[RFC linux 1/2] drivers: fsi: Add SBE client device driver

[Alistair Popple]

On Tue, 13 Dec 2016 10:37:53 AM Joel Stanley wrote:
> Hi Eddie,
> 
> On Tue, Dec 13, 2016 at 10:05 AM, Edward James <eajames at us.ibm.com> wrote:
> > Yes, the OCC driver will depend on the SBE driver for P9. I suppose the 
OCC
> > driver could handle all the SBE register access over FSI itself... but not
> > sure if that's a good approach.
> 
> Can you expand on why you think this not a good approach?
> 
> What abstraction does the SBE driver provide that the OCC driver
> benefits from? In other words, how does the code change in the OCC
> driver if we don't have a SBE driver in between?

I think it may have been mentioned earlier but it's probably not necessary to 
have drivers that simply provide read/write methods to the FSI address spaces 
at different offsets as these just add code to the kernel for no real benefit 
as it's easy to add those offsets to a /dev/cfam device access for example.

The thing the kernel can provide that userspace can't easily is atomic/shared 
access to multiple registers. So for example I seem to recall there was a 
discussion somewhere on using a FIFO to communicate with the SBE. It may make 
sense in this case for a kernel SBE client driver to provide read/write calls 
that would access that FIFO, but any userspace client that just needs to 
read/write an FSI/CFAM address should just do that via a single 
chardev/driver.

- Alistair

> Cheers,
> 
> Joel
> _______________________________________________
> openbmc mailing list
> openbmc at lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/openbmc
>