From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:58444) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1fEyIJ-0008T7-JC for qemu-devel@nongnu.org; Sat, 05 May 2018 10:36:17 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1fEyIH-0008Ew-3k for qemu-devel@nongnu.org; Sat, 05 May 2018 10:36:15 -0400 Received: from mail-pf0-x244.google.com ([2607:f8b0:400e:c00::244]:39642) by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_128_CBC_SHA1:16) (Exim 4.71) (envelope-from ) id 1fEyIG-0008EP-Q3 for qemu-devel@nongnu.org; Sat, 05 May 2018 10:36:13 -0400 Received: by mail-pf0-x244.google.com with SMTP id a22so5483179pfn.6 for ; Sat, 05 May 2018 07:36:12 -0700 (PDT) From: Lidong Chen Date: Sat, 5 May 2018 22:35:35 +0800 Message-Id: <1525530936-21835-6-git-send-email-lidongchen@tencent.com> In-Reply-To: <1525530936-21835-1-git-send-email-lidongchen@tencent.com> References: <1525530936-21835-1-git-send-email-lidongchen@tencent.com> Subject: [Qemu-devel] [PATCH v3 5/6] migration: implement bi-directional RDMA QIOChannel List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: quintela@redhat.com, dgilbert@redhat.com, berrange@redhat.com Cc: qemu-devel@nongnu.org, galsha@mellanox.com, aviadye@mellanox.com, adido@mellanox.com, Lidong Chen This patch implements bi-directional RDMA QIOChannel. Because different threads may access RDMAQIOChannel currently, this patch use RCU to protect it. Signed-off-by: Lidong Chen --- migration/colo.c | 2 + migration/migration.c | 2 + migration/postcopy-ram.c | 2 + migration/ram.c | 4 + migration/rdma.c | 196 ++++++++++++++++++++++++++++++++++++++++------- migration/savevm.c | 3 + 6 files changed, 183 insertions(+), 26 deletions(-) diff --git a/migration/colo.c b/migration/colo.c index 4381067..88936f5 100644 --- a/migration/colo.c +++ b/migration/colo.c @@ -534,6 +534,7 @@ void *colo_process_incoming_thread(void *opaque) uint64_t value; Error *local_err = NULL; + rcu_register_thread(); qemu_sem_init(&mis->colo_incoming_sem, 0); migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, @@ -666,5 +667,6 @@ out: } migration_incoming_exit_colo(); + rcu_unregister_thread(); return NULL; } diff --git a/migration/migration.c b/migration/migration.c index 0bdb28e..584666b 100644 --- a/migration/migration.c +++ b/migration/migration.c @@ -1787,6 +1787,7 @@ static void *source_return_path_thread(void *opaque) int res; trace_source_return_path_thread_entry(); + rcu_register_thread(); while (!ms->rp_state.error && !qemu_file_get_error(rp) && migration_is_setup_or_active(ms->state)) { trace_source_return_path_thread_loop_top(); @@ -1887,6 +1888,7 @@ static void *source_return_path_thread(void *opaque) out: ms->rp_state.from_dst_file = NULL; qemu_fclose(rp); + rcu_unregister_thread(); return NULL; } diff --git a/migration/postcopy-ram.c b/migration/postcopy-ram.c index 8ceeaa2..4e05966 100644 --- a/migration/postcopy-ram.c +++ b/migration/postcopy-ram.c @@ -842,6 +842,7 @@ static void *postcopy_ram_fault_thread(void *opaque) RAMBlock *rb = NULL; trace_postcopy_ram_fault_thread_entry(); + rcu_register_thread(); mis->last_rb = NULL; /* last RAMBlock we sent part of */ qemu_sem_post(&mis->fault_thread_sem); @@ -1013,6 +1014,7 @@ static void *postcopy_ram_fault_thread(void *opaque) } } } + rcu_unregister_thread(); trace_postcopy_ram_fault_thread_exit(); g_free(pfd); return NULL; diff --git a/migration/ram.c b/migration/ram.c index 912810c..9bc92fc 100644 --- a/migration/ram.c +++ b/migration/ram.c @@ -491,6 +491,7 @@ static void *multifd_send_thread(void *opaque) { MultiFDSendParams *p = opaque; + rcu_register_thread(); while (true) { qemu_mutex_lock(&p->mutex); if (p->quit) { @@ -500,6 +501,7 @@ static void *multifd_send_thread(void *opaque) qemu_mutex_unlock(&p->mutex); qemu_sem_wait(&p->sem); } + rcu_unregister_thread(); return NULL; } @@ -592,6 +594,7 @@ static void *multifd_recv_thread(void *opaque) { MultiFDRecvParams *p = opaque; + rcu_register_thread(); while (true) { qemu_mutex_lock(&p->mutex); if (p->quit) { @@ -601,6 +604,7 @@ static void *multifd_recv_thread(void *opaque) qemu_mutex_unlock(&p->mutex); qemu_sem_wait(&p->sem); } + rcu_unregister_thread(); return NULL; } diff --git a/migration/rdma.c b/migration/rdma.c index f5c1d02..854f355 100644 --- a/migration/rdma.c +++ b/migration/rdma.c @@ -86,6 +86,7 @@ static uint32_t known_capabilities = RDMA_CAPABILITY_PIN_ALL; " to abort!"); \ rdma->error_reported = 1; \ } \ + rcu_read_unlock(); \ return rdma->error_state; \ } \ } while (0) @@ -402,7 +403,8 @@ typedef struct QIOChannelRDMA QIOChannelRDMA; struct QIOChannelRDMA { QIOChannel parent; - RDMAContext *rdma; + RDMAContext *rdmain; + RDMAContext *rdmaout; QEMUFile *file; bool blocking; /* XXX we don't actually honour this yet */ }; @@ -2635,12 +2637,20 @@ static ssize_t qio_channel_rdma_writev(QIOChannel *ioc, { QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(ioc); QEMUFile *f = rioc->file; - RDMAContext *rdma = rioc->rdma; + RDMAContext *rdma; int ret; ssize_t done = 0; size_t i; size_t len = 0; + rcu_read_lock(); + rdma = atomic_rcu_read(&rioc->rdmaout); + + if (!rdma) { + rcu_read_unlock(); + return -EIO; + } + CHECK_ERROR_STATE(); /* @@ -2650,6 +2660,7 @@ static ssize_t qio_channel_rdma_writev(QIOChannel *ioc, ret = qemu_rdma_write_flush(f, rdma); if (ret < 0) { rdma->error_state = ret; + rcu_read_unlock(); return ret; } @@ -2669,6 +2680,7 @@ static ssize_t qio_channel_rdma_writev(QIOChannel *ioc, if (ret < 0) { rdma->error_state = ret; + rcu_read_unlock(); return ret; } @@ -2677,6 +2689,7 @@ static ssize_t qio_channel_rdma_writev(QIOChannel *ioc, } } + rcu_read_unlock(); return done; } @@ -2710,12 +2723,20 @@ static ssize_t qio_channel_rdma_readv(QIOChannel *ioc, Error **errp) { QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(ioc); - RDMAContext *rdma = rioc->rdma; + RDMAContext *rdma; RDMAControlHeader head; int ret = 0; ssize_t i; size_t done = 0; + rcu_read_lock(); + rdma = atomic_rcu_read(&rioc->rdmain); + + if (!rdma) { + rcu_read_unlock(); + return -EIO; + } + CHECK_ERROR_STATE(); for (i = 0; i < niov; i++) { @@ -2727,7 +2748,7 @@ static ssize_t qio_channel_rdma_readv(QIOChannel *ioc, * were given and dish out the bytes until we run * out of bytes. */ - ret = qemu_rdma_fill(rioc->rdma, data, want, 0); + ret = qemu_rdma_fill(rdma, data, want, 0); done += ret; want -= ret; /* Got what we needed, so go to next iovec */ @@ -2749,25 +2770,28 @@ static ssize_t qio_channel_rdma_readv(QIOChannel *ioc, if (ret < 0) { rdma->error_state = ret; + rcu_read_unlock(); return ret; } /* * SEND was received with new bytes, now try again. */ - ret = qemu_rdma_fill(rioc->rdma, data, want, 0); + ret = qemu_rdma_fill(rdma, data, want, 0); done += ret; want -= ret; /* Still didn't get enough, so lets just return */ if (want) { if (done == 0) { + rcu_read_unlock(); return QIO_CHANNEL_ERR_BLOCK; } else { break; } } } + rcu_read_unlock(); return done; } @@ -2819,15 +2843,29 @@ qio_channel_rdma_source_prepare(GSource *source, gint *timeout) { QIOChannelRDMASource *rsource = (QIOChannelRDMASource *)source; - RDMAContext *rdma = rsource->rioc->rdma; + RDMAContext *rdma; GIOCondition cond = 0; *timeout = -1; + rcu_read_lock(); + if (rsource->condition == G_IO_IN) { + rdma = atomic_rcu_read(&rsource->rioc->rdmain); + } else { + rdma = atomic_rcu_read(&rsource->rioc->rdmaout); + } + + if (!rdma) { + error_report("RDMAContext is NULL when prepare Gsource"); + rcu_read_unlock(); + return FALSE; + } + if (rdma->wr_data[0].control_len) { cond |= G_IO_IN; } cond |= G_IO_OUT; + rcu_read_unlock(); return cond & rsource->condition; } @@ -2835,14 +2873,28 @@ static gboolean qio_channel_rdma_source_check(GSource *source) { QIOChannelRDMASource *rsource = (QIOChannelRDMASource *)source; - RDMAContext *rdma = rsource->rioc->rdma; + RDMAContext *rdma; GIOCondition cond = 0; + rcu_read_lock(); + if (rsource->condition == G_IO_IN) { + rdma = atomic_rcu_read(&rsource->rioc->rdmain); + } else { + rdma = atomic_rcu_read(&rsource->rioc->rdmaout); + } + + if (!rdma) { + error_report("RDMAContext is NULL when check Gsource"); + rcu_read_unlock(); + return FALSE; + } + if (rdma->wr_data[0].control_len) { cond |= G_IO_IN; } cond |= G_IO_OUT; + rcu_read_unlock(); return cond & rsource->condition; } @@ -2853,14 +2905,28 @@ qio_channel_rdma_source_dispatch(GSource *source, { QIOChannelFunc func = (QIOChannelFunc)callback; QIOChannelRDMASource *rsource = (QIOChannelRDMASource *)source; - RDMAContext *rdma = rsource->rioc->rdma; + RDMAContext *rdma; GIOCondition cond = 0; + rcu_read_lock(); + if (rsource->condition == G_IO_IN) { + rdma = atomic_rcu_read(&rsource->rioc->rdmain); + } else { + rdma = atomic_rcu_read(&rsource->rioc->rdmaout); + } + + if (!rdma) { + error_report("RDMAContext is NULL when dispatch Gsource"); + rcu_read_unlock(); + return FALSE; + } + if (rdma->wr_data[0].control_len) { cond |= G_IO_IN; } cond |= G_IO_OUT; + rcu_read_unlock(); return (*func)(QIO_CHANNEL(rsource->rioc), (cond & rsource->condition), user_data); @@ -2905,15 +2971,32 @@ static int qio_channel_rdma_close(QIOChannel *ioc, Error **errp) { QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(ioc); + RDMAContext *rdmain, *rdmaout; trace_qemu_rdma_close(); - if (rioc->rdma) { - if (!rioc->rdma->error_state) { - rioc->rdma->error_state = qemu_file_get_error(rioc->file); - } - qemu_rdma_cleanup(rioc->rdma); - g_free(rioc->rdma); - rioc->rdma = NULL; + + rdmain = rioc->rdmain; + if (rdmain) { + atomic_rcu_set(&rioc->rdmain, NULL); + } + + rdmaout = rioc->rdmaout; + if (rdmaout) { + atomic_rcu_set(&rioc->rdmaout, NULL); } + + synchronize_rcu(); + + if (rdmain) { + qemu_rdma_cleanup(rdmain); + } + + if (rdmaout) { + qemu_rdma_cleanup(rdmaout); + } + + g_free(rdmain); + g_free(rdmaout); + return 0; } @@ -2956,12 +3039,21 @@ static size_t qemu_rdma_save_page(QEMUFile *f, void *opaque, size_t size, uint64_t *bytes_sent) { QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(opaque); - RDMAContext *rdma = rioc->rdma; + RDMAContext *rdma; int ret; + rcu_read_lock(); + rdma = atomic_rcu_read(&rioc->rdmaout); + + if (!rdma) { + rcu_read_unlock(); + return -EIO; + } + CHECK_ERROR_STATE(); if (migrate_get_current()->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { + rcu_read_unlock(); return RAM_SAVE_CONTROL_NOT_SUPP; } @@ -3046,9 +3138,11 @@ static size_t qemu_rdma_save_page(QEMUFile *f, void *opaque, } } + rcu_read_unlock(); return RAM_SAVE_CONTROL_DELAYED; err: rdma->error_state = ret; + rcu_read_unlock(); return ret; } @@ -3224,8 +3318,8 @@ static int qemu_rdma_registration_handle(QEMUFile *f, void *opaque) RDMAControlHeader blocks = { .type = RDMA_CONTROL_RAM_BLOCKS_RESULT, .repeat = 1 }; QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(opaque); - RDMAContext *rdma = rioc->rdma; - RDMALocalBlocks *local = &rdma->local_ram_blocks; + RDMAContext *rdma; + RDMALocalBlocks *local; RDMAControlHeader head; RDMARegister *reg, *registers; RDMACompress *comp; @@ -3238,8 +3332,17 @@ static int qemu_rdma_registration_handle(QEMUFile *f, void *opaque) int count = 0; int i = 0; + rcu_read_lock(); + rdma = atomic_rcu_read(&rioc->rdmain); + + if (!rdma) { + rcu_read_unlock(); + return -EIO; + } + CHECK_ERROR_STATE(); + local = &rdma->local_ram_blocks; do { trace_qemu_rdma_registration_handle_wait(); @@ -3469,6 +3572,7 @@ out: if (ret < 0) { rdma->error_state = ret; } + rcu_read_unlock(); return ret; } @@ -3482,10 +3586,18 @@ out: static int rdma_block_notification_handle(QIOChannelRDMA *rioc, const char *name) { - RDMAContext *rdma = rioc->rdma; + RDMAContext *rdma; int curr; int found = -1; + rcu_read_lock(); + rdma = atomic_rcu_read(&rioc->rdmain); + + if (!rdma) { + rcu_read_unlock(); + return -EIO; + } + /* 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)) { @@ -3496,6 +3608,7 @@ rdma_block_notification_handle(QIOChannelRDMA *rioc, const char *name) if (found == -1) { error_report("RAMBlock '%s' not found on destination", name); + rcu_read_unlock(); return -ENOENT; } @@ -3503,6 +3616,7 @@ rdma_block_notification_handle(QIOChannelRDMA *rioc, const char *name) trace_rdma_block_notification_handle(name, rdma->next_src_index); rdma->next_src_index++; + rcu_read_unlock(); return 0; } @@ -3525,11 +3639,19 @@ static int qemu_rdma_registration_start(QEMUFile *f, void *opaque, uint64_t flags, void *data) { QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(opaque); - RDMAContext *rdma = rioc->rdma; + RDMAContext *rdma; + + rcu_read_lock(); + rdma = atomic_rcu_read(&rioc->rdmaout); + if (!rdma) { + rcu_read_unlock(); + return -EIO; + } CHECK_ERROR_STATE(); if (migrate_get_current()->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { + rcu_read_unlock(); return 0; } @@ -3537,6 +3659,7 @@ static int qemu_rdma_registration_start(QEMUFile *f, void *opaque, qemu_put_be64(f, RAM_SAVE_FLAG_HOOK); qemu_fflush(f); + rcu_read_unlock(); return 0; } @@ -3549,13 +3672,21 @@ static int qemu_rdma_registration_stop(QEMUFile *f, void *opaque, { Error *local_err = NULL, **errp = &local_err; QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(opaque); - RDMAContext *rdma = rioc->rdma; + RDMAContext *rdma; RDMAControlHeader head = { .len = 0, .repeat = 1 }; int ret = 0; + rcu_read_lock(); + rdma = atomic_rcu_read(&rioc->rdmaout); + if (!rdma) { + rcu_read_unlock(); + return -EIO; + } + CHECK_ERROR_STATE(); if (migrate_get_current()->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { + rcu_read_unlock(); return 0; } @@ -3587,6 +3718,7 @@ static int qemu_rdma_registration_stop(QEMUFile *f, void *opaque, qemu_rdma_reg_whole_ram_blocks : NULL); if (ret < 0) { ERROR(errp, "receiving remote info!"); + rcu_read_unlock(); return ret; } @@ -3610,6 +3742,7 @@ static int qemu_rdma_registration_stop(QEMUFile *f, void *opaque, "not identical on both the source and destination.", local->nb_blocks, nb_dest_blocks); rdma->error_state = -EINVAL; + rcu_read_unlock(); return -EINVAL; } @@ -3626,6 +3759,7 @@ static int qemu_rdma_registration_stop(QEMUFile *f, void *opaque, local->block[i].length, rdma->dest_blocks[i].length); rdma->error_state = -EINVAL; + rcu_read_unlock(); return -EINVAL; } local->block[i].remote_host_addr = @@ -3643,9 +3777,11 @@ static int qemu_rdma_registration_stop(QEMUFile *f, void *opaque, goto err; } + rcu_read_unlock(); return 0; err: rdma->error_state = ret; + rcu_read_unlock(); return ret; } @@ -3663,10 +3799,15 @@ static const QEMUFileHooks rdma_write_hooks = { 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; + if (rioc->rdmain) { + qemu_rdma_cleanup(rioc->rdmain); + g_free(rioc->rdmain); + rioc->rdmain = NULL; + } + if (rioc->rdmaout) { + qemu_rdma_cleanup(rioc->rdmaout); + g_free(rioc->rdmaout); + rioc->rdmaout = NULL; } } @@ -3706,13 +3847,16 @@ static QEMUFile *qemu_fopen_rdma(RDMAContext *rdma, const char *mode) } rioc = QIO_CHANNEL_RDMA(object_new(TYPE_QIO_CHANNEL_RDMA)); - rioc->rdma = rdma; if (mode[0] == 'w') { rioc->file = qemu_fopen_channel_output(QIO_CHANNEL(rioc)); + rioc->rdmaout = rdma; + rioc->rdmain = rdma->return_path; qemu_file_set_hooks(rioc->file, &rdma_write_hooks); } else { rioc->file = qemu_fopen_channel_input(QIO_CHANNEL(rioc)); + rioc->rdmain = rdma; + rioc->rdmaout = rdma->return_path; qemu_file_set_hooks(rioc->file, &rdma_read_hooks); } diff --git a/migration/savevm.c b/migration/savevm.c index e2be02a..45ec809 100644 --- a/migration/savevm.c +++ b/migration/savevm.c @@ -1573,6 +1573,7 @@ static void *postcopy_ram_listen_thread(void *opaque) qemu_sem_post(&mis->listen_thread_sem); trace_postcopy_ram_listen_thread_start(); + rcu_register_thread(); /* * Because we're a thread and not a coroutine we can't yield * in qemu_file, and thus we must be blocking now. @@ -1605,6 +1606,7 @@ static void *postcopy_ram_listen_thread(void *opaque) * to leave the guest running and fire MCEs for pages that never * arrived as a desperate recovery step. */ + rcu_unregister_thread(); exit(EXIT_FAILURE); } @@ -1619,6 +1621,7 @@ static void *postcopy_ram_listen_thread(void *opaque) migration_incoming_state_destroy(); qemu_loadvm_state_cleanup(); + rcu_unregister_thread(); return NULL; } -- 1.8.3.1