rdma.c 113.4 KB
Newer Older
M
Michael R. Hines 已提交
1 2 3 4
/*
 * RDMA protocol and interfaces
 *
 * Copyright IBM, Corp. 2010-2013
5
 * Copyright Red Hat, Inc. 2015-2016
M
Michael R. Hines 已提交
6 7 8 9
 *
 * Authors:
 *  Michael R. Hines <mrhines@us.ibm.com>
 *  Jiuxing Liu <jl@us.ibm.com>
10
 *  Daniel P. Berrange <berrange@redhat.com>
M
Michael R. Hines 已提交
11 12 13 14 15
 *
 * This work is licensed under the terms of the GNU GPL, version 2 or
 * later.  See the COPYING file in the top-level directory.
 *
 */
P
Peter Maydell 已提交
16
#include "qemu/osdep.h"
17
#include "qapi/error.h"
M
Michael R. Hines 已提交
18
#include "qemu-common.h"
19
#include "qemu/cutils.h"
M
Michael R. Hines 已提交
20 21 22
#include "migration/migration.h"
#include "migration/qemu-file.h"
#include "exec/cpu-common.h"
23
#include "qemu/error-report.h"
M
Michael R. Hines 已提交
24 25 26
#include "qemu/main-loop.h"
#include "qemu/sockets.h"
#include "qemu/bitmap.h"
27
#include "qemu/coroutine.h"
M
Michael R. Hines 已提交
28 29 30 31
#include <sys/socket.h>
#include <netdb.h>
#include <arpa/inet.h>
#include <rdma/rdma_cma.h>
32
#include "trace.h"
M
Michael R. Hines 已提交
33 34 35 36 37 38

/*
 * Print and error on both the Monitor and the Log file.
 */
#define ERROR(errp, fmt, ...) \
    do { \
39
        fprintf(stderr, "RDMA ERROR: " fmt "\n", ## __VA_ARGS__); \
M
Michael R. Hines 已提交
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82
        if (errp && (*(errp) == NULL)) { \
            error_setg(errp, "RDMA ERROR: " fmt, ## __VA_ARGS__); \
        } \
    } while (0)

#define RDMA_RESOLVE_TIMEOUT_MS 10000

/* Do not merge data if larger than this. */
#define RDMA_MERGE_MAX (2 * 1024 * 1024)
#define RDMA_SIGNALED_SEND_MAX (RDMA_MERGE_MAX / 4096)

#define RDMA_REG_CHUNK_SHIFT 20 /* 1 MB */

/*
 * This is only for non-live state being migrated.
 * Instead of RDMA_WRITE messages, we use RDMA_SEND
 * messages for that state, which requires a different
 * delivery design than main memory.
 */
#define RDMA_SEND_INCREMENT 32768

/*
 * Maximum size infiniband SEND message
 */
#define RDMA_CONTROL_MAX_BUFFER (512 * 1024)
#define RDMA_CONTROL_MAX_COMMANDS_PER_MESSAGE 4096

#define RDMA_CONTROL_VERSION_CURRENT 1
/*
 * Capabilities for negotiation.
 */
#define RDMA_CAPABILITY_PIN_ALL 0x01

/*
 * Add the other flags above to this list of known capabilities
 * as they are introduced.
 */
static uint32_t known_capabilities = RDMA_CAPABILITY_PIN_ALL;

#define CHECK_ERROR_STATE() \
    do { \
        if (rdma->error_state) { \
            if (!rdma->error_reported) { \
83 84
                error_report("RDMA is in an error state waiting migration" \
                                " to abort!"); \
M
Michael R. Hines 已提交
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126
                rdma->error_reported = 1; \
            } \
            return rdma->error_state; \
        } \
    } while (0);

/*
 * A work request ID is 64-bits and we split up these bits
 * into 3 parts:
 *
 * bits 0-15 : type of control message, 2^16
 * bits 16-29: ram block index, 2^14
 * bits 30-63: ram block chunk number, 2^34
 *
 * The last two bit ranges are only used for RDMA writes,
 * in order to track their completion and potentially
 * also track unregistration status of the message.
 */
#define RDMA_WRID_TYPE_SHIFT  0UL
#define RDMA_WRID_BLOCK_SHIFT 16UL
#define RDMA_WRID_CHUNK_SHIFT 30UL

#define RDMA_WRID_TYPE_MASK \
    ((1UL << RDMA_WRID_BLOCK_SHIFT) - 1UL)

#define RDMA_WRID_BLOCK_MASK \
    (~RDMA_WRID_TYPE_MASK & ((1UL << RDMA_WRID_CHUNK_SHIFT) - 1UL))

#define RDMA_WRID_CHUNK_MASK (~RDMA_WRID_BLOCK_MASK & ~RDMA_WRID_TYPE_MASK)

/*
 * RDMA migration protocol:
 * 1. RDMA Writes (data messages, i.e. RAM)
 * 2. IB Send/Recv (control channel messages)
 */
enum {
    RDMA_WRID_NONE = 0,
    RDMA_WRID_RDMA_WRITE = 1,
    RDMA_WRID_SEND_CONTROL = 2000,
    RDMA_WRID_RECV_CONTROL = 4000,
};

127
static const char *wrid_desc[] = {
M
Michael R. Hines 已提交
128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165
    [RDMA_WRID_NONE] = "NONE",
    [RDMA_WRID_RDMA_WRITE] = "WRITE RDMA",
    [RDMA_WRID_SEND_CONTROL] = "CONTROL SEND",
    [RDMA_WRID_RECV_CONTROL] = "CONTROL RECV",
};

/*
 * Work request IDs for IB SEND messages only (not RDMA writes).
 * This is used by the migration protocol to transmit
 * control messages (such as device state and registration commands)
 *
 * We could use more WRs, but we have enough for now.
 */
enum {
    RDMA_WRID_READY = 0,
    RDMA_WRID_DATA,
    RDMA_WRID_CONTROL,
    RDMA_WRID_MAX,
};

/*
 * SEND/RECV IB Control Messages.
 */
enum {
    RDMA_CONTROL_NONE = 0,
    RDMA_CONTROL_ERROR,
    RDMA_CONTROL_READY,               /* ready to receive */
    RDMA_CONTROL_QEMU_FILE,           /* QEMUFile-transmitted bytes */
    RDMA_CONTROL_RAM_BLOCKS_REQUEST,  /* RAMBlock synchronization */
    RDMA_CONTROL_RAM_BLOCKS_RESULT,   /* RAMBlock synchronization */
    RDMA_CONTROL_COMPRESS,            /* page contains repeat values */
    RDMA_CONTROL_REGISTER_REQUEST,    /* dynamic page registration */
    RDMA_CONTROL_REGISTER_RESULT,     /* key to use after registration */
    RDMA_CONTROL_REGISTER_FINISHED,   /* current iteration finished */
    RDMA_CONTROL_UNREGISTER_REQUEST,  /* dynamic UN-registration */
    RDMA_CONTROL_UNREGISTER_FINISHED, /* unpinning finished */
};

166
static const char *control_desc[] = {
M
Michael R. Hines 已提交
167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219
    [RDMA_CONTROL_NONE] = "NONE",
    [RDMA_CONTROL_ERROR] = "ERROR",
    [RDMA_CONTROL_READY] = "READY",
    [RDMA_CONTROL_QEMU_FILE] = "QEMU FILE",
    [RDMA_CONTROL_RAM_BLOCKS_REQUEST] = "RAM BLOCKS REQUEST",
    [RDMA_CONTROL_RAM_BLOCKS_RESULT] = "RAM BLOCKS RESULT",
    [RDMA_CONTROL_COMPRESS] = "COMPRESS",
    [RDMA_CONTROL_REGISTER_REQUEST] = "REGISTER REQUEST",
    [RDMA_CONTROL_REGISTER_RESULT] = "REGISTER RESULT",
    [RDMA_CONTROL_REGISTER_FINISHED] = "REGISTER FINISHED",
    [RDMA_CONTROL_UNREGISTER_REQUEST] = "UNREGISTER REQUEST",
    [RDMA_CONTROL_UNREGISTER_FINISHED] = "UNREGISTER FINISHED",
};

/*
 * Memory and MR structures used to represent an IB Send/Recv work request.
 * This is *not* used for RDMA writes, only IB Send/Recv.
 */
typedef struct {
    uint8_t  control[RDMA_CONTROL_MAX_BUFFER]; /* actual buffer to register */
    struct   ibv_mr *control_mr;               /* registration metadata */
    size_t   control_len;                      /* length of the message */
    uint8_t *control_curr;                     /* start of unconsumed bytes */
} RDMAWorkRequestData;

/*
 * Negotiate RDMA capabilities during connection-setup time.
 */
typedef struct {
    uint32_t version;
    uint32_t flags;
} RDMACapabilities;

static void caps_to_network(RDMACapabilities *cap)
{
    cap->version = htonl(cap->version);
    cap->flags = htonl(cap->flags);
}

static void network_to_caps(RDMACapabilities *cap)
{
    cap->version = ntohl(cap->version);
    cap->flags = ntohl(cap->flags);
}

/*
 * Representation of a RAMBlock from an RDMA perspective.
 * This is not transmitted, only local.
 * This and subsequent structures cannot be linked lists
 * because we're using a single IB message to transmit
 * the information. It's small anyway, so a list is overkill.
 */
typedef struct RDMALocalBlock {
220 221 222 223 224 225 226 227 228 229
    char          *block_name;
    uint8_t       *local_host_addr; /* local virtual address */
    uint64_t       remote_host_addr; /* remote virtual address */
    uint64_t       offset;
    uint64_t       length;
    struct         ibv_mr **pmr;    /* MRs for chunk-level registration */
    struct         ibv_mr *mr;      /* MR for non-chunk-level registration */
    uint32_t      *remote_keys;     /* rkeys for chunk-level registration */
    uint32_t       remote_rkey;     /* rkeys for non-chunk-level registration */
    int            index;           /* which block are we */
230
    unsigned int   src_index;       /* (Only used on dest) */
231 232
    bool           is_ram_block;
    int            nb_chunks;
M
Michael R. Hines 已提交
233 234 235 236 237 238 239 240 241 242 243
    unsigned long *transit_bitmap;
    unsigned long *unregister_bitmap;
} RDMALocalBlock;

/*
 * Also represents a RAMblock, but only on the dest.
 * This gets transmitted by the dest during connection-time
 * to the source VM and then is used to populate the
 * corresponding RDMALocalBlock with
 * the information needed to perform the actual RDMA.
 */
244
typedef struct QEMU_PACKED RDMADestBlock {
M
Michael R. Hines 已提交
245 246 247 248 249
    uint64_t remote_host_addr;
    uint64_t offset;
    uint64_t length;
    uint32_t remote_rkey;
    uint32_t padding;
250
} RDMADestBlock;
M
Michael R. Hines 已提交
251 252 253 254 255 256 257 258 259 260 261 262 263 264 265

static uint64_t htonll(uint64_t v)
{
    union { uint32_t lv[2]; uint64_t llv; } u;
    u.lv[0] = htonl(v >> 32);
    u.lv[1] = htonl(v & 0xFFFFFFFFULL);
    return u.llv;
}

static uint64_t ntohll(uint64_t v) {
    union { uint32_t lv[2]; uint64_t llv; } u;
    u.llv = v;
    return ((uint64_t)ntohl(u.lv[0]) << 32) | (uint64_t) ntohl(u.lv[1]);
}

266
static void dest_block_to_network(RDMADestBlock *db)
M
Michael R. Hines 已提交
267
{
268 269 270 271
    db->remote_host_addr = htonll(db->remote_host_addr);
    db->offset = htonll(db->offset);
    db->length = htonll(db->length);
    db->remote_rkey = htonl(db->remote_rkey);
M
Michael R. Hines 已提交
272 273
}

274
static void network_to_dest_block(RDMADestBlock *db)
M
Michael R. Hines 已提交
275
{
276 277 278 279
    db->remote_host_addr = ntohll(db->remote_host_addr);
    db->offset = ntohll(db->offset);
    db->length = ntohll(db->length);
    db->remote_rkey = ntohl(db->remote_rkey);
M
Michael R. Hines 已提交
280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302
}

/*
 * Virtual address of the above structures used for transmitting
 * the RAMBlock descriptions at connection-time.
 * This structure is *not* transmitted.
 */
typedef struct RDMALocalBlocks {
    int nb_blocks;
    bool     init;             /* main memory init complete */
    RDMALocalBlock *block;
} RDMALocalBlocks;

/*
 * Main data structure for RDMA state.
 * While there is only one copy of this structure being allocated right now,
 * this is the place where one would start if you wanted to consider
 * having more than one RDMA connection open at the same time.
 */
typedef struct RDMAContext {
    char *host;
    int port;

303
    RDMAWorkRequestData wr_data[RDMA_WRID_MAX];
M
Michael R. Hines 已提交
304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336

    /*
     * This is used by *_exchange_send() to figure out whether or not
     * the initial "READY" message has already been received or not.
     * This is because other functions may potentially poll() and detect
     * the READY message before send() does, in which case we need to
     * know if it completed.
     */
    int control_ready_expected;

    /* number of outstanding writes */
    int nb_sent;

    /* store info about current buffer so that we can
       merge it with future sends */
    uint64_t current_addr;
    uint64_t current_length;
    /* index of ram block the current buffer belongs to */
    int current_index;
    /* index of the chunk in the current ram block */
    int current_chunk;

    bool pin_all;

    /*
     * infiniband-specific variables for opening the device
     * and maintaining connection state and so forth.
     *
     * cm_id also has ibv_context, rdma_event_channel, and ibv_qp in
     * cm_id->verbs, cm_id->channel, and cm_id->qp.
     */
    struct rdma_cm_id *cm_id;               /* connection manager ID */
    struct rdma_cm_id *listen_id;
337
    bool connected;
M
Michael R. Hines 已提交
338 339 340 341 342 343 344 345 346 347 348 349 350 351 352

    struct ibv_context          *verbs;
    struct rdma_event_channel   *channel;
    struct ibv_qp *qp;                      /* queue pair */
    struct ibv_comp_channel *comp_channel;  /* completion channel */
    struct ibv_pd *pd;                      /* protection domain */
    struct ibv_cq *cq;                      /* completion queue */

    /*
     * If a previous write failed (perhaps because of a failed
     * memory registration, then do not attempt any future work
     * and remember the error state.
     */
    int error_state;
    int error_reported;
353
    int received_error;
M
Michael R. Hines 已提交
354 355 356 357 358

    /*
     * Description of ram blocks used throughout the code.
     */
    RDMALocalBlocks local_ram_blocks;
359
    RDMADestBlock  *dest_blocks;
M
Michael R. Hines 已提交
360

361 362 363
    /* Index of the next RAMBlock received during block registration */
    unsigned int    next_src_index;

M
Michael R. Hines 已提交
364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379
    /*
     * Migration on *destination* started.
     * Then use coroutine yield function.
     * Source runs in a thread, so we don't care.
     */
    int migration_started_on_destination;

    int total_registrations;
    int total_writes;

    int unregister_current, unregister_next;
    uint64_t unregistrations[RDMA_SIGNALED_SEND_MAX];

    GHashTable *blockmap;
} RDMAContext;

380 381 382 383 384 385 386 387 388
#define TYPE_QIO_CHANNEL_RDMA "qio-channel-rdma"
#define QIO_CHANNEL_RDMA(obj)                                     \
    OBJECT_CHECK(QIOChannelRDMA, (obj), TYPE_QIO_CHANNEL_RDMA)

typedef struct QIOChannelRDMA QIOChannelRDMA;


struct QIOChannelRDMA {
    QIOChannel parent;
M
Michael R. Hines 已提交
389
    RDMAContext *rdma;
390
    QEMUFile *file;
M
Michael R. Hines 已提交
391
    size_t len;
392 393
    bool blocking; /* XXX we don't actually honour this yet */
};
M
Michael R. Hines 已提交
394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427

/*
 * Main structure for IB Send/Recv control messages.
 * This gets prepended at the beginning of every Send/Recv.
 */
typedef struct QEMU_PACKED {
    uint32_t len;     /* Total length of data portion */
    uint32_t type;    /* which control command to perform */
    uint32_t repeat;  /* number of commands in data portion of same type */
    uint32_t padding;
} RDMAControlHeader;

static void control_to_network(RDMAControlHeader *control)
{
    control->type = htonl(control->type);
    control->len = htonl(control->len);
    control->repeat = htonl(control->repeat);
}

static void network_to_control(RDMAControlHeader *control)
{
    control->type = ntohl(control->type);
    control->len = ntohl(control->len);
    control->repeat = ntohl(control->repeat);
}

/*
 * Register a single Chunk.
 * Information sent by the source VM to inform the dest
 * to register an single chunk of memory before we can perform
 * the actual RDMA operation.
 */
typedef struct QEMU_PACKED {
    union QEMU_PACKED {
428
        uint64_t current_addr;  /* offset into the ram_addr_t space */
M
Michael R. Hines 已提交
429 430 431 432 433 434 435
        uint64_t chunk;         /* chunk to lookup if unregistering */
    } key;
    uint32_t current_index; /* which ramblock the chunk belongs to */
    uint32_t padding;
    uint64_t chunks;            /* how many sequential chunks to register */
} RDMARegister;

436
static void register_to_network(RDMAContext *rdma, RDMARegister *reg)
M
Michael R. Hines 已提交
437
{
438 439 440 441 442 443 444 445 446 447 448
    RDMALocalBlock *local_block;
    local_block  = &rdma->local_ram_blocks.block[reg->current_index];

    if (local_block->is_ram_block) {
        /*
         * current_addr as passed in is an address in the local ram_addr_t
         * space, we need to translate this for the destination
         */
        reg->key.current_addr -= local_block->offset;
        reg->key.current_addr += rdma->dest_blocks[reg->current_index].offset;
    }
M
Michael R. Hines 已提交
449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
    reg->key.current_addr = htonll(reg->key.current_addr);
    reg->current_index = htonl(reg->current_index);
    reg->chunks = htonll(reg->chunks);
}

static void network_to_register(RDMARegister *reg)
{
    reg->key.current_addr = ntohll(reg->key.current_addr);
    reg->current_index = ntohl(reg->current_index);
    reg->chunks = ntohll(reg->chunks);
}

typedef struct QEMU_PACKED {
    uint32_t value;     /* if zero, we will madvise() */
    uint32_t block_idx; /* which ram block index */
464
    uint64_t offset;    /* Address in remote ram_addr_t space */
M
Michael R. Hines 已提交
465 466 467
    uint64_t length;    /* length of the chunk */
} RDMACompress;

468
static void compress_to_network(RDMAContext *rdma, RDMACompress *comp)
M
Michael R. Hines 已提交
469 470
{
    comp->value = htonl(comp->value);
471 472 473 474 475 476
    /*
     * comp->offset as passed in is an address in the local ram_addr_t
     * space, we need to translate this for the destination
     */
    comp->offset -= rdma->local_ram_blocks.block[comp->block_idx].offset;
    comp->offset += rdma->dest_blocks[comp->block_idx].offset;
M
Michael R. Hines 已提交
477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518
    comp->block_idx = htonl(comp->block_idx);
    comp->offset = htonll(comp->offset);
    comp->length = htonll(comp->length);
}

static void network_to_compress(RDMACompress *comp)
{
    comp->value = ntohl(comp->value);
    comp->block_idx = ntohl(comp->block_idx);
    comp->offset = ntohll(comp->offset);
    comp->length = ntohll(comp->length);
}

/*
 * The result of the dest's memory registration produces an "rkey"
 * which the source VM must reference in order to perform
 * the RDMA operation.
 */
typedef struct QEMU_PACKED {
    uint32_t rkey;
    uint32_t padding;
    uint64_t host_addr;
} RDMARegisterResult;

static void result_to_network(RDMARegisterResult *result)
{
    result->rkey = htonl(result->rkey);
    result->host_addr = htonll(result->host_addr);
};

static void network_to_result(RDMARegisterResult *result)
{
    result->rkey = ntohl(result->rkey);
    result->host_addr = ntohll(result->host_addr);
};

const char *print_wrid(int wrid);
static int qemu_rdma_exchange_send(RDMAContext *rdma, RDMAControlHeader *head,
                                   uint8_t *data, RDMAControlHeader *resp,
                                   int *resp_idx,
                                   int (*callback)(RDMAContext *rdma));

519 520
static inline uint64_t ram_chunk_index(const uint8_t *start,
                                       const uint8_t *host)
M
Michael R. Hines 已提交
521 522 523 524
{
    return ((uintptr_t) host - (uintptr_t) start) >> RDMA_REG_CHUNK_SHIFT;
}

525
static inline uint8_t *ram_chunk_start(const RDMALocalBlock *rdma_ram_block,
M
Michael R. Hines 已提交
526 527
                                       uint64_t i)
{
528 529
    return (uint8_t *)(uintptr_t)(rdma_ram_block->local_host_addr +
                                  (i << RDMA_REG_CHUNK_SHIFT));
M
Michael R. Hines 已提交
530 531
}

532 533
static inline uint8_t *ram_chunk_end(const RDMALocalBlock *rdma_ram_block,
                                     uint64_t i)
M
Michael R. Hines 已提交
534 535 536 537 538 539 540 541 542 543 544
{
    uint8_t *result = ram_chunk_start(rdma_ram_block, i) +
                                         (1UL << RDMA_REG_CHUNK_SHIFT);

    if (result > (rdma_ram_block->local_host_addr + rdma_ram_block->length)) {
        result = rdma_ram_block->local_host_addr + rdma_ram_block->length;
    }

    return result;
}

545 546
static int rdma_add_block(RDMAContext *rdma, const char *block_name,
                         void *host_addr,
M
Michael R. Hines 已提交
547 548 549
                         ram_addr_t block_offset, uint64_t length)
{
    RDMALocalBlocks *local = &rdma->local_ram_blocks;
D
Dr. David Alan Gilbert 已提交
550
    RDMALocalBlock *block;
M
Michael R. Hines 已提交
551 552
    RDMALocalBlock *old = local->block;

553
    local->block = g_new0(RDMALocalBlock, local->nb_blocks + 1);
M
Michael R. Hines 已提交
554 555 556 557

    if (local->nb_blocks) {
        int x;

D
Dr. David Alan Gilbert 已提交
558 559 560 561 562 563 564 565
        if (rdma->blockmap) {
            for (x = 0; x < local->nb_blocks; x++) {
                g_hash_table_remove(rdma->blockmap,
                                    (void *)(uintptr_t)old[x].offset);
                g_hash_table_insert(rdma->blockmap,
                                    (void *)(uintptr_t)old[x].offset,
                                    &local->block[x]);
            }
M
Michael R. Hines 已提交
566 567 568 569 570 571 572
        }
        memcpy(local->block, old, sizeof(RDMALocalBlock) * local->nb_blocks);
        g_free(old);
    }

    block = &local->block[local->nb_blocks];

573
    block->block_name = g_strdup(block_name);
M
Michael R. Hines 已提交
574 575 576 577
    block->local_host_addr = host_addr;
    block->offset = block_offset;
    block->length = length;
    block->index = local->nb_blocks;
578
    block->src_index = ~0U; /* Filled in by the receipt of the block list */
M
Michael R. Hines 已提交
579 580 581 582 583
    block->nb_chunks = ram_chunk_index(host_addr, host_addr + length) + 1UL;
    block->transit_bitmap = bitmap_new(block->nb_chunks);
    bitmap_clear(block->transit_bitmap, 0, block->nb_chunks);
    block->unregister_bitmap = bitmap_new(block->nb_chunks);
    bitmap_clear(block->unregister_bitmap, 0, block->nb_chunks);
584
    block->remote_keys = g_new0(uint32_t, block->nb_chunks);
M
Michael R. Hines 已提交
585 586 587

    block->is_ram_block = local->init ? false : true;

D
Dr. David Alan Gilbert 已提交
588
    if (rdma->blockmap) {
589
        g_hash_table_insert(rdma->blockmap, (void *)(uintptr_t)block_offset, block);
D
Dr. David Alan Gilbert 已提交
590
    }
M
Michael R. Hines 已提交
591

592 593
    trace_rdma_add_block(block_name, local->nb_blocks,
                         (uintptr_t) block->local_host_addr,
594
                         block->offset, block->length,
595
                         (uintptr_t) (block->local_host_addr + block->length),
596 597 598
                         BITS_TO_LONGS(block->nb_chunks) *
                             sizeof(unsigned long) * 8,
                         block->nb_chunks);
M
Michael R. Hines 已提交
599 600 601 602 603 604 605 606 607 608 609

    local->nb_blocks++;

    return 0;
}

/*
 * Memory regions need to be registered with the device and queue pairs setup
 * in advanced before the migration starts. This tells us where the RAM blocks
 * are so that we can register them individually.
 */
610
static int qemu_rdma_init_one_block(const char *block_name, void *host_addr,
M
Michael R. Hines 已提交
611 612
    ram_addr_t block_offset, ram_addr_t length, void *opaque)
{
613
    return rdma_add_block(opaque, block_name, host_addr, block_offset, length);
M
Michael R. Hines 已提交
614 615 616 617 618 619 620 621 622 623 624 625 626 627
}

/*
 * Identify the RAMBlocks and their quantity. They will be references to
 * identify chunk boundaries inside each RAMBlock and also be referenced
 * during dynamic page registration.
 */
static int qemu_rdma_init_ram_blocks(RDMAContext *rdma)
{
    RDMALocalBlocks *local = &rdma->local_ram_blocks;

    assert(rdma->blockmap == NULL);
    memset(local, 0, sizeof *local);
    qemu_ram_foreach_block(qemu_rdma_init_one_block, rdma);
628
    trace_qemu_rdma_init_ram_blocks(local->nb_blocks);
629 630
    rdma->dest_blocks = g_new0(RDMADestBlock,
                               rdma->local_ram_blocks.nb_blocks);
M
Michael R. Hines 已提交
631 632 633 634
    local->init = true;
    return 0;
}

635 636 637 638 639
/*
 * Note: If used outside of cleanup, the caller must ensure that the destination
 * block structures are also updated
 */
static int rdma_delete_block(RDMAContext *rdma, RDMALocalBlock *block)
M
Michael R. Hines 已提交
640 641 642 643 644
{
    RDMALocalBlocks *local = &rdma->local_ram_blocks;
    RDMALocalBlock *old = local->block;
    int x;

645 646 647
    if (rdma->blockmap) {
        g_hash_table_remove(rdma->blockmap, (void *)(uintptr_t)block->offset);
    }
M
Michael R. Hines 已提交
648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676
    if (block->pmr) {
        int j;

        for (j = 0; j < block->nb_chunks; j++) {
            if (!block->pmr[j]) {
                continue;
            }
            ibv_dereg_mr(block->pmr[j]);
            rdma->total_registrations--;
        }
        g_free(block->pmr);
        block->pmr = NULL;
    }

    if (block->mr) {
        ibv_dereg_mr(block->mr);
        rdma->total_registrations--;
        block->mr = NULL;
    }

    g_free(block->transit_bitmap);
    block->transit_bitmap = NULL;

    g_free(block->unregister_bitmap);
    block->unregister_bitmap = NULL;

    g_free(block->remote_keys);
    block->remote_keys = NULL;

677 678 679
    g_free(block->block_name);
    block->block_name = NULL;

680 681 682 683 684
    if (rdma->blockmap) {
        for (x = 0; x < local->nb_blocks; x++) {
            g_hash_table_remove(rdma->blockmap,
                                (void *)(uintptr_t)old[x].offset);
        }
M
Michael R. Hines 已提交
685 686 687 688
    }

    if (local->nb_blocks > 1) {

689
        local->block = g_new0(RDMALocalBlock, local->nb_blocks - 1);
M
Michael R. Hines 已提交
690 691 692 693 694 695 696 697 698 699 700 701 702 703 704

        if (block->index) {
            memcpy(local->block, old, sizeof(RDMALocalBlock) * block->index);
        }

        if (block->index < (local->nb_blocks - 1)) {
            memcpy(local->block + block->index, old + (block->index + 1),
                sizeof(RDMALocalBlock) *
                    (local->nb_blocks - (block->index + 1)));
        }
    } else {
        assert(block == local->block);
        local->block = NULL;
    }

705
    trace_rdma_delete_block(block, (uintptr_t)block->local_host_addr,
706
                           block->offset, block->length,
707
                            (uintptr_t)(block->local_host_addr + block->length),
708 709
                           BITS_TO_LONGS(block->nb_chunks) *
                               sizeof(unsigned long) * 8, block->nb_chunks);
M
Michael R. Hines 已提交
710 711 712 713 714

    g_free(old);

    local->nb_blocks--;

715
    if (local->nb_blocks && rdma->blockmap) {
M
Michael R. Hines 已提交
716
        for (x = 0; x < local->nb_blocks; x++) {
717 718 719
            g_hash_table_insert(rdma->blockmap,
                                (void *)(uintptr_t)local->block[x].offset,
                                &local->block[x]);
M
Michael R. Hines 已提交
720 721 722 723 724 725 726 727 728 729 730 731
        }
    }

    return 0;
}

/*
 * Put in the log file which RDMA device was opened and the details
 * associated with that device.
 */
static void qemu_rdma_dump_id(const char *who, struct ibv_context *verbs)
{
732 733 734
    struct ibv_port_attr port;

    if (ibv_query_port(verbs, 1, &port)) {
735
        error_report("Failed to query port information");
736 737 738
        return;
    }

M
Michael R. Hines 已提交
739 740
    printf("%s RDMA Device opened: kernel name %s "
           "uverbs device name %s, "
741 742 743
           "infiniband_verbs class device path %s, "
           "infiniband class device path %s, "
           "transport: (%d) %s\n",
M
Michael R. Hines 已提交
744 745 746 747
                who,
                verbs->device->name,
                verbs->device->dev_name,
                verbs->device->dev_path,
748 749 750
                verbs->device->ibdev_path,
                port.link_layer,
                (port.link_layer == IBV_LINK_LAYER_INFINIBAND) ? "Infiniband" :
751
                 ((port.link_layer == IBV_LINK_LAYER_ETHERNET)
752
                    ? "Ethernet" : "Unknown"));
M
Michael R. Hines 已提交
753 754 755 756 757 758 759 760 761 762 763 764 765
}

/*
 * Put in the log file the RDMA gid addressing information,
 * useful for folks who have trouble understanding the
 * RDMA device hierarchy in the kernel.
 */
static void qemu_rdma_dump_gid(const char *who, struct rdma_cm_id *id)
{
    char sgid[33];
    char dgid[33];
    inet_ntop(AF_INET6, &id->route.addr.addr.ibaddr.sgid, sgid, sizeof sgid);
    inet_ntop(AF_INET6, &id->route.addr.addr.ibaddr.dgid, dgid, sizeof dgid);
766
    trace_qemu_rdma_dump_gid(who, sgid, dgid);
M
Michael R. Hines 已提交
767 768
}

769 770 771 772 773 774 775 776 777 778 779 780
/*
 * As of now, IPv6 over RoCE / iWARP is not supported by linux.
 * We will try the next addrinfo struct, and fail if there are
 * no other valid addresses to bind against.
 *
 * If user is listening on '[::]', then we will not have a opened a device
 * yet and have no way of verifying if the device is RoCE or not.
 *
 * In this case, the source VM will throw an error for ALL types of
 * connections (both IPv4 and IPv6) if the destination machine does not have
 * a regular infiniband network available for use.
 *
781
 * The only way to guarantee that an error is thrown for broken kernels is
782 783 784 785
 * for the management software to choose a *specific* interface at bind time
 * and validate what time of hardware it is.
 *
 * Unfortunately, this puts the user in a fix:
786
 *
787 788
 *  If the source VM connects with an IPv4 address without knowing that the
 *  destination has bound to '[::]' the migration will unconditionally fail
789
 *  unless the management software is explicitly listening on the IPv4
790 791 792 793
 *  address while using a RoCE-based device.
 *
 *  If the source VM connects with an IPv6 address, then we're OK because we can
 *  throw an error on the source (and similarly on the destination).
794
 *
795 796 797 798 799
 *  But in mixed environments, this will be broken for a while until it is fixed
 *  inside linux.
 *
 * We do provide a *tiny* bit of help in this function: We can list all of the
 * devices in the system and check to see if all the devices are RoCE or
800
 * Infiniband.
801 802
 *
 * If we detect that we have a *pure* RoCE environment, then we can safely
803
 * thrown an error even if the management software has specified '[::]' as the
804 805 806 807 808 809 810 811
 * bind address.
 *
 * However, if there is are multiple hetergeneous devices, then we cannot make
 * this assumption and the user just has to be sure they know what they are
 * doing.
 *
 * Patches are being reviewed on linux-rdma.
 */
812
static int qemu_rdma_broken_ipv6_kernel(struct ibv_context *verbs, Error **errp)
813 814 815 816 817 818
{
    struct ibv_port_attr port_attr;

    /* This bug only exists in linux, to our knowledge. */
#ifdef CONFIG_LINUX

819
    /*
820
     * Verbs are only NULL if management has bound to '[::]'.
821
     *
822 823
     * Let's iterate through all the devices and see if there any pure IB
     * devices (non-ethernet).
824
     *
825
     * If not, then we can safely proceed with the migration.
826
     * Otherwise, there are no guarantees until the bug is fixed in linux.
827 828
     */
    if (!verbs) {
829
        int num_devices, x;
830 831 832 833 834 835
        struct ibv_device ** dev_list = ibv_get_device_list(&num_devices);
        bool roce_found = false;
        bool ib_found = false;

        for (x = 0; x < num_devices; x++) {
            verbs = ibv_open_device(dev_list[x]);
836 837 838 839 840 841 842
            if (!verbs) {
                if (errno == EPERM) {
                    continue;
                } else {
                    return -EINVAL;
                }
            }
843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880

            if (ibv_query_port(verbs, 1, &port_attr)) {
                ibv_close_device(verbs);
                ERROR(errp, "Could not query initial IB port");
                return -EINVAL;
            }

            if (port_attr.link_layer == IBV_LINK_LAYER_INFINIBAND) {
                ib_found = true;
            } else if (port_attr.link_layer == IBV_LINK_LAYER_ETHERNET) {
                roce_found = true;
            }

            ibv_close_device(verbs);

        }

        if (roce_found) {
            if (ib_found) {
                fprintf(stderr, "WARN: migrations may fail:"
                                " IPv6 over RoCE / iWARP in linux"
                                " is broken. But since you appear to have a"
                                " mixed RoCE / IB environment, be sure to only"
                                " migrate over the IB fabric until the kernel "
                                " fixes the bug.\n");
            } else {
                ERROR(errp, "You only have RoCE / iWARP devices in your systems"
                            " and your management software has specified '[::]'"
                            ", but IPv6 over RoCE / iWARP is not supported in Linux.");
                return -ENONET;
            }
        }

        return 0;
    }

    /*
     * If we have a verbs context, that means that some other than '[::]' was
881 882
     * used by the management software for binding. In which case we can
     * actually warn the user about a potentially broken kernel.
883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901
     */

    /* IB ports start with 1, not 0 */
    if (ibv_query_port(verbs, 1, &port_attr)) {
        ERROR(errp, "Could not query initial IB port");
        return -EINVAL;
    }

    if (port_attr.link_layer == IBV_LINK_LAYER_ETHERNET) {
        ERROR(errp, "Linux kernel's RoCE / iWARP does not support IPv6 "
                    "(but patches on linux-rdma in progress)");
        return -ENONET;
    }

#endif

    return 0;
}

M
Michael R. Hines 已提交
902 903 904 905 906 907 908 909
/*
 * Figure out which RDMA device corresponds to the requested IP hostname
 * Also create the initial connection manager identifiers for opening
 * the connection.
 */
static int qemu_rdma_resolve_host(RDMAContext *rdma, Error **errp)
{
    int ret;
910
    struct rdma_addrinfo *res;
M
Michael R. Hines 已提交
911 912 913
    char port_str[16];
    struct rdma_cm_event *cm_event;
    char ip[40] = "unknown";
914
    struct rdma_addrinfo *e;
M
Michael R. Hines 已提交
915 916

    if (rdma->host == NULL || !strcmp(rdma->host, "")) {
917
        ERROR(errp, "RDMA hostname has not been set");
918
        return -EINVAL;
M
Michael R. Hines 已提交
919 920 921 922 923
    }

    /* create CM channel */
    rdma->channel = rdma_create_event_channel();
    if (!rdma->channel) {
924
        ERROR(errp, "could not create CM channel");
925
        return -EINVAL;
M
Michael R. Hines 已提交
926 927 928 929 930
    }

    /* create CM id */
    ret = rdma_create_id(rdma->channel, &rdma->cm_id, NULL, RDMA_PS_TCP);
    if (ret) {
931
        ERROR(errp, "could not create channel id");
M
Michael R. Hines 已提交
932 933 934 935 936 937
        goto err_resolve_create_id;
    }

    snprintf(port_str, 16, "%d", rdma->port);
    port_str[15] = '\0';

938
    ret = rdma_getaddrinfo(rdma->host, port_str, NULL, &res);
M
Michael R. Hines 已提交
939
    if (ret < 0) {
940
        ERROR(errp, "could not rdma_getaddrinfo address %s", rdma->host);
M
Michael R. Hines 已提交
941 942 943
        goto err_resolve_get_addr;
    }

944 945
    for (e = res; e != NULL; e = e->ai_next) {
        inet_ntop(e->ai_family,
946
            &((struct sockaddr_in *) e->ai_dst_addr)->sin_addr, ip, sizeof ip);
947
        trace_qemu_rdma_resolve_host_trying(rdma->host, ip);
M
Michael R. Hines 已提交
948

949
        ret = rdma_resolve_addr(rdma->cm_id, NULL, e->ai_dst_addr,
950 951
                RDMA_RESOLVE_TIMEOUT_MS);
        if (!ret) {
M
Michael R. Hines 已提交
952
            if (e->ai_family == AF_INET6) {
953
                ret = qemu_rdma_broken_ipv6_kernel(rdma->cm_id->verbs, errp);
M
Michael R. Hines 已提交
954 955 956
                if (ret) {
                    continue;
                }
957
            }
958 959
            goto route;
        }
M
Michael R. Hines 已提交
960 961
    }

962 963 964 965
    ERROR(errp, "could not resolve address %s", rdma->host);
    goto err_resolve_get_addr;

route:
M
Michael R. Hines 已提交
966 967 968 969
    qemu_rdma_dump_gid("source_resolve_addr", rdma->cm_id);

    ret = rdma_get_cm_event(rdma->channel, &cm_event);
    if (ret) {
970
        ERROR(errp, "could not perform event_addr_resolved");
M
Michael R. Hines 已提交
971 972 973 974
        goto err_resolve_get_addr;
    }

    if (cm_event->event != RDMA_CM_EVENT_ADDR_RESOLVED) {
975
        ERROR(errp, "result not equal to event_addr_resolved %s",
M
Michael R. Hines 已提交
976 977
                rdma_event_str(cm_event->event));
        perror("rdma_resolve_addr");
G
Gonglei 已提交
978
        rdma_ack_cm_event(cm_event);
979
        ret = -EINVAL;
M
Michael R. Hines 已提交
980 981 982 983 984 985 986
        goto err_resolve_get_addr;
    }
    rdma_ack_cm_event(cm_event);

    /* resolve route */
    ret = rdma_resolve_route(rdma->cm_id, RDMA_RESOLVE_TIMEOUT_MS);
    if (ret) {
987
        ERROR(errp, "could not resolve rdma route");
M
Michael R. Hines 已提交
988 989 990 991 992
        goto err_resolve_get_addr;
    }

    ret = rdma_get_cm_event(rdma->channel, &cm_event);
    if (ret) {
993
        ERROR(errp, "could not perform event_route_resolved");
M
Michael R. Hines 已提交
994 995 996
        goto err_resolve_get_addr;
    }
    if (cm_event->event != RDMA_CM_EVENT_ROUTE_RESOLVED) {
997
        ERROR(errp, "result not equal to event_route_resolved: %s",
M
Michael R. Hines 已提交
998 999
                        rdma_event_str(cm_event->event));
        rdma_ack_cm_event(cm_event);
1000
        ret = -EINVAL;
M
Michael R. Hines 已提交
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014
        goto err_resolve_get_addr;
    }
    rdma_ack_cm_event(cm_event);
    rdma->verbs = rdma->cm_id->verbs;
    qemu_rdma_dump_id("source_resolve_host", rdma->cm_id->verbs);
    qemu_rdma_dump_gid("source_resolve_host", rdma->cm_id);
    return 0;

err_resolve_get_addr:
    rdma_destroy_id(rdma->cm_id);
    rdma->cm_id = NULL;
err_resolve_create_id:
    rdma_destroy_event_channel(rdma->channel);
    rdma->channel = NULL;
1015
    return ret;
M
Michael R. Hines 已提交
1016 1017 1018 1019 1020 1021 1022 1023 1024 1025
}

/*
 * Create protection domain and completion queues
 */
static int qemu_rdma_alloc_pd_cq(RDMAContext *rdma)
{
    /* allocate pd */
    rdma->pd = ibv_alloc_pd(rdma->verbs);
    if (!rdma->pd) {
1026
        error_report("failed to allocate protection domain");
M
Michael R. Hines 已提交
1027 1028 1029 1030 1031 1032
        return -1;
    }

    /* create completion channel */
    rdma->comp_channel = ibv_create_comp_channel(rdma->verbs);
    if (!rdma->comp_channel) {
1033
        error_report("failed to allocate completion channel");
M
Michael R. Hines 已提交
1034 1035 1036 1037 1038 1039 1040 1041 1042 1043
        goto err_alloc_pd_cq;
    }

    /*
     * Completion queue can be filled by both read and write work requests,
     * so must reflect the sum of both possible queue sizes.
     */
    rdma->cq = ibv_create_cq(rdma->verbs, (RDMA_SIGNALED_SEND_MAX * 3),
            NULL, rdma->comp_channel, 0);
    if (!rdma->cq) {
1044
        error_report("failed to allocate completion queue");
M
Michael R. Hines 已提交
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130
        goto err_alloc_pd_cq;
    }

    return 0;

err_alloc_pd_cq:
    if (rdma->pd) {
        ibv_dealloc_pd(rdma->pd);
    }
    if (rdma->comp_channel) {
        ibv_destroy_comp_channel(rdma->comp_channel);
    }
    rdma->pd = NULL;
    rdma->comp_channel = NULL;
    return -1;

}

/*
 * Create queue pairs.
 */
static int qemu_rdma_alloc_qp(RDMAContext *rdma)
{
    struct ibv_qp_init_attr attr = { 0 };
    int ret;

    attr.cap.max_send_wr = RDMA_SIGNALED_SEND_MAX;
    attr.cap.max_recv_wr = 3;
    attr.cap.max_send_sge = 1;
    attr.cap.max_recv_sge = 1;
    attr.send_cq = rdma->cq;
    attr.recv_cq = rdma->cq;
    attr.qp_type = IBV_QPT_RC;

    ret = rdma_create_qp(rdma->cm_id, rdma->pd, &attr);
    if (ret) {
        return -1;
    }

    rdma->qp = rdma->cm_id->qp;
    return 0;
}

static int qemu_rdma_reg_whole_ram_blocks(RDMAContext *rdma)
{
    int i;
    RDMALocalBlocks *local = &rdma->local_ram_blocks;

    for (i = 0; i < local->nb_blocks; i++) {
        local->block[i].mr =
            ibv_reg_mr(rdma->pd,
                    local->block[i].local_host_addr,
                    local->block[i].length,
                    IBV_ACCESS_LOCAL_WRITE |
                    IBV_ACCESS_REMOTE_WRITE
                    );
        if (!local->block[i].mr) {
            perror("Failed to register local dest ram block!\n");
            break;
        }
        rdma->total_registrations++;
    }

    if (i >= local->nb_blocks) {
        return 0;
    }

    for (i--; i >= 0; i--) {
        ibv_dereg_mr(local->block[i].mr);
        rdma->total_registrations--;
    }

    return -1;

}

/*
 * Find the ram block that corresponds to the page requested to be
 * transmitted by QEMU.
 *
 * Once the block is found, also identify which 'chunk' within that
 * block that the page belongs to.
 *
 * This search cannot fail or the migration will fail.
 */
static int qemu_rdma_search_ram_block(RDMAContext *rdma,
1131
                                      uintptr_t block_offset,
M
Michael R. Hines 已提交
1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158
                                      uint64_t offset,
                                      uint64_t length,
                                      uint64_t *block_index,
                                      uint64_t *chunk_index)
{
    uint64_t current_addr = block_offset + offset;
    RDMALocalBlock *block = g_hash_table_lookup(rdma->blockmap,
                                                (void *) block_offset);
    assert(block);
    assert(current_addr >= block->offset);
    assert((current_addr + length) <= (block->offset + block->length));

    *block_index = block->index;
    *chunk_index = ram_chunk_index(block->local_host_addr,
                block->local_host_addr + (current_addr - block->offset));

    return 0;
}

/*
 * Register a chunk with IB. If the chunk was already registered
 * previously, then skip.
 *
 * Also return the keys associated with the registration needed
 * to perform the actual RDMA operation.
 */
static int qemu_rdma_register_and_get_keys(RDMAContext *rdma,
1159
        RDMALocalBlock *block, uintptr_t host_addr,
M
Michael R. Hines 已提交
1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174
        uint32_t *lkey, uint32_t *rkey, int chunk,
        uint8_t *chunk_start, uint8_t *chunk_end)
{
    if (block->mr) {
        if (lkey) {
            *lkey = block->mr->lkey;
        }
        if (rkey) {
            *rkey = block->mr->rkey;
        }
        return 0;
    }

    /* allocate memory to store chunk MRs */
    if (!block->pmr) {
1175
        block->pmr = g_new0(struct ibv_mr *, block->nb_chunks);
M
Michael R. Hines 已提交
1176 1177 1178 1179 1180 1181 1182 1183 1184 1185
    }

    /*
     * If 'rkey', then we're the destination, so grant access to the source.
     *
     * If 'lkey', then we're the source VM, so grant access only to ourselves.
     */
    if (!block->pmr[chunk]) {
        uint64_t len = chunk_end - chunk_start;

1186
        trace_qemu_rdma_register_and_get_keys(len, chunk_start);
M
Michael R. Hines 已提交
1187 1188 1189 1190 1191 1192 1193 1194 1195

        block->pmr[chunk] = ibv_reg_mr(rdma->pd,
                chunk_start, len,
                (rkey ? (IBV_ACCESS_LOCAL_WRITE |
                        IBV_ACCESS_REMOTE_WRITE) : 0));

        if (!block->pmr[chunk]) {
            perror("Failed to register chunk!");
            fprintf(stderr, "Chunk details: block: %d chunk index %d"
1196 1197 1198 1199 1200 1201
                            " start %" PRIuPTR " end %" PRIuPTR
                            " host %" PRIuPTR
                            " local %" PRIuPTR " registrations: %d\n",
                            block->index, chunk, (uintptr_t)chunk_start,
                            (uintptr_t)chunk_end, host_addr,
                            (uintptr_t)block->local_host_addr,
M
Michael R. Hines 已提交
1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229
                            rdma->total_registrations);
            return -1;
        }
        rdma->total_registrations++;
    }

    if (lkey) {
        *lkey = block->pmr[chunk]->lkey;
    }
    if (rkey) {
        *rkey = block->pmr[chunk]->rkey;
    }
    return 0;
}

/*
 * Register (at connection time) the memory used for control
 * channel messages.
 */
static int qemu_rdma_reg_control(RDMAContext *rdma, int idx)
{
    rdma->wr_data[idx].control_mr = ibv_reg_mr(rdma->pd,
            rdma->wr_data[idx].control, RDMA_CONTROL_MAX_BUFFER,
            IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE);
    if (rdma->wr_data[idx].control_mr) {
        rdma->total_registrations++;
        return 0;
    }
1230
    error_report("qemu_rdma_reg_control failed");
M
Michael R. Hines 已提交
1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267
    return -1;
}

const char *print_wrid(int wrid)
{
    if (wrid >= RDMA_WRID_RECV_CONTROL) {
        return wrid_desc[RDMA_WRID_RECV_CONTROL];
    }
    return wrid_desc[wrid];
}

/*
 * RDMA requires memory registration (mlock/pinning), but this is not good for
 * overcommitment.
 *
 * In preparation for the future where LRU information or workload-specific
 * writable writable working set memory access behavior is available to QEMU
 * it would be nice to have in place the ability to UN-register/UN-pin
 * particular memory regions from the RDMA hardware when it is determine that
 * those regions of memory will likely not be accessed again in the near future.
 *
 * While we do not yet have such information right now, the following
 * compile-time option allows us to perform a non-optimized version of this
 * behavior.
 *
 * By uncommenting this option, you will cause *all* RDMA transfers to be
 * unregistered immediately after the transfer completes on both sides of the
 * connection. This has no effect in 'rdma-pin-all' mode, only regular mode.
 *
 * This will have a terrible impact on migration performance, so until future
 * workload information or LRU information is available, do not attempt to use
 * this feature except for basic testing.
 */
//#define RDMA_UNREGISTRATION_EXAMPLE

/*
 * Perform a non-optimized memory unregistration after every transfer
D
Dr. David Alan Gilbert 已提交
1268
 * for demonstration purposes, only if pin-all is not requested.
M
Michael R. Hines 已提交
1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295
 *
 * Potential optimizations:
 * 1. Start a new thread to run this function continuously
        - for bit clearing
        - and for receipt of unregister messages
 * 2. Use an LRU.
 * 3. Use workload hints.
 */
static int qemu_rdma_unregister_waiting(RDMAContext *rdma)
{
    while (rdma->unregistrations[rdma->unregister_current]) {
        int ret;
        uint64_t wr_id = rdma->unregistrations[rdma->unregister_current];
        uint64_t chunk =
            (wr_id & RDMA_WRID_CHUNK_MASK) >> RDMA_WRID_CHUNK_SHIFT;
        uint64_t index =
            (wr_id & RDMA_WRID_BLOCK_MASK) >> RDMA_WRID_BLOCK_SHIFT;
        RDMALocalBlock *block =
            &(rdma->local_ram_blocks.block[index]);
        RDMARegister reg = { .current_index = index };
        RDMAControlHeader resp = { .type = RDMA_CONTROL_UNREGISTER_FINISHED,
                                 };
        RDMAControlHeader head = { .len = sizeof(RDMARegister),
                                   .type = RDMA_CONTROL_UNREGISTER_REQUEST,
                                   .repeat = 1,
                                 };

1296 1297
        trace_qemu_rdma_unregister_waiting_proc(chunk,
                                                rdma->unregister_current);
M
Michael R. Hines 已提交
1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316

        rdma->unregistrations[rdma->unregister_current] = 0;
        rdma->unregister_current++;

        if (rdma->unregister_current == RDMA_SIGNALED_SEND_MAX) {
            rdma->unregister_current = 0;
        }


        /*
         * Unregistration is speculative (because migration is single-threaded
         * and we cannot break the protocol's inifinband message ordering).
         * Thus, if the memory is currently being used for transmission,
         * then abort the attempt to unregister and try again
         * later the next time a completion is received for this memory.
         */
        clear_bit(chunk, block->unregister_bitmap);

        if (test_bit(chunk, block->transit_bitmap)) {
1317
            trace_qemu_rdma_unregister_waiting_inflight(chunk);
M
Michael R. Hines 已提交
1318 1319 1320
            continue;
        }

1321
        trace_qemu_rdma_unregister_waiting_send(chunk);
M
Michael R. Hines 已提交
1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333

        ret = ibv_dereg_mr(block->pmr[chunk]);
        block->pmr[chunk] = NULL;
        block->remote_keys[chunk] = 0;

        if (ret != 0) {
            perror("unregistration chunk failed");
            return -ret;
        }
        rdma->total_registrations--;

        reg.key.chunk = chunk;
1334
        register_to_network(rdma, &reg);
M
Michael R. Hines 已提交
1335 1336 1337 1338 1339 1340
        ret = qemu_rdma_exchange_send(rdma, &head, (uint8_t *) &reg,
                                &resp, NULL, NULL);
        if (ret < 0) {
            return ret;
        }

1341
        trace_qemu_rdma_unregister_waiting_complete(chunk);
M
Michael R. Hines 已提交
1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365
    }

    return 0;
}

static uint64_t qemu_rdma_make_wrid(uint64_t wr_id, uint64_t index,
                                         uint64_t chunk)
{
    uint64_t result = wr_id & RDMA_WRID_TYPE_MASK;

    result |= (index << RDMA_WRID_BLOCK_SHIFT);
    result |= (chunk << RDMA_WRID_CHUNK_SHIFT);

    return result;
}

/*
 * Set bit for unregistration in the next iteration.
 * We cannot transmit right here, but will unpin later.
 */
static void qemu_rdma_signal_unregister(RDMAContext *rdma, uint64_t index,
                                        uint64_t chunk, uint64_t wr_id)
{
    if (rdma->unregistrations[rdma->unregister_next] != 0) {
1366
        error_report("rdma migration: queue is full");
M
Michael R. Hines 已提交
1367 1368 1369 1370
    } else {
        RDMALocalBlock *block = &(rdma->local_ram_blocks.block[index]);

        if (!test_and_set_bit(chunk, block->unregister_bitmap)) {
1371 1372
            trace_qemu_rdma_signal_unregister_append(chunk,
                                                     rdma->unregister_next);
M
Michael R. Hines 已提交
1373 1374 1375 1376 1377 1378 1379 1380

            rdma->unregistrations[rdma->unregister_next++] =
                    qemu_rdma_make_wrid(wr_id, index, chunk);

            if (rdma->unregister_next == RDMA_SIGNALED_SEND_MAX) {
                rdma->unregister_next = 0;
            }
        } else {
1381
            trace_qemu_rdma_signal_unregister_already(chunk);
M
Michael R. Hines 已提交
1382 1383 1384 1385 1386 1387 1388 1389 1390
        }
    }
}

/*
 * Consult the connection manager to see a work request
 * (of any kind) has completed.
 * Return the work request ID that completed.
 */
1391 1392
static uint64_t qemu_rdma_poll(RDMAContext *rdma, uint64_t *wr_id_out,
                               uint32_t *byte_len)
M
Michael R. Hines 已提交
1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405
{
    int ret;
    struct ibv_wc wc;
    uint64_t wr_id;

    ret = ibv_poll_cq(rdma->cq, 1, &wc);

    if (!ret) {
        *wr_id_out = RDMA_WRID_NONE;
        return 0;
    }

    if (ret < 0) {
1406
        error_report("ibv_poll_cq return %d", ret);
M
Michael R. Hines 已提交
1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421
        return ret;
    }

    wr_id = wc.wr_id & RDMA_WRID_TYPE_MASK;

    if (wc.status != IBV_WC_SUCCESS) {
        fprintf(stderr, "ibv_poll_cq wc.status=%d %s!\n",
                        wc.status, ibv_wc_status_str(wc.status));
        fprintf(stderr, "ibv_poll_cq wrid=%s!\n", wrid_desc[wr_id]);

        return -1;
    }

    if (rdma->control_ready_expected &&
        (wr_id >= RDMA_WRID_RECV_CONTROL)) {
1422
        trace_qemu_rdma_poll_recv(wrid_desc[RDMA_WRID_RECV_CONTROL],
M
Michael R. Hines 已提交
1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433
                  wr_id - RDMA_WRID_RECV_CONTROL, wr_id, rdma->nb_sent);
        rdma->control_ready_expected = 0;
    }

    if (wr_id == RDMA_WRID_RDMA_WRITE) {
        uint64_t chunk =
            (wc.wr_id & RDMA_WRID_CHUNK_MASK) >> RDMA_WRID_CHUNK_SHIFT;
        uint64_t index =
            (wc.wr_id & RDMA_WRID_BLOCK_MASK) >> RDMA_WRID_BLOCK_SHIFT;
        RDMALocalBlock *block = &(rdma->local_ram_blocks.block[index]);

1434
        trace_qemu_rdma_poll_write(print_wrid(wr_id), wr_id, rdma->nb_sent,
1435 1436
                                   index, chunk, block->local_host_addr,
                                   (void *)(uintptr_t)block->remote_host_addr);
M
Michael R. Hines 已提交
1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455

        clear_bit(chunk, block->transit_bitmap);

        if (rdma->nb_sent > 0) {
            rdma->nb_sent--;
        }

        if (!rdma->pin_all) {
            /*
             * FYI: If one wanted to signal a specific chunk to be unregistered
             * using LRU or workload-specific information, this is the function
             * you would call to do so. That chunk would then get asynchronously
             * unregistered later.
             */
#ifdef RDMA_UNREGISTRATION_EXAMPLE
            qemu_rdma_signal_unregister(rdma, index, chunk, wc.wr_id);
#endif
        }
    } else {
1456
        trace_qemu_rdma_poll_other(print_wrid(wr_id), wr_id, rdma->nb_sent);
M
Michael R. Hines 已提交
1457 1458 1459
    }

    *wr_id_out = wc.wr_id;
1460 1461 1462
    if (byte_len) {
        *byte_len = wc.byte_len;
    }
M
Michael R. Hines 已提交
1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479

    return  0;
}

/*
 * Block until the next work request has completed.
 *
 * First poll to see if a work request has already completed,
 * otherwise block.
 *
 * If we encounter completed work requests for IDs other than
 * the one we're interested in, then that's generally an error.
 *
 * The only exception is actual RDMA Write completions. These
 * completions only need to be recorded, but do not actually
 * need further processing.
 */
1480 1481
static int qemu_rdma_block_for_wrid(RDMAContext *rdma, int wrid_requested,
                                    uint32_t *byte_len)
M
Michael R. Hines 已提交
1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492
{
    int num_cq_events = 0, ret = 0;
    struct ibv_cq *cq;
    void *cq_ctx;
    uint64_t wr_id = RDMA_WRID_NONE, wr_id_in;

    if (ibv_req_notify_cq(rdma->cq, 0)) {
        return -1;
    }
    /* poll cq first */
    while (wr_id != wrid_requested) {
1493
        ret = qemu_rdma_poll(rdma, &wr_id_in, byte_len);
M
Michael R. Hines 已提交
1494 1495 1496 1497 1498 1499 1500 1501 1502 1503
        if (ret < 0) {
            return ret;
        }

        wr_id = wr_id_in & RDMA_WRID_TYPE_MASK;

        if (wr_id == RDMA_WRID_NONE) {
            break;
        }
        if (wr_id != wrid_requested) {
1504 1505
            trace_qemu_rdma_block_for_wrid_miss(print_wrid(wrid_requested),
                       wrid_requested, print_wrid(wr_id), wr_id);
M
Michael R. Hines 已提交
1506 1507 1508 1509 1510 1511 1512 1513 1514
        }
    }

    if (wr_id == wrid_requested) {
        return 0;
    }

    while (1) {
        /*
1515
         * Coroutine doesn't start until migration_fd_process_incoming()
M
Michael R. Hines 已提交
1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533
         * so don't yield unless we know we're running inside of a coroutine.
         */
        if (rdma->migration_started_on_destination) {
            yield_until_fd_readable(rdma->comp_channel->fd);
        }

        if (ibv_get_cq_event(rdma->comp_channel, &cq, &cq_ctx)) {
            perror("ibv_get_cq_event");
            goto err_block_for_wrid;
        }

        num_cq_events++;

        if (ibv_req_notify_cq(cq, 0)) {
            goto err_block_for_wrid;
        }

        while (wr_id != wrid_requested) {
1534
            ret = qemu_rdma_poll(rdma, &wr_id_in, byte_len);
M
Michael R. Hines 已提交
1535 1536 1537 1538 1539 1540 1541 1542 1543 1544
            if (ret < 0) {
                goto err_block_for_wrid;
            }

            wr_id = wr_id_in & RDMA_WRID_TYPE_MASK;

            if (wr_id == RDMA_WRID_NONE) {
                break;
            }
            if (wr_id != wrid_requested) {
1545 1546
                trace_qemu_rdma_block_for_wrid_miss(print_wrid(wrid_requested),
                                   wrid_requested, print_wrid(wr_id), wr_id);
M
Michael R. Hines 已提交
1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575
            }
        }

        if (wr_id == wrid_requested) {
            goto success_block_for_wrid;
        }
    }

success_block_for_wrid:
    if (num_cq_events) {
        ibv_ack_cq_events(cq, num_cq_events);
    }
    return 0;

err_block_for_wrid:
    if (num_cq_events) {
        ibv_ack_cq_events(cq, num_cq_events);
    }
    return ret;
}

/*
 * Post a SEND message work request for the control channel
 * containing some data and block until the post completes.
 */
static int qemu_rdma_post_send_control(RDMAContext *rdma, uint8_t *buf,
                                       RDMAControlHeader *head)
{
    int ret = 0;
1576
    RDMAWorkRequestData *wr = &rdma->wr_data[RDMA_WRID_CONTROL];
M
Michael R. Hines 已提交
1577 1578
    struct ibv_send_wr *bad_wr;
    struct ibv_sge sge = {
1579
                           .addr = (uintptr_t)(wr->control),
M
Michael R. Hines 已提交
1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590
                           .length = head->len + sizeof(RDMAControlHeader),
                           .lkey = wr->control_mr->lkey,
                         };
    struct ibv_send_wr send_wr = {
                                   .wr_id = RDMA_WRID_SEND_CONTROL,
                                   .opcode = IBV_WR_SEND,
                                   .send_flags = IBV_SEND_SIGNALED,
                                   .sg_list = &sge,
                                   .num_sge = 1,
                                };

1591
    trace_qemu_rdma_post_send_control(control_desc[head->type]);
M
Michael R. Hines 已提交
1592 1593 1594 1595 1596 1597 1598 1599 1600

    /*
     * We don't actually need to do a memcpy() in here if we used
     * the "sge" properly, but since we're only sending control messages
     * (not RAM in a performance-critical path), then its OK for now.
     *
     * The copy makes the RDMAControlHeader simpler to manipulate
     * for the time being.
     */
1601
    assert(head->len <= RDMA_CONTROL_MAX_BUFFER - sizeof(*head));
M
Michael R. Hines 已提交
1602 1603 1604 1605 1606 1607 1608 1609
    memcpy(wr->control, head, sizeof(RDMAControlHeader));
    control_to_network((void *) wr->control);

    if (buf) {
        memcpy(wr->control + sizeof(RDMAControlHeader), buf, head->len);
    }


M
Michael R. Hines 已提交
1610
    ret = ibv_post_send(rdma->qp, &send_wr, &bad_wr);
M
Michael R. Hines 已提交
1611

M
Michael R. Hines 已提交
1612
    if (ret > 0) {
1613
        error_report("Failed to use post IB SEND for control");
M
Michael R. Hines 已提交
1614
        return -ret;
M
Michael R. Hines 已提交
1615 1616
    }

1617
    ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_SEND_CONTROL, NULL);
M
Michael R. Hines 已提交
1618
    if (ret < 0) {
1619
        error_report("rdma migration: send polling control error");
M
Michael R. Hines 已提交
1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632
    }

    return ret;
}

/*
 * Post a RECV work request in anticipation of some future receipt
 * of data on the control channel.
 */
static int qemu_rdma_post_recv_control(RDMAContext *rdma, int idx)
{
    struct ibv_recv_wr *bad_wr;
    struct ibv_sge sge = {
1633
                            .addr = (uintptr_t)(rdma->wr_data[idx].control),
M
Michael R. Hines 已提交
1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657
                            .length = RDMA_CONTROL_MAX_BUFFER,
                            .lkey = rdma->wr_data[idx].control_mr->lkey,
                         };

    struct ibv_recv_wr recv_wr = {
                                    .wr_id = RDMA_WRID_RECV_CONTROL + idx,
                                    .sg_list = &sge,
                                    .num_sge = 1,
                                 };


    if (ibv_post_recv(rdma->qp, &recv_wr, &bad_wr)) {
        return -1;
    }

    return 0;
}

/*
 * Block and wait for a RECV control channel message to arrive.
 */
static int qemu_rdma_exchange_get_response(RDMAContext *rdma,
                RDMAControlHeader *head, int expecting, int idx)
{
1658 1659 1660
    uint32_t byte_len;
    int ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RECV_CONTROL + idx,
                                       &byte_len);
M
Michael R. Hines 已提交
1661 1662

    if (ret < 0) {
1663
        error_report("rdma migration: recv polling control error!");
M
Michael R. Hines 已提交
1664 1665 1666 1667 1668 1669
        return ret;
    }

    network_to_control((void *) rdma->wr_data[idx].control);
    memcpy(head, rdma->wr_data[idx].control, sizeof(RDMAControlHeader));

1670
    trace_qemu_rdma_exchange_get_response_start(control_desc[expecting]);
M
Michael R. Hines 已提交
1671 1672

    if (expecting == RDMA_CONTROL_NONE) {
1673 1674
        trace_qemu_rdma_exchange_get_response_none(control_desc[head->type],
                                             head->type);
M
Michael R. Hines 已提交
1675
    } else if (head->type != expecting || head->type == RDMA_CONTROL_ERROR) {
1676 1677
        error_report("Was expecting a %s (%d) control message"
                ", but got: %s (%d), length: %d",
M
Michael R. Hines 已提交
1678 1679
                control_desc[expecting], expecting,
                control_desc[head->type], head->type, head->len);
1680 1681 1682
        if (head->type == RDMA_CONTROL_ERROR) {
            rdma->received_error = true;
        }
M
Michael R. Hines 已提交
1683 1684
        return -EIO;
    }
1685
    if (head->len > RDMA_CONTROL_MAX_BUFFER - sizeof(*head)) {
1686
        error_report("too long length: %d", head->len);
1687 1688
        return -EINVAL;
    }
1689
    if (sizeof(*head) + head->len != byte_len) {
1690
        error_report("Malformed length: %d byte_len %d", head->len, byte_len);
1691 1692
        return -EINVAL;
    }
M
Michael R. Hines 已提交
1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751

    return 0;
}

/*
 * When a RECV work request has completed, the work request's
 * buffer is pointed at the header.
 *
 * This will advance the pointer to the data portion
 * of the control message of the work request's buffer that
 * was populated after the work request finished.
 */
static void qemu_rdma_move_header(RDMAContext *rdma, int idx,
                                  RDMAControlHeader *head)
{
    rdma->wr_data[idx].control_len = head->len;
    rdma->wr_data[idx].control_curr =
        rdma->wr_data[idx].control + sizeof(RDMAControlHeader);
}

/*
 * This is an 'atomic' high-level operation to deliver a single, unified
 * control-channel message.
 *
 * Additionally, if the user is expecting some kind of reply to this message,
 * they can request a 'resp' response message be filled in by posting an
 * additional work request on behalf of the user and waiting for an additional
 * completion.
 *
 * The extra (optional) response is used during registration to us from having
 * to perform an *additional* exchange of message just to provide a response by
 * instead piggy-backing on the acknowledgement.
 */
static int qemu_rdma_exchange_send(RDMAContext *rdma, RDMAControlHeader *head,
                                   uint8_t *data, RDMAControlHeader *resp,
                                   int *resp_idx,
                                   int (*callback)(RDMAContext *rdma))
{
    int ret = 0;

    /*
     * Wait until the dest is ready before attempting to deliver the message
     * by waiting for a READY message.
     */
    if (rdma->control_ready_expected) {
        RDMAControlHeader resp;
        ret = qemu_rdma_exchange_get_response(rdma,
                                    &resp, RDMA_CONTROL_READY, RDMA_WRID_READY);
        if (ret < 0) {
            return ret;
        }
    }

    /*
     * If the user is expecting a response, post a WR in anticipation of it.
     */
    if (resp) {
        ret = qemu_rdma_post_recv_control(rdma, RDMA_WRID_DATA);
        if (ret) {
1752
            error_report("rdma migration: error posting"
M
Michael R. Hines 已提交
1753 1754 1755 1756 1757 1758 1759 1760 1761 1762
                    " extra control recv for anticipated result!");
            return ret;
        }
    }

    /*
     * Post a WR to replace the one we just consumed for the READY message.
     */
    ret = qemu_rdma_post_recv_control(rdma, RDMA_WRID_READY);
    if (ret) {
1763
        error_report("rdma migration: error posting first control recv!");
M
Michael R. Hines 已提交
1764 1765 1766 1767 1768 1769 1770 1771 1772
        return ret;
    }

    /*
     * Deliver the control message that was requested.
     */
    ret = qemu_rdma_post_send_control(rdma, data, head);

    if (ret < 0) {
1773
        error_report("Failed to send control buffer!");
M
Michael R. Hines 已提交
1774 1775 1776 1777 1778 1779 1780 1781
        return ret;
    }

    /*
     * If we're expecting a response, block and wait for it.
     */
    if (resp) {
        if (callback) {
1782
            trace_qemu_rdma_exchange_send_issue_callback();
M
Michael R. Hines 已提交
1783 1784 1785 1786 1787 1788
            ret = callback(rdma);
            if (ret < 0) {
                return ret;
            }
        }

1789
        trace_qemu_rdma_exchange_send_waiting(control_desc[resp->type]);
M
Michael R. Hines 已提交
1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800
        ret = qemu_rdma_exchange_get_response(rdma, resp,
                                              resp->type, RDMA_WRID_DATA);

        if (ret < 0) {
            return ret;
        }

        qemu_rdma_move_header(rdma, RDMA_WRID_DATA, resp);
        if (resp_idx) {
            *resp_idx = RDMA_WRID_DATA;
        }
1801
        trace_qemu_rdma_exchange_send_received(control_desc[resp->type]);
M
Michael R. Hines 已提交
1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828
    }

    rdma->control_ready_expected = 1;

    return 0;
}

/*
 * This is an 'atomic' high-level operation to receive a single, unified
 * control-channel message.
 */
static int qemu_rdma_exchange_recv(RDMAContext *rdma, RDMAControlHeader *head,
                                int expecting)
{
    RDMAControlHeader ready = {
                                .len = 0,
                                .type = RDMA_CONTROL_READY,
                                .repeat = 1,
                              };
    int ret;

    /*
     * Inform the source that we're ready to receive a message.
     */
    ret = qemu_rdma_post_send_control(rdma, NULL, &ready);

    if (ret < 0) {
1829
        error_report("Failed to send control buffer!");
M
Michael R. Hines 已提交
1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849
        return ret;
    }

    /*
     * Block and wait for the message.
     */
    ret = qemu_rdma_exchange_get_response(rdma, head,
                                          expecting, RDMA_WRID_READY);

    if (ret < 0) {
        return ret;
    }

    qemu_rdma_move_header(rdma, RDMA_WRID_READY, head);

    /*
     * Post a new RECV work request to replace the one we just consumed.
     */
    ret = qemu_rdma_post_recv_control(rdma, RDMA_WRID_READY);
    if (ret) {
1850
        error_report("rdma migration: error posting second control recv!");
M
Michael R. Hines 已提交
1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882
        return ret;
    }

    return 0;
}

/*
 * Write an actual chunk of memory using RDMA.
 *
 * If we're using dynamic registration on the dest-side, we have to
 * send a registration command first.
 */
static int qemu_rdma_write_one(QEMUFile *f, RDMAContext *rdma,
                               int current_index, uint64_t current_addr,
                               uint64_t length)
{
    struct ibv_sge sge;
    struct ibv_send_wr send_wr = { 0 };
    struct ibv_send_wr *bad_wr;
    int reg_result_idx, ret, count = 0;
    uint64_t chunk, chunks;
    uint8_t *chunk_start, *chunk_end;
    RDMALocalBlock *block = &(rdma->local_ram_blocks.block[current_index]);
    RDMARegister reg;
    RDMARegisterResult *reg_result;
    RDMAControlHeader resp = { .type = RDMA_CONTROL_REGISTER_RESULT };
    RDMAControlHeader head = { .len = sizeof(RDMARegister),
                               .type = RDMA_CONTROL_REGISTER_REQUEST,
                               .repeat = 1,
                             };

retry:
1883
    sge.addr = (uintptr_t)(block->local_host_addr +
M
Michael R. Hines 已提交
1884 1885 1886
                            (current_addr - block->offset));
    sge.length = length;

1887 1888
    chunk = ram_chunk_index(block->local_host_addr,
                            (uint8_t *)(uintptr_t)sge.addr);
M
Michael R. Hines 已提交
1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904
    chunk_start = ram_chunk_start(block, chunk);

    if (block->is_ram_block) {
        chunks = length / (1UL << RDMA_REG_CHUNK_SHIFT);

        if (chunks && ((length % (1UL << RDMA_REG_CHUNK_SHIFT)) == 0)) {
            chunks--;
        }
    } else {
        chunks = block->length / (1UL << RDMA_REG_CHUNK_SHIFT);

        if (chunks && ((block->length % (1UL << RDMA_REG_CHUNK_SHIFT)) == 0)) {
            chunks--;
        }
    }

1905 1906 1907
    trace_qemu_rdma_write_one_top(chunks + 1,
                                  (chunks + 1) *
                                  (1UL << RDMA_REG_CHUNK_SHIFT) / 1024 / 1024);
M
Michael R. Hines 已提交
1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918

    chunk_end = ram_chunk_end(block, chunk + chunks);

    if (!rdma->pin_all) {
#ifdef RDMA_UNREGISTRATION_EXAMPLE
        qemu_rdma_unregister_waiting(rdma);
#endif
    }

    while (test_bit(chunk, block->transit_bitmap)) {
        (void)count;
1919
        trace_qemu_rdma_write_one_block(count++, current_index, chunk,
M
Michael R. Hines 已提交
1920 1921
                sge.addr, length, rdma->nb_sent, block->nb_chunks);

1922
        ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RDMA_WRITE, NULL);
M
Michael R. Hines 已提交
1923 1924

        if (ret < 0) {
1925
            error_report("Failed to Wait for previous write to complete "
M
Michael R. Hines 已提交
1926
                    "block %d chunk %" PRIu64
1927
                    " current %" PRIu64 " len %" PRIu64 " %d",
M
Michael R. Hines 已提交
1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940
                    current_index, chunk, sge.addr, length, rdma->nb_sent);
            return ret;
        }
    }

    if (!rdma->pin_all || !block->is_ram_block) {
        if (!block->remote_keys[chunk]) {
            /*
             * This chunk has not yet been registered, so first check to see
             * if the entire chunk is zero. If so, tell the other size to
             * memset() + madvise() the entire chunk without RDMA.
             */

1941
            if (buffer_is_zero((void *)(uintptr_t)sge.addr, length)) {
M
Michael R. Hines 已提交
1942 1943 1944 1945 1946 1947 1948 1949 1950 1951
                RDMACompress comp = {
                                        .offset = current_addr,
                                        .value = 0,
                                        .block_idx = current_index,
                                        .length = length,
                                    };

                head.len = sizeof(comp);
                head.type = RDMA_CONTROL_COMPRESS;

1952 1953
                trace_qemu_rdma_write_one_zero(chunk, sge.length,
                                               current_index, current_addr);
M
Michael R. Hines 已提交
1954

1955
                compress_to_network(rdma, &comp);
M
Michael R. Hines 已提交
1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978
                ret = qemu_rdma_exchange_send(rdma, &head,
                                (uint8_t *) &comp, NULL, NULL, NULL);

                if (ret < 0) {
                    return -EIO;
                }

                acct_update_position(f, sge.length, true);

                return 1;
            }

            /*
             * Otherwise, tell other side to register.
             */
            reg.current_index = current_index;
            if (block->is_ram_block) {
                reg.key.current_addr = current_addr;
            } else {
                reg.key.chunk = chunk;
            }
            reg.chunks = chunks;

1979 1980
            trace_qemu_rdma_write_one_sendreg(chunk, sge.length, current_index,
                                              current_addr);
M
Michael R. Hines 已提交
1981

1982
            register_to_network(rdma, &reg);
M
Michael R. Hines 已提交
1983 1984 1985 1986 1987 1988 1989
            ret = qemu_rdma_exchange_send(rdma, &head, (uint8_t *) &reg,
                                    &resp, &reg_result_idx, NULL);
            if (ret < 0) {
                return ret;
            }

            /* try to overlap this single registration with the one we sent. */
1990
            if (qemu_rdma_register_and_get_keys(rdma, block, sge.addr,
M
Michael R. Hines 已提交
1991 1992
                                                &sge.lkey, NULL, chunk,
                                                chunk_start, chunk_end)) {
1993
                error_report("cannot get lkey");
M
Michael R. Hines 已提交
1994 1995 1996 1997 1998 1999 2000 2001
                return -EINVAL;
            }

            reg_result = (RDMARegisterResult *)
                    rdma->wr_data[reg_result_idx].control_curr;

            network_to_result(reg_result);

2002 2003
            trace_qemu_rdma_write_one_recvregres(block->remote_keys[chunk],
                                                 reg_result->rkey, chunk);
M
Michael R. Hines 已提交
2004 2005 2006 2007 2008

            block->remote_keys[chunk] = reg_result->rkey;
            block->remote_host_addr = reg_result->host_addr;
        } else {
            /* already registered before */
2009
            if (qemu_rdma_register_and_get_keys(rdma, block, sge.addr,
M
Michael R. Hines 已提交
2010 2011
                                                &sge.lkey, NULL, chunk,
                                                chunk_start, chunk_end)) {
2012
                error_report("cannot get lkey!");
M
Michael R. Hines 已提交
2013 2014 2015 2016 2017 2018 2019 2020
                return -EINVAL;
            }
        }

        send_wr.wr.rdma.rkey = block->remote_keys[chunk];
    } else {
        send_wr.wr.rdma.rkey = block->remote_rkey;

2021
        if (qemu_rdma_register_and_get_keys(rdma, block, sge.addr,
M
Michael R. Hines 已提交
2022 2023
                                                     &sge.lkey, NULL, chunk,
                                                     chunk_start, chunk_end)) {
2024
            error_report("cannot get lkey!");
M
Michael R. Hines 已提交
2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044
            return -EINVAL;
        }
    }

    /*
     * Encode the ram block index and chunk within this wrid.
     * We will use this information at the time of completion
     * to figure out which bitmap to check against and then which
     * chunk in the bitmap to look for.
     */
    send_wr.wr_id = qemu_rdma_make_wrid(RDMA_WRID_RDMA_WRITE,
                                        current_index, chunk);

    send_wr.opcode = IBV_WR_RDMA_WRITE;
    send_wr.send_flags = IBV_SEND_SIGNALED;
    send_wr.sg_list = &sge;
    send_wr.num_sge = 1;
    send_wr.wr.rdma.remote_addr = block->remote_host_addr +
                                (current_addr - block->offset);

2045 2046
    trace_qemu_rdma_write_one_post(chunk, sge.addr, send_wr.wr.rdma.remote_addr,
                                   sge.length);
M
Michael R. Hines 已提交
2047 2048 2049 2050 2051 2052 2053 2054

    /*
     * ibv_post_send() does not return negative error numbers,
     * per the specification they are positive - no idea why.
     */
    ret = ibv_post_send(rdma->qp, &send_wr, &bad_wr);

    if (ret == ENOMEM) {
2055
        trace_qemu_rdma_write_one_queue_full();
2056
        ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RDMA_WRITE, NULL);
M
Michael R. Hines 已提交
2057
        if (ret < 0) {
2058 2059
            error_report("rdma migration: failed to make "
                         "room in full send queue! %d", ret);
M
Michael R. Hines 已提交
2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099
            return ret;
        }

        goto retry;

    } else if (ret > 0) {
        perror("rdma migration: post rdma write failed");
        return -ret;
    }

    set_bit(chunk, block->transit_bitmap);
    acct_update_position(f, sge.length, false);
    rdma->total_writes++;

    return 0;
}

/*
 * Push out any unwritten RDMA operations.
 *
 * We support sending out multiple chunks at the same time.
 * Not all of them need to get signaled in the completion queue.
 */
static int qemu_rdma_write_flush(QEMUFile *f, RDMAContext *rdma)
{
    int ret;

    if (!rdma->current_length) {
        return 0;
    }

    ret = qemu_rdma_write_one(f, rdma,
            rdma->current_index, rdma->current_addr, rdma->current_length);

    if (ret < 0) {
        return ret;
    }

    if (ret == 0) {
        rdma->nb_sent++;
2100
        trace_qemu_rdma_write_flush(rdma->nb_sent);
M
Michael R. Hines 已提交
2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111
    }

    rdma->current_length = 0;
    rdma->current_addr = 0;

    return 0;
}

static inline int qemu_rdma_buffer_mergable(RDMAContext *rdma,
                    uint64_t offset, uint64_t len)
{
2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126
    RDMALocalBlock *block;
    uint8_t *host_addr;
    uint8_t *chunk_end;

    if (rdma->current_index < 0) {
        return 0;
    }

    if (rdma->current_chunk < 0) {
        return 0;
    }

    block = &(rdma->local_ram_blocks.block[rdma->current_index]);
    host_addr = block->local_host_addr + (offset - block->offset);
    chunk_end = ram_chunk_end(block, rdma->current_chunk);
M
Michael R. Hines 已提交
2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184

    if (rdma->current_length == 0) {
        return 0;
    }

    /*
     * Only merge into chunk sequentially.
     */
    if (offset != (rdma->current_addr + rdma->current_length)) {
        return 0;
    }

    if (offset < block->offset) {
        return 0;
    }

    if ((offset + len) > (block->offset + block->length)) {
        return 0;
    }

    if ((host_addr + len) > chunk_end) {
        return 0;
    }

    return 1;
}

/*
 * We're not actually writing here, but doing three things:
 *
 * 1. Identify the chunk the buffer belongs to.
 * 2. If the chunk is full or the buffer doesn't belong to the current
 *    chunk, then start a new chunk and flush() the old chunk.
 * 3. To keep the hardware busy, we also group chunks into batches
 *    and only require that a batch gets acknowledged in the completion
 *    qeueue instead of each individual chunk.
 */
static int qemu_rdma_write(QEMUFile *f, RDMAContext *rdma,
                           uint64_t block_offset, uint64_t offset,
                           uint64_t len)
{
    uint64_t current_addr = block_offset + offset;
    uint64_t index = rdma->current_index;
    uint64_t chunk = rdma->current_chunk;
    int ret;

    /* If we cannot merge it, we flush the current buffer first. */
    if (!qemu_rdma_buffer_mergable(rdma, current_addr, len)) {
        ret = qemu_rdma_write_flush(f, rdma);
        if (ret) {
            return ret;
        }
        rdma->current_length = 0;
        rdma->current_addr = current_addr;

        ret = qemu_rdma_search_ram_block(rdma, block_offset,
                                         offset, len, &index, &chunk);
        if (ret) {
2185
            error_report("ram block search failed");
M
Michael R. Hines 已提交
2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207
            return ret;
        }
        rdma->current_index = index;
        rdma->current_chunk = chunk;
    }

    /* merge it */
    rdma->current_length += len;

    /* flush it if buffer is too large */
    if (rdma->current_length >= RDMA_MERGE_MAX) {
        return qemu_rdma_write_flush(f, rdma);
    }

    return 0;
}

static void qemu_rdma_cleanup(RDMAContext *rdma)
{
    struct rdma_cm_event *cm_event;
    int ret, idx;

2208
    if (rdma->cm_id && rdma->connected) {
2209
        if (rdma->error_state && !rdma->received_error) {
M
Michael R. Hines 已提交
2210 2211 2212 2213
            RDMAControlHeader head = { .len = 0,
                                       .type = RDMA_CONTROL_ERROR,
                                       .repeat = 1,
                                     };
2214
            error_report("Early error. Sending error.");
M
Michael R. Hines 已提交
2215 2216 2217 2218 2219
            qemu_rdma_post_send_control(rdma, NULL, &head);
        }

        ret = rdma_disconnect(rdma->cm_id);
        if (!ret) {
2220
            trace_qemu_rdma_cleanup_waiting_for_disconnect();
M
Michael R. Hines 已提交
2221 2222 2223 2224 2225
            ret = rdma_get_cm_event(rdma->channel, &cm_event);
            if (!ret) {
                rdma_ack_cm_event(cm_event);
            }
        }
2226
        trace_qemu_rdma_cleanup_disconnect();
2227
        rdma->connected = false;
M
Michael R. Hines 已提交
2228 2229
    }

2230 2231
    g_free(rdma->dest_blocks);
    rdma->dest_blocks = NULL;
M
Michael R. Hines 已提交
2232

2233
    for (idx = 0; idx < RDMA_WRID_MAX; idx++) {
M
Michael R. Hines 已提交
2234 2235 2236 2237 2238 2239 2240 2241 2242
        if (rdma->wr_data[idx].control_mr) {
            rdma->total_registrations--;
            ibv_dereg_mr(rdma->wr_data[idx].control_mr);
        }
        rdma->wr_data[idx].control_mr = NULL;
    }

    if (rdma->local_ram_blocks.block) {
        while (rdma->local_ram_blocks.nb_blocks) {
2243
            rdma_delete_block(rdma, &rdma->local_ram_blocks.block[0]);
M
Michael R. Hines 已提交
2244 2245 2246
        }
    }

2247 2248 2249 2250
    if (rdma->qp) {
        rdma_destroy_qp(rdma->cm_id);
        rdma->qp = NULL;
    }
M
Michael R. Hines 已提交
2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266
    if (rdma->cq) {
        ibv_destroy_cq(rdma->cq);
        rdma->cq = NULL;
    }
    if (rdma->comp_channel) {
        ibv_destroy_comp_channel(rdma->comp_channel);
        rdma->comp_channel = NULL;
    }
    if (rdma->pd) {
        ibv_dealloc_pd(rdma->pd);
        rdma->pd = NULL;
    }
    if (rdma->cm_id) {
        rdma_destroy_id(rdma->cm_id);
        rdma->cm_id = NULL;
    }
2267 2268 2269 2270
    if (rdma->listen_id) {
        rdma_destroy_id(rdma->listen_id);
        rdma->listen_id = NULL;
    }
M
Michael R. Hines 已提交
2271 2272 2273 2274
    if (rdma->channel) {
        rdma_destroy_event_channel(rdma->channel);
        rdma->channel = NULL;
    }
2275 2276
    g_free(rdma->host);
    rdma->host = NULL;
M
Michael R. Hines 已提交
2277 2278 2279
}


2280
static int qemu_rdma_source_init(RDMAContext *rdma, bool pin_all, Error **errp)
M
Michael R. Hines 已提交
2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299
{
    int ret, idx;
    Error *local_err = NULL, **temp = &local_err;

    /*
     * Will be validated against destination's actual capabilities
     * after the connect() completes.
     */
    rdma->pin_all = pin_all;

    ret = qemu_rdma_resolve_host(rdma, temp);
    if (ret) {
        goto err_rdma_source_init;
    }

    ret = qemu_rdma_alloc_pd_cq(rdma);
    if (ret) {
        ERROR(temp, "rdma migration: error allocating pd and cq! Your mlock()"
                    " limits may be too low. Please check $ ulimit -a # and "
2300
                    "search for 'ulimit -l' in the output");
M
Michael R. Hines 已提交
2301 2302 2303 2304 2305
        goto err_rdma_source_init;
    }

    ret = qemu_rdma_alloc_qp(rdma);
    if (ret) {
2306
        ERROR(temp, "rdma migration: error allocating qp!");
M
Michael R. Hines 已提交
2307 2308 2309 2310 2311
        goto err_rdma_source_init;
    }

    ret = qemu_rdma_init_ram_blocks(rdma);
    if (ret) {
2312
        ERROR(temp, "rdma migration: error initializing ram blocks!");
M
Michael R. Hines 已提交
2313 2314 2315
        goto err_rdma_source_init;
    }

D
Dr. David Alan Gilbert 已提交
2316 2317 2318 2319 2320 2321 2322 2323
    /* Build the hash that maps from offset to RAMBlock */
    rdma->blockmap = g_hash_table_new(g_direct_hash, g_direct_equal);
    for (idx = 0; idx < rdma->local_ram_blocks.nb_blocks; idx++) {
        g_hash_table_insert(rdma->blockmap,
                (void *)(uintptr_t)rdma->local_ram_blocks.block[idx].offset,
                &rdma->local_ram_blocks.block[idx]);
    }

2324
    for (idx = 0; idx < RDMA_WRID_MAX; idx++) {
M
Michael R. Hines 已提交
2325 2326
        ret = qemu_rdma_reg_control(rdma, idx);
        if (ret) {
2327
            ERROR(temp, "rdma migration: error registering %d control!",
M
Michael R. Hines 已提交
2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359
                                                            idx);
            goto err_rdma_source_init;
        }
    }

    return 0;

err_rdma_source_init:
    error_propagate(errp, local_err);
    qemu_rdma_cleanup(rdma);
    return -1;
}

static int qemu_rdma_connect(RDMAContext *rdma, Error **errp)
{
    RDMACapabilities cap = {
                                .version = RDMA_CONTROL_VERSION_CURRENT,
                                .flags = 0,
                           };
    struct rdma_conn_param conn_param = { .initiator_depth = 2,
                                          .retry_count = 5,
                                          .private_data = &cap,
                                          .private_data_len = sizeof(cap),
                                        };
    struct rdma_cm_event *cm_event;
    int ret;

    /*
     * Only negotiate the capability with destination if the user
     * on the source first requested the capability.
     */
    if (rdma->pin_all) {
2360
        trace_qemu_rdma_connect_pin_all_requested();
M
Michael R. Hines 已提交
2361 2362 2363 2364 2365 2366 2367 2368
        cap.flags |= RDMA_CAPABILITY_PIN_ALL;
    }

    caps_to_network(&cap);

    ret = rdma_connect(rdma->cm_id, &conn_param);
    if (ret) {
        perror("rdma_connect");
2369
        ERROR(errp, "connecting to destination!");
M
Michael R. Hines 已提交
2370 2371 2372 2373 2374 2375
        goto err_rdma_source_connect;
    }

    ret = rdma_get_cm_event(rdma->channel, &cm_event);
    if (ret) {
        perror("rdma_get_cm_event after rdma_connect");
2376
        ERROR(errp, "connecting to destination!");
M
Michael R. Hines 已提交
2377 2378 2379 2380 2381 2382
        rdma_ack_cm_event(cm_event);
        goto err_rdma_source_connect;
    }

    if (cm_event->event != RDMA_CM_EVENT_ESTABLISHED) {
        perror("rdma_get_cm_event != EVENT_ESTABLISHED after rdma_connect");
2383
        ERROR(errp, "connecting to destination!");
M
Michael R. Hines 已提交
2384 2385 2386
        rdma_ack_cm_event(cm_event);
        goto err_rdma_source_connect;
    }
2387
    rdma->connected = true;
M
Michael R. Hines 已提交
2388 2389 2390 2391 2392 2393 2394 2395 2396 2397

    memcpy(&cap, cm_event->param.conn.private_data, sizeof(cap));
    network_to_caps(&cap);

    /*
     * Verify that the *requested* capabilities are supported by the destination
     * and disable them otherwise.
     */
    if (rdma->pin_all && !(cap.flags & RDMA_CAPABILITY_PIN_ALL)) {
        ERROR(errp, "Server cannot support pinning all memory. "
2398
                        "Will register memory dynamically.");
M
Michael R. Hines 已提交
2399 2400 2401
        rdma->pin_all = false;
    }

2402
    trace_qemu_rdma_connect_pin_all_outcome(rdma->pin_all);
M
Michael R. Hines 已提交
2403 2404 2405

    rdma_ack_cm_event(cm_event);

I
Isaku Yamahata 已提交
2406
    ret = qemu_rdma_post_recv_control(rdma, RDMA_WRID_READY);
M
Michael R. Hines 已提交
2407
    if (ret) {
2408
        ERROR(errp, "posting second control recv!");
M
Michael R. Hines 已提交
2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422
        goto err_rdma_source_connect;
    }

    rdma->control_ready_expected = 1;
    rdma->nb_sent = 0;
    return 0;

err_rdma_source_connect:
    qemu_rdma_cleanup(rdma);
    return -1;
}

static int qemu_rdma_dest_init(RDMAContext *rdma, Error **errp)
{
2423
    int ret, idx;
M
Michael R. Hines 已提交
2424 2425
    struct rdma_cm_id *listen_id;
    char ip[40] = "unknown";
2426
    struct rdma_addrinfo *res, *e;
2427
    char port_str[16];
M
Michael R. Hines 已提交
2428

2429
    for (idx = 0; idx < RDMA_WRID_MAX; idx++) {
M
Michael R. Hines 已提交
2430 2431 2432 2433
        rdma->wr_data[idx].control_len = 0;
        rdma->wr_data[idx].control_curr = NULL;
    }

2434
    if (!rdma->host || !rdma->host[0]) {
2435
        ERROR(errp, "RDMA host is not set!");
M
Michael R. Hines 已提交
2436 2437 2438 2439 2440 2441
        rdma->error_state = -EINVAL;
        return -1;
    }
    /* create CM channel */
    rdma->channel = rdma_create_event_channel();
    if (!rdma->channel) {
2442
        ERROR(errp, "could not create rdma event channel");
M
Michael R. Hines 已提交
2443 2444 2445 2446 2447 2448 2449
        rdma->error_state = -EINVAL;
        return -1;
    }

    /* create CM id */
    ret = rdma_create_id(rdma->channel, &listen_id, NULL, RDMA_PS_TCP);
    if (ret) {
2450
        ERROR(errp, "could not create cm_id!");
M
Michael R. Hines 已提交
2451 2452 2453
        goto err_dest_init_create_listen_id;
    }

2454 2455
    snprintf(port_str, 16, "%d", rdma->port);
    port_str[15] = '\0';
M
Michael R. Hines 已提交
2456

2457 2458 2459 2460 2461
    ret = rdma_getaddrinfo(rdma->host, port_str, NULL, &res);
    if (ret < 0) {
        ERROR(errp, "could not rdma_getaddrinfo address %s", rdma->host);
        goto err_dest_init_bind_addr;
    }
2462

2463 2464 2465 2466 2467 2468 2469
    for (e = res; e != NULL; e = e->ai_next) {
        inet_ntop(e->ai_family,
            &((struct sockaddr_in *) e->ai_dst_addr)->sin_addr, ip, sizeof ip);
        trace_qemu_rdma_dest_init_trying(rdma->host, ip);
        ret = rdma_bind_addr(listen_id, e->ai_dst_addr);
        if (ret) {
            continue;
M
Michael R. Hines 已提交
2470
        }
2471
        if (e->ai_family == AF_INET6) {
2472
            ret = qemu_rdma_broken_ipv6_kernel(listen_id->verbs, errp);
2473 2474
            if (ret) {
                continue;
2475 2476
            }
        }
2477 2478
        break;
    }
2479

2480
    if (!e) {
2481 2482
        ERROR(errp, "Error: could not rdma_bind_addr!");
        goto err_dest_init_bind_addr;
M
Michael R. Hines 已提交
2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504
    }

    rdma->listen_id = listen_id;
    qemu_rdma_dump_gid("dest_init", listen_id);
    return 0;

err_dest_init_bind_addr:
    rdma_destroy_id(listen_id);
err_dest_init_create_listen_id:
    rdma_destroy_event_channel(rdma->channel);
    rdma->channel = NULL;
    rdma->error_state = ret;
    return ret;

}

static void *qemu_rdma_data_init(const char *host_port, Error **errp)
{
    RDMAContext *rdma = NULL;
    InetSocketAddress *addr;

    if (host_port) {
2505
        rdma = g_new0(RDMAContext, 1);
M
Michael R. Hines 已提交
2506 2507 2508
        rdma->current_index = -1;
        rdma->current_chunk = -1;

2509 2510
        addr = g_new(InetSocketAddress, 1);
        if (!inet_parse(addr, host_port, NULL)) {
M
Michael R. Hines 已提交
2511 2512 2513 2514 2515
            rdma->port = atoi(addr->port);
            rdma->host = g_strdup(addr->host);
        } else {
            ERROR(errp, "bad RDMA migration address '%s'", host_port);
            g_free(rdma);
M
Michael R. Hines 已提交
2516
            rdma = NULL;
M
Michael R. Hines 已提交
2517
        }
M
Michael R. Hines 已提交
2518 2519

        qapi_free_InetSocketAddress(addr);
M
Michael R. Hines 已提交
2520 2521 2522 2523 2524 2525 2526 2527
    }

    return rdma;
}

/*
 * QEMUFile interface to the control channel.
 * SEND messages for control only.
2528
 * VM's ram is handled with regular RDMA messages.
M
Michael R. Hines 已提交
2529
 */
2530 2531 2532 2533 2534 2535 2536 2537 2538 2539
static ssize_t qio_channel_rdma_writev(QIOChannel *ioc,
                                       const struct iovec *iov,
                                       size_t niov,
                                       int *fds,
                                       size_t nfds,
                                       Error **errp)
{
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(ioc);
    QEMUFile *f = rioc->file;
    RDMAContext *rdma = rioc->rdma;
M
Michael R. Hines 已提交
2540
    int ret;
2541 2542
    ssize_t done = 0;
    size_t i;
M
Michael R. Hines 已提交
2543 2544 2545 2546 2547

    CHECK_ERROR_STATE();

    /*
     * Push out any writes that
2548
     * we're queued up for VM's ram.
M
Michael R. Hines 已提交
2549 2550 2551 2552 2553 2554 2555
     */
    ret = qemu_rdma_write_flush(f, rdma);
    if (ret < 0) {
        rdma->error_state = ret;
        return ret;
    }

2556 2557 2558 2559 2560
    for (i = 0; i < niov; i++) {
        size_t remaining = iov[i].iov_len;
        uint8_t * data = (void *)iov[i].iov_base;
        while (remaining) {
            RDMAControlHeader head;
M
Michael R. Hines 已提交
2561

2562 2563
            rioc->len = MIN(remaining, RDMA_SEND_INCREMENT);
            remaining -= rioc->len;
M
Michael R. Hines 已提交
2564

2565 2566
            head.len = rioc->len;
            head.type = RDMA_CONTROL_QEMU_FILE;
M
Michael R. Hines 已提交
2567

2568
            ret = qemu_rdma_exchange_send(rdma, &head, data, NULL, NULL, NULL);
M
Michael R. Hines 已提交
2569

2570 2571 2572 2573
            if (ret < 0) {
                rdma->error_state = ret;
                return ret;
            }
M
Michael R. Hines 已提交
2574

2575 2576 2577
            data += rioc->len;
            done += rioc->len;
        }
M
Michael R. Hines 已提交
2578 2579
    }

2580
    return done;
M
Michael R. Hines 已提交
2581 2582 2583
}

static size_t qemu_rdma_fill(RDMAContext *rdma, uint8_t *buf,
2584
                             size_t size, int idx)
M
Michael R. Hines 已提交
2585 2586 2587 2588
{
    size_t len = 0;

    if (rdma->wr_data[idx].control_len) {
2589
        trace_qemu_rdma_fill(rdma->wr_data[idx].control_len, size);
M
Michael R. Hines 已提交
2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604

        len = MIN(size, rdma->wr_data[idx].control_len);
        memcpy(buf, rdma->wr_data[idx].control_curr, len);
        rdma->wr_data[idx].control_curr += len;
        rdma->wr_data[idx].control_len -= len;
    }

    return len;
}

/*
 * QEMUFile interface to the control channel.
 * RDMA links don't use bytestreams, so we have to
 * return bytes to QEMUFile opportunistically.
 */
2605 2606 2607 2608 2609 2610 2611 2612 2613
static ssize_t qio_channel_rdma_readv(QIOChannel *ioc,
                                      const struct iovec *iov,
                                      size_t niov,
                                      int **fds,
                                      size_t *nfds,
                                      Error **errp)
{
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(ioc);
    RDMAContext *rdma = rioc->rdma;
M
Michael R. Hines 已提交
2614 2615
    RDMAControlHeader head;
    int ret = 0;
2616 2617
    ssize_t i;
    size_t done = 0;
M
Michael R. Hines 已提交
2618 2619 2620

    CHECK_ERROR_STATE();

2621 2622 2623
    for (i = 0; i < niov; i++) {
        size_t want = iov[i].iov_len;
        uint8_t *data = (void *)iov[i].iov_base;
M
Michael R. Hines 已提交
2624

2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636
        /*
         * First, we hold on to the last SEND message we
         * were given and dish out the bytes until we run
         * out of bytes.
         */
        ret = qemu_rdma_fill(rioc->rdma, data, want, 0);
        done += ret;
        want -= ret;
        /* Got what we needed, so go to next iovec */
        if (want == 0) {
            continue;
        }
M
Michael R. Hines 已提交
2637

2638 2639 2640 2641 2642
        /* If we got any data so far, then don't wait
         * for more, just return what we have */
        if (done > 0) {
            break;
        }
M
Michael R. Hines 已提交
2643

2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672

        /* We've got nothing at all, so lets wait for
         * more to arrive
         */
        ret = qemu_rdma_exchange_recv(rdma, &head, RDMA_CONTROL_QEMU_FILE);

        if (ret < 0) {
            rdma->error_state = ret;
            return ret;
        }

        /*
         * SEND was received with new bytes, now try again.
         */
        ret = qemu_rdma_fill(rioc->rdma, data, want, 0);
        done += ret;
        want -= ret;

        /* Still didn't get enough, so lets just return */
        if (want) {
            if (done == 0) {
                return QIO_CHANNEL_ERR_BLOCK;
            } else {
                break;
            }
        }
    }
    rioc->len = done;
    return rioc->len;
M
Michael R. Hines 已提交
2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686
}

/*
 * Block until all the outstanding chunks have been delivered by the hardware.
 */
static int qemu_rdma_drain_cq(QEMUFile *f, RDMAContext *rdma)
{
    int ret;

    if (qemu_rdma_write_flush(f, rdma) < 0) {
        return -EIO;
    }

    while (rdma->nb_sent) {
2687
        ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RDMA_WRITE, NULL);
M
Michael R. Hines 已提交
2688
        if (ret < 0) {
2689
            error_report("rdma migration: complete polling error!");
M
Michael R. Hines 已提交
2690 2691 2692 2693 2694 2695 2696 2697 2698
            return -EIO;
        }
    }

    qemu_rdma_unregister_waiting(rdma);

    return 0;
}

2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786

static int qio_channel_rdma_set_blocking(QIOChannel *ioc,
                                         bool blocking,
                                         Error **errp)
{
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(ioc);
    /* XXX we should make readv/writev actually honour this :-) */
    rioc->blocking = blocking;
    return 0;
}


typedef struct QIOChannelRDMASource QIOChannelRDMASource;
struct QIOChannelRDMASource {
    GSource parent;
    QIOChannelRDMA *rioc;
    GIOCondition condition;
};

static gboolean
qio_channel_rdma_source_prepare(GSource *source,
                                gint *timeout)
{
    QIOChannelRDMASource *rsource = (QIOChannelRDMASource *)source;
    RDMAContext *rdma = rsource->rioc->rdma;
    GIOCondition cond = 0;
    *timeout = -1;

    if (rdma->wr_data[0].control_len) {
        cond |= G_IO_IN;
    }
    cond |= G_IO_OUT;

    return cond & rsource->condition;
}

static gboolean
qio_channel_rdma_source_check(GSource *source)
{
    QIOChannelRDMASource *rsource = (QIOChannelRDMASource *)source;
    RDMAContext *rdma = rsource->rioc->rdma;
    GIOCondition cond = 0;

    if (rdma->wr_data[0].control_len) {
        cond |= G_IO_IN;
    }
    cond |= G_IO_OUT;

    return cond & rsource->condition;
}

static gboolean
qio_channel_rdma_source_dispatch(GSource *source,
                                 GSourceFunc callback,
                                 gpointer user_data)
{
    QIOChannelFunc func = (QIOChannelFunc)callback;
    QIOChannelRDMASource *rsource = (QIOChannelRDMASource *)source;
    RDMAContext *rdma = rsource->rioc->rdma;
    GIOCondition cond = 0;

    if (rdma->wr_data[0].control_len) {
        cond |= G_IO_IN;
    }
    cond |= G_IO_OUT;

    return (*func)(QIO_CHANNEL(rsource->rioc),
                   (cond & rsource->condition),
                   user_data);
}

static void
qio_channel_rdma_source_finalize(GSource *source)
{
    QIOChannelRDMASource *ssource = (QIOChannelRDMASource *)source;

    object_unref(OBJECT(ssource->rioc));
}

GSourceFuncs qio_channel_rdma_source_funcs = {
    qio_channel_rdma_source_prepare,
    qio_channel_rdma_source_check,
    qio_channel_rdma_source_dispatch,
    qio_channel_rdma_source_finalize
};

static GSource *qio_channel_rdma_create_watch(QIOChannel *ioc,
                                              GIOCondition condition)
M
Michael R. Hines 已提交
2787
{
2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(ioc);
    QIOChannelRDMASource *ssource;
    GSource *source;

    source = g_source_new(&qio_channel_rdma_source_funcs,
                          sizeof(QIOChannelRDMASource));
    ssource = (QIOChannelRDMASource *)source;

    ssource->rioc = rioc;
    object_ref(OBJECT(rioc));

    ssource->condition = condition;

    return source;
}


static int qio_channel_rdma_close(QIOChannel *ioc,
                                  Error **errp)
{
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(ioc);
2809
    trace_qemu_rdma_close();
2810
    if (rioc->rdma) {
2811 2812 2813
        if (!rioc->rdma->error_state) {
            rioc->rdma->error_state = qemu_file_get_error(rioc->file);
        }
2814 2815 2816
        qemu_rdma_cleanup(rioc->rdma);
        g_free(rioc->rdma);
        rioc->rdma = NULL;
M
Michael R. Hines 已提交
2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838
    }
    return 0;
}

/*
 * Parameters:
 *    @offset == 0 :
 *        This means that 'block_offset' is a full virtual address that does not
 *        belong to a RAMBlock of the virtual machine and instead
 *        represents a private malloc'd memory area that the caller wishes to
 *        transfer.
 *
 *    @offset != 0 :
 *        Offset is an offset to be added to block_offset and used
 *        to also lookup the corresponding RAMBlock.
 *
 *    @size > 0 :
 *        Initiate an transfer this size.
 *
 *    @size == 0 :
 *        A 'hint' or 'advice' that means that we wish to speculatively
 *        and asynchronously unregister this memory. In this case, there is no
2839
 *        guarantee that the unregister will actually happen, for example,
M
Michael R. Hines 已提交
2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856
 *        if the memory is being actively transmitted. Additionally, the memory
 *        may be re-registered at any future time if a write within the same
 *        chunk was requested again, even if you attempted to unregister it
 *        here.
 *
 *    @size < 0 : TODO, not yet supported
 *        Unregister the memory NOW. This means that the caller does not
 *        expect there to be any future RDMA transfers and we just want to clean
 *        things up. This is used in case the upper layer owns the memory and
 *        cannot wait for qemu_fclose() to occur.
 *
 *    @bytes_sent : User-specificed pointer to indicate how many bytes were
 *                  sent. Usually, this will not be more than a few bytes of
 *                  the protocol because most transfers are sent asynchronously.
 */
static size_t qemu_rdma_save_page(QEMUFile *f, void *opaque,
                                  ram_addr_t block_offset, ram_addr_t offset,
2857
                                  size_t size, uint64_t *bytes_sent)
M
Michael R. Hines 已提交
2858
{
2859 2860
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(opaque);
    RDMAContext *rdma = rioc->rdma;
M
Michael R. Hines 已提交
2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874
    int ret;

    CHECK_ERROR_STATE();

    qemu_fflush(f);

    if (size > 0) {
        /*
         * Add this page to the current 'chunk'. If the chunk
         * is full, or the page doen't belong to the current chunk,
         * an actual RDMA write will occur and a new chunk will be formed.
         */
        ret = qemu_rdma_write(f, rdma, block_offset, offset, size);
        if (ret < 0) {
2875
            error_report("rdma migration: write error! %d", ret);
M
Michael R. Hines 已提交
2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907
            goto err;
        }

        /*
         * We always return 1 bytes because the RDMA
         * protocol is completely asynchronous. We do not yet know
         * whether an  identified chunk is zero or not because we're
         * waiting for other pages to potentially be merged with
         * the current chunk. So, we have to call qemu_update_position()
         * later on when the actual write occurs.
         */
        if (bytes_sent) {
            *bytes_sent = 1;
        }
    } else {
        uint64_t index, chunk;

        /* TODO: Change QEMUFileOps prototype to be signed: size_t => long
        if (size < 0) {
            ret = qemu_rdma_drain_cq(f, rdma);
            if (ret < 0) {
                fprintf(stderr, "rdma: failed to synchronously drain"
                                " completion queue before unregistration.\n");
                goto err;
            }
        }
        */

        ret = qemu_rdma_search_ram_block(rdma, block_offset,
                                         offset, size, &index, &chunk);

        if (ret) {
2908
            error_report("ram block search failed");
M
Michael R. Hines 已提交
2909 2910 2911 2912 2913 2914
            goto err;
        }

        qemu_rdma_signal_unregister(rdma, index, chunk, 0);

        /*
2915
         * TODO: Synchronous, guaranteed unregistration (should not occur during
M
Michael R. Hines 已提交
2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932
         * fast-path). Otherwise, unregisters will process on the next call to
         * qemu_rdma_drain_cq()
        if (size < 0) {
            qemu_rdma_unregister_waiting(rdma);
        }
        */
    }

    /*
     * Drain the Completion Queue if possible, but do not block,
     * just poll.
     *
     * If nothing to poll, the end of the iteration will do this
     * again to make sure we don't overflow the request queue.
     */
    while (1) {
        uint64_t wr_id, wr_id_in;
2933
        int ret = qemu_rdma_poll(rdma, &wr_id_in, NULL);
M
Michael R. Hines 已提交
2934
        if (ret < 0) {
2935
            error_report("rdma migration: polling error! %d", ret);
M
Michael R. Hines 已提交
2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979
            goto err;
        }

        wr_id = wr_id_in & RDMA_WRID_TYPE_MASK;

        if (wr_id == RDMA_WRID_NONE) {
            break;
        }
    }

    return RAM_SAVE_CONTROL_DELAYED;
err:
    rdma->error_state = ret;
    return ret;
}

static int qemu_rdma_accept(RDMAContext *rdma)
{
    RDMACapabilities cap;
    struct rdma_conn_param conn_param = {
                                            .responder_resources = 2,
                                            .private_data = &cap,
                                            .private_data_len = sizeof(cap),
                                         };
    struct rdma_cm_event *cm_event;
    struct ibv_context *verbs;
    int ret = -EINVAL;
    int idx;

    ret = rdma_get_cm_event(rdma->channel, &cm_event);
    if (ret) {
        goto err_rdma_dest_wait;
    }

    if (cm_event->event != RDMA_CM_EVENT_CONNECT_REQUEST) {
        rdma_ack_cm_event(cm_event);
        goto err_rdma_dest_wait;
    }

    memcpy(&cap, cm_event->param.conn.private_data, sizeof(cap));

    network_to_caps(&cap);

    if (cap.version < 1 || cap.version > RDMA_CONTROL_VERSION_CURRENT) {
2980
            error_report("Unknown source RDMA version: %d, bailing...",
M
Michael R. Hines 已提交
2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003
                            cap.version);
            rdma_ack_cm_event(cm_event);
            goto err_rdma_dest_wait;
    }

    /*
     * Respond with only the capabilities this version of QEMU knows about.
     */
    cap.flags &= known_capabilities;

    /*
     * Enable the ones that we do know about.
     * Add other checks here as new ones are introduced.
     */
    if (cap.flags & RDMA_CAPABILITY_PIN_ALL) {
        rdma->pin_all = true;
    }

    rdma->cm_id = cm_event->id;
    verbs = cm_event->id->verbs;

    rdma_ack_cm_event(cm_event);

3004
    trace_qemu_rdma_accept_pin_state(rdma->pin_all);
M
Michael R. Hines 已提交
3005 3006 3007

    caps_to_network(&cap);

3008
    trace_qemu_rdma_accept_pin_verbsc(verbs);
M
Michael R. Hines 已提交
3009 3010 3011 3012

    if (!rdma->verbs) {
        rdma->verbs = verbs;
    } else if (rdma->verbs != verbs) {
3013 3014
            error_report("ibv context not matching %p, %p!", rdma->verbs,
                         verbs);
M
Michael R. Hines 已提交
3015 3016 3017 3018 3019 3020 3021
            goto err_rdma_dest_wait;
    }

    qemu_rdma_dump_id("dest_init", verbs);

    ret = qemu_rdma_alloc_pd_cq(rdma);
    if (ret) {
3022
        error_report("rdma migration: error allocating pd and cq!");
M
Michael R. Hines 已提交
3023 3024 3025 3026 3027
        goto err_rdma_dest_wait;
    }

    ret = qemu_rdma_alloc_qp(rdma);
    if (ret) {
3028
        error_report("rdma migration: error allocating qp!");
M
Michael R. Hines 已提交
3029 3030 3031 3032 3033
        goto err_rdma_dest_wait;
    }

    ret = qemu_rdma_init_ram_blocks(rdma);
    if (ret) {
3034
        error_report("rdma migration: error initializing ram blocks!");
M
Michael R. Hines 已提交
3035 3036 3037
        goto err_rdma_dest_wait;
    }

3038
    for (idx = 0; idx < RDMA_WRID_MAX; idx++) {
M
Michael R. Hines 已提交
3039 3040
        ret = qemu_rdma_reg_control(rdma, idx);
        if (ret) {
3041
            error_report("rdma: error registering %d control", idx);
M
Michael R. Hines 已提交
3042 3043 3044 3045
            goto err_rdma_dest_wait;
        }
    }

3046
    qemu_set_fd_handler(rdma->channel->fd, NULL, NULL, NULL);
M
Michael R. Hines 已提交
3047 3048 3049

    ret = rdma_accept(rdma->cm_id, &conn_param);
    if (ret) {
3050
        error_report("rdma_accept returns %d", ret);
M
Michael R. Hines 已提交
3051 3052 3053 3054 3055
        goto err_rdma_dest_wait;
    }

    ret = rdma_get_cm_event(rdma->channel, &cm_event);
    if (ret) {
3056
        error_report("rdma_accept get_cm_event failed %d", ret);
M
Michael R. Hines 已提交
3057 3058 3059 3060
        goto err_rdma_dest_wait;
    }

    if (cm_event->event != RDMA_CM_EVENT_ESTABLISHED) {
3061
        error_report("rdma_accept not event established");
M
Michael R. Hines 已提交
3062 3063 3064 3065 3066
        rdma_ack_cm_event(cm_event);
        goto err_rdma_dest_wait;
    }

    rdma_ack_cm_event(cm_event);
3067
    rdma->connected = true;
M
Michael R. Hines 已提交
3068

I
Isaku Yamahata 已提交
3069
    ret = qemu_rdma_post_recv_control(rdma, RDMA_WRID_READY);
M
Michael R. Hines 已提交
3070
    if (ret) {
3071
        error_report("rdma migration: error posting second control recv");
M
Michael R. Hines 已提交
3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084
        goto err_rdma_dest_wait;
    }

    qemu_rdma_dump_gid("dest_connect", rdma->cm_id);

    return 0;

err_rdma_dest_wait:
    rdma->error_state = ret;
    qemu_rdma_cleanup(rdma);
    return ret;
}

3085 3086 3087 3088 3089 3090 3091 3092
static int dest_ram_sort_func(const void *a, const void *b)
{
    unsigned int a_index = ((const RDMALocalBlock *)a)->src_index;
    unsigned int b_index = ((const RDMALocalBlock *)b)->src_index;

    return (a_index < b_index) ? -1 : (a_index != b_index);
}

M
Michael R. Hines 已提交
3093 3094 3095 3096 3097 3098 3099 3100 3101
/*
 * During each iteration of the migration, we listen for instructions
 * by the source VM to perform dynamic page registrations before they
 * can perform RDMA operations.
 *
 * We respond with the 'rkey'.
 *
 * Keep doing this until the source tells us to stop.
 */
3102
static int qemu_rdma_registration_handle(QEMUFile *f, void *opaque)
M
Michael R. Hines 已提交
3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113
{
    RDMAControlHeader reg_resp = { .len = sizeof(RDMARegisterResult),
                               .type = RDMA_CONTROL_REGISTER_RESULT,
                               .repeat = 0,
                             };
    RDMAControlHeader unreg_resp = { .len = 0,
                               .type = RDMA_CONTROL_UNREGISTER_FINISHED,
                               .repeat = 0,
                             };
    RDMAControlHeader blocks = { .type = RDMA_CONTROL_RAM_BLOCKS_RESULT,
                                 .repeat = 1 };
3114 3115
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(opaque);
    RDMAContext *rdma = rioc->rdma;
M
Michael R. Hines 已提交
3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131
    RDMALocalBlocks *local = &rdma->local_ram_blocks;
    RDMAControlHeader head;
    RDMARegister *reg, *registers;
    RDMACompress *comp;
    RDMARegisterResult *reg_result;
    static RDMARegisterResult results[RDMA_CONTROL_MAX_COMMANDS_PER_MESSAGE];
    RDMALocalBlock *block;
    void *host_addr;
    int ret = 0;
    int idx = 0;
    int count = 0;
    int i = 0;

    CHECK_ERROR_STATE();

    do {
3132
        trace_qemu_rdma_registration_handle_wait();
M
Michael R. Hines 已提交
3133 3134 3135 3136 3137 3138 3139 3140

        ret = qemu_rdma_exchange_recv(rdma, &head, RDMA_CONTROL_NONE);

        if (ret < 0) {
            break;
        }

        if (head.repeat > RDMA_CONTROL_MAX_COMMANDS_PER_MESSAGE) {
3141 3142
            error_report("rdma: Too many requests in this message (%d)."
                            "Bailing.", head.repeat);
M
Michael R. Hines 已提交
3143 3144 3145 3146 3147 3148 3149 3150 3151
            ret = -EIO;
            break;
        }

        switch (head.type) {
        case RDMA_CONTROL_COMPRESS:
            comp = (RDMACompress *) rdma->wr_data[idx].control_curr;
            network_to_compress(comp);

3152 3153 3154
            trace_qemu_rdma_registration_handle_compress(comp->length,
                                                         comp->block_idx,
                                                         comp->offset);
3155 3156 3157 3158 3159
            if (comp->block_idx >= rdma->local_ram_blocks.nb_blocks) {
                error_report("rdma: 'compress' bad block index %u (vs %d)",
                             (unsigned int)comp->block_idx,
                             rdma->local_ram_blocks.nb_blocks);
                ret = -EIO;
D
Dr. David Alan Gilbert 已提交
3160
                goto out;
3161
            }
M
Michael R. Hines 已提交
3162 3163 3164 3165 3166 3167 3168 3169 3170
            block = &(rdma->local_ram_blocks.block[comp->block_idx]);

            host_addr = block->local_host_addr +
                            (comp->offset - block->offset);

            ram_handle_compressed(host_addr, comp->value, comp->length);
            break;

        case RDMA_CONTROL_REGISTER_FINISHED:
3171
            trace_qemu_rdma_registration_handle_finished();
M
Michael R. Hines 已提交
3172 3173 3174
            goto out;

        case RDMA_CONTROL_RAM_BLOCKS_REQUEST:
3175
            trace_qemu_rdma_registration_handle_ram_blocks();
M
Michael R. Hines 已提交
3176

3177 3178 3179 3180 3181 3182 3183
            /* Sort our local RAM Block list so it's the same as the source,
             * we can do this since we've filled in a src_index in the list
             * as we received the RAMBlock list earlier.
             */
            qsort(rdma->local_ram_blocks.block,
                  rdma->local_ram_blocks.nb_blocks,
                  sizeof(RDMALocalBlock), dest_ram_sort_func);
M
Michael R. Hines 已提交
3184 3185 3186
            if (rdma->pin_all) {
                ret = qemu_rdma_reg_whole_ram_blocks(rdma);
                if (ret) {
3187 3188
                    error_report("rdma migration: error dest "
                                    "registering ram blocks");
M
Michael R. Hines 已提交
3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199
                    goto out;
                }
            }

            /*
             * Dest uses this to prepare to transmit the RAMBlock descriptions
             * to the source VM after connection setup.
             * Both sides use the "remote" structure to communicate and update
             * their "local" descriptions with what was sent.
             */
            for (i = 0; i < local->nb_blocks; i++) {
3200
                rdma->dest_blocks[i].remote_host_addr =
3201
                    (uintptr_t)(local->block[i].local_host_addr);
M
Michael R. Hines 已提交
3202 3203

                if (rdma->pin_all) {
3204
                    rdma->dest_blocks[i].remote_rkey = local->block[i].mr->rkey;
M
Michael R. Hines 已提交
3205 3206
                }

3207 3208
                rdma->dest_blocks[i].offset = local->block[i].offset;
                rdma->dest_blocks[i].length = local->block[i].length;
M
Michael R. Hines 已提交
3209

3210
                dest_block_to_network(&rdma->dest_blocks[i]);
3211 3212 3213 3214 3215 3216
                trace_qemu_rdma_registration_handle_ram_blocks_loop(
                    local->block[i].block_name,
                    local->block[i].offset,
                    local->block[i].length,
                    local->block[i].local_host_addr,
                    local->block[i].src_index);
M
Michael R. Hines 已提交
3217 3218 3219
            }

            blocks.len = rdma->local_ram_blocks.nb_blocks
3220
                                                * sizeof(RDMADestBlock);
M
Michael R. Hines 已提交
3221 3222 3223


            ret = qemu_rdma_post_send_control(rdma,
3224
                                        (uint8_t *) rdma->dest_blocks, &blocks);
M
Michael R. Hines 已提交
3225 3226

            if (ret < 0) {
3227
                error_report("rdma migration: error sending remote info");
M
Michael R. Hines 已提交
3228 3229 3230 3231 3232
                goto out;
            }

            break;
        case RDMA_CONTROL_REGISTER_REQUEST:
3233
            trace_qemu_rdma_registration_handle_register(head.repeat);
M
Michael R. Hines 已提交
3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246

            reg_resp.repeat = head.repeat;
            registers = (RDMARegister *) rdma->wr_data[idx].control_curr;

            for (count = 0; count < head.repeat; count++) {
                uint64_t chunk;
                uint8_t *chunk_start, *chunk_end;

                reg = &registers[count];
                network_to_register(reg);

                reg_result = &results[count];

3247
                trace_qemu_rdma_registration_handle_register_loop(count,
M
Michael R. Hines 已提交
3248 3249
                         reg->current_index, reg->key.current_addr, reg->chunks);

3250 3251 3252 3253 3254
                if (reg->current_index >= rdma->local_ram_blocks.nb_blocks) {
                    error_report("rdma: 'register' bad block index %u (vs %d)",
                                 (unsigned int)reg->current_index,
                                 rdma->local_ram_blocks.nb_blocks);
                    ret = -ENOENT;
D
Dr. David Alan Gilbert 已提交
3255
                    goto out;
3256
                }
M
Michael R. Hines 已提交
3257 3258
                block = &(rdma->local_ram_blocks.block[reg->current_index]);
                if (block->is_ram_block) {
3259 3260 3261 3262 3263 3264
                    if (block->offset > reg->key.current_addr) {
                        error_report("rdma: bad register address for block %s"
                            " offset: %" PRIx64 " current_addr: %" PRIx64,
                            block->block_name, block->offset,
                            reg->key.current_addr);
                        ret = -ERANGE;
D
Dr. David Alan Gilbert 已提交
3265
                        goto out;
3266
                    }
M
Michael R. Hines 已提交
3267 3268 3269 3270 3271 3272 3273 3274
                    host_addr = (block->local_host_addr +
                                (reg->key.current_addr - block->offset));
                    chunk = ram_chunk_index(block->local_host_addr,
                                            (uint8_t *) host_addr);
                } else {
                    chunk = reg->key.chunk;
                    host_addr = block->local_host_addr +
                        (reg->key.chunk * (1UL << RDMA_REG_CHUNK_SHIFT));
3275 3276 3277 3278 3279 3280
                    /* Check for particularly bad chunk value */
                    if (host_addr < (void *)block->local_host_addr) {
                        error_report("rdma: bad chunk for block %s"
                            " chunk: %" PRIx64,
                            block->block_name, reg->key.chunk);
                        ret = -ERANGE;
D
Dr. David Alan Gilbert 已提交
3281
                        goto out;
3282
                    }
M
Michael R. Hines 已提交
3283 3284 3285 3286
                }
                chunk_start = ram_chunk_start(block, chunk);
                chunk_end = ram_chunk_end(block, chunk + reg->chunks);
                if (qemu_rdma_register_and_get_keys(rdma, block,
3287
                            (uintptr_t)host_addr, NULL, &reg_result->rkey,
M
Michael R. Hines 已提交
3288
                            chunk, chunk_start, chunk_end)) {
3289
                    error_report("cannot get rkey");
M
Michael R. Hines 已提交
3290 3291 3292 3293
                    ret = -EINVAL;
                    goto out;
                }

3294
                reg_result->host_addr = (uintptr_t)block->local_host_addr;
M
Michael R. Hines 已提交
3295

3296 3297
                trace_qemu_rdma_registration_handle_register_rkey(
                                                           reg_result->rkey);
M
Michael R. Hines 已提交
3298 3299 3300 3301 3302 3303 3304 3305

                result_to_network(reg_result);
            }

            ret = qemu_rdma_post_send_control(rdma,
                            (uint8_t *) results, &reg_resp);

            if (ret < 0) {
3306
                error_report("Failed to send control buffer");
M
Michael R. Hines 已提交
3307 3308 3309 3310
                goto out;
            }
            break;
        case RDMA_CONTROL_UNREGISTER_REQUEST:
3311
            trace_qemu_rdma_registration_handle_unregister(head.repeat);
M
Michael R. Hines 已提交
3312 3313 3314 3315 3316 3317 3318
            unreg_resp.repeat = head.repeat;
            registers = (RDMARegister *) rdma->wr_data[idx].control_curr;

            for (count = 0; count < head.repeat; count++) {
                reg = &registers[count];
                network_to_register(reg);

3319 3320
                trace_qemu_rdma_registration_handle_unregister_loop(count,
                           reg->current_index, reg->key.chunk);
M
Michael R. Hines 已提交
3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334

                block = &(rdma->local_ram_blocks.block[reg->current_index]);

                ret = ibv_dereg_mr(block->pmr[reg->key.chunk]);
                block->pmr[reg->key.chunk] = NULL;

                if (ret != 0) {
                    perror("rdma unregistration chunk failed");
                    ret = -ret;
                    goto out;
                }

                rdma->total_registrations--;

3335 3336
                trace_qemu_rdma_registration_handle_unregister_success(
                                                       reg->key.chunk);
M
Michael R. Hines 已提交
3337 3338 3339 3340 3341
            }

            ret = qemu_rdma_post_send_control(rdma, NULL, &unreg_resp);

            if (ret < 0) {
3342
                error_report("Failed to send control buffer");
M
Michael R. Hines 已提交
3343 3344 3345 3346
                goto out;
            }
            break;
        case RDMA_CONTROL_REGISTER_RESULT:
3347
            error_report("Invalid RESULT message at dest.");
M
Michael R. Hines 已提交
3348 3349 3350
            ret = -EIO;
            goto out;
        default:
3351
            error_report("Unknown control message %s", control_desc[head.type]);
M
Michael R. Hines 已提交
3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362
            ret = -EIO;
            goto out;
        }
    } while (1);
out:
    if (ret < 0) {
        rdma->error_state = ret;
    }
    return ret;
}

3363 3364 3365 3366 3367 3368 3369
/* Destination:
 * Called via a ram_control_load_hook during the initial RAM load section which
 * lists the RAMBlocks by name.  This lets us know the order of the RAMBlocks
 * on the source.
 * We've already built our local RAMBlock list, but not yet sent the list to
 * the source.
 */
3370 3371
static int
rdma_block_notification_handle(QIOChannelRDMA *rioc, const char *name)
3372
{
3373
    RDMAContext *rdma = rioc->rdma;
3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396
    int curr;
    int found = -1;

    /* Find the matching RAMBlock in our local list */
    for (curr = 0; curr < rdma->local_ram_blocks.nb_blocks; curr++) {
        if (!strcmp(rdma->local_ram_blocks.block[curr].block_name, name)) {
            found = curr;
            break;
        }
    }

    if (found == -1) {
        error_report("RAMBlock '%s' not found on destination", name);
        return -ENOENT;
    }

    rdma->local_ram_blocks.block[curr].src_index = rdma->next_src_index;
    trace_rdma_block_notification_handle(name, rdma->next_src_index);
    rdma->next_src_index++;

    return 0;
}

3397 3398 3399 3400
static int rdma_load_hook(QEMUFile *f, void *opaque, uint64_t flags, void *data)
{
    switch (flags) {
    case RAM_CONTROL_BLOCK_REG:
3401
        return rdma_block_notification_handle(opaque, data);
3402 3403 3404 3405 3406 3407 3408 3409 3410 3411

    case RAM_CONTROL_HOOK:
        return qemu_rdma_registration_handle(f, opaque);

    default:
        /* Shouldn't be called with any other values */
        abort();
    }
}

M
Michael R. Hines 已提交
3412
static int qemu_rdma_registration_start(QEMUFile *f, void *opaque,
3413
                                        uint64_t flags, void *data)
M
Michael R. Hines 已提交
3414
{
3415 3416
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(opaque);
    RDMAContext *rdma = rioc->rdma;
M
Michael R. Hines 已提交
3417 3418 3419

    CHECK_ERROR_STATE();

3420
    trace_qemu_rdma_registration_start(flags);
M
Michael R. Hines 已提交
3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431
    qemu_put_be64(f, RAM_SAVE_FLAG_HOOK);
    qemu_fflush(f);

    return 0;
}

/*
 * Inform dest that dynamic registrations are done for now.
 * First, flush writes, if any.
 */
static int qemu_rdma_registration_stop(QEMUFile *f, void *opaque,
3432
                                       uint64_t flags, void *data)
M
Michael R. Hines 已提交
3433 3434
{
    Error *local_err = NULL, **errp = &local_err;
3435 3436
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(opaque);
    RDMAContext *rdma = rioc->rdma;
M
Michael R. Hines 已提交
3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451
    RDMAControlHeader head = { .len = 0, .repeat = 1 };
    int ret = 0;

    CHECK_ERROR_STATE();

    qemu_fflush(f);
    ret = qemu_rdma_drain_cq(f, rdma);

    if (ret < 0) {
        goto err;
    }

    if (flags == RAM_CONTROL_SETUP) {
        RDMAControlHeader resp = {.type = RDMA_CONTROL_RAM_BLOCKS_RESULT };
        RDMALocalBlocks *local = &rdma->local_ram_blocks;
3452
        int reg_result_idx, i, nb_dest_blocks;
M
Michael R. Hines 已提交
3453 3454

        head.type = RDMA_CONTROL_RAM_BLOCKS_REQUEST;
3455
        trace_qemu_rdma_registration_stop_ram();
M
Michael R. Hines 已提交
3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468

        /*
         * Make sure that we parallelize the pinning on both sides.
         * For very large guests, doing this serially takes a really
         * long time, so we have to 'interleave' the pinning locally
         * with the control messages by performing the pinning on this
         * side before we receive the control response from the other
         * side that the pinning has completed.
         */
        ret = qemu_rdma_exchange_send(rdma, &head, NULL, &resp,
                    &reg_result_idx, rdma->pin_all ?
                    qemu_rdma_reg_whole_ram_blocks : NULL);
        if (ret < 0) {
3469
            ERROR(errp, "receiving remote info!");
M
Michael R. Hines 已提交
3470 3471 3472
            return ret;
        }

3473
        nb_dest_blocks = resp.len / sizeof(RDMADestBlock);
M
Michael R. Hines 已提交
3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486

        /*
         * The protocol uses two different sets of rkeys (mutually exclusive):
         * 1. One key to represent the virtual address of the entire ram block.
         *    (dynamic chunk registration disabled - pin everything with one rkey.)
         * 2. One to represent individual chunks within a ram block.
         *    (dynamic chunk registration enabled - pin individual chunks.)
         *
         * Once the capability is successfully negotiated, the destination transmits
         * the keys to use (or sends them later) including the virtual addresses
         * and then propagates the remote ram block descriptions to his local copy.
         */

3487
        if (local->nb_blocks != nb_dest_blocks) {
3488
            ERROR(errp, "ram blocks mismatch (Number of blocks %d vs %d) "
M
Michael R. Hines 已提交
3489
                        "Your QEMU command line parameters are probably "
3490 3491
                        "not identical on both the source and destination.",
                        local->nb_blocks, nb_dest_blocks);
3492
            rdma->error_state = -EINVAL;
M
Michael R. Hines 已提交
3493 3494 3495
            return -EINVAL;
        }

3496
        qemu_rdma_move_header(rdma, reg_result_idx, &resp);
3497
        memcpy(rdma->dest_blocks,
3498
            rdma->wr_data[reg_result_idx].control_curr, resp.len);
3499 3500
        for (i = 0; i < nb_dest_blocks; i++) {
            network_to_dest_block(&rdma->dest_blocks[i]);
M
Michael R. Hines 已提交
3501

3502 3503 3504 3505 3506 3507
            /* We require that the blocks are in the same order */
            if (rdma->dest_blocks[i].length != local->block[i].length) {
                ERROR(errp, "Block %s/%d has a different length %" PRIu64
                            "vs %" PRIu64, local->block[i].block_name, i,
                            local->block[i].length,
                            rdma->dest_blocks[i].length);
3508
                rdma->error_state = -EINVAL;
M
Michael R. Hines 已提交
3509 3510
                return -EINVAL;
            }
3511 3512 3513
            local->block[i].remote_host_addr =
                    rdma->dest_blocks[i].remote_host_addr;
            local->block[i].remote_rkey = rdma->dest_blocks[i].remote_rkey;
M
Michael R. Hines 已提交
3514 3515 3516
        }
    }

3517
    trace_qemu_rdma_registration_stop(flags);
M
Michael R. Hines 已提交
3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531

    head.type = RDMA_CONTROL_REGISTER_FINISHED;
    ret = qemu_rdma_exchange_send(rdma, &head, NULL, NULL, NULL, NULL);

    if (ret < 0) {
        goto err;
    }

    return 0;
err:
    rdma->error_state = ret;
    return ret;
}

3532
static const QEMUFileHooks rdma_read_hooks = {
3533
    .hook_ram_load = rdma_load_hook,
M
Michael R. Hines 已提交
3534 3535
};

3536
static const QEMUFileHooks rdma_write_hooks = {
M
Michael R. Hines 已提交
3537 3538 3539 3540 3541
    .before_ram_iterate = qemu_rdma_registration_start,
    .after_ram_iterate  = qemu_rdma_registration_stop,
    .save_page          = qemu_rdma_save_page,
};

3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580

static void qio_channel_rdma_finalize(Object *obj)
{
    QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(obj);
    if (rioc->rdma) {
        qemu_rdma_cleanup(rioc->rdma);
        g_free(rioc->rdma);
        rioc->rdma = NULL;
    }
}

static void qio_channel_rdma_class_init(ObjectClass *klass,
                                        void *class_data G_GNUC_UNUSED)
{
    QIOChannelClass *ioc_klass = QIO_CHANNEL_CLASS(klass);

    ioc_klass->io_writev = qio_channel_rdma_writev;
    ioc_klass->io_readv = qio_channel_rdma_readv;
    ioc_klass->io_set_blocking = qio_channel_rdma_set_blocking;
    ioc_klass->io_close = qio_channel_rdma_close;
    ioc_klass->io_create_watch = qio_channel_rdma_create_watch;
}

static const TypeInfo qio_channel_rdma_info = {
    .parent = TYPE_QIO_CHANNEL,
    .name = TYPE_QIO_CHANNEL_RDMA,
    .instance_size = sizeof(QIOChannelRDMA),
    .instance_finalize = qio_channel_rdma_finalize,
    .class_init = qio_channel_rdma_class_init,
};

static void qio_channel_rdma_register_types(void)
{
    type_register_static(&qio_channel_rdma_info);
}

type_init(qio_channel_rdma_register_types);

static QEMUFile *qemu_fopen_rdma(RDMAContext *rdma, const char *mode)
M
Michael R. Hines 已提交
3581
{
3582
    QIOChannelRDMA *rioc;
M
Michael R. Hines 已提交
3583 3584 3585 3586 3587

    if (qemu_file_mode_is_not_valid(mode)) {
        return NULL;
    }

3588 3589
    rioc = QIO_CHANNEL_RDMA(object_new(TYPE_QIO_CHANNEL_RDMA));
    rioc->rdma = rdma;
M
Michael R. Hines 已提交
3590 3591

    if (mode[0] == 'w') {
3592 3593
        rioc->file = qemu_fopen_channel_output(QIO_CHANNEL(rioc));
        qemu_file_set_hooks(rioc->file, &rdma_write_hooks);
M
Michael R. Hines 已提交
3594
    } else {
3595 3596
        rioc->file = qemu_fopen_channel_input(QIO_CHANNEL(rioc));
        qemu_file_set_hooks(rioc->file, &rdma_read_hooks);
M
Michael R. Hines 已提交
3597 3598
    }

3599
    return rioc->file;
M
Michael R. Hines 已提交
3600 3601 3602 3603 3604 3605 3606 3607 3608
}

static void rdma_accept_incoming_migration(void *opaque)
{
    RDMAContext *rdma = opaque;
    int ret;
    QEMUFile *f;
    Error *local_err = NULL, **errp = &local_err;

D
Dr. David Alan Gilbert 已提交
3609
    trace_qemu_rdma_accept_incoming_migration();
M
Michael R. Hines 已提交
3610 3611 3612
    ret = qemu_rdma_accept(rdma);

    if (ret) {
3613
        ERROR(errp, "RDMA Migration initialization failed!");
M
Michael R. Hines 已提交
3614 3615 3616
        return;
    }

D
Dr. David Alan Gilbert 已提交
3617
    trace_qemu_rdma_accept_incoming_migration_accepted();
M
Michael R. Hines 已提交
3618 3619 3620

    f = qemu_fopen_rdma(rdma, "rb");
    if (f == NULL) {
3621
        ERROR(errp, "could not qemu_fopen_rdma!");
M
Michael R. Hines 已提交
3622 3623 3624 3625 3626
        qemu_rdma_cleanup(rdma);
        return;
    }

    rdma->migration_started_on_destination = 1;
3627
    migration_fd_process_incoming(f);
M
Michael R. Hines 已提交
3628 3629 3630 3631 3632 3633 3634 3635
}

void rdma_start_incoming_migration(const char *host_port, Error **errp)
{
    int ret;
    RDMAContext *rdma;
    Error *local_err = NULL;

3636
    trace_rdma_start_incoming_migration();
M
Michael R. Hines 已提交
3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648
    rdma = qemu_rdma_data_init(host_port, &local_err);

    if (rdma == NULL) {
        goto err;
    }

    ret = qemu_rdma_dest_init(rdma, &local_err);

    if (ret) {
        goto err;
    }

3649
    trace_rdma_start_incoming_migration_after_dest_init();
M
Michael R. Hines 已提交
3650 3651 3652 3653

    ret = rdma_listen(rdma->listen_id, 5);

    if (ret) {
3654
        ERROR(errp, "listening on socket!");
M
Michael R. Hines 已提交
3655 3656 3657
        goto err;
    }

3658
    trace_rdma_start_incoming_migration_after_rdma_listen();
M
Michael R. Hines 已提交
3659

3660 3661
    qemu_set_fd_handler(rdma->channel->fd, rdma_accept_incoming_migration,
                        NULL, (void *)(intptr_t)rdma);
M
Michael R. Hines 已提交
3662 3663 3664 3665 3666 3667 3668 3669 3670 3671
    return;
err:
    error_propagate(errp, local_err);
    g_free(rdma);
}

void rdma_start_outgoing_migration(void *opaque,
                            const char *host_port, Error **errp)
{
    MigrationState *s = opaque;
3672
    RDMAContext *rdma = qemu_rdma_data_init(host_port, errp);
M
Michael R. Hines 已提交
3673 3674 3675 3676 3677 3678
    int ret = 0;

    if (rdma == NULL) {
        goto err;
    }

3679 3680
    ret = qemu_rdma_source_init(rdma,
        s->enabled_capabilities[MIGRATION_CAPABILITY_RDMA_PIN_ALL], errp);
M
Michael R. Hines 已提交
3681 3682 3683 3684 3685

    if (ret) {
        goto err;
    }

3686
    trace_rdma_start_outgoing_migration_after_rdma_source_init();
3687
    ret = qemu_rdma_connect(rdma, errp);
M
Michael R. Hines 已提交
3688 3689 3690 3691 3692

    if (ret) {
        goto err;
    }

3693
    trace_rdma_start_outgoing_migration_after_rdma_connect();
M
Michael R. Hines 已提交
3694

3695
    s->to_dst_file = qemu_fopen_rdma(rdma, "wb");
M
Michael R. Hines 已提交
3696 3697 3698 3699 3700
    migrate_fd_connect(s);
    return;
err:
    g_free(rdma);
}