Hi Bryan,

On Wed, May 04, 2016 at 10:51:17AM +0100, Bryan O'Donoghue wrote:

On Wed, 2016-05-04 at 12:42 +0300, Andy Shevchenko wrote:

quoted

On Wed, May 4, 2016 at 12:31 PM, Bryan O'Donoghue
[off-list ref] wrote:

quoted

Andy,

If you are going to start removing working PCI devices from the PCI
config table in favour of a shim in SERIAL_8250_LPSS then the very
minimum should be some sort of dependency link between
SERIAL_8250_LPSS
and CONFIG_SERIAL_8250_PCI in kconfig.

A user could reasonably read the QRK datasheet - switch on
CONFIG_SERIAL_8250_PCI and then wonder why no console output
happened
on boot. S/he shouldn't have to know that devices were moved from
the
PCI driver to an LPSS shim driver or that the 8250_lpss driver now
needs to be selected instead of the intuitively correct 8250_pci
driver.

That is taken care of since default is set to SERIAL_8250 (you even
don't need to have PCI driver enabled!).
Doesn't work for you?

The default may be set to SERIAL_8250 but, without the QRK specific
entry in 8250_pci.c you won't get console output.

So if you are going to remove the QRK entry from 8250_pci.c and stuff
it into 8250_lpss.c then 8250_lpss needs to be selected by
CONFIG_SERIAL_8250_PCI.

Otherwise the person doing the config needs to know that stuff was
moved from one file to another - even though it's a PCI device (not an
LPSS/ACPI enumerated device) - which seems like an unreasonable level
of knowledge to assume on the part of the user.

The only way the user get's to de-select CONFIG_SERIAL_8250_PCI or
CONFIG_SERIAL_8250_LPSS is if S/he also selects CONFIG_EXPERT, and if
S/he does that then we can assume S/he has the knowledge.

I'm against binding these separated drivers to CONFIG_SERIAL_8250_PCI
because doing that will very fast mean that we also remove the
possibility to de-select CONFIG_SERIAL_8250_PCI when only, for
example, CONFIG_SERIAL_8250_LPSS was wanted.


Thanks,

-- 
heikki