This series tried to introduce a new IRQ_ROUTING type named "eventfd". This work is totally RFC, and the idea is invoked when I was trying to cope with a VT-d vIOMMU error handling case when guest setup incorrect IRTE. Since this idea is quite interesting to me (which looks like a reversed version of irqfd, since both of them are doing convertions of virq <-> eventfd, just in different directions), and I see it relatively safe to have (it touches little other part of the codes), so here came the series. I posted this series mainly wanted to know whether we like this, and also I'm curious about whether this feature can be leveraged in other use cases outside my "vIOMMU problem". If we think this is something safe/good to have, I can continue to work on RFC v2 to provide workable userspace test program (or QEMU branch) that leverages this new feature. If not, any kind of NACK is also welcomed. :-) Please kindly review. Thanks, Peter Xu (2): kvm: x86: add new gsi route type EVENTFD kvm: add new cap KVM_CAP_GSI_EVENTFD arch/x86/kvm/irq_comm.c | 33 +++++++++++++++++++++++++++++++++ arch/x86/kvm/x86.c | 1 + include/linux/kvm_host.h | 7 +++++++ include/uapi/linux/kvm.h | 7 +++++++ virt/kvm/irqchip.c | 6 ++++++ 5 files changed, 54 insertions(+) -- 2.7.4
