Re: [PATCH-for-5.2 0/4] hw/char/serial: Use the Clock API to feed the UART reference clock

[Philippe Mathieu-Daudé]

On 9/1/20 7:02 PM, Paolo Bonzini wrote:
> On 06/08/20 15:03, Philippe Mathieu-Daudé wrote:
>> This series improve tracing of multiple UART device in the same
>> chipset, and allow to use the Clock API to feed each device with
>> an (updatable) input clock.
>>
>> Based-on: <20200730165900.7030-1-philmd@redhat.com>
>> "hw/char: Remove TYPE_SERIAL_IO"
>>
>> Philippe Mathieu-Daudé (4):
>>   hw/char/serial: Replace commented DPRINTF() by trace event
>>   hw/char/serial: Remove old DEBUG_SERIAL commented code
>>   hw/char/serial: Let SerialState have an 'id' field
>>   hw/char/serial: Use the Clock API to feed the UART reference clock
>>
>>  include/hw/char/serial.h |  4 +++
>>  hw/char/serial.c         | 55 +++++++++++++++++++++++-----------------
>>  hw/char/trace-events     |  5 ++--
>>  3 files changed, 39 insertions(+), 25 deletions(-)
>>
> 
> Acked-by: Paolo Bonzini <pbonzini@redhat.com>
> 
> Are you planning to deprecate the baudbase property, and instead setting
> up the clock already in serial_mm_init?

I'd love to get there, but unfortunately most of serial_mm_init() have
the baudrate/frequency argument wrong, because it has never been very
clear it was the uart input clock, so most users sets the default
baudrate expected by their guest.

Maybe a sane way to get this slowly fixed is to add a new
serial_mm_init2() documenting it expects an input clock rate
argument (or even better, a connected clock), use it where possible,
deprecate serial_mm_init(), let the maintainer eventually adapt...

I'm not comfortable adding yet another deprecation that will stay
during years.

> 
> Thanks,
> 
> Paolo
> 
>