Re: [PATCH] drivers: soc: xilinx: fix firmware driver Kconfig dependency

From: Arnd Bergmann
Date: Thu Apr 09 2020 - 05:09:59 EST


On Thu, Apr 9, 2020 at 8:37 AM Michal Simek <michal.simek@xxxxxxxxxx> wrote:
>
> On 08. 04. 20 17:52, Arnd Bergmann wrote:
> > The firmware driver is optional, but the power driver depends on it,
> > which needs to be reflected in Kconfig to avoid link errors:
> >
> > aarch64-linux-ld: drivers/soc/xilinx/zynqmp_power.o: in function `zynqmp_pm_isr':
> > zynqmp_power.c:(.text+0x284): undefined reference to `zynqmp_pm_invoke_fn'
> >
> > The firmware driver can probably be allowed for compile-testing as
> > well, so it's best to drop the dependency on the ZYNQ platform
> > here and allow building as long as the firmware code is built-in.
> >
> > Fixes: ab272643d723 ("drivers: soc: xilinx: Add ZynqMP PM driver")
> > Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
> > ---
> > drivers/soc/xilinx/Kconfig | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/soc/xilinx/Kconfig b/drivers/soc/xilinx/Kconfig
> > index 223f1f9d0922..646512d7276f 100644
> > --- a/drivers/soc/xilinx/Kconfig
> > +++ b/drivers/soc/xilinx/Kconfig
> > @@ -19,7 +19,7 @@ config XILINX_VCU
> >
> > config ZYNQMP_POWER
> > bool "Enable Xilinx Zynq MPSoC Power Management driver"
> > - depends on PM && ARCH_ZYNQMP
> > + depends on PM && ZYNQMP_FIRMWARE
> > default y
> > select MAILBOX
> > select ZYNQMP_IPI_MBOX
> > @@ -35,7 +35,7 @@ config ZYNQMP_POWER
> > config ZYNQMP_PM_DOMAINS
> > bool "Enable Zynq MPSoC generic PM domains"
> > default y
> > - depends on PM && ARCH_ZYNQMP && ZYNQMP_FIRMWARE
> > + depends on PM && ZYNQMP_FIRMWARE
> > select PM_GENERIC_DOMAINS
> > help
> > Say yes to enable device power management through PM domains
> >
>
> The same issue is likely with others drivers dependencies too which
> depends on ARCH_ZYNQMP.
>
> It means all drivers which includes "linux/firmware/xlnx-zynqmp.h" and
> call zynqmp_pm_get_eemi_ops() should depend on ZYNQMP_FIRMWARE instead
> of ARCH_ZYNQMP.

The only one I see that has a hard dependency on ARCH_ZYNQMP
without allowing compile-testing at the moment is drivers/edac/synopsys_edac.c
but that doesn't use the firmware interface.

What I see in the header are declarations for exported functions:

int zynqmp_pm_invoke_fn(u32 pm_api_id, u32 arg0, u32 arg1,
u32 arg2, u32 arg3, u32 *ret_payload);
#if IS_REACHABLE(CONFIG_ZYNQMP_FIRMWARE)
const struct zynqmp_eemi_ops *zynqmp_pm_get_eemi_ops(void);
#else
static inline struct zynqmp_eemi_ops *zynqmp_pm_get_eemi_ops(void)
{
return ERR_PTR(-ENODEV);
}
#endif

The second one already allows compile-testing by turning into an
inline stub, but zynqmp_pm_invoke_fn() does not, and this is the
one causing the problem here.

I still think my patch is a good fix for that issue, but if you want to
handle both interfaces the same way, we can also do that, either
removing the stub and using a proper dependency, or using
the same stub trick for both.

Arnd