Hi Andy,
On Mon, 21 Oct 2019 at 01:52, Andy Shevchenko andy.shevchenko@gmail.com wrote:
On Mon, Oct 21, 2019 at 7:24 AM Simon Glass sjg@chromium.org wrote:
In TPL we try to minimise code size so do not include the PCI subsystem. We can use fixed BARs and drivers can directly program the devices that they need.
However we do need to bind the devices on the PCI bus and without PCI this does not ordinarily happen. As a work-around, define a fake PCI bus which does this binding, but no other PCI operations. This is a convenient way to ensure that we can use the same device tree for TPL, SPL and U-Boot proper:
TPL - CONFIG_TPL_PCI is not set (no auto-config, fake PCI bus) SPL - CONFIG_SPL_PCI is set (no auto-config but with real PCI bus) U-Boot - CONFIG_PCI is set (full auto-config after relocation)
PCI(e) bus is present in a lot of SoCs (not exclusively x86). Perhaps better idea is to have something like lib/pci.c with minimum support for PCI type 1 and probably PCI type 2 accessors and other very basic functions.
I don't know of any use case for PCI in TPL on other platforms.
x86 is I think unique in that it requires PCI to do anything. For other platforms I am familiar with they can boot a fair way without it.
I do want to avoid the premature-optimisation problem, i.e. inventing new use cases that no one uses. The only thing we really know right now is that we need this for newer x86 platforms.
Regards, Simon