Hardware • Re: Cherry keyboards and device descriptor read/64, error -71

• device descriptor read/64, error -71:

  On Mon, 27 Dec 2004, Christian Heimanns wrote:
  
  > Thanks for the tipp: I've compiled the USBcore as a module now. Before
  > it was compiled into the kernel. In the file /etc/modprobe.conf I've
  > added the line: > options usbcore old_scheme_first=y <.
  > It seems to work now Can you please explain in a few sentences the
  > difference between the "new" and the "old" USB scheme?
  
  I'll try...
  
  The "old scheme" is the way Linux worked before 2.6.10. When a new device
  is plugged in, the system first assigns it an address, then reads the
  initial 8 bytes of the device descriptor, then reads the entire 18-byte
  device descriptor.
  
  The "new scheme" is basically the way Windows works. (Not surprisingly,
  some devices won't work any other way.) When a new device is plugged in,
  the system first issues a 64-byte read request for the device descriptor,
  then resets the device, then assigns it an address, and then reads the
  actual 18-byte device descriptor.
  
  The reason for these shenanigans is that with a full-speed device, the
  maximum packet size for endpoint 0 (ep0maxpacket) isn't known beforehand.
  It could be 8, 16, 32, or 64 bytes. (Low-speed devices must use 8, and
  high-speed devices must use 64.) The ep0maxpacket value is stored in the
  initial 8 bytes of the device descriptor, but to read the device
  descriptor you have to use endpoint 0!
  
  The two schemes above are the two common solutions to this chicken-and-egg
  problem. The old scheme is the one recommended by the USB Implementors
  Forum (which makes it the "Standard"); the new scheme is the one used by
  Microsoft (which makes it another kind of "Standard"!). A well-designed
  standards-compliant device will work okay with either scheme.
  Unfortunately it seems that no matter which scheme you pick, some
  badly-designed non-compliant devices won't work. There's an additional
  usbcore.ko module parameter people can use in especially bad cases:
  
  use_both_schemes=y
  
  This will cause the system to try one of the schemes, and if it fails then
  try the other scheme. (Maybe that should always be the default...)
  
  Alan Stern

Statistics: Posted by Aki — 2025-03-02 14:32