Alyssa Ross
Alyssa Ross
> Don't wait around for queued darwin builds that are not `meta.available` Some of the common failures happen on Linux too.
Does this happen consistently? Is it likely to be possible to reproduce this without any special hardware?
I think we need https://github.com/NixOS/nixpkgs/pull/340089, and we could say to run with `dontAddPrefix=1 setOutputFlags=`.
https://github.com/NixOS/nixpkgs/pull/349360 would possibly make this workable with the new interface.
I don't know. I do know that making them different happened on purpose (https://github.com/NixOS/nixpkgs/pull/331167), but maybe this wasn't considered. cc @emilazy @mmlb
The fix was https://github.com/rust-vmm/vhost/pull/243. Why do you need to use an old Cloud Hypervisor with a new virtiofsd?
Ah, I was mistaken about which version was LTS — that makes sense. I agree that this should be fixed on the LTS.
I am interested in implementing this, but only once Cloud Hypervisor has a solution for interrupt remapping on x86_64. (On x86_64, generally the IOMMU is expected to handle interrupt remapping,...
> Ah, do you have a link to the documentation? Does QEMU implement it? There is an implementation for QEMU that has never been upstreamed, and crosvm previously implemented it...
I'm not familiar with this API, so I might have missed something, but why does the caller need to guarantee the validity of `slot`? What's unsafe about passing an invalid...