Thanks to Pat Mochel for finding out where the error was for this bug.

Based on the 2.4 version of the driver.

One person working on a mass-storage driver (the usb protocol
side, not the block subsystem side) ran into a bug in how a
bit of net2280 dma automagic was handled.

This patch fixes it by calling existing dma chain patchup
code when the dma engine was forced to "hiccup" by having
a not-yet-valid entry in it.  The hiccup is needed in this
case since the IN data stage mustn't terminate with a
short transfer (zero length packet); but the status stage
is always a short packet.  The "terminate with short packet"
bit is endpoint state, not request state, so IN dma queues
sometimes need this kind of fixup.

This information was provided by Anthony Arkles <aja042@mail.usask.ca>.
Please apply to both 2.4 and 2.6.

into kroah.com:/home/greg/linux/BK/gregkh-2.6

This is a straightforward change matching ones sent in
for ehci (last month) and ohci (earlier this week).

It abstracts the reset operation into something that's
called before the pci glue does much to the hardware.

It also arranges to kick the BIOS off the hardware before
it resets it (not after) ... so there's no confusion at
any time about what driver "owns" that hardware.  (Again
matching what ehci and ohci drivers do.)

This fixes two small and unrelated bugs in the current OHCI code:

  - Certain initialization sequences had problems with IRQs.
    Fixed last month in EHCI, but this ohci patch didn't seem
    needed back then.  OK, so now I saw the same bug in OHCI.
    (I could believe UHCI needs it too, sigh.)

  - When restarting endpoint i/o after a queue fault, the HC
    needs to be told the control (or bulk) list filled (CLF/BLF).
    Likely this wasn't common (usbtest test10 subcase7 fault
    recovery reproduced it nicely).

Please merge.  Lack of the first one might make trouble for
some people.

This makes the new CDC Ethernet code handle more devices:

  - Uses the active config, not just the default one, if it's
    coping "descriptors in wrong place" quirk.  (bugfix)

  - Uses usb_ifnum_to_if() to get interfaces.  (bugfix)

  - AMBIT USB cable modems have bogus CDC Union descriptors;
    workaround by switching master and slave.  (add quirk)

  - To make it easier the next time we run into firmware
    that violates the class spec, add debug messages saying
    exactly why it's giving up on a given CDC device.

Net result, this code now handles at least one more
cable modem design.

Small updates:

  - don't try chiprev 0100 erratum 0114 workaround on
    newer chips; and (mostly) revert it when clearing
    endpoint halt feature.  (bugfix)

  - add missing define for the "force crc error" bit;
    I guess those #defines were generated from old chip
    specs!  potentially useful with test software.

  - sysfs register dump includes chiprev and decodes some
    of the more interesting endpoint response bits.

  - makes a sysfs "gadget" node, representing the gadget
    itself.  (decided against the class_device or bus_type
    approaches, until their value outweighs their costs.)

This goes on top of Dave Hollis' patch, and makes the front matter
match the slightly revised role -- and mention that new support.
It also eliminates a potentially confusing name and corrects an
omission (Zaurus framing wasn't printed).

This removes the compiler warning.

This removes the compiler warning from this driver.

Now it's up to the individual USB drivers to implement suspend() and
resume() if they want to.

Mahlzeit

I attached you the diff for 2.6.0-test4. It does there also one program run
without any error, but not more. I hope this issue will be resolved soon,
but I do not know yet how.

 On Fri, Aug 22, Greg KH wrote:
> On Sat, Aug 16, 2003 at 01:41:01PM +0200, Olaf Hering wrote:
> >
> > I sent you a patch for 2.4 once to make that FOO_MSR_RI, it seems that
> > was not applied to 2.6
> Care to send me a patch for 2.6 then?

How about that one:

Small kerneldoc clarifications:

  - more endpoint halt clearing info:
      * some hardware can't do it; which causes problems with drivers
        that want to use multiple interfaces or altsettings.
      * it doesn't affect queuing of data (should help usb-storage
        gadget driver, plus it's more sensible this way);
  - disconnect() callback not guaranteed: some hardware can't tell

Mostly this captures answers to questions I've been asked.

Minor cleanup.  This call no longer needs exporting for root hubs,
they have their own API to use instead.

David T Hollis wrote:
> This patch adds support to Kconfig to build ax8817x support into
> usbnet.  This renames the config option for the existing standalone
> ax8817x driver to CONFIG_USB_AX8817X_STANDALONE.

Please merge this version instead.  It includes all of David's
patch, plus it makes the descriptive info match the updated
role of this driver.  So given this, only the webpage still
needs updates; the in-tree docs are now consistent.

It needs to be "Copyright (C)" not "Copyright (c)" according to the lawyers
who know these things...

Another minor cleanup:
- if usb_register fails report back its return code rather than -1

Changes:
- if tty_register_driver report back its return code rather than -1
- if usb_register fails report back its return code rather than -1

Another better audit:
- If usb_register failes report back its return value
  rather than -1

- if usb_register failes report back its return code rather than -1

- audit usb_register in hiddev_init

- audit hiddev_init in hid_init
- audit usb_register in hid_init

- if usb_register fails report back its return code rather than 0

- if usb_register fails report back its return code rather than -1

- if usb_register fails report back its return code rather than -1
- static declarations for module_init/cleanup functions

- if usb_register fails report back its return code rather than -1