Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0+ |
| 2 | /* |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 3 | * (C) 2007-2008 Samuel Thibault. |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 4 | * (C) Copyright 2020 EPAM Systems Inc. |
| 5 | */ |
| 6 | #include <blk.h> |
| 7 | #include <common.h> |
| 8 | #include <dm.h> |
| 9 | #include <dm/device-internal.h> |
Anastasiia Lukianenko | a999313 | 2020-08-06 12:42:56 +0300 | [diff] [blame] | 10 | #include <malloc.h> |
| 11 | #include <part.h> |
| 12 | |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 13 | #include <asm/armv8/mmu.h> |
| 14 | #include <asm/io.h> |
| 15 | #include <asm/xen/system.h> |
| 16 | |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 17 | #include <linux/bug.h> |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 18 | #include <linux/compat.h> |
| 19 | |
| 20 | #include <xen/events.h> |
| 21 | #include <xen/gnttab.h> |
| 22 | #include <xen/hvm.h> |
Anastasiia Lukianenko | a999313 | 2020-08-06 12:42:56 +0300 | [diff] [blame] | 23 | #include <xen/xenbus.h> |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 24 | |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 25 | #include <xen/interface/io/ring.h> |
| 26 | #include <xen/interface/io/blkif.h> |
| 27 | #include <xen/interface/io/protocols.h> |
| 28 | |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 29 | #define DRV_NAME "pvblock" |
| 30 | #define DRV_NAME_BLK "pvblock_blk" |
| 31 | |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 32 | #define O_RDONLY 00 |
| 33 | #define O_RDWR 02 |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 34 | #define WAIT_RING_TO_MS 10 |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 35 | |
| 36 | struct blkfront_info { |
| 37 | u64 sectors; |
| 38 | unsigned int sector_size; |
| 39 | int mode; |
| 40 | int info; |
| 41 | int barrier; |
| 42 | int flush; |
| 43 | }; |
| 44 | |
| 45 | /** |
| 46 | * struct blkfront_dev - Struct representing blkfront device |
| 47 | * @dom: Domain id |
| 48 | * @ring: Front_ring structure |
| 49 | * @ring_ref: The grant reference, allowing us to grant access |
| 50 | * to the ring to the other end/domain |
| 51 | * @evtchn: Event channel used to signal ring events |
| 52 | * @handle: Events handle |
| 53 | * @nodename: Device XenStore path in format "device/vbd/" + @devid |
| 54 | * @backend: Backend XenStore path |
| 55 | * @info: Private data |
| 56 | * @devid: Device id |
| 57 | */ |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 58 | struct blkfront_dev { |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 59 | domid_t dom; |
| 60 | |
| 61 | struct blkif_front_ring ring; |
| 62 | grant_ref_t ring_ref; |
| 63 | evtchn_port_t evtchn; |
| 64 | blkif_vdev_t handle; |
| 65 | |
| 66 | char *nodename; |
| 67 | char *backend; |
| 68 | struct blkfront_info info; |
| 69 | unsigned int devid; |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 70 | u8 *bounce_buffer; |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 71 | }; |
| 72 | |
Anastasiia Lukianenko | a999313 | 2020-08-06 12:42:56 +0300 | [diff] [blame] | 73 | struct blkfront_platdata { |
| 74 | unsigned int devid; |
| 75 | }; |
| 76 | |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 77 | /** |
| 78 | * struct blkfront_aiocb - AIO сontrol block |
| 79 | * @aio_dev: Blockfront device |
| 80 | * @aio_buf: Memory buffer, which must be sector-aligned for |
| 81 | * @aio_dev sector |
| 82 | * @aio_nbytes: Size of AIO, which must be less than @aio_dev |
| 83 | * sector-sized amounts |
| 84 | * @aio_offset: Offset, which must not go beyond @aio_dev |
| 85 | * sector-aligned location |
| 86 | * @data: Data used to receiving response from ring |
| 87 | * @gref: Array of grant references |
| 88 | * @n: Number of segments |
| 89 | * @aio_cb: Represents one I/O request. |
| 90 | */ |
| 91 | struct blkfront_aiocb { |
| 92 | struct blkfront_dev *aio_dev; |
| 93 | u8 *aio_buf; |
| 94 | size_t aio_nbytes; |
| 95 | off_t aio_offset; |
| 96 | void *data; |
| 97 | |
| 98 | grant_ref_t gref[BLKIF_MAX_SEGMENTS_PER_REQUEST]; |
| 99 | int n; |
| 100 | |
| 101 | void (*aio_cb)(struct blkfront_aiocb *aiocb, int ret); |
| 102 | }; |
| 103 | |
| 104 | static void blkfront_sync(struct blkfront_dev *dev); |
| 105 | |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 106 | static void free_blkfront(struct blkfront_dev *dev) |
| 107 | { |
| 108 | mask_evtchn(dev->evtchn); |
| 109 | free(dev->backend); |
| 110 | |
| 111 | gnttab_end_access(dev->ring_ref); |
| 112 | free(dev->ring.sring); |
| 113 | |
| 114 | unbind_evtchn(dev->evtchn); |
| 115 | |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 116 | free(dev->bounce_buffer); |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 117 | free(dev->nodename); |
| 118 | free(dev); |
| 119 | } |
| 120 | |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 121 | static int init_blkfront(unsigned int devid, struct blkfront_dev *dev) |
| 122 | { |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 123 | xenbus_transaction_t xbt; |
| 124 | char *err = NULL; |
| 125 | char *message = NULL; |
| 126 | struct blkif_sring *s; |
| 127 | int retry = 0; |
| 128 | char *msg = NULL; |
| 129 | char *c; |
| 130 | char nodename[32]; |
| 131 | char path[ARRAY_SIZE(nodename) + strlen("/backend-id") + 1]; |
| 132 | |
| 133 | sprintf(nodename, "device/vbd/%d", devid); |
| 134 | |
| 135 | memset(dev, 0, sizeof(*dev)); |
| 136 | dev->nodename = strdup(nodename); |
| 137 | dev->devid = devid; |
| 138 | |
| 139 | snprintf(path, sizeof(path), "%s/backend-id", nodename); |
| 140 | dev->dom = xenbus_read_integer(path); |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 141 | evtchn_alloc_unbound(dev->dom, NULL, dev, &dev->evtchn); |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 142 | |
| 143 | s = (struct blkif_sring *)memalign(PAGE_SIZE, PAGE_SIZE); |
| 144 | if (!s) { |
| 145 | printf("Failed to allocate shared ring\n"); |
| 146 | goto error; |
| 147 | } |
| 148 | |
| 149 | SHARED_RING_INIT(s); |
| 150 | FRONT_RING_INIT(&dev->ring, s, PAGE_SIZE); |
| 151 | |
| 152 | dev->ring_ref = gnttab_grant_access(dev->dom, virt_to_pfn(s), 0); |
| 153 | |
| 154 | again: |
| 155 | err = xenbus_transaction_start(&xbt); |
| 156 | if (err) { |
| 157 | printf("starting transaction\n"); |
| 158 | free(err); |
| 159 | } |
| 160 | |
| 161 | err = xenbus_printf(xbt, nodename, "ring-ref", "%u", dev->ring_ref); |
| 162 | if (err) { |
| 163 | message = "writing ring-ref"; |
| 164 | goto abort_transaction; |
| 165 | } |
| 166 | err = xenbus_printf(xbt, nodename, "event-channel", "%u", dev->evtchn); |
| 167 | if (err) { |
| 168 | message = "writing event-channel"; |
| 169 | goto abort_transaction; |
| 170 | } |
| 171 | err = xenbus_printf(xbt, nodename, "protocol", "%s", |
| 172 | XEN_IO_PROTO_ABI_NATIVE); |
| 173 | if (err) { |
| 174 | message = "writing protocol"; |
| 175 | goto abort_transaction; |
| 176 | } |
| 177 | |
| 178 | snprintf(path, sizeof(path), "%s/state", nodename); |
| 179 | err = xenbus_switch_state(xbt, path, XenbusStateConnected); |
| 180 | if (err) { |
| 181 | message = "switching state"; |
| 182 | goto abort_transaction; |
| 183 | } |
| 184 | |
| 185 | err = xenbus_transaction_end(xbt, 0, &retry); |
| 186 | free(err); |
| 187 | if (retry) { |
| 188 | goto again; |
| 189 | printf("completing transaction\n"); |
| 190 | } |
| 191 | |
| 192 | goto done; |
| 193 | |
| 194 | abort_transaction: |
| 195 | free(err); |
| 196 | err = xenbus_transaction_end(xbt, 1, &retry); |
| 197 | printf("Abort transaction %s\n", message); |
| 198 | goto error; |
| 199 | |
| 200 | done: |
| 201 | snprintf(path, sizeof(path), "%s/backend", nodename); |
| 202 | msg = xenbus_read(XBT_NIL, path, &dev->backend); |
| 203 | if (msg) { |
| 204 | printf("Error %s when reading the backend path %s\n", |
| 205 | msg, path); |
| 206 | goto error; |
| 207 | } |
| 208 | |
| 209 | dev->handle = strtoul(strrchr(nodename, '/') + 1, NULL, 0); |
| 210 | |
| 211 | { |
| 212 | XenbusState state; |
| 213 | char path[strlen(dev->backend) + |
| 214 | strlen("/feature-flush-cache") + 1]; |
| 215 | |
| 216 | snprintf(path, sizeof(path), "%s/mode", dev->backend); |
| 217 | msg = xenbus_read(XBT_NIL, path, &c); |
| 218 | if (msg) { |
| 219 | printf("Error %s when reading the mode\n", msg); |
| 220 | goto error; |
| 221 | } |
| 222 | if (*c == 'w') |
| 223 | dev->info.mode = O_RDWR; |
| 224 | else |
| 225 | dev->info.mode = O_RDONLY; |
| 226 | free(c); |
| 227 | |
| 228 | snprintf(path, sizeof(path), "%s/state", dev->backend); |
| 229 | |
| 230 | msg = NULL; |
| 231 | state = xenbus_read_integer(path); |
| 232 | while (!msg && state < XenbusStateConnected) |
| 233 | msg = xenbus_wait_for_state_change(path, &state); |
| 234 | if (msg || state != XenbusStateConnected) { |
| 235 | printf("backend not available, state=%d\n", state); |
| 236 | goto error; |
| 237 | } |
| 238 | |
| 239 | snprintf(path, sizeof(path), "%s/info", dev->backend); |
| 240 | dev->info.info = xenbus_read_integer(path); |
| 241 | |
| 242 | snprintf(path, sizeof(path), "%s/sectors", dev->backend); |
| 243 | /* |
| 244 | * FIXME: read_integer returns an int, so disk size |
| 245 | * limited to 1TB for now |
| 246 | */ |
| 247 | dev->info.sectors = xenbus_read_integer(path); |
| 248 | |
| 249 | snprintf(path, sizeof(path), "%s/sector-size", dev->backend); |
| 250 | dev->info.sector_size = xenbus_read_integer(path); |
| 251 | |
| 252 | snprintf(path, sizeof(path), "%s/feature-barrier", |
| 253 | dev->backend); |
| 254 | dev->info.barrier = xenbus_read_integer(path); |
| 255 | |
| 256 | snprintf(path, sizeof(path), "%s/feature-flush-cache", |
| 257 | dev->backend); |
| 258 | dev->info.flush = xenbus_read_integer(path); |
| 259 | } |
| 260 | unmask_evtchn(dev->evtchn); |
| 261 | |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 262 | dev->bounce_buffer = memalign(dev->info.sector_size, |
| 263 | dev->info.sector_size); |
| 264 | if (!dev->bounce_buffer) { |
| 265 | printf("Failed to allocate bouncing buffer\n"); |
| 266 | goto error; |
| 267 | } |
| 268 | |
| 269 | debug("%llu sectors of %u bytes, bounce buffer at %p\n", |
| 270 | dev->info.sectors, dev->info.sector_size, |
| 271 | dev->bounce_buffer); |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 272 | |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 273 | return 0; |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 274 | |
| 275 | error: |
| 276 | free(msg); |
| 277 | free(err); |
| 278 | free_blkfront(dev); |
| 279 | return -ENODEV; |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 280 | } |
| 281 | |
| 282 | static void shutdown_blkfront(struct blkfront_dev *dev) |
| 283 | { |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 284 | char *err = NULL, *err2; |
| 285 | XenbusState state; |
| 286 | |
| 287 | char path[strlen(dev->backend) + strlen("/state") + 1]; |
| 288 | char nodename[strlen(dev->nodename) + strlen("/event-channel") + 1]; |
| 289 | |
| 290 | debug("Close " DRV_NAME ", device ID %d\n", dev->devid); |
| 291 | |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 292 | blkfront_sync(dev); |
| 293 | |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 294 | snprintf(path, sizeof(path), "%s/state", dev->backend); |
| 295 | snprintf(nodename, sizeof(nodename), "%s/state", dev->nodename); |
| 296 | |
Anastasiia Lukianenko | 3337b29 | 2020-08-21 12:10:04 +0300 | [diff] [blame] | 297 | err = xenbus_switch_state(XBT_NIL, nodename, XenbusStateClosing); |
| 298 | if (err) { |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 299 | printf("%s: error changing state to %d: %s\n", __func__, |
| 300 | XenbusStateClosing, err); |
| 301 | goto close; |
| 302 | } |
| 303 | |
| 304 | state = xenbus_read_integer(path); |
| 305 | while (!err && state < XenbusStateClosing) |
| 306 | err = xenbus_wait_for_state_change(path, &state); |
| 307 | free(err); |
| 308 | |
Anastasiia Lukianenko | 3337b29 | 2020-08-21 12:10:04 +0300 | [diff] [blame] | 309 | err = xenbus_switch_state(XBT_NIL, nodename, XenbusStateClosed); |
| 310 | if (err) { |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 311 | printf("%s: error changing state to %d: %s\n", __func__, |
| 312 | XenbusStateClosed, err); |
| 313 | goto close; |
| 314 | } |
| 315 | |
| 316 | state = xenbus_read_integer(path); |
| 317 | while (state < XenbusStateClosed) { |
| 318 | err = xenbus_wait_for_state_change(path, &state); |
| 319 | free(err); |
| 320 | } |
| 321 | |
Anastasiia Lukianenko | 3337b29 | 2020-08-21 12:10:04 +0300 | [diff] [blame] | 322 | err = xenbus_switch_state(XBT_NIL, nodename, XenbusStateInitialising); |
| 323 | if (err) { |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 324 | printf("%s: error changing state to %d: %s\n", __func__, |
| 325 | XenbusStateInitialising, err); |
| 326 | goto close; |
| 327 | } |
| 328 | |
| 329 | state = xenbus_read_integer(path); |
| 330 | while (!err && |
| 331 | (state < XenbusStateInitWait || state >= XenbusStateClosed)) |
| 332 | err = xenbus_wait_for_state_change(path, &state); |
| 333 | |
| 334 | close: |
| 335 | free(err); |
| 336 | |
| 337 | snprintf(nodename, sizeof(nodename), "%s/ring-ref", dev->nodename); |
| 338 | err2 = xenbus_rm(XBT_NIL, nodename); |
| 339 | free(err2); |
| 340 | snprintf(nodename, sizeof(nodename), "%s/event-channel", dev->nodename); |
| 341 | err2 = xenbus_rm(XBT_NIL, nodename); |
| 342 | free(err2); |
| 343 | |
| 344 | if (!err) |
| 345 | free_blkfront(dev); |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 346 | } |
| 347 | |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 348 | /** |
| 349 | * blkfront_aio_poll() - AIO polling function. |
| 350 | * @dev: Blkfront device |
| 351 | * |
| 352 | * Here we receive response from the ring and check its status. This happens |
| 353 | * until we read all data from the ring. We read the data from consumed pointer |
| 354 | * to the response pointer. Then increase consumed pointer to make it clear that |
| 355 | * the data has been read. |
| 356 | * |
| 357 | * Return: Number of consumed bytes. |
| 358 | */ |
| 359 | static int blkfront_aio_poll(struct blkfront_dev *dev) |
| 360 | { |
| 361 | RING_IDX rp, cons; |
| 362 | struct blkif_response *rsp; |
| 363 | int more; |
| 364 | int nr_consumed; |
| 365 | |
| 366 | moretodo: |
| 367 | rp = dev->ring.sring->rsp_prod; |
| 368 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
| 369 | cons = dev->ring.rsp_cons; |
| 370 | |
| 371 | nr_consumed = 0; |
| 372 | while ((cons != rp)) { |
| 373 | struct blkfront_aiocb *aiocbp; |
| 374 | int status; |
| 375 | |
| 376 | rsp = RING_GET_RESPONSE(&dev->ring, cons); |
| 377 | nr_consumed++; |
| 378 | |
| 379 | aiocbp = (void *)(uintptr_t)rsp->id; |
| 380 | status = rsp->status; |
| 381 | |
| 382 | switch (rsp->operation) { |
| 383 | case BLKIF_OP_READ: |
| 384 | case BLKIF_OP_WRITE: |
| 385 | { |
| 386 | int j; |
| 387 | |
| 388 | if (status != BLKIF_RSP_OKAY) |
| 389 | printf("%s error %d on %s at offset %llu, num bytes %llu\n", |
| 390 | rsp->operation == BLKIF_OP_READ ? |
| 391 | "read" : "write", |
| 392 | status, aiocbp->aio_dev->nodename, |
| 393 | (unsigned long long)aiocbp->aio_offset, |
| 394 | (unsigned long long)aiocbp->aio_nbytes); |
| 395 | |
| 396 | for (j = 0; j < aiocbp->n; j++) |
| 397 | gnttab_end_access(aiocbp->gref[j]); |
| 398 | |
| 399 | break; |
| 400 | } |
| 401 | |
| 402 | case BLKIF_OP_WRITE_BARRIER: |
| 403 | if (status != BLKIF_RSP_OKAY) |
| 404 | printf("write barrier error %d\n", status); |
| 405 | break; |
| 406 | case BLKIF_OP_FLUSH_DISKCACHE: |
| 407 | if (status != BLKIF_RSP_OKAY) |
| 408 | printf("flush error %d\n", status); |
| 409 | break; |
| 410 | |
| 411 | default: |
| 412 | printf("unrecognized block operation %d response (status %d)\n", |
| 413 | rsp->operation, status); |
| 414 | break; |
| 415 | } |
| 416 | |
| 417 | dev->ring.rsp_cons = ++cons; |
| 418 | /* Nota: callback frees aiocbp itself */ |
| 419 | if (aiocbp && aiocbp->aio_cb) |
| 420 | aiocbp->aio_cb(aiocbp, status ? -EIO : 0); |
| 421 | if (dev->ring.rsp_cons != cons) |
| 422 | /* We reentered, we must not continue here */ |
| 423 | break; |
| 424 | } |
| 425 | |
| 426 | RING_FINAL_CHECK_FOR_RESPONSES(&dev->ring, more); |
| 427 | if (more) |
| 428 | goto moretodo; |
| 429 | |
| 430 | return nr_consumed; |
| 431 | } |
| 432 | |
| 433 | static void blkfront_wait_slot(struct blkfront_dev *dev) |
| 434 | { |
| 435 | /* Wait for a slot */ |
| 436 | if (RING_FULL(&dev->ring)) { |
| 437 | while (true) { |
| 438 | blkfront_aio_poll(dev); |
| 439 | if (!RING_FULL(&dev->ring)) |
| 440 | break; |
| 441 | wait_event_timeout(NULL, !RING_FULL(&dev->ring), |
| 442 | WAIT_RING_TO_MS); |
| 443 | } |
| 444 | } |
| 445 | } |
| 446 | |
| 447 | /** |
| 448 | * blkfront_aio_poll() - Issue an aio. |
| 449 | * @aiocbp: AIO control block structure |
| 450 | * @write: Describes is it read or write operation |
| 451 | * 0 - read |
| 452 | * 1 - write |
| 453 | * |
| 454 | * We check whether the AIO parameters meet the requirements of the device. |
| 455 | * Then receive request from ring and define its arguments. After this we |
| 456 | * grant access to the grant references. The last step is notifying about AIO |
| 457 | * via event channel. |
| 458 | */ |
| 459 | static void blkfront_aio(struct blkfront_aiocb *aiocbp, int write) |
| 460 | { |
| 461 | struct blkfront_dev *dev = aiocbp->aio_dev; |
| 462 | struct blkif_request *req; |
| 463 | RING_IDX i; |
| 464 | int notify; |
| 465 | int n, j; |
| 466 | uintptr_t start, end; |
| 467 | |
| 468 | /* Can't io at non-sector-aligned location */ |
| 469 | BUG_ON(aiocbp->aio_offset & (dev->info.sector_size - 1)); |
| 470 | /* Can't io non-sector-sized amounts */ |
| 471 | BUG_ON(aiocbp->aio_nbytes & (dev->info.sector_size - 1)); |
| 472 | /* Can't io non-sector-aligned buffer */ |
| 473 | BUG_ON(((uintptr_t)aiocbp->aio_buf & (dev->info.sector_size - 1))); |
| 474 | |
| 475 | start = (uintptr_t)aiocbp->aio_buf & PAGE_MASK; |
| 476 | end = ((uintptr_t)aiocbp->aio_buf + aiocbp->aio_nbytes + |
| 477 | PAGE_SIZE - 1) & PAGE_MASK; |
| 478 | n = (end - start) / PAGE_SIZE; |
| 479 | aiocbp->n = n; |
| 480 | |
| 481 | BUG_ON(n > BLKIF_MAX_SEGMENTS_PER_REQUEST); |
| 482 | |
| 483 | blkfront_wait_slot(dev); |
| 484 | i = dev->ring.req_prod_pvt; |
| 485 | req = RING_GET_REQUEST(&dev->ring, i); |
| 486 | |
| 487 | req->operation = write ? BLKIF_OP_WRITE : BLKIF_OP_READ; |
| 488 | req->nr_segments = n; |
| 489 | req->handle = dev->handle; |
| 490 | req->id = (uintptr_t)aiocbp; |
| 491 | req->sector_number = aiocbp->aio_offset / dev->info.sector_size; |
| 492 | |
| 493 | for (j = 0; j < n; j++) { |
| 494 | req->seg[j].first_sect = 0; |
| 495 | req->seg[j].last_sect = PAGE_SIZE / dev->info.sector_size - 1; |
| 496 | } |
| 497 | req->seg[0].first_sect = ((uintptr_t)aiocbp->aio_buf & ~PAGE_MASK) / |
| 498 | dev->info.sector_size; |
| 499 | req->seg[n - 1].last_sect = (((uintptr_t)aiocbp->aio_buf + |
| 500 | aiocbp->aio_nbytes - 1) & ~PAGE_MASK) / dev->info.sector_size; |
| 501 | for (j = 0; j < n; j++) { |
| 502 | uintptr_t data = start + j * PAGE_SIZE; |
| 503 | |
| 504 | if (!write) { |
| 505 | /* Trigger CoW if needed */ |
| 506 | *(char *)(data + (req->seg[j].first_sect * |
| 507 | dev->info.sector_size)) = 0; |
| 508 | barrier(); |
| 509 | } |
| 510 | req->seg[j].gref = gnttab_grant_access(dev->dom, |
| 511 | virt_to_pfn((void *)data), |
| 512 | write); |
| 513 | aiocbp->gref[j] = req->seg[j].gref; |
| 514 | } |
| 515 | |
| 516 | dev->ring.req_prod_pvt = i + 1; |
| 517 | |
| 518 | wmb(); |
| 519 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&dev->ring, notify); |
| 520 | |
| 521 | if (notify) |
| 522 | notify_remote_via_evtchn(dev->evtchn); |
| 523 | } |
| 524 | |
| 525 | static void blkfront_aio_cb(struct blkfront_aiocb *aiocbp, int ret) |
| 526 | { |
| 527 | aiocbp->data = (void *)1; |
| 528 | aiocbp->aio_cb = NULL; |
| 529 | } |
| 530 | |
| 531 | static void blkfront_io(struct blkfront_aiocb *aiocbp, int write) |
| 532 | { |
| 533 | aiocbp->aio_cb = blkfront_aio_cb; |
| 534 | blkfront_aio(aiocbp, write); |
| 535 | aiocbp->data = NULL; |
| 536 | |
| 537 | while (true) { |
| 538 | blkfront_aio_poll(aiocbp->aio_dev); |
| 539 | if (aiocbp->data) |
| 540 | break; |
| 541 | cpu_relax(); |
| 542 | } |
| 543 | } |
| 544 | |
| 545 | static void blkfront_push_operation(struct blkfront_dev *dev, u8 op, |
| 546 | uint64_t id) |
| 547 | { |
| 548 | struct blkif_request *req; |
| 549 | int notify, i; |
| 550 | |
| 551 | blkfront_wait_slot(dev); |
| 552 | i = dev->ring.req_prod_pvt; |
| 553 | req = RING_GET_REQUEST(&dev->ring, i); |
| 554 | req->operation = op; |
| 555 | req->nr_segments = 0; |
| 556 | req->handle = dev->handle; |
| 557 | req->id = id; |
| 558 | req->sector_number = 0; |
| 559 | dev->ring.req_prod_pvt = i + 1; |
| 560 | wmb(); |
| 561 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&dev->ring, notify); |
| 562 | if (notify) |
| 563 | notify_remote_via_evtchn(dev->evtchn); |
| 564 | } |
| 565 | |
| 566 | static void blkfront_sync(struct blkfront_dev *dev) |
| 567 | { |
| 568 | if (dev->info.mode == O_RDWR) { |
| 569 | if (dev->info.barrier == 1) |
| 570 | blkfront_push_operation(dev, |
| 571 | BLKIF_OP_WRITE_BARRIER, 0); |
| 572 | |
| 573 | if (dev->info.flush == 1) |
| 574 | blkfront_push_operation(dev, |
| 575 | BLKIF_OP_FLUSH_DISKCACHE, 0); |
| 576 | } |
| 577 | |
| 578 | while (true) { |
| 579 | blkfront_aio_poll(dev); |
| 580 | if (RING_FREE_REQUESTS(&dev->ring) == RING_SIZE(&dev->ring)) |
| 581 | break; |
| 582 | cpu_relax(); |
| 583 | } |
| 584 | } |
| 585 | |
| 586 | /** |
| 587 | * pvblock_iop() - Issue an aio. |
| 588 | * @udev: Pvblock device |
| 589 | * @blknr: Block number to read from / write to |
| 590 | * @blkcnt: Amount of blocks to read / write |
| 591 | * @buffer: Memory buffer with data to be read / write |
| 592 | * @write: Describes is it read or write operation |
| 593 | * 0 - read |
| 594 | * 1 - write |
| 595 | * |
| 596 | * Depending on the operation - reading or writing, data is read / written from the |
| 597 | * specified address (@buffer) to the sector (@blknr). |
| 598 | */ |
| 599 | static ulong pvblock_iop(struct udevice *udev, lbaint_t blknr, |
| 600 | lbaint_t blkcnt, void *buffer, int write) |
| 601 | { |
| 602 | struct blkfront_dev *blk_dev = dev_get_priv(udev); |
| 603 | struct blk_desc *desc = dev_get_uclass_platdata(udev); |
| 604 | struct blkfront_aiocb aiocb; |
| 605 | lbaint_t blocks_todo; |
| 606 | bool unaligned; |
| 607 | |
| 608 | if (blkcnt == 0) |
| 609 | return 0; |
| 610 | |
| 611 | if ((blknr + blkcnt) > desc->lba) { |
| 612 | printf(DRV_NAME ": block number 0x" LBAF " exceeds max(0x" LBAF ")\n", |
| 613 | blknr + blkcnt, desc->lba); |
| 614 | return 0; |
| 615 | } |
| 616 | |
| 617 | unaligned = (uintptr_t)buffer & (blk_dev->info.sector_size - 1); |
| 618 | |
| 619 | aiocb.aio_dev = blk_dev; |
| 620 | aiocb.aio_offset = blknr * desc->blksz; |
| 621 | aiocb.aio_cb = NULL; |
| 622 | aiocb.data = NULL; |
| 623 | blocks_todo = blkcnt; |
| 624 | do { |
| 625 | aiocb.aio_buf = unaligned ? blk_dev->bounce_buffer : buffer; |
| 626 | |
| 627 | if (write && unaligned) |
| 628 | memcpy(blk_dev->bounce_buffer, buffer, desc->blksz); |
| 629 | |
| 630 | aiocb.aio_nbytes = unaligned ? desc->blksz : |
| 631 | min((size_t)(BLKIF_MAX_SEGMENTS_PER_REQUEST * PAGE_SIZE), |
| 632 | (size_t)(blocks_todo * desc->blksz)); |
| 633 | |
| 634 | blkfront_io(&aiocb, write); |
| 635 | |
| 636 | if (!write && unaligned) |
| 637 | memcpy(buffer, blk_dev->bounce_buffer, desc->blksz); |
| 638 | |
| 639 | aiocb.aio_offset += aiocb.aio_nbytes; |
| 640 | buffer += aiocb.aio_nbytes; |
| 641 | blocks_todo -= aiocb.aio_nbytes / desc->blksz; |
| 642 | } while (blocks_todo > 0); |
| 643 | |
| 644 | return blkcnt; |
| 645 | } |
| 646 | |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 647 | ulong pvblock_blk_read(struct udevice *udev, lbaint_t blknr, lbaint_t blkcnt, |
| 648 | void *buffer) |
| 649 | { |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 650 | return pvblock_iop(udev, blknr, blkcnt, buffer, 0); |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 651 | } |
| 652 | |
| 653 | ulong pvblock_blk_write(struct udevice *udev, lbaint_t blknr, lbaint_t blkcnt, |
| 654 | const void *buffer) |
| 655 | { |
Anastasiia Lukianenko | 3a739cc | 2020-08-06 12:42:58 +0300 | [diff] [blame] | 656 | return pvblock_iop(udev, blknr, blkcnt, (void *)buffer, 1); |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 657 | } |
| 658 | |
| 659 | static int pvblock_blk_bind(struct udevice *udev) |
| 660 | { |
Anastasiia Lukianenko | a999313 | 2020-08-06 12:42:56 +0300 | [diff] [blame] | 661 | struct blk_desc *desc = dev_get_uclass_platdata(udev); |
| 662 | int devnum; |
| 663 | |
| 664 | desc->if_type = IF_TYPE_PVBLOCK; |
| 665 | /* |
| 666 | * Initialize the devnum to -ENODEV. This is to make sure that |
| 667 | * blk_next_free_devnum() works as expected, since the default |
| 668 | * value 0 is a valid devnum. |
| 669 | */ |
| 670 | desc->devnum = -ENODEV; |
| 671 | devnum = blk_next_free_devnum(IF_TYPE_PVBLOCK); |
| 672 | if (devnum < 0) |
| 673 | return devnum; |
| 674 | desc->devnum = devnum; |
| 675 | desc->part_type = PART_TYPE_UNKNOWN; |
| 676 | desc->bdev = udev; |
| 677 | |
| 678 | strncpy(desc->vendor, "Xen", sizeof(desc->vendor)); |
| 679 | strncpy(desc->revision, "1", sizeof(desc->revision)); |
| 680 | strncpy(desc->product, "Virtual disk", sizeof(desc->product)); |
| 681 | |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 682 | return 0; |
| 683 | } |
| 684 | |
| 685 | static int pvblock_blk_probe(struct udevice *udev) |
| 686 | { |
| 687 | struct blkfront_dev *blk_dev = dev_get_priv(udev); |
Anastasiia Lukianenko | a999313 | 2020-08-06 12:42:56 +0300 | [diff] [blame] | 688 | struct blkfront_platdata *platdata = dev_get_platdata(udev); |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 689 | struct blk_desc *desc = dev_get_uclass_platdata(udev); |
Anastasiia Lukianenko | a999313 | 2020-08-06 12:42:56 +0300 | [diff] [blame] | 690 | int ret, devid; |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 691 | |
Anastasiia Lukianenko | a999313 | 2020-08-06 12:42:56 +0300 | [diff] [blame] | 692 | devid = platdata->devid; |
| 693 | free(platdata); |
| 694 | |
| 695 | ret = init_blkfront(devid, blk_dev); |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 696 | if (ret < 0) |
| 697 | return ret; |
Anastasiia Lukianenko | 17c96f8 | 2020-08-06 12:42:57 +0300 | [diff] [blame] | 698 | |
| 699 | desc->blksz = blk_dev->info.sector_size; |
| 700 | desc->lba = blk_dev->info.sectors; |
| 701 | desc->log2blksz = LOG2(blk_dev->info.sector_size); |
| 702 | |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 703 | return 0; |
| 704 | } |
| 705 | |
| 706 | static int pvblock_blk_remove(struct udevice *udev) |
| 707 | { |
| 708 | struct blkfront_dev *blk_dev = dev_get_priv(udev); |
| 709 | |
| 710 | shutdown_blkfront(blk_dev); |
| 711 | return 0; |
| 712 | } |
| 713 | |
| 714 | static const struct blk_ops pvblock_blk_ops = { |
| 715 | .read = pvblock_blk_read, |
| 716 | .write = pvblock_blk_write, |
| 717 | }; |
| 718 | |
| 719 | U_BOOT_DRIVER(pvblock_blk) = { |
| 720 | .name = DRV_NAME_BLK, |
| 721 | .id = UCLASS_BLK, |
| 722 | .ops = &pvblock_blk_ops, |
| 723 | .bind = pvblock_blk_bind, |
| 724 | .probe = pvblock_blk_probe, |
| 725 | .remove = pvblock_blk_remove, |
| 726 | .priv_auto_alloc_size = sizeof(struct blkfront_dev), |
| 727 | .flags = DM_FLAG_OS_PREPARE, |
| 728 | }; |
| 729 | |
| 730 | /******************************************************************************* |
| 731 | * Para-virtual block device class |
| 732 | *******************************************************************************/ |
| 733 | |
Anastasiia Lukianenko | a999313 | 2020-08-06 12:42:56 +0300 | [diff] [blame] | 734 | typedef int (*enum_vbd_callback)(struct udevice *parent, unsigned int devid); |
| 735 | |
| 736 | static int on_new_vbd(struct udevice *parent, unsigned int devid) |
| 737 | { |
| 738 | struct driver_info info; |
| 739 | struct udevice *udev; |
| 740 | struct blkfront_platdata *platdata; |
| 741 | int ret; |
| 742 | |
| 743 | debug("New " DRV_NAME_BLK ", device ID %d\n", devid); |
| 744 | |
| 745 | platdata = malloc(sizeof(struct blkfront_platdata)); |
| 746 | if (!platdata) { |
| 747 | printf("Failed to allocate platform data\n"); |
| 748 | return -ENOMEM; |
| 749 | } |
| 750 | |
| 751 | platdata->devid = devid; |
| 752 | |
| 753 | info.name = DRV_NAME_BLK; |
| 754 | info.platdata = platdata; |
| 755 | |
| 756 | ret = device_bind_by_name(parent, false, &info, &udev); |
| 757 | if (ret < 0) { |
| 758 | printf("Failed to bind " DRV_NAME_BLK " to device with ID %d, ret: %d\n", |
| 759 | devid, ret); |
| 760 | free(platdata); |
| 761 | } |
| 762 | return ret; |
| 763 | } |
| 764 | |
| 765 | static int xenbus_enumerate_vbd(struct udevice *udev, enum_vbd_callback clb) |
| 766 | { |
| 767 | char **dirs, *msg; |
| 768 | int i, ret; |
| 769 | |
| 770 | msg = xenbus_ls(XBT_NIL, "device/vbd", &dirs); |
| 771 | if (msg) { |
| 772 | printf("Failed to read device/vbd directory: %s\n", msg); |
| 773 | free(msg); |
| 774 | return -ENODEV; |
| 775 | } |
| 776 | |
| 777 | for (i = 0; dirs[i]; i++) { |
| 778 | int devid; |
| 779 | |
| 780 | sscanf(dirs[i], "%d", &devid); |
| 781 | ret = clb(udev, devid); |
| 782 | if (ret < 0) |
| 783 | goto fail; |
| 784 | |
| 785 | free(dirs[i]); |
| 786 | } |
| 787 | ret = 0; |
| 788 | |
| 789 | fail: |
| 790 | for (; dirs[i]; i++) |
| 791 | free(dirs[i]); |
| 792 | free(dirs); |
| 793 | return ret; |
| 794 | } |
| 795 | |
Anastasiia Lukianenko | 53d725c | 2020-08-06 12:42:59 +0300 | [diff] [blame] | 796 | static void print_pvblock_devices(void) |
| 797 | { |
| 798 | struct udevice *udev; |
| 799 | bool first = true; |
| 800 | const char *class_name; |
| 801 | |
| 802 | class_name = uclass_get_name(UCLASS_PVBLOCK); |
| 803 | for (blk_first_device(IF_TYPE_PVBLOCK, &udev); udev; |
| 804 | blk_next_device(&udev), first = false) { |
| 805 | struct blk_desc *desc = dev_get_uclass_platdata(udev); |
| 806 | |
| 807 | if (!first) |
| 808 | puts(", "); |
| 809 | printf("%s: %d", class_name, desc->devnum); |
| 810 | } |
| 811 | printf("\n"); |
| 812 | } |
| 813 | |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 814 | void pvblock_init(void) |
| 815 | { |
| 816 | struct driver_info info; |
| 817 | struct udevice *udev; |
| 818 | struct uclass *uc; |
| 819 | int ret; |
| 820 | |
| 821 | /* |
| 822 | * At this point Xen drivers have already initialized, |
| 823 | * so we can instantiate the class driver and enumerate |
| 824 | * virtual block devices. |
| 825 | */ |
| 826 | info.name = DRV_NAME; |
| 827 | ret = device_bind_by_name(gd->dm_root, false, &info, &udev); |
| 828 | if (ret < 0) |
| 829 | printf("Failed to bind " DRV_NAME ", ret: %d\n", ret); |
| 830 | |
| 831 | /* Bootstrap virtual block devices class driver */ |
| 832 | ret = uclass_get(UCLASS_PVBLOCK, &uc); |
| 833 | if (ret) |
| 834 | return; |
| 835 | uclass_foreach_dev_probe(UCLASS_PVBLOCK, udev); |
Anastasiia Lukianenko | 53d725c | 2020-08-06 12:42:59 +0300 | [diff] [blame] | 836 | |
| 837 | print_pvblock_devices(); |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 838 | } |
| 839 | |
| 840 | static int pvblock_probe(struct udevice *udev) |
| 841 | { |
Anastasiia Lukianenko | a999313 | 2020-08-06 12:42:56 +0300 | [diff] [blame] | 842 | struct uclass *uc; |
| 843 | int ret; |
| 844 | |
| 845 | if (xenbus_enumerate_vbd(udev, on_new_vbd) < 0) |
| 846 | return -ENODEV; |
| 847 | |
| 848 | ret = uclass_get(UCLASS_BLK, &uc); |
| 849 | if (ret) |
| 850 | return ret; |
| 851 | uclass_foreach_dev_probe(UCLASS_BLK, udev) { |
| 852 | if (_ret) |
| 853 | return _ret; |
| 854 | }; |
Anastasiia Lukianenko | 722bc5b | 2020-08-06 12:42:55 +0300 | [diff] [blame] | 855 | return 0; |
| 856 | } |
| 857 | |
| 858 | U_BOOT_DRIVER(pvblock_drv) = { |
| 859 | .name = DRV_NAME, |
| 860 | .id = UCLASS_PVBLOCK, |
| 861 | .probe = pvblock_probe, |
| 862 | }; |
| 863 | |
| 864 | UCLASS_DRIVER(pvblock) = { |
| 865 | .name = DRV_NAME, |
| 866 | .id = UCLASS_PVBLOCK, |
| 867 | }; |