Re: [PATCH] net: broadcom: Add PTP_1588_CLOCK_OPTIONAL dependency for BCMGENET under ARCH_BCM2835

[Arnd Bergmann]

On Tue, Nov 29, 2022, at 04:18, Jakub Kicinski wrote:
> On Fri, 25 Nov 2022 19:50:03 +0800 YueHaibing wrote:
>> diff --git a/drivers/net/ethernet/broadcom/Kconfig b/drivers/net/ethernet/broadcom/Kconfig
>> index 55dfdb34e37b..f4ca0c6c0f51 100644
>> --- a/drivers/net/ethernet/broadcom/Kconfig
>> +++ b/drivers/net/ethernet/broadcom/Kconfig
>> @@ -71,13 +71,14 @@ config BCM63XX_ENET
>>  config BCMGENET
>>  	tristate "Broadcom GENET internal MAC support"
>>  	depends on HAS_IOMEM
>> +	depends on PTP_1588_CLOCK_OPTIONAL || !ARCH_BCM2835
>>  	select MII
>>  	select PHYLIB
>>  	select FIXED_PHY
>>  	select BCM7XXX_PHY
>>  	select MDIO_BCM_UNIMAC
>>  	select DIMLIB
>> -	select BROADCOM_PHY if (ARCH_BCM2835 && PTP_1588_CLOCK_OPTIONAL)
>> +	select BROADCOM_PHY if ARCH_BCM2835
>>  	help
>>  	  This driver supports the built-in Ethernet MACs found in the
>>  	  Broadcom BCM7xxx Set Top Box family chipset.
>
> What's the code path that leads to the failure? I want to double check
> that the driver is handling the PTP registration return codes correctly.
> IIUC this is a source of misunderstandings in the PTP API.
>
> Richard, here's the original report:
> https://lore.kernel.org/all/CA+G9fYvKfbJHcMZtybf_0Ru3+6fKPg9HwWTOhdCLrOBXMaeG1A@xxxxxxxxxxxxxx

The original report was for a different bug that resulted in the
BROADCOM_PHY driver not being selectable at all.

The remaining problem here is this configuration:

CONFIG_ARM=y
CONFIG_BCM2835=y
CONFIG_BCMGENET=y
CONFIG_PTP_1588_CLOCK=m
CONFIG_PTP_1588_CLOCK_OPTIONAL=m
CONFIG_BROADCOM_PHY=m

In this case, BCMGENET should 'select BROADCOM_PHY' to make the
driver work correctly, but fails to do this because of the
dependency. During early boot, this means it cannot access the
PHY because that is in a loadable module, despite commit
99addbe31f55 ("net: broadcom: Select BROADCOM_PHY for BCMGENET")
trying to ensure that it could.

Note that many other ethernet drivers don't have this
particular 'select' statement and just rely on the .config
to contain a sensible set of drivers. In particular that
is true when running 64-bit kernels on the same chip,
which is now the normal configuration.

The alternative to YueHaibing's fix would be to just revert
99addbe31f55 ("net: broadcom: Select BROADCOM_PHY for BCMGENET")
and instead change the defconfig file to include the phy driver,
as we do elsewhere.

    Arnd