This ought to fix the NS9750 init issue, and make the AMD756 case
at least somewhat better.  It makes the init go "by the book" in more
ways, and formalizes one quirk.

Various OHCI init/reset cleanups for different silicion environments:

 - Reset a bit more "by the book".

    * Define a new quirk flag for the SiS and OPTi problem seen earlier.
      Since 2.4 we've always worked around that quirk, even though we've
      not seen that on other chips; but it's "wrong" and doesn't work on
      some chips (notably NetSilicon NS9750).

      The quirk still seems to be needed for SiS, but either this test
      machine is too fast for the OPTi problem to show up, or the frame
      timing setup problem there came from a now-fixed bug.

    * Look at the HC state before resetting it; depending on whether
      it was previously owned by BIOS, SMM, an OS, or nobody, different
      USB signaling (and timings) might be needed.

    * Re-init the frame timings right after soft reset, rather than
      later (potentially too much later).

    * Restore a reset in the PCI startup code, so this logic more closely
      resembles the non-PCI paths (future code sharing).  It also makes it
      easier to guarantee a 1-millisecond ceiling between reset and "go".

      An earlier reset is being done to help workaround BIOS-related
      problems on some boards, but we may need an even earlier one
      (as a PCI quirk, before IRQs get reconfigured).

 - Add an explicit #define to disable the BIOS/SMM handoff; it's
   not just HPPA, many embedded chips don't expect BIOS either.

 - There are reports of AMD 756 machines disliking the OHCI suspend
   patch of a few months back.  Erratum #10 partly explains that, so
   now root hubs won't autosuspend on those Slot-A era chips.

 - Other minor fixes

    * We've got lots of non-PCI OHCI now too, so comments shouldn't
      be assuming all-is-pci!

    * Hey, it's unsafe to call hc_reset() in IRQ (after unrecoverable
      error); so just force a soft reset, don't do the whole thing.

Tested on half a dozen different OHCI versions, but maybe some other
versions of OHCI will be sensitive to one of these changes.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>