[Xen-devel] [PATCH] Fix QEMU HVM hotplug race in QEMU traditional (Xen 4.1, Xen 4.2, and Xen 4.3) (v1).

[Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>]

Hey Ian and Stefano,

Please see this thread: 
http://lists.xen.org/archives/html/xen-devel/2013-05/msg01053.html
for the debug patches and some of the discussion.

These three patches fix an race that has been in QEMU traditional for a long 
time.

The guts of the bug is that if you have a guest with these options:

vcpus=1
maxvcpus=32

and do 'xl vcpu-set <guest> 32' the guest OS only seems to hotplug 13 or 14 of 
the vCPUs.
To hotplug the rest I have to play games by offlining/onlining arbitrary number 
of 
them.

Or I can do 'xl vcpu-set <guest> 8', then '16', '24', etc. In groups of eight.

These patches fix this and should be considered for backport.
 hw/piix4acpi.c | 18 ++++++++-------
 monitor.c      |  4 ++--
 sysemu.h       |  4 +++-
 xenstore.c     | 73 ++++++++++++++++++++++++++++++++++++++++++++++++++++------
 4 files changed, 81 insertions(+), 18 deletions(-)

Konrad Rzeszutek Wilk (3):
      piix4acpi, xen, vcpu hotplug: Split the notification from the changes.
      piix4acpi, xen: Clarify that the qemu_set_irq calls just do an IRQ pulse.
      piix4acpi, xen, hotplug: Fix race with ACPI AML code and hotplug.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel