> > There might be some root hub interaction problems in the latest
> > code.  (Just like there's a compiler problem; sometime after the
> > 2.4.10 integration, someone changed those hub.c fields so the two
> > variable length bitmaps are now represented as two fields, not
> > one.  Sigh -- it was better the other way, the second field can
> > never be referenced by name.)
> 
> Do you mean the PortPwrCtrlMask fields?
> Why can't they be referenced?

Yes... all code referencing it needs to compute an
offset from the first field, based on the number of
ports the particular hub is reporting.  But the name
in the "hub.h" file has a fixed offset for that field,
which doesn't work for all size hubs.

Not that very many people access that field; it's
ignored by "hub.c" and (as I seem to recall) is
there only for backwards compat with the very
earliest USB 1.0 hubs.  (USB 1.1 fixed the bits
to be all ones.)   In the kernel, only virtual root
hubs touch that field; and user mode code can
see it through usbdevfs, but it should avoid it.


> And I am the one to blame for that update from the -ac tree.

Or maybe me, because after I carefully removed that stuff
from the AC tree for the 2.4.10 integration of usb 2.0
framework code into Linus' tree, I didn't submit a patch to
Alan to obliterate it from his tree.  Having two trees just
sucks totally ... :)

- Dave




_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to