mbox series

[kvmtool,v2,0/9] Disk fixes and AIO reset

Message ID 20190404132050.37309-1-jean-philippe.brucker@arm.com (mailing list archive)
Headers show
Series Disk fixes and AIO reset | expand

Message

Jean-Philippe Brucker April 4, 2019, 1:20 p.m. UTC
Changes since v1 [1]:
* Added review tags
* Fixed lkvm-static build, using $(OBJS_STATOPT/DYNOPT) in the makefile
  rather than $(OBJS).

Patches 1-7 fix a few issues found while testing disk backends, and
patches 8-9 implement reset for the AIO backend. By draining the AIO
queue on reset, we avoid writing I/O completions to the used ring after
the guest reclaimed the ring's pages.

[1] https://www.spinics.net/lists/kvm/msg182324.html

Jean-Philippe Brucker (9):
  qcow: Fix qcow1 exit fault
  virtio/blk: Set VIRTIO_BLK_F_RO when the disk is read-only
  guest: sync disk before shutting down
  disk/aio: Refactor AIO code
  disk/aio: Fix use of disk->async
  disk/aio: Fix AIO thread
  disk/aio: Cancel AIO thread on cleanup
  disk/aio: Add wait() disk operation
  virtio/blk: sync I/O on reset

 Makefile                 |   2 +
 disk/aio.c               | 150 +++++++++++++++++++++++++++++++++++++++
 disk/blk.c               |  10 +--
 disk/core.c              |  69 ++++++++----------
 disk/qcow.c              |   3 +-
 disk/raw.c               |  56 +++------------
 guest/init.c             |   1 +
 include/kvm/disk-image.h |  53 ++++++++++++--
 include/kvm/read-write.h |  11 ---
 util/read-write.c        |  36 ----------
 virtio/blk.c             |   7 +-
 11 files changed, 249 insertions(+), 149 deletions(-)
 create mode 100644 disk/aio.c

Comments

Will Deacon April 26, 2019, 2:09 p.m. UTC | #1
On Thu, Apr 04, 2019 at 02:20:41PM +0100, Jean-Philippe Brucker wrote:
> Changes since v1 [1]:
> * Added review tags
> * Fixed lkvm-static build, using $(OBJS_STATOPT/DYNOPT) in the makefile
>   rather than $(OBJS).
> 
> Patches 1-7 fix a few issues found while testing disk backends, and
> patches 8-9 implement reset for the AIO backend. By draining the AIO
> queue on reset, we avoid writing I/O completions to the used ring after
> the guest reclaimed the ring's pages.

Cheers, I've applied the series.

Will