Re: [Xen-devel] RT Xen on ARM - R-Car series

[Julien Grall <julien.grall@xxxxxxxxx>]

Sorry for the formatting.

On Fri, 15 Feb 2019, 18:30 Andrii Anisov, <andrii.anisov@xxxxxxxxx> wrote:

Hello Julien,

On 15.02.19 18:31, Julien Grall wrote:
> Why? Is it because you want to be cache-aligned?  If so, requiring the
> structure to be 64-bytes is not enough.

I did not mean caches.

What is the reason then?

> You also want the address to
> be 64-bytes aligned.

I would keep it as a hint for static/dynamic allocations in VMs, hoping the address would be normally 64 bytes aligned.
I hope it might be stronger than, only commenting it should not cross a page boundary. E.g. like `struct vcpu_register_vcpu_info` is commented.

I've got this idea after looking at runstate definition as per-cpu in Linux [1]

It is not obvious why it would be 64-bytes alignment from the definition. Can you please explain the rationale to impose that alignment?

I really appreciate you suggest ideas/patches but  it would be helpful if you provide rationale at the same time. This would avoid a round of e-mails just for asking the reasons and delay the interesting bits.

> If an OS cares about it, then it can aligned itself here.
I suppose we can hint the OS by structure alignment in the interface header, and require it from OS verifying it on hypercall handling

.

[1] https://elixir.bootlin.com/linux/v5.0-rc6/source/drivers/xen/time.c#L22

--
Sincerely,
Andrii Anisov.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel