Re: [edk2-devel] [PATCH v3 1/1] MdeModulePkg: Add the EFI_RESOURCE_ATTRIBUTE_SPECIAL_PURPOSE attribute

["Oliver Smith-Denny" <osde@linux.microsoft.com>]

Hi Mike,

This conversation has gotten a little fragmented across this thread and
Dhaval put up a PR I commented on :).

On 6/11/2024 10:30 AM, Michael D Kinney wrote:
> Hi Oliver,
> 
> The DXE Core already has a policy to auto convert untested memory to
> tested memory if the DXE Core runs out of memory.
> 

Yeah, this was my suggestion on Dhaval's PR, that we follow a similar
memory promotion with EFI_MEMORY_SP as we do with untested memory. I
was planning to write this up, but haven't gotten to it yet.

> Should we have a different memory type to allocate memory with the
> EFI_MEMORY_SP attribute?

There is a lot of ambiguity in the PI/UEFI specs around what memory
types EFI_MEMORY_SP can be. This is further complicated by the CDAT
spec which uses EDKII-isms in defining the memory type (which it
probably shouldn't).

Amongst the various groups I have been chatting with (OS folks, this
mailing thread, firmware folks), there does seem to be a consensus that
the UEFI spec, at least, needs clarification here, as currently it just
says this attribute means special memory, which of course in and of
itself means nothing :).

However, these systems don't really exist yet. As these discussions
continue, I am changing my opinion to say less is more, for now. I think
we can't do too much architecting yet without knowing what these systems
are going to look like and what OSes will expect. If we get to a point
where we have systems with only remote attached memory (or the vast
majority remote attached memory) then having a separate memory type
probably does make sense. It would also solve the unenlightened OS
case where it wouldn't use SP memory if it was a new memory type,
whereas just the attribute on EfiConventionalMemory will have an
unenlightened OS use the memory.

> 
> What would be the side effects of performing general allocations from
> memory with the EFI_MEMORY_SP attribute?  If there are potentially bad
> side effects, then the DXE Core should never allocate from those
> ranges and it is up to the platform to make sure there is enough normal
> memory to boot.
> 

Right, I think we don't know enough yet, and so defaulting to UEFI not
using the memory and letting these systems mature a bit makes sense.
Inevitably if we build something now before these systems are fleshed
out, we will have built something that doesn't quite work.

Thanks,
Oliver



-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#119553): https://edk2.groups.io/g/devel/message/119553
Mute This Topic: https://groups.io/mt/106165072/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-