bjorn at mork.no
Fri May 9 06:07:05 PDT 2014
Dmytro Milinevskyy <dmilinevskyy at sequans.com> writes:
> That said nobody is interested in my patches for kernel-space driver?
I can only speak for myself...
I think the changes to usbnet are interesting, but they are also the
part that probably needs most work to complete. There are likely a
number of corner cases which must be tested and verified, expecially in
the suspend/resume paths. I don't claim to have full overview of this.
But if you decide to drop the character device part, then there aren't
any users of the facility however interesting it is.
As I said before you started this work: I do not think it is a good idea
to implement the character device mapping in the kernel, as long as this
can be done perfectly well in userspace using the existing kernel APIs.
And I said that already having written and thrown away a couple of such
implementations... None as nice and complete as yours, though, but still.
There is no shortage of APIs letting you create a pipe to a netdev. An
af_packet socket is only one of them. You can also bridge to a tap
interface for example. Take a look at your favourite hypervisor
environment and I'm sure you'll find lots of ways such pipes can be
implemented in userspace.
More information about the libmbim-devel