Messages in this thread | | | Date | Fri, 18 Jul 1997 17:55:19 -0400 (EDT) | From | linux kernel account <> | Subject | Re: The i2o Bus: A Conspiracy Against Free Software? (fwd) |
| |
What if the OS&Drivers have the responceibility of programming the I2O chip? also, does the I2O spec tie you into using the i960? Assuming someone clean roomed the I2O spec, could I produce a MB that used a TI-C6 DSP, A sharc, or a 68K coldfire? and have any OS look on it the same as the i960 in a Intel MB?
What I2O boils down to is offloading the host CPU, it's a device-coprocessor. Many devices already have their own processors, and since PCI devices can access the entire memory space, I dont see any reason why a combination of good hardware and drivers couldn't accomplish the same thing.. (See SCSI cards that do mailboxing, like my BT-948 with it's spiffy 25mhz 80186) So, this seems to me as a crutch which limits choice and flexibility.
I think Intel figured out they made a mistake making PCI largly platform independant... Would you have believed it in the 80's if someone told you that you could use the same videocard in a mac, a dec supercomputer, and a Pc? Now, Intel regrets it and they are trying to fix it.. Would someone be so kind as to create an anti-I2O mailing list and subscribe me?
Thanks, Gregory Maxwell
On Fri, 18 Jul 1997, Benjamin Saller Bender wrote:
> Dan Hollis <mailto:goemon@sasami.anime.net> writes: > >Someone should make a comprehensive web page about why I2O is wrong, why > >the NDA is wrong, the real reason behind the NDA, why I2O is nothing new, > >compare I2O to MCA (that should scare a lot of management types), etc. > > > As I understand it, and I've only glanced at the design overview, I2O > is designed so that hardware vendors can write drivers to a layer falling > between the OS and the devices. This layer is controlled by a i960 running > something along the lines of vxWorks. The OS (linux in this case) only needs > to be able to talk to the interface exported by the i960. We don't need the > detailed specs of the underlying communications between the i960 and the > device being controlled. > So you say, how do we obtain the protocol to talk to the generic > communications layer? Here is my (maybe naive) thought on the matter. If we > don't get it from the hardware vendors there sales suffer. Thats the bottom > line. Linux users and the other free OS are no longer totally ignorable from > the view of the all powerful dollar. It is in the best interest of the > hardware vendors to make there drivers work on a number of OS, even if M$ gets > a little peeved. The cost of driver development for hardware vendors drops if > they only need to write it once for it to work on 95/98/NT, OS2, Linux, etc. > If this is the way of things they will have to make the spec to talk to said > drivers public. >
| |