Patching stubdom

General Discussion
1

Can someone help to realize:

  1. Why it is still needed to patch stubdom in order to be able to provide HVMs with devices attached to it (mostly GPU as I understand) with more than 3.5GB RAM? Why that wouldn’t be already implemented in kernel? I know this isn’t Qubes specific but it impacts Qubes in a huge way.
  2. How to track down why some succeed to allocate more than 3.5GB RAM to HVM, while others don’t even after applying correctly the same patch?

Thanks in advance.

1 Like
2

It’s not kernel patching, but stubdom patching to pass additional qemu argument.
Read the comments in this issue:
AppVM with GPU pass-through crashes when more than 3.5 GB (3584MB) of RAM is assigned to it · Issue #4321 · QubesOS/qubes-issues · GitHub

1 Like
3

Thank you for the correcting me, of course I meant on stubdom. Time for another coffee! :grinning: :coffee:
The 2nd question still remains, though.

4

Any idea how to track down why some succeed to allocate more than 3.5GB RAM to HVM, while others don’t even after applying correctly the same patch?

5

I think it’d be better to ask this in the github issue above as people there are trying to figure it out as well.

6

If it’s better there, then I wouldn’t see the purpose of this forum.

Anyway, thanks, but I saw very few topics about it there. It’s more about passthrough, and the ones who succeeded, didn’t mention it, except again very few without any response on the issue there as well.