Re: [PATCH v3 2/2] serial: 8250: Apply FSL workarounds also without SERIAL_8250_CONSOLE

[Ilpo Järvinen]

On Mon, 5 Jun 2023, Uwe Kleine-König wrote:

> Hello Ilpo,
> 
> On Mon, Jun 05, 2023 at 04:44:08PM +0300, Ilpo Järvinen wrote:
> > On Mon, 5 Jun 2023, Uwe Kleine-König wrote:
> > > On Mon, Jun 05, 2023 at 04:22:55PM +0300, Ilpo Järvinen wrote:
> > > > On Mon, 5 Jun 2023, Uwe Kleine-König wrote:
> > > > 
> > > > > The need to handle the FSL variant of 8250 in a special way is also
> > > > > present without console support. So soften the dependency for
> > > > > SERIAL_8250_FSL accordingly. Note that with the 8250 driver compiled as
> > > > > a module, some devices still might not make use of the needed
> > > > > workarounds. That affects the ports instantiated in
> > > > > arch/powerpc/kernel/legacy_serial.c.
> > > > > 
> > > > > This issue was identified by Dominik Andreas Schorpp.
> > > > > 
> > > > > To cope for CONFIG_SERIAL_8250=m + CONFIG_SERIAL_8250_FSL=y, 8250_fsl.o
> > > > > must be put in the same compilation unit as 8250_port.o because the
> > > > > latter defines some functions needed in the former and so 8250_fsl.o
> > > > > must not be built-in if 8250_port.o is available in a module.
> > > > > 
> > > > > Acked-by: Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx>
> > > > > Link: https://lore.kernel.org/r/20230531083230.2702181-1-u.kleine-koenig@xxxxxxxxxxxxxx
> > > > > Signed-off-by: Uwe Kleine-König <u.kleine-koenig@xxxxxxxxxxxxxx>
> > > > > ---
> > > > >  drivers/tty/serial/8250/Kconfig  | 2 +-
> > > > >  drivers/tty/serial/8250/Makefile | 2 +-
> > > > >  2 files changed, 2 insertions(+), 2 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/tty/serial/8250/Kconfig b/drivers/tty/serial/8250/Kconfig
> > > > > index 5313aa31930f..10c09b19c871 100644
> > > > > --- a/drivers/tty/serial/8250/Kconfig
> > > > > +++ b/drivers/tty/serial/8250/Kconfig
> > > > > @@ -378,7 +378,7 @@ config SERIAL_8250_BCM2835AUX
> > > > >  
> > > > >  config SERIAL_8250_FSL
> > > > >  	bool "Freescale 16550 UART support" if COMPILE_TEST && !(PPC || ARM || ARM64)
> > > > > -	depends on SERIAL_8250_CONSOLE
> > > > > +	depends on SERIAL_8250
> > > > 
> > > > Just one additional thought: After the adding the arch side 
> > > > workaround/hack, SERIAL_8250_FSL could become a tristate?
> > > 
> > > I see no benefit for a module separate from 8250_base.ko. There are
> > > dependencies in both directions between 8250_port.o and 8250_fsl.o[1].
> > > So in my book a bool SERIAL_8250_FSL that modifies 8250_base.ko (with
> > > SERIAL_8250=m) is fine.
> > > 
> > > Best regards
> > > Uwe
> > > 
> > > [1] 8250_port.o uses fsl8250_handle_irq() from 8250_fsl.o
> > 
> > Is that after some fix which isn't in tty-next? I see only these:
> > 
> > $ git grep -l fsl8250_handle_irq
> > arch/powerpc/kernel/legacy_serial.c
> > drivers/tty/serial/8250/8250_fsl.c
> > drivers/tty/serial/8250/8250_of.c
> > include/linux/serial_8250.h
> > 
> > No users of fsl8250_handle_irq in 8250_port.c.
> 
> Ah right, I was too quick:
> 
> 	8250_of.o uses fsl8250_handle_irq() from 8250_fsl.o
> 	8250_fsl.o uses serial8250_modem_status() from 8250_port.o (which is in 8250_base.o)
> 
> 
> However linking 8250_fsl.o in 8250_of.o isn't a good solution either as
> 8250_fsl.o should also be available with CONFIG_SERIAL_OF_PLATFORM
> disabled to provide the ACPI driver. And as 8250_of.o already depends on
> 8250_port.o (e.g. via serial8250_em485_config()) adding 8250_fsl.o
> together with 8250_port.o into 8250_base.ko is fine and doesn't add new
> dependencies.

So we have dependencies one-way only:

8250_port

/\    |\
        \
8250_fsl \
         /
/\      /

8250_of

There's no loop here, both they be indepedent modules and configured 
independently (with a correct IS_*() in 8250_of.c).

-- 
 i.