> On 04/12/2021 8:23 AM Yoshihiro Shimoda <yoshihiro.shimoda.uh@xxxxxxxxxxx> wrote: > Hmm, when we use HSCIF with 10 bits, 3000000 baud and 128 bytes FIFO, > the rx_timeout value will be set to 1536 (us). So, if we set rx_timeout > to 20000 (us) as a minimum value, the sh-sci' behavior will be back to > non hrtimer support, IIUC. > > Perhaps, describing uart_update_timeout() and the jiffies value of > uart_port->timeout with 115200 baud here may cause misreading?? > I didn't understand the purpose of uart_port->timeout yet thought. > But, at least, the current driver uses hrtimer to improve latency > for HSCIF, the driver should not set 20000 (us) as a minimum value. Not having looked at this stuff in a while, I was under the impression that the rx timeout is an error condition, when it is in fact part of normal (DMA) operation. I think it was indeed the reference to uart_update_timeout() that threw me off... So if my understanding is correct now, we should scrap the minimum timeout code entirely because the condition it is supposed to prevent cannot occur any longer due to the switch to hrtimers. Did I get that right? CU Uli
