>>> On 06.09.16 at 13:22, <xuqu...@huawei.com> wrote: > On September 06, 2016 5:05 PM, Jan Beulich <jbeul...@suse.com> >>>>> On 06.09.16 at 10:53, <xuqu...@huawei.com> wrote: >>> to be honest, my fix is also a tradeoff.. w/ my patch, multiple >>> pending pt interrupt instances may be combined as one injection of >>> guest timer interrupt.. >>> but as Yang said, It should be ok since it happens rarely and even >>> native OS will encounter the same problem if it disable the interrupt for >>> too >>long. >> >>And one OS may properly deal with that situation while another might not. For >>example, let me remind you of issues the hypervisor had in earlier days when >>no >>event/softirq processing happened for long enough, resulting in time >>management issues due to missed platform timer overflow handling. IOW an OS >>may avoid that situation by simply not disabling IRQs for too long a time. >> > Thanks for your reminding, > Jan, do you have any suggestion to fix this PT and apicv issue? > What do you think about Kevin's?
Kevin's suggestion sounded reasonable, iirc, and I'd rely on his and your better VMX knowledge for the low level details. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org https://lists.xen.org/xen-devel