Q: MTD and NIC Roms...

[ebiederman@lnxi.com (Eric W. Biederman)]

Jeff Garzik <jgarzik@pobox.com> writes:

> Eric W. Biederman wrote:
> > Currently I have a patch to eepro100.c that adds an MTD map driver so
> > the onboard rom can be written.  Making code like etherboot easier to
> > flash etc.
> [...]
> > I am currently looking for ideas on ways to cleanly get this code
> > into the kernel, and I am looking for ideas.  The map driver is
> 
> 
> Well... this functionality has existed for a while, and it doesn't need to be in
> the kernel :)
> 
> Donald Becker's diag suite can do flashing. ftp://www.scyld.com/pub/diag/  He
> provides means to program the flash from userspace.

Not on the eepro100, it does look like one or two other kinds of nic
are supported though.  His libflash.c is quite deficient when it comes
to the number of flash chips supported, the correctness of the
implementation of the cfi command set 2, and the completeness of it's
probe routine.

None of which goes into the races, or the portability problems
that arise from doing this in user space.

The linux mtd layer with it's larger user base, and the fact it sits
in the does not have any of those problems with handling flash chips.
And it steadily gets fewer problems as more kinds of flash chips are
looked at, and the problems in the code are addressed generically.

> And I think that's the best place for it.  We _could_ bloat up the kernel code
> by adding the ability flash -- but how many users is that going to serve, that
> are not already served by existing programs?  So, I disagree with getting this
> stuff into the kernel at all.

Given the lack of existing programs for the eepro100 every user served
is a new one.  Plus with the better support libraries provided by the 
linux mtd layer it is easier to do a quality job in the kernel.

I totally agree that this is not day to day functionality, and so it
should not burden the fast common paths of the kernel.  The code was
enclosed in a config option.  It is worth noting one of the busiest
booths at LinuxWorld was the etherboot booth.  And by other counts
as well there are quite a large number of users network booting.   So
the potential user base is significant.

And as David said it really is not that much code.

Eric