Patch Set 3:

Not quite right. The only reason it was introduced is to solve the
name collision we have with PCI config accessors. We needed a way
to have the same source code file built for different stages; but
different stages use different prototypes for functions like
pci_config_read/write().

[Subrata] Wondering why that might have designed at first place. Might be to avoid use of complex pointer in those pre_ram stages ?

Yep, remainer of the days before cache-as-ram, when romstage was built with romcc that could not deal with structs or pointers at all. Also romstage does not have a "complete" devicetree but that part can be sorted out.

 > 
 > With the PCI config accessors now inlined, there is no perfomance
 > penalty if we move romstage away from passing pci_devfn_t to struct
 > device *. As matter of fact, for code sections that move data from
 > devicetree.cb to hardware, latter probably creates more tighter
 > assembly output.
 > 
 > https://review.coreboot.org/q/topic:devicetree

[Subrata] thanks for your CL, we can get rid of __SIMPLE_DEVICE__ once your topic branch get landed completely. Do you have an ETA when you think it can be done.

No ETA; nobody is reviewing my commits on that topic or other ones either so everything is piled up.

For now we want Coreboot_lite implementation being independent so we can speed up.

Well I think you can just skip this particular commit, I don't see why you would need to change anything connected to __SIMPLE_DEVICE__ for that.

View Change

To view, visit change 32765. To unsubscribe, or for help writing mail filters, visit settings.