Re: E820 memory allocation issue on Threadripper platforms

[Branden Sherrell <sherrellbc@xxxxxxxxx>]

Sorry about that. The thread quickly diverged from the core issue to helping 
diagnose GPU issues, so I thought it best to circle back to the ticket 
reference to realign.

In any event, if there is any aid I can provide on this front please let me 
know. I have worked around this issue for now by rebuilding my x86 dom0 
(Debian) kernel from source with link at 2MiB rather than 16MiB (default). But 
this is a rather annoying requirement when updating the dom0 kernel. If there 
is any information you need from such an offending platform as this one feel 
free to let me know and I can test or otherwise provide relevant information as 
needed.

Branden.

> On Jun 24, 2024, at 9:40 AM, Jan Beulich <jbeulich@xxxxxxxx> wrote:
> 
> On 24.06.2024 15:07, Branden Sherrell wrote:
>> What is the reasoning that this fix be applied only to PVH domains? Taking a 
>> look at the fix logic it appears to walk the E820 to find a suitable range 
>> of memory to load the kernel into (assuming it can be determined that the 
>> kernel is also relocatable). Why can this logic not be applied to dom0 
>> kernel load in general?
> 
> Because PV requirements are different, first and foremost because there
> we have pseudo-physical and machine memory maps to deal with. As you can
> see from [1] I've raised the topic on how to deal with PV there, but so
> far there was no reply helping the issue towards resolution.
> 
> Btw - please don't top-post.
> 
> Jan
> 
> [1] https://lists.xen.org/archives/html/xen-devel/2024-06/msg00831.html