[Libusb-devel] Windows libusb0.sys back end considerations.

[Graeme G. <gr...@ar...>]

Thought I'd summarise some of my recent discoveries with respect to
the MSWindows back end & libusb0.sys:

  * CancelIoEx (used in poll_windows.c on Vista/Win7) seems to be unreliable
    (at least with the libusb0.sys back end).
    I've had to disable it in my code. I don't know why that is, although this
    may be a hint: <http://msdn.microsoft.com/en-us/library/aa480216.aspx>
    "Note that this does not guarantee that these requests will be honored by underlying
     drivers. Drivers that can take a long time to complete create requests must implement
     cancellation support. We developed driver I/O completion/cancellation guidelines as
     well as a new option for Vista's driver verifier v2 to support testing completion and
     cancellation behaviors (see Driver Hang Verifier in Related Links)."

  * The original libusb0.sys has it's set config on start code commented
    out for compatibility with running as a filter driver. This makes it
    (and copies of it such as the signed ptlibusb0.sys) not work
    with the MSWindows back end, since it assumes that the device has been
    configured.

    I had to introduce a patch to work around this problem, and get
    the MSWindows back end working with the signed ptlibusb0.sys.
    (patch is attached).

    (The libusb0.sys created from the source of my mod. does do
     a setconfig, so does work without a patch).

  * The ptlibusb0.sys file itself isn't signed, so won't work
    on its own with Vista/Win7 64 bit. This is because it's the catalog
    file that comes with it that's signed. It turns out that
    if you use that catalog file with the ptlibusb0.sys driver,
    Vista/Win7 64 bit will load the driver, even though the
    catalog signature doesn't match any of the the other files (.dll, .inf etc.).
    (This seems an obscure fact about Microsoft's code signing).

Graeme Gill.