libusual interface [was Re: libusual and ub]

[Pete Zaitcev]

On Thu, 15 Dec 2005 09:00:33 -0500, Jeremy Katz <katzj at redhat.com> wrote:

> Hey, to be fair, my patch was just probing for the devices once ub
> already got loaded.  I ignored the question of "how does ub get loaded"
> since I can guarantee that it happens in anaconda.

BTW... Just to make sure that we're on the same page, if there's no
working hotplug, you have to load both ub and usb-storage, because
with libusual there is no way to know which is required for the particular
device. In most cases, usb-storage is not needed, and it's a waste of
about 30KB of RAM or so. But I see no (easy) way around it.

> [...] So if we're going to build libusual
> in statically, having the sysfs method to change the bias becomes
> incredibly important (... as then I can just write to the file from the
> loader and not worry about it)

OK, that was a half an hour job. See attached, I'll ask Dave to include
it and send it to upstream now. Note that there's no bind/unbind here,
so it has to be set before the back end drivers are loaded.

Out of curiousity, what part of the code makes these decisions:
 a) If to load USB storage drivers at all, whatever they are
 b) What of the available CD-ROMs has the ISO
I would like to look.

With ub, it is not obvious that the media is indeed a CD-ROM (or a DVD).
Since ub issues no MODE SENSE command, everything is an array of blocks.
So, suppose that everything works, how would you know that we're
installing from /dev/uba or whatever? Do you try to read a header
from everything that looks vaguely like a block device? That may cause
interesting effects if you are attached to a SAN, so obviously you do not.
Very curious.

-- Pete