Re: serial8250: can not change baudrate while the controller is busy

["Ilpo Järvinen" <ilpo.jarvinen@linux.intel.com>]

On Fri, 14 Apr 2023, qianfan wrote:

> Hi:
> 
> My custom board is based on allwinner R40, the uart is compatibled with
> serial8250. Based on it's datasheet:
> 
> > When TX transmit data, or RX receives data, or TX FIFO is not empty, then
> the
> > BUSY flag bit can be set to 1 by hardware, which indicates the UART
> > controller is busy.
> 
> We cannot write LCR and DLL to update UART params such as baudrate and partity
> while the UART is busy, however `serial8250_do_set_termios` is a void
> function,
> the upper level always assume the uart params is updated.
> 
> The upper level `uart_set_termios` do noting if ktermios params is not
> changed,
> it will not update when the user space program running tcsetattr set a same
> baudrate again.
>
> So we can not fix the baudrate when
> `serial8250_do_set_termios`
> failed.
> 
> Allwinner R40's datasheet provided a way for this case.
> 
> > CHCFG_AT_BUSY(configure at busy): Enable the bit, software can also set UART
> > controller when UART is busy, such as the LCR, DLH, DLL register.
> > CHANGE_UPDATE(change update): If CHCFG_AT_BUSY is enabled, and CHANGE_UPDATE
> > is written to 1, the configuration of UART controller can be updated.
> > After completed update, the bit is cleared to 0 automatically.
> 
> I can't know this feature is expanded by allwinner, or it is a common
> functiton
> of serial8250. Perhaps the serial8250 driver need this.

tcsetattr() can be given a flag which enforces TX empty condition before 
core calls into the lower layer HW set_termios function. Would that be 
enough to solve the case you're interested in?

Obviously, nothing can prevent Rx from occuring as it's not under local 
UART's control (e.g. a busy flag check would still be racy). But does 
writing those registers actually break something or just corrupts the 
character under Tx/Rx (which can be handled by flushing)?

-- 
 i.