Subject: myrinet drivers
To: None <email@example.com>
From: None <firstname.lastname@example.org>
Date: 04/11/2003 09:54:04
So, if the code produced by Myrinet is not compatible with the NetBSD
kernel, it would seem that we have to write one ourselves.
Is there anyone with experience with Myrinet boards who would be
interested in such a project?
Is there anyone without experience with Myrinet boards, but with
NetBSD device driver experience, who would be interested in learning
enough about the boards to work on a driver?
If I were to offer a prize for successful completion and integration
of a driver and packages (i.e., pkgsrc) of associated software, would
anyone be interested?
Although I am not really a device driver programmer, it appears that
the relevant information for the hardware is available at:
Essentially, the boards provide a heap of memory, a RISC processor for
which there is a compiler available, a DMA interface to the PCI bus,
and a packet interface to the network. I'm guessing that the driver
would need to define some ioctls for retrieving/modifying information,
for mapping board memory into the process and/or kernel space, and for
downloading an image for the RISC processor to run; presumably that
image would not need to be part of the kernel and could include code
provided by Myrinet. I would hope that licensing issues would not
prevent that use of the code. (Right or wrong?)
The other aspect I can identify at this point is that the higher level
software (e.g., the GM system) would have to be modified to issue the
right set of ioctls for the driver. My survey of the code suggests
that much of this is rather straightforward, as it involves lots of
little functions, each intended to handle one piece of information.
Replacing the function bodies with appropriate ioctls or whatever
would be reasonably straightforward. I also think this code could lie
outside the domain of NetBSD, even in pkgsrc, so licensing shouldn't
be a problem.
Now is the moment for all of you who know a lot more about this than
me, which is likely everyone, to correct my mistaken assumptions. I
hope, though, that at least one of you is interested enough to step up
and try to make this happen. As I said, I am willing to offer a
nominal prize if that would spur on the activity.
Thanks for your help.