Hi Jan,
On 13/06/2019 14:40, Jan Beulich wrote:
On 13.06.19 at 15:14, <julien.gr...@arm.com> wrote:
I am not asking you to spend a lot of time on it. But if you have a gut feeling
this can be done, then a little help would be extremely useful...
... I thought I had already outlined a model: Allow cross-vCPU updates
only while the target vCPU is still offline. Once online, a vCPU can only
itself update its runstate area address. I think you can get away
without any locks in this case; there may be a corner case with a vCPU
being onlined right at that point in time, so there may need to be a more
strict condition (like "only one online vCPU" instead of "the target vCPU
is offline").
Sorry I may have missed it. We can't really restrict the usage of the current
hypercall (it is pretty lax). So I think any lockless solution would require to
allow the hypercall
to be used together (which I want to avoid).
If we agree to allow the two hypercalls to be used together, then if we protect
the update with domain_lock() then you should be able to avoid any race with the
update path as onlining a vCPU requires to take the domain_lock() (see
do_vcpu_op for x86 and do_common_cpu_on for Arm).
Cheers,
--
Julien Grall
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
