Re: [PATCH 03/22] arm: switch boards to "default y"

[Fabiano Rosas]

Philippe Mathieu-Daudé <philmd@linaro.org> writes:

> Hi Fabiano,
>
> On 23/4/24 20:02, Fabiano Rosas wrote:
>> Paolo Bonzini <pbonzini@redhat.com> writes:
>> 
>>> For ARM targets, boards that require TCG are already using "default y".
>>> Switch ARM_VIRT to the same selection mechanism.
>>>
>>> No changes to generated config-devices.mak file.
>>>
>>> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>>> ---
>>>   configs/devices/arm-softmmu/default.mak | 3 ++-
>>>   hw/arm/Kconfig                          | 2 ++
>>>   2 files changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/configs/devices/arm-softmmu/default.mak 
>>> b/configs/devices/arm-softmmu/default.mak
>>> index c1cfb3bcf75..31f77c20269 100644
>>> --- a/configs/devices/arm-softmmu/default.mak
>>> +++ b/configs/devices/arm-softmmu/default.mak
>>> @@ -5,7 +5,8 @@
>>>   # CONFIG_PCI_DEVICES=n
>>>   # CONFIG_TEST_DEVICES=n
>>>   
>>> -CONFIG_ARM_VIRT=y
>>> +# Boards are selected by default, uncomment to keep out of the build.
>>> +# CONFIG_ARM_VIRT=n
>>>   
>>>   # These are selected by default when TCG is enabled, uncomment them to
>>>   # keep out of the build.
>>> diff --git a/hw/arm/Kconfig b/hw/arm/Kconfig
>>> index 893a7bff66b..1e7cd01087f 100644
>>> --- a/hw/arm/Kconfig
>>> +++ b/hw/arm/Kconfig
>>> @@ -1,5 +1,7 @@
>>>   config ARM_VIRT
>>>       bool
>>> +    default y
>>> +    depends on ARM
>>>       imply PCI_DEVICES
>> 
>> We lose pci.c when building --without-default-devices:
>> 
>> $ arch
>> aarch64
>> $ ../configure --target-list=aarch64-softmmu,arm-softmmu
>> --disable-linux-user --without-default-devices
>> $ make
>> ...
>> libqemu-aarch64-softmmu.fa.p/target_arm_kvm.c.o: in function 
>> `kvm_arch_fixup_msi_route':
>> ../target/arm/kvm.c:1548: undefined reference to 
>> `pci_device_iommu_address_space'
>
> IIRC QEMU isn't yet ready to be built with --without-default-devices
> but without any machine, you need to explicit at least one. For
> example you can use --with-devices-aarch64=minimal, or define a
> config with only the Virt machine.

Could we detect this kind of breakage somehow and report it properly to
the user? As in "you chose an impossible configuration, select a board
and try again".

I see this comes from kvm.c, perhaps on arm CONFIG_KVM needs to depend
on VIRT?

>
> Regards,
>
> Phil.