| 1 | /* |
| 2 | * Copyright (c) 2018 Citrix Systems Inc. |
| 3 | * (c) Gerd Hoffmann <kraxel@redhat.com> |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or modify |
| 6 | * it under the terms of the GNU General Public License as published by |
| 7 | * the Free Software Foundation; under version 2 of the License. |
| 8 | * |
| 9 | * This program is distributed in the hope that it will be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
| 13 | * |
| 14 | * You should have received a copy of the GNU General Public License along |
| 15 | * with this program; if not, see <http://www.gnu.org/licenses/>. |
| 16 | * |
| 17 | * Contributions after 2012-01-13 are licensed under the terms of the |
| 18 | * GNU GPL, version 2 or (at your option) any later version. |
| 19 | */ |
| 20 | |
| 21 | #include "qemu/osdep.h" |
| 22 | #include "qemu/defer-call.h" |
| 23 | #include "qemu/error-report.h" |
| 24 | #include "qemu/main-loop.h" |
| 25 | #include "qemu/memalign.h" |
| 26 | #include "qapi/error.h" |
| 27 | #include "hw/xen/xen.h" |
| 28 | #include "hw/block/xen_blkif.h" |
| 29 | #include "hw/xen/interface/io/ring.h" |
| 30 | #include "system/block-backend.h" |
| 31 | #include "system/iothread.h" |
| 32 | #include "xen-block.h" |
| 33 | |
| 34 | typedef struct XenBlockRequest { |
| 35 | blkif_request_t req; |
| 36 | int16_t status; |
| 37 | off_t start; |
| 38 | QEMUIOVector v; |
| 39 | void *buf; |
| 40 | size_t size; |
| 41 | int presync; |
| 42 | int aio_inflight; |
| 43 | int aio_errors; |
| 44 | XenBlockDataPlane *dataplane; |
| 45 | QLIST_ENTRY(XenBlockRequest) list; |
| 46 | BlockAcctCookie acct; |
| 47 | } XenBlockRequest; |
| 48 | |
| 49 | struct XenBlockDataPlane { |
| 50 | XenDevice *xendev; |
| 51 | XenEventChannel *event_channel; |
| 52 | unsigned int *ring_ref; |
| 53 | unsigned int nr_ring_ref; |
| 54 | void *sring; |
| 55 | int protocol; |
| 56 | blkif_back_rings_t rings; |
| 57 | int more_work; |
| 58 | QLIST_HEAD(inflight_head, XenBlockRequest) inflight; |
| 59 | QLIST_HEAD(freelist_head, XenBlockRequest) freelist; |
| 60 | int requests_total; |
| 61 | int requests_inflight; |
| 62 | unsigned int max_requests; |
| 63 | BlockBackend *blk; |
| 64 | unsigned int sector_size; |
| 65 | QEMUBH *bh; |
| 66 | IOThread *iothread; |
| 67 | AioContext *ctx; |
| 68 | }; |
| 69 | |
| 70 | static int xen_block_send_response(XenBlockRequest *request); |
| 71 | |
| 72 | static void reset_request(XenBlockRequest *request) |
| 73 | { |
| 74 | memset(&request->req, 0, sizeof(request->req)); |
| 75 | request->status = 0; |
| 76 | request->start = 0; |
| 77 | request->size = 0; |
| 78 | request->presync = 0; |
| 79 | |
| 80 | request->aio_inflight = 0; |
| 81 | request->aio_errors = 0; |
| 82 | |
| 83 | request->dataplane = NULL; |
| 84 | memset(&request->list, 0, sizeof(request->list)); |
| 85 | memset(&request->acct, 0, sizeof(request->acct)); |
| 86 | |
| 87 | qemu_iovec_reset(&request->v); |
| 88 | } |
| 89 | |
| 90 | static XenBlockRequest *xen_block_start_request(XenBlockDataPlane *dataplane) |
| 91 | { |
| 92 | XenBlockRequest *request = NULL; |
| 93 | |
| 94 | if (QLIST_EMPTY(&dataplane->freelist)) { |
| 95 | if (dataplane->requests_total >= dataplane->max_requests) { |
| 96 | goto out; |
| 97 | } |
| 98 | /* allocate new struct */ |
| 99 | request = g_malloc0(sizeof(*request)); |
| 100 | request->dataplane = dataplane; |
| 101 | /* |
| 102 | * We cannot need more pages per requests than this, and since we |
| 103 | * re-use requests, allocate the memory once here. It will be freed |
| 104 | * xen_block_dataplane_destroy() when the request list is freed. |
| 105 | */ |
| 106 | request->buf = qemu_memalign(XEN_PAGE_SIZE, |
| 107 | BLKIF_MAX_SEGMENTS_PER_REQUEST * |
| 108 | XEN_PAGE_SIZE); |
| 109 | dataplane->requests_total++; |
| 110 | qemu_iovec_init(&request->v, 1); |
| 111 | } else { |
| 112 | /* get one from freelist */ |
| 113 | request = QLIST_FIRST(&dataplane->freelist); |
| 114 | QLIST_REMOVE(request, list); |
| 115 | } |
| 116 | QLIST_INSERT_HEAD(&dataplane->inflight, request, list); |
| 117 | dataplane->requests_inflight++; |
| 118 | |
| 119 | out: |
| 120 | return request; |
| 121 | } |
| 122 | |
| 123 | static void xen_block_complete_request(XenBlockRequest *request) |
| 124 | { |
| 125 | XenBlockDataPlane *dataplane = request->dataplane; |
| 126 | |
| 127 | if (xen_block_send_response(request)) { |
| 128 | Error *local_err = NULL; |
| 129 | |
| 130 | xen_device_notify_event_channel(dataplane->xendev, |
| 131 | dataplane->event_channel, |
| 132 | &local_err); |
| 133 | if (local_err) { |
| 134 | error_report_err(local_err); |
| 135 | } |
| 136 | } |
| 137 | |
| 138 | QLIST_REMOVE(request, list); |
| 139 | dataplane->requests_inflight--; |
| 140 | reset_request(request); |
| 141 | request->dataplane = dataplane; |
| 142 | QLIST_INSERT_HEAD(&dataplane->freelist, request, list); |
| 143 | } |
| 144 | |
| 145 | /* |
| 146 | * translate request into iovec + start offset |
| 147 | * do sanity checks along the way |
| 148 | */ |
| 149 | static int xen_block_parse_request(XenBlockRequest *request) |
| 150 | { |
| 151 | XenBlockDataPlane *dataplane = request->dataplane; |
| 152 | size_t len; |
| 153 | int i; |
| 154 | |
| 155 | switch (request->req.operation) { |
| 156 | case BLKIF_OP_READ: |
| 157 | break; |
| 158 | case BLKIF_OP_FLUSH_DISKCACHE: |
| 159 | request->presync = 1; |
| 160 | if (!request->req.nr_segments) { |
| 161 | return 0; |
| 162 | } |
| 163 | /* fall through */ |
| 164 | case BLKIF_OP_WRITE: |
| 165 | break; |
| 166 | case BLKIF_OP_DISCARD: |
| 167 | return 0; |
| 168 | default: |
| 169 | error_report("error: unknown operation (%d)", request->req.operation); |
| 170 | goto err; |
| 171 | }; |
| 172 | |
| 173 | if (request->req.operation != BLKIF_OP_READ && |
| 174 | !blk_is_writable(dataplane->blk)) { |
| 175 | error_report("error: write req for ro device"); |
| 176 | goto err; |
| 177 | } |
| 178 | |
| 179 | request->start = request->req.sector_number * dataplane->sector_size; |
| 180 | for (i = 0; i < request->req.nr_segments; i++) { |
| 181 | if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) { |
| 182 | error_report("error: nr_segments too big"); |
| 183 | goto err; |
| 184 | } |
| 185 | if (request->req.seg[i].first_sect > request->req.seg[i].last_sect) { |
| 186 | error_report("error: first > last sector"); |
| 187 | goto err; |
| 188 | } |
| 189 | if (request->req.seg[i].last_sect * dataplane->sector_size >= |
| 190 | XEN_PAGE_SIZE) { |
| 191 | error_report("error: page crossing"); |
| 192 | goto err; |
| 193 | } |
| 194 | |
| 195 | len = (request->req.seg[i].last_sect - |
| 196 | request->req.seg[i].first_sect + 1) * dataplane->sector_size; |
| 197 | request->size += len; |
| 198 | } |
| 199 | if (request->start + request->size > blk_getlength(dataplane->blk)) { |
| 200 | error_report("error: access beyond end of file"); |
| 201 | goto err; |
| 202 | } |
| 203 | return 0; |
| 204 | |
| 205 | err: |
| 206 | request->status = BLKIF_RSP_ERROR; |
| 207 | return -1; |
| 208 | } |
| 209 | |
| 210 | static int xen_block_copy_request(XenBlockRequest *request) |
| 211 | { |
| 212 | XenBlockDataPlane *dataplane = request->dataplane; |
| 213 | XenDevice *xendev = dataplane->xendev; |
| 214 | XenDeviceGrantCopySegment segs[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
| 215 | int i, count; |
| 216 | bool to_domain = (request->req.operation == BLKIF_OP_READ); |
| 217 | void *virt = request->buf; |
| 218 | Error *local_err = NULL; |
| 219 | |
| 220 | if (request->req.nr_segments == 0) { |
| 221 | return 0; |
| 222 | } |
| 223 | |
| 224 | count = request->req.nr_segments; |
| 225 | |
| 226 | for (i = 0; i < count; i++) { |
| 227 | if (to_domain) { |
| 228 | segs[i].dest.foreign.ref = request->req.seg[i].gref; |
| 229 | segs[i].dest.foreign.offset = request->req.seg[i].first_sect * |
| 230 | dataplane->sector_size; |
| 231 | segs[i].source.virt = virt; |
| 232 | } else { |
| 233 | segs[i].source.foreign.ref = request->req.seg[i].gref; |
| 234 | segs[i].source.foreign.offset = request->req.seg[i].first_sect * |
| 235 | dataplane->sector_size; |
| 236 | segs[i].dest.virt = virt; |
| 237 | } |
| 238 | segs[i].len = (request->req.seg[i].last_sect - |
| 239 | request->req.seg[i].first_sect + 1) * |
| 240 | dataplane->sector_size; |
| 241 | virt += segs[i].len; |
| 242 | } |
| 243 | |
| 244 | xen_device_copy_grant_refs(xendev, to_domain, segs, count, &local_err); |
| 245 | |
| 246 | if (local_err) { |
| 247 | error_reportf_err(local_err, "failed to copy data: "); |
| 248 | |
| 249 | request->aio_errors++; |
| 250 | return -1; |
| 251 | } |
| 252 | |
| 253 | return 0; |
| 254 | } |
| 255 | |
| 256 | static int xen_block_do_aio(XenBlockRequest *request); |
| 257 | |
| 258 | static void xen_block_complete_aio(void *opaque, int ret) |
| 259 | { |
| 260 | XenBlockRequest *request = opaque; |
| 261 | XenBlockDataPlane *dataplane = request->dataplane; |
| 262 | |
| 263 | if (ret != 0) { |
| 264 | error_report("%s I/O error", |
| 265 | request->req.operation == BLKIF_OP_READ ? |
| 266 | "read" : "write"); |
| 267 | request->aio_errors++; |
| 268 | } |
| 269 | |
| 270 | request->aio_inflight--; |
| 271 | if (request->presync) { |
| 272 | request->presync = 0; |
| 273 | xen_block_do_aio(request); |
| 274 | return; |
| 275 | } |
| 276 | if (request->aio_inflight > 0) { |
| 277 | return; |
| 278 | } |
| 279 | |
| 280 | switch (request->req.operation) { |
| 281 | case BLKIF_OP_READ: |
| 282 | /* in case of failure request->aio_errors is increased */ |
| 283 | if (ret == 0) { |
| 284 | xen_block_copy_request(request); |
| 285 | } |
| 286 | break; |
| 287 | case BLKIF_OP_WRITE: |
| 288 | case BLKIF_OP_FLUSH_DISKCACHE: |
| 289 | default: |
| 290 | break; |
| 291 | } |
| 292 | |
| 293 | request->status = request->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY; |
| 294 | |
| 295 | switch (request->req.operation) { |
| 296 | case BLKIF_OP_WRITE: |
| 297 | case BLKIF_OP_FLUSH_DISKCACHE: |
| 298 | if (!request->req.nr_segments) { |
| 299 | break; |
| 300 | } |
| 301 | /* fall through */ |
| 302 | case BLKIF_OP_READ: |
| 303 | if (request->status == BLKIF_RSP_OKAY) { |
| 304 | block_acct_done(blk_get_stats(dataplane->blk), &request->acct); |
| 305 | } else { |
| 306 | block_acct_failed(blk_get_stats(dataplane->blk), &request->acct); |
| 307 | } |
| 308 | break; |
| 309 | case BLKIF_OP_DISCARD: |
| 310 | default: |
| 311 | break; |
| 312 | } |
| 313 | |
| 314 | xen_block_complete_request(request); |
| 315 | |
| 316 | if (dataplane->more_work) { |
| 317 | qemu_bh_schedule(dataplane->bh); |
| 318 | } |
| 319 | } |
| 320 | |
| 321 | static bool xen_block_split_discard(XenBlockRequest *request, |
| 322 | blkif_sector_t sector_number, |
| 323 | uint64_t nr_sectors) |
| 324 | { |
| 325 | XenBlockDataPlane *dataplane = request->dataplane; |
| 326 | int64_t byte_offset; |
| 327 | int byte_chunk; |
| 328 | uint64_t byte_remaining; |
| 329 | uint64_t sec_start = sector_number; |
| 330 | uint64_t sec_count = nr_sectors; |
| 331 | |
| 332 | /* Wrap around, or overflowing byte limit? */ |
| 333 | if (sec_start + sec_count < sec_count || |
| 334 | sec_start + sec_count > INT64_MAX / dataplane->sector_size) { |
| 335 | return false; |
| 336 | } |
| 337 | |
| 338 | byte_offset = sec_start * dataplane->sector_size; |
| 339 | byte_remaining = sec_count * dataplane->sector_size; |
| 340 | |
| 341 | do { |
| 342 | byte_chunk = byte_remaining > BDRV_REQUEST_MAX_BYTES ? |
| 343 | BDRV_REQUEST_MAX_BYTES : byte_remaining; |
| 344 | request->aio_inflight++; |
| 345 | blk_aio_pdiscard(dataplane->blk, byte_offset, byte_chunk, |
| 346 | xen_block_complete_aio, request); |
| 347 | byte_remaining -= byte_chunk; |
| 348 | byte_offset += byte_chunk; |
| 349 | } while (byte_remaining > 0); |
| 350 | |
| 351 | return true; |
| 352 | } |
| 353 | |
| 354 | static int xen_block_do_aio(XenBlockRequest *request) |
| 355 | { |
| 356 | XenBlockDataPlane *dataplane = request->dataplane; |
| 357 | |
| 358 | if (request->req.nr_segments && |
| 359 | (request->req.operation == BLKIF_OP_WRITE || |
| 360 | request->req.operation == BLKIF_OP_FLUSH_DISKCACHE) && |
| 361 | xen_block_copy_request(request)) { |
| 362 | goto err; |
| 363 | } |
| 364 | |
| 365 | request->aio_inflight++; |
| 366 | if (request->presync) { |
| 367 | blk_aio_flush(request->dataplane->blk, xen_block_complete_aio, |
| 368 | request); |
| 369 | return 0; |
| 370 | } |
| 371 | |
| 372 | switch (request->req.operation) { |
| 373 | case BLKIF_OP_READ: |
| 374 | qemu_iovec_add(&request->v, request->buf, request->size); |
| 375 | block_acct_start(blk_get_stats(dataplane->blk), &request->acct, |
| 376 | request->v.size, BLOCK_ACCT_READ); |
| 377 | request->aio_inflight++; |
| 378 | blk_aio_preadv(dataplane->blk, request->start, &request->v, 0, |
| 379 | xen_block_complete_aio, request); |
| 380 | break; |
| 381 | case BLKIF_OP_WRITE: |
| 382 | case BLKIF_OP_FLUSH_DISKCACHE: |
| 383 | if (!request->req.nr_segments) { |
| 384 | break; |
| 385 | } |
| 386 | |
| 387 | qemu_iovec_add(&request->v, request->buf, request->size); |
| 388 | block_acct_start(blk_get_stats(dataplane->blk), &request->acct, |
| 389 | request->v.size, |
| 390 | request->req.operation == BLKIF_OP_WRITE ? |
| 391 | BLOCK_ACCT_WRITE : BLOCK_ACCT_FLUSH); |
| 392 | request->aio_inflight++; |
| 393 | blk_aio_pwritev(dataplane->blk, request->start, &request->v, 0, |
| 394 | xen_block_complete_aio, request); |
| 395 | break; |
| 396 | case BLKIF_OP_DISCARD: |
| 397 | { |
| 398 | struct blkif_request_discard *req = (void *)&request->req; |
| 399 | if (!xen_block_split_discard(request, req->sector_number, |
| 400 | req->nr_sectors)) { |
| 401 | goto err; |
| 402 | } |
| 403 | break; |
| 404 | } |
| 405 | default: |
| 406 | /* unknown operation (shouldn't happen -- parse catches this) */ |
| 407 | goto err; |
| 408 | } |
| 409 | |
| 410 | xen_block_complete_aio(request, 0); |
| 411 | |
| 412 | return 0; |
| 413 | |
| 414 | err: |
| 415 | request->status = BLKIF_RSP_ERROR; |
| 416 | xen_block_complete_request(request); |
| 417 | return -1; |
| 418 | } |
| 419 | |
| 420 | static int xen_block_send_response(XenBlockRequest *request) |
| 421 | { |
| 422 | XenBlockDataPlane *dataplane = request->dataplane; |
| 423 | int send_notify = 0; |
| 424 | int have_requests = 0; |
| 425 | blkif_response_t *resp; |
| 426 | |
| 427 | /* Place on the response ring for the relevant domain. */ |
| 428 | switch (dataplane->protocol) { |
| 429 | case BLKIF_PROTOCOL_NATIVE: |
| 430 | resp = (blkif_response_t *)RING_GET_RESPONSE( |
| 431 | &dataplane->rings.native, |
| 432 | dataplane->rings.native.rsp_prod_pvt); |
| 433 | break; |
| 434 | case BLKIF_PROTOCOL_X86_32: |
| 435 | resp = (blkif_response_t *)RING_GET_RESPONSE( |
| 436 | &dataplane->rings.x86_32_part, |
| 437 | dataplane->rings.x86_32_part.rsp_prod_pvt); |
| 438 | break; |
| 439 | case BLKIF_PROTOCOL_X86_64: |
| 440 | resp = (blkif_response_t *)RING_GET_RESPONSE( |
| 441 | &dataplane->rings.x86_64_part, |
| 442 | dataplane->rings.x86_64_part.rsp_prod_pvt); |
| 443 | break; |
| 444 | default: |
| 445 | return 0; |
| 446 | } |
| 447 | |
| 448 | resp->id = request->req.id; |
| 449 | resp->operation = request->req.operation; |
| 450 | resp->status = request->status; |
| 451 | |
| 452 | dataplane->rings.common.rsp_prod_pvt++; |
| 453 | |
| 454 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&dataplane->rings.common, |
| 455 | send_notify); |
| 456 | if (dataplane->rings.common.rsp_prod_pvt == |
| 457 | dataplane->rings.common.req_cons) { |
| 458 | /* |
| 459 | * Tail check for pending requests. Allows frontend to avoid |
| 460 | * notifications if requests are already in flight (lower |
| 461 | * overheads and promotes batching). |
| 462 | */ |
| 463 | RING_FINAL_CHECK_FOR_REQUESTS(&dataplane->rings.common, |
| 464 | have_requests); |
| 465 | } else if (RING_HAS_UNCONSUMED_REQUESTS(&dataplane->rings.common)) { |
| 466 | have_requests = 1; |
| 467 | } |
| 468 | |
| 469 | if (have_requests) { |
| 470 | dataplane->more_work++; |
| 471 | } |
| 472 | return send_notify; |
| 473 | } |
| 474 | |
| 475 | static int xen_block_get_request(XenBlockDataPlane *dataplane, |
| 476 | XenBlockRequest *request, RING_IDX rc) |
| 477 | { |
| 478 | switch (dataplane->protocol) { |
| 479 | case BLKIF_PROTOCOL_NATIVE: { |
| 480 | blkif_request_t *req = |
| 481 | RING_GET_REQUEST(&dataplane->rings.native, rc); |
| 482 | |
| 483 | memcpy(&request->req, req, sizeof(request->req)); |
| 484 | break; |
| 485 | } |
| 486 | case BLKIF_PROTOCOL_X86_32: { |
| 487 | blkif_x86_32_request_t *req = |
| 488 | RING_GET_REQUEST(&dataplane->rings.x86_32_part, rc); |
| 489 | |
| 490 | blkif_get_x86_32_req(&request->req, req); |
| 491 | break; |
| 492 | } |
| 493 | case BLKIF_PROTOCOL_X86_64: { |
| 494 | blkif_x86_64_request_t *req = |
| 495 | RING_GET_REQUEST(&dataplane->rings.x86_64_part, rc); |
| 496 | |
| 497 | blkif_get_x86_64_req(&request->req, req); |
| 498 | break; |
| 499 | } |
| 500 | } |
| 501 | /* Prevent the compiler from accessing the on-ring fields instead. */ |
| 502 | barrier(); |
| 503 | return 0; |
| 504 | } |
| 505 | |
| 506 | /* |
| 507 | * Threshold of in-flight requests above which we will start using |
| 508 | * defer_call_begin()/defer_call_end() to batch requests. |
| 509 | */ |
| 510 | #define IO_PLUG_THRESHOLD 1 |
| 511 | |
| 512 | static bool xen_block_handle_requests(XenBlockDataPlane *dataplane) |
| 513 | { |
| 514 | RING_IDX rc, rp; |
| 515 | XenBlockRequest *request; |
| 516 | int inflight_atstart = dataplane->requests_inflight; |
| 517 | int batched = 0; |
| 518 | bool done_something = false; |
| 519 | |
| 520 | dataplane->more_work = 0; |
| 521 | |
| 522 | rc = dataplane->rings.common.req_cons; |
| 523 | rp = dataplane->rings.common.sring->req_prod; |
| 524 | xen_rmb(); /* Ensure we see queued requests up to 'rp'. */ |
| 525 | |
| 526 | /* |
| 527 | * If there was more than IO_PLUG_THRESHOLD requests in flight |
| 528 | * when we got here, this is an indication that there the bottleneck |
| 529 | * is below us, so it's worth beginning to batch up I/O requests |
| 530 | * rather than submitting them immediately. The maximum number |
| 531 | * of requests we're willing to batch is the number already in |
| 532 | * flight, so it can grow up to max_requests when the bottleneck |
| 533 | * is below us. |
| 534 | */ |
| 535 | if (inflight_atstart > IO_PLUG_THRESHOLD) { |
| 536 | defer_call_begin(); |
| 537 | } |
| 538 | while (rc != rp) { |
| 539 | /* pull request from ring */ |
| 540 | if (RING_REQUEST_CONS_OVERFLOW(&dataplane->rings.common, rc)) { |
| 541 | break; |
| 542 | } |
| 543 | request = xen_block_start_request(dataplane); |
| 544 | if (request == NULL) { |
| 545 | dataplane->more_work++; |
| 546 | break; |
| 547 | } |
| 548 | xen_block_get_request(dataplane, request, rc); |
| 549 | dataplane->rings.common.req_cons = ++rc; |
| 550 | done_something = true; |
| 551 | |
| 552 | /* parse them */ |
| 553 | if (xen_block_parse_request(request) != 0) { |
| 554 | switch (request->req.operation) { |
| 555 | case BLKIF_OP_READ: |
| 556 | block_acct_invalid(blk_get_stats(dataplane->blk), |
| 557 | BLOCK_ACCT_READ); |
| 558 | break; |
| 559 | case BLKIF_OP_WRITE: |
| 560 | block_acct_invalid(blk_get_stats(dataplane->blk), |
| 561 | BLOCK_ACCT_WRITE); |
| 562 | break; |
| 563 | case BLKIF_OP_FLUSH_DISKCACHE: |
| 564 | block_acct_invalid(blk_get_stats(dataplane->blk), |
| 565 | BLOCK_ACCT_FLUSH); |
| 566 | default: |
| 567 | break; |
| 568 | }; |
| 569 | |
| 570 | xen_block_complete_request(request); |
| 571 | continue; |
| 572 | } |
| 573 | |
| 574 | if (inflight_atstart > IO_PLUG_THRESHOLD && |
| 575 | batched >= inflight_atstart) { |
| 576 | defer_call_end(); |
| 577 | } |
| 578 | xen_block_do_aio(request); |
| 579 | if (inflight_atstart > IO_PLUG_THRESHOLD) { |
| 580 | if (batched >= inflight_atstart) { |
| 581 | defer_call_begin(); |
| 582 | batched = 0; |
| 583 | } else { |
| 584 | batched++; |
| 585 | } |
| 586 | } |
| 587 | } |
| 588 | if (inflight_atstart > IO_PLUG_THRESHOLD) { |
| 589 | defer_call_end(); |
| 590 | } |
| 591 | |
| 592 | return done_something; |
| 593 | } |
| 594 | |
| 595 | static void xen_block_dataplane_bh(void *opaque) |
| 596 | { |
| 597 | XenBlockDataPlane *dataplane = opaque; |
| 598 | |
| 599 | xen_block_handle_requests(dataplane); |
| 600 | } |
| 601 | |
| 602 | static bool xen_block_dataplane_event(void *opaque) |
| 603 | { |
| 604 | XenBlockDataPlane *dataplane = opaque; |
| 605 | |
| 606 | return xen_block_handle_requests(dataplane); |
| 607 | } |
| 608 | |
| 609 | XenBlockDataPlane *xen_block_dataplane_create(XenDevice *xendev, |
| 610 | BlockBackend *blk, |
| 611 | unsigned int sector_size, |
| 612 | IOThread *iothread) |
| 613 | { |
| 614 | XenBlockDataPlane *dataplane = g_new0(XenBlockDataPlane, 1); |
| 615 | |
| 616 | dataplane->xendev = xendev; |
| 617 | dataplane->blk = blk; |
| 618 | dataplane->sector_size = sector_size; |
| 619 | |
| 620 | QLIST_INIT(&dataplane->inflight); |
| 621 | QLIST_INIT(&dataplane->freelist); |
| 622 | |
| 623 | if (iothread) { |
| 624 | dataplane->iothread = iothread; |
| 625 | object_ref(OBJECT(dataplane->iothread)); |
| 626 | dataplane->ctx = iothread_get_aio_context(dataplane->iothread); |
| 627 | } else { |
| 628 | dataplane->ctx = qemu_get_aio_context(); |
| 629 | } |
| 630 | dataplane->bh = aio_bh_new_guarded(dataplane->ctx, xen_block_dataplane_bh, |
| 631 | dataplane, |
| 632 | &DEVICE(xendev)->mem_reentrancy_guard); |
| 633 | |
| 634 | return dataplane; |
| 635 | } |
| 636 | |
| 637 | void xen_block_dataplane_destroy(XenBlockDataPlane *dataplane) |
| 638 | { |
| 639 | XenBlockRequest *request; |
| 640 | |
| 641 | if (!dataplane) { |
| 642 | return; |
| 643 | } |
| 644 | |
| 645 | while (!QLIST_EMPTY(&dataplane->freelist)) { |
| 646 | request = QLIST_FIRST(&dataplane->freelist); |
| 647 | QLIST_REMOVE(request, list); |
| 648 | qemu_iovec_destroy(&request->v); |
| 649 | qemu_vfree(request->buf); |
| 650 | g_free(request); |
| 651 | } |
| 652 | |
| 653 | qemu_bh_delete(dataplane->bh); |
| 654 | if (dataplane->iothread) { |
| 655 | object_unref(OBJECT(dataplane->iothread)); |
| 656 | } |
| 657 | |
| 658 | g_free(dataplane); |
| 659 | } |
| 660 | |
| 661 | void xen_block_dataplane_detach(XenBlockDataPlane *dataplane) |
| 662 | { |
| 663 | if (!dataplane || !dataplane->event_channel) { |
| 664 | return; |
| 665 | } |
| 666 | |
| 667 | /* Only reason for failure is a NULL channel */ |
| 668 | xen_device_set_event_channel_context(dataplane->xendev, |
| 669 | dataplane->event_channel, |
| 670 | NULL, &error_abort); |
| 671 | } |
| 672 | |
| 673 | void xen_block_dataplane_attach(XenBlockDataPlane *dataplane) |
| 674 | { |
| 675 | if (!dataplane || !dataplane->event_channel) { |
| 676 | return; |
| 677 | } |
| 678 | |
| 679 | /* Only reason for failure is a NULL channel */ |
| 680 | xen_device_set_event_channel_context(dataplane->xendev, |
| 681 | dataplane->event_channel, |
| 682 | dataplane->ctx, &error_abort); |
| 683 | } |
| 684 | |
| 685 | void xen_block_dataplane_stop(XenBlockDataPlane *dataplane) |
| 686 | { |
| 687 | XenDevice *xendev; |
| 688 | |
| 689 | if (!dataplane) { |
| 690 | return; |
| 691 | } |
| 692 | |
| 693 | xendev = dataplane->xendev; |
| 694 | |
| 695 | if (!blk_in_drain(dataplane->blk)) { |
| 696 | xen_block_dataplane_detach(dataplane); |
| 697 | } |
| 698 | |
| 699 | /* Xen doesn't have multiple users for nodes, so this can't fail */ |
| 700 | blk_set_aio_context(dataplane->blk, qemu_get_aio_context(), &error_abort); |
| 701 | |
| 702 | /* |
| 703 | * Now that the context has been moved onto the main thread, cancel |
| 704 | * further processing. |
| 705 | */ |
| 706 | qemu_bh_cancel(dataplane->bh); |
| 707 | |
| 708 | if (dataplane->event_channel) { |
| 709 | Error *local_err = NULL; |
| 710 | |
| 711 | xen_device_unbind_event_channel(xendev, dataplane->event_channel, |
| 712 | &local_err); |
| 713 | dataplane->event_channel = NULL; |
| 714 | |
| 715 | if (local_err) { |
| 716 | error_report_err(local_err); |
| 717 | } |
| 718 | } |
| 719 | |
| 720 | if (dataplane->sring) { |
| 721 | Error *local_err = NULL; |
| 722 | |
| 723 | xen_device_unmap_grant_refs(xendev, dataplane->sring, |
| 724 | dataplane->ring_ref, |
| 725 | dataplane->nr_ring_ref, &local_err); |
| 726 | dataplane->sring = NULL; |
| 727 | |
| 728 | if (local_err) { |
| 729 | error_report_err(local_err); |
| 730 | } |
| 731 | } |
| 732 | |
| 733 | g_free(dataplane->ring_ref); |
| 734 | dataplane->ring_ref = NULL; |
| 735 | } |
| 736 | |
| 737 | void xen_block_dataplane_start(XenBlockDataPlane *dataplane, |
| 738 | const unsigned int ring_ref[], |
| 739 | unsigned int nr_ring_ref, |
| 740 | unsigned int event_channel, |
| 741 | unsigned int protocol, |
| 742 | Error **errp) |
| 743 | { |
| 744 | ERRP_GUARD(); |
| 745 | XenDevice *xendev = dataplane->xendev; |
| 746 | unsigned int ring_size; |
| 747 | unsigned int i; |
| 748 | |
| 749 | dataplane->nr_ring_ref = nr_ring_ref; |
| 750 | dataplane->ring_ref = g_new(unsigned int, nr_ring_ref); |
| 751 | |
| 752 | for (i = 0; i < nr_ring_ref; i++) { |
| 753 | dataplane->ring_ref[i] = ring_ref[i]; |
| 754 | } |
| 755 | |
| 756 | dataplane->protocol = protocol; |
| 757 | |
| 758 | ring_size = XEN_PAGE_SIZE * dataplane->nr_ring_ref; |
| 759 | switch (dataplane->protocol) { |
| 760 | case BLKIF_PROTOCOL_NATIVE: |
| 761 | { |
| 762 | dataplane->max_requests = __CONST_RING_SIZE(blkif, ring_size); |
| 763 | break; |
| 764 | } |
| 765 | case BLKIF_PROTOCOL_X86_32: |
| 766 | { |
| 767 | dataplane->max_requests = __CONST_RING_SIZE(blkif_x86_32, ring_size); |
| 768 | break; |
| 769 | } |
| 770 | case BLKIF_PROTOCOL_X86_64: |
| 771 | { |
| 772 | dataplane->max_requests = __CONST_RING_SIZE(blkif_x86_64, ring_size); |
| 773 | break; |
| 774 | } |
| 775 | default: |
| 776 | error_setg(errp, "unknown protocol %u", dataplane->protocol); |
| 777 | return; |
| 778 | } |
| 779 | |
| 780 | xen_device_set_max_grant_refs(xendev, dataplane->nr_ring_ref, |
| 781 | errp); |
| 782 | if (*errp) { |
| 783 | goto stop; |
| 784 | } |
| 785 | |
| 786 | dataplane->sring = xen_device_map_grant_refs(xendev, |
| 787 | dataplane->ring_ref, |
| 788 | dataplane->nr_ring_ref, |
| 789 | PROT_READ | PROT_WRITE, |
| 790 | errp); |
| 791 | if (*errp) { |
| 792 | goto stop; |
| 793 | } |
| 794 | |
| 795 | switch (dataplane->protocol) { |
| 796 | case BLKIF_PROTOCOL_NATIVE: |
| 797 | { |
| 798 | blkif_sring_t *sring_native = dataplane->sring; |
| 799 | |
| 800 | BACK_RING_INIT(&dataplane->rings.native, sring_native, ring_size); |
| 801 | break; |
| 802 | } |
| 803 | case BLKIF_PROTOCOL_X86_32: |
| 804 | { |
| 805 | blkif_x86_32_sring_t *sring_x86_32 = dataplane->sring; |
| 806 | |
| 807 | BACK_RING_INIT(&dataplane->rings.x86_32_part, sring_x86_32, |
| 808 | ring_size); |
| 809 | break; |
| 810 | } |
| 811 | case BLKIF_PROTOCOL_X86_64: |
| 812 | { |
| 813 | blkif_x86_64_sring_t *sring_x86_64 = dataplane->sring; |
| 814 | |
| 815 | BACK_RING_INIT(&dataplane->rings.x86_64_part, sring_x86_64, |
| 816 | ring_size); |
| 817 | break; |
| 818 | } |
| 819 | } |
| 820 | |
| 821 | dataplane->event_channel = |
| 822 | xen_device_bind_event_channel(xendev, event_channel, |
| 823 | xen_block_dataplane_event, dataplane, |
| 824 | errp); |
| 825 | if (*errp) { |
| 826 | goto stop; |
| 827 | } |
| 828 | |
| 829 | /* If other users keep the BlockBackend in the iothread, that's ok */ |
| 830 | blk_set_aio_context(dataplane->blk, dataplane->ctx, NULL); |
| 831 | |
| 832 | if (!blk_in_drain(dataplane->blk)) { |
| 833 | xen_block_dataplane_attach(dataplane); |
| 834 | } |
| 835 | |
| 836 | return; |
| 837 | |
| 838 | stop: |
| 839 | xen_block_dataplane_stop(dataplane); |
| 840 | } |