The test for data partially transferred is incorrectec: chan->xfrd != xfrlen
Should be chan->xfrd > 0

From Ronly XLN.

OHCI:  Fix length calculation in all OHCI drivers:  CBP==0 means that the entire buffer was transferred, not that a null packet was tranaferred

In go_os_start that sets the IDLE thread stack coloration, mov does not set condition codes which are tested by the following beq.  Need to use movs instead.  Noted by David Sidrane

 kl: enable the clocks to UART1 and UART2. The previous version would cause a hard fault on startup due to the modules not being clocked.  Also drop the GPIO configuration as it's done during kl_start().  From Michael Hope.

kl: fix PWM debugging. TPM1 and TPM2 have two channels instead of six and will hard fault if you try to read the missing channels.  From Michael Hope.

Misc improvements to the CDC/ACM host driver, mostly related to parsing the configuration descriptor.  Still does not work.

Add support for the KL25Z64.  The KL25Z64 is a lower memory variant of the KL25Z128 and is used on the Teensy LC.  From Michael as SourceForge patch 50.

LPC17 USB host: Fix some compile errors when bulk endpoints + USB debug enabled.  Refresh a configuration

USB host:  Modify the transfer() and asynch() methods so that the actual size of  the transfer is returned.  Unverified on initial commit.

SAMA5D OHCI: Remove a local variable that is unused when debug is disabled; refresh a configuration.  Update a README

Olimex LPC1766STK: The USB host waiter stack should be configurable because it needs to be a lot depending when using a hub

LPC31 and SAMA5Dx EHCI drivers: Fix cache related problem.  All buffers are now aligned with the cache line size in both starting address and in length.  This cause major problems in unlucky builds where the USB host buffers where unaligned and abbutting other data.  The cache flush and invalidate operations could be subverted by acceses to adjacent data or could have unexpected side effects.  This bug has been in the ECHI drivers forever, but was only revealed due to unlucky memory allocations during the integration of the hub feature.

USB host: If the implementation of the disconnect method frees the EP0 endpoint, then it must nullify it in the port structure

STM32 USB device.  Add protection in the event that out-of-bound enpoint numbers are received.  From David Sidrane.

USB host:  Need to include the port structure when disconnecting, otherwise may destroy a root hub port

LPC17 GPIO:  Reorder steps when an output GPIO is configured in order to avoid transient bad values from being output.  From Hal Glenn.

SAM PIO: sam_pio.h needs to include chip/sam_memorymap.h in order to resolve some conditional logic properly