| 1 | /* |
| 2 | * Copyright (c) 2009, Microsoft Corporation. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or modify it |
| 5 | * under the terms and conditions of the GNU General Public License, |
| 6 | * version 2, as published by the Free Software Foundation. |
| 7 | * |
| 8 | * This program is distributed in the hope it will be useful, but WITHOUT |
| 9 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 10 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for |
| 11 | * more details. |
| 12 | * |
| 13 | * You should have received a copy of the GNU General Public License along with |
| 14 | * this program; if not, write to the Free Software Foundation, Inc., 59 Temple |
| 15 | * Place - Suite 330, Boston, MA 02111-1307 USA. |
| 16 | * |
| 17 | * Authors: |
| 18 | * Haiyang Zhang <haiyangz@microsoft.com> |
| 19 | * Hank Janssen <hjanssen@microsoft.com> |
| 20 | * K. Y. Srinivasan <kys@microsoft.com> |
| 21 | */ |
| 22 | |
| 23 | #include <linux/kernel.h> |
| 24 | #include <linux/wait.h> |
| 25 | #include <linux/sched.h> |
| 26 | #include <linux/completion.h> |
| 27 | #include <linux/string.h> |
| 28 | #include <linux/mm.h> |
| 29 | #include <linux/delay.h> |
| 30 | #include <linux/init.h> |
| 31 | #include <linux/slab.h> |
| 32 | #include <linux/module.h> |
| 33 | #include <linux/device.h> |
| 34 | #include <linux/hyperv.h> |
| 35 | #include <linux/mempool.h> |
| 36 | #include <scsi/scsi.h> |
| 37 | #include <scsi/scsi_cmnd.h> |
| 38 | #include <scsi/scsi_host.h> |
| 39 | #include <scsi/scsi_device.h> |
| 40 | #include <scsi/scsi_tcq.h> |
| 41 | #include <scsi/scsi_eh.h> |
| 42 | #include <scsi/scsi_devinfo.h> |
| 43 | #include <scsi/scsi_dbg.h> |
| 44 | |
| 45 | /* |
| 46 | * All wire protocol details (storage protocol between the guest and the host) |
| 47 | * are consolidated here. |
| 48 | * |
| 49 | * Begin protocol definitions. |
| 50 | */ |
| 51 | |
| 52 | /* |
| 53 | * Version history: |
| 54 | * V1 Beta: 0.1 |
| 55 | * V1 RC < 2008/1/31: 1.0 |
| 56 | * V1 RC > 2008/1/31: 2.0 |
| 57 | * Win7: 4.2 |
| 58 | */ |
| 59 | |
| 60 | #define VMSTOR_CURRENT_MAJOR 4 |
| 61 | #define VMSTOR_CURRENT_MINOR 2 |
| 62 | |
| 63 | |
| 64 | /* Packet structure describing virtual storage requests. */ |
| 65 | enum vstor_packet_operation { |
| 66 | VSTOR_OPERATION_COMPLETE_IO = 1, |
| 67 | VSTOR_OPERATION_REMOVE_DEVICE = 2, |
| 68 | VSTOR_OPERATION_EXECUTE_SRB = 3, |
| 69 | VSTOR_OPERATION_RESET_LUN = 4, |
| 70 | VSTOR_OPERATION_RESET_ADAPTER = 5, |
| 71 | VSTOR_OPERATION_RESET_BUS = 6, |
| 72 | VSTOR_OPERATION_BEGIN_INITIALIZATION = 7, |
| 73 | VSTOR_OPERATION_END_INITIALIZATION = 8, |
| 74 | VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9, |
| 75 | VSTOR_OPERATION_QUERY_PROPERTIES = 10, |
| 76 | VSTOR_OPERATION_ENUMERATE_BUS = 11, |
| 77 | VSTOR_OPERATION_MAXIMUM = 11 |
| 78 | }; |
| 79 | |
| 80 | /* |
| 81 | * Platform neutral description of a scsi request - |
| 82 | * this remains the same across the write regardless of 32/64 bit |
| 83 | * note: it's patterned off the SCSI_PASS_THROUGH structure |
| 84 | */ |
| 85 | #define STORVSC_MAX_CMD_LEN 0x10 |
| 86 | #define STORVSC_SENSE_BUFFER_SIZE 0x12 |
| 87 | #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14 |
| 88 | |
| 89 | struct vmscsi_request { |
| 90 | u16 length; |
| 91 | u8 srb_status; |
| 92 | u8 scsi_status; |
| 93 | |
| 94 | u8 port_number; |
| 95 | u8 path_id; |
| 96 | u8 target_id; |
| 97 | u8 lun; |
| 98 | |
| 99 | u8 cdb_length; |
| 100 | u8 sense_info_length; |
| 101 | u8 data_in; |
| 102 | u8 reserved; |
| 103 | |
| 104 | u32 data_transfer_length; |
| 105 | |
| 106 | union { |
| 107 | u8 cdb[STORVSC_MAX_CMD_LEN]; |
| 108 | u8 sense_data[STORVSC_SENSE_BUFFER_SIZE]; |
| 109 | u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING]; |
| 110 | }; |
| 111 | } __attribute((packed)); |
| 112 | |
| 113 | |
| 114 | /* |
| 115 | * This structure is sent during the intialization phase to get the different |
| 116 | * properties of the channel. |
| 117 | */ |
| 118 | struct vmstorage_channel_properties { |
| 119 | u16 protocol_version; |
| 120 | u8 path_id; |
| 121 | u8 target_id; |
| 122 | |
| 123 | /* Note: port number is only really known on the client side */ |
| 124 | u32 port_number; |
| 125 | u32 flags; |
| 126 | u32 max_transfer_bytes; |
| 127 | |
| 128 | /* |
| 129 | * This id is unique for each channel and will correspond with |
| 130 | * vendor specific data in the inquiry data. |
| 131 | */ |
| 132 | |
| 133 | u64 unique_id; |
| 134 | } __packed; |
| 135 | |
| 136 | /* This structure is sent during the storage protocol negotiations. */ |
| 137 | struct vmstorage_protocol_version { |
| 138 | /* Major (MSW) and minor (LSW) version numbers. */ |
| 139 | u16 major_minor; |
| 140 | |
| 141 | /* |
| 142 | * Revision number is auto-incremented whenever this file is changed |
| 143 | * (See FILL_VMSTOR_REVISION macro above). Mismatch does not |
| 144 | * definitely indicate incompatibility--but it does indicate mismatched |
| 145 | * builds. |
| 146 | * This is only used on the windows side. Just set it to 0. |
| 147 | */ |
| 148 | u16 revision; |
| 149 | } __packed; |
| 150 | |
| 151 | /* Channel Property Flags */ |
| 152 | #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1 |
| 153 | #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2 |
| 154 | |
| 155 | struct vstor_packet { |
| 156 | /* Requested operation type */ |
| 157 | enum vstor_packet_operation operation; |
| 158 | |
| 159 | /* Flags - see below for values */ |
| 160 | u32 flags; |
| 161 | |
| 162 | /* Status of the request returned from the server side. */ |
| 163 | u32 status; |
| 164 | |
| 165 | /* Data payload area */ |
| 166 | union { |
| 167 | /* |
| 168 | * Structure used to forward SCSI commands from the |
| 169 | * client to the server. |
| 170 | */ |
| 171 | struct vmscsi_request vm_srb; |
| 172 | |
| 173 | /* Structure used to query channel properties. */ |
| 174 | struct vmstorage_channel_properties storage_channel_properties; |
| 175 | |
| 176 | /* Used during version negotiations. */ |
| 177 | struct vmstorage_protocol_version version; |
| 178 | }; |
| 179 | } __packed; |
| 180 | |
| 181 | /* |
| 182 | * Packet Flags: |
| 183 | * |
| 184 | * This flag indicates that the server should send back a completion for this |
| 185 | * packet. |
| 186 | */ |
| 187 | |
| 188 | #define REQUEST_COMPLETION_FLAG 0x1 |
| 189 | |
| 190 | /* Matches Windows-end */ |
| 191 | enum storvsc_request_type { |
| 192 | WRITE_TYPE = 0, |
| 193 | READ_TYPE, |
| 194 | UNKNOWN_TYPE, |
| 195 | }; |
| 196 | |
| 197 | /* |
| 198 | * SRB status codes and masks; a subset of the codes used here. |
| 199 | */ |
| 200 | |
| 201 | #define SRB_STATUS_AUTOSENSE_VALID 0x80 |
| 202 | #define SRB_STATUS_INVALID_LUN 0x20 |
| 203 | #define SRB_STATUS_SUCCESS 0x01 |
| 204 | #define SRB_STATUS_ABORTED 0x02 |
| 205 | #define SRB_STATUS_ERROR 0x04 |
| 206 | |
| 207 | /* |
| 208 | * This is the end of Protocol specific defines. |
| 209 | */ |
| 210 | |
| 211 | |
| 212 | /* |
| 213 | * We setup a mempool to allocate request structures for this driver |
| 214 | * on a per-lun basis. The following define specifies the number of |
| 215 | * elements in the pool. |
| 216 | */ |
| 217 | |
| 218 | #define STORVSC_MIN_BUF_NR 64 |
| 219 | static int storvsc_ringbuffer_size = (20 * PAGE_SIZE); |
| 220 | |
| 221 | module_param(storvsc_ringbuffer_size, int, S_IRUGO); |
| 222 | MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)"); |
| 223 | |
| 224 | #define STORVSC_MAX_IO_REQUESTS 128 |
| 225 | |
| 226 | /* |
| 227 | * In Hyper-V, each port/path/target maps to 1 scsi host adapter. In |
| 228 | * reality, the path/target is not used (ie always set to 0) so our |
| 229 | * scsi host adapter essentially has 1 bus with 1 target that contains |
| 230 | * up to 256 luns. |
| 231 | */ |
| 232 | #define STORVSC_MAX_LUNS_PER_TARGET 64 |
| 233 | #define STORVSC_MAX_TARGETS 1 |
| 234 | #define STORVSC_MAX_CHANNELS 1 |
| 235 | |
| 236 | |
| 237 | |
| 238 | struct storvsc_cmd_request { |
| 239 | struct list_head entry; |
| 240 | struct scsi_cmnd *cmd; |
| 241 | |
| 242 | unsigned int bounce_sgl_count; |
| 243 | struct scatterlist *bounce_sgl; |
| 244 | |
| 245 | struct hv_device *device; |
| 246 | |
| 247 | /* Synchronize the request/response if needed */ |
| 248 | struct completion wait_event; |
| 249 | |
| 250 | unsigned char *sense_buffer; |
| 251 | struct hv_multipage_buffer data_buffer; |
| 252 | struct vstor_packet vstor_packet; |
| 253 | }; |
| 254 | |
| 255 | |
| 256 | /* A storvsc device is a device object that contains a vmbus channel */ |
| 257 | struct storvsc_device { |
| 258 | struct hv_device *device; |
| 259 | |
| 260 | bool destroy; |
| 261 | bool drain_notify; |
| 262 | atomic_t num_outstanding_req; |
| 263 | struct Scsi_Host *host; |
| 264 | |
| 265 | wait_queue_head_t waiting_to_drain; |
| 266 | |
| 267 | /* |
| 268 | * Each unique Port/Path/Target represents 1 channel ie scsi |
| 269 | * controller. In reality, the pathid, targetid is always 0 |
| 270 | * and the port is set by us |
| 271 | */ |
| 272 | unsigned int port_number; |
| 273 | unsigned char path_id; |
| 274 | unsigned char target_id; |
| 275 | |
| 276 | /* Used for vsc/vsp channel reset process */ |
| 277 | struct storvsc_cmd_request init_request; |
| 278 | struct storvsc_cmd_request reset_request; |
| 279 | }; |
| 280 | |
| 281 | struct stor_mem_pools { |
| 282 | struct kmem_cache *request_pool; |
| 283 | mempool_t *request_mempool; |
| 284 | }; |
| 285 | |
| 286 | struct hv_host_device { |
| 287 | struct hv_device *dev; |
| 288 | unsigned int port; |
| 289 | unsigned char path; |
| 290 | unsigned char target; |
| 291 | }; |
| 292 | |
| 293 | struct storvsc_scan_work { |
| 294 | struct work_struct work; |
| 295 | struct Scsi_Host *host; |
| 296 | uint lun; |
| 297 | }; |
| 298 | |
| 299 | static void storvsc_device_scan(struct work_struct *work) |
| 300 | { |
| 301 | struct storvsc_scan_work *wrk; |
| 302 | uint lun; |
| 303 | struct scsi_device *sdev; |
| 304 | |
| 305 | wrk = container_of(work, struct storvsc_scan_work, work); |
| 306 | lun = wrk->lun; |
| 307 | |
| 308 | sdev = scsi_device_lookup(wrk->host, 0, 0, lun); |
| 309 | if (!sdev) |
| 310 | goto done; |
| 311 | scsi_rescan_device(&sdev->sdev_gendev); |
| 312 | scsi_device_put(sdev); |
| 313 | |
| 314 | done: |
| 315 | kfree(wrk); |
| 316 | } |
| 317 | |
| 318 | static void storvsc_bus_scan(struct work_struct *work) |
| 319 | { |
| 320 | struct storvsc_scan_work *wrk; |
| 321 | int id, order_id; |
| 322 | |
| 323 | wrk = container_of(work, struct storvsc_scan_work, work); |
| 324 | for (id = 0; id < wrk->host->max_id; ++id) { |
| 325 | if (wrk->host->reverse_ordering) |
| 326 | order_id = wrk->host->max_id - id - 1; |
| 327 | else |
| 328 | order_id = id; |
| 329 | |
| 330 | scsi_scan_target(&wrk->host->shost_gendev, 0, |
| 331 | order_id, SCAN_WILD_CARD, 1); |
| 332 | } |
| 333 | kfree(wrk); |
| 334 | } |
| 335 | |
| 336 | static void storvsc_remove_lun(struct work_struct *work) |
| 337 | { |
| 338 | struct storvsc_scan_work *wrk; |
| 339 | struct scsi_device *sdev; |
| 340 | |
| 341 | wrk = container_of(work, struct storvsc_scan_work, work); |
| 342 | if (!scsi_host_get(wrk->host)) |
| 343 | goto done; |
| 344 | |
| 345 | sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun); |
| 346 | |
| 347 | if (sdev) { |
| 348 | scsi_remove_device(sdev); |
| 349 | scsi_device_put(sdev); |
| 350 | } |
| 351 | scsi_host_put(wrk->host); |
| 352 | |
| 353 | done: |
| 354 | kfree(wrk); |
| 355 | } |
| 356 | |
| 357 | /* |
| 358 | * Major/minor macros. Minor version is in LSB, meaning that earlier flat |
| 359 | * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1). |
| 360 | */ |
| 361 | |
| 362 | static inline u16 storvsc_get_version(u8 major, u8 minor) |
| 363 | { |
| 364 | u16 version; |
| 365 | |
| 366 | version = ((major << 8) | minor); |
| 367 | return version; |
| 368 | } |
| 369 | |
| 370 | /* |
| 371 | * We can get incoming messages from the host that are not in response to |
| 372 | * messages that we have sent out. An example of this would be messages |
| 373 | * received by the guest to notify dynamic addition/removal of LUNs. To |
| 374 | * deal with potential race conditions where the driver may be in the |
| 375 | * midst of being unloaded when we might receive an unsolicited message |
| 376 | * from the host, we have implemented a mechanism to gurantee sequential |
| 377 | * consistency: |
| 378 | * |
| 379 | * 1) Once the device is marked as being destroyed, we will fail all |
| 380 | * outgoing messages. |
| 381 | * 2) We permit incoming messages when the device is being destroyed, |
| 382 | * only to properly account for messages already sent out. |
| 383 | */ |
| 384 | |
| 385 | static inline struct storvsc_device *get_out_stor_device( |
| 386 | struct hv_device *device) |
| 387 | { |
| 388 | struct storvsc_device *stor_device; |
| 389 | |
| 390 | stor_device = hv_get_drvdata(device); |
| 391 | |
| 392 | if (stor_device && stor_device->destroy) |
| 393 | stor_device = NULL; |
| 394 | |
| 395 | return stor_device; |
| 396 | } |
| 397 | |
| 398 | |
| 399 | static inline void storvsc_wait_to_drain(struct storvsc_device *dev) |
| 400 | { |
| 401 | dev->drain_notify = true; |
| 402 | wait_event(dev->waiting_to_drain, |
| 403 | atomic_read(&dev->num_outstanding_req) == 0); |
| 404 | dev->drain_notify = false; |
| 405 | } |
| 406 | |
| 407 | static inline struct storvsc_device *get_in_stor_device( |
| 408 | struct hv_device *device) |
| 409 | { |
| 410 | struct storvsc_device *stor_device; |
| 411 | |
| 412 | stor_device = hv_get_drvdata(device); |
| 413 | |
| 414 | if (!stor_device) |
| 415 | goto get_in_err; |
| 416 | |
| 417 | /* |
| 418 | * If the device is being destroyed; allow incoming |
| 419 | * traffic only to cleanup outstanding requests. |
| 420 | */ |
| 421 | |
| 422 | if (stor_device->destroy && |
| 423 | (atomic_read(&stor_device->num_outstanding_req) == 0)) |
| 424 | stor_device = NULL; |
| 425 | |
| 426 | get_in_err: |
| 427 | return stor_device; |
| 428 | |
| 429 | } |
| 430 | |
| 431 | static void destroy_bounce_buffer(struct scatterlist *sgl, |
| 432 | unsigned int sg_count) |
| 433 | { |
| 434 | int i; |
| 435 | struct page *page_buf; |
| 436 | |
| 437 | for (i = 0; i < sg_count; i++) { |
| 438 | page_buf = sg_page((&sgl[i])); |
| 439 | if (page_buf != NULL) |
| 440 | __free_page(page_buf); |
| 441 | } |
| 442 | |
| 443 | kfree(sgl); |
| 444 | } |
| 445 | |
| 446 | static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count) |
| 447 | { |
| 448 | int i; |
| 449 | |
| 450 | /* No need to check */ |
| 451 | if (sg_count < 2) |
| 452 | return -1; |
| 453 | |
| 454 | /* We have at least 2 sg entries */ |
| 455 | for (i = 0; i < sg_count; i++) { |
| 456 | if (i == 0) { |
| 457 | /* make sure 1st one does not have hole */ |
| 458 | if (sgl[i].offset + sgl[i].length != PAGE_SIZE) |
| 459 | return i; |
| 460 | } else if (i == sg_count - 1) { |
| 461 | /* make sure last one does not have hole */ |
| 462 | if (sgl[i].offset != 0) |
| 463 | return i; |
| 464 | } else { |
| 465 | /* make sure no hole in the middle */ |
| 466 | if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0) |
| 467 | return i; |
| 468 | } |
| 469 | } |
| 470 | return -1; |
| 471 | } |
| 472 | |
| 473 | static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl, |
| 474 | unsigned int sg_count, |
| 475 | unsigned int len, |
| 476 | int write) |
| 477 | { |
| 478 | int i; |
| 479 | int num_pages; |
| 480 | struct scatterlist *bounce_sgl; |
| 481 | struct page *page_buf; |
| 482 | unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE); |
| 483 | |
| 484 | num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT; |
| 485 | |
| 486 | bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC); |
| 487 | if (!bounce_sgl) |
| 488 | return NULL; |
| 489 | |
| 490 | sg_init_table(bounce_sgl, num_pages); |
| 491 | for (i = 0; i < num_pages; i++) { |
| 492 | page_buf = alloc_page(GFP_ATOMIC); |
| 493 | if (!page_buf) |
| 494 | goto cleanup; |
| 495 | sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0); |
| 496 | } |
| 497 | |
| 498 | return bounce_sgl; |
| 499 | |
| 500 | cleanup: |
| 501 | destroy_bounce_buffer(bounce_sgl, num_pages); |
| 502 | return NULL; |
| 503 | } |
| 504 | |
| 505 | /* Disgusting wrapper functions */ |
| 506 | static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx) |
| 507 | { |
| 508 | void *addr = kmap_atomic(sg_page(sgl + idx)); |
| 509 | return (unsigned long)addr; |
| 510 | } |
| 511 | |
| 512 | static inline void sg_kunmap_atomic(unsigned long addr) |
| 513 | { |
| 514 | kunmap_atomic((void *)addr); |
| 515 | } |
| 516 | |
| 517 | |
| 518 | /* Assume the original sgl has enough room */ |
| 519 | static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl, |
| 520 | struct scatterlist *bounce_sgl, |
| 521 | unsigned int orig_sgl_count, |
| 522 | unsigned int bounce_sgl_count) |
| 523 | { |
| 524 | int i; |
| 525 | int j = 0; |
| 526 | unsigned long src, dest; |
| 527 | unsigned int srclen, destlen, copylen; |
| 528 | unsigned int total_copied = 0; |
| 529 | unsigned long bounce_addr = 0; |
| 530 | unsigned long dest_addr = 0; |
| 531 | unsigned long flags; |
| 532 | |
| 533 | local_irq_save(flags); |
| 534 | |
| 535 | for (i = 0; i < orig_sgl_count; i++) { |
| 536 | dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset; |
| 537 | dest = dest_addr; |
| 538 | destlen = orig_sgl[i].length; |
| 539 | |
| 540 | if (bounce_addr == 0) |
| 541 | bounce_addr = sg_kmap_atomic(bounce_sgl,j); |
| 542 | |
| 543 | while (destlen) { |
| 544 | src = bounce_addr + bounce_sgl[j].offset; |
| 545 | srclen = bounce_sgl[j].length - bounce_sgl[j].offset; |
| 546 | |
| 547 | copylen = min(srclen, destlen); |
| 548 | memcpy((void *)dest, (void *)src, copylen); |
| 549 | |
| 550 | total_copied += copylen; |
| 551 | bounce_sgl[j].offset += copylen; |
| 552 | destlen -= copylen; |
| 553 | dest += copylen; |
| 554 | |
| 555 | if (bounce_sgl[j].offset == bounce_sgl[j].length) { |
| 556 | /* full */ |
| 557 | sg_kunmap_atomic(bounce_addr); |
| 558 | j++; |
| 559 | |
| 560 | /* |
| 561 | * It is possible that the number of elements |
| 562 | * in the bounce buffer may not be equal to |
| 563 | * the number of elements in the original |
| 564 | * scatter list. Handle this correctly. |
| 565 | */ |
| 566 | |
| 567 | if (j == bounce_sgl_count) { |
| 568 | /* |
| 569 | * We are done; cleanup and return. |
| 570 | */ |
| 571 | sg_kunmap_atomic(dest_addr - orig_sgl[i].offset); |
| 572 | local_irq_restore(flags); |
| 573 | return total_copied; |
| 574 | } |
| 575 | |
| 576 | /* if we need to use another bounce buffer */ |
| 577 | if (destlen || i != orig_sgl_count - 1) |
| 578 | bounce_addr = sg_kmap_atomic(bounce_sgl,j); |
| 579 | } else if (destlen == 0 && i == orig_sgl_count - 1) { |
| 580 | /* unmap the last bounce that is < PAGE_SIZE */ |
| 581 | sg_kunmap_atomic(bounce_addr); |
| 582 | } |
| 583 | } |
| 584 | |
| 585 | sg_kunmap_atomic(dest_addr - orig_sgl[i].offset); |
| 586 | } |
| 587 | |
| 588 | local_irq_restore(flags); |
| 589 | |
| 590 | return total_copied; |
| 591 | } |
| 592 | |
| 593 | /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */ |
| 594 | static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl, |
| 595 | struct scatterlist *bounce_sgl, |
| 596 | unsigned int orig_sgl_count) |
| 597 | { |
| 598 | int i; |
| 599 | int j = 0; |
| 600 | unsigned long src, dest; |
| 601 | unsigned int srclen, destlen, copylen; |
| 602 | unsigned int total_copied = 0; |
| 603 | unsigned long bounce_addr = 0; |
| 604 | unsigned long src_addr = 0; |
| 605 | unsigned long flags; |
| 606 | |
| 607 | local_irq_save(flags); |
| 608 | |
| 609 | for (i = 0; i < orig_sgl_count; i++) { |
| 610 | src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset; |
| 611 | src = src_addr; |
| 612 | srclen = orig_sgl[i].length; |
| 613 | |
| 614 | if (bounce_addr == 0) |
| 615 | bounce_addr = sg_kmap_atomic(bounce_sgl,j); |
| 616 | |
| 617 | while (srclen) { |
| 618 | /* assume bounce offset always == 0 */ |
| 619 | dest = bounce_addr + bounce_sgl[j].length; |
| 620 | destlen = PAGE_SIZE - bounce_sgl[j].length; |
| 621 | |
| 622 | copylen = min(srclen, destlen); |
| 623 | memcpy((void *)dest, (void *)src, copylen); |
| 624 | |
| 625 | total_copied += copylen; |
| 626 | bounce_sgl[j].length += copylen; |
| 627 | srclen -= copylen; |
| 628 | src += copylen; |
| 629 | |
| 630 | if (bounce_sgl[j].length == PAGE_SIZE) { |
| 631 | /* full..move to next entry */ |
| 632 | sg_kunmap_atomic(bounce_addr); |
| 633 | j++; |
| 634 | |
| 635 | /* if we need to use another bounce buffer */ |
| 636 | if (srclen || i != orig_sgl_count - 1) |
| 637 | bounce_addr = sg_kmap_atomic(bounce_sgl,j); |
| 638 | |
| 639 | } else if (srclen == 0 && i == orig_sgl_count - 1) { |
| 640 | /* unmap the last bounce that is < PAGE_SIZE */ |
| 641 | sg_kunmap_atomic(bounce_addr); |
| 642 | } |
| 643 | } |
| 644 | |
| 645 | sg_kunmap_atomic(src_addr - orig_sgl[i].offset); |
| 646 | } |
| 647 | |
| 648 | local_irq_restore(flags); |
| 649 | |
| 650 | return total_copied; |
| 651 | } |
| 652 | |
| 653 | static int storvsc_channel_init(struct hv_device *device) |
| 654 | { |
| 655 | struct storvsc_device *stor_device; |
| 656 | struct storvsc_cmd_request *request; |
| 657 | struct vstor_packet *vstor_packet; |
| 658 | int ret, t; |
| 659 | |
| 660 | stor_device = get_out_stor_device(device); |
| 661 | if (!stor_device) |
| 662 | return -ENODEV; |
| 663 | |
| 664 | request = &stor_device->init_request; |
| 665 | vstor_packet = &request->vstor_packet; |
| 666 | |
| 667 | /* |
| 668 | * Now, initiate the vsc/vsp initialization protocol on the open |
| 669 | * channel |
| 670 | */ |
| 671 | memset(request, 0, sizeof(struct storvsc_cmd_request)); |
| 672 | init_completion(&request->wait_event); |
| 673 | vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION; |
| 674 | vstor_packet->flags = REQUEST_COMPLETION_FLAG; |
| 675 | |
| 676 | ret = vmbus_sendpacket(device->channel, vstor_packet, |
| 677 | sizeof(struct vstor_packet), |
| 678 | (unsigned long)request, |
| 679 | VM_PKT_DATA_INBAND, |
| 680 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); |
| 681 | if (ret != 0) |
| 682 | goto cleanup; |
| 683 | |
| 684 | t = wait_for_completion_timeout(&request->wait_event, 5*HZ); |
| 685 | if (t == 0) { |
| 686 | ret = -ETIMEDOUT; |
| 687 | goto cleanup; |
| 688 | } |
| 689 | |
| 690 | if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || |
| 691 | vstor_packet->status != 0) |
| 692 | goto cleanup; |
| 693 | |
| 694 | |
| 695 | /* reuse the packet for version range supported */ |
| 696 | memset(vstor_packet, 0, sizeof(struct vstor_packet)); |
| 697 | vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION; |
| 698 | vstor_packet->flags = REQUEST_COMPLETION_FLAG; |
| 699 | |
| 700 | vstor_packet->version.major_minor = |
| 701 | storvsc_get_version(VMSTOR_CURRENT_MAJOR, VMSTOR_CURRENT_MINOR); |
| 702 | |
| 703 | /* |
| 704 | * The revision number is only used in Windows; set it to 0. |
| 705 | */ |
| 706 | vstor_packet->version.revision = 0; |
| 707 | |
| 708 | ret = vmbus_sendpacket(device->channel, vstor_packet, |
| 709 | sizeof(struct vstor_packet), |
| 710 | (unsigned long)request, |
| 711 | VM_PKT_DATA_INBAND, |
| 712 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); |
| 713 | if (ret != 0) |
| 714 | goto cleanup; |
| 715 | |
| 716 | t = wait_for_completion_timeout(&request->wait_event, 5*HZ); |
| 717 | if (t == 0) { |
| 718 | ret = -ETIMEDOUT; |
| 719 | goto cleanup; |
| 720 | } |
| 721 | |
| 722 | if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || |
| 723 | vstor_packet->status != 0) |
| 724 | goto cleanup; |
| 725 | |
| 726 | |
| 727 | memset(vstor_packet, 0, sizeof(struct vstor_packet)); |
| 728 | vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES; |
| 729 | vstor_packet->flags = REQUEST_COMPLETION_FLAG; |
| 730 | vstor_packet->storage_channel_properties.port_number = |
| 731 | stor_device->port_number; |
| 732 | |
| 733 | ret = vmbus_sendpacket(device->channel, vstor_packet, |
| 734 | sizeof(struct vstor_packet), |
| 735 | (unsigned long)request, |
| 736 | VM_PKT_DATA_INBAND, |
| 737 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); |
| 738 | |
| 739 | if (ret != 0) |
| 740 | goto cleanup; |
| 741 | |
| 742 | t = wait_for_completion_timeout(&request->wait_event, 5*HZ); |
| 743 | if (t == 0) { |
| 744 | ret = -ETIMEDOUT; |
| 745 | goto cleanup; |
| 746 | } |
| 747 | |
| 748 | if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || |
| 749 | vstor_packet->status != 0) |
| 750 | goto cleanup; |
| 751 | |
| 752 | stor_device->path_id = vstor_packet->storage_channel_properties.path_id; |
| 753 | stor_device->target_id |
| 754 | = vstor_packet->storage_channel_properties.target_id; |
| 755 | |
| 756 | memset(vstor_packet, 0, sizeof(struct vstor_packet)); |
| 757 | vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION; |
| 758 | vstor_packet->flags = REQUEST_COMPLETION_FLAG; |
| 759 | |
| 760 | ret = vmbus_sendpacket(device->channel, vstor_packet, |
| 761 | sizeof(struct vstor_packet), |
| 762 | (unsigned long)request, |
| 763 | VM_PKT_DATA_INBAND, |
| 764 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); |
| 765 | |
| 766 | if (ret != 0) |
| 767 | goto cleanup; |
| 768 | |
| 769 | t = wait_for_completion_timeout(&request->wait_event, 5*HZ); |
| 770 | if (t == 0) { |
| 771 | ret = -ETIMEDOUT; |
| 772 | goto cleanup; |
| 773 | } |
| 774 | |
| 775 | if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || |
| 776 | vstor_packet->status != 0) |
| 777 | goto cleanup; |
| 778 | |
| 779 | |
| 780 | cleanup: |
| 781 | return ret; |
| 782 | } |
| 783 | |
| 784 | static void storvsc_handle_error(struct vmscsi_request *vm_srb, |
| 785 | struct scsi_cmnd *scmnd, |
| 786 | struct Scsi_Host *host, |
| 787 | u8 asc, u8 ascq) |
| 788 | { |
| 789 | struct storvsc_scan_work *wrk; |
| 790 | void (*process_err_fn)(struct work_struct *work); |
| 791 | bool do_work = false; |
| 792 | |
| 793 | switch (vm_srb->srb_status) { |
| 794 | case SRB_STATUS_ERROR: |
| 795 | /* |
| 796 | * If there is an error; offline the device since all |
| 797 | * error recovery strategies would have already been |
| 798 | * deployed on the host side. However, if the command |
| 799 | * were a pass-through command deal with it appropriately. |
| 800 | */ |
| 801 | switch (scmnd->cmnd[0]) { |
| 802 | case ATA_16: |
| 803 | case ATA_12: |
| 804 | set_host_byte(scmnd, DID_PASSTHROUGH); |
| 805 | break; |
| 806 | default: |
| 807 | set_host_byte(scmnd, DID_TARGET_FAILURE); |
| 808 | } |
| 809 | break; |
| 810 | case SRB_STATUS_INVALID_LUN: |
| 811 | do_work = true; |
| 812 | process_err_fn = storvsc_remove_lun; |
| 813 | break; |
| 814 | case (SRB_STATUS_ABORTED | SRB_STATUS_AUTOSENSE_VALID): |
| 815 | if ((asc == 0x2a) && (ascq == 0x9)) { |
| 816 | do_work = true; |
| 817 | process_err_fn = storvsc_device_scan; |
| 818 | /* |
| 819 | * Retry the I/O that trigerred this. |
| 820 | */ |
| 821 | set_host_byte(scmnd, DID_REQUEUE); |
| 822 | } |
| 823 | break; |
| 824 | } |
| 825 | |
| 826 | if (!do_work) |
| 827 | return; |
| 828 | |
| 829 | /* |
| 830 | * We need to schedule work to process this error; schedule it. |
| 831 | */ |
| 832 | wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC); |
| 833 | if (!wrk) { |
| 834 | set_host_byte(scmnd, DID_TARGET_FAILURE); |
| 835 | return; |
| 836 | } |
| 837 | |
| 838 | wrk->host = host; |
| 839 | wrk->lun = vm_srb->lun; |
| 840 | INIT_WORK(&wrk->work, process_err_fn); |
| 841 | schedule_work(&wrk->work); |
| 842 | } |
| 843 | |
| 844 | |
| 845 | static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request) |
| 846 | { |
| 847 | struct scsi_cmnd *scmnd = cmd_request->cmd; |
| 848 | struct hv_host_device *host_dev = shost_priv(scmnd->device->host); |
| 849 | void (*scsi_done_fn)(struct scsi_cmnd *); |
| 850 | struct scsi_sense_hdr sense_hdr; |
| 851 | struct vmscsi_request *vm_srb; |
| 852 | struct stor_mem_pools *memp = scmnd->device->hostdata; |
| 853 | struct Scsi_Host *host; |
| 854 | struct storvsc_device *stor_dev; |
| 855 | struct hv_device *dev = host_dev->dev; |
| 856 | |
| 857 | stor_dev = get_in_stor_device(dev); |
| 858 | host = stor_dev->host; |
| 859 | |
| 860 | vm_srb = &cmd_request->vstor_packet.vm_srb; |
| 861 | if (cmd_request->bounce_sgl_count) { |
| 862 | if (vm_srb->data_in == READ_TYPE) |
| 863 | copy_from_bounce_buffer(scsi_sglist(scmnd), |
| 864 | cmd_request->bounce_sgl, |
| 865 | scsi_sg_count(scmnd), |
| 866 | cmd_request->bounce_sgl_count); |
| 867 | destroy_bounce_buffer(cmd_request->bounce_sgl, |
| 868 | cmd_request->bounce_sgl_count); |
| 869 | } |
| 870 | |
| 871 | scmnd->result = vm_srb->scsi_status; |
| 872 | |
| 873 | if (scmnd->result) { |
| 874 | if (scsi_normalize_sense(scmnd->sense_buffer, |
| 875 | SCSI_SENSE_BUFFERSIZE, &sense_hdr)) |
| 876 | scsi_print_sense_hdr("storvsc", &sense_hdr); |
| 877 | } |
| 878 | |
| 879 | if (vm_srb->srb_status != SRB_STATUS_SUCCESS) |
| 880 | storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc, |
| 881 | sense_hdr.ascq); |
| 882 | |
| 883 | scsi_set_resid(scmnd, |
| 884 | cmd_request->data_buffer.len - |
| 885 | vm_srb->data_transfer_length); |
| 886 | |
| 887 | scsi_done_fn = scmnd->scsi_done; |
| 888 | |
| 889 | scmnd->host_scribble = NULL; |
| 890 | scmnd->scsi_done = NULL; |
| 891 | |
| 892 | scsi_done_fn(scmnd); |
| 893 | |
| 894 | mempool_free(cmd_request, memp->request_mempool); |
| 895 | } |
| 896 | |
| 897 | static void storvsc_on_io_completion(struct hv_device *device, |
| 898 | struct vstor_packet *vstor_packet, |
| 899 | struct storvsc_cmd_request *request) |
| 900 | { |
| 901 | struct storvsc_device *stor_device; |
| 902 | struct vstor_packet *stor_pkt; |
| 903 | |
| 904 | stor_device = hv_get_drvdata(device); |
| 905 | stor_pkt = &request->vstor_packet; |
| 906 | |
| 907 | /* |
| 908 | * The current SCSI handling on the host side does |
| 909 | * not correctly handle: |
| 910 | * INQUIRY command with page code parameter set to 0x80 |
| 911 | * MODE_SENSE command with cmd[2] == 0x1c |
| 912 | * |
| 913 | * Setup srb and scsi status so this won't be fatal. |
| 914 | * We do this so we can distinguish truly fatal failues |
| 915 | * (srb status == 0x4) and off-line the device in that case. |
| 916 | */ |
| 917 | |
| 918 | if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) || |
| 919 | (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) { |
| 920 | vstor_packet->vm_srb.scsi_status = 0; |
| 921 | vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS; |
| 922 | } |
| 923 | |
| 924 | |
| 925 | /* Copy over the status...etc */ |
| 926 | stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status; |
| 927 | stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status; |
| 928 | stor_pkt->vm_srb.sense_info_length = |
| 929 | vstor_packet->vm_srb.sense_info_length; |
| 930 | |
| 931 | if (vstor_packet->vm_srb.scsi_status != 0 || |
| 932 | vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS){ |
| 933 | dev_warn(&device->device, |
| 934 | "cmd 0x%x scsi status 0x%x srb status 0x%x\n", |
| 935 | stor_pkt->vm_srb.cdb[0], |
| 936 | vstor_packet->vm_srb.scsi_status, |
| 937 | vstor_packet->vm_srb.srb_status); |
| 938 | } |
| 939 | |
| 940 | if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) { |
| 941 | /* CHECK_CONDITION */ |
| 942 | if (vstor_packet->vm_srb.srb_status & |
| 943 | SRB_STATUS_AUTOSENSE_VALID) { |
| 944 | /* autosense data available */ |
| 945 | dev_warn(&device->device, |
| 946 | "stor pkt %p autosense data valid - len %d\n", |
| 947 | request, |
| 948 | vstor_packet->vm_srb.sense_info_length); |
| 949 | |
| 950 | memcpy(request->sense_buffer, |
| 951 | vstor_packet->vm_srb.sense_data, |
| 952 | vstor_packet->vm_srb.sense_info_length); |
| 953 | |
| 954 | } |
| 955 | } |
| 956 | |
| 957 | stor_pkt->vm_srb.data_transfer_length = |
| 958 | vstor_packet->vm_srb.data_transfer_length; |
| 959 | |
| 960 | storvsc_command_completion(request); |
| 961 | |
| 962 | if (atomic_dec_and_test(&stor_device->num_outstanding_req) && |
| 963 | stor_device->drain_notify) |
| 964 | wake_up(&stor_device->waiting_to_drain); |
| 965 | |
| 966 | |
| 967 | } |
| 968 | |
| 969 | static void storvsc_on_receive(struct hv_device *device, |
| 970 | struct vstor_packet *vstor_packet, |
| 971 | struct storvsc_cmd_request *request) |
| 972 | { |
| 973 | struct storvsc_scan_work *work; |
| 974 | struct storvsc_device *stor_device; |
| 975 | |
| 976 | switch (vstor_packet->operation) { |
| 977 | case VSTOR_OPERATION_COMPLETE_IO: |
| 978 | storvsc_on_io_completion(device, vstor_packet, request); |
| 979 | break; |
| 980 | |
| 981 | case VSTOR_OPERATION_REMOVE_DEVICE: |
| 982 | case VSTOR_OPERATION_ENUMERATE_BUS: |
| 983 | stor_device = get_in_stor_device(device); |
| 984 | work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC); |
| 985 | if (!work) |
| 986 | return; |
| 987 | |
| 988 | INIT_WORK(&work->work, storvsc_bus_scan); |
| 989 | work->host = stor_device->host; |
| 990 | schedule_work(&work->work); |
| 991 | break; |
| 992 | |
| 993 | default: |
| 994 | break; |
| 995 | } |
| 996 | } |
| 997 | |
| 998 | static void storvsc_on_channel_callback(void *context) |
| 999 | { |
| 1000 | struct hv_device *device = (struct hv_device *)context; |
| 1001 | struct storvsc_device *stor_device; |
| 1002 | u32 bytes_recvd; |
| 1003 | u64 request_id; |
| 1004 | unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)]; |
| 1005 | struct storvsc_cmd_request *request; |
| 1006 | int ret; |
| 1007 | |
| 1008 | |
| 1009 | stor_device = get_in_stor_device(device); |
| 1010 | if (!stor_device) |
| 1011 | return; |
| 1012 | |
| 1013 | do { |
| 1014 | ret = vmbus_recvpacket(device->channel, packet, |
| 1015 | ALIGN(sizeof(struct vstor_packet), 8), |
| 1016 | &bytes_recvd, &request_id); |
| 1017 | if (ret == 0 && bytes_recvd > 0) { |
| 1018 | |
| 1019 | request = (struct storvsc_cmd_request *) |
| 1020 | (unsigned long)request_id; |
| 1021 | |
| 1022 | if ((request == &stor_device->init_request) || |
| 1023 | (request == &stor_device->reset_request)) { |
| 1024 | |
| 1025 | memcpy(&request->vstor_packet, packet, |
| 1026 | sizeof(struct vstor_packet)); |
| 1027 | complete(&request->wait_event); |
| 1028 | } else { |
| 1029 | storvsc_on_receive(device, |
| 1030 | (struct vstor_packet *)packet, |
| 1031 | request); |
| 1032 | } |
| 1033 | } else { |
| 1034 | break; |
| 1035 | } |
| 1036 | } while (1); |
| 1037 | |
| 1038 | return; |
| 1039 | } |
| 1040 | |
| 1041 | static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size) |
| 1042 | { |
| 1043 | struct vmstorage_channel_properties props; |
| 1044 | int ret; |
| 1045 | |
| 1046 | memset(&props, 0, sizeof(struct vmstorage_channel_properties)); |
| 1047 | |
| 1048 | ret = vmbus_open(device->channel, |
| 1049 | ring_size, |
| 1050 | ring_size, |
| 1051 | (void *)&props, |
| 1052 | sizeof(struct vmstorage_channel_properties), |
| 1053 | storvsc_on_channel_callback, device); |
| 1054 | |
| 1055 | if (ret != 0) |
| 1056 | return ret; |
| 1057 | |
| 1058 | ret = storvsc_channel_init(device); |
| 1059 | |
| 1060 | return ret; |
| 1061 | } |
| 1062 | |
| 1063 | static int storvsc_dev_remove(struct hv_device *device) |
| 1064 | { |
| 1065 | struct storvsc_device *stor_device; |
| 1066 | unsigned long flags; |
| 1067 | |
| 1068 | stor_device = hv_get_drvdata(device); |
| 1069 | |
| 1070 | spin_lock_irqsave(&device->channel->inbound_lock, flags); |
| 1071 | stor_device->destroy = true; |
| 1072 | spin_unlock_irqrestore(&device->channel->inbound_lock, flags); |
| 1073 | |
| 1074 | /* |
| 1075 | * At this point, all outbound traffic should be disable. We |
| 1076 | * only allow inbound traffic (responses) to proceed so that |
| 1077 | * outstanding requests can be completed. |
| 1078 | */ |
| 1079 | |
| 1080 | storvsc_wait_to_drain(stor_device); |
| 1081 | |
| 1082 | /* |
| 1083 | * Since we have already drained, we don't need to busy wait |
| 1084 | * as was done in final_release_stor_device() |
| 1085 | * Note that we cannot set the ext pointer to NULL until |
| 1086 | * we have drained - to drain the outgoing packets, we need to |
| 1087 | * allow incoming packets. |
| 1088 | */ |
| 1089 | spin_lock_irqsave(&device->channel->inbound_lock, flags); |
| 1090 | hv_set_drvdata(device, NULL); |
| 1091 | spin_unlock_irqrestore(&device->channel->inbound_lock, flags); |
| 1092 | |
| 1093 | /* Close the channel */ |
| 1094 | vmbus_close(device->channel); |
| 1095 | |
| 1096 | kfree(stor_device); |
| 1097 | return 0; |
| 1098 | } |
| 1099 | |
| 1100 | static int storvsc_do_io(struct hv_device *device, |
| 1101 | struct storvsc_cmd_request *request) |
| 1102 | { |
| 1103 | struct storvsc_device *stor_device; |
| 1104 | struct vstor_packet *vstor_packet; |
| 1105 | int ret = 0; |
| 1106 | |
| 1107 | vstor_packet = &request->vstor_packet; |
| 1108 | stor_device = get_out_stor_device(device); |
| 1109 | |
| 1110 | if (!stor_device) |
| 1111 | return -ENODEV; |
| 1112 | |
| 1113 | |
| 1114 | request->device = device; |
| 1115 | |
| 1116 | |
| 1117 | vstor_packet->flags |= REQUEST_COMPLETION_FLAG; |
| 1118 | |
| 1119 | vstor_packet->vm_srb.length = sizeof(struct vmscsi_request); |
| 1120 | |
| 1121 | |
| 1122 | vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE; |
| 1123 | |
| 1124 | |
| 1125 | vstor_packet->vm_srb.data_transfer_length = |
| 1126 | request->data_buffer.len; |
| 1127 | |
| 1128 | vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB; |
| 1129 | |
| 1130 | if (request->data_buffer.len) { |
| 1131 | ret = vmbus_sendpacket_multipagebuffer(device->channel, |
| 1132 | &request->data_buffer, |
| 1133 | vstor_packet, |
| 1134 | sizeof(struct vstor_packet), |
| 1135 | (unsigned long)request); |
| 1136 | } else { |
| 1137 | ret = vmbus_sendpacket(device->channel, vstor_packet, |
| 1138 | sizeof(struct vstor_packet), |
| 1139 | (unsigned long)request, |
| 1140 | VM_PKT_DATA_INBAND, |
| 1141 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); |
| 1142 | } |
| 1143 | |
| 1144 | if (ret != 0) |
| 1145 | return ret; |
| 1146 | |
| 1147 | atomic_inc(&stor_device->num_outstanding_req); |
| 1148 | |
| 1149 | return ret; |
| 1150 | } |
| 1151 | |
| 1152 | static int storvsc_device_alloc(struct scsi_device *sdevice) |
| 1153 | { |
| 1154 | struct stor_mem_pools *memp; |
| 1155 | int number = STORVSC_MIN_BUF_NR; |
| 1156 | |
| 1157 | memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL); |
| 1158 | if (!memp) |
| 1159 | return -ENOMEM; |
| 1160 | |
| 1161 | memp->request_pool = |
| 1162 | kmem_cache_create(dev_name(&sdevice->sdev_dev), |
| 1163 | sizeof(struct storvsc_cmd_request), 0, |
| 1164 | SLAB_HWCACHE_ALIGN, NULL); |
| 1165 | |
| 1166 | if (!memp->request_pool) |
| 1167 | goto err0; |
| 1168 | |
| 1169 | memp->request_mempool = mempool_create(number, mempool_alloc_slab, |
| 1170 | mempool_free_slab, |
| 1171 | memp->request_pool); |
| 1172 | |
| 1173 | if (!memp->request_mempool) |
| 1174 | goto err1; |
| 1175 | |
| 1176 | sdevice->hostdata = memp; |
| 1177 | |
| 1178 | return 0; |
| 1179 | |
| 1180 | err1: |
| 1181 | kmem_cache_destroy(memp->request_pool); |
| 1182 | |
| 1183 | err0: |
| 1184 | kfree(memp); |
| 1185 | return -ENOMEM; |
| 1186 | } |
| 1187 | |
| 1188 | static void storvsc_device_destroy(struct scsi_device *sdevice) |
| 1189 | { |
| 1190 | struct stor_mem_pools *memp = sdevice->hostdata; |
| 1191 | |
| 1192 | mempool_destroy(memp->request_mempool); |
| 1193 | kmem_cache_destroy(memp->request_pool); |
| 1194 | kfree(memp); |
| 1195 | sdevice->hostdata = NULL; |
| 1196 | } |
| 1197 | |
| 1198 | static int storvsc_device_configure(struct scsi_device *sdevice) |
| 1199 | { |
| 1200 | scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG, |
| 1201 | STORVSC_MAX_IO_REQUESTS); |
| 1202 | |
| 1203 | blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE); |
| 1204 | |
| 1205 | blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY); |
| 1206 | |
| 1207 | sdevice->no_write_same = 1; |
| 1208 | |
| 1209 | return 0; |
| 1210 | } |
| 1211 | |
| 1212 | static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev, |
| 1213 | sector_t capacity, int *info) |
| 1214 | { |
| 1215 | sector_t nsect = capacity; |
| 1216 | sector_t cylinders = nsect; |
| 1217 | int heads, sectors_pt; |
| 1218 | |
| 1219 | /* |
| 1220 | * We are making up these values; let us keep it simple. |
| 1221 | */ |
| 1222 | heads = 0xff; |
| 1223 | sectors_pt = 0x3f; /* Sectors per track */ |
| 1224 | sector_div(cylinders, heads * sectors_pt); |
| 1225 | if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect) |
| 1226 | cylinders = 0xffff; |
| 1227 | |
| 1228 | info[0] = heads; |
| 1229 | info[1] = sectors_pt; |
| 1230 | info[2] = (int)cylinders; |
| 1231 | |
| 1232 | return 0; |
| 1233 | } |
| 1234 | |
| 1235 | static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd) |
| 1236 | { |
| 1237 | struct hv_host_device *host_dev = shost_priv(scmnd->device->host); |
| 1238 | struct hv_device *device = host_dev->dev; |
| 1239 | |
| 1240 | struct storvsc_device *stor_device; |
| 1241 | struct storvsc_cmd_request *request; |
| 1242 | struct vstor_packet *vstor_packet; |
| 1243 | int ret, t; |
| 1244 | |
| 1245 | |
| 1246 | stor_device = get_out_stor_device(device); |
| 1247 | if (!stor_device) |
| 1248 | return FAILED; |
| 1249 | |
| 1250 | request = &stor_device->reset_request; |
| 1251 | vstor_packet = &request->vstor_packet; |
| 1252 | |
| 1253 | init_completion(&request->wait_event); |
| 1254 | |
| 1255 | vstor_packet->operation = VSTOR_OPERATION_RESET_BUS; |
| 1256 | vstor_packet->flags = REQUEST_COMPLETION_FLAG; |
| 1257 | vstor_packet->vm_srb.path_id = stor_device->path_id; |
| 1258 | |
| 1259 | ret = vmbus_sendpacket(device->channel, vstor_packet, |
| 1260 | sizeof(struct vstor_packet), |
| 1261 | (unsigned long)&stor_device->reset_request, |
| 1262 | VM_PKT_DATA_INBAND, |
| 1263 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); |
| 1264 | if (ret != 0) |
| 1265 | return FAILED; |
| 1266 | |
| 1267 | t = wait_for_completion_timeout(&request->wait_event, 5*HZ); |
| 1268 | if (t == 0) |
| 1269 | return TIMEOUT_ERROR; |
| 1270 | |
| 1271 | |
| 1272 | /* |
| 1273 | * At this point, all outstanding requests in the adapter |
| 1274 | * should have been flushed out and return to us |
| 1275 | * There is a potential race here where the host may be in |
| 1276 | * the process of responding when we return from here. |
| 1277 | * Just wait for all in-transit packets to be accounted for |
| 1278 | * before we return from here. |
| 1279 | */ |
| 1280 | storvsc_wait_to_drain(stor_device); |
| 1281 | |
| 1282 | return SUCCESS; |
| 1283 | } |
| 1284 | |
| 1285 | static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd) |
| 1286 | { |
| 1287 | bool allowed = true; |
| 1288 | u8 scsi_op = scmnd->cmnd[0]; |
| 1289 | |
| 1290 | switch (scsi_op) { |
| 1291 | /* the host does not handle WRITE_SAME, log accident usage */ |
| 1292 | case WRITE_SAME: |
| 1293 | /* |
| 1294 | * smartd sends this command and the host does not handle |
| 1295 | * this. So, don't send it. |
| 1296 | */ |
| 1297 | case SET_WINDOW: |
| 1298 | scmnd->result = ILLEGAL_REQUEST << 16; |
| 1299 | allowed = false; |
| 1300 | break; |
| 1301 | default: |
| 1302 | break; |
| 1303 | } |
| 1304 | return allowed; |
| 1305 | } |
| 1306 | |
| 1307 | static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd) |
| 1308 | { |
| 1309 | int ret; |
| 1310 | struct hv_host_device *host_dev = shost_priv(host); |
| 1311 | struct hv_device *dev = host_dev->dev; |
| 1312 | struct storvsc_cmd_request *cmd_request; |
| 1313 | unsigned int request_size = 0; |
| 1314 | int i; |
| 1315 | struct scatterlist *sgl; |
| 1316 | unsigned int sg_count = 0; |
| 1317 | struct vmscsi_request *vm_srb; |
| 1318 | struct stor_mem_pools *memp = scmnd->device->hostdata; |
| 1319 | |
| 1320 | if (!storvsc_scsi_cmd_ok(scmnd)) { |
| 1321 | scmnd->scsi_done(scmnd); |
| 1322 | return 0; |
| 1323 | } |
| 1324 | |
| 1325 | request_size = sizeof(struct storvsc_cmd_request); |
| 1326 | |
| 1327 | cmd_request = mempool_alloc(memp->request_mempool, |
| 1328 | GFP_ATOMIC); |
| 1329 | |
| 1330 | /* |
| 1331 | * We might be invoked in an interrupt context; hence |
| 1332 | * mempool_alloc() can fail. |
| 1333 | */ |
| 1334 | if (!cmd_request) |
| 1335 | return SCSI_MLQUEUE_DEVICE_BUSY; |
| 1336 | |
| 1337 | memset(cmd_request, 0, sizeof(struct storvsc_cmd_request)); |
| 1338 | |
| 1339 | /* Setup the cmd request */ |
| 1340 | cmd_request->cmd = scmnd; |
| 1341 | |
| 1342 | scmnd->host_scribble = (unsigned char *)cmd_request; |
| 1343 | |
| 1344 | vm_srb = &cmd_request->vstor_packet.vm_srb; |
| 1345 | |
| 1346 | |
| 1347 | /* Build the SRB */ |
| 1348 | switch (scmnd->sc_data_direction) { |
| 1349 | case DMA_TO_DEVICE: |
| 1350 | vm_srb->data_in = WRITE_TYPE; |
| 1351 | break; |
| 1352 | case DMA_FROM_DEVICE: |
| 1353 | vm_srb->data_in = READ_TYPE; |
| 1354 | break; |
| 1355 | default: |
| 1356 | vm_srb->data_in = UNKNOWN_TYPE; |
| 1357 | break; |
| 1358 | } |
| 1359 | |
| 1360 | |
| 1361 | vm_srb->port_number = host_dev->port; |
| 1362 | vm_srb->path_id = scmnd->device->channel; |
| 1363 | vm_srb->target_id = scmnd->device->id; |
| 1364 | vm_srb->lun = scmnd->device->lun; |
| 1365 | |
| 1366 | vm_srb->cdb_length = scmnd->cmd_len; |
| 1367 | |
| 1368 | memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length); |
| 1369 | |
| 1370 | cmd_request->sense_buffer = scmnd->sense_buffer; |
| 1371 | |
| 1372 | |
| 1373 | cmd_request->data_buffer.len = scsi_bufflen(scmnd); |
| 1374 | if (scsi_sg_count(scmnd)) { |
| 1375 | sgl = (struct scatterlist *)scsi_sglist(scmnd); |
| 1376 | sg_count = scsi_sg_count(scmnd); |
| 1377 | |
| 1378 | /* check if we need to bounce the sgl */ |
| 1379 | if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) { |
| 1380 | cmd_request->bounce_sgl = |
| 1381 | create_bounce_buffer(sgl, scsi_sg_count(scmnd), |
| 1382 | scsi_bufflen(scmnd), |
| 1383 | vm_srb->data_in); |
| 1384 | if (!cmd_request->bounce_sgl) { |
| 1385 | ret = SCSI_MLQUEUE_HOST_BUSY; |
| 1386 | goto queue_error; |
| 1387 | } |
| 1388 | |
| 1389 | cmd_request->bounce_sgl_count = |
| 1390 | ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >> |
| 1391 | PAGE_SHIFT; |
| 1392 | |
| 1393 | if (vm_srb->data_in == WRITE_TYPE) |
| 1394 | copy_to_bounce_buffer(sgl, |
| 1395 | cmd_request->bounce_sgl, |
| 1396 | scsi_sg_count(scmnd)); |
| 1397 | |
| 1398 | sgl = cmd_request->bounce_sgl; |
| 1399 | sg_count = cmd_request->bounce_sgl_count; |
| 1400 | } |
| 1401 | |
| 1402 | cmd_request->data_buffer.offset = sgl[0].offset; |
| 1403 | |
| 1404 | for (i = 0; i < sg_count; i++) |
| 1405 | cmd_request->data_buffer.pfn_array[i] = |
| 1406 | page_to_pfn(sg_page((&sgl[i]))); |
| 1407 | |
| 1408 | } else if (scsi_sglist(scmnd)) { |
| 1409 | cmd_request->data_buffer.offset = |
| 1410 | virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1); |
| 1411 | cmd_request->data_buffer.pfn_array[0] = |
| 1412 | virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT; |
| 1413 | } |
| 1414 | |
| 1415 | /* Invokes the vsc to start an IO */ |
| 1416 | ret = storvsc_do_io(dev, cmd_request); |
| 1417 | |
| 1418 | if (ret == -EAGAIN) { |
| 1419 | /* no more space */ |
| 1420 | |
| 1421 | if (cmd_request->bounce_sgl_count) { |
| 1422 | destroy_bounce_buffer(cmd_request->bounce_sgl, |
| 1423 | cmd_request->bounce_sgl_count); |
| 1424 | |
| 1425 | ret = SCSI_MLQUEUE_DEVICE_BUSY; |
| 1426 | goto queue_error; |
| 1427 | } |
| 1428 | } |
| 1429 | |
| 1430 | return 0; |
| 1431 | |
| 1432 | queue_error: |
| 1433 | mempool_free(cmd_request, memp->request_mempool); |
| 1434 | scmnd->host_scribble = NULL; |
| 1435 | return ret; |
| 1436 | } |
| 1437 | |
| 1438 | static struct scsi_host_template scsi_driver = { |
| 1439 | .module = THIS_MODULE, |
| 1440 | .name = "storvsc_host_t", |
| 1441 | .bios_param = storvsc_get_chs, |
| 1442 | .queuecommand = storvsc_queuecommand, |
| 1443 | .eh_host_reset_handler = storvsc_host_reset_handler, |
| 1444 | .slave_alloc = storvsc_device_alloc, |
| 1445 | .slave_destroy = storvsc_device_destroy, |
| 1446 | .slave_configure = storvsc_device_configure, |
| 1447 | .cmd_per_lun = 1, |
| 1448 | /* 64 max_queue * 1 target */ |
| 1449 | .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS, |
| 1450 | .this_id = -1, |
| 1451 | /* no use setting to 0 since ll_blk_rw reset it to 1 */ |
| 1452 | /* currently 32 */ |
| 1453 | .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT, |
| 1454 | .use_clustering = DISABLE_CLUSTERING, |
| 1455 | /* Make sure we dont get a sg segment crosses a page boundary */ |
| 1456 | .dma_boundary = PAGE_SIZE-1, |
| 1457 | .no_write_same = 1, |
| 1458 | }; |
| 1459 | |
| 1460 | enum { |
| 1461 | SCSI_GUID, |
| 1462 | IDE_GUID, |
| 1463 | }; |
| 1464 | |
| 1465 | static const struct hv_vmbus_device_id id_table[] = { |
| 1466 | /* SCSI guid */ |
| 1467 | { HV_SCSI_GUID, |
| 1468 | .driver_data = SCSI_GUID |
| 1469 | }, |
| 1470 | /* IDE guid */ |
| 1471 | { HV_IDE_GUID, |
| 1472 | .driver_data = IDE_GUID |
| 1473 | }, |
| 1474 | { }, |
| 1475 | }; |
| 1476 | |
| 1477 | MODULE_DEVICE_TABLE(vmbus, id_table); |
| 1478 | |
| 1479 | static int storvsc_probe(struct hv_device *device, |
| 1480 | const struct hv_vmbus_device_id *dev_id) |
| 1481 | { |
| 1482 | int ret; |
| 1483 | struct Scsi_Host *host; |
| 1484 | struct hv_host_device *host_dev; |
| 1485 | bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false); |
| 1486 | int target = 0; |
| 1487 | struct storvsc_device *stor_device; |
| 1488 | |
| 1489 | host = scsi_host_alloc(&scsi_driver, |
| 1490 | sizeof(struct hv_host_device)); |
| 1491 | if (!host) |
| 1492 | return -ENOMEM; |
| 1493 | |
| 1494 | host_dev = shost_priv(host); |
| 1495 | memset(host_dev, 0, sizeof(struct hv_host_device)); |
| 1496 | |
| 1497 | host_dev->port = host->host_no; |
| 1498 | host_dev->dev = device; |
| 1499 | |
| 1500 | |
| 1501 | stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL); |
| 1502 | if (!stor_device) { |
| 1503 | ret = -ENOMEM; |
| 1504 | goto err_out0; |
| 1505 | } |
| 1506 | |
| 1507 | stor_device->destroy = false; |
| 1508 | init_waitqueue_head(&stor_device->waiting_to_drain); |
| 1509 | stor_device->device = device; |
| 1510 | stor_device->host = host; |
| 1511 | hv_set_drvdata(device, stor_device); |
| 1512 | |
| 1513 | stor_device->port_number = host->host_no; |
| 1514 | ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size); |
| 1515 | if (ret) |
| 1516 | goto err_out1; |
| 1517 | |
| 1518 | host_dev->path = stor_device->path_id; |
| 1519 | host_dev->target = stor_device->target_id; |
| 1520 | |
| 1521 | /* max # of devices per target */ |
| 1522 | host->max_lun = STORVSC_MAX_LUNS_PER_TARGET; |
| 1523 | /* max # of targets per channel */ |
| 1524 | host->max_id = STORVSC_MAX_TARGETS; |
| 1525 | /* max # of channels */ |
| 1526 | host->max_channel = STORVSC_MAX_CHANNELS - 1; |
| 1527 | /* max cmd length */ |
| 1528 | host->max_cmd_len = STORVSC_MAX_CMD_LEN; |
| 1529 | |
| 1530 | /* Register the HBA and start the scsi bus scan */ |
| 1531 | ret = scsi_add_host(host, &device->device); |
| 1532 | if (ret != 0) |
| 1533 | goto err_out2; |
| 1534 | |
| 1535 | if (!dev_is_ide) { |
| 1536 | scsi_scan_host(host); |
| 1537 | } else { |
| 1538 | target = (device->dev_instance.b[5] << 8 | |
| 1539 | device->dev_instance.b[4]); |
| 1540 | ret = scsi_add_device(host, 0, target, 0); |
| 1541 | if (ret) { |
| 1542 | scsi_remove_host(host); |
| 1543 | goto err_out2; |
| 1544 | } |
| 1545 | } |
| 1546 | return 0; |
| 1547 | |
| 1548 | err_out2: |
| 1549 | /* |
| 1550 | * Once we have connected with the host, we would need to |
| 1551 | * to invoke storvsc_dev_remove() to rollback this state and |
| 1552 | * this call also frees up the stor_device; hence the jump around |
| 1553 | * err_out1 label. |
| 1554 | */ |
| 1555 | storvsc_dev_remove(device); |
| 1556 | goto err_out0; |
| 1557 | |
| 1558 | err_out1: |
| 1559 | kfree(stor_device); |
| 1560 | |
| 1561 | err_out0: |
| 1562 | scsi_host_put(host); |
| 1563 | return ret; |
| 1564 | } |
| 1565 | |
| 1566 | static int storvsc_remove(struct hv_device *dev) |
| 1567 | { |
| 1568 | struct storvsc_device *stor_device = hv_get_drvdata(dev); |
| 1569 | struct Scsi_Host *host = stor_device->host; |
| 1570 | |
| 1571 | scsi_remove_host(host); |
| 1572 | storvsc_dev_remove(dev); |
| 1573 | scsi_host_put(host); |
| 1574 | |
| 1575 | return 0; |
| 1576 | } |
| 1577 | |
| 1578 | static struct hv_driver storvsc_drv = { |
| 1579 | .name = KBUILD_MODNAME, |
| 1580 | .id_table = id_table, |
| 1581 | .probe = storvsc_probe, |
| 1582 | .remove = storvsc_remove, |
| 1583 | }; |
| 1584 | |
| 1585 | static int __init storvsc_drv_init(void) |
| 1586 | { |
| 1587 | u32 max_outstanding_req_per_channel; |
| 1588 | |
| 1589 | /* |
| 1590 | * Divide the ring buffer data size (which is 1 page less |
| 1591 | * than the ring buffer size since that page is reserved for |
| 1592 | * the ring buffer indices) by the max request size (which is |
| 1593 | * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64) |
| 1594 | */ |
| 1595 | max_outstanding_req_per_channel = |
| 1596 | ((storvsc_ringbuffer_size - PAGE_SIZE) / |
| 1597 | ALIGN(MAX_MULTIPAGE_BUFFER_PACKET + |
| 1598 | sizeof(struct vstor_packet) + sizeof(u64), |
| 1599 | sizeof(u64))); |
| 1600 | |
| 1601 | if (max_outstanding_req_per_channel < |
| 1602 | STORVSC_MAX_IO_REQUESTS) |
| 1603 | return -EINVAL; |
| 1604 | |
| 1605 | return vmbus_driver_register(&storvsc_drv); |
| 1606 | } |
| 1607 | |
| 1608 | static void __exit storvsc_drv_exit(void) |
| 1609 | { |
| 1610 | vmbus_driver_unregister(&storvsc_drv); |
| 1611 | } |
| 1612 | |
| 1613 | MODULE_LICENSE("GPL"); |
| 1614 | MODULE_VERSION(HV_DRV_VERSION); |
| 1615 | MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver"); |
| 1616 | module_init(storvsc_drv_init); |
| 1617 | module_exit(storvsc_drv_exit); |