Re: How open is Intel?

2006-10-19 Thread Damian Wiest
On Thu, Oct 19, 2006 at 11:14:20AM +1000, Jonathan Gray wrote:

[snip]

 For the longest time it was quite hard to get documentation out
 of the networking side of Intel, but it recent years they
 publish reasonably detailed manuals for 10/100 (fxp) and
 10/100/1000 (em) controllers and some PHYs.  I have not
 been able to find any 10GbE (ixgb) manuals and suspect
 they don't publish them.  Anything at all to do with
 wireless there is no documentation from Intel at all also.
 
 fxp
 http://www.intel.com/design/network/manuals/8255x_opensdm.htm
 
 em
 http://www.intel.com/design/network/manuals/8254x_GBe_SDM.htm

Funny you should mention this as I just grabbed some docs for the 
82563EB (Intel Pro/1000).

Device driver writing is an area I haven't yet explored, but the
datasheet for the Pro/1000 looks pretty detailed and includes
block diagrams, pin descriptions, signalling, timing specs, etc.
Is this sort of document sufficient for device driver writers?

I apologize if this is a stupid question, but I really would like
to learn more about writing device drivers.

-Damian



Re: How open is Intel?

2006-10-18 Thread Theo de Raadt
 For example, recently Intel was very boastful about demonstrating
 their ``ongoing commitment to providing free software drivers for
 Intel hardware''[1].  When I first read the announcement, I was
 excited, but after re-reading it, I caught on that nowhere did they
 mention providing documentation---just an open source driver.  I
 emailed Keith Packard about this, but never got a reply. 

Keith Packard works on X, but he also works for Intel, right?  No man
can have two masters.  But he does.  You can expect that X will become
more closed.  It is becoming a serious problem -- many senior X
developers now work for video card vendors.

 I also found some technical documentation on intel.com about the G965
 chipset[2], but it does not appear complete.  It seems to explain how
 to setup DMA to communicate with the card, but not what data should be
 sent over DMA.  Of course, because of my lack of expertise in this
 field, I may just be looking in the wrong places.

That's right -- sufficient documentation has not been made available.
 
 Another example appears to be the Intel PRO/1000 MT card.  Intel has
 an open source driver for it, but when I search their web site the
 most I find are product briefs and white papers[3].  (I know the link
 is for their PRO/1000 XF card, but that is the page I was directed to
 when I clicked on ``Technical Documents'' from the PRO/1000 GT page.)

Intel, Broadcom, and Marvell are the only three vendors still left who
fail to publish documentation for their ethernet chips.  (OK, there
are a few others, but they are mostly for very rare new products, and
people like jsg will soon win them over).

These vendors do help with fixing the drivers in some limited scope.
And that's the problem.  They only help a little.  If they wrote
complete bug-free drivers and promised to maintain them for us forever
it would be one story, but that is not what is going on.

 On the other hand, there appears to perhaps be sufficient technical
 documentation on their I/O Controller Hubs for OpenBSD to support them
 soon after introduction... or maybe they are just easy to reverse
 engineer?

Whatever division makes the host bridges does a fairly good job of
releasing information.  Of course there are two documents for these.
One is the primary programming document, which they make available.
The other is an extended document for BIOS writers, and this is never
made available.  Still, the primary document is largely good enough
for our needs.  This is the only division of Intel that is open.

 So how open is Intel?  Which chipsets do they provide sufficient
 documentation to fully support?  Which chipsets do they provide some
 documentation, but omit important parts (and what are these parts)?
 And which chipsets are they completely unproviding for?

Intel only really provides the pci host bridge documentation.
Everything else is closed.

Intel hopes that if it keeps telling saying we are open that
the Linux people will keep believing them.

Let's prove them wrong.



Re: How open is Intel?

2006-10-18 Thread Jonathan Gray
On Wed, Oct 18, 2006 at 01:31:58PM -0600, Theo de Raadt wrote:
  Another example appears to be the Intel PRO/1000 MT card.  Intel has
  an open source driver for it, but when I search their web site the
  most I find are product briefs and white papers[3].  (I know the link
  is for their PRO/1000 XF card, but that is the page I was directed to
  when I clicked on ``Technical Documents'' from the PRO/1000 GT page.)
 
 Intel, Broadcom, and Marvell are the only three vendors still left who
 fail to publish documentation for their ethernet chips.  (OK, there
 are a few others, but they are mostly for very rare new products, and
 people like jsg will soon win them over).
 
 These vendors do help with fixing the drivers in some limited scope.
 And that's the problem.  They only help a little.  If they wrote
 complete bug-free drivers and promised to maintain them for us forever
 it would be one story, but that is not what is going on.

For the longest time it was quite hard to get documentation out
of the networking side of Intel, but it recent years they
publish reasonably detailed manuals for 10/100 (fxp) and
10/100/1000 (em) controllers and some PHYs.  I have not
been able to find any 10GbE (ixgb) manuals and suspect
they don't publish them.  Anything at all to do with
wireless there is no documentation from Intel at all also.

fxp
http://www.intel.com/design/network/manuals/8255x_opensdm.htm

em
http://www.intel.com/design/network/manuals/8254x_GBe_SDM.htm



How open is Intel?

2006-10-17 Thread Matthew R. Dempsky
Lately, I have been in several discussions regarding Intel's stance
towards the open source community, and the topic of providing hardware
documentation frequently arises.  However, since I am not much of a
kernel hacker, I do not have a good perspective on what documentation
is necessary.

For example, recently Intel was very boastful about demonstrating
their ``ongoing commitment to providing free software drivers for
Intel hardware''[1].  When I first read the announcement, I was
excited, but after re-reading it, I caught on that nowhere did they
mention providing documentation---just an open source driver.  I
emailed Keith Packard about this, but never got a reply. 

I also found some technical documentation on intel.com about the G965
chipset[2], but it does not appear complete.  It seems to explain how
to setup DMA to communicate with the card, but not what data should be
sent over DMA.  Of course, because of my lack of expertise in this
field, I may just be looking in the wrong places.

Another example appears to be the Intel PRO/1000 MT card.  Intel has
an open source driver for it, but when I search their web site the
most I find are product briefs and white papers[3].  (I know the link
is for their PRO/1000 XF card, but that is the page I was directed to
when I clicked on ``Technical Documents'' from the PRO/1000 GT page.)

On the other hand, there appears to perhaps be sufficient technical
documentation on their I/O Controller Hubs for OpenBSD to support them
soon after introduction... or maybe they are just easy to reverse
engineer?

So how open is Intel?  Which chipsets do they provide sufficient
documentation to fully support?  Which chipsets do they provide some
documentation, but omit important parts (and what are these parts)?
And which chipsets are they completely unproviding for?

Thanks.

[1] http://lists.freedesktop.org/archives/xorg/2006-August/017404.html
[2] http://www.intel.com/design/chipsets/datashts/313053.htm
[3] 
http://www.intel.com/network/connectivity/products/pro1000xf_server_adapter_docs.htm