Re: [edk2-devel] [PATCH v5 0/7] MdeModulePkg: add support for dispatching foreign arch PE/COFF images

["Ard Biesheuvel" <ard.biesheuvel@linaro.org>]

On Thu, 11 Apr 2019 at 11:58, Ard Biesheuvel via Groups.Io
<ard.biesheuvel=linaro.org@groups.io> wrote:
>
> Add the basic plumbing to DXE core, the PCI bus driver and the boot manager
> to allow PE/COFF images to be dispatched that target an architecture that is
> not native for the platform, but which is supported by one of potentially
> several available emulators.
>
> One implementation of such an emulator can be found here:
> https://github.com/ardbiesheuvel/X86EmulatorPkg/tree/upstream-v4
>
> This also allows us to get rid of the special treatment of EBC images in
> core code. Instead, the EbcDxe driver is augmented with an implementation
> of the EDK2 PE/COFF image emulator protocol so that internal knowledge of
> how EBC is implemented (I-cache flushing, thunks) is removed from the DXE
> core.
>
> Changes since v4:
> - Fix an issue in the protocol notify handler pointed out by Mike Kinney (#2)
>
> Changes since v3:
> - Simplify the handling of option ROMs and Driver#### images, by simply
>   deferring to the LoadImage() boot service to decide whether an image
>   can be supported or not - this removes some redundant checks from the
>   BDS layer and the PCI bus driver.
> - Move the machine type supported by the emulator into the protocol struct,
>   so we can optimize away calls into the emulator for each image loaded.
>   Instead, the LoadImage() code will only invoke the IsSupported() method for
>   images that are known to have a matching machine type.
>
> Note that I have considered, but ultimately dismissed the suggestion to
> register and unregister emulators via a new protocol. The main issue is
> that registering and unregistering structs containing sets of function
> pointers is awfully similar to managing a protocol database, and we already
> have the code to do that in EDK2.
>
> So instead, I have removed all the code that iterates over a handle buffer
> of emu protocols and invokes each one to see if it will support the image.
> Instead, this is all done by CoreLoadImage().
>
> Changes since v2:
> - incorporate feedback from Andrew Fish (delivered in person):
>   * pass a device path into the IsImageSupported() protocol method so that an
>     implementation can blacklist or whitelist certain devices, or implement
>     other policies that depend on the device where the driver originated
>   * allow the emulator to supersede the native loading of the image - this
>     permits things like X86 on X86 emulators for security sandboxing or debug
>
> Changes since v1:
> - subsume the EBC handling into the EDK2 emulator protocol and abstract
>   away from EBC specifics in core code.
> - allow multiple emulator implementations to co-exist
> - incorporate Star's review feedback
>


In order not to spam everyone with another revision within a couple of
hours, I have applied the feedback from Mike for the VS build errors
and pushed the result here:

https://github.com/ardbiesheuvel/edk2/pull/new/pecoff-emu-v5+

in case the MdeModulePkg maintainers want to pull the code and test it
during review.