On 09.11.24 05:59, Hyman Huang wrote:
The first iteration's RAMBlock dirty sync can be omitted because QEMU
always initializes the RAMBlock's bmap to all 1s by default.
Signed-off-by: Hyman Huang <yong.hu...@smartx.com>
---
migration/cpu-throttle.c | 2 +-
migration/ram.c | 11 ++++++++---
2 files changed, 9 insertions(+), 4 deletions(-)
diff --git a/migration/cpu-throttle.c b/migration/cpu-throttle.c
index 5179019e33..674dc2004e 100644
--- a/migration/cpu-throttle.c
+++ b/migration/cpu-throttle.c
@@ -141,7 +141,7 @@ void cpu_throttle_dirty_sync_timer_tick(void *opaque)
* effect on guest performance, therefore omit it to avoid
* paying extra for the sync penalty.
*/
- if (sync_cnt <= 1) {
+ if (!sync_cnt) {
goto end;
}
diff --git a/migration/ram.c b/migration/ram.c
index 05ff9eb328..571dba10b7 100644
--- a/migration/ram.c
+++ b/migration/ram.c
@@ -2718,7 +2718,7 @@ static void ram_list_init_bitmaps(void)
{
MigrationState *ms = migrate_get_current();
RAMBlock *block;
- unsigned long pages;
+ unsigned long pages, clear_bmap_pages;
uint8_t shift;
/* Skip setting bitmap if there is no RAM */
@@ -2736,6 +2736,7 @@ static void ram_list_init_bitmaps(void)
RAMBLOCK_FOREACH_NOT_IGNORED(block) {
pages = block->max_length >> TARGET_PAGE_BITS;
+ clear_bmap_pages = clear_bmap_size(pages, shift);
/*
* The initial dirty bitmap for migration must be set with all
* ones to make sure we'll migrate every guest RAM page to
@@ -2751,7 +2752,12 @@ static void ram_list_init_bitmaps(void)
block->file_bmap = bitmap_new(pages);
}
block->clear_bmap_shift = shift;
- block->clear_bmap = bitmap_new(clear_bmap_size(pages, shift));
+ block->clear_bmap = bitmap_new(clear_bmap_pages);
+ /*
+ * Set clear_bmap to 1 unconditionally, as we always set bmap
+ * to all 1s by default.
+ */
+ bitmap_set(block->clear_bmap, 0, clear_bmap_pages);
}
}
}
@@ -2783,7 +2789,6 @@ static bool ram_init_bitmaps(RAMState *rs, Error **errp)
if (!ret) {
goto out_unlock;
}
- migration_bitmap_sync_precopy(false);
}
}
out_unlock:
For virtio-mem, we rely on the migration_bitmap_clear_discarded_pages()
call to clear all bits that correspond to unplugged memory ranges.
If we ommit the sync, we can likely have bits of unplugged ranges still
set to "1", meaning we would try migrate them later, although we shouldn't?
Or is that handled differently?
--
Cheers,
David / dhildenb
