Re: [Qemu-trivial] [Qemu-arm] [PATCH 2/4] hw/core: Add a config switch for the "or-irq" device

[Philippe Mathieu-Daudé]

On 5/14/19 12:25 PM, Philippe Mathieu-Daudé wrote:
> On 5/14/19 12:06 PM, Peter Maydell wrote:
>> On Tue, 14 May 2019 at 11:00, Thomas Huth <address@hidden> wrote:
>>>
>>> The "or-irq" device is only used by certain machines. Let's add
>>> a proper config switch for it so that it only gets compiled when we
>>> really need it.
>>>
>>> Signed-off-by: Thomas Huth <address@hidden>
>>> ---
>>>  hw/arm/Kconfig        | 2 ++
>>>  hw/core/Kconfig       | 3 +++
>>>  hw/core/Makefile.objs | 2 +-
>>>  hw/pci-host/Kconfig   | 3 ++-
>>>  4 files changed, 8 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/hw/arm/Kconfig b/hw/arm/Kconfig
>>> index af8cffde9c..0bb3bbe9d3 100644
>>> --- a/hw/arm/Kconfig
>>> +++ b/hw/arm/Kconfig
>>> @@ -277,6 +277,7 @@ config RASPI
>>>  config STM32F205_SOC
>>>      bool
>>>      select ARM_V7M
>>> +    select OR_IRQ
>>>      select STM32F2XX_TIMER
>>>      select STM32F2XX_USART
>>>      select STM32F2XX_SYSCFG
>>> @@ -424,6 +425,7 @@ config ARMSSE
>>>      select IOTKIT_SECCTL
>>>      select IOTKIT_SYSCTL
>>>      select IOTKIT_SYSINFO
>>> +    select OR_IRQ
>>>      select TZ_MPC
>>>      select TZ_MSC
>>>      select TZ_PPC
>>
>> In cases like this where a device is used both by
>> an SoC and also directly by the board code that uses
>> that SoC, should we put the select OR_IRQ only in
>> the SoC's config, or also in the board model's config
>> (ie, in "config MPS2" as well as "config ARMSSE") ?
> 
> Someone should be able to work on the board without having to look at
> the SoC code/config, so both :) The idea of Kconfig is you only worry
> about a specific device, and the qgraph sort the rest out.

Hypothetical example if you only use the selector in the SoC config:
If you replace the SoC OR_IRQ by a more complex device or extended one,
the board will lack the OR_IRQ selector.

Using Kconfig selectors in all place a dependency is explicit also ease
backports.

> 
> So having in both place is safer, and helps to visualize dependencies in
> the graph tree (I'm slowly working on this feature to help new-comer to
> understand model dependencies).
>