&blkif->drain)) {
747 complete(&blkif->drain_complete);
748 }
749 xen_blkif_put(blkif);
750 }
751
752 static void xen_blkbk_unmap_and_respond(struct pending_req *req)
753 {
754 struct gntab_unmap_queue_data* work = &req->gnttab_unmap_data;
755 struct xen_blkif_ring *ring = req->ring;
756 struct grant_page **pages = req->segments;
757 unsigned int invcount;
758
759 invcount = xen_blkbk_unmap_prepare(ring, pages, req->nr_segs,
760 req->unmap, req->unmap_pages);
761
762 work->data = req;
763 work->done = xen_blkbk_unmap_and_respond_callback;
764 work->unmap_ops = req->unmap;
765 work->kunmap_ops = NULL;
766 work->pages = req->unmap_pages;
767 work->count = invcount;
768
769 gnttab_unmap_refs_async(&req->gnttab_unmap_data);
770 }
771
772
773
774
775
776
777
778
779
780 static void xen_blkbk_unmap(struct xen_blkif_ring *ring,
781 struct grant_page *pages[],
782 int num)
783 {
784 struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
785 struct page *unmap_pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
786 unsigned int invcount = 0;
787 int ret;
788
789 while (num) {
790 unsigned int batch = min(num, BLKIF_MAX_SEGMENTS_PER_REQUEST);
791
792 invcount = xen_blkbk_unmap_prepare(ring, pages, batch,
793 unmap, unmap_pages);
794 if (invcount) {
795 ret = gnttab_unmap_refs(unmap, NULL, unmap_pages, invcount);
796 BUG_ON(ret);
797 put_free_pages(ring, unmap_pages, invcount);
798 }
799 pages += batch;
800 num -= batch;
801 }
802 }
803
804 static int xen_blkbk_map(struct xen_blkif_ring *ring,
805 struct grant_page *pages[],
806 int num, bool ro)
807 {
808 struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST];
809 struct page *pages_to_gnt[BLKIF_MAX_SEGMENTS_PER_REQUEST];
810 struct persistent_gnt *persistent_gnt = NULL;
811 phys_addr_t addr = 0;
812 int i, seg_idx, new_map_idx;
813 int segs_to_map = 0;
814 int ret = 0;
815 int last_map = 0, map_until = 0;
816 int use_persistent_gnts;
817 struct xen_blkif *blkif = ring->blkif;
818
819 use_persistent_gnts = (blkif->vbd.feature_gnt_persistent);
820
821
822
823
824
825
826 again:
827 for (i = map_until; i < num; i++) {
828 uint32_t flags;
829
830 if (use_persistent_gnts) {
831 persistent_gnt = get_persistent_gnt(
832 ring,
833 pages[i]->gref);
834 }
835
836 if (persistent_gnt) {
837
838
839
840
841 pages[i]->page = persistent_gnt->page;
842 pages[i]->persistent_gnt = persistent_gnt;
843 } else {
844 if (get_free_page(ring, &pages[i]->page))
845 goto out_of_memory;
846 addr = vaddr(pages[i]->page);
847 pages_to_gnt[segs_to_map] = pages[i]->page;
848 pages[i]->persistent_gnt = NULL;
849 flags = GNTMAP_host_map;
850 if (!use_persistent_gnts && ro)
851 flags |= GNTMAP_readonly;
852 gnttab_set_map_op(&map[segs_to_map++], addr,
853 flags, pages[i]->gref,
854 blkif->domid);
855 }
856 map_until = i + 1;
857 if (segs_to_map == BLKIF_MAX_SEGMENTS_PER_REQUEST)
858 break;
859 }
860
861 if (segs_to_map) {
862 ret = gnttab_map_refs(map, NULL, pages_to_gnt, segs_to_map);
863 BUG_ON(ret);
864 }
865
866
867
868
869
870
871 for (seg_idx = last_map, new_map_idx = 0; seg_idx < map_until; seg_idx++) {
872 if (!pages[seg_idx]->persistent_gnt) {
873
874 BUG_ON(new_map_idx >= segs_to_map);
875 if (unlikely(map[new_map_idx].status != 0)) {
876 pr_debug("invalid buffer -- could not remap it\n");
877 put_free_pages(ring, &pages[seg_idx]->page, 1);
878 pages[seg_idx]->handle = BLKBACK_INVALID_HANDLE;
879 ret |= 1;
880 goto next;
881 }
882 pages[seg_idx]->handle = map[new_map_idx].handle;
883 } else {
884 continue;
885 }
886 if (use_persistent_gnts &&
887 ring->persistent_gnt_c < xen_blkif_max_pgrants) {
888
889
890
891
892 persistent_gnt = kmalloc(sizeof(struct persistent_gnt),
893 GFP_KERNEL);
894 if (!persistent_gnt) {
895
896
897
898
899
900 goto next;
901 }
902 persistent_gnt->gnt = map[new_map_idx].ref;
903 persistent_gnt->handle = map[new_map_idx].handle;
904 persistent_gnt->page = pages[seg_idx]->page;
905 if (add_persistent_gnt(ring,
906 persistent_gnt)) {
907 kfree(persistent_gnt);
908 persistent_gnt = NULL;
909 goto next;
910 }
911 pages[seg_idx]->persistent_gnt = persistent_gnt;
912 pr_debug("grant %u added to the tree of persistent grants, using %u/%u\n",
913 persistent_gnt->gnt, ring->persistent_gnt_c,
914 xen_blkif_max_pgrants);
915 goto next;
916 }
917 if (use_persistent_gnts && !blkif->vbd.overflow_max_grants) {
918 blkif->vbd.overflow_max_grants = 1;
919 pr_debug("domain %u, device %#x is using maximum number of persistent grants\n",
920 blkif->domid, blkif->vbd.handle);
921 }
922
923
924
925
926 next:
927 new_map_idx++;
928 }
929 segs_to_map = 0;
930 last_map = map_until;
931 if (map_until != num)
932 goto again;
933
934 return ret;
935
936 out_of_memory:
937 pr_alert("%s: out of memory\n", __func__);
938 put_free_pages(ring, pages_to_gnt, segs_to_map);
939 for (i = last_map; i < num; i++)
940 pages[i]->handle = BLKBACK_INVALID_HANDLE;
941 return -ENOMEM;
942 }
943
944 static int xen_blkbk_map_seg(struct pending_req *pending_req)
945 {
946 int rc;
947
948 rc = xen_blkbk_map(pending_req->ring, pending_req->segments,
949 pending_req->nr_segs,
950 (pending_req->operation != BLKIF_OP_READ));
951
952 return rc;
953 }
954
955 static int xen_blkbk_parse_indirect(struct blkif_request *req,
956 struct pending_req *pending_req,
957 struct seg_buf seg[],
958 struct phys_req *preq)
959 {
960 struct grant_page **pages = pending_req->indirect_pages;
961 struct xen_blkif_ring *ring = pending_req->ring;
962 int indirect_grefs, rc, n, nseg, i;
963 struct blkif_request_segment *segments = NULL;
964
965 nseg = pending_req->nr_segs;
966 indirect_grefs = INDIRECT_PAGES(nseg);
967 BUG_ON(indirect_grefs > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
968
969 for (i = 0; i < indirect_grefs; i++)
970 pages[i]->gref = req->u.indirect.indirect_grefs[i];
971
972 rc = xen_blkbk_map(ring, pages, indirect_grefs, true);
973 if (rc)
974 goto unmap;
975
976 for (n = 0, i = 0; n < nseg; n++) {
977 uint8_t first_sect, last_sect;
978
979 if ((n % SEGS_PER_INDIRECT_FRAME) == 0) {
980
981 if (segments)
982 kunmap_atomic(segments);
983 segments = kmap_atomic(pages[n/SEGS_PER_INDIRECT_FRAME]->page);
984 }
985 i = n % SEGS_PER_INDIRECT_FRAME;
986
987 pending_req->segments[n]->gref = segments[i].gref;
988
989 first_sect = READ_ONCE(segments[i].first_sect);
990 last_sect = READ_ONCE(segments[i].last_sect);
991 if (last_sect >= (XEN_PAGE_SIZE >> 9) || last_sect < first_sect) {
992 rc = -EINVAL;
993 goto unmap;
994 }
995
996 seg[n].nsec = last_sect - first_sect + 1;
997 seg[n].offset = first_sect << 9;
998 preq->nr_sects += seg[n].nsec;
999 }
1000
1001 unmap:
1002 if (segments)
1003 kunmap_atomic(segments);
1004 xen_blkbk_unmap(ring, pages, indirect_grefs);
1005 return rc;
1006 }
1007
1008 static int dispatch_discard_io(struct xen_blkif_ring *ring,
1009 struct blkif_request *req)
1010 {
1011 int err = 0;
1012 int status = BLKIF_RSP_OKAY;
1013 struct xen_blkif *blkif = ring->blkif;
1014 struct block_device *bdev = blkif->vbd.bdev;
1015 unsigned long secure;
1016 struct phys_req preq;
1017
1018 xen_blkif_get(blkif);
1019
1020 preq.sector_number = req->u.discard.sector_number;
1021 preq.nr_sects = req->u.discard.nr_sectors;
1022
1023 err = xen_vbd_translate(&preq, blkif, REQ_OP_WRITE);
1024 if (err) {
1025 pr_warn("access denied: DISCARD [%llu->%llu] on dev=%04x\n",
1026 preq.sector_number,
1027 preq.sector_number + preq.nr_sects, blkif->vbd.pdevice);
1028 goto fail_response;
1029 }
1030 ring->st_ds_req++;
1031
1032 secure = (blkif->vbd.discard_secure &&
1033 (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ?
1034 BLKDEV_DISCARD_SECURE : 0;
1035
1036 err = blkdev_issue_discard(bdev, req->u.discard.sector_number,
1037 req->u.discard.nr_sectors,
1038 GFP_KERNEL, secure);
1039 fail_response:
1040 if (err == -EOPNOTSUPP) {
1041 pr_debug("discard op failed, not supported\n");
1042 status = BLKIF_RSP_EOPNOTSUPP;
1043 } else if (err)
1044 status = BLKIF_RSP_ERROR;
1045
1046 make_response(ring, req->u.discard.id, req->operation, status);
1047 xen_blkif_put(blkif);
1048 return err;
1049 }
1050
1051 static int dispatch_other_io(struct xen_blkif_ring *ring,
1052 struct blkif_request *req,
1053 struct pending_req *pending_req)
1054 {
1055 free_req(ring, pending_req);
1056 make_response(ring, req->u.other.id, req->operation,
1057 BLKIF_RSP_EOPNOTSUPP);
1058 return -EIO;
1059 }
1060
1061 static void xen_blk_drain_io(struct xen_blkif_ring *ring)
1062 {
1063 struct xen_blkif *blkif = ring->blkif;
1064
1065 atomic_set(&blkif->drain, 1);
1066 do {
1067 if (atomic_read(&ring->inflight) == 0)
1068 break;
1069 wait_for_completion_interruptible_timeout(
1070 &blkif->drain_complete, HZ);
1071
1072 if (!atomic_read(&blkif->drain))
1073 break;
1074 } while (!kthread_should_stop());
1075 atomic_set(&blkif->drain, 0);
1076 }
1077
1078 static void __end_block_io_op(struct pending_req *pending_req,
1079 blk_status_t error)
1080 {
1081
1082 if (pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE &&
1083 error == BLK_STS_NOTSUPP) {
1084 pr_debug("flush diskcache op failed, not supported\n");
1085 xen_blkbk_flush_diskcache(XBT_NIL, pending_req->ring->blkif->be, 0);
1086 pending_req->status = BLKIF_RSP_EOPNOTSUPP;
1087 } else if (pending_req->operation == BLKIF_OP_WRITE_BARRIER &&
1088 error == BLK_STS_NOTSUPP) {
1089 pr_debug("write barrier op failed, not supported\n");
1090 xen_blkbk_barrier(XBT_NIL, pending_req->ring->blkif->be, 0);
1091 pending_req->status = BLKIF_RSP_EOPNOTSUPP;
1092 } else if (error) {
1093 pr_debug("Buffer not up-to-date at end of operation,"
1094 " error=%d\n", error);
1095 pending_req->status = BLKIF_RSP_ERROR;
1096 }
1097
1098
1099
1100
1101
1102
1103 if (atomic_dec_and_test(&pending_req->pendcnt))
1104 xen_blkbk_unmap_and_respond(pending_req);
1105 }
1106
1107
1108
1109
1110 static void end_block_io_op(struct bio *bio)
1111 {
1112 __end_block_io_op(bio->bi_private, bio->bi_status);
1113 bio_put(bio);
1114 }
1115
1116
1117
1118
1119
1120
1121
1122
1123 static int
1124 __do_block_io_op(struct xen_blkif_ring *ring)
1125 {
1126 union blkif_back_rings *blk_rings = &ring->blk_rings;
1127 struct blkif_request req;
1128 struct pending_req *pending_req;
1129 RING_IDX rc, rp;
1130 int more_to_do = 0;
1131
1132 rc = blk_rings->common.req_cons;
1133 rp = blk_rings->common.sring->req_prod;
1134 rmb();
1135
1136 if (RING_REQUEST_PROD_OVERFLOW(&blk_rings->common, rp)) {
1137 rc = blk_rings->common.rsp_prod_pvt;
1138 pr_warn("Frontend provided bogus ring requests (%d - %d = %d). Halting ring processing on dev=%04x\n",
1139 rp, rc, rp - rc, ring->blkif->vbd.pdevice);
1140 return -EACCES;
1141 }
1142 while (rc != rp) {
1143
1144 if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc))
1145 break;
1146
1147 if (kthread_should_stop()) {
1148 more_to_do = 1;
1149 break;
1150 }
1151
1152 pending_req = alloc_req(ring);
1153 if (NULL == pending_req) {
1154 ring->st_oo_req++;
1155 more_to_do = 1;
1156 break;
1157 }
1158
1159 switch (ring->blkif->blk_protocol) {
1160 case BLKIF_PROTOCOL_NATIVE:
1161 memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req));
1162 break;
1163 case BLKIF_PROTOCOL_X86_32:
1164 blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc));
1165 break;
1166 case BLKIF_PROTOCOL_X86_64:
1167 blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc));
1168 break;
1169 default:
1170 BUG();
1171 }
1172 blk_rings->common.req_cons = ++rc;
1173
1174
1175 barrier();
1176
1177 switch (req.operation) {
1178 case BLKIF_OP_READ:
1179 case BLKIF_OP_WRITE:
1180 case BLKIF_OP_WRITE_BARRIER:
1181 case BLKIF_OP_FLUSH_DISKCACHE:
1182 case BLKIF_OP_INDIRECT:
1183 if (dispatch_rw_block_io(ring, &req, pending_req))
1184 goto done;
1185 break;
1186 case BLKIF_OP_DISCARD:
1187 free_req(ring, pending_req);
1188 if (dispatch_discard_io(ring, &req))
1189 goto done;
1190 break;
1191 default:
1192 if (dispatch_other_io(ring, &req, pending_req))
1193 goto done;
1194 break;
1195 }
1196
1197
1198 cond_resched();
1199 }
1200 done:
1201 return more_to_do;
1202 }
1203
1204 static int
1205 do_block_io_op(struct xen_blkif_ring *ring)
1206 {
1207 union blkif_back_rings *blk_rings = &ring->blk_rings;
1208 int more_to_do;
1209
1210 do {
1211 more_to_do = __do_block_io_op(ring);
1212 if (more_to_do)
1213 break;
1214
1215 RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do);
1216 } while (more_to_do);
1217
1218 return more_to_do;
1219 }
1220
1221
1222
1223
1224 static int dispatch_rw_block_io(struct xen_blkif_ring *ring,
1225 struct blkif_request *req,
1226 struct pending_req *pending_req)
1227 {
1228 struct phys_req preq;
1229 struct seg_buf *seg = pending_req->seg;
1230 unsigned int nseg;
1231 struct bio *bio = NULL;
1232 struct bio **biolist = pending_req->biolist;
1233 int i, nbio = 0;
1234 int operation;
1235 int operation_flags = 0;
1236 struct blk_plug plug;
1237 bool drain = false;
1238 struct grant_page **pages = pending_req->segments;
1239 unsigned short req_operation;
1240
1241 req_operation = req->operation == BLKIF_OP_INDIRECT ?
1242 req->u.indirect.indirect_op : req->operation;
1243
1244 if ((req->operation == BLKIF_OP_INDIRECT) &&
1245 (req_operation != BLKIF_OP_READ) &&
1246 (req_operation != BLKIF_OP_WRITE)) {
1247 pr_debug("Invalid indirect operation (%u)\n", req_operation);
1248 goto fail_response;
1249 }
1250
1251 switch (req_operation) {
1252 case BLKIF_OP_READ:
1253 ring->st_rd_req++;
1254 operation = REQ_OP_READ;
1255 break;
1256 case BLKIF_OP_WRITE:
1257 ring->st_wr_req++;
1258 operation = REQ_OP_WRITE;
1259 operation_flags = REQ_SYNC | REQ_IDLE;
1260 break;
1261 case BLKIF_OP_WRITE_BARRIER:
1262 drain = true;
1263
1264 case BLKIF_OP_FLUSH_DISKCACHE:
1265 ring->st_f_req++;
1266 operation = REQ_OP_WRITE;
1267 operation_flags = REQ_PREFLUSH;
1268 break;
1269 default:
1270 operation = 0;
1271 goto fail_response;
1272 break;
1273 }
1274
1275
1276 nseg = req->operation == BLKIF_OP_INDIRECT ?
1277 req->u.indirect.nr_segments : req->u.rw.nr_segments;
1278
1279 if (unlikely(nseg == 0 && operation_flags != REQ_PREFLUSH) ||
1280 unlikely((req->operation != BLKIF_OP_INDIRECT) &&
1281 (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) ||
1282 unlikely((req->operation == BLKIF_OP_INDIRECT) &&
1283 (nseg > MAX_INDIRECT_SEGMENTS))) {
1284 pr_debug("Bad number of segments in request (%d)\n", nseg);
1285
1286 goto fail_response;
1287 }
1288
1289 preq.nr_sects = 0;
1290
1291 pending_req->ring = ring;
1292 pending_req->id = req->u.rw.id;
1293 pending_req->operation = req_operation;
1294 pending_req->status = BLKIF_RSP_OKAY;
1295 pending_req->nr_segs = nseg;
1296
1297 if (req->operation != BLKIF_OP_INDIRECT) {
1298 preq.dev = req->u.rw.handle;
1299 preq.sector_number = req->u.rw.sector_number;
1300 for (i = 0; i < nseg; i++) {
1301 pages[i]->gref = req->u.rw.seg[i].gref;
1302 seg[i].nsec = req->u.rw.seg[i].last_sect -
1303 req->u.rw.seg[i].first_sect + 1;
1304 seg[i].offset = (req->u.rw.seg[i].first_sect << 9);
1305 if ((req->u.rw.seg[i].last_sect >= (XEN_PAGE_SIZE >> 9)) ||
1306 (req->u.rw.seg[i].last_sect <
1307 req->u.rw.seg[i].first_sect))
1308 goto fail_response;
1309 preq.nr_sects += seg[i].nsec;
1310 }
1311 } else {
1312 preq.dev = req->u.indirect.handle;
1313 preq.sector_number = req->u.indirect.sector_number;
1314 if (xen_blkbk_parse_indirect(req, pending_req, seg, &preq))
1315 goto fail_response;
1316 }
1317
1318 if (xen_vbd_translate(&preq, ring->blkif, operation) != 0) {
1319 pr_debug("access denied: %s of [%llu,%llu] on dev=%04x\n",
1320 operation == REQ_OP_READ ? "read" : "write",
1321 preq.sector_number,
1322 preq.sector_number + preq.nr_sects,
1323 ring->blkif->vbd.pdevice);
1324 goto fail_response;
1325 }
1326
1327
1328
1329
1330
1331 for (i = 0; i < nseg; i++) {
1332 if (((int)preq.sector_number|(int)seg[i].nsec) &
1333 ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) {
1334 pr_debug("Misaligned I/O request from domain %d\n",
1335 ring->blkif->domid);
1336 goto fail_response;
1337 }
1338 }
1339
1340
1341
1342
1343 if (drain)
1344 xen_blk_drain_io(pending_req->ring);
1345
1346
1347
1348
1349
1350
1351
1352 if (xen_blkbk_map_seg(pending_req))
1353 goto fail_flush;
1354
1355
1356
1357
1358
1359 xen_blkif_get(ring->blkif);
1360 atomic_inc(&ring->inflight);
1361
1362 for (i = 0; i < nseg; i++) {
1363 while ((bio == NULL) ||
1364 (bio_add_page(bio,
1365 pages[i]->page,
1366 seg[i].nsec << 9,
1367 seg[i].offset) == 0)) {
1368
1369 int nr_iovecs = min_t(int, (nseg-i), BIO_MAX_PAGES);
1370 bio = bio_alloc(GFP_KERNEL, nr_iovecs);
1371 if (unlikely(bio == NULL))
1372 goto fail_put_bio;
1373
1374 biolist[nbio++] = bio;
1375 bio_set_dev(bio, preq.bdev);
1376 bio->bi_private = pending_req;
1377 bio->bi_end_io = end_block_io_op;
1378 bio->bi_iter.bi_sector = preq.sector_number;
1379 bio_set_op_attrs(bio, operation, operation_flags);
1380 }
1381
1382 preq.sector_number += seg[i].nsec;
1383 }
1384
1385
1386 if (!bio) {
1387 BUG_ON(operation_flags != REQ_PREFLUSH);
1388
1389 bio = bio_alloc(GFP_KERNEL, 0);
1390 if (unlikely(bio == NULL))
1391 goto fail_put_bio;
1392
1393 biolist[nbio++] = bio;
1394 bio_set_dev(bio, preq.bdev);
1395 bio->bi_private = pending_req;
1396 bio->bi_end_io = end_block_io_op;
1397 bio_set_op_attrs(bio, operation, operation_flags);
1398 }
1399
1400 atomic_set(&pending_req->pendcnt, nbio);
1401 blk_start_plug(&plug);
1402
1403 for (i = 0; i < nbio; i++)
1404 submit_bio(biolist[i]);
1405
1406
1407 blk_finish_plug(&plug);
1408
1409 if (operation == REQ_OP_READ)
1410 ring->st_rd_sect += preq.nr_sects;
1411 else if (operation == REQ_OP_WRITE)
1412 ring->st_wr_sect += preq.nr_sects;
1413
1414 return 0;
1415
1416 fail_flush:
1417 xen_blkbk_unmap(ring, pending_req->segments,
1418 pending_req->nr_segs);
1419 fail_response:
1420
1421 make_response(ring, req->u.rw.id, req_operation, BLKIF_RSP_ERROR);
1422 free_req(ring, pending_req);
1423 msleep(1);
1424 return -EIO;
1425
1426 fail_put_bio:
1427 for (i = 0; i < nbio; i++)
1428 bio_put(biolist[i]);
1429 atomic_set(&pending_req->pendcnt, 1);
1430 __end_block_io_op(pending_req, BLK_STS_RESOURCE);
1431 msleep(1);
1432 return -EIO;
1433 }
1434
1435
1436
1437
1438
1439
1440 static void make_response(struct xen_blkif_ring *ring, u64 id,
1441 unsigned short op, int st)
1442 {
1443 struct blkif_response *resp;
1444 unsigned long flags;
1445 union blkif_back_rings *blk_rings;
1446 int notify;
1447
1448 spin_lock_irqsave(&ring->blk_ring_lock, flags);
1449 blk_rings = &ring->blk_rings;
1450
1451 switch (ring->blkif->blk_protocol) {
1452 case BLKIF_PROTOCOL_NATIVE:
1453 resp = RING_GET_RESPONSE(&blk_rings->native,
1454 blk_rings->native.rsp_prod_pvt);
1455 break;
1456 case BLKIF_PROTOCOL_X86_32:
1457 resp = RING_GET_RESPONSE(&blk_rings->x86_32,
1458 blk_rings->x86_32.rsp_prod_pvt);
1459 break;
1460 case BLKIF_PROTOCOL_X86_64:
1461 resp = RING_GET_RESPONSE(&blk_rings->x86_64,
1462 blk_rings->x86_64.rsp_prod_pvt);
1463 break;
1464 default:
1465 BUG();
1466 }
1467
1468 resp->id = id;
1469 resp->operation = op;
1470 resp->status = st;
1471
1472 blk_rings->common.rsp_prod_pvt++;
1473 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify);
1474 spin_unlock_irqrestore(&ring->blk_ring_lock, flags);
1475 if (notify)
1476 notify_remote_via_irq(ring->irq);
1477 }
1478
1479 static int __init xen_blkif_init(void)
1480 {
1481 int rc = 0;
1482
1483 if (!xen_domain())
1484 return -ENODEV;
1485
1486 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
1487 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
1488 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
1489 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
1490 }
1491
1492 if (xenblk_max_queues == 0)
1493 xenblk_max_queues = num_online_cpus();
1494
1495 rc = xen_blkif_interface_init();
1496 if (rc)
1497 goto failed_init;
1498
1499 rc = xen_blkif_xenbus_init();
1500 if (rc)
1501 goto failed_init;
1502
1503 failed_init:
1504 return rc;
1505 }
1506
1507 module_init(xen_blkif_init);
1508
1509 MODULE_LICENSE("Dual BSD/GPL");
1510 MODULE_ALIAS("xen-backend:vbd");