Hi Rafał.

On Sun, 23 Nov 2014, Rafał Miłecki wrote:

> After Broadcom switched from MIPS to ARM for their home routers we need
> to have NVRAM driver in some common place (not arch/mips/).

So this is a driver for a chunk of NVRAM that's embedded in the SoC, 
hanging off an I/O bus?

Is this actually some kind of RAM, or is it flash, or something else?

> We were thinking about putting it in bus directory, however there are
> two possible buses for MIPS: drivers/ssb/ and drivers/bcma/. So this
> won't fit there neither.
> This is why I would like to move this driver to the drivers/soc/

Can't speak for anyone else, but I personally consider drivers/soc to be 
primarily intended for so-called "integration" IP blocks.  Those are used 
for SoC control functions.

So low-level NVRAM drivers would probably go somewhere else.  Here are 
some likely possibilities, where it can live with others of its kind 
(assuming it is something similar to RAM):

1. the PC "CMOS memory" NVRAM driver appears to be under drivers/char, as 
drivers/char/nvram.c.  

2. there's a generic SRAM driver, drivers/misc/sram.c

3. while people have put some of the eFuse code under drivers/soc, in my 
opinion, the low-level fuse access code should really go under 
drivers/misc/fuse.

... 

Looking at arch/mips/bcm47xx/nvram.c: if the low-level NVRAM probe code 
were moved elsewhere, the higher-level NVRAM "interpretation" functions 
still remain: bcm47xx_nvram_getenv() and bcm47xx_nvram_gpio_pin().  Those 
seem to be intended to parse device configuration data, yes?  And this 
device configuration data is organized this way by platform software 
convention -- there's no hardware requirement to store data this way in 
the NVRAM, right?

Unfortunately the way that we organize device data parsing code in the 
kernel is quite frankly ... anarchic.  In some kind of ideal world, we'd 
have a standard place for device data storage and parsing functions, and 
any combination of DT/ACPI/pdata/PCI/ISAPNP/whatever could be placed in 
use, depending on the software environment.  But instead we have 
directories like drivers/acpi and drivers/of and drivers/base/platform and 
drivers/pnp and drivers/platform/chrome.

So if the answer to the above two questions is "yes," meaning that this 
NVRAM is used to store device probing data by software convention, then it 
would seem to me that proposing some place like 
drivers/devicedata/bcm47xx-nvram is a better bet (where "bcm47xx-nvram" is 
just some kind of opaque token).  Looking at those two functions, it 
doesn't seem like there's really anything MIPS or Broadcom or NVRAM or 
even SoC-specific about key-value pair parsing and GPIO device data?  Or 
am I missing something?

Actually, considering that bcm47xx_nvram_gpio_pin() isn't even used, can 
we just drop it and save the hassle?  :-)


- Paul