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

Re: [Xen-devel] [PATCH v14 for-xen-4.5 11/21] x86/VPMU: Interface for setting PMU mode and flags

>>> On 29.10.14 at 15:22, <boris.ostrovsky@xxxxxxxxxx> wrote:
> On 10/29/2014 04:14 AM, Jan Beulich wrote:
>> A static variable may indeed be insufficient here. Did you look at
>> the XSA-97 change at all, trying to mirror its logic here?
> You mean storing this in domain structure? I don't want to add new 
> fields to such a common structure for an operation that is exceedingly 
> inferquent.

No matter how (in)frequent, you'd have to go that route if you
want to stick to that mechanism.

>> However, I wonder whether a model without continuations (and
>> hence not along the lines of what we did for XSA-97) might not be
>> better here after all:
>> 1) Considering that you don't need access to the hypercall
>> arguments after initial evaluation, continue_hypercall_on_cpu()
>> would seem usable here: Once you visited all CPUs, you can be
>> certain a context switch occurred everywhere.
>> 2) You could pause the current vCPU after scheduling all tasklets
>> and have the last one unpause it and do the necessary cleanup.
> This sounds simpler than what I have now.
> I don't think I will need the tasklets with this approach: they are all 
> part of continue_hypercall_on_cpu()?


> As for pausing the VCPU? Won't the continue_hypercall_on_cpu() keep it 
> asleep until everyone has completed?

This was a second possible approach I suggested, not connected
to the first one.


Xen-devel mailing list



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