drivers/gpu/drm/bridge/samsung-dsim.c link error

[Randy Dunlap]

On 8/3/23 19:26, Adam Ford wrote:
> On Thu, Aug 3, 2023 at 9:01 PM Randy Dunlap <rdunlap at infradead.org> wrote:
>>
>> Hi--
>>
>> On 8/3/23 08:24, Adam Ford wrote:
>>> On Thu, Aug 3, 2023 at 9:42 AM Aleksandr Nogikh <nogikh at google.com> wrote:
>>>>
>>>> Hello,
>>>>
>>>> We've been seing the following linker error on arm64 syzbot instances:
>>>>
>>>> ld.lld: error: undefined symbol: phy_mipi_dphy_get_default_config_for_hsclk
>>>>>>> referenced by samsung-dsim.c:731 (drivers/gpu/drm/bridge/samsung-dsim.c:731)
>>>>>>>               drivers/gpu/drm/bridge/samsung-dsim.o:(samsung_dsim_init) in archive vmlinux.a
>>>>
>>>> Steps to reproduce on the latest linux-next:
>>>>
>>>> $ git checkout next-20230803
>>>> $ wget -O '.config' 'https://raw.githubusercontent.com/google/syzkaller/master/dashboard/config/linux/upstream-arm64-kasan.config'
>>>> $ make CC=clang ARCH=arm64 LD=ld.lld CROSS_COMPILE=aarch64-linux-gnu- olddefconfig
>>>>
>>>> This also prints:
>>>>
>>>> WARNING: unmet direct dependencies detected for GENERIC_PHY_MIPI_DPHY
>>>>   Depends on [n]: GENERIC_PHY [=n]
>>>>   Selected by [y]:
>>>>   - DRM_NWL_MIPI_DSI [=y] && DRM_BRIDGE [=y] && DRM [=y] && COMMON_CLK [=y] && OF [=y] && HAS_IOMEM [=y]
>>>>   - DRM_SAMSUNG_DSIM [=y] && DRM [=y] && DRM_BRIDGE [=y] && COMMON_CLK [=y] && OF [=y] && HAS_IOMEM [=y]
>>>>
>>>> $ make CC=clang ARCH=arm64 LD=ld.lld CROSS_COMPILE=aarch64-linux-gnu- -j$(nproc)
>>>>
>>>> The kernel should have still compiled fine even despite the message above, right?
>>
>> No. See drivers/Makefile:
>> obj-$(CONFIG_GENERIC_PHY)       += phy/
>>
>> so the drivers/phy/ subdir is only built when CONFIG_GENERIC_PHY is set,
>> but it's not set.
>>
>>>> Could you please take a look?
>>>
>>> GENERIC_PHY_MIPI_DPHY was enabled to use
>>> phy_mipi_dphy_get_default_config() which takes in the pixel clock,
>>> bits-per-pixel, number of lanes and phy structure.  It applies a bunch
>>> of math based on the info passed and fills in the structure, but that
>>> function itself doesn't appear to be referencing phy code, so it's
>>> likely safe.
>>>
>>> I think this can be resolved by enabling GENERIC_PHY.  I just checked
>>> linux-next and when I built the arm64 defconfig, it enables
>>> GENERIC_PHY=y, so I don't think this is an issue.  I also checked the
>>
>> defconfig doesn't matter in this case. The .config that was provided has:
>> # CONFIG_GENERIC_PHY is not set
>> CONFIG_GENERIC_PHY_MIPI_DPHY=y
>>
> 
> Where/how was the .config generated?
> 

Aleksandr posted a link to the config file above.

> Are you building linux-next or something else?  The .config file
> generated when I build the arm64 defconfig  show both enabled:

linux-next.


> $ grep GENERIC_PHY .config
> CONFIG_GENERIC_PHY=y
> CONFIG_GENERIC_PHY_MIPI_DPHY=y
> 

Yes, this is not a defconfig file.

> 
>> but yes, selecting GENERIC_PHY (needed in 3 places) does fix the warnings
>> and build error.  2 instance in drm/bridge/Kconfig and one in
>> drm/bridge/cadence/Kconfig (found by inspection).
>>
>>
>>
>>> multi_v7_defconfig for ARM and it also sets GENERIC_PHY=y for 32-bit
>>> ARM people using some of the Exynos boards.
>>>
>>> I don't know what version of Linux you're trying to build, but I can't
>>> replicate your issue.
>>
>> I had no problem replicating the kconfig warnings and build error.
> 
> If you can replicate the problem, I'd suggest submitting a patch.

Sure, I'll do that.

-- 
~Randy