Re: [edk2-devel] Running OVMF on crosvm

["Pedro Falcato" <pedro.falcato@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2343 bytes --]

Hi Mike,

What's the point of upstreaming firmware for crosvm? Isn't crosvm supposed
to be a "thin" hypervisor that runs the kernel directly?

Thanks,
Pedro

On Fri, Jun 17, 2022 at 12:40 AM Mike Gerow via groups.io <gerow=
google.com@groups.io> wrote:

> Hi Folks,
>
> For the last few months I've been using OVMF with crosvm
> <https://chromium.googlesource.com/chromiumos/platform/crosvm/+/HEAD/README.md#:~:text=crosvm%20is%20a%20virtual%20machine,on%20the%20Chrome%20OS%20platform.>
> to give it the ability to boot a UEFI OS image in a UEFI environment, and
> it's been working fantastically! Eventually I'd like to get secure boot
> working, but I haven't fully begun exploring that yet. I just wanted to
> post here to figure out how I might go about making crosvm properly
> supported within OVMF.
>
> As you might guess, OVMF didn't boot out of the box and I had to do quite
> a few hacks to get it to work. All the changes are attached as a diff, but
> I'll list the main points out here:
>
> * Switch TimerLib from the ACPI-based ones
> like OvmfPkg/Library/AcpiTimerLib/BaseAcpiTimerLib.inf
> to MdePkg/Library/SecPeiDxeTimerLibCpu/SecPeiDxeTimerLibCpu.inf since
> crosvm doesn't provide these methods
> * Hack out the RTC dependency completely, since for reasons I haven't yet
> figured out crosvm hangs OVMF tries to access it
> * Disable bus enumeration since crosvm doesn't support moving PCI BARs
> around, which causes it to hang
> * Remove VirtioFsDxe dependency as crosvm seems to choke on certain things
> it's doing too
> * Manually searching for ACPI tables instead of using PVH
>
> And one major theme of some of the other changes: removing conditionals
> that checked if we're cloud-hypervisor and just doing things the
> cloud-hypervisor way (which happens to work under crosvm, for reasons that
> are readily apparent given the nature of cloud-hypervisor and crosvm).
>
> Clearly this isn't upstreamable at this point, but I'd like eventually go
> that direction. I think a lot of these are just bugs in crosvm that'll need
> to be fixed, but for the cloud-hypervisor conditionals case I'd need to
> teach OVMF to detect when we're running under crosvm and act appropriately.
>
> In either case, I appreciate your attention and any feedback you might
> have on what I'm trying to do here.
> 
>
>

-- 
Pedro Falcato

[-- Attachment #2: Type: text/html, Size: 2944 bytes --]