Commit | Line | Data |
---|---|---|
4d05a28d KRW |
1 | /****************************************************************************** |
2 | * arch/xen/drivers/blkif/backend/main.c | |
3 | * | |
4 | * Back-end of the driver for virtual block devices. This portion of the | |
5 | * driver exports a 'unified' block-device interface that can be accessed | |
6 | * by any operating system that implements a compatible front end. A | |
7 | * reference front-end implementation can be found in: | |
8 | * arch/xen/drivers/blkif/frontend | |
9 | * | |
10 | * Copyright (c) 2003-2004, Keir Fraser & Steve Hand | |
11 | * Copyright (c) 2005, Christopher Clark | |
12 | * | |
13 | * This program is free software; you can redistribute it and/or | |
14 | * modify it under the terms of the GNU General Public License version 2 | |
15 | * as published by the Free Software Foundation; or, when distributed | |
16 | * separately from the Linux kernel or incorporated into other | |
17 | * software packages, subject to the following license: | |
18 | * | |
19 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
20 | * of this source file (the "Software"), to deal in the Software without | |
21 | * restriction, including without limitation the rights to use, copy, modify, | |
22 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
23 | * and to permit persons to whom the Software is furnished to do so, subject to | |
24 | * the following conditions: | |
25 | * | |
26 | * The above copyright notice and this permission notice shall be included in | |
27 | * all copies or substantial portions of the Software. | |
28 | * | |
29 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
30 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
31 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
32 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
33 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
34 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
35 | * IN THE SOFTWARE. | |
36 | */ | |
37 | ||
38 | #include <linux/spinlock.h> | |
39 | #include <linux/kthread.h> | |
40 | #include <linux/list.h> | |
41 | #include <linux/delay.h> | |
88122933 | 42 | #include <linux/freezer.h> |
afd91d07 | 43 | |
4d05a28d | 44 | #include <xen/balloon.h> |
88122933 JF |
45 | #include <xen/events.h> |
46 | #include <xen/page.h> | |
47 | #include <asm/xen/hypervisor.h> | |
48 | #include <asm/xen/hypercall.h> | |
4d05a28d KRW |
49 | #include "common.h" |
50 | ||
51 | /* | |
52 | * These are rather arbitrary. They are fairly large because adjacent requests | |
53 | * pulled from a communication ring are quite likely to end up being part of | |
54 | * the same scatter/gather request at the disc. | |
55 | * | |
56 | * ** TRY INCREASING 'blkif_reqs' IF WRITE SPEEDS SEEM TOO LOW ** | |
57 | * | |
58 | * This will increase the chances of being able to write whole tracks. | |
59 | * 64 should be enough to keep us competitive with Linux. | |
60 | */ | |
61 | static int blkif_reqs = 64; | |
62 | module_param_named(reqs, blkif_reqs, int, 0); | |
63 | MODULE_PARM_DESC(reqs, "Number of blkback requests to allocate"); | |
64 | ||
65 | /* Run-time switchable: /sys/module/blkback/parameters/ */ | |
66 | static unsigned int log_stats = 0; | |
67 | static unsigned int debug_lvl = 0; | |
68 | module_param(log_stats, int, 0644); | |
69 | module_param(debug_lvl, int, 0644); | |
70 | ||
71 | /* | |
72 | * Each outstanding request that we've passed to the lower device layers has a | |
73 | * 'pending_req' allocated to it. Each buffer_head that completes decrements | |
74 | * the pendcnt towards zero. When it hits zero, the specified domain has a | |
75 | * response queued for it, with the saved 'id' passed back. | |
76 | */ | |
77 | typedef struct { | |
78 | blkif_t *blkif; | |
79 | u64 id; | |
80 | int nr_pages; | |
81 | atomic_t pendcnt; | |
82 | unsigned short operation; | |
83 | int status; | |
84 | struct list_head free_list; | |
85 | } pending_req_t; | |
86 | ||
87 | static pending_req_t *pending_reqs; | |
88 | static struct list_head pending_free; | |
89 | static DEFINE_SPINLOCK(pending_free_lock); | |
90 | static DECLARE_WAIT_QUEUE_HEAD(pending_free_wq); | |
91 | ||
92 | #define BLKBACK_INVALID_HANDLE (~0) | |
93 | ||
94 | static struct page **pending_pages; | |
95 | static grant_handle_t *pending_grant_handles; | |
96 | ||
97 | static inline int vaddr_pagenr(pending_req_t *req, int seg) | |
98 | { | |
99 | return (req - pending_reqs) * BLKIF_MAX_SEGMENTS_PER_REQUEST + seg; | |
100 | } | |
101 | ||
102 | static inline unsigned long vaddr(pending_req_t *req, int seg) | |
103 | { | |
104 | unsigned long pfn = page_to_pfn(pending_pages[vaddr_pagenr(req, seg)]); | |
105 | return (unsigned long)pfn_to_kaddr(pfn); | |
106 | } | |
107 | ||
108 | #define pending_handle(_req, _seg) \ | |
109 | (pending_grant_handles[vaddr_pagenr(_req, _seg)]) | |
110 | ||
111 | ||
112 | static int do_block_io_op(blkif_t *blkif); | |
113 | static void dispatch_rw_block_io(blkif_t *blkif, | |
88122933 | 114 | struct blkif_request *req, |
4d05a28d KRW |
115 | pending_req_t *pending_req); |
116 | static void make_response(blkif_t *blkif, u64 id, | |
117 | unsigned short op, int st); | |
118 | ||
119 | /****************************************************************** | |
120 | * misc small helpers | |
121 | */ | |
122 | static pending_req_t* alloc_req(void) | |
123 | { | |
124 | pending_req_t *req = NULL; | |
125 | unsigned long flags; | |
126 | ||
127 | spin_lock_irqsave(&pending_free_lock, flags); | |
128 | if (!list_empty(&pending_free)) { | |
129 | req = list_entry(pending_free.next, pending_req_t, free_list); | |
130 | list_del(&req->free_list); | |
131 | } | |
132 | spin_unlock_irqrestore(&pending_free_lock, flags); | |
133 | return req; | |
134 | } | |
135 | ||
136 | static void free_req(pending_req_t *req) | |
137 | { | |
138 | unsigned long flags; | |
139 | int was_empty; | |
140 | ||
141 | spin_lock_irqsave(&pending_free_lock, flags); | |
142 | was_empty = list_empty(&pending_free); | |
143 | list_add(&req->free_list, &pending_free); | |
144 | spin_unlock_irqrestore(&pending_free_lock, flags); | |
145 | if (was_empty) | |
146 | wake_up(&pending_free_wq); | |
147 | } | |
148 | ||
149 | static void unplug_queue(blkif_t *blkif) | |
150 | { | |
151 | if (blkif->plug == NULL) | |
152 | return; | |
153 | if (blkif->plug->unplug_fn) | |
154 | blkif->plug->unplug_fn(blkif->plug); | |
155 | blk_put_queue(blkif->plug); | |
156 | blkif->plug = NULL; | |
157 | } | |
158 | ||
159 | static void plug_queue(blkif_t *blkif, struct block_device *bdev) | |
160 | { | |
88122933 | 161 | struct request_queue *q = bdev_get_queue(bdev); |
4d05a28d KRW |
162 | |
163 | if (q == blkif->plug) | |
164 | return; | |
165 | unplug_queue(blkif); | |
166 | blk_get_queue(q); | |
167 | blkif->plug = q; | |
168 | } | |
169 | ||
170 | static void fast_flush_area(pending_req_t *req) | |
171 | { | |
172 | struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
173 | unsigned int i, invcount = 0; | |
174 | grant_handle_t handle; | |
175 | int ret; | |
176 | ||
177 | for (i = 0; i < req->nr_pages; i++) { | |
178 | handle = pending_handle(req, i); | |
179 | if (handle == BLKBACK_INVALID_HANDLE) | |
180 | continue; | |
181 | gnttab_set_unmap_op(&unmap[invcount], vaddr(req, i), | |
182 | GNTMAP_host_map, handle); | |
183 | pending_handle(req, i) = BLKBACK_INVALID_HANDLE; | |
184 | invcount++; | |
185 | } | |
186 | ||
187 | ret = HYPERVISOR_grant_table_op( | |
188 | GNTTABOP_unmap_grant_ref, unmap, invcount); | |
189 | BUG_ON(ret); | |
190 | } | |
191 | ||
192 | /****************************************************************** | |
193 | * SCHEDULER FUNCTIONS | |
194 | */ | |
195 | ||
196 | static void print_stats(blkif_t *blkif) | |
197 | { | |
198 | printk(KERN_DEBUG "%s: oo %3d | rd %4d | wr %4d | br %4d\n", | |
199 | current->comm, blkif->st_oo_req, | |
200 | blkif->st_rd_req, blkif->st_wr_req, blkif->st_br_req); | |
201 | blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000); | |
202 | blkif->st_rd_req = 0; | |
203 | blkif->st_wr_req = 0; | |
204 | blkif->st_oo_req = 0; | |
205 | } | |
206 | ||
207 | int blkif_schedule(void *arg) | |
208 | { | |
209 | blkif_t *blkif = arg; | |
210 | ||
211 | blkif_get(blkif); | |
212 | ||
213 | if (debug_lvl) | |
214 | printk(KERN_DEBUG "%s: started\n", current->comm); | |
215 | ||
216 | while (!kthread_should_stop()) { | |
217 | if (try_to_freeze()) | |
218 | continue; | |
219 | ||
220 | wait_event_interruptible( | |
221 | blkif->wq, | |
222 | blkif->waiting_reqs || kthread_should_stop()); | |
223 | wait_event_interruptible( | |
224 | pending_free_wq, | |
225 | !list_empty(&pending_free) || kthread_should_stop()); | |
226 | ||
227 | blkif->waiting_reqs = 0; | |
228 | smp_mb(); /* clear flag *before* checking for work */ | |
229 | ||
230 | if (do_block_io_op(blkif)) | |
231 | blkif->waiting_reqs = 1; | |
232 | unplug_queue(blkif); | |
233 | ||
234 | if (log_stats && time_after(jiffies, blkif->st_print)) | |
235 | print_stats(blkif); | |
236 | } | |
237 | ||
238 | if (log_stats) | |
239 | print_stats(blkif); | |
240 | if (debug_lvl) | |
241 | printk(KERN_DEBUG "%s: exiting\n", current->comm); | |
242 | ||
243 | blkif->xenblkd = NULL; | |
244 | blkif_put(blkif); | |
245 | ||
246 | return 0; | |
247 | } | |
248 | ||
249 | /****************************************************************** | |
250 | * COMPLETION CALLBACK -- Called as bh->b_end_io() | |
251 | */ | |
252 | ||
253 | static void __end_block_io_op(pending_req_t *pending_req, int error) | |
254 | { | |
255 | /* An error fails the entire request. */ | |
256 | if ((pending_req->operation == BLKIF_OP_WRITE_BARRIER) && | |
257 | (error == -EOPNOTSUPP)) { | |
258 | DPRINTK("blkback: write barrier op failed, not supported\n"); | |
259 | blkback_barrier(XBT_NIL, pending_req->blkif->be, 0); | |
260 | pending_req->status = BLKIF_RSP_EOPNOTSUPP; | |
261 | } else if (error) { | |
262 | DPRINTK("Buffer not up-to-date at end of operation, " | |
263 | "error=%d\n", error); | |
264 | pending_req->status = BLKIF_RSP_ERROR; | |
265 | } | |
266 | ||
267 | if (atomic_dec_and_test(&pending_req->pendcnt)) { | |
268 | fast_flush_area(pending_req); | |
269 | make_response(pending_req->blkif, pending_req->id, | |
270 | pending_req->operation, pending_req->status); | |
271 | blkif_put(pending_req->blkif); | |
272 | free_req(pending_req); | |
273 | } | |
274 | } | |
275 | ||
88122933 | 276 | static void end_block_io_op(struct bio *bio, int error) |
4d05a28d | 277 | { |
4d05a28d KRW |
278 | __end_block_io_op(bio->bi_private, error); |
279 | bio_put(bio); | |
4d05a28d KRW |
280 | } |
281 | ||
282 | ||
283 | /****************************************************************************** | |
284 | * NOTIFICATION FROM GUEST OS. | |
285 | */ | |
286 | ||
287 | static void blkif_notify_work(blkif_t *blkif) | |
288 | { | |
289 | blkif->waiting_reqs = 1; | |
290 | wake_up(&blkif->wq); | |
291 | } | |
292 | ||
88122933 | 293 | irqreturn_t blkif_be_int(int irq, void *dev_id) |
4d05a28d KRW |
294 | { |
295 | blkif_notify_work(dev_id); | |
296 | return IRQ_HANDLED; | |
297 | } | |
298 | ||
299 | ||
300 | ||
301 | /****************************************************************** | |
302 | * DOWNWARD CALLS -- These interface with the block-device layer proper. | |
303 | */ | |
304 | ||
305 | static int do_block_io_op(blkif_t *blkif) | |
306 | { | |
88122933 JF |
307 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
308 | struct blkif_request req; | |
4d05a28d KRW |
309 | pending_req_t *pending_req; |
310 | RING_IDX rc, rp; | |
311 | int more_to_do = 0; | |
312 | ||
313 | rc = blk_rings->common.req_cons; | |
314 | rp = blk_rings->common.sring->req_prod; | |
315 | rmb(); /* Ensure we see queued requests up to 'rp'. */ | |
316 | ||
317 | while (rc != rp) { | |
318 | ||
319 | if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc)) | |
320 | break; | |
321 | ||
8270b45b | 322 | if (kthread_should_stop()) { |
4d05a28d KRW |
323 | more_to_do = 1; |
324 | break; | |
325 | } | |
326 | ||
8270b45b KF |
327 | pending_req = alloc_req(); |
328 | if (NULL == pending_req) { | |
329 | blkif->st_oo_req++; | |
4d05a28d KRW |
330 | more_to_do = 1; |
331 | break; | |
332 | } | |
333 | ||
334 | switch (blkif->blk_protocol) { | |
335 | case BLKIF_PROTOCOL_NATIVE: | |
336 | memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req)); | |
337 | break; | |
338 | case BLKIF_PROTOCOL_X86_32: | |
339 | blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc)); | |
340 | break; | |
341 | case BLKIF_PROTOCOL_X86_64: | |
342 | blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc)); | |
343 | break; | |
344 | default: | |
345 | BUG(); | |
346 | } | |
347 | blk_rings->common.req_cons = ++rc; /* before make_response() */ | |
348 | ||
349 | /* Apply all sanity checks to /private copy/ of request. */ | |
350 | barrier(); | |
351 | ||
352 | switch (req.operation) { | |
353 | case BLKIF_OP_READ: | |
354 | blkif->st_rd_req++; | |
355 | dispatch_rw_block_io(blkif, &req, pending_req); | |
356 | break; | |
357 | case BLKIF_OP_WRITE_BARRIER: | |
358 | blkif->st_br_req++; | |
359 | /* fall through */ | |
360 | case BLKIF_OP_WRITE: | |
361 | blkif->st_wr_req++; | |
362 | dispatch_rw_block_io(blkif, &req, pending_req); | |
363 | break; | |
364 | default: | |
365 | /* A good sign something is wrong: sleep for a while to | |
366 | * avoid excessive CPU consumption by a bad guest. */ | |
367 | msleep(1); | |
368 | DPRINTK("error: unknown block io operation [%d]\n", | |
369 | req.operation); | |
370 | make_response(blkif, req.id, req.operation, | |
371 | BLKIF_RSP_ERROR); | |
372 | free_req(pending_req); | |
373 | break; | |
374 | } | |
375 | ||
376 | /* Yield point for this unbounded loop. */ | |
377 | cond_resched(); | |
378 | } | |
379 | ||
380 | return more_to_do; | |
381 | } | |
382 | ||
383 | static void dispatch_rw_block_io(blkif_t *blkif, | |
88122933 | 384 | struct blkif_request *req, |
4d05a28d KRW |
385 | pending_req_t *pending_req) |
386 | { | |
4d05a28d KRW |
387 | struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
388 | struct phys_req preq; | |
389 | struct { | |
390 | unsigned long buf; unsigned int nsec; | |
391 | } seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; | |
392 | unsigned int nseg; | |
393 | struct bio *bio = NULL; | |
394 | int ret, i; | |
395 | int operation; | |
396 | ||
397 | switch (req->operation) { | |
398 | case BLKIF_OP_READ: | |
399 | operation = READ; | |
400 | break; | |
401 | case BLKIF_OP_WRITE: | |
402 | operation = WRITE; | |
403 | break; | |
404 | case BLKIF_OP_WRITE_BARRIER: | |
405 | operation = WRITE_BARRIER; | |
406 | break; | |
407 | default: | |
408 | operation = 0; /* make gcc happy */ | |
409 | BUG(); | |
410 | } | |
411 | ||
412 | /* Check that number of segments is sane. */ | |
413 | nseg = req->nr_segments; | |
414 | if (unlikely(nseg == 0 && operation != WRITE_BARRIER) || | |
415 | unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) { | |
416 | DPRINTK("Bad number of segments in request (%d)\n", nseg); | |
417 | goto fail_response; | |
418 | } | |
419 | ||
420 | preq.dev = req->handle; | |
421 | preq.sector_number = req->sector_number; | |
422 | preq.nr_sects = 0; | |
423 | ||
424 | pending_req->blkif = blkif; | |
425 | pending_req->id = req->id; | |
426 | pending_req->operation = req->operation; | |
427 | pending_req->status = BLKIF_RSP_OKAY; | |
428 | pending_req->nr_pages = nseg; | |
429 | ||
430 | for (i = 0; i < nseg; i++) { | |
431 | uint32_t flags; | |
432 | ||
433 | seg[i].nsec = req->seg[i].last_sect - | |
434 | req->seg[i].first_sect + 1; | |
435 | ||
436 | if ((req->seg[i].last_sect >= (PAGE_SIZE >> 9)) || | |
437 | (req->seg[i].last_sect < req->seg[i].first_sect)) | |
438 | goto fail_response; | |
439 | preq.nr_sects += seg[i].nsec; | |
440 | ||
441 | flags = GNTMAP_host_map; | |
442 | if (operation != READ) | |
443 | flags |= GNTMAP_readonly; | |
444 | gnttab_set_map_op(&map[i], vaddr(pending_req, i), flags, | |
445 | req->seg[i].gref, blkif->domid); | |
446 | } | |
447 | ||
448 | ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, map, nseg); | |
449 | BUG_ON(ret); | |
450 | ||
451 | for (i = 0; i < nseg; i++) { | |
452 | if (unlikely(map[i].status != 0)) { | |
453 | DPRINTK("invalid buffer -- could not remap it\n"); | |
454 | map[i].handle = BLKBACK_INVALID_HANDLE; | |
455 | ret |= 1; | |
456 | } | |
457 | ||
458 | pending_handle(pending_req, i) = map[i].handle; | |
459 | ||
460 | if (ret) | |
461 | continue; | |
462 | ||
463 | set_phys_to_machine(__pa(vaddr( | |
464 | pending_req, i)) >> PAGE_SHIFT, | |
465 | FOREIGN_FRAME(map[i].dev_bus_addr >> PAGE_SHIFT)); | |
466 | seg[i].buf = map[i].dev_bus_addr | | |
467 | (req->seg[i].first_sect << 9); | |
468 | } | |
469 | ||
470 | if (ret) | |
471 | goto fail_flush; | |
472 | ||
473 | if (vbd_translate(&preq, blkif, operation) != 0) { | |
474 | DPRINTK("access denied: %s of [%llu,%llu] on dev=%04x\n", | |
475 | operation == READ ? "read" : "write", | |
476 | preq.sector_number, | |
477 | preq.sector_number + preq.nr_sects, preq.dev); | |
478 | goto fail_flush; | |
479 | } | |
480 | ||
481 | plug_queue(blkif, preq.bdev); | |
482 | atomic_set(&pending_req->pendcnt, 1); | |
483 | blkif_get(blkif); | |
484 | ||
485 | for (i = 0; i < nseg; i++) { | |
486 | if (((int)preq.sector_number|(int)seg[i].nsec) & | |
05d43865 | 487 | ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) { |
4d05a28d KRW |
488 | DPRINTK("Misaligned I/O request from domain %d", |
489 | blkif->domid); | |
490 | goto fail_put_bio; | |
491 | } | |
492 | ||
493 | while ((bio == NULL) || | |
494 | (bio_add_page(bio, | |
495 | virt_to_page(vaddr(pending_req, i)), | |
496 | seg[i].nsec << 9, | |
497 | seg[i].buf & ~PAGE_MASK) == 0)) { | |
498 | if (bio) { | |
499 | atomic_inc(&pending_req->pendcnt); | |
500 | submit_bio(operation, bio); | |
501 | } | |
502 | ||
503 | bio = bio_alloc(GFP_KERNEL, nseg-i); | |
504 | if (unlikely(bio == NULL)) | |
505 | goto fail_put_bio; | |
506 | ||
507 | bio->bi_bdev = preq.bdev; | |
508 | bio->bi_private = pending_req; | |
509 | bio->bi_end_io = end_block_io_op; | |
510 | bio->bi_sector = preq.sector_number; | |
511 | } | |
512 | ||
513 | preq.sector_number += seg[i].nsec; | |
514 | } | |
515 | ||
516 | if (!bio) { | |
517 | BUG_ON(operation != WRITE_BARRIER); | |
518 | bio = bio_alloc(GFP_KERNEL, 0); | |
519 | if (unlikely(bio == NULL)) | |
520 | goto fail_put_bio; | |
521 | ||
522 | bio->bi_bdev = preq.bdev; | |
523 | bio->bi_private = pending_req; | |
524 | bio->bi_end_io = end_block_io_op; | |
525 | bio->bi_sector = -1; | |
526 | } | |
527 | ||
528 | submit_bio(operation, bio); | |
529 | ||
530 | if (operation == READ) | |
531 | blkif->st_rd_sect += preq.nr_sects; | |
532 | else if (operation == WRITE || operation == WRITE_BARRIER) | |
533 | blkif->st_wr_sect += preq.nr_sects; | |
534 | ||
535 | return; | |
536 | ||
537 | fail_flush: | |
538 | fast_flush_area(pending_req); | |
539 | fail_response: | |
540 | make_response(blkif, req->id, req->operation, BLKIF_RSP_ERROR); | |
541 | free_req(pending_req); | |
542 | msleep(1); /* back off a bit */ | |
543 | return; | |
544 | ||
545 | fail_put_bio: | |
546 | __end_block_io_op(pending_req, -EINVAL); | |
547 | if (bio) | |
548 | bio_put(bio); | |
549 | unplug_queue(blkif); | |
550 | msleep(1); /* back off a bit */ | |
551 | return; | |
552 | } | |
553 | ||
554 | ||
555 | ||
556 | /****************************************************************** | |
557 | * MISCELLANEOUS SETUP / TEARDOWN / DEBUGGING | |
558 | */ | |
559 | ||
560 | ||
561 | static void make_response(blkif_t *blkif, u64 id, | |
562 | unsigned short op, int st) | |
563 | { | |
88122933 | 564 | struct blkif_response resp; |
4d05a28d | 565 | unsigned long flags; |
88122933 | 566 | union blkif_back_rings *blk_rings = &blkif->blk_rings; |
4d05a28d KRW |
567 | int more_to_do = 0; |
568 | int notify; | |
569 | ||
570 | resp.id = id; | |
571 | resp.operation = op; | |
572 | resp.status = st; | |
573 | ||
574 | spin_lock_irqsave(&blkif->blk_ring_lock, flags); | |
575 | /* Place on the response ring for the relevant domain. */ | |
576 | switch (blkif->blk_protocol) { | |
577 | case BLKIF_PROTOCOL_NATIVE: | |
578 | memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt), | |
579 | &resp, sizeof(resp)); | |
580 | break; | |
581 | case BLKIF_PROTOCOL_X86_32: | |
582 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt), | |
583 | &resp, sizeof(resp)); | |
584 | break; | |
585 | case BLKIF_PROTOCOL_X86_64: | |
586 | memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt), | |
587 | &resp, sizeof(resp)); | |
588 | break; | |
589 | default: | |
590 | BUG(); | |
591 | } | |
592 | blk_rings->common.rsp_prod_pvt++; | |
593 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify); | |
594 | if (blk_rings->common.rsp_prod_pvt == blk_rings->common.req_cons) { | |
595 | /* | |
596 | * Tail check for pending requests. Allows frontend to avoid | |
597 | * notifications if requests are already in flight (lower | |
598 | * overheads and promotes batching). | |
599 | */ | |
600 | RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do); | |
601 | ||
602 | } else if (RING_HAS_UNCONSUMED_REQUESTS(&blk_rings->common)) { | |
603 | more_to_do = 1; | |
604 | } | |
605 | ||
606 | spin_unlock_irqrestore(&blkif->blk_ring_lock, flags); | |
607 | ||
608 | if (more_to_do) | |
609 | blkif_notify_work(blkif); | |
610 | if (notify) | |
611 | notify_remote_via_irq(blkif->irq); | |
612 | } | |
613 | ||
614 | static int __init blkif_init(void) | |
615 | { | |
616 | int i, mmap_pages; | |
617 | ||
88122933 | 618 | if (!xen_pv_domain()) |
4d05a28d KRW |
619 | return -ENODEV; |
620 | ||
621 | mmap_pages = blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST; | |
622 | ||
623 | pending_reqs = kmalloc(sizeof(pending_reqs[0]) * | |
624 | blkif_reqs, GFP_KERNEL); | |
625 | pending_grant_handles = kmalloc(sizeof(pending_grant_handles[0]) * | |
626 | mmap_pages, GFP_KERNEL); | |
627 | pending_pages = alloc_empty_pages_and_pagevec(mmap_pages); | |
628 | ||
629 | if (!pending_reqs || !pending_grant_handles || !pending_pages) | |
630 | goto out_of_memory; | |
631 | ||
632 | for (i = 0; i < mmap_pages; i++) | |
633 | pending_grant_handles[i] = BLKBACK_INVALID_HANDLE; | |
634 | ||
635 | blkif_interface_init(); | |
636 | ||
637 | memset(pending_reqs, 0, sizeof(pending_reqs)); | |
638 | INIT_LIST_HEAD(&pending_free); | |
639 | ||
640 | for (i = 0; i < blkif_reqs; i++) | |
641 | list_add_tail(&pending_reqs[i].free_list, &pending_free); | |
642 | ||
643 | blkif_xenbus_init(); | |
644 | ||
645 | return 0; | |
646 | ||
647 | out_of_memory: | |
648 | kfree(pending_reqs); | |
649 | kfree(pending_grant_handles); | |
650 | free_empty_pages_and_pagevec(pending_pages, mmap_pages); | |
651 | printk("%s: out of memory\n", __FUNCTION__); | |
652 | return -ENOMEM; | |
653 | } | |
654 | ||
655 | module_init(blkif_init); | |
656 | ||
657 | MODULE_LICENSE("Dual BSD/GPL"); |