> >> And while we're at it, it would be helpful to have a userspace > >> program to do it and return a completion code (and maybe even do > >> the CP LINK as well). > > > >I would disagree about the CP LINK part. > >More significantly, this may be an ioctl() type of call. > >Punching strings to /proc pseudo files is fine, but not always best.
> I don't mind if you disagree, but could you do it a bit sooner so > that I still remember the line of reasoning. ;-) Sorry. Also, I may have misunderstood, but I *thought* that you were suggesting that the "add device" logic would make a kernel call to [H]CP LINK of a related VM minidisk. > Clearly /proc is not the easiest interface if you want to make a > robust program to do the work. An ioctl() to issue the LINK and > update the tables would be useful. "violent agreement" is a term that comes to mind. ;-) I would suggest that (in general) whether by script or by compiled executable, the HCP LINK part (hypervisor) be handled separately from the "add device" part (Linux kernel). I would also suggest that both 'hcp' and whatever "add device" function we have do more of their work via ioctl() than via read() and write(). But I'm an idealist. > Rob
