Hello Aaron,

On Fri, 23 Oct 2026, at 01:48, Aaron Pop via groups.io wrote:
> Hi everyone, 
> 
> Requesting feedback on a recent edk2 change that is functionally 
> correct, but is breaking some Video option ROMs. 
>
>
>
> The PR:
> MdeModulePkg/PciBusDxe: Preserve Physical BAR numbering by apop5 · Pull 
> Request #13258 · tianocore/edk2 
> <https://github.com/tianocore/edk2/pull/13258>
>
>
> It was intended to make the PciIo function according to UEFI 
> specification: 14. Protocols — PCI Bus Support — UEFI Specification 
> 2.11 documentation 
> <https://uefi.org/specs/UEFI/2.11/14_Protocols_PCI_Bus_Support.html#efi-pci-io-protocol-mem-write>
>
> Specifically
> “The BAR index of the standard PCI Configuration header to use as the 
> base address for the memory operation to perform. This allows all 
> drivers to use BAR relative addressing. The legal range for this field 
> is 0..5.”
> 
>
> Prior to this change, the internal PciBus.h structure PCI_IO_DEVICE  
> would pack BARs when placing them into `PCI_BAR  PciBar[PCI_MAX_BAR]`
> This had impacts when accessing via PciIo->Mem.Read/Write, which 
> accepts as a parameter the BAR index it is attempting to access. 
>
> example:
>
> Pci Device A
>   BAR0 – 32 bit memory, Placed in PciBar[0] in internal structure
>   BAR1 – 32 bit memory, Placed in PciBar[1] in internal structure
>   BAR2 – 64 bit memory, Placed in PciBar[2] in internal structure
>
> PCI Device B
>   BAR0 – 64 bit memory, Placed in PciBar[0] in internal structure
>   BAR1 – nothing, upper 32 bits of BAR0
>   BAR2 – 64 bit memory, Placed in PciBar[1] in internal structure
> 
> 
> When attempting to access the BARs through PciIo->Mem.Read
>  PciIo->Mem.Read (
>                      PciIo,
>                      EfiPciIoWidthUint32,
>                      PCI_BAR_IDX2,
>                      0,
>                      1,
>                      Value
>                      );
> 
> Prior to the PR, Pci Device A would access internal structure offset [2]
> Prior to the PR, Pci Device B would return EFI_UNSUPPORTED because 
> there would be nothing at index 2 in the internal structure.
>
> After the change, Pci Device A would work the same.
> After the change Pci Device B would correctly access BAR2 and be able 
> to read into the memory space.
>
>
> Since this has gone into the project, there have been 2 reported 
> problems. Both with Video Option Roms.
> From what can be collected, the video option roms, working around the 
> existing problems in the PCI driver, the option roms would pass in 
> PCI_BAR_IDX1 when attempting to access the second 64 bit BAR.
> 
> 

We can't just renumber BARs - this potentially breaks all drivers for
devices with 64-bit BARs appearing at index > 0.

So we have to revert it, and update the spec accordingly.

Please create a GitHub issue for this, and put Liming and the stewards on cc.



-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#122204): https://edk2.groups.io/g/devel/message/122204
Mute This Topic: https://groups.io/mt/121893816/21656
Group Owner: [email protected]
Unsubscribe: https://edk2.groups.io/g/devel/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-


Reply via email to