]> www.infradead.org Git - users/dwmw2/qemu.git/commit
hw/xen: select kernel mode for per-vCPU event channel upcall vector
authorDavid Woodhouse <dwmw@amazon.co.uk>
Wed, 11 Oct 2023 23:06:26 +0000 (00:06 +0100)
committerDavid Woodhouse <dwmw@amazon.co.uk>
Wed, 25 Oct 2023 11:45:22 +0000 (12:45 +0100)
commit19357da98e196c7135e9a8e3bee5d6aec52fe77c
tree8ab31c01e86717fb8f1ffda2b34d3730c58d8a26
parentf03de53962271dd03d4b33553a460fc7c61f9a4f
hw/xen: select kernel mode for per-vCPU event channel upcall vector

A guest which has configured the per-vCPU upcall vector may set the
HVM_PARAM_CALLBACK_IRQ param to fairly much anything other than zero.

For example, Linux v6.0+ after commit b1c3497e604 ("x86/xen: Add support
for HVMOP_set_evtchn_upcall_vector") will just do this after setting the
vector:

       /* Trick toolstack to think we are enlightened. */
       if (!cpu)
               rc = xen_set_callback_via(1);

That's explicitly setting the delivery to GSI#1, but it's supposed to be
overridden by the per-vCPU vector setting. This mostly works in Qemu
*except* for the logic to enable the in-kernel handling of event channels,
which falsely determines that the kernel cannot accelerate GSI delivery
in this case.

Add a kvm_xen_has_vcpu_callback_vector() to report whether vCPU#0 has
the vector set, and use that in xen_evtchn_set_callback_param() to
enable the kernel acceleration features even when the param *appears*
to be set to target a GSI.

Preserve the Xen behaviour that when HVM_PARAM_CALLBACK_IRQ is set to
*zero* the event channel delivery is disabled completely. (Which is
what that bizarre guest behaviour is working round in the first place.)

Fixes: 91cce756179 ("hw/xen: Add xen_evtchn device for event channel emulation")
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>
hw/i386/kvm/xen_evtchn.c
include/sysemu/kvm_xen.h
target/i386/kvm/xen-emu.c