[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] [PATCH v6 00/10] migration: objectify MigrationState
From: |
Peter Xu |
Subject: |
[Qemu-devel] [PATCH v6 00/10] migration: objectify MigrationState |
Date: |
Tue, 27 Jun 2017 12:10:09 +0800 |
v6:
- patch 1: fix comment, and add more comment for
register_compat_props() [Eduardo]
- patch 2: add more comments for register_compat_props_array() and the
new AccelClass.global_props [Eduardo]
- patch 3: rename user_register_compat_props() into
user_register_global_props, and some update on comments [Eduardo]
- patch 5: let xen_compat_props be static, and use a better "end of
list" for it [Eduardo] (since the change is trivial, I didn't remove
Juan's r-b)
v5:
- add r-b for Juan on patch 4,5,7-10
- patch 2: introduce register_compat_props_array() that may be further
used by machine codes [Eduardo]
- patch 2: fix english error [Eduardo]
v4:
- dropped lots of patches related to AccelState.global_props. Now I am
using AccelClass.global_props, and only use it in Xen codes.
- one patch is added to clean up the global property registerations,
with some comment to show the ordering.
- one patch is added to dump migration globals in "info migrate"
unconditionally (last patch).
- changed all the following xen_init() patches to use
AccelClass.global_props.
- (since most of the patches are either touched-up again, or threw
away, so no much r-b added, only one for Juan on the only_migratable
patch, anyway, thanks for reviewing the old codes!)
------------
The main goal of this series is to let MigrationState be a QDev.
It helps in many use cases.
First of all, we can remove many legacy tricky functions. To name
some: savevm_skip_section_footers(), savevm_skip_configuration(), etc.
They didn't do much thing but setup a bool value. If MigrationState
can be a QDev, then these things can be setup by the HW_COMPAT_* magic
with some lines like:
{
.driver = "migration",
.property = "send-configuration",
.value = "off",
}
Next, if this can be merged and okay, we can move on to convert more
things into properties for migration. A very attractive use case of it
is, we will be able to do this for migration:
-global migration.postcopy=on
Then we don't need to use either HMP/QMP interface to enable it. It
greatly simplifies the migration test scripts.
Why QDev not QObject? The answer is simple: all the magic that we want
for migration is bound to QDev (HW_COMPAT, "-global"). If one day we
want to move these features from QDev to QObject, that'll be fine and
easy for MigrationState. But before that, let's have MigrationState a
QDev. :-)
Please kindly review. Thanks.
Peter Xu (10):
machine: export register_compat_prop()
accel: introduce AccelClass.global_props
vl: clean up global property registerations
migration: let MigrationState be a qdev
migration: move global_state.optional out
migration: move only_migratable to MigrationState
migration: move skip_configuration out
migration: move skip_section_footers
migration: merge enforce_config_section somewhat
migration: hmp: dump globals
accel/accel.c | 6 ++
hmp.c | 3 +
hw/core/machine.c | 13 -----
hw/core/qdev-properties.c | 21 +++++++
hw/i386/pc_piix.c | 3 -
hw/ppc/spapr.c | 3 -
hw/xen/xen-common.c | 25 +++++++--
include/hw/compat.h | 12 ++++
include/hw/qdev-properties.h | 29 ++++++++++
include/migration/global_state.h | 1 -
include/migration/misc.h | 6 +-
include/sysemu/accel.h | 11 ++++
include/sysemu/sysemu.h | 1 -
migration/global_state.c | 9 +--
migration/migration.c | 118 +++++++++++++++++++++++++++++++--------
migration/migration.h | 33 +++++++++++
migration/savevm.c | 32 ++---------
vl.c | 41 ++++++++++++--
18 files changed, 277 insertions(+), 90 deletions(-)
--
2.7.4