On 27 November 2014 at 20:56, Paul Walmsley <paul@xxxxxxxxx> wrote: > On Tue, 25 Nov 2014, Rafał Miłecki wrote: >> I understand your arguments against drivers/soc/, but on the other hand >> I have no idea where else this driver could go. > > After looking around the tree to find out where similar code is located, > it looks like drivers/firmware is the right place. These days, > drivers/firmware is mainly used for drivers that parse EFI bootloader > data, DMI data, that sort of thing. Quite similar to the CFE-provided > data that the bcm47xx-nvram code deals with. So, by functional analogy, > drivers/firmware appears to be the right place to stash this device > data-probing code. > >> I guess DT is older than CFE, but Broadcom decided to invent own >> solution called NVRAM anyway. This is a bit messy, because it actually >> stores hardware details (CPU, RAM, switch) as well as user settings >> (e.g. LEDs behavior). I can't say why Broadcom decided to implement it >> this way. > > Yep, based on what the other drivers in drivers/firmware are used for, I > think drivers/firmware is the right place for the CFE parsing code. The problem is I can't find MAINTAINER of the drivers/firmware/. Is there someone responsible for that? Some mailing list maybe? Who could give us an ACK to move bcm47xx_nvram there? >> > It sounds to me like this code is a combination of three >> > pieces: >> > >> > 1. code that autoprobes the size of the "nvram" partition in the Broadcom >> > platform flash, by reading various locations in the MMIO flash aperture, >> > configured by some other system entity >> >> That's right, on MIPS we simply detect flash type (drivers/ssb & >> driver/bcma) and using that we init NVRAM passing memory offset where >> the flash is mapped. > > OK. > > So (as a side issue), I would suggest that when you move this code out of > arch/mips, the MIPS-isms in it should be removed, like KSEG1ADDR(), etc., > and replaced by the standard ioremap()-type approach. After all, Broadcom > could build CFE for ARM, and then we'd want to use this same code to parse > the CFE-provided data. > > Also I would suggest getting rid of the #ifdefs for the flash type, and > probing it dynamically instead. The flash setup code under drivers/ssb/ > and drivers/bcma/ sets up platform_devices for the flash, right? If so > then it would be best if this code could run after the bus setup code, > query the Linux device model for the type of platform flash in use, and > then extract the appropriate address space to probe from that data. I'm pretty sure you look at some old version of arch/bcm47xx/nvram.c. I wouldn't dare to move such a MIPS-focused driver to some common place ;) Please check for the version of nvram.c in Ralf's upstream-sfr tree. I think you'll like it much more. Hopefully you will even consider it ready for moving to the drivers/firmware/ or whatever :) -- Rafał
