So you are trying to work around the "problem" that distro kernels provides drivers for everything by adding additional kernel complexity? What prevents you from using a sane, stripped down, kernel image for these vms which would keep things sane and simpler and easier to audit and most importantly, prove, that the image is not doing something you don't expect it to do? Why not just make distros that want to support this type of platform, also provide these tiny kernel images? Why are you pushing this work on the kernel community instead?
Greg, I'm surprised by your comment. Traditionally we've been tried to support all platforms in unified binary kernels and gone to considerable complications to do so. That has been standard Linux practice for at least the 90ies. In some cases we went to considerable pain to support that, for example for the 5 level page tables or for paravirt ops.
Imagine there were 10 new features or platforms and they would all ask distributions to produce custom kernels for them, they would need to maintain 10 different kernel packages forever for all these different cases. It's totally reasonable that they don't want to do that.
Also even if they were willing to do custom configs it's not clear how this could be maintained and distributed. We would either have a standard TDX config and get everyone to agree on it (very difficult). Or some enforcement mechanism at the Kconfig level that forces most drivers to be disabled when TDX is on, which would be also a considerable new mechanism and complication. In addition there are drivers which are not covered by Kconfig today (like quite a few of the basic platform drivers), but which we still want to filter. So to implement a full build time mechanism would likely need more changes than this relatively straight forward run time mechanism based on the driver model.
And of course there other use cases for a run time filter mechanism. For example let's say you want filtering for Thunderbolt security. In this case it has to be done at runtime because it's not practical to have a kernel that is only built for Thunderbolt drivers, but doesn't support anything else that is on the SOC. The only sane way to handle such a case is to make a runtime distinction.
And what's wrong with the current method of removing drivers from devices that you do not want them to be bound to? We offer that support for all busses now that want to do it, what driver types are you needing to "control" here that does not take advantage of the existing infrastructure that we currently have for this type of thing?
I'm not sure what mechanism you're referring to here, but in general don't want the drivers to initialize at all because they might get exploited in any code that they execute.The intention is to disable all drivers except for a small allow list, because it's not practical to harden all 25M lines of Linux code.
-Andi
