Re: Interface for SCSI transactions ?

[olafBuddenhagen]

Hi,

On Wed, Sep 07, 2011 at 11:51:15PM +0200, Thomas Schmitt wrote:

> The idea came to me to have a generic RPC with two-way parameter
> transmission and a function code. Similar to the job of Unix ioctl().
> It would reduce future work when other new kernel calls shall be
> implemented.

No no, that wouldn't be good. ioctl() is basically a generic kernel RPC
mechanism -- but we already have one! :-) No need to implement RPCs on
top of RPCs... Defining special-purpose RPCs as we need them is The
Right Way (TM).

Also we will be able to reuse this RPC directly when we implement the
userspace drivers; only substituting the device parameter with a file
one :-)

> Actually libburn is a userspace driver with a library API. This has
> advantages, like a development cycle without reboot, and it has
> disadvantages, like udev beginning to show allergic reactions when my
> programs burn on DVD+RW. (As driver of the drive, i really could need
> reliable locking.)

Well, a mechanism to temporarily claim exclusive control of the device
may indeed be desirable -- but that's really orthogonal to the question
whether the driver runs in kernel space or not :-)

-antrik-