Re: [edk2-devel] How does EDK2 detect virtio-blk-pci device as a boot device

["Laszlo Ersek" <lersek@redhat.com>]

On 03/11/21 15:48, Xiaohe Yang wrote:
> Hello, I am using EDK2  OvmfPkgX64 at commit 37568365, and have a
> problem. The following is the detail.
> 
> 
> The Image disk that I use has EFI system partition, and is configured as
> a virtio-blk-pci device. Bootorder file-entry is not added in fw_cfg
> device. The log of EDK2 shows that
> 
> (1)  VirtioBlkDxe.efi is successfully loaded at DXE stage.
> BootScriptExecutorDxe drivers "was discovered but not loaded".
> 
> (2) PCI enumeration has finished at the start of BDS stage,
> virtio-blk-pci device is discovered. And one Option Rom's address is 0/0.
> 
> (3) Before entering Shell, log shows "map: No mapping found".
> 
> (4) After entering Shell, I choose “Boot From File”, the virtio-blk-pci
> device is not displayed.
> 
> 
> So I wonder what are the requirements for UEFI to detect the
> virtio-blk-pci device as a boot device, except having EFI system partition?

If you have a firmware log with DEBUG_VERBOSE enabled, I could take a look.

Assuming you are *not* on QEMU (e.g., you are developing a new
virtio-blk device, and running OVMF on your non-QEMU emulator), it could
be anything, really. Try to add DEBUG messages to the following drivers
/ functions:

- VirtioPciDeviceDxe
  - VirtioPciDeviceBindingSupported
  - VirtioPciDeviceBindingStart

- Virtio10Dxe
  - Virtio10BindingSupported
  - Virtio10BindingStart

- VirtioBlkDxe
  - VirtioBlkDriverBindingSupported
  - VirtioBlkDriverBindingStart

These drivers all follow the UEFI driver model. So, if you find no
mapping for the virtio-blk-pci device in the UEFI shell, then they fail
to bind your device implementation for some reason.

You mention that there is no "bootorder" fw_cfg file, so
PlatformBdsConnectSequence() will see ConnectDevicesFromQemu() fail, and
then EfiBootManagerConnectAll() will be called. If the virtio-blk-pci
devices are not connected under those circumstances, then they don't do
something that the virtio PCI transport drivers (0.9.5 or 1.0) expect,
or the virtio-blk device driver expects. There's no silver bullet, you
may have to add DEBUGs to various spots in these drivers, to see what fails.

Thanks
Laszlo