Re: [PATCH v2 3/5] MdeModulePkg: implement generic PCI I/O driver for non-discoverable devices

["Ni, Ruiyu" <ruiyu.ni@intel.com>]

Ok. I didn't realize your PCI access is different from the standard way. I agree the PciLib suggestion is not good.

发自我的 iPhone

> 在 2016年11月16日，下午7:43，Ard Biesheuvel <ard.biesheuvel@linaro.org> 写道：
> 
>> On 15 November 2016 at 11:30, Ard Biesheuvel <ard.biesheuvel@linaro.org> wrote:
>>> On 15 November 2016 at 08:40, Ni, Ruiyu <ruiyu.ni@intel.com> wrote:
>>> Ard,
>>> Can you check whether PciLib can be used to replace the implementation in
>>> NonDiscoverablePciDeviceIo.c?
>>> 
>> 
>> OK
> 
> I am not sure how this should work. This driver needs to coexist with
> real PCI host bridge drivers, and the config space access method is
> fully defined by the requirements of the driver, i.e., each PCI config
> space access should be served from the fake PCI config space in
> memory, and never be forwarded to a real device.
> 
> So /providing/ a PciLib implementation, and requiring that that
> particular implementation must be used with this driver makes no
> sense, I think.
> And depending on a PciLib implementation does not make sense either,
> since the platform cannot choose which method to use.
> 
> Note that we are not emulating a PCI bus hierarchy, where a single
> contiguous chunk of memory emulates the entire config space of several
> devices. Each device is fully independent, and provides a minimal
> config space to describe the type of device, and the BARs
> 
> Or did I miss something?