If we do not yeld in ISR when we write to queue/give semaphore, the scheduler will not know of the change and will not check the queue untill the next OS tick. This change causes the task to be called immediately and makes communication many times faster.
In some rare ocasions (bad cable, noise, etc.) data transfer might timeout and hang the endpoint, unless the interrupt flag is cleared.
This pull request targets to solve that case.
ESP32-S2 has only 5 available endpoint-in FIFOs (including EP0) but 7 available EP IN numbers. This change decouples the fifo number from the endpoint number, providing FIFO numbers until they reach the limit, at which point it will return false and assert an error that too many endpoints were allocated.
me-no-dev