[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [RFC 1/6] xen/arm: Re-enable interrupt later in the trap path

Hi Andrii,

On 02/08/2019 13:24, Andrii Anisov wrote:

On 02.08.19 12:03, Julien Grall wrote:
A fair amount of leave_hypervisor_tail() deal with the guest itself (i.e vGIC, P2M...)

All that stuff is what hypervisor does for the guest. And does behind the guest's back.
Please define "guest's back".

Technically a guest accessing an IO does not know that the access will be emulated. So this should also take into account as "guest's back".

An hypercall is initiated by the guest directly, so I agree this is not done on guest's back.

Some of the work done in leave_hypervisor_tail() is an extension of IO emulation. They are not done directly in the IO emulation because they may take a long time and get preemption. So I don't see any difference with "IO emulation".

Regarding the vGIC. This is a bit more a grey area. While it is an overhead of virtualization, this is indirectly initiated by the guest. Indeed, you would only configure the vGIC if you receive an interrupt generated by one of the device assigned.

, so I think they should be accounted to the guest time.
This point is arguable. That's why we have a discussion here to agree on the time accounting approach, what will directly affect scheduling accuracy.

Note the "I think" in my answer. So this is my opinion and your input is 


Julien Grall

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.