On Apr 28, 2021, at 6:03 AM, Ard Biesheuvel <ardb@kernel.org> wrote:

On Tue, 27 Apr 2021 at 21:31, Kinney, Michael D
<michael.d.kinney@intel.com> wrote:

This is an example of another approach.  This module uses PCI I/O or IoLib based on the type of device.



https://github.com/tianocore/edk2/tree/master/MdeModulePkg/Bus/Pci/PciSioSerialDxe



No additional protocols or lib classes/instances.  Instead, the register access APIs are included in the module and based on the type of device detected, it uses PCI I/O or IoLib:



https://github.com/tianocore/edk2/blob/5b90b8abb4049e2d98040f548ad23b6ab22d5d19/MdeModulePkg/Bus/Pci/PciSioSerialDxe/SerialIo.c#L1327

https://github.com/tianocore/edk2/blob/5b90b8abb4049e2d98040f548ad23b6ab22d5d19/MdeModulePkg/Bus/Pci/PciSioSerialDxe/SerialIo.c#L1358



This technique could for PCI I/O vs MMIO register access.  You would need to add more APIs for the use of PCI I/O or DmaLib for DMA access.

Is it really worth the effort to rewrite this code?

This patch has been circulating for a while now, and I fail to see the
point of refactoring and splitting up this code, given how unlikely it
is that DesignWare will ever put a real PCI frontend on this IP. The
SD/MMC override protocol was intended for implementations that are
almost SDHCI compliant, but have some quirks that need to be worked
around.