Skip to contentSkip to conversation

migration: do not sent zero_pages in bulk stage

RFC PATCH In review Peter Lieven 2 participants 2 messages Mar 12, 2013

lore
Share & tools
lore
#0 lore raw

RFC PATCH 8/9 migration: do not sent zero_pages in bulk stage

if the dup_page is a zero_page do not send it at all. the memory at the destination reads as zero anyway.

---
 arch_init.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
arch_init.c +4 -2
index 04c82e4..abd54a7 100644
@@ -447,9 +447,11 @@ static int ram_save_block(QEMUFile *f, bool last_stage)
             bytes_sent = -1;
             if (ram_bulk_stage && is_dup_page(p)) {
                 acct_info.dup_pages++;
-                bytes_sent = save_block_hdr(f, block, offset, cont,
+                if (p[0]) {
+                    bytes_sent = save_block_hdr(f, block, offset, cont,
                                             RAM_SAVE_FLAG_COMPRESS);
-                qemu_put_byte(f, *p);
+                    qemu_put_byte(f, *p);
+                }
                 bytes_sent += 1;
             } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
                 current_addr = block->offset + offset;
Signature · 1 line
1.7.9.5
#1 lore raw
Quoted 2 lines · Peter Lieven

if the dup_page is a zero_page do not send it at all. the memory at the destination reads as zero anyway.

It might help to comment that this is only true for the bulk migration stage. If the guest is actively zeroing pages, then later stages might encounter an all-zero page that was not all zeros in the bulk migration, so we must transmit the page in that case.

Quoted 11 lines
---
 arch_init.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
arch_init.c +0 -0
index 04c82e4..abd54a7 100644

arch_init.c static int ram_save_block(QEMUFile *f, bool · line 451

Quoted 6 lines
arch_init.c +1 -1
@@ -447,9 +447,11 @@ static int ram_save_block(QEMUFile *f, bool
 last_stage)
             bytes_sent = -1;
             if (ram_bulk_stage && is_dup_page(p)) {
                 acct_info.dup_pages++;
-                bytes_sent = save_block_hdr(f, block, offset, cont,
+                if (p[0]) {

Your code has it right, because of the ram_bulk_stage check guarding this if body, but I did a double take when reading just your commit message.

Quoted 8 lines
arch_init.c +3 -1
+                    bytes_sent = save_block_hdr(f, block, offset, cont,
                                             RAM_SAVE_FLAG_COMPRESS);
-                qemu_put_byte(f, *p);
+                    qemu_put_byte(f, *p);
+                }
                 bytes_sent += 1;
             } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
                 current_addr = block->offset + offset;
Signature · 2 lines
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org