[Prev][Next][Index][Thread]
Re: Quickcam device driver for Linux
> >The device driver is either compatible or minimal in this case, but not
> >both.
> Huh?
All I meant: the more compatibility (i.e., API features) you provide, the
bigger the driver gets, especially if it has to start buffering and
translating 300K images. There are a lot more resources available to user
space than to kernel space.
I would not put Quickcam code inline with my kernel because it's big, it's
slow, and it's often buggy (and perfecting the software only fixes one of
those problems). I would imagine that there are similar reasons that we
don't see kernel devices for specific printers or SVGA cards. Imagine
/dev/matrox, a 3MB X11 server, in kernel memory.
As much as I hate the portability headaches, I would support an effort to
create a kernel driver to pass bytes directly (no buffering, translation,
etc) from the camera to a read() call. Any (de)compression, dithering, or
resizing should be done in userspace, where malloc() is cheap and
processes can be preempted.
--Patrick
Follow-Ups:
References: