Re: [edk2-devel] [PATCH 1/1] MdeModulePkg: Load Serial driver earlier in DXE

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

On 2/25/24 17:18, Laszlo Ersek wrote:

> ... Ah, okay! This is the point where I re-read your steps, and now I
> see the actual breakage.
> 
> It is your step #2.
> 
> Placing the PciSioSerialDxe driver -- which is a UEFI driver that
> follows the UEFI driver model, and has a long list of dependencies on
> the architectural protocols -- in the APRIORI DXE file of the firmware
> volume(s) *completely violates* the DXE dispatch order.
> 
> I wrote above:
> 
>     How do you ensure that step (4) happens *both* early enough *and*
>     late enough?
> 
>     You must do step (4) *late enough* for PciSioSerialDxe to have a
>     fleeting chance to be dispatched by the DXE dispatcher (due to arch
>     protocol deps), and *early enough* for your key modules to find
>     SERIAL_IO_PROTOCOL (from PciSioSerialDxe) via DebugLib, in order to
>     produce actual debug messages.
> 
> And your step#2 is the answer to that: you violate the standard DXE
> dispatch order, by *overriding* the UEFI arch protocols depex in
> PciSioSerialDxe, by placing the driver in the APRIORI DXE file.
> 
> The APRIORI DXE file is *only* appropriate to use when you have
> *platform DXE* drivers that, for some reason, cannot be ordered
> deterministically correctly, by way of depexes, against other platform
> DXE drivers. APRIORI DXE is always a last resort; it serves as a crutch
> when the DEPEX language is simply not expressive enough for a given
> platform; APRIORI DXE is not license to *override* existent (and
> standard!) depexes.

I've looked again at the patch. It introduces a variant of
PciSioSerialDxe that is a DXE_DRIVER, and has

[Depex]
  TRUE

This means that step#2 (= APRIORI DXE) doesn't actually override a UEFI
driver's depex -- therefore I must agree that step#2 in itself does not
violate DXE dispatch order.

However, I continue to object to step (4), that is, to
ConnectController() being called by a platform module way before BDS is
entered. That's not how the driver binding protocol and the
ConnectController() boot service were intended, to my understanding.

Laszlo



-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#115932): https://edk2.groups.io/g/devel/message/115932
Mute This Topic: https://groups.io/mt/104469297/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-