From 0393031a16735835a441b6d6e0495a1bd14adb90 Mon Sep 17 00:00:00 2001 From: zhanghailiang Date: Mon, 24 Feb 2020 14:54:10 +0800 Subject: [PATCH] COLO: Optimize memory back-up process This patch will reduce the downtime of VM for the initial process, Previously, we copied all these memory in preparing stage of COLO while we need to stop VM, which is a time-consuming process. Here we optimize it by a trick, back-up every page while in migration process while COLO is enabled, though it affects the speed of the migration, but it obviously reduce the downtime of back-up all SVM'S memory in COLO preparing stage. Signed-off-by: zhanghailiang Message-Id: <20200224065414.36524-5-zhang.zhanghailiang@huawei.com> Signed-off-by: Dr. David Alan Gilbert minor typo fixes --- migration/colo.c | 3 +++ migration/ram.c | 66 +++++++++++++++++++++++++++++++++++------------- migration/ram.h | 1 + 3 files changed, 53 insertions(+), 17 deletions(-) diff --git a/migration/colo.c b/migration/colo.c index 93c5a452fb..44942c4e23 100644 --- a/migration/colo.c +++ b/migration/colo.c @@ -26,6 +26,7 @@ #include "qemu/main-loop.h" #include "qemu/rcu.h" #include "migration/failover.h" +#include "migration/ram.h" #ifdef CONFIG_REPLICATION #include "replication.h" #endif @@ -845,6 +846,8 @@ void *colo_process_incoming_thread(void *opaque) */ qemu_file_set_blocking(mis->from_src_file, true); + colo_incoming_start_dirty_log(); + bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE); fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc)); object_unref(OBJECT(bioc)); diff --git a/migration/ram.c b/migration/ram.c index 02cfd76d19..31f4a9dc09 100644 --- a/migration/ram.c +++ b/migration/ram.c @@ -2994,7 +2994,6 @@ int colo_init_ram_cache(void) } return -errno; } - memcpy(block->colo_cache, block->host, block->used_length); } } @@ -3008,19 +3007,36 @@ int colo_init_ram_cache(void) RAMBLOCK_FOREACH_NOT_IGNORED(block) { unsigned long pages = block->max_length >> TARGET_PAGE_BITS; - block->bmap = bitmap_new(pages); - bitmap_set(block->bmap, 0, pages); } } - ram_state = g_new0(RAMState, 1); - ram_state->migration_dirty_pages = 0; - qemu_mutex_init(&ram_state->bitmap_mutex); - memory_global_dirty_log_start(); + ram_state_init(&ram_state); return 0; } +/* TODO: duplicated with ram_init_bitmaps */ +void colo_incoming_start_dirty_log(void) +{ + RAMBlock *block = NULL; + /* For memory_global_dirty_log_start below. */ + qemu_mutex_lock_iothread(); + qemu_mutex_lock_ramlist(); + + memory_global_dirty_log_sync(); + WITH_RCU_READ_LOCK_GUARD() { + RAMBLOCK_FOREACH_NOT_IGNORED(block) { + ramblock_sync_dirty_bitmap(ram_state, block); + /* Discard this dirty bitmap record */ + bitmap_zero(block->bmap, block->max_length >> TARGET_PAGE_BITS); + } + memory_global_dirty_log_start(); + } + ram_state->migration_dirty_pages = 0; + qemu_mutex_unlock_ramlist(); + qemu_mutex_unlock_iothread(); +} + /* It is need to hold the global lock to call this helper */ void colo_release_ram_cache(void) { @@ -3040,9 +3056,7 @@ void colo_release_ram_cache(void) } } } - qemu_mutex_destroy(&ram_state->bitmap_mutex); - g_free(ram_state); - ram_state = NULL; + ram_state_cleanup(&ram_state); } /** @@ -3356,7 +3370,7 @@ static int ram_load_precopy(QEMUFile *f) while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { ram_addr_t addr, total_ram_bytes; - void *host = NULL; + void *host = NULL, *host_bak = NULL; uint8_t ch; /* @@ -3387,20 +3401,35 @@ static int ram_load_precopy(QEMUFile *f) RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) { RAMBlock *block = ram_block_from_stream(f, flags); + host = host_from_ram_block_offset(block, addr); /* - * After going into COLO, we should load the Page into colo_cache. + * After going into COLO stage, we should not load the page + * into SVM's memory directly, we put them into colo_cache firstly. + * NOTE: We need to keep a copy of SVM's ram in colo_cache. + * Previously, we copied all these memory in preparing stage of COLO + * while we need to stop VM, which is a time-consuming process. + * Here we optimize it by a trick, back-up every page while in + * migration process while COLO is enabled, though it affects the + * speed of the migration, but it obviously reduce the downtime of + * back-up all SVM'S memory in COLO preparing stage. */ - if (migration_incoming_in_colo_state()) { - host = colo_cache_from_block_offset(block, addr); - } else { - host = host_from_ram_block_offset(block, addr); + if (migration_incoming_colo_enabled()) { + if (migration_incoming_in_colo_state()) { + /* In COLO stage, put all pages into cache temporarily */ + host = colo_cache_from_block_offset(block, addr); + } else { + /* + * In migration stage but before COLO stage, + * Put all pages into both cache and SVM's memory. + */ + host_bak = colo_cache_from_block_offset(block, addr); + } } if (!host) { error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); ret = -EINVAL; break; } - if (!migration_incoming_in_colo_state()) { ramblock_recv_bitmap_set(block, host); } @@ -3514,6 +3543,9 @@ static int ram_load_precopy(QEMUFile *f) if (!ret) { ret = qemu_file_get_error(f); } + if (!ret && host_bak) { + memcpy(host_bak, host, TARGET_PAGE_SIZE); + } } ret |= wait_for_decompress_done(); diff --git a/migration/ram.h b/migration/ram.h index a553d40751..5ceaff7cb4 100644 --- a/migration/ram.h +++ b/migration/ram.h @@ -66,5 +66,6 @@ int ram_dirty_bitmap_reload(MigrationState *s, RAMBlock *rb); /* ram cache */ int colo_init_ram_cache(void); void colo_release_ram_cache(void); +void colo_incoming_start_dirty_log(void); #endif