On 12/6/2010 7:13 AM, Jan Kiszka wrote:
> Hi Chris,
>
> as I'm already running around, telling people that Tile might be the
> next arch to gain KVM support, I wanted to back this derived [1]
> information with some more details. Can you share some of your plans
> regarding this, either officially (LKML, kvm-devel) or yet privately?
> - What will be the level of support in the first version and long-term
> (CPU virtualization + I/O emulation, also I/O virtualization/
> pass-though)?
> - What use cases do you target, and why do you plan to use KVM for
> them?
> - What use cases may not fit a KVM-based approach?
>
> The background of this questionnaire is not (yet) a concrete project
> based on a Tile processor and KVM. Right now I'm primarily promoting KVM
> for use cases beyond classic x86 server scenarios, both in-house as well
> as in the community.
>
> TiA!
>
> Best regards,
> Jan Kiszka
>
> [1] http://permalink.gmane.org/gmane.linux.kernel/1048568
We already have a hypervisor that is used for Tile, which allows us to do
client isolation and spatial multiplexing (i.e. splitting the cores among
different supervisors), and smooths over some of the more nitty-gritty
hardware issues to present an easier API to the client supervisor, e.g.
Linux. The supervisor is paravirtualized, i.e. aware of the hypervisor API
for page-table management and I/O access.
But moving forward there is some appeal to using a standard virtualization
technology, and we picked KVM as the target that seemed best for us to
support. Some of the things this will facilitate for us include dynamic
reconfiguration of supervisor domains, sharing I/O devices between
supervisors, providing virtual devices to supervisors, virtual machine
migration/snapshots, etc. And, we'd like to support a standard management
interface such as the KVM interface, so our customers don't have to learn
how to manage the Tilera-specific hypervisor software.
None of this is committed to any particular release schedule yet, but this
is the direction we are currently planning to head.
--
Chris Metcalf, Tilera Corp.
http://www.tilera.com
Am 06.12.2010 16:59, Chris Metcalf wrote:
> On 12/6/2010 7:13 AM, Jan Kiszka wrote:
>> Hi Chris,
>>
>> as I'm already running around, telling people that Tile might be the
>> next arch to gain KVM support, I wanted to back this derived [1]
>> information with some more details. Can you share some of your plans
>> regarding this, either officially (LKML, kvm-devel) or yet privately?
>> - What will be the level of support in the first version and long-term
>> (CPU virtualization + I/O emulation, also I/O virtualization/
>> pass-though)?
>> - What use cases do you target, and why do you plan to use KVM for
>> them?
>> - What use cases may not fit a KVM-based approach?
>>
>> The background of this questionnaire is not (yet) a concrete project
>> based on a Tile processor and KVM. Right now I'm primarily promoting KVM
>> for use cases beyond classic x86 server scenarios, both in-house as well
>> as in the community.
>>
>> TiA!
>>
>> Best regards,
>> Jan Kiszka
>>
>> [1] http://permalink.gmane.org/gmane.linux.kernel/1048568
>
> We already have a hypervisor that is used for Tile, which allows us to do
> client isolation and spatial multiplexing (i.e. splitting the cores among
> different supervisors), and smooths over some of the more nitty-gritty
> hardware issues to present an easier API to the client supervisor, e.g.
> Linux. The supervisor is paravirtualized, i.e. aware of the hypervisor API
> for page-table management and I/O access.
>
> But moving forward there is some appeal to using a standard virtualization
> technology, and we picked KVM as the target that seemed best for us to
> support. Some of the things this will facilitate for us include dynamic
> reconfiguration of supervisor domains, sharing I/O devices between
> supervisors, providing virtual devices to supervisors, virtual machine
> migration/snapshots, etc. And, we'd like to support a standard management
> interface such as the KVM interface, so our customers don't have to learn
> how to manage the Tilera-specific hypervisor software.
>
> None of this is committed to any particular release schedule yet, but this
> is the direction we are currently planning to head.
>
Thanks for the information! Sound thrilling, looking forward seeing this
materializing.
Jan
--
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux