Hi, On 15.11.23 07:50, Guntupalli, Manikanta wrote: >> From: Lino Sanfilippo <LinoSanfilippo@xxxxxx> >> >> So, this will be executed each time (including the rts_before_send delay) the >> core wants to send data? This is not how it is supposed to work: The tx setup >> (and the delay before send) has to be done once when transmission starts. >> Note that when sending a bulk of data the core may call cdns_uart_start_tx() >> several times before it eventually calls cdns_uart_stop_tx() to stop the >> transmission. > We have tested bulk transfer (5000 bytes) and observed that cdns_uart_start_tx() is getting called only once. Can you please suggest how to reproduce the case where the core may call cdns_uart_start_tx() several times before it eventually calls cdns_uart_stop_tx(). Thats strange. Normally the uart_ports start_tx() function is called whenever there is new data in the tty circular buffer (see uart_write()) and the writing process is suspended after that (see n_tty_write() in case of N_TTY line discipline). The writing process is woken up via uart_write_wakeup() called from the uart driver and then it writes new data into the circular buffer which results in another call to the uart_ports start_tx(). There is no stop_tx() until all data from the passed userspace buffer is written. But there is a start_tx for every new bulk of data that is available in the circular buffer. This is at least what I can observe in my test setup (using a PL011 UART with the amba driver). If I write a test buffer of 9212 bytes to the tty device using one single write() I can see 10 consecutive calls of tx_start() before tx_stop() eventually is called. What does your test setup look like? Regards, Lino
