SMP: Add logic to avoid a deadlock condition when CPU1 is hung waiting for g_cpu_irqlock and CPU0 is waitin for g_cpu_paused

The examples/qencoder app was trying to init the encoder by a direct call into the board, cheating in a local header to declare the normally unavailable function prototype.

Remove a assertion condition that appears to rarely cause false-alarm assertions.  Teported by Petteri Aimonen

Spinlocks:  Added capability to provide architecture-specific memory barriers.  This was for i.MX6 but does not help with the SMP problems.  It is still a good feature.

This commit adds a new internal interfaces and fixes a problem with three APIs in the SMP configuration.  The new internal interface is sched_cpu_pause(tcb).  This function will pause a CPU if the task associated with 'tcb' is running on that CPU.  This allows a different CPU to modify that OS data stuctures associated with the CPU.  When the other CPU is resumed, those modifications can safely take place.

The three fixes are to handle cases in the SMP configuration where one CPU does need to make modifications to TCB and data structures on a task that could be running running on another CPU.  Those three cases are task_delete(), task_restart(), and execution of signal handles.  In all three cases the solutions is basically the same:  (1) Call sched_cpu_pause(tcb) to pause the CPU on which the task is running, (2) perform the necessary operations, then (3) call up_cpu_resume() to restart the paused CPU.

sched/clock:  Correct calculation for the case of Tickless mode with a 32-bit timer.  In that case, the calculation was returning millisecond accuracy.  That is not good when the timer accuracy is < 1 msec.

Refresh all networking configurations due to changes in work-queue related configuration variables for Ethernet drivers.

arch/:  Add option to use low-priority work queue to all Ethernet drivers in arch that support CONFIG_NET_NOINTS.

vfork(): Fix a race condition in the SMP case.  Existing logic depended on the fact that the child would not run until waitpid was called because the child had the same priority as the parent.  BUT in the SMP case that is not true... the child may run immediately on a different CPU.

Master

STM32: STM32F303xB and STM32F303xC chips have 4 ADCs

The flag is not necessary. The state of the endpoint can be determined
using 'epstate' instead.

We need to add a delay between setting and clearing the endpoint reset
bit in SAM_UDP_RSTEP. Without the delay the USB controller will (may?)
not reset the endpoint.

If the endpoint is not being reset, the Data Toggle (DTGLE) bit will
not to be cleared which will cause the next transaction to fail if
DTGLE is 1. If that happens the host will time-out and reset the bus.

Adding this delay may also fix the USBMSC_STALL_RACEWAR in
usbmsc_scsi.c, however this has not been verified yet.