Intel Controllers [Rant]

[Edward O'Callaghan]

On Tue, 10 Dec 2013 12:16:57 -0600
Mark Felder <feld at FreeBSD.org> wrote:

> 
> On Dec 10, 2013, at 12:13, Jim Thompson <jim at netgate.com> wrote:
> 
> > On Tue, Dec 10, 2013 at 11:47 AM, Mark Felder <feld at freebsd.org>
> > wrote:
> >> On Mon, Dec 9, 2013, at 18:20, Barney Cordoba wrote:
> >>> Why is it that every time I get a new MB it has yet-another intel
> >>> controller on it? How are you supposed to write good drivers that
> >>> support 9000 different controllers? As much as I appreciate the
> >>> progression of CPU technology, they
> >>> would serve the community better by making up their damn minds
> >>> and just build 1 or 2 controller for each
> >>> target market.
> >>> 
> >> 
> >> I'm pretty sure an entirely new driver doesn't need to be written
> >> for each new controller. I assume like with most things (network
> >> cards, wifi chips, etc) it's usually a matter of adding its
> >> identifier to the driver code.
> > 
> > It goes beyond that.   In the simplest case, then yes, things are
> > as you say.
> > 
> > But chips have bugs, which require work-arounds, or other slightly
> > strange stuff.
> > 
> > Case(s) in-point:
> > 
> > The 82575 assigns queues using vectors via a bitmask
> > The Intel 82576 chip uses a table that essentially consists of 2
> > columns with 8 rows.  The ordering is column-major (like Fortran
> > arrays).
> > On 82580 and newer adapters the scheme is similar to 82576, however
> > instead of ordering column-major, the ordering is row-major (like C
> > or Pascal).
> > 
> > These chips are all supported via the igb driver.
> > 
> > Some parts supported by the igb driver use MSI-X interrupts, others
> > don't.
> > 
> > On i350, i354, i210, and i211, loopback VLAN packets have the tag
> > byte-swapped.
> > 
> > The list of issues is by no means limited to the above.
> > 
> > Jim
> 
> That sounds tedious. Thanks to those of you that tirelessly hack away
> at these quirks.
> 
> When I first replied to this post I thought I was reading another
> list and the discussion was about disk controllers which don't seem
> to be so troublesome
> 
> 

Another thing that can happen, given some hw rev. is that you in one
rev. if you write values into registers in the wrong order you can
actually get the hardware into undocumented states. This ends badly,
typically locking up the hardware requiring soft-resets, although on
occasion that may not even be sufficient ! Other times you may need to
wait some time after banging on a register, typically overly long
periods with Intel gear. In some cases you can actually irreparable
damage the hw just because you did not wait and/or order some register
twiddling correctly O_o

Basically x86 sucks. Both Sun, IBM & Sgi designed things correctly with
things like PROM's. The Intel UEFI firmware attempts to replicate these
late 80's, early 90's ideas and fails horribly. The only hope aside
from IBM POWER has been left with ARM, however that is quickly becoming
a fragmented mess..

Cheers,


-- 
______________________________________________________________________

Edward O'Callaghan, Director and Principal consultant.

https://www.alterapraxis.com | eocallaghan at alterapraxis.com

Altera Praxis Pty Ltd,  // Discretions in cybersecurity
ACN 165 424 064.       // reconnaissance and adversary mitigation.
______________________________________________________________________
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-net/attachments/20131212/a71e68c5/attachment.sig>