xen/x86: Update topology map for PV VCPUs
authorBoris Ostrovsky <boris.ostrovsky@oracle.com>
Wed, 5 Oct 2016 17:09:33 +0000 (13:09 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 16 Oct 2016 16:03:38 +0000 (18:03 +0200)
commit88540ad0820ddfb05626e0136c0e5a79cea85fd1
treeb9e4cdc2b5ca91c2e3d03dd864f810c6a035ebb9
parentc64c7600e60086b658250b9f80b18cb959ed7b0f
xen/x86: Update topology map for PV VCPUs

commit a6a198bc60e6c980a56eca24d33dc7f29139f8ea upstream.

Early during boot topology_update_package_map() computes
logical_pkg_ids for all present processors.

Later, when processors are brought up, identify_cpu() updates
these values based on phys_pkg_id which is a function of
initial_apicid. On PV guests the latter may point to a
non-existing node, causing logical_pkg_ids to be set to -1.

Intel's RAPL uses logical_pkg_id (as topology_logical_package_id())
to index its arrays and therefore in this case will point to index
65535 (since logical_pkg_id is a u16). This could lead to either a
crash or may actually access random memory location.

As a workaround, we recompute topology during CPU bringup to reset
logical_pkg_id to a valid value.

(The reason for initial_apicid being bogus is because it is
initial_apicid of the processor from which the guest is launched.
This value is CPUID(1).EBX[31:24])

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/x86/xen/smp.c
This page took 0.025993 seconds and 5 git commands to generate.