Re: [9fans] TODO lists for Plan 9

[Nigel Roles <nigel@9fs.org>]

lucio@proxima.alt.za wrote:

>>I think OHCI and UHCI were competing USB 1 interfaces.  One was from
>>Intel and appears on lots of motherboards; I'm guessing that that's
>>UHCI, since that's what Plan 9 currently supports.  EHCI must be the
>>USB 2 interface I was thinking of.
>>
>>
>
>I'm sure there are others whose knowledge is greater than mine, but
>they must be asleep as no one has shouted at me yet :-)
>
>I have both UHCI and OHCI specs (and EHCI, believe or not).  I haven't
>yet been able to read both (either, really, I only superficially read
>the UHCI one) and so I don't know how they differ: Universal (Intel)
>and Open (Compaq, of all names, but all USB designers at the time seem
>to have had a hand in this one).  I have a feeling that OHCI has
>superseded UHCI in practice, possibly because of its appearance in all
>EHCI implementations.  But I'm guessing and my experience is not very
>broad.
>
>As for EHCI, by volume the specification does not seem much greater
>than OHCI, but I'm sure there lurks some serious complication
>somewhere in its innards.
>
>
>
OHCI has not superceded UHCI. It's simply that UHCI came first and was
designed by Intel so appeared in the chipsets first. OHCI came later,
and took
longer to make silicon. VIA in particular made OHCI based chips, so bit
by bit OHCI took off. It is a better specification, requiring less
activity in
the interrupt handlers to decide what to do. Also, bandwidth limiting
(if you
want to get into that) is easier to do with OHCI. Or, rather, UHCI is
missing
a level of sophistication compared to OHCI, so the first step in writing a
UHCI driver is to introduce some OHCI-like structures to make up the
deficit.

There is no fundamental reason why UHCI style silicon could not support
USB 2.0, if you turn the clock speed up. I imagine, though I don't
know, that those organisations keen to promote USB 2.0 just preferred
to develop OHCI because the already had silicon IP in that style.

I've not looked at EHCI, though I doubt there is much which changes
things.

I've written an OHCI driver for non PCI (i.e. embedded) silicon for an
operating system not unlike Plan 9; it's not mine to give away. The first
thing I would lay my hands on is a USB bus analyser. Really, honestly,
you will save an order of magnitude in development time and your sanity.

Despite the bad impression early drivers gave (i.e. Win95), USB is
easy. It does what is says on the tin, and there are few gotchas. The
bus just works.

However handling hot plugging, and attachment of device specific
drivers in a manner which is bomb-proof is harder. Unless things have
changed recently, the Plan 9 approach with the namespace and usbd
is too simplistic on account of having been developed in the early days of
USB when things were very clean.

Choosing which of the multiple interfaces and configurations
available to use is unfortunately unavoidable, so I have felt for a time
that a new driver architecture is required, whether or not the underlying
hardware is UHCI, OHCI, EHCI, or just ICKY.