Re: About runtime service for Xen on ARM

Re: About runtime service for Xen on ARM

[Julien Grall]

On 19/08/2015 23:27, Shannon Zhao wrote:
> Hi Julien,

Hi Shannon,

Sorry for the late answer I was on holiday. In general it's better to CC 
multiple person and the mailing list as we may not be able to answer in 
short time or not know the answer.

I've CCed the ML and few people which know better than me the status of 
EFI runtime.

> Do you know the status of runtime service for Xen on ARM? Does it
> support runtime service booting with DT? If not, is there a plan to
> support it?

There is no support of runtime EFI on ARM for now and I'm not aware of 
any short plan for this.

> In addition, if you know this, does it support runtime service for Xen
> on x86?

There is some on-going work for EFI runtime on x86. I'm not sure if it 
has been upstream (CCed Daniel for this).

Although, one of there is a fundamental difference between the UEFI boot 
on x86 and ARM. While the code is the same, on ARM the UEFI act as a 
stub and should not be able to communicate via exported variable with 
Xen (we clobber BSS just after the UEFI stub jump on the normal boot start).

I know that the ARM EFI stub is using some shared variable with Xen (I 
have in mind the memory banks see bootinfo). We are very lucky that it's 
working because the variable is living in initdata which somehow is part 
of text (I'm not sure why) and not BSS.
So we should either decide that the stub and xen should be able to 
communicate together or not.

I would be in favor of the former but I don't know what are the view of 
the others.

Regards,

-- 
Julien Grall

Re: About runtime service for Xen on ARM

[Jan Beulich]

>>> On 26.08.15 at 05:24, <julien.grall@citrix.com> wrote:
> On 19/08/2015 23:27, Shannon Zhao wrote:
>> In addition, if you know this, does it support runtime service for Xen
>> on x86?
> 
> There is some on-going work for EFI runtime on x86. I'm not sure if it 
> has been upstream (CCed Daniel for this).

Daniel is working on boot improvements; runtime services have been
available from the beginning of EFI support on the x86 side.

Jan

Re: About runtime service for Xen on ARM

[Stefano Stabellini]

On Wed, 26 Aug 2015, Jan Beulich wrote:
> >>> On 26.08.15 at 05:24, <julien.grall@citrix.com> wrote:
> > On 19/08/2015 23:27, Shannon Zhao wrote:
> >> In addition, if you know this, does it support runtime service for Xen
> >> on x86?
> > 
> > There is some on-going work for EFI runtime on x86. I'm not sure if it 
> > has been upstream (CCed Daniel for this).
> 
> Daniel is working on boot improvements; runtime services have been
> available from the beginning of EFI support on the x86 side.

I am pretty sure that we can reuse the same code to do runtime services on arm too.

Re: About runtime service for Xen on ARM

[Shannon Zhao]

Hi Julien,

Thanks for your reply. :)

On 2015/8/26 11:24, Julien Grall wrote:
>
>
> On 19/08/2015 23:27, Shannon Zhao wrote:
>> Hi Julien,
>
> Hi Shannon,
>
> Sorry for the late answer I was on holiday. In general it's better to CC
> multiple person and the mailing list as we may not be able to answer in
> short time or not know the answer.
>
> I've CCed the ML and few people which know better than me the status of
> EFI runtime.
>
>> Do you know the status of runtime service for Xen on ARM? Does it
>> support runtime service booting with DT? If not, is there a plan to
>> support it?
>
> There is no support of runtime EFI on ARM for now and I'm not aware of
> any short plan for this.
>
>> In addition, if you know this, does it support runtime service for Xen
>> on x86?
>
> There is some on-going work for EFI runtime on x86. I'm not sure if it
> has been upstream (CCed Daniel for this).
>
> Although, one of there is a fundamental difference between the UEFI boot
> on x86 and ARM. While the code is the same, on ARM the UEFI act as a
> stub and should not be able to communicate via exported variable with
> Xen (we clobber BSS just after the UEFI stub jump on the normal boot
> start).
>
> I know that the ARM EFI stub is using some shared variable with Xen (I
> have in mind the memory banks see bootinfo). We are very lucky that it's
> working because the variable is living in initdata which somehow is part
> of text (I'm not sure why) and not BSS.
> So we should either decide that the stub and xen should be able to
> communicate together or not.
>
> I would be in favor of the former but I don't know what are the view of
> the others.
>
> Regards,
>

-- 
Shannon